alistair23-linux/drivers/scsi/lpfc/lpfc_sli.c

19320 lines
588 KiB
C
Raw Normal View History

/*******************************************************************
* This file is part of the Emulex Linux Device Driver for *
* Fibre Channel Host Bus Adapters. *
* Copyright (C) 2017-2018 Broadcom. All Rights Reserved. The term *
* Broadcom refers to Broadcom Limited and/or its subsidiaries. *
* Copyright (C) 2004-2016 Emulex. All rights reserved. *
* EMULEX and SLI are trademarks of Emulex. *
* www.broadcom.com *
* Portions Copyright (C) 2004-2005 Christoph Hellwig *
* *
* This program is free software; you can redistribute it and/or *
* modify it under the terms of version 2 of the GNU General *
* Public License as published by the Free Software Foundation. *
* This program is distributed in the hope that it will be useful. *
* ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND *
* WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE *
* DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
* TO BE LEGALLY INVALID. See the GNU General Public License for *
* more details, a copy of which can be found in the file COPYING *
* included with this package. *
*******************************************************************/
#include <linux/blkdev.h>
#include <linux/pci.h>
#include <linux/interrupt.h>
#include <linux/delay.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 02:04:11 -06:00
#include <linux/slab.h>
#include <linux/lockdep.h>
#include <scsi/scsi.h>
#include <scsi/scsi_cmnd.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi_transport_fc.h>
#include <scsi/fc/fc_fs.h>
#include <linux/aer.h>
#ifdef CONFIG_X86
#include <asm/set_memory.h>
#endif
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
#include <linux/nvme-fc-driver.h>
#include "lpfc_hw4.h"
#include "lpfc_hw.h"
#include "lpfc_sli.h"
#include "lpfc_sli4.h"
#include "lpfc_nl.h"
#include "lpfc_disc.h"
#include "lpfc.h"
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
#include "lpfc_scsi.h"
#include "lpfc_nvme.h"
#include "lpfc_nvmet.h"
#include "lpfc_crtn.h"
#include "lpfc_logmsg.h"
#include "lpfc_compat.h"
#include "lpfc_debugfs.h"
#include "lpfc_vport.h"
#include "lpfc_version.h"
/* There are only four IOCB completion types. */
typedef enum _lpfc_iocb_type {
LPFC_UNKNOWN_IOCB,
LPFC_UNSOL_IOCB,
LPFC_SOL_IOCB,
LPFC_ABORT_IOCB
} lpfc_iocb_type;
/* Provide function prototypes local to this module. */
static int lpfc_sli_issue_mbox_s4(struct lpfc_hba *, LPFC_MBOXQ_t *,
uint32_t);
static int lpfc_sli4_read_rev(struct lpfc_hba *, LPFC_MBOXQ_t *,
uint8_t *, uint32_t *);
static struct lpfc_iocbq *lpfc_sli4_els_wcqe_to_rspiocbq(struct lpfc_hba *,
struct lpfc_iocbq *);
static void lpfc_sli4_send_seq_to_ulp(struct lpfc_vport *,
struct hbq_dmabuf *);
static void lpfc_sli4_handle_mds_loopback(struct lpfc_vport *vport,
struct hbq_dmabuf *dmabuf);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
static int lpfc_sli4_fp_handle_cqe(struct lpfc_hba *, struct lpfc_queue *,
struct lpfc_cqe *);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
static int lpfc_sli4_post_sgl_list(struct lpfc_hba *, struct list_head *,
int);
static void lpfc_sli4_hba_handle_eqe(struct lpfc_hba *phba,
struct lpfc_eqe *eqe, uint32_t qidx);
static bool lpfc_sli4_mbox_completions_pending(struct lpfc_hba *phba);
static bool lpfc_sli4_process_missed_mbox_completions(struct lpfc_hba *phba);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
static int lpfc_sli4_abort_nvme_io(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring,
struct lpfc_iocbq *cmdiocb);
static IOCB_t *
lpfc_get_iocb_from_iocbq(struct lpfc_iocbq *iocbq)
{
return &iocbq->iocb;
}
/**
* lpfc_sli4_wq_put - Put a Work Queue Entry on an Work Queue
* @q: The Work Queue to operate on.
* @wqe: The work Queue Entry to put on the Work queue.
*
* This routine will copy the contents of @wqe to the next available entry on
* the @q. This function will then ring the Work Queue Doorbell to signal the
* HBA to start processing the Work Queue Entry. This function returns 0 if
* successful. If no entries are available on @q then this function will return
* -ENOMEM.
* The caller is expected to hold the hbalock when calling this routine.
**/
static int
lpfc_sli4_wq_put(struct lpfc_queue *q, union lpfc_wqe128 *wqe)
{
union lpfc_wqe *temp_wqe;
struct lpfc_register doorbell;
uint32_t host_index;
uint32_t idx;
uint32_t i = 0;
uint8_t *tmp;
/* sanity check on queue memory */
if (unlikely(!q))
return -ENOMEM;
temp_wqe = q->qe[q->host_index].wqe;
/* If the host has not yet processed the next entry then we are done */
idx = ((q->host_index + 1) % q->entry_count);
if (idx == q->hba_index) {
q->WQ_overflow++;
return -EBUSY;
}
q->WQ_posted++;
/* set consumption flag every once in a while */
if (!((q->host_index + 1) % q->entry_repost))
bf_set(wqe_wqec, &wqe->generic.wqe_com, 1);
else
bf_set(wqe_wqec, &wqe->generic.wqe_com, 0);
if (q->phba->sli3_options & LPFC_SLI4_PHWQ_ENABLED)
bf_set(wqe_wqid, &wqe->generic.wqe_com, q->queue_id);
lpfc_sli_pcimem_bcopy(wqe, temp_wqe, q->entry_size);
if (q->dpp_enable && q->phba->cfg_enable_dpp) {
/* write to DPP aperture taking advatage of Combined Writes */
tmp = (uint8_t *)temp_wqe;
#ifdef __raw_writeq
for (i = 0; i < q->entry_size; i += sizeof(uint64_t))
__raw_writeq(*((uint64_t *)(tmp + i)),
q->dpp_regaddr + i);
#else
for (i = 0; i < q->entry_size; i += sizeof(uint32_t))
__raw_writel(*((uint32_t *)(tmp + i)),
q->dpp_regaddr + i);
#endif
}
/* ensure WQE bcopy and DPP flushed before doorbell write */
wmb();
/* Update the host index before invoking device */
host_index = q->host_index;
q->host_index = idx;
/* Ring Doorbell */
doorbell.word0 = 0;
if (q->db_format == LPFC_DB_LIST_FORMAT) {
if (q->dpp_enable && q->phba->cfg_enable_dpp) {
bf_set(lpfc_if6_wq_db_list_fm_num_posted, &doorbell, 1);
bf_set(lpfc_if6_wq_db_list_fm_dpp, &doorbell, 1);
bf_set(lpfc_if6_wq_db_list_fm_dpp_id, &doorbell,
q->dpp_id);
bf_set(lpfc_if6_wq_db_list_fm_id, &doorbell,
q->queue_id);
} else {
bf_set(lpfc_wq_db_list_fm_num_posted, &doorbell, 1);
bf_set(lpfc_wq_db_list_fm_index, &doorbell, host_index);
bf_set(lpfc_wq_db_list_fm_id, &doorbell, q->queue_id);
}
} else if (q->db_format == LPFC_DB_RING_FORMAT) {
bf_set(lpfc_wq_db_ring_fm_num_posted, &doorbell, 1);
bf_set(lpfc_wq_db_ring_fm_id, &doorbell, q->queue_id);
} else {
return -EINVAL;
}
writel(doorbell.word0, q->db_regaddr);
return 0;
}
/**
* lpfc_sli4_wq_release - Updates internal hba index for WQ
* @q: The Work Queue to operate on.
* @index: The index to advance the hba index to.
*
* This routine will update the HBA index of a queue to reflect consumption of
* Work Queue Entries by the HBA. When the HBA indicates that it has consumed
* an entry the host calls this function to update the queue's internal
* pointers. This routine returns the number of entries that were consumed by
* the HBA.
**/
static uint32_t
lpfc_sli4_wq_release(struct lpfc_queue *q, uint32_t index)
{
uint32_t released = 0;
/* sanity check on queue memory */
if (unlikely(!q))
return 0;
if (q->hba_index == index)
return 0;
do {
q->hba_index = ((q->hba_index + 1) % q->entry_count);
released++;
} while (q->hba_index != index);
return released;
}
/**
* lpfc_sli4_mq_put - Put a Mailbox Queue Entry on an Mailbox Queue
* @q: The Mailbox Queue to operate on.
* @wqe: The Mailbox Queue Entry to put on the Work queue.
*
* This routine will copy the contents of @mqe to the next available entry on
* the @q. This function will then ring the Work Queue Doorbell to signal the
* HBA to start processing the Work Queue Entry. This function returns 0 if
* successful. If no entries are available on @q then this function will return
* -ENOMEM.
* The caller is expected to hold the hbalock when calling this routine.
**/
static uint32_t
lpfc_sli4_mq_put(struct lpfc_queue *q, struct lpfc_mqe *mqe)
{
struct lpfc_mqe *temp_mqe;
struct lpfc_register doorbell;
/* sanity check on queue memory */
if (unlikely(!q))
return -ENOMEM;
temp_mqe = q->qe[q->host_index].mqe;
/* If the host has not yet processed the next entry then we are done */
if (((q->host_index + 1) % q->entry_count) == q->hba_index)
return -ENOMEM;
lpfc_sli_pcimem_bcopy(mqe, temp_mqe, q->entry_size);
/* Save off the mailbox pointer for completion */
q->phba->mbox = (MAILBOX_t *)temp_mqe;
/* Update the host index before invoking device */
q->host_index = ((q->host_index + 1) % q->entry_count);
/* Ring Doorbell */
doorbell.word0 = 0;
bf_set(lpfc_mq_doorbell_num_posted, &doorbell, 1);
bf_set(lpfc_mq_doorbell_id, &doorbell, q->queue_id);
writel(doorbell.word0, q->phba->sli4_hba.MQDBregaddr);
return 0;
}
/**
* lpfc_sli4_mq_release - Updates internal hba index for MQ
* @q: The Mailbox Queue to operate on.
*
* This routine will update the HBA index of a queue to reflect consumption of
* a Mailbox Queue Entry by the HBA. When the HBA indicates that it has consumed
* an entry the host calls this function to update the queue's internal
* pointers. This routine returns the number of entries that were consumed by
* the HBA.
**/
static uint32_t
lpfc_sli4_mq_release(struct lpfc_queue *q)
{
/* sanity check on queue memory */
if (unlikely(!q))
return 0;
/* Clear the mailbox pointer for completion */
q->phba->mbox = NULL;
q->hba_index = ((q->hba_index + 1) % q->entry_count);
return 1;
}
/**
* lpfc_sli4_eq_get - Gets the next valid EQE from a EQ
* @q: The Event Queue to get the first valid EQE from
*
* This routine will get the first valid Event Queue Entry from @q, update
* the queue's internal hba index, and return the EQE. If no valid EQEs are in
* the Queue (no more work to do), or the Queue is full of EQEs that have been
* processed, but not popped back to the HBA then this routine will return NULL.
**/
static struct lpfc_eqe *
lpfc_sli4_eq_get(struct lpfc_queue *q)
{
struct lpfc_hba *phba;
struct lpfc_eqe *eqe;
uint32_t idx;
/* sanity check on queue memory */
if (unlikely(!q))
return NULL;
phba = q->phba;
eqe = q->qe[q->hba_index].eqe;
/* If the next EQE is not valid then we are done */
if (bf_get_le32(lpfc_eqe_valid, eqe) != q->qe_valid)
return NULL;
/* If the host has not yet processed the next entry then we are done */
idx = ((q->hba_index + 1) % q->entry_count);
if (idx == q->host_index)
return NULL;
q->hba_index = idx;
/* if the index wrapped around, toggle the valid bit */
if (phba->sli4_hba.pc_sli4_params.eqav && !q->hba_index)
q->qe_valid = (q->qe_valid) ? 0 : 1;
/*
* insert barrier for instruction interlock : data from the hardware
* must have the valid bit checked before it can be copied and acted
* upon. Speculative instructions were allowing a bcopy at the start
* of lpfc_sli4_fp_handle_wcqe(), which is called immediately
* after our return, to copy data before the valid bit check above
* was done. As such, some of the copied data was stale. The barrier
* ensures the check is before any data is copied.
*/
mb();
return eqe;
}
/**
* lpfc_sli4_eq_clr_intr - Turn off interrupts from this EQ
* @q: The Event Queue to disable interrupts
*
**/
inline void
lpfc_sli4_eq_clr_intr(struct lpfc_queue *q)
{
struct lpfc_register doorbell;
doorbell.word0 = 0;
bf_set(lpfc_eqcq_doorbell_eqci, &doorbell, 1);
bf_set(lpfc_eqcq_doorbell_qt, &doorbell, LPFC_QUEUE_TYPE_EVENT);
bf_set(lpfc_eqcq_doorbell_eqid_hi, &doorbell,
(q->queue_id >> LPFC_EQID_HI_FIELD_SHIFT));
bf_set(lpfc_eqcq_doorbell_eqid_lo, &doorbell, q->queue_id);
writel(doorbell.word0, q->phba->sli4_hba.EQDBregaddr);
}
/**
* lpfc_sli4_if6_eq_clr_intr - Turn off interrupts from this EQ
* @q: The Event Queue to disable interrupts
*
**/
inline void
lpfc_sli4_if6_eq_clr_intr(struct lpfc_queue *q)
{
struct lpfc_register doorbell;
doorbell.word0 = 0;
bf_set(lpfc_eqcq_doorbell_eqci, &doorbell, 1);
bf_set(lpfc_eqcq_doorbell_qt, &doorbell, LPFC_QUEUE_TYPE_EVENT);
bf_set(lpfc_eqcq_doorbell_eqid_hi, &doorbell,
(q->queue_id >> LPFC_EQID_HI_FIELD_SHIFT));
bf_set(lpfc_eqcq_doorbell_eqid_lo, &doorbell, q->queue_id);
writel(doorbell.word0, q->phba->sli4_hba.EQDBregaddr);
}
/**
* lpfc_sli4_eq_release - Indicates the host has finished processing an EQ
* @q: The Event Queue that the host has completed processing for.
* @arm: Indicates whether the host wants to arms this CQ.
*
* This routine will mark all Event Queue Entries on @q, from the last
* known completed entry to the last entry that was processed, as completed
* by clearing the valid bit for each completion queue entry. Then it will
* notify the HBA, by ringing the doorbell, that the EQEs have been processed.
* The internal host index in the @q will be updated by this routine to indicate
* that the host has finished processing the entries. The @arm parameter
* indicates that the queue should be rearmed when ringing the doorbell.
*
* This function will return the number of EQEs that were popped.
**/
uint32_t
lpfc_sli4_eq_release(struct lpfc_queue *q, bool arm)
{
uint32_t released = 0;
struct lpfc_hba *phba;
struct lpfc_eqe *temp_eqe;
struct lpfc_register doorbell;
/* sanity check on queue memory */
if (unlikely(!q))
return 0;
phba = q->phba;
/* while there are valid entries */
while (q->hba_index != q->host_index) {
if (!phba->sli4_hba.pc_sli4_params.eqav) {
temp_eqe = q->qe[q->host_index].eqe;
bf_set_le32(lpfc_eqe_valid, temp_eqe, 0);
}
released++;
q->host_index = ((q->host_index + 1) % q->entry_count);
}
if (unlikely(released == 0 && !arm))
return 0;
/* ring doorbell for number popped */
doorbell.word0 = 0;
if (arm) {
bf_set(lpfc_eqcq_doorbell_arm, &doorbell, 1);
bf_set(lpfc_eqcq_doorbell_eqci, &doorbell, 1);
}
bf_set(lpfc_eqcq_doorbell_num_released, &doorbell, released);
bf_set(lpfc_eqcq_doorbell_qt, &doorbell, LPFC_QUEUE_TYPE_EVENT);
bf_set(lpfc_eqcq_doorbell_eqid_hi, &doorbell,
(q->queue_id >> LPFC_EQID_HI_FIELD_SHIFT));
bf_set(lpfc_eqcq_doorbell_eqid_lo, &doorbell, q->queue_id);
writel(doorbell.word0, q->phba->sli4_hba.EQDBregaddr);
/* PCI read to flush PCI pipeline on re-arming for INTx mode */
if ((q->phba->intr_type == INTx) && (arm == LPFC_QUEUE_REARM))
readl(q->phba->sli4_hba.EQDBregaddr);
return released;
}
/**
* lpfc_sli4_if6_eq_release - Indicates the host has finished processing an EQ
* @q: The Event Queue that the host has completed processing for.
* @arm: Indicates whether the host wants to arms this CQ.
*
* This routine will mark all Event Queue Entries on @q, from the last
* known completed entry to the last entry that was processed, as completed
* by clearing the valid bit for each completion queue entry. Then it will
* notify the HBA, by ringing the doorbell, that the EQEs have been processed.
* The internal host index in the @q will be updated by this routine to indicate
* that the host has finished processing the entries. The @arm parameter
* indicates that the queue should be rearmed when ringing the doorbell.
*
* This function will return the number of EQEs that were popped.
**/
uint32_t
lpfc_sli4_if6_eq_release(struct lpfc_queue *q, bool arm)
{
uint32_t released = 0;
struct lpfc_hba *phba;
struct lpfc_eqe *temp_eqe;
struct lpfc_register doorbell;
/* sanity check on queue memory */
if (unlikely(!q))
return 0;
phba = q->phba;
/* while there are valid entries */
while (q->hba_index != q->host_index) {
if (!phba->sli4_hba.pc_sli4_params.eqav) {
temp_eqe = q->qe[q->host_index].eqe;
bf_set_le32(lpfc_eqe_valid, temp_eqe, 0);
}
released++;
q->host_index = ((q->host_index + 1) % q->entry_count);
}
if (unlikely(released == 0 && !arm))
return 0;
/* ring doorbell for number popped */
doorbell.word0 = 0;
if (arm)
bf_set(lpfc_if6_eq_doorbell_arm, &doorbell, 1);
bf_set(lpfc_if6_eq_doorbell_num_released, &doorbell, released);
bf_set(lpfc_if6_eq_doorbell_eqid, &doorbell, q->queue_id);
writel(doorbell.word0, q->phba->sli4_hba.EQDBregaddr);
/* PCI read to flush PCI pipeline on re-arming for INTx mode */
if ((q->phba->intr_type == INTx) && (arm == LPFC_QUEUE_REARM))
readl(q->phba->sli4_hba.EQDBregaddr);
return released;
}
/**
* lpfc_sli4_cq_get - Gets the next valid CQE from a CQ
* @q: The Completion Queue to get the first valid CQE from
*
* This routine will get the first valid Completion Queue Entry from @q, update
* the queue's internal hba index, and return the CQE. If no valid CQEs are in
* the Queue (no more work to do), or the Queue is full of CQEs that have been
* processed, but not popped back to the HBA then this routine will return NULL.
**/
static struct lpfc_cqe *
lpfc_sli4_cq_get(struct lpfc_queue *q)
{
struct lpfc_hba *phba;
struct lpfc_cqe *cqe;
uint32_t idx;
/* sanity check on queue memory */
if (unlikely(!q))
return NULL;
phba = q->phba;
cqe = q->qe[q->hba_index].cqe;
/* If the next CQE is not valid then we are done */
if (bf_get_le32(lpfc_cqe_valid, cqe) != q->qe_valid)
return NULL;
/* If the host has not yet processed the next entry then we are done */
idx = ((q->hba_index + 1) % q->entry_count);
if (idx == q->host_index)
return NULL;
q->hba_index = idx;
/* if the index wrapped around, toggle the valid bit */
if (phba->sli4_hba.pc_sli4_params.cqav && !q->hba_index)
q->qe_valid = (q->qe_valid) ? 0 : 1;
/*
* insert barrier for instruction interlock : data from the hardware
* must have the valid bit checked before it can be copied and acted
* upon. Given what was seen in lpfc_sli4_cq_get() of speculative
* instructions allowing action on content before valid bit checked,
* add barrier here as well. May not be needed as "content" is a
* single 32-bit entity here (vs multi word structure for cq's).
*/
mb();
return cqe;
}
/**
* lpfc_sli4_cq_release - Indicates the host has finished processing a CQ
* @q: The Completion Queue that the host has completed processing for.
* @arm: Indicates whether the host wants to arms this CQ.
*
* This routine will mark all Completion queue entries on @q, from the last
* known completed entry to the last entry that was processed, as completed
* by clearing the valid bit for each completion queue entry. Then it will
* notify the HBA, by ringing the doorbell, that the CQEs have been processed.
* The internal host index in the @q will be updated by this routine to indicate
* that the host has finished processing the entries. The @arm parameter
* indicates that the queue should be rearmed when ringing the doorbell.
*
* This function will return the number of CQEs that were released.
**/
uint32_t
lpfc_sli4_cq_release(struct lpfc_queue *q, bool arm)
{
uint32_t released = 0;
struct lpfc_hba *phba;
struct lpfc_cqe *temp_qe;
struct lpfc_register doorbell;
/* sanity check on queue memory */
if (unlikely(!q))
return 0;
phba = q->phba;
/* while there are valid entries */
while (q->hba_index != q->host_index) {
if (!phba->sli4_hba.pc_sli4_params.cqav) {
temp_qe = q->qe[q->host_index].cqe;
bf_set_le32(lpfc_cqe_valid, temp_qe, 0);
}
released++;
q->host_index = ((q->host_index + 1) % q->entry_count);
}
if (unlikely(released == 0 && !arm))
return 0;
/* ring doorbell for number popped */
doorbell.word0 = 0;
if (arm)
bf_set(lpfc_eqcq_doorbell_arm, &doorbell, 1);
bf_set(lpfc_eqcq_doorbell_num_released, &doorbell, released);
bf_set(lpfc_eqcq_doorbell_qt, &doorbell, LPFC_QUEUE_TYPE_COMPLETION);
bf_set(lpfc_eqcq_doorbell_cqid_hi, &doorbell,
(q->queue_id >> LPFC_CQID_HI_FIELD_SHIFT));
bf_set(lpfc_eqcq_doorbell_cqid_lo, &doorbell, q->queue_id);
writel(doorbell.word0, q->phba->sli4_hba.CQDBregaddr);
return released;
}
/**
* lpfc_sli4_if6_cq_release - Indicates the host has finished processing a CQ
* @q: The Completion Queue that the host has completed processing for.
* @arm: Indicates whether the host wants to arms this CQ.
*
* This routine will mark all Completion queue entries on @q, from the last
* known completed entry to the last entry that was processed, as completed
* by clearing the valid bit for each completion queue entry. Then it will
* notify the HBA, by ringing the doorbell, that the CQEs have been processed.
* The internal host index in the @q will be updated by this routine to indicate
* that the host has finished processing the entries. The @arm parameter
* indicates that the queue should be rearmed when ringing the doorbell.
*
* This function will return the number of CQEs that were released.
**/
uint32_t
lpfc_sli4_if6_cq_release(struct lpfc_queue *q, bool arm)
{
uint32_t released = 0;
struct lpfc_hba *phba;
struct lpfc_cqe *temp_qe;
struct lpfc_register doorbell;
/* sanity check on queue memory */
if (unlikely(!q))
return 0;
phba = q->phba;
/* while there are valid entries */
while (q->hba_index != q->host_index) {
if (!phba->sli4_hba.pc_sli4_params.cqav) {
temp_qe = q->qe[q->host_index].cqe;
bf_set_le32(lpfc_cqe_valid, temp_qe, 0);
}
released++;
q->host_index = ((q->host_index + 1) % q->entry_count);
}
if (unlikely(released == 0 && !arm))
return 0;
/* ring doorbell for number popped */
doorbell.word0 = 0;
if (arm)
bf_set(lpfc_if6_cq_doorbell_arm, &doorbell, 1);
bf_set(lpfc_if6_cq_doorbell_num_released, &doorbell, released);
bf_set(lpfc_if6_cq_doorbell_cqid, &doorbell, q->queue_id);
writel(doorbell.word0, q->phba->sli4_hba.CQDBregaddr);
return released;
}
/**
* lpfc_sli4_rq_put - Put a Receive Buffer Queue Entry on a Receive Queue
* @q: The Header Receive Queue to operate on.
* @wqe: The Receive Queue Entry to put on the Receive queue.
*
* This routine will copy the contents of @wqe to the next available entry on
* the @q. This function will then ring the Receive Queue Doorbell to signal the
* HBA to start processing the Receive Queue Entry. This function returns the
* index that the rqe was copied to if successful. If no entries are available
* on @q then this function will return -ENOMEM.
* The caller is expected to hold the hbalock when calling this routine.
**/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int
lpfc_sli4_rq_put(struct lpfc_queue *hq, struct lpfc_queue *dq,
struct lpfc_rqe *hrqe, struct lpfc_rqe *drqe)
{
struct lpfc_rqe *temp_hrqe;
struct lpfc_rqe *temp_drqe;
struct lpfc_register doorbell;
int hq_put_index;
int dq_put_index;
/* sanity check on queue memory */
if (unlikely(!hq) || unlikely(!dq))
return -ENOMEM;
hq_put_index = hq->host_index;
dq_put_index = dq->host_index;
temp_hrqe = hq->qe[hq_put_index].rqe;
temp_drqe = dq->qe[dq_put_index].rqe;
if (hq->type != LPFC_HRQ || dq->type != LPFC_DRQ)
return -EINVAL;
if (hq_put_index != dq_put_index)
return -EINVAL;
/* If the host has not yet processed the next entry then we are done */
if (((hq_put_index + 1) % hq->entry_count) == hq->hba_index)
return -EBUSY;
lpfc_sli_pcimem_bcopy(hrqe, temp_hrqe, hq->entry_size);
lpfc_sli_pcimem_bcopy(drqe, temp_drqe, dq->entry_size);
/* Update the host index to point to the next slot */
hq->host_index = ((hq_put_index + 1) % hq->entry_count);
dq->host_index = ((dq_put_index + 1) % dq->entry_count);
hq->RQ_buf_posted++;
/* Ring The Header Receive Queue Doorbell */
2011-10-10 19:32:10 -06:00
if (!(hq->host_index % hq->entry_repost)) {
doorbell.word0 = 0;
if (hq->db_format == LPFC_DB_RING_FORMAT) {
bf_set(lpfc_rq_db_ring_fm_num_posted, &doorbell,
hq->entry_repost);
bf_set(lpfc_rq_db_ring_fm_id, &doorbell, hq->queue_id);
} else if (hq->db_format == LPFC_DB_LIST_FORMAT) {
bf_set(lpfc_rq_db_list_fm_num_posted, &doorbell,
hq->entry_repost);
bf_set(lpfc_rq_db_list_fm_index, &doorbell,
hq->host_index);
bf_set(lpfc_rq_db_list_fm_id, &doorbell, hq->queue_id);
} else {
return -EINVAL;
}
writel(doorbell.word0, hq->db_regaddr);
}
return hq_put_index;
}
/**
* lpfc_sli4_rq_release - Updates internal hba index for RQ
* @q: The Header Receive Queue to operate on.
*
* This routine will update the HBA index of a queue to reflect consumption of
* one Receive Queue Entry by the HBA. When the HBA indicates that it has
* consumed an entry the host calls this function to update the queue's
* internal pointers. This routine returns the number of entries that were
* consumed by the HBA.
**/
static uint32_t
lpfc_sli4_rq_release(struct lpfc_queue *hq, struct lpfc_queue *dq)
{
/* sanity check on queue memory */
if (unlikely(!hq) || unlikely(!dq))
return 0;
if ((hq->type != LPFC_HRQ) || (dq->type != LPFC_DRQ))
return 0;
hq->hba_index = ((hq->hba_index + 1) % hq->entry_count);
dq->hba_index = ((dq->hba_index + 1) % dq->entry_count);
return 1;
}
/**
* lpfc_cmd_iocb - Get next command iocb entry in the ring
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function returns pointer to next command iocb entry
* in the command ring. The caller must hold hbalock to prevent
* other threads consume the next command iocb.
* SLI-2/SLI-3 provide different sized iocbs.
**/
static inline IOCB_t *
lpfc_cmd_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
return (IOCB_t *) (((char *) pring->sli.sli3.cmdringaddr) +
pring->sli.sli3.cmdidx * phba->iocb_cmd_size);
}
/**
* lpfc_resp_iocb - Get next response iocb entry in the ring
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function returns pointer to next response iocb entry
* in the response ring. The caller must hold hbalock to make sure
* that no other thread consume the next response iocb.
* SLI-2/SLI-3 provide different sized iocbs.
**/
static inline IOCB_t *
lpfc_resp_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
return (IOCB_t *) (((char *) pring->sli.sli3.rspringaddr) +
pring->sli.sli3.rspidx * phba->iocb_rsp_size);
}
/**
* __lpfc_sli_get_iocbq - Allocates an iocb object from iocb pool
* @phba: Pointer to HBA context object.
*
* This function is called with hbalock held. This function
* allocates a new driver iocb object from the iocb pool. If the
* allocation is successful, it returns pointer to the newly
* allocated iocb object else it returns NULL.
**/
struct lpfc_iocbq *
__lpfc_sli_get_iocbq(struct lpfc_hba *phba)
{
struct list_head *lpfc_iocb_list = &phba->lpfc_iocb_list;
struct lpfc_iocbq * iocbq = NULL;
lockdep_assert_held(&phba->hbalock);
list_remove_head(lpfc_iocb_list, iocbq, struct lpfc_iocbq, list);
if (iocbq)
phba->iocb_cnt++;
if (phba->iocb_cnt > phba->iocb_max)
phba->iocb_max = phba->iocb_cnt;
return iocbq;
}
/**
* __lpfc_clear_active_sglq - Remove the active sglq for this XRI.
* @phba: Pointer to HBA context object.
* @xritag: XRI value.
*
* This function clears the sglq pointer from the array of acive
* sglq's. The xritag that is passed in is used to index into the
* array. Before the xritag can be used it needs to be adjusted
* by subtracting the xribase.
*
* Returns sglq ponter = success, NULL = Failure.
**/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sglq *
__lpfc_clear_active_sglq(struct lpfc_hba *phba, uint16_t xritag)
{
struct lpfc_sglq *sglq;
sglq = phba->sli4_hba.lpfc_sglq_active_list[xritag];
phba->sli4_hba.lpfc_sglq_active_list[xritag] = NULL;
return sglq;
}
/**
* __lpfc_get_active_sglq - Get the active sglq for this XRI.
* @phba: Pointer to HBA context object.
* @xritag: XRI value.
*
* This function returns the sglq pointer from the array of acive
* sglq's. The xritag that is passed in is used to index into the
* array. Before the xritag can be used it needs to be adjusted
* by subtracting the xribase.
*
* Returns sglq ponter = success, NULL = Failure.
**/
struct lpfc_sglq *
__lpfc_get_active_sglq(struct lpfc_hba *phba, uint16_t xritag)
{
struct lpfc_sglq *sglq;
sglq = phba->sli4_hba.lpfc_sglq_active_list[xritag];
return sglq;
}
/**
* lpfc_clr_rrq_active - Clears RRQ active bit in xri_bitmap.
* @phba: Pointer to HBA context object.
* @xritag: xri used in this exchange.
* @rrq: The RRQ to be cleared.
*
**/
void
lpfc_clr_rrq_active(struct lpfc_hba *phba,
uint16_t xritag,
struct lpfc_node_rrq *rrq)
{
struct lpfc_nodelist *ndlp = NULL;
if ((rrq->vport) && NLP_CHK_NODE_ACT(rrq->ndlp))
ndlp = lpfc_findnode_did(rrq->vport, rrq->nlp_DID);
/* The target DID could have been swapped (cable swap)
* we should use the ndlp from the findnode if it is
* available.
*/
if ((!ndlp) && rrq->ndlp)
ndlp = rrq->ndlp;
if (!ndlp)
goto out;
if (test_and_clear_bit(xritag, ndlp->active_rrqs_xri_bitmap)) {
rrq->send_rrq = 0;
rrq->xritag = 0;
rrq->rrq_stop_time = 0;
}
out:
mempool_free(rrq, phba->rrq_pool);
}
/**
* lpfc_handle_rrq_active - Checks if RRQ has waithed RATOV.
* @phba: Pointer to HBA context object.
*
* This function is called with hbalock held. This function
* Checks if stop_time (ratov from setting rrq active) has
* been reached, if it has and the send_rrq flag is set then
* it will call lpfc_send_rrq. If the send_rrq flag is not set
* then it will just call the routine to clear the rrq and
* free the rrq resource.
* The timer is set to the next rrq that is going to expire before
* leaving the routine.
*
**/
void
lpfc_handle_rrq_active(struct lpfc_hba *phba)
{
struct lpfc_node_rrq *rrq;
struct lpfc_node_rrq *nextrrq;
unsigned long next_time;
unsigned long iflags;
LIST_HEAD(send_rrq);
spin_lock_irqsave(&phba->hbalock, iflags);
phba->hba_flag &= ~HBA_RRQ_ACTIVE;
next_time = jiffies + msecs_to_jiffies(1000 * (phba->fc_ratov + 1));
list_for_each_entry_safe(rrq, nextrrq,
&phba->active_rrq_list, list) {
if (time_after(jiffies, rrq->rrq_stop_time))
list_move(&rrq->list, &send_rrq);
else if (time_before(rrq->rrq_stop_time, next_time))
next_time = rrq->rrq_stop_time;
}
spin_unlock_irqrestore(&phba->hbalock, iflags);
if ((!list_empty(&phba->active_rrq_list)) &&
(!(phba->pport->load_flag & FC_UNLOADING)))
mod_timer(&phba->rrq_tmr, next_time);
list_for_each_entry_safe(rrq, nextrrq, &send_rrq, list) {
list_del(&rrq->list);
if (!rrq->send_rrq)
/* this call will free the rrq */
lpfc_clr_rrq_active(phba, rrq->xritag, rrq);
else if (lpfc_send_rrq(phba, rrq)) {
/* if we send the rrq then the completion handler
* will clear the bit in the xribitmap.
*/
lpfc_clr_rrq_active(phba, rrq->xritag,
rrq);
}
}
}
/**
* lpfc_get_active_rrq - Get the active RRQ for this exchange.
* @vport: Pointer to vport context object.
* @xri: The xri used in the exchange.
* @did: The targets DID for this exchange.
*
* returns NULL = rrq not found in the phba->active_rrq_list.
* rrq = rrq for this xri and target.
**/
struct lpfc_node_rrq *
lpfc_get_active_rrq(struct lpfc_vport *vport, uint16_t xri, uint32_t did)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_node_rrq *rrq;
struct lpfc_node_rrq *nextrrq;
unsigned long iflags;
if (phba->sli_rev != LPFC_SLI_REV4)
return NULL;
spin_lock_irqsave(&phba->hbalock, iflags);
list_for_each_entry_safe(rrq, nextrrq, &phba->active_rrq_list, list) {
if (rrq->vport == vport && rrq->xritag == xri &&
rrq->nlp_DID == did){
list_del(&rrq->list);
spin_unlock_irqrestore(&phba->hbalock, iflags);
return rrq;
}
}
spin_unlock_irqrestore(&phba->hbalock, iflags);
return NULL;
}
/**
* lpfc_cleanup_vports_rrqs - Remove and clear the active RRQ for this vport.
* @vport: Pointer to vport context object.
* @ndlp: Pointer to the lpfc_node_list structure.
* If ndlp is NULL Remove all active RRQs for this vport from the
* phba->active_rrq_list and clear the rrq.
* If ndlp is not NULL then only remove rrqs for this vport & this ndlp.
**/
void
lpfc_cleanup_vports_rrqs(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_node_rrq *rrq;
struct lpfc_node_rrq *nextrrq;
unsigned long iflags;
LIST_HEAD(rrq_list);
if (phba->sli_rev != LPFC_SLI_REV4)
return;
if (!ndlp) {
lpfc_sli4_vport_delete_els_xri_aborted(vport);
lpfc_sli4_vport_delete_fcp_xri_aborted(vport);
}
spin_lock_irqsave(&phba->hbalock, iflags);
list_for_each_entry_safe(rrq, nextrrq, &phba->active_rrq_list, list)
if ((rrq->vport == vport) && (!ndlp || rrq->ndlp == ndlp))
list_move(&rrq->list, &rrq_list);
spin_unlock_irqrestore(&phba->hbalock, iflags);
list_for_each_entry_safe(rrq, nextrrq, &rrq_list, list) {
list_del(&rrq->list);
lpfc_clr_rrq_active(phba, rrq->xritag, rrq);
}
}
/**
* lpfc_test_rrq_active - Test RRQ bit in xri_bitmap.
* @phba: Pointer to HBA context object.
* @ndlp: Targets nodelist pointer for this exchange.
* @xritag the xri in the bitmap to test.
*
* This function is called with hbalock held. This function
* returns 0 = rrq not active for this xri
* 1 = rrq is valid for this xri.
**/
int
lpfc_test_rrq_active(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
uint16_t xritag)
{
lockdep_assert_held(&phba->hbalock);
if (!ndlp)
return 0;
if (!ndlp->active_rrqs_xri_bitmap)
return 0;
if (test_bit(xritag, ndlp->active_rrqs_xri_bitmap))
return 1;
else
return 0;
}
/**
* lpfc_set_rrq_active - set RRQ active bit in xri_bitmap.
* @phba: Pointer to HBA context object.
* @ndlp: nodelist pointer for this target.
* @xritag: xri used in this exchange.
* @rxid: Remote Exchange ID.
* @send_rrq: Flag used to determine if we should send rrq els cmd.
*
* This function takes the hbalock.
* The active bit is always set in the active rrq xri_bitmap even
* if there is no slot avaiable for the other rrq information.
*
* returns 0 rrq actived for this xri
* < 0 No memory or invalid ndlp.
**/
int
lpfc_set_rrq_active(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
uint16_t xritag, uint16_t rxid, uint16_t send_rrq)
{
unsigned long iflags;
struct lpfc_node_rrq *rrq;
int empty;
if (!ndlp)
return -EINVAL;
if (!phba->cfg_enable_rrq)
return -EINVAL;
spin_lock_irqsave(&phba->hbalock, iflags);
if (phba->pport->load_flag & FC_UNLOADING) {
phba->hba_flag &= ~HBA_RRQ_ACTIVE;
goto out;
}
/*
* set the active bit even if there is no mem available.
*/
if (NLP_CHK_FREE_REQ(ndlp))
goto out;
if (ndlp->vport && (ndlp->vport->load_flag & FC_UNLOADING))
goto out;
if (!ndlp->active_rrqs_xri_bitmap)
goto out;
if (test_and_set_bit(xritag, ndlp->active_rrqs_xri_bitmap))
goto out;
spin_unlock_irqrestore(&phba->hbalock, iflags);
rrq = mempool_alloc(phba->rrq_pool, GFP_KERNEL);
if (!rrq) {
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"3155 Unable to allocate RRQ xri:0x%x rxid:0x%x"
" DID:0x%x Send:%d\n",
xritag, rxid, ndlp->nlp_DID, send_rrq);
return -EINVAL;
}
if (phba->cfg_enable_rrq == 1)
rrq->send_rrq = send_rrq;
else
rrq->send_rrq = 0;
rrq->xritag = xritag;
rrq->rrq_stop_time = jiffies +
msecs_to_jiffies(1000 * (phba->fc_ratov + 1));
rrq->ndlp = ndlp;
rrq->nlp_DID = ndlp->nlp_DID;
rrq->vport = ndlp->vport;
rrq->rxid = rxid;
spin_lock_irqsave(&phba->hbalock, iflags);
empty = list_empty(&phba->active_rrq_list);
list_add_tail(&rrq->list, &phba->active_rrq_list);
phba->hba_flag |= HBA_RRQ_ACTIVE;
if (empty)
lpfc_worker_wake_up(phba);
spin_unlock_irqrestore(&phba->hbalock, iflags);
return 0;
out:
spin_unlock_irqrestore(&phba->hbalock, iflags);
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"2921 Can't set rrq active xri:0x%x rxid:0x%x"
" DID:0x%x Send:%d\n",
xritag, rxid, ndlp->nlp_DID, send_rrq);
return -EINVAL;
}
/**
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* __lpfc_sli_get_els_sglq - Allocates an iocb object from sgl pool
* @phba: Pointer to HBA context object.
* @piocb: Pointer to the iocbq.
*
* This function is called with the ring lock held. This function
* gets a new driver sglq object from the sglq list. If the
* list is not empty then it is successful, it returns pointer to the newly
* allocated sglq object else it returns NULL.
**/
static struct lpfc_sglq *
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
__lpfc_sli_get_els_sglq(struct lpfc_hba *phba, struct lpfc_iocbq *piocbq)
{
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct list_head *lpfc_els_sgl_list = &phba->sli4_hba.lpfc_els_sgl_list;
struct lpfc_sglq *sglq = NULL;
struct lpfc_sglq *start_sglq = NULL;
struct lpfc_scsi_buf *lpfc_cmd;
struct lpfc_nodelist *ndlp;
int found = 0;
lockdep_assert_held(&phba->hbalock);
if (piocbq->iocb_flag & LPFC_IO_FCP) {
lpfc_cmd = (struct lpfc_scsi_buf *) piocbq->context1;
ndlp = lpfc_cmd->rdata->pnode;
} else if ((piocbq->iocb.ulpCommand == CMD_GEN_REQUEST64_CR) &&
!(piocbq->iocb_flag & LPFC_IO_LIBDFC)) {
ndlp = piocbq->context_un.ndlp;
} else if (piocbq->iocb_flag & LPFC_IO_LIBDFC) {
if (piocbq->iocb_flag & LPFC_IO_LOOPBACK)
ndlp = NULL;
else
ndlp = piocbq->context_un.ndlp;
} else {
ndlp = piocbq->context1;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_lock(&phba->sli4_hba.sgl_list_lock);
list_remove_head(lpfc_els_sgl_list, sglq, struct lpfc_sglq, list);
start_sglq = sglq;
while (!found) {
if (!sglq)
break;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (ndlp && ndlp->active_rrqs_xri_bitmap &&
test_bit(sglq->sli4_lxritag,
ndlp->active_rrqs_xri_bitmap)) {
/* This xri has an rrq outstanding for this DID.
* put it back in the list and get another xri.
*/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
list_add_tail(&sglq->list, lpfc_els_sgl_list);
sglq = NULL;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
list_remove_head(lpfc_els_sgl_list, sglq,
struct lpfc_sglq, list);
if (sglq == start_sglq) {
list_add_tail(&sglq->list, lpfc_els_sgl_list);
sglq = NULL;
break;
} else
continue;
}
sglq->ndlp = ndlp;
found = 1;
phba->sli4_hba.lpfc_sglq_active_list[sglq->sli4_lxritag] = sglq;
sglq->state = SGL_ALLOCATED;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_unlock(&phba->sli4_hba.sgl_list_lock);
return sglq;
}
/**
* __lpfc_sli_get_nvmet_sglq - Allocates an iocb object from sgl pool
* @phba: Pointer to HBA context object.
* @piocb: Pointer to the iocbq.
*
* This function is called with the sgl_list lock held. This function
* gets a new driver sglq object from the sglq list. If the
* list is not empty then it is successful, it returns pointer to the newly
* allocated sglq object else it returns NULL.
**/
struct lpfc_sglq *
__lpfc_sli_get_nvmet_sglq(struct lpfc_hba *phba, struct lpfc_iocbq *piocbq)
{
struct list_head *lpfc_nvmet_sgl_list;
struct lpfc_sglq *sglq = NULL;
lpfc_nvmet_sgl_list = &phba->sli4_hba.lpfc_nvmet_sgl_list;
lockdep_assert_held(&phba->sli4_hba.sgl_list_lock);
list_remove_head(lpfc_nvmet_sgl_list, sglq, struct lpfc_sglq, list);
if (!sglq)
return NULL;
phba->sli4_hba.lpfc_sglq_active_list[sglq->sli4_lxritag] = sglq;
sglq->state = SGL_ALLOCATED;
return sglq;
}
/**
* lpfc_sli_get_iocbq - Allocates an iocb object from iocb pool
* @phba: Pointer to HBA context object.
*
* This function is called with no lock held. This function
* allocates a new driver iocb object from the iocb pool. If the
* allocation is successful, it returns pointer to the newly
* allocated iocb object else it returns NULL.
**/
struct lpfc_iocbq *
lpfc_sli_get_iocbq(struct lpfc_hba *phba)
{
struct lpfc_iocbq * iocbq = NULL;
unsigned long iflags;
spin_lock_irqsave(&phba->hbalock, iflags);
iocbq = __lpfc_sli_get_iocbq(phba);
spin_unlock_irqrestore(&phba->hbalock, iflags);
return iocbq;
}
/**
* __lpfc_sli_release_iocbq_s4 - Release iocb to the iocb pool
* @phba: Pointer to HBA context object.
* @iocbq: Pointer to driver iocb object.
*
* This function is called with hbalock held to release driver
* iocb object to the iocb pool. The iotag in the iocb object
* does not change for each use of the iocb object. This function
* clears all other fields of the iocb object when it is freed.
* The sqlq structure that holds the xritag and phys and virtual
* mappings for the scatter gather list is retrieved from the
* active array of sglq. The get of the sglq pointer also clears
* the entry in the array. If the status of the IO indiactes that
* this IO was aborted then the sglq entry it put on the
* lpfc_abts_els_sgl_list until the CQ_ABORTED_XRI is received. If the
* IO has good status or fails for any other reason then the sglq
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* entry is added to the free list (lpfc_els_sgl_list).
**/
static void
__lpfc_sli_release_iocbq_s4(struct lpfc_hba *phba, struct lpfc_iocbq *iocbq)
{
struct lpfc_sglq *sglq;
size_t start_clean = offsetof(struct lpfc_iocbq, iocb);
unsigned long iflag = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *pring;
lockdep_assert_held(&phba->hbalock);
if (iocbq->sli4_xritag == NO_XRI)
sglq = NULL;
else
sglq = __lpfc_clear_active_sglq(phba, iocbq->sli4_lxritag);
if (sglq) {
if (iocbq->iocb_flag & LPFC_IO_NVMET) {
spin_lock_irqsave(&phba->sli4_hba.sgl_list_lock,
iflag);
sglq->state = SGL_FREED;
sglq->ndlp = NULL;
list_add_tail(&sglq->list,
&phba->sli4_hba.lpfc_nvmet_sgl_list);
spin_unlock_irqrestore(
&phba->sli4_hba.sgl_list_lock, iflag);
goto out;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = phba->sli4_hba.els_wq->pring;
if ((iocbq->iocb_flag & LPFC_EXCHANGE_BUSY) &&
(sglq->state != SGL_XRI_ABORTED)) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_lock_irqsave(&phba->sli4_hba.sgl_list_lock,
iflag);
list_add(&sglq->list,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
&phba->sli4_hba.lpfc_abts_els_sgl_list);
spin_unlock_irqrestore(
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
&phba->sli4_hba.sgl_list_lock, iflag);
} else {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_lock_irqsave(&phba->sli4_hba.sgl_list_lock,
iflag);
sglq->state = SGL_FREED;
sglq->ndlp = NULL;
list_add_tail(&sglq->list,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
&phba->sli4_hba.lpfc_els_sgl_list);
spin_unlock_irqrestore(
&phba->sli4_hba.sgl_list_lock, iflag);
/* Check if TXQ queue needs to be serviced */
if (!list_empty(&pring->txq))
lpfc_worker_wake_up(phba);
}
}
out:
/*
* Clean all volatile data fields, preserve iotag and node struct.
*/
memset((char *)iocbq + start_clean, 0, sizeof(*iocbq) - start_clean);
iocbq->sli4_lxritag = NO_XRI;
iocbq->sli4_xritag = NO_XRI;
iocbq->iocb_flag &= ~(LPFC_IO_NVME | LPFC_IO_NVMET |
LPFC_IO_NVME_LS);
list_add_tail(&iocbq->list, &phba->lpfc_iocb_list);
}
/**
* __lpfc_sli_release_iocbq_s3 - Release iocb to the iocb pool
* @phba: Pointer to HBA context object.
* @iocbq: Pointer to driver iocb object.
*
* This function is called with hbalock held to release driver
* iocb object to the iocb pool. The iotag in the iocb object
* does not change for each use of the iocb object. This function
* clears all other fields of the iocb object when it is freed.
**/
static void
__lpfc_sli_release_iocbq_s3(struct lpfc_hba *phba, struct lpfc_iocbq *iocbq)
{
size_t start_clean = offsetof(struct lpfc_iocbq, iocb);
lockdep_assert_held(&phba->hbalock);
/*
* Clean all volatile data fields, preserve iotag and node struct.
*/
memset((char*)iocbq + start_clean, 0, sizeof(*iocbq) - start_clean);
iocbq->sli4_xritag = NO_XRI;
list_add_tail(&iocbq->list, &phba->lpfc_iocb_list);
}
/**
* __lpfc_sli_release_iocbq - Release iocb to the iocb pool
* @phba: Pointer to HBA context object.
* @iocbq: Pointer to driver iocb object.
*
* This function is called with hbalock held to release driver
* iocb object to the iocb pool. The iotag in the iocb object
* does not change for each use of the iocb object. This function
* clears all other fields of the iocb object when it is freed.
**/
static void
__lpfc_sli_release_iocbq(struct lpfc_hba *phba, struct lpfc_iocbq *iocbq)
{
lockdep_assert_held(&phba->hbalock);
phba->__lpfc_sli_release_iocbq(phba, iocbq);
phba->iocb_cnt--;
}
/**
* lpfc_sli_release_iocbq - Release iocb to the iocb pool
* @phba: Pointer to HBA context object.
* @iocbq: Pointer to driver iocb object.
*
* This function is called with no lock held to release the iocb to
* iocb pool.
**/
void
lpfc_sli_release_iocbq(struct lpfc_hba *phba, struct lpfc_iocbq *iocbq)
{
unsigned long iflags;
/*
* Clean all volatile data fields, preserve iotag and node struct.
*/
spin_lock_irqsave(&phba->hbalock, iflags);
__lpfc_sli_release_iocbq(phba, iocbq);
spin_unlock_irqrestore(&phba->hbalock, iflags);
}
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/**
* lpfc_sli_cancel_iocbs - Cancel all iocbs from a list.
* @phba: Pointer to HBA context object.
* @iocblist: List of IOCBs.
* @ulpstatus: ULP status in IOCB command field.
* @ulpWord4: ULP word-4 in IOCB command field.
*
* This function is called with a list of IOCBs to cancel. It cancels the IOCB
* on the list by invoking the complete callback function associated with the
* IOCB with the provided @ulpstatus and @ulpword4 set to the IOCB commond
* fields.
**/
void
lpfc_sli_cancel_iocbs(struct lpfc_hba *phba, struct list_head *iocblist,
uint32_t ulpstatus, uint32_t ulpWord4)
{
struct lpfc_iocbq *piocb;
while (!list_empty(iocblist)) {
list_remove_head(iocblist, piocb, struct lpfc_iocbq, list);
if (!piocb->iocb_cmpl)
lpfc_sli_release_iocbq(phba, piocb);
else {
piocb->iocb.ulpStatus = ulpstatus;
piocb->iocb.un.ulpWord[4] = ulpWord4;
(piocb->iocb_cmpl) (phba, piocb, piocb);
}
}
return;
}
/**
* lpfc_sli_iocb_cmd_type - Get the iocb type
* @iocb_cmnd: iocb command code.
*
* This function is called by ring event handler function to get the iocb type.
* This function translates the iocb command to an iocb command type used to
* decide the final disposition of each completed IOCB.
* The function returns
* LPFC_UNKNOWN_IOCB if it is an unsupported iocb
* LPFC_SOL_IOCB if it is a solicited iocb completion
* LPFC_ABORT_IOCB if it is an abort iocb
* LPFC_UNSOL_IOCB if it is an unsolicited iocb
*
* The caller is not required to hold any lock.
**/
static lpfc_iocb_type
lpfc_sli_iocb_cmd_type(uint8_t iocb_cmnd)
{
lpfc_iocb_type type = LPFC_UNKNOWN_IOCB;
if (iocb_cmnd > CMD_MAX_IOCB_CMD)
return 0;
switch (iocb_cmnd) {
case CMD_XMIT_SEQUENCE_CR:
case CMD_XMIT_SEQUENCE_CX:
case CMD_XMIT_BCAST_CN:
case CMD_XMIT_BCAST_CX:
case CMD_ELS_REQUEST_CR:
case CMD_ELS_REQUEST_CX:
case CMD_CREATE_XRI_CR:
case CMD_CREATE_XRI_CX:
case CMD_GET_RPI_CN:
case CMD_XMIT_ELS_RSP_CX:
case CMD_GET_RPI_CR:
case CMD_FCP_IWRITE_CR:
case CMD_FCP_IWRITE_CX:
case CMD_FCP_IREAD_CR:
case CMD_FCP_IREAD_CX:
case CMD_FCP_ICMND_CR:
case CMD_FCP_ICMND_CX:
case CMD_FCP_TSEND_CX:
case CMD_FCP_TRSP_CX:
case CMD_FCP_TRECEIVE_CX:
case CMD_FCP_AUTO_TRSP_CX:
case CMD_ADAPTER_MSG:
case CMD_ADAPTER_DUMP:
case CMD_XMIT_SEQUENCE64_CR:
case CMD_XMIT_SEQUENCE64_CX:
case CMD_XMIT_BCAST64_CN:
case CMD_XMIT_BCAST64_CX:
case CMD_ELS_REQUEST64_CR:
case CMD_ELS_REQUEST64_CX:
case CMD_FCP_IWRITE64_CR:
case CMD_FCP_IWRITE64_CX:
case CMD_FCP_IREAD64_CR:
case CMD_FCP_IREAD64_CX:
case CMD_FCP_ICMND64_CR:
case CMD_FCP_ICMND64_CX:
case CMD_FCP_TSEND64_CX:
case CMD_FCP_TRSP64_CX:
case CMD_FCP_TRECEIVE64_CX:
case CMD_GEN_REQUEST64_CR:
case CMD_GEN_REQUEST64_CX:
case CMD_XMIT_ELS_RSP64_CX:
case DSSCMD_IWRITE64_CR:
case DSSCMD_IWRITE64_CX:
case DSSCMD_IREAD64_CR:
case DSSCMD_IREAD64_CX:
type = LPFC_SOL_IOCB;
break;
case CMD_ABORT_XRI_CN:
case CMD_ABORT_XRI_CX:
case CMD_CLOSE_XRI_CN:
case CMD_CLOSE_XRI_CX:
case CMD_XRI_ABORTED_CX:
case CMD_ABORT_MXRI64_CN:
case CMD_XMIT_BLS_RSP64_CX:
type = LPFC_ABORT_IOCB;
break;
case CMD_RCV_SEQUENCE_CX:
case CMD_RCV_ELS_REQ_CX:
case CMD_RCV_SEQUENCE64_CX:
case CMD_RCV_ELS_REQ64_CX:
case CMD_ASYNC_STATUS:
case CMD_IOCB_RCV_SEQ64_CX:
case CMD_IOCB_RCV_ELS64_CX:
case CMD_IOCB_RCV_CONT64_CX:
case CMD_IOCB_RET_XRI64_CX:
type = LPFC_UNSOL_IOCB;
break;
case CMD_IOCB_XMIT_MSEQ64_CR:
case CMD_IOCB_XMIT_MSEQ64_CX:
case CMD_IOCB_RCV_SEQ_LIST64_CX:
case CMD_IOCB_RCV_ELS_LIST64_CX:
case CMD_IOCB_CLOSE_EXTENDED_CN:
case CMD_IOCB_ABORT_EXTENDED_CN:
case CMD_IOCB_RET_HBQE64_CN:
case CMD_IOCB_FCP_IBIDIR64_CR:
case CMD_IOCB_FCP_IBIDIR64_CX:
case CMD_IOCB_FCP_ITASKMGT64_CX:
case CMD_IOCB_LOGENTRY_CN:
case CMD_IOCB_LOGENTRY_ASYNC_CN:
printk("%s - Unhandled SLI-3 Command x%x\n",
__func__, iocb_cmnd);
type = LPFC_UNKNOWN_IOCB;
break;
default:
type = LPFC_UNKNOWN_IOCB;
break;
}
return type;
}
/**
* lpfc_sli_ring_map - Issue config_ring mbox for all rings
* @phba: Pointer to HBA context object.
*
* This function is called from SLI initialization code
* to configure every ring of the HBA's SLI interface. The
* caller is not required to hold any lock. This function issues
* a config_ring mailbox command for each ring.
* This function returns zero if successful else returns a negative
* error code.
**/
static int
lpfc_sli_ring_map(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
LPFC_MBOXQ_t *pmb;
MAILBOX_t *pmbox;
int i, rc, ret = 0;
pmb = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!pmb)
return -ENOMEM;
pmbox = &pmb->u.mb;
phba->link_state = LPFC_INIT_MBX_CMDS;
for (i = 0; i < psli->num_rings; i++) {
lpfc_config_ring(phba, i, pmb);
rc = lpfc_sli_issue_mbox(phba, pmb, MBX_POLL);
if (rc != MBX_SUCCESS) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0446 Adapter failed to init (%d), "
"mbxCmd x%x CFG_RING, mbxStatus x%x, "
"ring %d\n",
rc, pmbox->mbxCommand,
pmbox->mbxStatus, i);
phba->link_state = LPFC_HBA_ERROR;
ret = -ENXIO;
break;
}
}
mempool_free(pmb, phba->mbox_mem_pool);
return ret;
}
/**
* lpfc_sli_ringtxcmpl_put - Adds new iocb to the txcmplq
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @piocb: Pointer to the driver iocb object.
*
* This function is called with hbalock held. The function adds the
* new iocb to txcmplq of the given ring. This function always returns
* 0. If this function is called for ELS ring, this function checks if
* there is a vport associated with the ELS command. This function also
* starts els_tmofunc timer if this is an ELS command.
**/
static int
lpfc_sli_ringtxcmpl_put(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *piocb)
{
lockdep_assert_held(&phba->hbalock);
scsi: lpfc: fix oops/BUG in lpfc_sli_ringtxcmpl_put() The BUG_ON() recently introduced in lpfc_sli_ringtxcmpl_put() is hit in the lpfc_els_abort() > lpfc_sli_issue_abort_iotag() > lpfc_sli_abort_iotag_issue() function path [similar names], due to 'piocb->vport == NULL': BUG_ON(!piocb || !piocb->vport); This happens because lpfc_sli_abort_iotag_issue() doesn't set the 'abtsiocbp->vport' pointer -- but this is not the problem. Previously, lpfc_sli_ringtxcmpl_put() accessed 'piocb->vport' only if 'piocb->iocb.ulpCommand' is neither CMD_ABORT_XRI_CN nor CMD_CLOSE_XRI_CN, which are the only possible values for lpfc_sli_abort_iotag_issue(): lpfc_sli_ringtxcmpl_put(): if ((unlikely(pring->ringno == LPFC_ELS_RING)) && (piocb->iocb.ulpCommand != CMD_ABORT_XRI_CN) && (piocb->iocb.ulpCommand != CMD_CLOSE_XRI_CN) && (!(piocb->vport->load_flag & FC_UNLOADING))) lpfc_sli_abort_iotag_issue(): if (phba->link_state >= LPFC_LINK_UP) iabt->ulpCommand = CMD_ABORT_XRI_CN; else iabt->ulpCommand = CMD_CLOSE_XRI_CN; So, this function path would not have hit this possible NULL pointer dereference before. In order to fix this regression, move the second part of the BUG_ON() check prior to the pointer dereference that it does check for. For reference, this is the stack trace observed. The problem happened because an unsolicited event was received - a PLOGI was received after our PLOGI was issued but not yet complete, so the discovery state machine goes on to sw-abort our PLOGI. kernel BUG at drivers/scsi/lpfc/lpfc_sli.c:1326! Oops: Exception in kernel mode, sig: 5 [#1] <...> NIP [...] lpfc_sli_ringtxcmpl_put+0x1c/0xf0 [lpfc] LR [...] __lpfc_sli_issue_iocb_s4+0x188/0x200 [lpfc] Call Trace: [...] [...] __lpfc_sli_issue_iocb_s4+0xb0/0x200 [lpfc] (unreliable) [...] [...] lpfc_sli_issue_abort_iotag+0x2b4/0x350 [lpfc] [...] [...] lpfc_els_abort+0x1a8/0x4a0 [lpfc] [...] [...] lpfc_rcv_plogi+0x6d4/0x700 [lpfc] [...] [...] lpfc_rcv_plogi_plogi_issue+0xd8/0x1d0 [lpfc] [...] [...] lpfc_disc_state_machine+0xc0/0x2b0 [lpfc] [...] [...] lpfc_els_unsol_buffer+0xcc0/0x26c0 [lpfc] [...] [...] lpfc_els_unsol_event+0xa8/0x220 [lpfc] [...] [...] lpfc_complete_unsol_iocb+0xb8/0x138 [lpfc] [...] [...] lpfc_sli4_handle_received_buffer+0x6a0/0xec0 [lpfc] [...] [...] lpfc_sli_handle_slow_ring_event_s4+0x1c4/0x240 [lpfc] [...] [...] lpfc_sli_handle_slow_ring_event+0x24/0x40 [lpfc] [...] [...] lpfc_do_work+0xd88/0x1970 [lpfc] [...] [...] kthread+0x108/0x130 [...] [...] ret_from_kernel_thread+0x5c/0xbc <...> Cc: stable@vger.kernel.org # v4.8 Fixes: 22466da5b4b7 ("lpfc: Fix possible NULL pointer dereference") Reported-by: Harsha Thyagaraja <hathyaga@in.ibm.com> Signed-off-by: Mauricio Faria de Oliveira <mauricfo@linux.vnet.ibm.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-11-23 05:33:19 -07:00
BUG_ON(!piocb);
list_add_tail(&piocb->list, &pring->txcmplq);
piocb->iocb_flag |= LPFC_IO_ON_TXCMPLQ;
if ((unlikely(pring->ringno == LPFC_ELS_RING)) &&
(piocb->iocb.ulpCommand != CMD_ABORT_XRI_CN) &&
scsi: lpfc: fix oops/BUG in lpfc_sli_ringtxcmpl_put() The BUG_ON() recently introduced in lpfc_sli_ringtxcmpl_put() is hit in the lpfc_els_abort() > lpfc_sli_issue_abort_iotag() > lpfc_sli_abort_iotag_issue() function path [similar names], due to 'piocb->vport == NULL': BUG_ON(!piocb || !piocb->vport); This happens because lpfc_sli_abort_iotag_issue() doesn't set the 'abtsiocbp->vport' pointer -- but this is not the problem. Previously, lpfc_sli_ringtxcmpl_put() accessed 'piocb->vport' only if 'piocb->iocb.ulpCommand' is neither CMD_ABORT_XRI_CN nor CMD_CLOSE_XRI_CN, which are the only possible values for lpfc_sli_abort_iotag_issue(): lpfc_sli_ringtxcmpl_put(): if ((unlikely(pring->ringno == LPFC_ELS_RING)) && (piocb->iocb.ulpCommand != CMD_ABORT_XRI_CN) && (piocb->iocb.ulpCommand != CMD_CLOSE_XRI_CN) && (!(piocb->vport->load_flag & FC_UNLOADING))) lpfc_sli_abort_iotag_issue(): if (phba->link_state >= LPFC_LINK_UP) iabt->ulpCommand = CMD_ABORT_XRI_CN; else iabt->ulpCommand = CMD_CLOSE_XRI_CN; So, this function path would not have hit this possible NULL pointer dereference before. In order to fix this regression, move the second part of the BUG_ON() check prior to the pointer dereference that it does check for. For reference, this is the stack trace observed. The problem happened because an unsolicited event was received - a PLOGI was received after our PLOGI was issued but not yet complete, so the discovery state machine goes on to sw-abort our PLOGI. kernel BUG at drivers/scsi/lpfc/lpfc_sli.c:1326! Oops: Exception in kernel mode, sig: 5 [#1] <...> NIP [...] lpfc_sli_ringtxcmpl_put+0x1c/0xf0 [lpfc] LR [...] __lpfc_sli_issue_iocb_s4+0x188/0x200 [lpfc] Call Trace: [...] [...] __lpfc_sli_issue_iocb_s4+0xb0/0x200 [lpfc] (unreliable) [...] [...] lpfc_sli_issue_abort_iotag+0x2b4/0x350 [lpfc] [...] [...] lpfc_els_abort+0x1a8/0x4a0 [lpfc] [...] [...] lpfc_rcv_plogi+0x6d4/0x700 [lpfc] [...] [...] lpfc_rcv_plogi_plogi_issue+0xd8/0x1d0 [lpfc] [...] [...] lpfc_disc_state_machine+0xc0/0x2b0 [lpfc] [...] [...] lpfc_els_unsol_buffer+0xcc0/0x26c0 [lpfc] [...] [...] lpfc_els_unsol_event+0xa8/0x220 [lpfc] [...] [...] lpfc_complete_unsol_iocb+0xb8/0x138 [lpfc] [...] [...] lpfc_sli4_handle_received_buffer+0x6a0/0xec0 [lpfc] [...] [...] lpfc_sli_handle_slow_ring_event_s4+0x1c4/0x240 [lpfc] [...] [...] lpfc_sli_handle_slow_ring_event+0x24/0x40 [lpfc] [...] [...] lpfc_do_work+0xd88/0x1970 [lpfc] [...] [...] kthread+0x108/0x130 [...] [...] ret_from_kernel_thread+0x5c/0xbc <...> Cc: stable@vger.kernel.org # v4.8 Fixes: 22466da5b4b7 ("lpfc: Fix possible NULL pointer dereference") Reported-by: Harsha Thyagaraja <hathyaga@in.ibm.com> Signed-off-by: Mauricio Faria de Oliveira <mauricfo@linux.vnet.ibm.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-11-23 05:33:19 -07:00
(piocb->iocb.ulpCommand != CMD_CLOSE_XRI_CN)) {
BUG_ON(!piocb->vport);
if (!(piocb->vport->load_flag & FC_UNLOADING))
mod_timer(&piocb->vport->els_tmofunc,
jiffies +
msecs_to_jiffies(1000 * (phba->fc_ratov << 1)));
}
return 0;
}
/**
* lpfc_sli_ringtx_get - Get first element of the txq
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function is called with hbalock held to get next
* iocb in txq of the given ring. If there is any iocb in
* the txq, the function returns first iocb in the list after
* removing the iocb from the list, else it returns NULL.
**/
struct lpfc_iocbq *
lpfc_sli_ringtx_get(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
struct lpfc_iocbq *cmd_iocb;
lockdep_assert_held(&phba->hbalock);
list_remove_head((&pring->txq), cmd_iocb, struct lpfc_iocbq, list);
return cmd_iocb;
}
/**
* lpfc_sli_next_iocb_slot - Get next iocb slot in the ring
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function is called with hbalock held and the caller must post the
* iocb without releasing the lock. If the caller releases the lock,
* iocb slot returned by the function is not guaranteed to be available.
* The function returns pointer to the next available iocb slot if there
* is available slot in the ring, else it returns NULL.
* If the get index of the ring is ahead of the put index, the function
* will post an error attention event to the worker thread to take the
* HBA to offline state.
**/
static IOCB_t *
lpfc_sli_next_iocb_slot (struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
struct lpfc_pgp *pgp = &phba->port_gp[pring->ringno];
uint32_t max_cmd_idx = pring->sli.sli3.numCiocb;
lockdep_assert_held(&phba->hbalock);
if ((pring->sli.sli3.next_cmdidx == pring->sli.sli3.cmdidx) &&
(++pring->sli.sli3.next_cmdidx >= max_cmd_idx))
pring->sli.sli3.next_cmdidx = 0;
if (unlikely(pring->sli.sli3.local_getidx ==
pring->sli.sli3.next_cmdidx)) {
pring->sli.sli3.local_getidx = le32_to_cpu(pgp->cmdGetInx);
if (unlikely(pring->sli.sli3.local_getidx >= max_cmd_idx)) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0315 Ring %d issue: portCmdGet %d "
"is bigger than cmd ring %d\n",
pring->ringno,
pring->sli.sli3.local_getidx,
max_cmd_idx);
phba->link_state = LPFC_HBA_ERROR;
/*
* All error attention handlers are posted to
* worker thread
*/
phba->work_ha |= HA_ERATT;
phba->work_hs = HS_FFER3;
lpfc_worker_wake_up(phba);
return NULL;
}
if (pring->sli.sli3.local_getidx == pring->sli.sli3.next_cmdidx)
return NULL;
}
return lpfc_cmd_iocb(phba, pring);
}
/**
* lpfc_sli_next_iotag - Get an iotag for the iocb
* @phba: Pointer to HBA context object.
* @iocbq: Pointer to driver iocb object.
*
* This function gets an iotag for the iocb. If there is no unused iotag and
* the iocbq_lookup_len < 0xffff, this function allocates a bigger iotag_lookup
* array and assigns a new iotag.
* The function returns the allocated iotag if successful, else returns zero.
* Zero is not a valid iotag.
* The caller is not required to hold any lock.
**/
uint16_t
lpfc_sli_next_iotag(struct lpfc_hba *phba, struct lpfc_iocbq *iocbq)
{
struct lpfc_iocbq **new_arr;
struct lpfc_iocbq **old_arr;
size_t new_len;
struct lpfc_sli *psli = &phba->sli;
uint16_t iotag;
spin_lock_irq(&phba->hbalock);
iotag = psli->last_iotag;
if(++iotag < psli->iocbq_lookup_len) {
psli->last_iotag = iotag;
psli->iocbq_lookup[iotag] = iocbq;
spin_unlock_irq(&phba->hbalock);
iocbq->iotag = iotag;
return iotag;
} else if (psli->iocbq_lookup_len < (0xffff
- LPFC_IOCBQ_LOOKUP_INCREMENT)) {
new_len = psli->iocbq_lookup_len + LPFC_IOCBQ_LOOKUP_INCREMENT;
spin_unlock_irq(&phba->hbalock);
new_arr = kzalloc(new_len * sizeof (struct lpfc_iocbq *),
GFP_KERNEL);
if (new_arr) {
spin_lock_irq(&phba->hbalock);
old_arr = psli->iocbq_lookup;
if (new_len <= psli->iocbq_lookup_len) {
/* highly unprobable case */
kfree(new_arr);
iotag = psli->last_iotag;
if(++iotag < psli->iocbq_lookup_len) {
psli->last_iotag = iotag;
psli->iocbq_lookup[iotag] = iocbq;
spin_unlock_irq(&phba->hbalock);
iocbq->iotag = iotag;
return iotag;
}
spin_unlock_irq(&phba->hbalock);
return 0;
}
if (psli->iocbq_lookup)
memcpy(new_arr, old_arr,
((psli->last_iotag + 1) *
sizeof (struct lpfc_iocbq *)));
psli->iocbq_lookup = new_arr;
psli->iocbq_lookup_len = new_len;
psli->last_iotag = iotag;
psli->iocbq_lookup[iotag] = iocbq;
spin_unlock_irq(&phba->hbalock);
iocbq->iotag = iotag;
kfree(old_arr);
return iotag;
}
} else
spin_unlock_irq(&phba->hbalock);
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0318 Failed to allocate IOTAG.last IOTAG is %d\n",
psli->last_iotag);
return 0;
}
/**
* lpfc_sli_submit_iocb - Submit an iocb to the firmware
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @iocb: Pointer to iocb slot in the ring.
* @nextiocb: Pointer to driver iocb object which need to be
* posted to firmware.
*
* This function is called with hbalock held to post a new iocb to
* the firmware. This function copies the new iocb to ring iocb slot and
* updates the ring pointers. It adds the new iocb to txcmplq if there is
* a completion call back for this iocb else the function will free the
* iocb object.
**/
static void
lpfc_sli_submit_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
IOCB_t *iocb, struct lpfc_iocbq *nextiocb)
{
lockdep_assert_held(&phba->hbalock);
/*
* Set up an iotag
*/
nextiocb->iocb.ulpIoTag = (nextiocb->iocb_cmpl) ? nextiocb->iotag : 0;
if (pring->ringno == LPFC_ELS_RING) {
lpfc_debugfs_slow_ring_trc(phba,
"IOCB cmd ring: wd4:x%08x wd6:x%08x wd7:x%08x",
*(((uint32_t *) &nextiocb->iocb) + 4),
*(((uint32_t *) &nextiocb->iocb) + 6),
*(((uint32_t *) &nextiocb->iocb) + 7));
}
/*
* Issue iocb command to adapter
*/
lpfc_sli_pcimem_bcopy(&nextiocb->iocb, iocb, phba->iocb_cmd_size);
wmb();
pring->stats.iocb_cmd++;
/*
* If there is no completion routine to call, we can release the
* IOCB buffer back right now. For IOCBs, like QUE_RING_BUF,
* that have no rsp ring completion, iocb_cmpl MUST be NULL.
*/
if (nextiocb->iocb_cmpl)
lpfc_sli_ringtxcmpl_put(phba, pring, nextiocb);
else
__lpfc_sli_release_iocbq(phba, nextiocb);
/*
* Let the HBA know what IOCB slot will be the next one the
* driver will put a command into.
*/
pring->sli.sli3.cmdidx = pring->sli.sli3.next_cmdidx;
writel(pring->sli.sli3.cmdidx, &phba->host_gp[pring->ringno].cmdPutInx);
}
/**
* lpfc_sli_update_full_ring - Update the chip attention register
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* The caller is not required to hold any lock for calling this function.
* This function updates the chip attention bits for the ring to inform firmware
* that there are pending work to be done for this ring and requests an
* interrupt when there is space available in the ring. This function is
* called when the driver is unable to post more iocbs to the ring due
* to unavailability of space in the ring.
**/
static void
lpfc_sli_update_full_ring(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
int ringno = pring->ringno;
pring->flag |= LPFC_CALL_RING_AVAILABLE;
wmb();
/*
* Set ring 'ringno' to SET R0CE_REQ in Chip Att register.
* The HBA will tell us when an IOCB entry is available.
*/
writel((CA_R0ATT|CA_R0CE_REQ) << (ringno*4), phba->CAregaddr);
readl(phba->CAregaddr); /* flush */
pring->stats.iocb_cmd_full++;
}
/**
* lpfc_sli_update_ring - Update chip attention register
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function updates the chip attention register bit for the
* given ring to inform HBA that there is more work to be done
* in this ring. The caller is not required to hold any lock.
**/
static void
lpfc_sli_update_ring(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
int ringno = pring->ringno;
/*
* Tell the HBA that there is work to do in this ring.
*/
if (!(phba->sli3_options & LPFC_SLI3_CRP_ENABLED)) {
wmb();
writel(CA_R0ATT << (ringno * 4), phba->CAregaddr);
readl(phba->CAregaddr); /* flush */
}
}
/**
* lpfc_sli_resume_iocb - Process iocbs in the txq
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function is called with hbalock held to post pending iocbs
* in the txq to the firmware. This function is called when driver
* detects space available in the ring.
**/
static void
lpfc_sli_resume_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
IOCB_t *iocb;
struct lpfc_iocbq *nextiocb;
lockdep_assert_held(&phba->hbalock);
/*
* Check to see if:
* (a) there is anything on the txq to send
* (b) link is up
* (c) link attention events can be processed (fcp ring only)
* (d) IOCB processing is not blocked by the outstanding mbox command.
*/
if (lpfc_is_link_up(phba) &&
(!list_empty(&pring->txq)) &&
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
(pring->ringno != LPFC_FCP_RING ||
phba->sli.sli_flag & LPFC_PROCESS_LA)) {
while ((iocb = lpfc_sli_next_iocb_slot(phba, pring)) &&
(nextiocb = lpfc_sli_ringtx_get(phba, pring)))
lpfc_sli_submit_iocb(phba, pring, iocb, nextiocb);
if (iocb)
lpfc_sli_update_ring(phba, pring);
else
lpfc_sli_update_full_ring(phba, pring);
}
return;
}
/**
* lpfc_sli_next_hbq_slot - Get next hbq entry for the HBQ
* @phba: Pointer to HBA context object.
* @hbqno: HBQ number.
*
* This function is called with hbalock held to get the next
* available slot for the given HBQ. If there is free slot
* available for the HBQ it will return pointer to the next available
* HBQ entry else it will return NULL.
**/
static struct lpfc_hbq_entry *
lpfc_sli_next_hbq_slot(struct lpfc_hba *phba, uint32_t hbqno)
{
struct hbq_s *hbqp = &phba->hbqs[hbqno];
lockdep_assert_held(&phba->hbalock);
if (hbqp->next_hbqPutIdx == hbqp->hbqPutIdx &&
++hbqp->next_hbqPutIdx >= hbqp->entry_count)
hbqp->next_hbqPutIdx = 0;
if (unlikely(hbqp->local_hbqGetIdx == hbqp->next_hbqPutIdx)) {
uint32_t raw_index = phba->hbq_get[hbqno];
uint32_t getidx = le32_to_cpu(raw_index);
hbqp->local_hbqGetIdx = getidx;
if (unlikely(hbqp->local_hbqGetIdx >= hbqp->entry_count)) {
lpfc_printf_log(phba, KERN_ERR,
LOG_SLI | LOG_VPORT,
"1802 HBQ %d: local_hbqGetIdx "
"%u is > than hbqp->entry_count %u\n",
hbqno, hbqp->local_hbqGetIdx,
hbqp->entry_count);
phba->link_state = LPFC_HBA_ERROR;
return NULL;
}
if (hbqp->local_hbqGetIdx == hbqp->next_hbqPutIdx)
return NULL;
}
return (struct lpfc_hbq_entry *) phba->hbqs[hbqno].hbq_virt +
hbqp->hbqPutIdx;
}
/**
* lpfc_sli_hbqbuf_free_all - Free all the hbq buffers
* @phba: Pointer to HBA context object.
*
* This function is called with no lock held to free all the
* hbq buffers while uninitializing the SLI interface. It also
* frees the HBQ buffers returned by the firmware but not yet
* processed by the upper layers.
**/
void
lpfc_sli_hbqbuf_free_all(struct lpfc_hba *phba)
{
struct lpfc_dmabuf *dmabuf, *next_dmabuf;
struct hbq_dmabuf *hbq_buf;
unsigned long flags;
int i, hbq_count;
hbq_count = lpfc_sli_hbq_count();
/* Return all memory used by all HBQs */
spin_lock_irqsave(&phba->hbalock, flags);
for (i = 0; i < hbq_count; ++i) {
list_for_each_entry_safe(dmabuf, next_dmabuf,
&phba->hbqs[i].hbq_buffer_list, list) {
hbq_buf = container_of(dmabuf, struct hbq_dmabuf, dbuf);
list_del(&hbq_buf->dbuf.list);
(phba->hbqs[i].hbq_free_buffer)(phba, hbq_buf);
}
phba->hbqs[i].buffer_count = 0;
}
/* Mark the HBQs not in use */
phba->hbq_in_use = 0;
spin_unlock_irqrestore(&phba->hbalock, flags);
}
/**
* lpfc_sli_hbq_to_firmware - Post the hbq buffer to firmware
* @phba: Pointer to HBA context object.
* @hbqno: HBQ number.
* @hbq_buf: Pointer to HBQ buffer.
*
* This function is called with the hbalock held to post a
* hbq buffer to the firmware. If the function finds an empty
* slot in the HBQ, it will post the buffer. The function will return
* pointer to the hbq entry if it successfully post the buffer
* else it will return NULL.
**/
static int
lpfc_sli_hbq_to_firmware(struct lpfc_hba *phba, uint32_t hbqno,
struct hbq_dmabuf *hbq_buf)
{
lockdep_assert_held(&phba->hbalock);
return phba->lpfc_sli_hbq_to_firmware(phba, hbqno, hbq_buf);
}
/**
* lpfc_sli_hbq_to_firmware_s3 - Post the hbq buffer to SLI3 firmware
* @phba: Pointer to HBA context object.
* @hbqno: HBQ number.
* @hbq_buf: Pointer to HBQ buffer.
*
* This function is called with the hbalock held to post a hbq buffer to the
* firmware. If the function finds an empty slot in the HBQ, it will post the
* buffer and place it on the hbq_buffer_list. The function will return zero if
* it successfully post the buffer else it will return an error.
**/
static int
lpfc_sli_hbq_to_firmware_s3(struct lpfc_hba *phba, uint32_t hbqno,
struct hbq_dmabuf *hbq_buf)
{
struct lpfc_hbq_entry *hbqe;
dma_addr_t physaddr = hbq_buf->dbuf.phys;
lockdep_assert_held(&phba->hbalock);
/* Get next HBQ entry slot to use */
hbqe = lpfc_sli_next_hbq_slot(phba, hbqno);
if (hbqe) {
struct hbq_s *hbqp = &phba->hbqs[hbqno];
hbqe->bde.addrHigh = le32_to_cpu(putPaddrHigh(physaddr));
hbqe->bde.addrLow = le32_to_cpu(putPaddrLow(physaddr));
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
hbqe->bde.tus.f.bdeSize = hbq_buf->total_size;
hbqe->bde.tus.f.bdeFlags = 0;
hbqe->bde.tus.w = le32_to_cpu(hbqe->bde.tus.w);
hbqe->buffer_tag = le32_to_cpu(hbq_buf->tag);
/* Sync SLIM */
hbqp->hbqPutIdx = hbqp->next_hbqPutIdx;
writel(hbqp->hbqPutIdx, phba->hbq_put + hbqno);
/* flush */
readl(phba->hbq_put + hbqno);
list_add_tail(&hbq_buf->dbuf.list, &hbqp->hbq_buffer_list);
return 0;
} else
return -ENOMEM;
}
/**
* lpfc_sli_hbq_to_firmware_s4 - Post the hbq buffer to SLI4 firmware
* @phba: Pointer to HBA context object.
* @hbqno: HBQ number.
* @hbq_buf: Pointer to HBQ buffer.
*
* This function is called with the hbalock held to post an RQE to the SLI4
* firmware. If able to post the RQE to the RQ it will queue the hbq entry to
* the hbq_buffer_list and return zero, otherwise it will return an error.
**/
static int
lpfc_sli_hbq_to_firmware_s4(struct lpfc_hba *phba, uint32_t hbqno,
struct hbq_dmabuf *hbq_buf)
{
int rc;
struct lpfc_rqe hrqe;
struct lpfc_rqe drqe;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_queue *hrq;
struct lpfc_queue *drq;
if (hbqno != LPFC_ELS_HBQ)
return 1;
hrq = phba->sli4_hba.hdr_rq;
drq = phba->sli4_hba.dat_rq;
lockdep_assert_held(&phba->hbalock);
hrqe.address_lo = putPaddrLow(hbq_buf->hbuf.phys);
hrqe.address_hi = putPaddrHigh(hbq_buf->hbuf.phys);
drqe.address_lo = putPaddrLow(hbq_buf->dbuf.phys);
drqe.address_hi = putPaddrHigh(hbq_buf->dbuf.phys);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
rc = lpfc_sli4_rq_put(hrq, drq, &hrqe, &drqe);
if (rc < 0)
return rc;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
hbq_buf->tag = (rc | (hbqno << 16));
list_add_tail(&hbq_buf->dbuf.list, &phba->hbqs[hbqno].hbq_buffer_list);
return 0;
}
/* HBQ for ELS and CT traffic. */
static struct lpfc_hbq_init lpfc_els_hbq = {
.rn = 1,
.entry_count = 256,
.mask_count = 0,
.profile = 0,
.ring_mask = (1 << LPFC_ELS_RING),
.buffer_count = 0,
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
.init_count = 40,
.add_count = 40,
};
/* Array of HBQs */
struct lpfc_hbq_init *lpfc_hbq_defs[] = {
&lpfc_els_hbq,
};
/**
* lpfc_sli_hbqbuf_fill_hbqs - Post more hbq buffers to HBQ
* @phba: Pointer to HBA context object.
* @hbqno: HBQ number.
* @count: Number of HBQ buffers to be posted.
*
* This function is called with no lock held to post more hbq buffers to the
* given HBQ. The function returns the number of HBQ buffers successfully
* posted.
**/
static int
lpfc_sli_hbqbuf_fill_hbqs(struct lpfc_hba *phba, uint32_t hbqno, uint32_t count)
{
uint32_t i, posted = 0;
unsigned long flags;
struct hbq_dmabuf *hbq_buffer;
LIST_HEAD(hbq_buf_list);
if (!phba->hbqs[hbqno].hbq_alloc_buffer)
return 0;
if ((phba->hbqs[hbqno].buffer_count + count) >
lpfc_hbq_defs[hbqno]->entry_count)
count = lpfc_hbq_defs[hbqno]->entry_count -
phba->hbqs[hbqno].buffer_count;
if (!count)
return 0;
/* Allocate HBQ entries */
for (i = 0; i < count; i++) {
hbq_buffer = (phba->hbqs[hbqno].hbq_alloc_buffer)(phba);
if (!hbq_buffer)
break;
list_add_tail(&hbq_buffer->dbuf.list, &hbq_buf_list);
}
/* Check whether HBQ is still in use */
spin_lock_irqsave(&phba->hbalock, flags);
if (!phba->hbq_in_use)
goto err;
while (!list_empty(&hbq_buf_list)) {
list_remove_head(&hbq_buf_list, hbq_buffer, struct hbq_dmabuf,
dbuf.list);
hbq_buffer->tag = (phba->hbqs[hbqno].buffer_count |
(hbqno << 16));
if (!lpfc_sli_hbq_to_firmware(phba, hbqno, hbq_buffer)) {
phba->hbqs[hbqno].buffer_count++;
posted++;
} else
(phba->hbqs[hbqno].hbq_free_buffer)(phba, hbq_buffer);
}
spin_unlock_irqrestore(&phba->hbalock, flags);
return posted;
err:
spin_unlock_irqrestore(&phba->hbalock, flags);
while (!list_empty(&hbq_buf_list)) {
list_remove_head(&hbq_buf_list, hbq_buffer, struct hbq_dmabuf,
dbuf.list);
(phba->hbqs[hbqno].hbq_free_buffer)(phba, hbq_buffer);
}
return 0;
}
/**
* lpfc_sli_hbqbuf_add_hbqs - Post more HBQ buffers to firmware
* @phba: Pointer to HBA context object.
* @qno: HBQ number.
*
* This function posts more buffers to the HBQ. This function
* is called with no lock held. The function returns the number of HBQ entries
* successfully allocated.
**/
int
lpfc_sli_hbqbuf_add_hbqs(struct lpfc_hba *phba, uint32_t qno)
{
if (phba->sli_rev == LPFC_SLI_REV4)
return 0;
else
return lpfc_sli_hbqbuf_fill_hbqs(phba, qno,
lpfc_hbq_defs[qno]->add_count);
}
/**
* lpfc_sli_hbqbuf_init_hbqs - Post initial buffers to the HBQ
* @phba: Pointer to HBA context object.
* @qno: HBQ queue number.
*
* This function is called from SLI initialization code path with
* no lock held to post initial HBQ buffers to firmware. The
* function returns the number of HBQ entries successfully allocated.
**/
static int
lpfc_sli_hbqbuf_init_hbqs(struct lpfc_hba *phba, uint32_t qno)
{
if (phba->sli_rev == LPFC_SLI_REV4)
return lpfc_sli_hbqbuf_fill_hbqs(phba, qno,
2011-10-10 19:32:10 -06:00
lpfc_hbq_defs[qno]->entry_count);
else
return lpfc_sli_hbqbuf_fill_hbqs(phba, qno,
lpfc_hbq_defs[qno]->init_count);
}
/**
* lpfc_sli_hbqbuf_get - Remove the first hbq off of an hbq list
* @phba: Pointer to HBA context object.
* @hbqno: HBQ number.
*
* This function removes the first hbq buffer on an hbq list and returns a
* pointer to that buffer. If it finds no buffers on the list it returns NULL.
**/
static struct hbq_dmabuf *
lpfc_sli_hbqbuf_get(struct list_head *rb_list)
{
struct lpfc_dmabuf *d_buf;
list_remove_head(rb_list, d_buf, struct lpfc_dmabuf, list);
if (!d_buf)
return NULL;
return container_of(d_buf, struct hbq_dmabuf, dbuf);
}
/**
* lpfc_sli_rqbuf_get - Remove the first dma buffer off of an RQ list
* @phba: Pointer to HBA context object.
* @hbqno: HBQ number.
*
* This function removes the first RQ buffer on an RQ buffer list and returns a
* pointer to that buffer. If it finds no buffers on the list it returns NULL.
**/
static struct rqb_dmabuf *
lpfc_sli_rqbuf_get(struct lpfc_hba *phba, struct lpfc_queue *hrq)
{
struct lpfc_dmabuf *h_buf;
struct lpfc_rqb *rqbp;
rqbp = hrq->rqbp;
list_remove_head(&rqbp->rqb_buffer_list, h_buf,
struct lpfc_dmabuf, list);
if (!h_buf)
return NULL;
rqbp->buffer_count--;
return container_of(h_buf, struct rqb_dmabuf, hbuf);
}
/**
* lpfc_sli_hbqbuf_find - Find the hbq buffer associated with a tag
* @phba: Pointer to HBA context object.
* @tag: Tag of the hbq buffer.
*
* This function searches for the hbq buffer associated with the given tag in
* the hbq buffer list. If it finds the hbq buffer, it returns the hbq_buffer
* otherwise it returns NULL.
**/
static struct hbq_dmabuf *
lpfc_sli_hbqbuf_find(struct lpfc_hba *phba, uint32_t tag)
{
struct lpfc_dmabuf *d_buf;
struct hbq_dmabuf *hbq_buf;
uint32_t hbqno;
hbqno = tag >> 16;
if (hbqno >= LPFC_MAX_HBQS)
return NULL;
spin_lock_irq(&phba->hbalock);
list_for_each_entry(d_buf, &phba->hbqs[hbqno].hbq_buffer_list, list) {
hbq_buf = container_of(d_buf, struct hbq_dmabuf, dbuf);
if (hbq_buf->tag == tag) {
spin_unlock_irq(&phba->hbalock);
return hbq_buf;
}
}
spin_unlock_irq(&phba->hbalock);
lpfc_printf_log(phba, KERN_ERR, LOG_SLI | LOG_VPORT,
"1803 Bad hbq tag. Data: x%x x%x\n",
tag, phba->hbqs[tag >> 16].buffer_count);
return NULL;
}
/**
* lpfc_sli_free_hbq - Give back the hbq buffer to firmware
* @phba: Pointer to HBA context object.
* @hbq_buffer: Pointer to HBQ buffer.
*
* This function is called with hbalock. This function gives back
* the hbq buffer to firmware. If the HBQ does not have space to
* post the buffer, it will free the buffer.
**/
void
lpfc_sli_free_hbq(struct lpfc_hba *phba, struct hbq_dmabuf *hbq_buffer)
{
uint32_t hbqno;
if (hbq_buffer) {
hbqno = hbq_buffer->tag >> 16;
if (lpfc_sli_hbq_to_firmware(phba, hbqno, hbq_buffer))
(phba->hbqs[hbqno].hbq_free_buffer)(phba, hbq_buffer);
}
}
/**
* lpfc_sli_chk_mbx_command - Check if the mailbox is a legitimate mailbox
* @mbxCommand: mailbox command code.
*
* This function is called by the mailbox event handler function to verify
* that the completed mailbox command is a legitimate mailbox command. If the
* completed mailbox is not known to the function, it will return MBX_SHUTDOWN
* and the mailbox event handler will take the HBA offline.
**/
static int
lpfc_sli_chk_mbx_command(uint8_t mbxCommand)
{
uint8_t ret;
switch (mbxCommand) {
case MBX_LOAD_SM:
case MBX_READ_NV:
case MBX_WRITE_NV:
case MBX_WRITE_VPARMS:
case MBX_RUN_BIU_DIAG:
case MBX_INIT_LINK:
case MBX_DOWN_LINK:
case MBX_CONFIG_LINK:
case MBX_CONFIG_RING:
case MBX_RESET_RING:
case MBX_READ_CONFIG:
case MBX_READ_RCONFIG:
case MBX_READ_SPARM:
case MBX_READ_STATUS:
case MBX_READ_RPI:
case MBX_READ_XRI:
case MBX_READ_REV:
case MBX_READ_LNK_STAT:
case MBX_REG_LOGIN:
case MBX_UNREG_LOGIN:
case MBX_CLEAR_LA:
case MBX_DUMP_MEMORY:
case MBX_DUMP_CONTEXT:
case MBX_RUN_DIAGS:
case MBX_RESTART:
case MBX_UPDATE_CFG:
case MBX_DOWN_LOAD:
case MBX_DEL_LD_ENTRY:
case MBX_RUN_PROGRAM:
case MBX_SET_MASK:
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
case MBX_SET_VARIABLE:
case MBX_UNREG_D_ID:
case MBX_KILL_BOARD:
case MBX_CONFIG_FARP:
case MBX_BEACON:
case MBX_LOAD_AREA:
case MBX_RUN_BIU_DIAG64:
case MBX_CONFIG_PORT:
case MBX_READ_SPARM64:
case MBX_READ_RPI64:
case MBX_REG_LOGIN64:
2010-11-20 21:11:48 -07:00
case MBX_READ_TOPOLOGY:
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
case MBX_WRITE_WWN:
case MBX_SET_DEBUG:
case MBX_LOAD_EXP_ROM:
case MBX_ASYNCEVT_ENABLE:
case MBX_REG_VPI:
case MBX_UNREG_VPI:
case MBX_HEARTBEAT:
case MBX_PORT_CAPABILITIES:
case MBX_PORT_IOV_CONTROL:
case MBX_SLI4_CONFIG:
case MBX_SLI4_REQ_FTRS:
case MBX_REG_FCFI:
case MBX_UNREG_FCFI:
case MBX_REG_VFI:
case MBX_UNREG_VFI:
case MBX_INIT_VPI:
case MBX_INIT_VFI:
case MBX_RESUME_RPI:
case MBX_READ_EVENT_LOG_STATUS:
case MBX_READ_EVENT_LOG:
case MBX_SECURITY_MGMT:
case MBX_AUTH_PORT:
case MBX_ACCESS_VDATA:
ret = mbxCommand;
break;
default:
ret = MBX_SHUTDOWN;
break;
}
return ret;
}
/**
* lpfc_sli_wake_mbox_wait - lpfc_sli_issue_mbox_wait mbox completion handler
* @phba: Pointer to HBA context object.
* @pmboxq: Pointer to mailbox command.
*
* This is completion handler function for mailbox commands issued from
* lpfc_sli_issue_mbox_wait function. This function is called by the
* mailbox event handler function with no lock held. This function
* will wake up thread waiting on the wait queue pointed by context1
* of the mailbox.
**/
void
lpfc_sli_wake_mbox_wait(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmboxq)
{
unsigned long drvr_flag;
struct completion *pmbox_done;
/*
* If pmbox_done is empty, the driver thread gave up waiting and
* continued running.
*/
pmboxq->mbox_flag |= LPFC_MBX_WAKE;
spin_lock_irqsave(&phba->hbalock, drvr_flag);
pmbox_done = (struct completion *)pmboxq->context3;
if (pmbox_done)
complete(pmbox_done);
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
return;
}
/**
* lpfc_sli_def_mbox_cmpl - Default mailbox completion handler
* @phba: Pointer to HBA context object.
* @pmb: Pointer to mailbox object.
*
* This function is the default mailbox completion handler. It
* frees the memory resources associated with the completed mailbox
* command. If the completed command is a REG_LOGIN mailbox command,
* this function will issue a UREG_LOGIN to re-claim the RPI.
**/
void
lpfc_sli_def_mbox_cmpl(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
{
struct lpfc_vport *vport = pmb->vport;
struct lpfc_dmabuf *mp;
struct lpfc_nodelist *ndlp;
struct Scsi_Host *shost;
uint16_t rpi, vpi;
int rc;
mp = (struct lpfc_dmabuf *) (pmb->context1);
if (mp) {
lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
}
/*
* If a REG_LOGIN succeeded after node is destroyed or node
* is in re-discovery driver need to cleanup the RPI.
*/
if (!(phba->pport->load_flag & FC_UNLOADING) &&
pmb->u.mb.mbxCommand == MBX_REG_LOGIN64 &&
!pmb->u.mb.mbxStatus) {
rpi = pmb->u.mb.un.varWords[0];
vpi = pmb->u.mb.un.varRegLogin.vpi;
lpfc_unreg_login(phba, vpi, rpi, pmb);
pmb->vport = vport;
pmb->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
rc = lpfc_sli_issue_mbox(phba, pmb, MBX_NOWAIT);
if (rc != MBX_NOT_FINISHED)
return;
}
if ((pmb->u.mb.mbxCommand == MBX_REG_VPI) &&
!(phba->pport->load_flag & FC_UNLOADING) &&
!pmb->u.mb.mbxStatus) {
shost = lpfc_shost_from_vport(vport);
spin_lock_irq(shost->host_lock);
vport->vpi_state |= LPFC_VPI_REGISTERED;
vport->fc_flag &= ~FC_VPORT_NEEDS_REG_VPI;
spin_unlock_irq(shost->host_lock);
}
if (pmb->u.mb.mbxCommand == MBX_REG_LOGIN64) {
ndlp = (struct lpfc_nodelist *)pmb->context2;
lpfc_nlp_put(ndlp);
pmb->context2 = NULL;
}
/* Check security permission status on INIT_LINK mailbox command */
if ((pmb->u.mb.mbxCommand == MBX_INIT_LINK) &&
(pmb->u.mb.mbxStatus == MBXERR_SEC_NO_PERMISSION))
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"2860 SLI authentication is required "
"for INIT_LINK but has not done yet\n");
if (bf_get(lpfc_mqe_command, &pmb->u.mqe) == MBX_SLI4_CONFIG)
lpfc_sli4_mbox_cmd_free(phba, pmb);
else
mempool_free(pmb, phba->mbox_mem_pool);
}
/**
* lpfc_sli4_unreg_rpi_cmpl_clr - mailbox completion handler
* @phba: Pointer to HBA context object.
* @pmb: Pointer to mailbox object.
*
* This function is the unreg rpi mailbox completion handler. It
* frees the memory resources associated with the completed mailbox
* command. An additional refrenece is put on the ndlp to prevent
* lpfc_nlp_release from freeing the rpi bit in the bitmask before
* the unreg mailbox command completes, this routine puts the
* reference back.
*
**/
void
lpfc_sli4_unreg_rpi_cmpl_clr(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
{
struct lpfc_vport *vport = pmb->vport;
struct lpfc_nodelist *ndlp;
ndlp = pmb->context1;
if (pmb->u.mb.mbxCommand == MBX_UNREG_LOGIN) {
if (phba->sli_rev == LPFC_SLI_REV4 &&
(bf_get(lpfc_sli_intf_if_type,
&phba->sli4_hba.sli_intf) >=
LPFC_SLI_INTF_IF_TYPE_2)) {
if (ndlp) {
lpfc_printf_vlog(vport, KERN_INFO, LOG_SLI,
"0010 UNREG_LOGIN vpi:%x "
"rpi:%x DID:%x map:%x %p\n",
vport->vpi, ndlp->nlp_rpi,
ndlp->nlp_DID,
ndlp->nlp_usg_map, ndlp);
ndlp->nlp_flag &= ~NLP_LOGO_ACC;
lpfc_nlp_put(ndlp);
}
}
}
mempool_free(pmb, phba->mbox_mem_pool);
}
/**
* lpfc_sli_handle_mb_event - Handle mailbox completions from firmware
* @phba: Pointer to HBA context object.
*
* This function is called with no lock held. This function processes all
* the completed mailbox commands and gives it to upper layers. The interrupt
* service routine processes mailbox completion interrupt and adds completed
* mailbox commands to the mboxq_cmpl queue and signals the worker thread.
* Worker thread call lpfc_sli_handle_mb_event, which will return the
* completed mailbox commands in mboxq_cmpl queue to the upper layers. This
* function returns the mailbox commands to the upper layer by calling the
* completion handler function of each mailbox.
**/
int
lpfc_sli_handle_mb_event(struct lpfc_hba *phba)
{
MAILBOX_t *pmbox;
LPFC_MBOXQ_t *pmb;
int rc;
LIST_HEAD(cmplq);
phba->sli.slistat.mbox_event++;
/* Get all completed mailboxe buffers into the cmplq */
spin_lock_irq(&phba->hbalock);
list_splice_init(&phba->sli.mboxq_cmpl, &cmplq);
spin_unlock_irq(&phba->hbalock);
/* Get a Mailbox buffer to setup mailbox commands for callback */
do {
list_remove_head(&cmplq, pmb, LPFC_MBOXQ_t, list);
if (pmb == NULL)
break;
pmbox = &pmb->u.mb;
if (pmbox->mbxCommand != MBX_HEARTBEAT) {
if (pmb->vport) {
lpfc_debugfs_disc_trc(pmb->vport,
LPFC_DISC_TRC_MBOX_VPORT,
"MBOX cmpl vport: cmd:x%x mb:x%x x%x",
(uint32_t)pmbox->mbxCommand,
pmbox->un.varWords[0],
pmbox->un.varWords[1]);
}
else {
lpfc_debugfs_disc_trc(phba->pport,
LPFC_DISC_TRC_MBOX,
"MBOX cmpl: cmd:x%x mb:x%x x%x",
(uint32_t)pmbox->mbxCommand,
pmbox->un.varWords[0],
pmbox->un.varWords[1]);
}
}
/*
* It is a fatal error if unknown mbox command completion.
*/
if (lpfc_sli_chk_mbx_command(pmbox->mbxCommand) ==
MBX_SHUTDOWN) {
/* Unknown mailbox command compl */
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):0323 Unknown Mailbox command "
"x%x (x%x/x%x) Cmpl\n",
pmb->vport ? pmb->vport->vpi : 0,
pmbox->mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba,
pmb),
lpfc_sli_config_mbox_opcode_get(phba,
pmb));
phba->link_state = LPFC_HBA_ERROR;
phba->work_hs = HS_FFER3;
lpfc_handle_eratt(phba);
continue;
}
if (pmbox->mbxStatus) {
phba->sli.slistat.mbox_stat_err++;
if (pmbox->mbxStatus == MBXERR_NO_RESOURCES) {
/* Mbox cmd cmpl error - RETRYing */
lpfc_printf_log(phba, KERN_INFO,
LOG_MBOX | LOG_SLI,
"(%d):0305 Mbox cmd cmpl "
"error - RETRYing Data: x%x "
"(x%x/x%x) x%x x%x x%x\n",
pmb->vport ? pmb->vport->vpi : 0,
pmbox->mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba,
pmb),
lpfc_sli_config_mbox_opcode_get(phba,
pmb),
pmbox->mbxStatus,
pmbox->un.varWords[0],
pmb->vport->port_state);
pmbox->mbxStatus = 0;
pmbox->mbxOwner = OWN_HOST;
rc = lpfc_sli_issue_mbox(phba, pmb, MBX_NOWAIT);
if (rc != MBX_NOT_FINISHED)
continue;
}
}
/* Mailbox cmd <cmd> Cmpl <cmpl> */
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):0307 Mailbox cmd x%x (x%x/x%x) Cmpl x%p "
"Data: x%x x%x x%x x%x x%x x%x x%x x%x x%x "
"x%x x%x x%x\n",
pmb->vport ? pmb->vport->vpi : 0,
pmbox->mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba, pmb),
lpfc_sli_config_mbox_opcode_get(phba, pmb),
pmb->mbox_cmpl,
*((uint32_t *) pmbox),
pmbox->un.varWords[0],
pmbox->un.varWords[1],
pmbox->un.varWords[2],
pmbox->un.varWords[3],
pmbox->un.varWords[4],
pmbox->un.varWords[5],
pmbox->un.varWords[6],
pmbox->un.varWords[7],
pmbox->un.varWords[8],
pmbox->un.varWords[9],
pmbox->un.varWords[10]);
if (pmb->mbox_cmpl)
pmb->mbox_cmpl(phba,pmb);
} while (1);
return 0;
}
/**
* lpfc_sli_get_buff - Get the buffer associated with the buffer tag
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @tag: buffer tag.
*
* This function is called with no lock held. When QUE_BUFTAG_BIT bit
* is set in the tag the buffer is posted for a particular exchange,
* the function will return the buffer without replacing the buffer.
* If the buffer is for unsolicited ELS or CT traffic, this function
* returns the buffer and also posts another buffer to the firmware.
**/
static struct lpfc_dmabuf *
lpfc_sli_get_buff(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring,
uint32_t tag)
{
struct hbq_dmabuf *hbq_entry;
if (tag & QUE_BUFTAG_BIT)
return lpfc_sli_ring_taggedbuf_get(phba, pring, tag);
hbq_entry = lpfc_sli_hbqbuf_find(phba, tag);
if (!hbq_entry)
return NULL;
return &hbq_entry->dbuf;
}
/**
* lpfc_complete_unsol_iocb - Complete an unsolicited sequence
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @saveq: Pointer to the iocbq struct representing the sequence starting frame.
* @fch_r_ctl: the r_ctl for the first frame of the sequence.
* @fch_type: the type for the first frame of the sequence.
*
* This function is called with no lock held. This function uses the r_ctl and
* type of the received sequence to find the correct callback function to call
* to process the sequence.
**/
static int
lpfc_complete_unsol_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *saveq, uint32_t fch_r_ctl,
uint32_t fch_type)
{
int i;
switch (fch_type) {
case FC_TYPE_NVME:
lpfc_nvmet_unsol_ls_event(phba, pring, saveq);
return 1;
default:
break;
}
/* unSolicited Responses */
if (pring->prt[0].profile) {
if (pring->prt[0].lpfc_sli_rcv_unsol_event)
(pring->prt[0].lpfc_sli_rcv_unsol_event) (phba, pring,
saveq);
return 1;
}
/* We must search, based on rctl / type
for the right routine */
for (i = 0; i < pring->num_mask; i++) {
if ((pring->prt[i].rctl == fch_r_ctl) &&
(pring->prt[i].type == fch_type)) {
if (pring->prt[i].lpfc_sli_rcv_unsol_event)
(pring->prt[i].lpfc_sli_rcv_unsol_event)
(phba, pring, saveq);
return 1;
}
}
return 0;
}
/**
* lpfc_sli_process_unsol_iocb - Unsolicited iocb handler
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @saveq: Pointer to the unsolicited iocb.
*
* This function is called with no lock held by the ring event handler
* when there is an unsolicited iocb posted to the response ring by the
* firmware. This function gets the buffer associated with the iocbs
* and calls the event handler for the ring. This function handles both
* qring buffers and hbq buffers.
* When the function returns 1 the caller can free the iocb object otherwise
* upper layer functions will free the iocb objects.
**/
static int
lpfc_sli_process_unsol_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *saveq)
{
IOCB_t * irsp;
WORD5 * w5p;
uint32_t Rctl, Type;
struct lpfc_iocbq *iocbq;
struct lpfc_dmabuf *dmzbuf;
irsp = &(saveq->iocb);
if (irsp->ulpCommand == CMD_ASYNC_STATUS) {
if (pring->lpfc_sli_rcv_async_status)
pring->lpfc_sli_rcv_async_status(phba, pring, saveq);
else
lpfc_printf_log(phba,
KERN_WARNING,
LOG_SLI,
"0316 Ring %d handler: unexpected "
"ASYNC_STATUS iocb received evt_code "
"0x%x\n",
pring->ringno,
irsp->un.asyncstat.evt_code);
return 1;
}
if ((irsp->ulpCommand == CMD_IOCB_RET_XRI64_CX) &&
(phba->sli3_options & LPFC_SLI3_HBQ_ENABLED)) {
if (irsp->ulpBdeCount > 0) {
dmzbuf = lpfc_sli_get_buff(phba, pring,
irsp->un.ulpWord[3]);
lpfc_in_buf_free(phba, dmzbuf);
}
if (irsp->ulpBdeCount > 1) {
dmzbuf = lpfc_sli_get_buff(phba, pring,
irsp->unsli3.sli3Words[3]);
lpfc_in_buf_free(phba, dmzbuf);
}
if (irsp->ulpBdeCount > 2) {
dmzbuf = lpfc_sli_get_buff(phba, pring,
irsp->unsli3.sli3Words[7]);
lpfc_in_buf_free(phba, dmzbuf);
}
return 1;
}
if (phba->sli3_options & LPFC_SLI3_HBQ_ENABLED) {
if (irsp->ulpBdeCount != 0) {
saveq->context2 = lpfc_sli_get_buff(phba, pring,
irsp->un.ulpWord[3]);
if (!saveq->context2)
lpfc_printf_log(phba,
KERN_ERR,
LOG_SLI,
"0341 Ring %d Cannot find buffer for "
"an unsolicited iocb. tag 0x%x\n",
pring->ringno,
irsp->un.ulpWord[3]);
}
if (irsp->ulpBdeCount == 2) {
saveq->context3 = lpfc_sli_get_buff(phba, pring,
irsp->unsli3.sli3Words[7]);
if (!saveq->context3)
lpfc_printf_log(phba,
KERN_ERR,
LOG_SLI,
"0342 Ring %d Cannot find buffer for an"
" unsolicited iocb. tag 0x%x\n",
pring->ringno,
irsp->unsli3.sli3Words[7]);
}
list_for_each_entry(iocbq, &saveq->list, list) {
irsp = &(iocbq->iocb);
if (irsp->ulpBdeCount != 0) {
iocbq->context2 = lpfc_sli_get_buff(phba, pring,
irsp->un.ulpWord[3]);
if (!iocbq->context2)
lpfc_printf_log(phba,
KERN_ERR,
LOG_SLI,
"0343 Ring %d Cannot find "
"buffer for an unsolicited iocb"
". tag 0x%x\n", pring->ringno,
irsp->un.ulpWord[3]);
}
if (irsp->ulpBdeCount == 2) {
iocbq->context3 = lpfc_sli_get_buff(phba, pring,
irsp->unsli3.sli3Words[7]);
if (!iocbq->context3)
lpfc_printf_log(phba,
KERN_ERR,
LOG_SLI,
"0344 Ring %d Cannot find "
"buffer for an unsolicited "
"iocb. tag 0x%x\n",
pring->ringno,
irsp->unsli3.sli3Words[7]);
}
}
}
if (irsp->ulpBdeCount != 0 &&
(irsp->ulpCommand == CMD_IOCB_RCV_CONT64_CX ||
irsp->ulpStatus == IOSTAT_INTERMED_RSP)) {
int found = 0;
/* search continue save q for same XRI */
list_for_each_entry(iocbq, &pring->iocb_continue_saveq, clist) {
if (iocbq->iocb.unsli3.rcvsli3.ox_id ==
saveq->iocb.unsli3.rcvsli3.ox_id) {
list_add_tail(&saveq->list, &iocbq->list);
found = 1;
break;
}
}
if (!found)
list_add_tail(&saveq->clist,
&pring->iocb_continue_saveq);
if (saveq->iocb.ulpStatus != IOSTAT_INTERMED_RSP) {
list_del_init(&iocbq->clist);
saveq = iocbq;
irsp = &(saveq->iocb);
} else
return 0;
}
if ((irsp->ulpCommand == CMD_RCV_ELS_REQ64_CX) ||
(irsp->ulpCommand == CMD_RCV_ELS_REQ_CX) ||
(irsp->ulpCommand == CMD_IOCB_RCV_ELS64_CX)) {
Rctl = FC_RCTL_ELS_REQ;
Type = FC_TYPE_ELS;
} else {
w5p = (WORD5 *)&(saveq->iocb.un.ulpWord[5]);
Rctl = w5p->hcsw.Rctl;
Type = w5p->hcsw.Type;
/* Firmware Workaround */
if ((Rctl == 0) && (pring->ringno == LPFC_ELS_RING) &&
(irsp->ulpCommand == CMD_RCV_SEQUENCE64_CX ||
irsp->ulpCommand == CMD_IOCB_RCV_SEQ64_CX)) {
Rctl = FC_RCTL_ELS_REQ;
Type = FC_TYPE_ELS;
w5p->hcsw.Rctl = Rctl;
w5p->hcsw.Type = Type;
}
}
if (!lpfc_complete_unsol_iocb(phba, pring, saveq, Rctl, Type))
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0313 Ring %d handler: unexpected Rctl x%x "
"Type x%x received\n",
pring->ringno, Rctl, Type);
return 1;
}
/**
* lpfc_sli_iocbq_lookup - Find command iocb for the given response iocb
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @prspiocb: Pointer to response iocb object.
*
* This function looks up the iocb_lookup table to get the command iocb
* corresponding to the given response iocb using the iotag of the
* response iocb. This function is called with the hbalock held
* for sli3 devices or the ring_lock for sli4 devices.
* This function returns the command iocb object if it finds the command
* iocb else returns NULL.
**/
static struct lpfc_iocbq *
lpfc_sli_iocbq_lookup(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring,
struct lpfc_iocbq *prspiocb)
{
struct lpfc_iocbq *cmd_iocb = NULL;
uint16_t iotag;
lockdep_assert_held(&phba->hbalock);
iotag = prspiocb->iocb.ulpIoTag;
if (iotag != 0 && iotag <= phba->sli.last_iotag) {
cmd_iocb = phba->sli.iocbq_lookup[iotag];
if (cmd_iocb->iocb_flag & LPFC_IO_ON_TXCMPLQ) {
/* remove from txcmpl queue list */
list_del_init(&cmd_iocb->list);
cmd_iocb->iocb_flag &= ~LPFC_IO_ON_TXCMPLQ;
return cmd_iocb;
}
}
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0317 iotag x%x is out of "
"range: max iotag x%x wd0 x%x\n",
iotag, phba->sli.last_iotag,
*(((uint32_t *) &prspiocb->iocb) + 7));
return NULL;
}
/**
* lpfc_sli_iocbq_lookup_by_tag - Find command iocb for the iotag
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @iotag: IOCB tag.
*
* This function looks up the iocb_lookup table to get the command iocb
* corresponding to the given iotag. This function is called with the
* hbalock held.
* This function returns the command iocb object if it finds the command
* iocb else returns NULL.
**/
static struct lpfc_iocbq *
lpfc_sli_iocbq_lookup_by_tag(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring, uint16_t iotag)
{
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_iocbq *cmd_iocb = NULL;
lockdep_assert_held(&phba->hbalock);
if (iotag != 0 && iotag <= phba->sli.last_iotag) {
cmd_iocb = phba->sli.iocbq_lookup[iotag];
if (cmd_iocb->iocb_flag & LPFC_IO_ON_TXCMPLQ) {
/* remove from txcmpl queue list */
list_del_init(&cmd_iocb->list);
cmd_iocb->iocb_flag &= ~LPFC_IO_ON_TXCMPLQ;
return cmd_iocb;
}
}
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
"0372 iotag x%x lookup error: max iotag (x%x) "
"iocb_flag x%x\n",
iotag, phba->sli.last_iotag,
cmd_iocb ? cmd_iocb->iocb_flag : 0xffff);
return NULL;
}
/**
* lpfc_sli_process_sol_iocb - process solicited iocb completion
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @saveq: Pointer to the response iocb to be processed.
*
* This function is called by the ring event handler for non-fcp
* rings when there is a new response iocb in the response ring.
* The caller is not required to hold any locks. This function
* gets the command iocb associated with the response iocb and
* calls the completion handler for the command iocb. If there
* is no completion handler, the function will free the resources
* associated with command iocb. If the response iocb is for
* an already aborted command iocb, the status of the completion
* is changed to IOSTAT_LOCAL_REJECT/IOERR_SLI_ABORTED.
* This function always returns 1.
**/
static int
lpfc_sli_process_sol_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *saveq)
{
struct lpfc_iocbq *cmdiocbp;
int rc = 1;
unsigned long iflag;
/* Based on the iotag field, get the cmd IOCB from the txcmplq */
if (phba->sli_rev == LPFC_SLI_REV4)
spin_lock_irqsave(&pring->ring_lock, iflag);
else
spin_lock_irqsave(&phba->hbalock, iflag);
cmdiocbp = lpfc_sli_iocbq_lookup(phba, pring, saveq);
if (phba->sli_rev == LPFC_SLI_REV4)
spin_unlock_irqrestore(&pring->ring_lock, iflag);
else
spin_unlock_irqrestore(&phba->hbalock, iflag);
if (cmdiocbp) {
if (cmdiocbp->iocb_cmpl) {
/*
* If an ELS command failed send an event to mgmt
* application.
*/
if (saveq->iocb.ulpStatus &&
(pring->ringno == LPFC_ELS_RING) &&
(cmdiocbp->iocb.ulpCommand ==
CMD_ELS_REQUEST64_CR))
lpfc_send_els_failure_event(phba,
cmdiocbp, saveq);
/*
* Post all ELS completions to the worker thread.
* All other are passed to the completion callback.
*/
if (pring->ringno == LPFC_ELS_RING) {
if ((phba->sli_rev < LPFC_SLI_REV4) &&
(cmdiocbp->iocb_flag &
LPFC_DRIVER_ABORTED)) {
spin_lock_irqsave(&phba->hbalock,
iflag);
cmdiocbp->iocb_flag &=
~LPFC_DRIVER_ABORTED;
spin_unlock_irqrestore(&phba->hbalock,
iflag);
saveq->iocb.ulpStatus =
IOSTAT_LOCAL_REJECT;
saveq->iocb.un.ulpWord[4] =
IOERR_SLI_ABORTED;
/* Firmware could still be in progress
* of DMAing payload, so don't free data
* buffer till after a hbeat.
*/
spin_lock_irqsave(&phba->hbalock,
iflag);
saveq->iocb_flag |= LPFC_DELAY_MEM_FREE;
spin_unlock_irqrestore(&phba->hbalock,
iflag);
}
if (phba->sli_rev == LPFC_SLI_REV4) {
if (saveq->iocb_flag &
LPFC_EXCHANGE_BUSY) {
/* Set cmdiocb flag for the
* exchange busy so sgl (xri)
* will not be released until
* the abort xri is received
* from hba.
*/
spin_lock_irqsave(
&phba->hbalock, iflag);
cmdiocbp->iocb_flag |=
LPFC_EXCHANGE_BUSY;
spin_unlock_irqrestore(
&phba->hbalock, iflag);
}
if (cmdiocbp->iocb_flag &
LPFC_DRIVER_ABORTED) {
/*
* Clear LPFC_DRIVER_ABORTED
* bit in case it was driver
* initiated abort.
*/
spin_lock_irqsave(
&phba->hbalock, iflag);
cmdiocbp->iocb_flag &=
~LPFC_DRIVER_ABORTED;
spin_unlock_irqrestore(
&phba->hbalock, iflag);
cmdiocbp->iocb.ulpStatus =
IOSTAT_LOCAL_REJECT;
cmdiocbp->iocb.un.ulpWord[4] =
IOERR_ABORT_REQUESTED;
/*
* For SLI4, irsiocb contains
* NO_XRI in sli_xritag, it
* shall not affect releasing
* sgl (xri) process.
*/
saveq->iocb.ulpStatus =
IOSTAT_LOCAL_REJECT;
saveq->iocb.un.ulpWord[4] =
IOERR_SLI_ABORTED;
spin_lock_irqsave(
&phba->hbalock, iflag);
saveq->iocb_flag |=
LPFC_DELAY_MEM_FREE;
spin_unlock_irqrestore(
&phba->hbalock, iflag);
}
}
}
(cmdiocbp->iocb_cmpl) (phba, cmdiocbp, saveq);
} else
lpfc_sli_release_iocbq(phba, cmdiocbp);
} else {
/*
* Unknown initiating command based on the response iotag.
* This could be the case on the ELS ring because of
* lpfc_els_abort().
*/
if (pring->ringno != LPFC_ELS_RING) {
/*
* Ring <ringno> handler: unexpected completion IoTag
* <IoTag>
*/
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0322 Ring %d handler: "
"unexpected completion IoTag x%x "
"Data: x%x x%x x%x x%x\n",
pring->ringno,
saveq->iocb.ulpIoTag,
saveq->iocb.ulpStatus,
saveq->iocb.un.ulpWord[4],
saveq->iocb.ulpCommand,
saveq->iocb.ulpContext);
}
}
return rc;
}
/**
* lpfc_sli_rsp_pointers_error - Response ring pointer error handler
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function is called from the iocb ring event handlers when
* put pointer is ahead of the get pointer for a ring. This function signal
* an error attention condition to the worker thread and the worker
* thread will transition the HBA to offline state.
**/
static void
lpfc_sli_rsp_pointers_error(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
{
struct lpfc_pgp *pgp = &phba->port_gp[pring->ringno];
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
/*
* Ring <ringno> handler: portRspPut <portRspPut> is bigger than
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
* rsp ring <portRspMax>
*/
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0312 Ring %d handler: portRspPut %d "
"is bigger than rsp ring %d\n",
pring->ringno, le32_to_cpu(pgp->rspPutInx),
pring->sli.sli3.numRiocb);
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
phba->link_state = LPFC_HBA_ERROR;
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
/*
* All error attention handlers are posted to
* worker thread
*/
phba->work_ha |= HA_ERATT;
phba->work_hs = HS_FFER3;
lpfc_worker_wake_up(phba);
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
return;
}
/**
* lpfc_poll_eratt - Error attention polling timer timeout handler
* @ptr: Pointer to address of HBA context object.
*
* This function is invoked by the Error Attention polling timer when the
* timer times out. It will check the SLI Error Attention register for
* possible attention events. If so, it will post an Error Attention event
* and wake up worker thread to process it. Otherwise, it will set up the
* Error Attention polling timer for the next poll.
**/
void lpfc_poll_eratt(struct timer_list *t)
{
struct lpfc_hba *phba;
uint32_t eratt = 0;
uint64_t sli_intr, cnt;
phba = from_timer(phba, t, eratt_poll);
/* Here we will also keep track of interrupts per sec of the hba */
sli_intr = phba->sli.slistat.sli_intr;
if (phba->sli.slistat.sli_prev_intr > sli_intr)
cnt = (((uint64_t)(-1) - phba->sli.slistat.sli_prev_intr) +
sli_intr);
else
cnt = (sli_intr - phba->sli.slistat.sli_prev_intr);
/* 64-bit integer division not supported on 32-bit x86 - use do_div */
do_div(cnt, phba->eratt_poll_interval);
phba->sli.slistat.sli_ips = cnt;
phba->sli.slistat.sli_prev_intr = sli_intr;
/* Check chip HA register for error event */
eratt = lpfc_sli_check_eratt(phba);
if (eratt)
/* Tell the worker thread there is work to do */
lpfc_worker_wake_up(phba);
else
/* Restart the timer for next eratt poll */
mod_timer(&phba->eratt_poll,
jiffies +
msecs_to_jiffies(1000 * phba->eratt_poll_interval));
return;
}
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
/**
* lpfc_sli_handle_fast_ring_event - Handle ring events on FCP ring
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @mask: Host attention register mask for this ring.
*
* This function is called from the interrupt context when there is a ring
* event for the fcp ring. The caller does not hold any lock.
* The function processes each response iocb in the response ring until it
* finds an iocb with LE bit set and chains all the iocbs up to the iocb with
* LE bit set. The function will call the completion handler of the command iocb
* if the response iocb indicates a completion for a command iocb or it is
* an abort completion. The function will call lpfc_sli_process_unsol_iocb
* function if this is an unsolicited iocb.
* This routine presumes LPFC_FCP_RING handling and doesn't bother
* to check it explicitly.
*/
int
lpfc_sli_handle_fast_ring_event(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring, uint32_t mask)
{
struct lpfc_pgp *pgp = &phba->port_gp[pring->ringno];
IOCB_t *irsp = NULL;
IOCB_t *entry = NULL;
struct lpfc_iocbq *cmdiocbq = NULL;
struct lpfc_iocbq rspiocbq;
uint32_t status;
uint32_t portRspPut, portRspMax;
int rc = 1;
lpfc_iocb_type type;
unsigned long iflag;
uint32_t rsp_cmpl = 0;
spin_lock_irqsave(&phba->hbalock, iflag);
pring->stats.iocb_event++;
/*
* The next available response entry should never exceed the maximum
* entries. If it does, treat it as an adapter hardware error.
*/
portRspMax = pring->sli.sli3.numRiocb;
portRspPut = le32_to_cpu(pgp->rspPutInx);
if (unlikely(portRspPut >= portRspMax)) {
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
lpfc_sli_rsp_pointers_error(phba, pring);
spin_unlock_irqrestore(&phba->hbalock, iflag);
return 1;
}
if (phba->fcp_ring_in_use) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
return 1;
} else
phba->fcp_ring_in_use = 1;
rmb();
while (pring->sli.sli3.rspidx != portRspPut) {
/*
* Fetch an entry off the ring and copy it into a local data
* structure. The copy involves a byte-swap since the
* network byte order and pci byte orders are different.
*/
entry = lpfc_resp_iocb(phba, pring);
phba->last_completion_time = jiffies;
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
if (++pring->sli.sli3.rspidx >= portRspMax)
pring->sli.sli3.rspidx = 0;
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
lpfc_sli_pcimem_bcopy((uint32_t *) entry,
(uint32_t *) &rspiocbq.iocb,
phba->iocb_rsp_size);
INIT_LIST_HEAD(&(rspiocbq.list));
irsp = &rspiocbq.iocb;
type = lpfc_sli_iocb_cmd_type(irsp->ulpCommand & CMD_IOCB_MASK);
pring->stats.iocb_rsp++;
rsp_cmpl++;
if (unlikely(irsp->ulpStatus)) {
/*
* If resource errors reported from HBA, reduce
* queuedepths of the SCSI device.
*/
if ((irsp->ulpStatus == IOSTAT_LOCAL_REJECT) &&
((irsp->un.ulpWord[4] & IOERR_PARAM_MASK) ==
IOERR_NO_RESOURCES)) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
phba->lpfc_rampdown_queue_depth(phba);
spin_lock_irqsave(&phba->hbalock, iflag);
}
/* Rsp ring <ringno> error: IOCB */
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0336 Rsp Ring %d error: IOCB Data: "
"x%x x%x x%x x%x x%x x%x x%x x%x\n",
pring->ringno,
irsp->un.ulpWord[0],
irsp->un.ulpWord[1],
irsp->un.ulpWord[2],
irsp->un.ulpWord[3],
irsp->un.ulpWord[4],
irsp->un.ulpWord[5],
*(uint32_t *)&irsp->un1,
*((uint32_t *)&irsp->un1 + 1));
}
switch (type) {
case LPFC_ABORT_IOCB:
case LPFC_SOL_IOCB:
/*
* Idle exchange closed via ABTS from port. No iocb
* resources need to be recovered.
*/
if (unlikely(irsp->ulpCommand == CMD_XRI_ABORTED_CX)) {
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0333 IOCB cmd 0x%x"
" processed. Skipping"
" completion\n",
irsp->ulpCommand);
break;
}
cmdiocbq = lpfc_sli_iocbq_lookup(phba, pring,
&rspiocbq);
if (unlikely(!cmdiocbq))
break;
if (cmdiocbq->iocb_flag & LPFC_DRIVER_ABORTED)
cmdiocbq->iocb_flag &= ~LPFC_DRIVER_ABORTED;
if (cmdiocbq->iocb_cmpl) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
(cmdiocbq->iocb_cmpl)(phba, cmdiocbq,
&rspiocbq);
spin_lock_irqsave(&phba->hbalock, iflag);
}
break;
case LPFC_UNSOL_IOCB:
spin_unlock_irqrestore(&phba->hbalock, iflag);
lpfc_sli_process_unsol_iocb(phba, pring, &rspiocbq);
spin_lock_irqsave(&phba->hbalock, iflag);
break;
default:
if (irsp->ulpCommand == CMD_ADAPTER_MSG) {
char adaptermsg[LPFC_MAX_ADPTMSG];
memset(adaptermsg, 0, LPFC_MAX_ADPTMSG);
memcpy(&adaptermsg[0], (uint8_t *) irsp,
MAX_MSG_DATA);
dev_warn(&((phba->pcidev)->dev),
"lpfc%d: %s\n",
phba->brd_no, adaptermsg);
} else {
/* Unknown IOCB command */
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0334 Unknown IOCB command "
"Data: x%x, x%x x%x x%x x%x\n",
type, irsp->ulpCommand,
irsp->ulpStatus,
irsp->ulpIoTag,
irsp->ulpContext);
}
break;
}
/*
* The response IOCB has been processed. Update the ring
* pointer in SLIM. If the port response put pointer has not
* been updated, sync the pgp->rspPutInx and fetch the new port
* response put pointer.
*/
writel(pring->sli.sli3.rspidx,
&phba->host_gp[pring->ringno].rspGetInx);
if (pring->sli.sli3.rspidx == portRspPut)
portRspPut = le32_to_cpu(pgp->rspPutInx);
}
if ((rsp_cmpl > 0) && (mask & HA_R0RE_REQ)) {
pring->stats.iocb_rsp_full++;
status = ((CA_R0ATT | CA_R0RE_RSP) << (pring->ringno * 4));
writel(status, phba->CAregaddr);
readl(phba->CAregaddr);
}
if ((mask & HA_R0CE_RSP) && (pring->flag & LPFC_CALL_RING_AVAILABLE)) {
pring->flag &= ~LPFC_CALL_RING_AVAILABLE;
pring->stats.iocb_cmd_empty++;
/* Force update of the local copy of cmdGetInx */
pring->sli.sli3.local_getidx = le32_to_cpu(pgp->cmdGetInx);
lpfc_sli_resume_iocb(phba, pring);
if ((pring->lpfc_sli_cmd_available))
(pring->lpfc_sli_cmd_available) (phba, pring);
}
phba->fcp_ring_in_use = 0;
spin_unlock_irqrestore(&phba->hbalock, iflag);
return rc;
}
/**
* lpfc_sli_sp_handle_rspiocb - Handle slow-path response iocb
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @rspiocbp: Pointer to driver response IOCB object.
*
* This function is called from the worker thread when there is a slow-path
* response IOCB to process. This function chains all the response iocbs until
* seeing the iocb with the LE bit set. The function will call
* lpfc_sli_process_sol_iocb function if the response iocb indicates a
* completion of a command iocb. The function will call the
* lpfc_sli_process_unsol_iocb function if this is an unsolicited iocb.
* The function frees the resources or calls the completion handler if this
* iocb is an abort completion. The function returns NULL when the response
* iocb has the LE bit set and all the chained iocbs are processed, otherwise
* this function shall chain the iocb on to the iocb_continueq and return the
* response iocb passed in.
**/
static struct lpfc_iocbq *
lpfc_sli_sp_handle_rspiocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *rspiocbp)
{
struct lpfc_iocbq *saveq;
struct lpfc_iocbq *cmdiocbp;
struct lpfc_iocbq *next_iocb;
IOCB_t *irsp = NULL;
uint32_t free_saveq;
uint8_t iocb_cmd_type;
lpfc_iocb_type type;
unsigned long iflag;
int rc;
spin_lock_irqsave(&phba->hbalock, iflag);
/* First add the response iocb to the countinueq list */
list_add_tail(&rspiocbp->list, &(pring->iocb_continueq));
pring->iocb_continueq_cnt++;
/* Now, determine whether the list is completed for processing */
irsp = &rspiocbp->iocb;
if (irsp->ulpLe) {
/*
* By default, the driver expects to free all resources
* associated with this iocb completion.
*/
free_saveq = 1;
saveq = list_get_first(&pring->iocb_continueq,
struct lpfc_iocbq, list);
irsp = &(saveq->iocb);
list_del_init(&pring->iocb_continueq);
pring->iocb_continueq_cnt = 0;
pring->stats.iocb_rsp++;
/*
* If resource errors reported from HBA, reduce
* queuedepths of the SCSI device.
*/
if ((irsp->ulpStatus == IOSTAT_LOCAL_REJECT) &&
((irsp->un.ulpWord[4] & IOERR_PARAM_MASK) ==
IOERR_NO_RESOURCES)) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
phba->lpfc_rampdown_queue_depth(phba);
spin_lock_irqsave(&phba->hbalock, iflag);
}
if (irsp->ulpStatus) {
/* Rsp ring <ringno> error: IOCB */
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0328 Rsp Ring %d error: "
"IOCB Data: "
"x%x x%x x%x x%x "
"x%x x%x x%x x%x "
"x%x x%x x%x x%x "
"x%x x%x x%x x%x\n",
pring->ringno,
irsp->un.ulpWord[0],
irsp->un.ulpWord[1],
irsp->un.ulpWord[2],
irsp->un.ulpWord[3],
irsp->un.ulpWord[4],
irsp->un.ulpWord[5],
*(((uint32_t *) irsp) + 6),
*(((uint32_t *) irsp) + 7),
*(((uint32_t *) irsp) + 8),
*(((uint32_t *) irsp) + 9),
*(((uint32_t *) irsp) + 10),
*(((uint32_t *) irsp) + 11),
*(((uint32_t *) irsp) + 12),
*(((uint32_t *) irsp) + 13),
*(((uint32_t *) irsp) + 14),
*(((uint32_t *) irsp) + 15));
}
/*
* Fetch the IOCB command type and call the correct completion
* routine. Solicited and Unsolicited IOCBs on the ELS ring
* get freed back to the lpfc_iocb_list by the discovery
* kernel thread.
*/
iocb_cmd_type = irsp->ulpCommand & CMD_IOCB_MASK;
type = lpfc_sli_iocb_cmd_type(iocb_cmd_type);
switch (type) {
case LPFC_SOL_IOCB:
spin_unlock_irqrestore(&phba->hbalock, iflag);
rc = lpfc_sli_process_sol_iocb(phba, pring, saveq);
spin_lock_irqsave(&phba->hbalock, iflag);
break;
case LPFC_UNSOL_IOCB:
spin_unlock_irqrestore(&phba->hbalock, iflag);
rc = lpfc_sli_process_unsol_iocb(phba, pring, saveq);
spin_lock_irqsave(&phba->hbalock, iflag);
if (!rc)
free_saveq = 0;
break;
case LPFC_ABORT_IOCB:
cmdiocbp = NULL;
if (irsp->ulpCommand != CMD_XRI_ABORTED_CX)
cmdiocbp = lpfc_sli_iocbq_lookup(phba, pring,
saveq);
if (cmdiocbp) {
/* Call the specified completion routine */
if (cmdiocbp->iocb_cmpl) {
spin_unlock_irqrestore(&phba->hbalock,
iflag);
(cmdiocbp->iocb_cmpl)(phba, cmdiocbp,
saveq);
spin_lock_irqsave(&phba->hbalock,
iflag);
} else
__lpfc_sli_release_iocbq(phba,
cmdiocbp);
}
break;
case LPFC_UNKNOWN_IOCB:
if (irsp->ulpCommand == CMD_ADAPTER_MSG) {
char adaptermsg[LPFC_MAX_ADPTMSG];
memset(adaptermsg, 0, LPFC_MAX_ADPTMSG);
memcpy(&adaptermsg[0], (uint8_t *)irsp,
MAX_MSG_DATA);
dev_warn(&((phba->pcidev)->dev),
"lpfc%d: %s\n",
phba->brd_no, adaptermsg);
} else {
/* Unknown IOCB command */
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0335 Unknown IOCB "
"command Data: x%x "
"x%x x%x x%x\n",
irsp->ulpCommand,
irsp->ulpStatus,
irsp->ulpIoTag,
irsp->ulpContext);
}
break;
}
if (free_saveq) {
list_for_each_entry_safe(rspiocbp, next_iocb,
&saveq->list, list) {
list_del_init(&rspiocbp->list);
__lpfc_sli_release_iocbq(phba, rspiocbp);
}
__lpfc_sli_release_iocbq(phba, saveq);
}
rspiocbp = NULL;
}
spin_unlock_irqrestore(&phba->hbalock, iflag);
return rspiocbp;
}
/**
* lpfc_sli_handle_slow_ring_event - Wrapper func for handling slow-path iocbs
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @mask: Host attention register mask for this ring.
*
* This routine wraps the actual slow_ring event process routine from the
* API jump table function pointer from the lpfc_hba struct.
**/
void
lpfc_sli_handle_slow_ring_event(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring, uint32_t mask)
{
phba->lpfc_sli_handle_slow_ring_event(phba, pring, mask);
}
/**
* lpfc_sli_handle_slow_ring_event_s3 - Handle SLI3 ring event for non-FCP rings
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @mask: Host attention register mask for this ring.
*
* This function is called from the worker thread when there is a ring event
* for non-fcp rings. The caller does not hold any lock. The function will
* remove each response iocb in the response ring and calls the handle
* response iocb routine (lpfc_sli_sp_handle_rspiocb) to process it.
**/
static void
lpfc_sli_handle_slow_ring_event_s3(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring, uint32_t mask)
{
struct lpfc_pgp *pgp;
IOCB_t *entry;
IOCB_t *irsp = NULL;
struct lpfc_iocbq *rspiocbp = NULL;
uint32_t portRspPut, portRspMax;
unsigned long iflag;
uint32_t status;
pgp = &phba->port_gp[pring->ringno];
spin_lock_irqsave(&phba->hbalock, iflag);
pring->stats.iocb_event++;
/*
* The next available response entry should never exceed the maximum
* entries. If it does, treat it as an adapter hardware error.
*/
portRspMax = pring->sli.sli3.numRiocb;
portRspPut = le32_to_cpu(pgp->rspPutInx);
if (portRspPut >= portRspMax) {
/*
* Ring <ringno> handler: portRspPut <portRspPut> is bigger than
* rsp ring <portRspMax>
*/
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0303 Ring %d handler: portRspPut %d "
"is bigger than rsp ring %d\n",
pring->ringno, portRspPut, portRspMax);
phba->link_state = LPFC_HBA_ERROR;
spin_unlock_irqrestore(&phba->hbalock, iflag);
phba->work_hs = HS_FFER3;
lpfc_handle_eratt(phba);
return;
}
rmb();
while (pring->sli.sli3.rspidx != portRspPut) {
/*
* Build a completion list and call the appropriate handler.
* The process is to get the next available response iocb, get
* a free iocb from the list, copy the response data into the
* free iocb, insert to the continuation list, and update the
* next response index to slim. This process makes response
* iocb's in the ring available to DMA as fast as possible but
* pays a penalty for a copy operation. Since the iocb is
* only 32 bytes, this penalty is considered small relative to
* the PCI reads for register values and a slim write. When
* the ulpLe field is set, the entire Command has been
* received.
*/
entry = lpfc_resp_iocb(phba, pring);
phba->last_completion_time = jiffies;
rspiocbp = __lpfc_sli_get_iocbq(phba);
if (rspiocbp == NULL) {
printk(KERN_ERR "%s: out of buffers! Failing "
"completion.\n", __func__);
break;
}
lpfc_sli_pcimem_bcopy(entry, &rspiocbp->iocb,
phba->iocb_rsp_size);
irsp = &rspiocbp->iocb;
if (++pring->sli.sli3.rspidx >= portRspMax)
pring->sli.sli3.rspidx = 0;
if (pring->ringno == LPFC_ELS_RING) {
lpfc_debugfs_slow_ring_trc(phba,
"IOCB rsp ring: wd4:x%08x wd6:x%08x wd7:x%08x",
*(((uint32_t *) irsp) + 4),
*(((uint32_t *) irsp) + 6),
*(((uint32_t *) irsp) + 7));
}
writel(pring->sli.sli3.rspidx,
&phba->host_gp[pring->ringno].rspGetInx);
spin_unlock_irqrestore(&phba->hbalock, iflag);
/* Handle the response IOCB */
rspiocbp = lpfc_sli_sp_handle_rspiocb(phba, pring, rspiocbp);
spin_lock_irqsave(&phba->hbalock, iflag);
/*
* If the port response put pointer has not been updated, sync
* the pgp->rspPutInx in the MAILBOX_tand fetch the new port
* response put pointer.
*/
if (pring->sli.sli3.rspidx == portRspPut) {
portRspPut = le32_to_cpu(pgp->rspPutInx);
}
} /* while (pring->sli.sli3.rspidx != portRspPut) */
if ((rspiocbp != NULL) && (mask & HA_R0RE_REQ)) {
/* At least one response entry has been freed */
pring->stats.iocb_rsp_full++;
/* SET RxRE_RSP in Chip Att register */
status = ((CA_R0ATT | CA_R0RE_RSP) << (pring->ringno * 4));
writel(status, phba->CAregaddr);
readl(phba->CAregaddr); /* flush */
}
if ((mask & HA_R0CE_RSP) && (pring->flag & LPFC_CALL_RING_AVAILABLE)) {
pring->flag &= ~LPFC_CALL_RING_AVAILABLE;
pring->stats.iocb_cmd_empty++;
/* Force update of the local copy of cmdGetInx */
pring->sli.sli3.local_getidx = le32_to_cpu(pgp->cmdGetInx);
lpfc_sli_resume_iocb(phba, pring);
if ((pring->lpfc_sli_cmd_available))
(pring->lpfc_sli_cmd_available) (phba, pring);
}
spin_unlock_irqrestore(&phba->hbalock, iflag);
return;
}
/**
* lpfc_sli_handle_slow_ring_event_s4 - Handle SLI4 slow-path els events
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @mask: Host attention register mask for this ring.
*
* This function is called from the worker thread when there is a pending
* ELS response iocb on the driver internal slow-path response iocb worker
* queue. The caller does not hold any lock. The function will remove each
* response iocb from the response worker queue and calls the handle
* response iocb routine (lpfc_sli_sp_handle_rspiocb) to process it.
**/
static void
lpfc_sli_handle_slow_ring_event_s4(struct lpfc_hba *phba,
struct lpfc_sli_ring *pring, uint32_t mask)
{
struct lpfc_iocbq *irspiocbq;
struct hbq_dmabuf *dmabuf;
struct lpfc_cq_event *cq_event;
unsigned long iflag;
spin_lock_irqsave(&phba->hbalock, iflag);
phba->hba_flag &= ~HBA_SP_QUEUE_EVT;
spin_unlock_irqrestore(&phba->hbalock, iflag);
while (!list_empty(&phba->sli4_hba.sp_queue_event)) {
/* Get the response iocb from the head of work queue */
spin_lock_irqsave(&phba->hbalock, iflag);
list_remove_head(&phba->sli4_hba.sp_queue_event,
cq_event, struct lpfc_cq_event, list);
spin_unlock_irqrestore(&phba->hbalock, iflag);
switch (bf_get(lpfc_wcqe_c_code, &cq_event->cqe.wcqe_cmpl)) {
case CQE_CODE_COMPL_WQE:
irspiocbq = container_of(cq_event, struct lpfc_iocbq,
cq_event);
/* Translate ELS WCQE to response IOCBQ */
irspiocbq = lpfc_sli4_els_wcqe_to_rspiocbq(phba,
irspiocbq);
if (irspiocbq)
lpfc_sli_sp_handle_rspiocb(phba, pring,
irspiocbq);
break;
case CQE_CODE_RECEIVE:
case CQE_CODE_RECEIVE_V1:
dmabuf = container_of(cq_event, struct hbq_dmabuf,
cq_event);
lpfc_sli4_handle_received_buffer(phba, dmabuf);
break;
default:
break;
}
}
}
/**
* lpfc_sli_abort_iocb_ring - Abort all iocbs in the ring
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function aborts all iocbs in the given ring and frees all the iocb
* objects in txq. This function issues an abort iocb for all the iocb commands
* in txcmplq. The iocbs in the txcmplq is not guaranteed to complete before
* the return of this function. The caller is not required to hold any locks.
**/
void
lpfc_sli_abort_iocb_ring(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
LIST_HEAD(completions);
struct lpfc_iocbq *iocb, *next_iocb;
if (pring->ringno == LPFC_ELS_RING) {
lpfc_fabric_abort_hba(phba);
}
/* Error everything on txq and txcmplq
* First do the txq.
*/
if (phba->sli_rev >= LPFC_SLI_REV4) {
spin_lock_irq(&pring->ring_lock);
list_splice_init(&pring->txq, &completions);
pring->txq_cnt = 0;
spin_unlock_irq(&pring->ring_lock);
spin_lock_irq(&phba->hbalock);
/* Next issue ABTS for everything on the txcmplq */
list_for_each_entry_safe(iocb, next_iocb, &pring->txcmplq, list)
lpfc_sli_issue_abort_iotag(phba, pring, iocb);
spin_unlock_irq(&phba->hbalock);
} else {
spin_lock_irq(&phba->hbalock);
list_splice_init(&pring->txq, &completions);
pring->txq_cnt = 0;
/* Next issue ABTS for everything on the txcmplq */
list_for_each_entry_safe(iocb, next_iocb, &pring->txcmplq, list)
lpfc_sli_issue_abort_iotag(phba, pring, iocb);
spin_unlock_irq(&phba->hbalock);
}
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/* Cancel all the IOCBs from the completions list */
lpfc_sli_cancel_iocbs(phba, &completions, IOSTAT_LOCAL_REJECT,
IOERR_SLI_ABORTED);
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/**
* lpfc_sli_abort_wqe_ring - Abort all iocbs in the ring
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function aborts all iocbs in the given ring and frees all the iocb
* objects in txq. This function issues an abort iocb for all the iocb commands
* in txcmplq. The iocbs in the txcmplq is not guaranteed to complete before
* the return of this function. The caller is not required to hold any locks.
**/
void
lpfc_sli_abort_wqe_ring(struct lpfc_hba *phba, struct lpfc_sli_ring *pring)
{
LIST_HEAD(completions);
struct lpfc_iocbq *iocb, *next_iocb;
if (pring->ringno == LPFC_ELS_RING)
lpfc_fabric_abort_hba(phba);
spin_lock_irq(&phba->hbalock);
/* Next issue ABTS for everything on the txcmplq */
list_for_each_entry_safe(iocb, next_iocb, &pring->txcmplq, list)
lpfc_sli4_abort_nvme_io(phba, pring, iocb);
spin_unlock_irq(&phba->hbalock);
}
/**
* lpfc_sli_abort_fcp_rings - Abort all iocbs in all FCP rings
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
*
* This function aborts all iocbs in FCP rings and frees all the iocb
* objects in txq. This function issues an abort iocb for all the iocb commands
* in txcmplq. The iocbs in the txcmplq is not guaranteed to complete before
* the return of this function. The caller is not required to hold any locks.
**/
void
lpfc_sli_abort_fcp_rings(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
struct lpfc_sli_ring *pring;
uint32_t i;
/* Look on all the FCP Rings for the iotag */
if (phba->sli_rev >= LPFC_SLI_REV4) {
for (i = 0; i < phba->cfg_fcp_io_channel; i++) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = phba->sli4_hba.fcp_wq[i]->pring;
lpfc_sli_abort_iocb_ring(phba, pring);
}
} else {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &psli->sli3_ring[LPFC_FCP_RING];
lpfc_sli_abort_iocb_ring(phba, pring);
}
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/**
* lpfc_sli_abort_nvme_rings - Abort all wqes in all NVME rings
* @phba: Pointer to HBA context object.
*
* This function aborts all wqes in NVME rings. This function issues an
* abort wqe for all the outstanding IO commands in txcmplq. The iocbs in
* the txcmplq is not guaranteed to complete before the return of this
* function. The caller is not required to hold any locks.
**/
void
lpfc_sli_abort_nvme_rings(struct lpfc_hba *phba)
{
struct lpfc_sli_ring *pring;
uint32_t i;
if (phba->sli_rev < LPFC_SLI_REV4)
return;
/* Abort all IO on each NVME ring. */
for (i = 0; i < phba->cfg_nvme_io_channel; i++) {
pring = phba->sli4_hba.nvme_wq[i]->pring;
lpfc_sli_abort_wqe_ring(phba, pring);
}
}
/**
* lpfc_sli_flush_fcp_rings - flush all iocbs in the fcp ring
* @phba: Pointer to HBA context object.
*
* This function flushes all iocbs in the fcp ring and frees all the iocb
* objects in txq and txcmplq. This function will not issue abort iocbs
* for all the iocb commands in txcmplq, they will just be returned with
* IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
* slot has been permanently disabled.
**/
void
lpfc_sli_flush_fcp_rings(struct lpfc_hba *phba)
{
LIST_HEAD(txq);
LIST_HEAD(txcmplq);
struct lpfc_sli *psli = &phba->sli;
struct lpfc_sli_ring *pring;
uint32_t i;
struct lpfc_iocbq *piocb, *next_iocb;
spin_lock_irq(&phba->hbalock);
/* Indicate the I/O queues are flushed */
phba->hba_flag |= HBA_FCP_IOQ_FLUSH;
spin_unlock_irq(&phba->hbalock);
/* Look on all the FCP Rings for the iotag */
if (phba->sli_rev >= LPFC_SLI_REV4) {
for (i = 0; i < phba->cfg_fcp_io_channel; i++) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = phba->sli4_hba.fcp_wq[i]->pring;
spin_lock_irq(&pring->ring_lock);
/* Retrieve everything on txq */
list_splice_init(&pring->txq, &txq);
list_for_each_entry_safe(piocb, next_iocb,
&pring->txcmplq, list)
piocb->iocb_flag &= ~LPFC_IO_ON_TXCMPLQ;
/* Retrieve everything on the txcmplq */
list_splice_init(&pring->txcmplq, &txcmplq);
pring->txq_cnt = 0;
pring->txcmplq_cnt = 0;
spin_unlock_irq(&pring->ring_lock);
/* Flush the txq */
lpfc_sli_cancel_iocbs(phba, &txq,
IOSTAT_LOCAL_REJECT,
IOERR_SLI_DOWN);
/* Flush the txcmpq */
lpfc_sli_cancel_iocbs(phba, &txcmplq,
IOSTAT_LOCAL_REJECT,
IOERR_SLI_DOWN);
}
} else {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &psli->sli3_ring[LPFC_FCP_RING];
spin_lock_irq(&phba->hbalock);
/* Retrieve everything on txq */
list_splice_init(&pring->txq, &txq);
list_for_each_entry_safe(piocb, next_iocb,
&pring->txcmplq, list)
piocb->iocb_flag &= ~LPFC_IO_ON_TXCMPLQ;
/* Retrieve everything on the txcmplq */
list_splice_init(&pring->txcmplq, &txcmplq);
pring->txq_cnt = 0;
pring->txcmplq_cnt = 0;
spin_unlock_irq(&phba->hbalock);
/* Flush the txq */
lpfc_sli_cancel_iocbs(phba, &txq, IOSTAT_LOCAL_REJECT,
IOERR_SLI_DOWN);
/* Flush the txcmpq */
lpfc_sli_cancel_iocbs(phba, &txcmplq, IOSTAT_LOCAL_REJECT,
IOERR_SLI_DOWN);
}
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/**
* lpfc_sli_flush_nvme_rings - flush all wqes in the nvme rings
* @phba: Pointer to HBA context object.
*
* This function flushes all wqes in the nvme rings and frees all resources
* in the txcmplq. This function does not issue abort wqes for the IO
* commands in txcmplq, they will just be returned with
* IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
* slot has been permanently disabled.
**/
void
lpfc_sli_flush_nvme_rings(struct lpfc_hba *phba)
{
LIST_HEAD(txcmplq);
struct lpfc_sli_ring *pring;
uint32_t i;
struct lpfc_iocbq *piocb, *next_iocb;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->sli_rev < LPFC_SLI_REV4)
return;
/* Hint to other driver operations that a flush is in progress. */
spin_lock_irq(&phba->hbalock);
phba->hba_flag |= HBA_NVME_IOQ_FLUSH;
spin_unlock_irq(&phba->hbalock);
/* Cycle through all NVME rings and complete each IO with
* a local driver reason code. This is a flush so no
* abort exchange to FW.
*/
for (i = 0; i < phba->cfg_nvme_io_channel; i++) {
pring = phba->sli4_hba.nvme_wq[i]->pring;
spin_lock_irq(&pring->ring_lock);
list_for_each_entry_safe(piocb, next_iocb,
&pring->txcmplq, list)
piocb->iocb_flag &= ~LPFC_IO_ON_TXCMPLQ;
/* Retrieve everything on the txcmplq */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
list_splice_init(&pring->txcmplq, &txcmplq);
pring->txcmplq_cnt = 0;
spin_unlock_irq(&pring->ring_lock);
/* Flush the txcmpq &&&PAE */
lpfc_sli_cancel_iocbs(phba, &txcmplq,
IOSTAT_LOCAL_REJECT,
IOERR_SLI_DOWN);
}
}
/**
* lpfc_sli_brdready_s3 - Check for sli3 host ready status
* @phba: Pointer to HBA context object.
* @mask: Bit mask to be checked.
*
* This function reads the host status register and compares
* with the provided bit mask to check if HBA completed
* the restart. This function will wait in a loop for the
* HBA to complete restart. If the HBA does not restart within
* 15 iterations, the function will reset the HBA again. The
* function returns 1 when HBA fail to restart otherwise returns
* zero.
**/
static int
lpfc_sli_brdready_s3(struct lpfc_hba *phba, uint32_t mask)
{
uint32_t status;
int i = 0;
int retval = 0;
/* Read the HBA Host Status Register */
if (lpfc_readl(phba->HSregaddr, &status))
return 1;
/*
* Check status register every 100ms for 5 retries, then every
* 500ms for 5, then every 2.5 sec for 5, then reset board and
* every 2.5 sec for 4.
* Break our of the loop if errors occurred during init.
*/
while (((status & mask) != mask) &&
!(status & HS_FFERM) &&
i++ < 20) {
if (i <= 5)
msleep(10);
else if (i <= 10)
msleep(500);
else
msleep(2500);
if (i == 15) {
/* Do post */
phba->pport->port_state = LPFC_VPORT_UNKNOWN;
lpfc_sli_brdrestart(phba);
}
/* Read the HBA Host Status Register */
if (lpfc_readl(phba->HSregaddr, &status)) {
retval = 1;
break;
}
}
/* Check to see if any errors occurred during init */
if ((status & HS_FFERM) || (i >= 20)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2751 Adapter failed to restart, "
"status reg x%x, FW Data: A8 x%x AC x%x\n",
status,
readl(phba->MBslimaddr + 0xa8),
readl(phba->MBslimaddr + 0xac));
phba->link_state = LPFC_HBA_ERROR;
retval = 1;
}
return retval;
}
/**
* lpfc_sli_brdready_s4 - Check for sli4 host ready status
* @phba: Pointer to HBA context object.
* @mask: Bit mask to be checked.
*
* This function checks the host status register to check if HBA is
* ready. This function will wait in a loop for the HBA to be ready
* If the HBA is not ready , the function will will reset the HBA PCI
* function again. The function returns 1 when HBA fail to be ready
* otherwise returns zero.
**/
static int
lpfc_sli_brdready_s4(struct lpfc_hba *phba, uint32_t mask)
{
uint32_t status;
int retval = 0;
/* Read the HBA Host Status Register */
status = lpfc_sli4_post_status_check(phba);
if (status) {
phba->pport->port_state = LPFC_VPORT_UNKNOWN;
lpfc_sli_brdrestart(phba);
status = lpfc_sli4_post_status_check(phba);
}
/* Check to see if any errors occurred during init */
if (status) {
phba->link_state = LPFC_HBA_ERROR;
retval = 1;
} else
phba->sli4_hba.intr_enable = 0;
return retval;
}
/**
* lpfc_sli_brdready - Wrapper func for checking the hba readyness
* @phba: Pointer to HBA context object.
* @mask: Bit mask to be checked.
*
* This routine wraps the actual SLI3 or SLI4 hba readyness check routine
* from the API jump table function pointer from the lpfc_hba struct.
**/
int
lpfc_sli_brdready(struct lpfc_hba *phba, uint32_t mask)
{
return phba->lpfc_sli_brdready(phba, mask);
}
#define BARRIER_TEST_PATTERN (0xdeadbeef)
/**
* lpfc_reset_barrier - Make HBA ready for HBA reset
* @phba: Pointer to HBA context object.
*
* This function is called before resetting an HBA. This function is called
* with hbalock held and requests HBA to quiesce DMAs before a reset.
**/
void lpfc_reset_barrier(struct lpfc_hba *phba)
{
uint32_t __iomem *resp_buf;
uint32_t __iomem *mbox_buf;
volatile uint32_t mbox;
uint32_t hc_copy, ha_copy, resp_data;
int i;
uint8_t hdrtype;
lockdep_assert_held(&phba->hbalock);
pci_read_config_byte(phba->pcidev, PCI_HEADER_TYPE, &hdrtype);
if (hdrtype != 0x80 ||
(FC_JEDEC_ID(phba->vpd.rev.biuRev) != HELIOS_JEDEC_ID &&
FC_JEDEC_ID(phba->vpd.rev.biuRev) != THOR_JEDEC_ID))
return;
/*
* Tell the other part of the chip to suspend temporarily all
* its DMA activity.
*/
resp_buf = phba->MBslimaddr;
/* Disable the error attention */
if (lpfc_readl(phba->HCregaddr, &hc_copy))
return;
writel((hc_copy & ~HC_ERINT_ENA), phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
phba->link_flag |= LS_IGNORE_ERATT;
if (lpfc_readl(phba->HAregaddr, &ha_copy))
return;
if (ha_copy & HA_ERATT) {
/* Clear Chip error bit */
writel(HA_ERATT, phba->HAregaddr);
phba->pport->stopped = 1;
}
mbox = 0;
((MAILBOX_t *)&mbox)->mbxCommand = MBX_KILL_BOARD;
((MAILBOX_t *)&mbox)->mbxOwner = OWN_CHIP;
writel(BARRIER_TEST_PATTERN, (resp_buf + 1));
mbox_buf = phba->MBslimaddr;
writel(mbox, mbox_buf);
for (i = 0; i < 50; i++) {
if (lpfc_readl((resp_buf + 1), &resp_data))
return;
if (resp_data != ~(BARRIER_TEST_PATTERN))
mdelay(1);
else
break;
}
resp_data = 0;
if (lpfc_readl((resp_buf + 1), &resp_data))
return;
if (resp_data != ~(BARRIER_TEST_PATTERN)) {
if (phba->sli.sli_flag & LPFC_SLI_ACTIVE ||
phba->pport->stopped)
goto restore_hc;
else
goto clear_errat;
}
((MAILBOX_t *)&mbox)->mbxOwner = OWN_HOST;
resp_data = 0;
for (i = 0; i < 500; i++) {
if (lpfc_readl(resp_buf, &resp_data))
return;
if (resp_data != mbox)
mdelay(1);
else
break;
}
clear_errat:
while (++i < 500) {
if (lpfc_readl(phba->HAregaddr, &ha_copy))
return;
if (!(ha_copy & HA_ERATT))
mdelay(1);
else
break;
}
if (readl(phba->HAregaddr) & HA_ERATT) {
writel(HA_ERATT, phba->HAregaddr);
phba->pport->stopped = 1;
}
restore_hc:
phba->link_flag &= ~LS_IGNORE_ERATT;
writel(hc_copy, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
}
/**
* lpfc_sli_brdkill - Issue a kill_board mailbox command
* @phba: Pointer to HBA context object.
*
* This function issues a kill_board mailbox command and waits for
* the error attention interrupt. This function is called for stopping
* the firmware processing. The caller is not required to hold any
* locks. This function calls lpfc_hba_down_post function to free
* any pending commands after the kill. The function will return 1 when it
* fails to kill the board else will return 0.
**/
int
lpfc_sli_brdkill(struct lpfc_hba *phba)
{
struct lpfc_sli *psli;
LPFC_MBOXQ_t *pmb;
uint32_t status;
uint32_t ha_copy;
int retval;
int i = 0;
psli = &phba->sli;
/* Kill HBA */
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0329 Kill HBA Data: x%x x%x\n",
phba->pport->port_state, psli->sli_flag);
pmb = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!pmb)
return 1;
/* Disable the error attention */
spin_lock_irq(&phba->hbalock);
if (lpfc_readl(phba->HCregaddr, &status)) {
spin_unlock_irq(&phba->hbalock);
mempool_free(pmb, phba->mbox_mem_pool);
return 1;
}
status &= ~HC_ERINT_ENA;
writel(status, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
phba->link_flag |= LS_IGNORE_ERATT;
spin_unlock_irq(&phba->hbalock);
lpfc_kill_board(phba, pmb);
pmb->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
retval = lpfc_sli_issue_mbox(phba, pmb, MBX_NOWAIT);
if (retval != MBX_SUCCESS) {
if (retval != MBX_BUSY)
mempool_free(pmb, phba->mbox_mem_pool);
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2752 KILL_BOARD command failed retval %d\n",
retval);
spin_lock_irq(&phba->hbalock);
phba->link_flag &= ~LS_IGNORE_ERATT;
spin_unlock_irq(&phba->hbalock);
return 1;
}
spin_lock_irq(&phba->hbalock);
psli->sli_flag &= ~LPFC_SLI_ACTIVE;
spin_unlock_irq(&phba->hbalock);
mempool_free(pmb, phba->mbox_mem_pool);
/* There is no completion for a KILL_BOARD mbox cmd. Check for an error
* attention every 100ms for 3 seconds. If we don't get ERATT after
* 3 seconds we still set HBA_ERROR state because the status of the
* board is now undefined.
*/
if (lpfc_readl(phba->HAregaddr, &ha_copy))
return 1;
while ((i++ < 30) && !(ha_copy & HA_ERATT)) {
mdelay(100);
if (lpfc_readl(phba->HAregaddr, &ha_copy))
return 1;
}
del_timer_sync(&psli->mbox_tmo);
if (ha_copy & HA_ERATT) {
writel(HA_ERATT, phba->HAregaddr);
phba->pport->stopped = 1;
}
spin_lock_irq(&phba->hbalock);
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
psli->mbox_active = NULL;
phba->link_flag &= ~LS_IGNORE_ERATT;
spin_unlock_irq(&phba->hbalock);
lpfc_hba_down_post(phba);
phba->link_state = LPFC_HBA_ERROR;
return ha_copy & HA_ERATT ? 0 : 1;
}
/**
* lpfc_sli_brdreset - Reset a sli-2 or sli-3 HBA
* @phba: Pointer to HBA context object.
*
* This function resets the HBA by writing HC_INITFF to the control
* register. After the HBA resets, this function resets all the iocb ring
* indices. This function disables PCI layer parity checking during
* the reset.
* This function returns 0 always.
* The caller is not required to hold any locks.
**/
int
lpfc_sli_brdreset(struct lpfc_hba *phba)
{
struct lpfc_sli *psli;
struct lpfc_sli_ring *pring;
uint16_t cfg_value;
int i;
psli = &phba->sli;
/* Reset HBA */
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0325 Reset HBA Data: x%x x%x\n",
scsi: lpfc: Fix panic on BFS configuration To select the appropriate shost template, the driver is issuing a mailbox command to retrieve the wwn. Turns out the sending of the command precedes the reset of the function. On SLI-4 adapters, this is inconsequential as the mailbox command location is specified by dma via the BMBX register. However, on SLI-3 adapters, the location of the mailbox command submission area changes. When the function is first powered on or reset, the cmd is submitted via PCI bar memory. Later the driver changes the function config to use host memory and DMA. The request to start a mailbox command is the same, a simple doorbell write, regardless of submission area. So.. if there has not been a boot driver run against the adapter, the mailbox command works as defaults are ok. But, if the boot driver has configured the card and, and if no platform pci function/slot reset occurs as the os starts, the mailbox command will fail. The SLI-3 device will use the stale boot driver dma location. This can cause PCI eeh errors. Fix is to reset the sli-3 function before sending the mailbox command, thus synchronizing the function/driver on mailbox location. Note: The fix uses routines that are typically invoked later in the call flow to reset the sli-3 device. The issue in using those routines is that the normal (non-fix) flow does additional initialization, namely the allocation of the pport structure. So, rather than significantly reworking the initialization flow so that the pport is alloc'd first, pointer checks are added to work around it. Checks are limited to the routines invoked by a sli-3 adapter (s3 routines) as this fix/early call is only invoked on a sli3 adapter. Nothing changes post the fix. Subsequent initialization, and another adapter reset, still occur - both on sli-3 and sli-4 adapters. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Fixes: 96418b5e2c88 ("scsi: lpfc: Fix eh_deadline setting for sli3 adapters.") Cc: stable@vger.kernel.org # v4.11+ Reviewed-by: Ewan D. Milne <emilne@redhat.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-04-27 16:08:26 -06:00
(phba->pport) ? phba->pport->port_state : 0,
psli->sli_flag);
/* perform board reset */
phba->fc_eventTag = 0;
phba->link_events = 0;
scsi: lpfc: Fix panic on BFS configuration To select the appropriate shost template, the driver is issuing a mailbox command to retrieve the wwn. Turns out the sending of the command precedes the reset of the function. On SLI-4 adapters, this is inconsequential as the mailbox command location is specified by dma via the BMBX register. However, on SLI-3 adapters, the location of the mailbox command submission area changes. When the function is first powered on or reset, the cmd is submitted via PCI bar memory. Later the driver changes the function config to use host memory and DMA. The request to start a mailbox command is the same, a simple doorbell write, regardless of submission area. So.. if there has not been a boot driver run against the adapter, the mailbox command works as defaults are ok. But, if the boot driver has configured the card and, and if no platform pci function/slot reset occurs as the os starts, the mailbox command will fail. The SLI-3 device will use the stale boot driver dma location. This can cause PCI eeh errors. Fix is to reset the sli-3 function before sending the mailbox command, thus synchronizing the function/driver on mailbox location. Note: The fix uses routines that are typically invoked later in the call flow to reset the sli-3 device. The issue in using those routines is that the normal (non-fix) flow does additional initialization, namely the allocation of the pport structure. So, rather than significantly reworking the initialization flow so that the pport is alloc'd first, pointer checks are added to work around it. Checks are limited to the routines invoked by a sli-3 adapter (s3 routines) as this fix/early call is only invoked on a sli3 adapter. Nothing changes post the fix. Subsequent initialization, and another adapter reset, still occur - both on sli-3 and sli-4 adapters. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Fixes: 96418b5e2c88 ("scsi: lpfc: Fix eh_deadline setting for sli3 adapters.") Cc: stable@vger.kernel.org # v4.11+ Reviewed-by: Ewan D. Milne <emilne@redhat.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-04-27 16:08:26 -06:00
if (phba->pport) {
phba->pport->fc_myDID = 0;
phba->pport->fc_prevDID = 0;
}
/* Turn off parity checking and serr during the physical reset */
pci_read_config_word(phba->pcidev, PCI_COMMAND, &cfg_value);
pci_write_config_word(phba->pcidev, PCI_COMMAND,
(cfg_value &
~(PCI_COMMAND_PARITY | PCI_COMMAND_SERR)));
psli->sli_flag &= ~(LPFC_SLI_ACTIVE | LPFC_PROCESS_LA);
/* Now toggle INITFF bit in the Host Control Register */
writel(HC_INITFF, phba->HCregaddr);
mdelay(1);
readl(phba->HCregaddr); /* flush */
writel(0, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
/* Restore PCI cmd register */
pci_write_config_word(phba->pcidev, PCI_COMMAND, cfg_value);
/* Initialize relevant SLI info */
for (i = 0; i < psli->num_rings; i++) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &psli->sli3_ring[i];
pring->flag = 0;
pring->sli.sli3.rspidx = 0;
pring->sli.sli3.next_cmdidx = 0;
pring->sli.sli3.local_getidx = 0;
pring->sli.sli3.cmdidx = 0;
pring->missbufcnt = 0;
}
phba->link_state = LPFC_WARM_START;
return 0;
}
/**
* lpfc_sli4_brdreset - Reset a sli-4 HBA
* @phba: Pointer to HBA context object.
*
* This function resets a SLI4 HBA. This function disables PCI layer parity
* checking during resets the device. The caller is not required to hold
* any locks.
*
* This function returns 0 always.
**/
int
lpfc_sli4_brdreset(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
uint16_t cfg_value;
int rc = 0;
/* Reset HBA */
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0295 Reset HBA Data: x%x x%x x%x\n",
phba->pport->port_state, psli->sli_flag,
phba->hba_flag);
/* perform board reset */
phba->fc_eventTag = 0;
phba->link_events = 0;
phba->pport->fc_myDID = 0;
phba->pport->fc_prevDID = 0;
spin_lock_irq(&phba->hbalock);
psli->sli_flag &= ~(LPFC_PROCESS_LA);
phba->fcf.fcf_flag = 0;
spin_unlock_irq(&phba->hbalock);
/* SLI4 INTF 2: if FW dump is being taken skip INIT_PORT */
if (phba->hba_flag & HBA_FW_DUMP_OP) {
phba->hba_flag &= ~HBA_FW_DUMP_OP;
return rc;
}
/* Now physically reset the device */
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"0389 Performing PCI function reset!\n");
/* Turn off parity checking and serr during the physical reset */
pci_read_config_word(phba->pcidev, PCI_COMMAND, &cfg_value);
pci_write_config_word(phba->pcidev, PCI_COMMAND, (cfg_value &
~(PCI_COMMAND_PARITY | PCI_COMMAND_SERR)));
/* Perform FCoE PCI function reset before freeing queue memory */
rc = lpfc_pci_function_reset(phba);
/* Restore PCI cmd register */
pci_write_config_word(phba->pcidev, PCI_COMMAND, cfg_value);
return rc;
}
/**
* lpfc_sli_brdrestart_s3 - Restart a sli-3 hba
* @phba: Pointer to HBA context object.
*
* This function is called in the SLI initialization code path to
* restart the HBA. The caller is not required to hold any lock.
* This function writes MBX_RESTART mailbox command to the SLIM and
* resets the HBA. At the end of the function, it calls lpfc_hba_down_post
* function to free any pending commands. The function enables
* POST only during the first initialization. The function returns zero.
* The function does not guarantee completion of MBX_RESTART mailbox
* command before the return of this function.
**/
static int
lpfc_sli_brdrestart_s3(struct lpfc_hba *phba)
{
MAILBOX_t *mb;
struct lpfc_sli *psli;
volatile uint32_t word0;
void __iomem *to_slim;
uint32_t hba_aer_enabled;
spin_lock_irq(&phba->hbalock);
/* Take PCIe device Advanced Error Reporting (AER) state */
hba_aer_enabled = phba->hba_flag & HBA_AER_ENABLED;
psli = &phba->sli;
/* Restart HBA */
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0337 Restart HBA Data: x%x x%x\n",
scsi: lpfc: Fix panic on BFS configuration To select the appropriate shost template, the driver is issuing a mailbox command to retrieve the wwn. Turns out the sending of the command precedes the reset of the function. On SLI-4 adapters, this is inconsequential as the mailbox command location is specified by dma via the BMBX register. However, on SLI-3 adapters, the location of the mailbox command submission area changes. When the function is first powered on or reset, the cmd is submitted via PCI bar memory. Later the driver changes the function config to use host memory and DMA. The request to start a mailbox command is the same, a simple doorbell write, regardless of submission area. So.. if there has not been a boot driver run against the adapter, the mailbox command works as defaults are ok. But, if the boot driver has configured the card and, and if no platform pci function/slot reset occurs as the os starts, the mailbox command will fail. The SLI-3 device will use the stale boot driver dma location. This can cause PCI eeh errors. Fix is to reset the sli-3 function before sending the mailbox command, thus synchronizing the function/driver on mailbox location. Note: The fix uses routines that are typically invoked later in the call flow to reset the sli-3 device. The issue in using those routines is that the normal (non-fix) flow does additional initialization, namely the allocation of the pport structure. So, rather than significantly reworking the initialization flow so that the pport is alloc'd first, pointer checks are added to work around it. Checks are limited to the routines invoked by a sli-3 adapter (s3 routines) as this fix/early call is only invoked on a sli3 adapter. Nothing changes post the fix. Subsequent initialization, and another adapter reset, still occur - both on sli-3 and sli-4 adapters. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Fixes: 96418b5e2c88 ("scsi: lpfc: Fix eh_deadline setting for sli3 adapters.") Cc: stable@vger.kernel.org # v4.11+ Reviewed-by: Ewan D. Milne <emilne@redhat.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-04-27 16:08:26 -06:00
(phba->pport) ? phba->pport->port_state : 0,
psli->sli_flag);
word0 = 0;
mb = (MAILBOX_t *) &word0;
mb->mbxCommand = MBX_RESTART;
mb->mbxHc = 1;
lpfc_reset_barrier(phba);
to_slim = phba->MBslimaddr;
writel(*(uint32_t *) mb, to_slim);
readl(to_slim); /* flush */
/* Only skip post after fc_ffinit is completed */
scsi: lpfc: Fix panic on BFS configuration To select the appropriate shost template, the driver is issuing a mailbox command to retrieve the wwn. Turns out the sending of the command precedes the reset of the function. On SLI-4 adapters, this is inconsequential as the mailbox command location is specified by dma via the BMBX register. However, on SLI-3 adapters, the location of the mailbox command submission area changes. When the function is first powered on or reset, the cmd is submitted via PCI bar memory. Later the driver changes the function config to use host memory and DMA. The request to start a mailbox command is the same, a simple doorbell write, regardless of submission area. So.. if there has not been a boot driver run against the adapter, the mailbox command works as defaults are ok. But, if the boot driver has configured the card and, and if no platform pci function/slot reset occurs as the os starts, the mailbox command will fail. The SLI-3 device will use the stale boot driver dma location. This can cause PCI eeh errors. Fix is to reset the sli-3 function before sending the mailbox command, thus synchronizing the function/driver on mailbox location. Note: The fix uses routines that are typically invoked later in the call flow to reset the sli-3 device. The issue in using those routines is that the normal (non-fix) flow does additional initialization, namely the allocation of the pport structure. So, rather than significantly reworking the initialization flow so that the pport is alloc'd first, pointer checks are added to work around it. Checks are limited to the routines invoked by a sli-3 adapter (s3 routines) as this fix/early call is only invoked on a sli3 adapter. Nothing changes post the fix. Subsequent initialization, and another adapter reset, still occur - both on sli-3 and sli-4 adapters. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Fixes: 96418b5e2c88 ("scsi: lpfc: Fix eh_deadline setting for sli3 adapters.") Cc: stable@vger.kernel.org # v4.11+ Reviewed-by: Ewan D. Milne <emilne@redhat.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-04-27 16:08:26 -06:00
if (phba->pport && phba->pport->port_state)
word0 = 1; /* This is really setting up word1 */
else
word0 = 0; /* This is really setting up word1 */
to_slim = phba->MBslimaddr + sizeof (uint32_t);
writel(*(uint32_t *) mb, to_slim);
readl(to_slim); /* flush */
lpfc_sli_brdreset(phba);
scsi: lpfc: Fix panic on BFS configuration To select the appropriate shost template, the driver is issuing a mailbox command to retrieve the wwn. Turns out the sending of the command precedes the reset of the function. On SLI-4 adapters, this is inconsequential as the mailbox command location is specified by dma via the BMBX register. However, on SLI-3 adapters, the location of the mailbox command submission area changes. When the function is first powered on or reset, the cmd is submitted via PCI bar memory. Later the driver changes the function config to use host memory and DMA. The request to start a mailbox command is the same, a simple doorbell write, regardless of submission area. So.. if there has not been a boot driver run against the adapter, the mailbox command works as defaults are ok. But, if the boot driver has configured the card and, and if no platform pci function/slot reset occurs as the os starts, the mailbox command will fail. The SLI-3 device will use the stale boot driver dma location. This can cause PCI eeh errors. Fix is to reset the sli-3 function before sending the mailbox command, thus synchronizing the function/driver on mailbox location. Note: The fix uses routines that are typically invoked later in the call flow to reset the sli-3 device. The issue in using those routines is that the normal (non-fix) flow does additional initialization, namely the allocation of the pport structure. So, rather than significantly reworking the initialization flow so that the pport is alloc'd first, pointer checks are added to work around it. Checks are limited to the routines invoked by a sli-3 adapter (s3 routines) as this fix/early call is only invoked on a sli3 adapter. Nothing changes post the fix. Subsequent initialization, and another adapter reset, still occur - both on sli-3 and sli-4 adapters. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Fixes: 96418b5e2c88 ("scsi: lpfc: Fix eh_deadline setting for sli3 adapters.") Cc: stable@vger.kernel.org # v4.11+ Reviewed-by: Ewan D. Milne <emilne@redhat.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-04-27 16:08:26 -06:00
if (phba->pport)
phba->pport->stopped = 0;
phba->link_state = LPFC_INIT_START;
phba->hba_flag = 0;
spin_unlock_irq(&phba->hbalock);
memset(&psli->lnk_stat_offsets, 0, sizeof(psli->lnk_stat_offsets));
psli->stats_start = get_seconds();
/* Give the INITFF and Post time to settle. */
mdelay(100);
/* Reset HBA AER if it was enabled, note hba_flag was reset above */
if (hba_aer_enabled)
pci_disable_pcie_error_reporting(phba->pcidev);
lpfc_hba_down_post(phba);
return 0;
}
/**
* lpfc_sli_brdrestart_s4 - Restart the sli-4 hba
* @phba: Pointer to HBA context object.
*
* This function is called in the SLI initialization code path to restart
* a SLI4 HBA. The caller is not required to hold any lock.
* At the end of the function, it calls lpfc_hba_down_post function to
* free any pending commands.
**/
static int
lpfc_sli_brdrestart_s4(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
uint32_t hba_aer_enabled;
int rc;
/* Restart HBA */
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0296 Restart HBA Data: x%x x%x\n",
phba->pport->port_state, psli->sli_flag);
/* Take PCIe device Advanced Error Reporting (AER) state */
hba_aer_enabled = phba->hba_flag & HBA_AER_ENABLED;
rc = lpfc_sli4_brdreset(phba);
spin_lock_irq(&phba->hbalock);
phba->pport->stopped = 0;
phba->link_state = LPFC_INIT_START;
phba->hba_flag = 0;
spin_unlock_irq(&phba->hbalock);
memset(&psli->lnk_stat_offsets, 0, sizeof(psli->lnk_stat_offsets));
psli->stats_start = get_seconds();
/* Reset HBA AER if it was enabled, note hba_flag was reset above */
if (hba_aer_enabled)
pci_disable_pcie_error_reporting(phba->pcidev);
lpfc_hba_down_post(phba);
lpfc_sli4_queue_destroy(phba);
return rc;
}
/**
* lpfc_sli_brdrestart - Wrapper func for restarting hba
* @phba: Pointer to HBA context object.
*
* This routine wraps the actual SLI3 or SLI4 hba restart routine from the
* API jump table function pointer from the lpfc_hba struct.
**/
int
lpfc_sli_brdrestart(struct lpfc_hba *phba)
{
return phba->lpfc_sli_brdrestart(phba);
}
/**
* lpfc_sli_chipset_init - Wait for the restart of the HBA after a restart
* @phba: Pointer to HBA context object.
*
* This function is called after a HBA restart to wait for successful
* restart of the HBA. Successful restart of the HBA is indicated by
* HS_FFRDY and HS_MBRDY bits. If the HBA fails to restart even after 15
* iteration, the function will restart the HBA again. The function returns
* zero if HBA successfully restarted else returns negative error code.
**/
scsi: lpfc: Fix panic on BFS configuration To select the appropriate shost template, the driver is issuing a mailbox command to retrieve the wwn. Turns out the sending of the command precedes the reset of the function. On SLI-4 adapters, this is inconsequential as the mailbox command location is specified by dma via the BMBX register. However, on SLI-3 adapters, the location of the mailbox command submission area changes. When the function is first powered on or reset, the cmd is submitted via PCI bar memory. Later the driver changes the function config to use host memory and DMA. The request to start a mailbox command is the same, a simple doorbell write, regardless of submission area. So.. if there has not been a boot driver run against the adapter, the mailbox command works as defaults are ok. But, if the boot driver has configured the card and, and if no platform pci function/slot reset occurs as the os starts, the mailbox command will fail. The SLI-3 device will use the stale boot driver dma location. This can cause PCI eeh errors. Fix is to reset the sli-3 function before sending the mailbox command, thus synchronizing the function/driver on mailbox location. Note: The fix uses routines that are typically invoked later in the call flow to reset the sli-3 device. The issue in using those routines is that the normal (non-fix) flow does additional initialization, namely the allocation of the pport structure. So, rather than significantly reworking the initialization flow so that the pport is alloc'd first, pointer checks are added to work around it. Checks are limited to the routines invoked by a sli-3 adapter (s3 routines) as this fix/early call is only invoked on a sli3 adapter. Nothing changes post the fix. Subsequent initialization, and another adapter reset, still occur - both on sli-3 and sli-4 adapters. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Fixes: 96418b5e2c88 ("scsi: lpfc: Fix eh_deadline setting for sli3 adapters.") Cc: stable@vger.kernel.org # v4.11+ Reviewed-by: Ewan D. Milne <emilne@redhat.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-04-27 16:08:26 -06:00
int
lpfc_sli_chipset_init(struct lpfc_hba *phba)
{
uint32_t status, i = 0;
/* Read the HBA Host Status Register */
if (lpfc_readl(phba->HSregaddr, &status))
return -EIO;
/* Check status register to see what current state is */
i = 0;
while ((status & (HS_FFRDY | HS_MBRDY)) != (HS_FFRDY | HS_MBRDY)) {
/* Check every 10ms for 10 retries, then every 100ms for 90
* retries, then every 1 sec for 50 retires for a total of
* ~60 seconds before reset the board again and check every
* 1 sec for 50 retries. The up to 60 seconds before the
* board ready is required by the Falcon FIPS zeroization
* complete, and any reset the board in between shall cause
* restart of zeroization, further delay the board ready.
*/
if (i++ >= 200) {
/* Adapter failed to init, timeout, status reg
<status> */
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0436 Adapter failed to init, "
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
"timeout, status reg x%x, "
"FW Data: A8 x%x AC x%x\n", status,
readl(phba->MBslimaddr + 0xa8),
readl(phba->MBslimaddr + 0xac));
phba->link_state = LPFC_HBA_ERROR;
return -ETIMEDOUT;
}
/* Check to see if any errors occurred during init */
if (status & HS_FFERM) {
/* ERROR: During chipset initialization */
/* Adapter failed to init, chipset, status reg
<status> */
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0437 Adapter failed to init, "
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
"chipset, status reg x%x, "
"FW Data: A8 x%x AC x%x\n", status,
readl(phba->MBslimaddr + 0xa8),
readl(phba->MBslimaddr + 0xac));
phba->link_state = LPFC_HBA_ERROR;
return -EIO;
}
if (i <= 10)
msleep(10);
else if (i <= 100)
msleep(100);
else
msleep(1000);
if (i == 150) {
/* Do post */
phba->pport->port_state = LPFC_VPORT_UNKNOWN;
lpfc_sli_brdrestart(phba);
}
/* Read the HBA Host Status Register */
if (lpfc_readl(phba->HSregaddr, &status))
return -EIO;
}
/* Check to see if any errors occurred during init */
if (status & HS_FFERM) {
/* ERROR: During chipset initialization */
/* Adapter failed to init, chipset, status reg <status> */
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0438 Adapter failed to init, chipset, "
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
"status reg x%x, "
"FW Data: A8 x%x AC x%x\n", status,
readl(phba->MBslimaddr + 0xa8),
readl(phba->MBslimaddr + 0xac));
phba->link_state = LPFC_HBA_ERROR;
return -EIO;
}
/* Clear all interrupt enable conditions */
writel(0, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
/* setup host attn register */
writel(0xffffffff, phba->HAregaddr);
readl(phba->HAregaddr); /* flush */
return 0;
}
/**
* lpfc_sli_hbq_count - Get the number of HBQs to be configured
*
* This function calculates and returns the number of HBQs required to be
* configured.
**/
int
lpfc_sli_hbq_count(void)
{
return ARRAY_SIZE(lpfc_hbq_defs);
}
/**
* lpfc_sli_hbq_entry_count - Calculate total number of hbq entries
*
* This function adds the number of hbq entries in every HBQ to get
* the total number of hbq entries required for the HBA and returns
* the total count.
**/
static int
lpfc_sli_hbq_entry_count(void)
{
int hbq_count = lpfc_sli_hbq_count();
int count = 0;
int i;
for (i = 0; i < hbq_count; ++i)
count += lpfc_hbq_defs[i]->entry_count;
return count;
}
/**
* lpfc_sli_hbq_size - Calculate memory required for all hbq entries
*
* This function calculates amount of memory required for all hbq entries
* to be configured and returns the total memory required.
**/
int
lpfc_sli_hbq_size(void)
{
return lpfc_sli_hbq_entry_count() * sizeof(struct lpfc_hbq_entry);
}
/**
* lpfc_sli_hbq_setup - configure and initialize HBQs
* @phba: Pointer to HBA context object.
*
* This function is called during the SLI initialization to configure
* all the HBQs and post buffers to the HBQ. The caller is not
* required to hold any locks. This function will return zero if successful
* else it will return negative error code.
**/
static int
lpfc_sli_hbq_setup(struct lpfc_hba *phba)
{
int hbq_count = lpfc_sli_hbq_count();
LPFC_MBOXQ_t *pmb;
MAILBOX_t *pmbox;
uint32_t hbqno;
uint32_t hbq_entry_index;
/* Get a Mailbox buffer to setup mailbox
* commands for HBA initialization
*/
pmb = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!pmb)
return -ENOMEM;
pmbox = &pmb->u.mb;
/* Initialize the struct lpfc_sli_hbq structure for each hbq */
phba->link_state = LPFC_INIT_MBX_CMDS;
phba->hbq_in_use = 1;
hbq_entry_index = 0;
for (hbqno = 0; hbqno < hbq_count; ++hbqno) {
phba->hbqs[hbqno].next_hbqPutIdx = 0;
phba->hbqs[hbqno].hbqPutIdx = 0;
phba->hbqs[hbqno].local_hbqGetIdx = 0;
phba->hbqs[hbqno].entry_count =
lpfc_hbq_defs[hbqno]->entry_count;
lpfc_config_hbq(phba, hbqno, lpfc_hbq_defs[hbqno],
hbq_entry_index, pmb);
hbq_entry_index += phba->hbqs[hbqno].entry_count;
if (lpfc_sli_issue_mbox(phba, pmb, MBX_POLL) != MBX_SUCCESS) {
/* Adapter failed to init, mbxCmd <cmd> CFG_RING,
mbxStatus <status>, ring <num> */
lpfc_printf_log(phba, KERN_ERR,
LOG_SLI | LOG_VPORT,
"1805 Adapter failed to init. "
"Data: x%x x%x x%x\n",
pmbox->mbxCommand,
pmbox->mbxStatus, hbqno);
phba->link_state = LPFC_HBA_ERROR;
mempool_free(pmb, phba->mbox_mem_pool);
return -ENXIO;
}
}
phba->hbq_count = hbq_count;
mempool_free(pmb, phba->mbox_mem_pool);
/* Initially populate or replenish the HBQs */
for (hbqno = 0; hbqno < hbq_count; ++hbqno)
lpfc_sli_hbqbuf_init_hbqs(phba, hbqno);
return 0;
}
/**
* lpfc_sli4_rb_setup - Initialize and post RBs to HBA
* @phba: Pointer to HBA context object.
*
* This function is called during the SLI initialization to configure
* all the HBQs and post buffers to the HBQ. The caller is not
* required to hold any locks. This function will return zero if successful
* else it will return negative error code.
**/
static int
lpfc_sli4_rb_setup(struct lpfc_hba *phba)
{
phba->hbq_in_use = 1;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
phba->hbqs[LPFC_ELS_HBQ].entry_count =
lpfc_hbq_defs[LPFC_ELS_HBQ]->entry_count;
phba->hbq_count = 1;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli_hbqbuf_init_hbqs(phba, LPFC_ELS_HBQ);
/* Initially populate or replenish the HBQs */
return 0;
}
/**
* lpfc_sli_config_port - Issue config port mailbox command
* @phba: Pointer to HBA context object.
* @sli_mode: sli mode - 2/3
*
* This function is called by the sli initialization code path
* to issue config_port mailbox command. This function restarts the
* HBA firmware and issues a config_port mailbox command to configure
* the SLI interface in the sli mode specified by sli_mode
* variable. The caller is not required to hold any locks.
* The function returns 0 if successful, else returns negative error
* code.
**/
int
lpfc_sli_config_port(struct lpfc_hba *phba, int sli_mode)
{
LPFC_MBOXQ_t *pmb;
uint32_t resetcount = 0, rc = 0, done = 0;
pmb = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!pmb) {
phba->link_state = LPFC_HBA_ERROR;
return -ENOMEM;
}
phba->sli_rev = sli_mode;
while (resetcount < 2 && !done) {
spin_lock_irq(&phba->hbalock);
phba->sli.sli_flag |= LPFC_SLI_MBOX_ACTIVE;
spin_unlock_irq(&phba->hbalock);
phba->pport->port_state = LPFC_VPORT_UNKNOWN;
lpfc_sli_brdrestart(phba);
rc = lpfc_sli_chipset_init(phba);
if (rc)
break;
spin_lock_irq(&phba->hbalock);
phba->sli.sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
spin_unlock_irq(&phba->hbalock);
resetcount++;
/* Call pre CONFIG_PORT mailbox command initialization. A
* value of 0 means the call was successful. Any other
* nonzero value is a failure, but if ERESTART is returned,
* the driver may reset the HBA and try again.
*/
rc = lpfc_config_port_prep(phba);
if (rc == -ERESTART) {
phba->link_state = LPFC_LINK_UNKNOWN;
continue;
} else if (rc)
break;
phba->link_state = LPFC_INIT_MBX_CMDS;
lpfc_config_port(phba, pmb);
rc = lpfc_sli_issue_mbox(phba, pmb, MBX_POLL);
phba->sli3_options &= ~(LPFC_SLI3_NPIV_ENABLED |
LPFC_SLI3_HBQ_ENABLED |
LPFC_SLI3_CRP_ENABLED |
LPFC_SLI3_BG_ENABLED |
LPFC_SLI3_DSS_ENABLED);
if (rc != MBX_SUCCESS) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0442 Adapter failed to init, mbxCmd x%x "
"CONFIG_PORT, mbxStatus x%x Data: x%x\n",
pmb->u.mb.mbxCommand, pmb->u.mb.mbxStatus, 0);
spin_lock_irq(&phba->hbalock);
phba->sli.sli_flag &= ~LPFC_SLI_ACTIVE;
spin_unlock_irq(&phba->hbalock);
rc = -ENXIO;
} else {
/* Allow asynchronous mailbox command to go through */
spin_lock_irq(&phba->hbalock);
phba->sli.sli_flag &= ~LPFC_SLI_ASYNC_MBX_BLK;
spin_unlock_irq(&phba->hbalock);
done = 1;
if ((pmb->u.mb.un.varCfgPort.casabt == 1) &&
(pmb->u.mb.un.varCfgPort.gasabt == 0))
lpfc_printf_log(phba, KERN_WARNING, LOG_INIT,
"3110 Port did not grant ASABT\n");
}
}
if (!done) {
rc = -EINVAL;
goto do_prep_failed;
}
if (pmb->u.mb.un.varCfgPort.sli_mode == 3) {
if (!pmb->u.mb.un.varCfgPort.cMA) {
rc = -ENXIO;
goto do_prep_failed;
}
if (phba->max_vpi && pmb->u.mb.un.varCfgPort.gmv) {
phba->sli3_options |= LPFC_SLI3_NPIV_ENABLED;
phba->max_vpi = pmb->u.mb.un.varCfgPort.max_vpi;
phba->max_vports = (phba->max_vpi > phba->max_vports) ?
phba->max_vpi : phba->max_vports;
} else
phba->max_vpi = 0;
phba->fips_level = 0;
phba->fips_spec_rev = 0;
if (pmb->u.mb.un.varCfgPort.gdss) {
phba->sli3_options |= LPFC_SLI3_DSS_ENABLED;
phba->fips_level = pmb->u.mb.un.varCfgPort.fips_level;
phba->fips_spec_rev = pmb->u.mb.un.varCfgPort.fips_rev;
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2850 Security Crypto Active. FIPS x%d "
"(Spec Rev: x%d)",
phba->fips_level, phba->fips_spec_rev);
}
if (pmb->u.mb.un.varCfgPort.sec_err) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2856 Config Port Security Crypto "
"Error: x%x ",
pmb->u.mb.un.varCfgPort.sec_err);
}
if (pmb->u.mb.un.varCfgPort.gerbm)
phba->sli3_options |= LPFC_SLI3_HBQ_ENABLED;
if (pmb->u.mb.un.varCfgPort.gcrp)
phba->sli3_options |= LPFC_SLI3_CRP_ENABLED;
phba->hbq_get = phba->mbox->us.s3_pgp.hbq_get;
phba->port_gp = phba->mbox->us.s3_pgp.port;
if (phba->sli3_options & LPFC_SLI3_BG_ENABLED) {
if (pmb->u.mb.un.varCfgPort.gbg == 0) {
phba->cfg_enable_bg = 0;
phba->sli3_options &= ~LPFC_SLI3_BG_ENABLED;
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0443 Adapter did not grant "
"BlockGuard\n");
}
}
} else {
phba->hbq_get = NULL;
phba->port_gp = phba->mbox->us.s2.port;
phba->max_vpi = 0;
}
do_prep_failed:
mempool_free(pmb, phba->mbox_mem_pool);
return rc;
}
/**
* lpfc_sli_hba_setup - SLI initialization function
* @phba: Pointer to HBA context object.
*
* This function is the main SLI initialization function. This function
* is called by the HBA initialization code, HBA reset code and HBA
* error attention handler code. Caller is not required to hold any
* locks. This function issues config_port mailbox command to configure
* the SLI, setup iocb rings and HBQ rings. In the end the function
* calls the config_port_post function to issue init_link mailbox
* command and to start the discovery. The function will return zero
* if successful, else it will return negative error code.
**/
int
lpfc_sli_hba_setup(struct lpfc_hba *phba)
{
uint32_t rc;
int mode = 3, i;
int longs;
switch (phba->cfg_sli_mode) {
case 2:
if (phba->cfg_enable_npiv) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT | LOG_VPORT,
"1824 NPIV enabled: Override sli_mode "
"parameter (%d) to auto (0).\n",
phba->cfg_sli_mode);
break;
}
mode = 2;
break;
case 0:
case 3:
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_INIT | LOG_VPORT,
"1819 Unrecognized sli_mode parameter: %d.\n",
phba->cfg_sli_mode);
break;
}
phba->fcp_embed_io = 0; /* SLI4 FC support only */
rc = lpfc_sli_config_port(phba, mode);
if (rc && phba->cfg_sli_mode == 3)
lpfc_printf_log(phba, KERN_ERR, LOG_INIT | LOG_VPORT,
"1820 Unable to select SLI-3. "
"Not supported by adapter.\n");
if (rc && mode != 2)
rc = lpfc_sli_config_port(phba, 2);
else if (rc && mode == 2)
rc = lpfc_sli_config_port(phba, 3);
if (rc)
goto lpfc_sli_hba_setup_error;
/* Enable PCIe device Advanced Error Reporting (AER) if configured */
if (phba->cfg_aer_support == 1 && !(phba->hba_flag & HBA_AER_ENABLED)) {
rc = pci_enable_pcie_error_reporting(phba->pcidev);
if (!rc) {
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2709 This device supports "
"Advanced Error Reporting (AER)\n");
spin_lock_irq(&phba->hbalock);
phba->hba_flag |= HBA_AER_ENABLED;
spin_unlock_irq(&phba->hbalock);
} else {
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2708 This device does not support "
"Advanced Error Reporting (AER): %d\n",
rc);
phba->cfg_aer_support = 0;
}
}
if (phba->sli_rev == 3) {
phba->iocb_cmd_size = SLI3_IOCB_CMD_SIZE;
phba->iocb_rsp_size = SLI3_IOCB_RSP_SIZE;
} else {
phba->iocb_cmd_size = SLI2_IOCB_CMD_SIZE;
phba->iocb_rsp_size = SLI2_IOCB_RSP_SIZE;
phba->sli3_options = 0;
}
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"0444 Firmware in SLI %x mode. Max_vpi %d\n",
phba->sli_rev, phba->max_vpi);
rc = lpfc_sli_ring_map(phba);
if (rc)
goto lpfc_sli_hba_setup_error;
/* Initialize VPIs. */
if (phba->sli_rev == LPFC_SLI_REV3) {
/*
* The VPI bitmask and physical ID array are allocated
* and initialized once only - at driver load. A port
* reset doesn't need to reinitialize this memory.
*/
if ((phba->vpi_bmask == NULL) && (phba->vpi_ids == NULL)) {
longs = (phba->max_vpi + BITS_PER_LONG) / BITS_PER_LONG;
phba->vpi_bmask = kzalloc(longs * sizeof(unsigned long),
GFP_KERNEL);
if (!phba->vpi_bmask) {
rc = -ENOMEM;
goto lpfc_sli_hba_setup_error;
}
phba->vpi_ids = kzalloc(
(phba->max_vpi+1) * sizeof(uint16_t),
GFP_KERNEL);
if (!phba->vpi_ids) {
kfree(phba->vpi_bmask);
rc = -ENOMEM;
goto lpfc_sli_hba_setup_error;
}
for (i = 0; i < phba->max_vpi; i++)
phba->vpi_ids[i] = i;
}
}
/* Init HBQs */
if (phba->sli3_options & LPFC_SLI3_HBQ_ENABLED) {
rc = lpfc_sli_hbq_setup(phba);
if (rc)
goto lpfc_sli_hba_setup_error;
}
spin_lock_irq(&phba->hbalock);
phba->sli.sli_flag |= LPFC_PROCESS_LA;
spin_unlock_irq(&phba->hbalock);
rc = lpfc_config_port_post(phba);
if (rc)
goto lpfc_sli_hba_setup_error;
return rc;
lpfc_sli_hba_setup_error:
phba->link_state = LPFC_HBA_ERROR;
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0445 Firmware initialization failed\n");
return rc;
}
/**
* lpfc_sli4_read_fcoe_params - Read fcoe params from conf region
* @phba: Pointer to HBA context object.
* @mboxq: mailbox pointer.
* This function issue a dump mailbox command to read config region
* 23 and parse the records in the region and populate driver
* data structure.
**/
static int
lpfc_sli4_read_fcoe_params(struct lpfc_hba *phba)
{
LPFC_MBOXQ_t *mboxq;
struct lpfc_dmabuf *mp;
struct lpfc_mqe *mqe;
uint32_t data_length;
int rc;
/* Program the default value of vlan_id and fc_map */
phba->valid_vlan = 0;
phba->fc_map[0] = LPFC_FCOE_FCF_MAP0;
phba->fc_map[1] = LPFC_FCOE_FCF_MAP1;
phba->fc_map[2] = LPFC_FCOE_FCF_MAP2;
mboxq = (LPFC_MBOXQ_t *)mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq)
return -ENOMEM;
mqe = &mboxq->u.mqe;
if (lpfc_sli4_dump_cfg_rg23(phba, mboxq)) {
rc = -ENOMEM;
goto out_free_mboxq;
}
mp = (struct lpfc_dmabuf *) mboxq->context1;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):2571 Mailbox cmd x%x Status x%x "
"Data: x%x x%x x%x x%x x%x x%x x%x x%x x%x "
"x%x x%x x%x x%x x%x x%x x%x x%x x%x "
"CQ: x%x x%x x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0,
bf_get(lpfc_mqe_command, mqe),
bf_get(lpfc_mqe_status, mqe),
mqe->un.mb_words[0], mqe->un.mb_words[1],
mqe->un.mb_words[2], mqe->un.mb_words[3],
mqe->un.mb_words[4], mqe->un.mb_words[5],
mqe->un.mb_words[6], mqe->un.mb_words[7],
mqe->un.mb_words[8], mqe->un.mb_words[9],
mqe->un.mb_words[10], mqe->un.mb_words[11],
mqe->un.mb_words[12], mqe->un.mb_words[13],
mqe->un.mb_words[14], mqe->un.mb_words[15],
mqe->un.mb_words[16], mqe->un.mb_words[50],
mboxq->mcqe.word0,
mboxq->mcqe.mcqe_tag0, mboxq->mcqe.mcqe_tag1,
mboxq->mcqe.trailer);
if (rc) {
lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
rc = -EIO;
goto out_free_mboxq;
}
data_length = mqe->un.mb_words[5];
if (data_length > DMP_RGN23_SIZE) {
lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
rc = -EIO;
goto out_free_mboxq;
}
lpfc_parse_fcoe_conf(phba, mp->virt, data_length);
lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
rc = 0;
out_free_mboxq:
mempool_free(mboxq, phba->mbox_mem_pool);
return rc;
}
/**
* lpfc_sli4_read_rev - Issue READ_REV and collect vpd data
* @phba: pointer to lpfc hba data structure.
* @mboxq: pointer to the LPFC_MBOXQ_t structure.
* @vpd: pointer to the memory to hold resulting port vpd data.
* @vpd_size: On input, the number of bytes allocated to @vpd.
* On output, the number of data bytes in @vpd.
*
* This routine executes a READ_REV SLI4 mailbox command. In
* addition, this routine gets the port vpd data.
*
* Return codes
* 0 - successful
* -ENOMEM - could not allocated memory.
**/
static int
lpfc_sli4_read_rev(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq,
uint8_t *vpd, uint32_t *vpd_size)
{
int rc = 0;
uint32_t dma_size;
struct lpfc_dmabuf *dmabuf;
struct lpfc_mqe *mqe;
dmabuf = kzalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
if (!dmabuf)
return -ENOMEM;
/*
* Get a DMA buffer for the vpd data resulting from the READ_REV
* mailbox command.
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
*/
dma_size = *vpd_size;
dmabuf->virt = dma_zalloc_coherent(&phba->pcidev->dev, dma_size,
&dmabuf->phys, GFP_KERNEL);
if (!dmabuf->virt) {
kfree(dmabuf);
return -ENOMEM;
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
}
/*
* The SLI4 implementation of READ_REV conflicts at word1,
* bits 31:16 and SLI4 adds vpd functionality not present
* in SLI3. This code corrects the conflicts.
*/
lpfc_read_rev(phba, mboxq);
mqe = &mboxq->u.mqe;
mqe->un.read_rev.vpd_paddr_high = putPaddrHigh(dmabuf->phys);
mqe->un.read_rev.vpd_paddr_low = putPaddrLow(dmabuf->phys);
mqe->un.read_rev.word1 &= 0x0000FFFF;
bf_set(lpfc_mbx_rd_rev_vpd, &mqe->un.read_rev, 1);
bf_set(lpfc_mbx_rd_rev_avail_len, &mqe->un.read_rev, dma_size);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc) {
dma_free_coherent(&phba->pcidev->dev, dma_size,
dmabuf->virt, dmabuf->phys);
kfree(dmabuf);
return -EIO;
}
/*
* The available vpd length cannot be bigger than the
* DMA buffer passed to the port. Catch the less than
* case and update the caller's size.
*/
if (mqe->un.read_rev.avail_vpd_len < *vpd_size)
*vpd_size = mqe->un.read_rev.avail_vpd_len;
memcpy(vpd, dmabuf->virt, *vpd_size);
dma_free_coherent(&phba->pcidev->dev, dma_size,
dmabuf->virt, dmabuf->phys);
kfree(dmabuf);
return 0;
}
/**
* lpfc_sli4_retrieve_pport_name - Retrieve SLI4 device physical port name
* @phba: pointer to lpfc hba data structure.
*
* This routine retrieves SLI4 device physical port name this PCI function
* is attached to.
*
* Return codes
* 0 - successful
* otherwise - failed to retrieve physical port name
**/
static int
lpfc_sli4_retrieve_pport_name(struct lpfc_hba *phba)
{
LPFC_MBOXQ_t *mboxq;
struct lpfc_mbx_get_cntl_attributes *mbx_cntl_attr;
struct lpfc_controller_attribute *cntl_attr;
struct lpfc_mbx_get_port_name *get_port_name;
void *virtaddr = NULL;
uint32_t alloclen, reqlen;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
char cport_name = 0;
int rc;
/* We assume nothing at this point */
phba->sli4_hba.lnk_info.lnk_dv = LPFC_LNK_DAT_INVAL;
phba->sli4_hba.pport_name_sta = LPFC_SLI4_PPNAME_NON;
mboxq = (LPFC_MBOXQ_t *)mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq)
return -ENOMEM;
/* obtain link type and link number via READ_CONFIG */
phba->sli4_hba.lnk_info.lnk_dv = LPFC_LNK_DAT_INVAL;
lpfc_sli4_read_config(phba);
if (phba->sli4_hba.lnk_info.lnk_dv == LPFC_LNK_DAT_VAL)
goto retrieve_ppname;
/* obtain link type and link number via COMMON_GET_CNTL_ATTRIBUTES */
reqlen = sizeof(struct lpfc_mbx_get_cntl_attributes);
alloclen = lpfc_sli4_config(phba, mboxq, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_GET_CNTL_ATTRIBUTES, reqlen,
LPFC_SLI4_MBX_NEMBED);
if (alloclen < reqlen) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3084 Allocated DMA memory size (%d) is "
"less than the requested DMA memory size "
"(%d)\n", alloclen, reqlen);
rc = -ENOMEM;
goto out_free_mboxq;
}
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
virtaddr = mboxq->sge_array->addr[0];
mbx_cntl_attr = (struct lpfc_mbx_get_cntl_attributes *)virtaddr;
shdr = &mbx_cntl_attr->cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"3085 Mailbox x%x (x%x/x%x) failed, "
"rc:x%x, status:x%x, add_status:x%x\n",
bf_get(lpfc_mqe_command, &mboxq->u.mqe),
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
rc, shdr_status, shdr_add_status);
rc = -ENXIO;
goto out_free_mboxq;
}
cntl_attr = &mbx_cntl_attr->cntl_attr;
phba->sli4_hba.lnk_info.lnk_dv = LPFC_LNK_DAT_VAL;
phba->sli4_hba.lnk_info.lnk_tp =
bf_get(lpfc_cntl_attr_lnk_type, cntl_attr);
phba->sli4_hba.lnk_info.lnk_no =
bf_get(lpfc_cntl_attr_lnk_numb, cntl_attr);
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"3086 lnk_type:%d, lnk_numb:%d\n",
phba->sli4_hba.lnk_info.lnk_tp,
phba->sli4_hba.lnk_info.lnk_no);
retrieve_ppname:
lpfc_sli4_config(phba, mboxq, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_GET_PORT_NAME,
sizeof(struct lpfc_mbx_get_port_name) -
sizeof(struct lpfc_sli4_cfg_mhdr),
LPFC_SLI4_MBX_EMBED);
get_port_name = &mboxq->u.mqe.un.get_port_name;
shdr = (union lpfc_sli4_cfg_shdr *)&get_port_name->header.cfg_shdr;
bf_set(lpfc_mbox_hdr_version, &shdr->request, LPFC_OPCODE_VERSION_1);
bf_set(lpfc_mbx_get_port_name_lnk_type, &get_port_name->u.request,
phba->sli4_hba.lnk_info.lnk_tp);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"3087 Mailbox x%x (x%x/x%x) failed: "
"rc:x%x, status:x%x, add_status:x%x\n",
bf_get(lpfc_mqe_command, &mboxq->u.mqe),
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
rc, shdr_status, shdr_add_status);
rc = -ENXIO;
goto out_free_mboxq;
}
switch (phba->sli4_hba.lnk_info.lnk_no) {
case LPFC_LINK_NUMBER_0:
cport_name = bf_get(lpfc_mbx_get_port_name_name0,
&get_port_name->u.response);
phba->sli4_hba.pport_name_sta = LPFC_SLI4_PPNAME_GET;
break;
case LPFC_LINK_NUMBER_1:
cport_name = bf_get(lpfc_mbx_get_port_name_name1,
&get_port_name->u.response);
phba->sli4_hba.pport_name_sta = LPFC_SLI4_PPNAME_GET;
break;
case LPFC_LINK_NUMBER_2:
cport_name = bf_get(lpfc_mbx_get_port_name_name2,
&get_port_name->u.response);
phba->sli4_hba.pport_name_sta = LPFC_SLI4_PPNAME_GET;
break;
case LPFC_LINK_NUMBER_3:
cport_name = bf_get(lpfc_mbx_get_port_name_name3,
&get_port_name->u.response);
phba->sli4_hba.pport_name_sta = LPFC_SLI4_PPNAME_GET;
break;
default:
break;
}
if (phba->sli4_hba.pport_name_sta == LPFC_SLI4_PPNAME_GET) {
phba->Port[0] = cport_name;
phba->Port[1] = '\0';
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"3091 SLI get port name: %s\n", phba->Port);
}
out_free_mboxq:
if (rc != MBX_TIMEOUT) {
if (bf_get(lpfc_mqe_command, &mboxq->u.mqe) == MBX_SLI4_CONFIG)
lpfc_sli4_mbox_cmd_free(phba, mboxq);
else
mempool_free(mboxq, phba->mbox_mem_pool);
}
return rc;
}
/**
* lpfc_sli4_arm_cqeq_intr - Arm sli-4 device completion and event queues
* @phba: pointer to lpfc hba data structure.
*
* This routine is called to explicitly arm the SLI4 device's completion and
* event queues
**/
static void
lpfc_sli4_arm_cqeq_intr(struct lpfc_hba *phba)
{
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int qidx;
struct lpfc_sli4_hba *sli4_hba = &phba->sli4_hba;
sli4_hba->sli4_cq_release(sli4_hba->mbx_cq, LPFC_QUEUE_REARM);
sli4_hba->sli4_cq_release(sli4_hba->els_cq, LPFC_QUEUE_REARM);
if (sli4_hba->nvmels_cq)
sli4_hba->sli4_cq_release(sli4_hba->nvmels_cq,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
LPFC_QUEUE_REARM);
if (sli4_hba->fcp_cq)
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (qidx = 0; qidx < phba->cfg_fcp_io_channel; qidx++)
sli4_hba->sli4_cq_release(sli4_hba->fcp_cq[qidx],
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
LPFC_QUEUE_REARM);
if (sli4_hba->nvme_cq)
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (qidx = 0; qidx < phba->cfg_nvme_io_channel; qidx++)
sli4_hba->sli4_cq_release(sli4_hba->nvme_cq[qidx],
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
LPFC_QUEUE_REARM);
if (phba->cfg_fof)
sli4_hba->sli4_cq_release(sli4_hba->oas_cq, LPFC_QUEUE_REARM);
if (sli4_hba->hba_eq)
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (qidx = 0; qidx < phba->io_channel_irqs; qidx++)
sli4_hba->sli4_eq_release(sli4_hba->hba_eq[qidx],
LPFC_QUEUE_REARM);
if (phba->nvmet_support) {
for (qidx = 0; qidx < phba->cfg_nvmet_mrq; qidx++) {
sli4_hba->sli4_cq_release(
sli4_hba->nvmet_cqset[qidx],
LPFC_QUEUE_REARM);
}
}
if (phba->cfg_fof)
sli4_hba->sli4_eq_release(sli4_hba->fof_eq, LPFC_QUEUE_REARM);
}
/**
* lpfc_sli4_get_avail_extnt_rsrc - Get available resource extent count.
* @phba: Pointer to HBA context object.
* @type: The resource extent type.
* @extnt_count: buffer to hold port available extent count.
* @extnt_size: buffer to hold element count per extent.
*
* This function calls the port and retrievs the number of available
* extents and their size for a particular extent type.
*
* Returns: 0 if successful. Nonzero otherwise.
**/
int
lpfc_sli4_get_avail_extnt_rsrc(struct lpfc_hba *phba, uint16_t type,
uint16_t *extnt_count, uint16_t *extnt_size)
{
int rc = 0;
uint32_t length;
uint32_t mbox_tmo;
struct lpfc_mbx_get_rsrc_extent_info *rsrc_info;
LPFC_MBOXQ_t *mbox;
mbox = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
/* Find out how many extents are available for this resource type */
length = (sizeof(struct lpfc_mbx_get_rsrc_extent_info) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_GET_RSRC_EXTENT_INFO,
length, LPFC_SLI4_MBX_EMBED);
/* Send an extents count of 0 - the GET doesn't use it. */
rc = lpfc_sli4_mbox_rsrc_extent(phba, mbox, 0, type,
LPFC_SLI4_MBX_EMBED);
if (unlikely(rc)) {
rc = -EIO;
goto err_exit;
}
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
if (unlikely(rc)) {
rc = -EIO;
goto err_exit;
}
rsrc_info = &mbox->u.mqe.un.rsrc_extent_info;
if (bf_get(lpfc_mbox_hdr_status,
&rsrc_info->header.cfg_shdr.response)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_INIT,
"2930 Failed to get resource extents "
"Status 0x%x Add'l Status 0x%x\n",
bf_get(lpfc_mbox_hdr_status,
&rsrc_info->header.cfg_shdr.response),
bf_get(lpfc_mbox_hdr_add_status,
&rsrc_info->header.cfg_shdr.response));
rc = -EIO;
goto err_exit;
}
*extnt_count = bf_get(lpfc_mbx_get_rsrc_extent_info_cnt,
&rsrc_info->u.rsp);
*extnt_size = bf_get(lpfc_mbx_get_rsrc_extent_info_size,
&rsrc_info->u.rsp);
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"3162 Retrieved extents type-%d from port: count:%d, "
"size:%d\n", type, *extnt_count, *extnt_size);
err_exit:
mempool_free(mbox, phba->mbox_mem_pool);
return rc;
}
/**
* lpfc_sli4_chk_avail_extnt_rsrc - Check for available SLI4 resource extents.
* @phba: Pointer to HBA context object.
* @type: The extent type to check.
*
* This function reads the current available extents from the port and checks
* if the extent count or extent size has changed since the last access.
* Callers use this routine post port reset to understand if there is a
* extent reprovisioning requirement.
*
* Returns:
* -Error: error indicates problem.
* 1: Extent count or size has changed.
* 0: No changes.
**/
static int
lpfc_sli4_chk_avail_extnt_rsrc(struct lpfc_hba *phba, uint16_t type)
{
uint16_t curr_ext_cnt, rsrc_ext_cnt;
uint16_t size_diff, rsrc_ext_size;
int rc = 0;
struct lpfc_rsrc_blks *rsrc_entry;
struct list_head *rsrc_blk_list = NULL;
size_diff = 0;
curr_ext_cnt = 0;
rc = lpfc_sli4_get_avail_extnt_rsrc(phba, type,
&rsrc_ext_cnt,
&rsrc_ext_size);
if (unlikely(rc))
return -EIO;
switch (type) {
case LPFC_RSC_TYPE_FCOE_RPI:
rsrc_blk_list = &phba->sli4_hba.lpfc_rpi_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_VPI:
rsrc_blk_list = &phba->lpfc_vpi_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_XRI:
rsrc_blk_list = &phba->sli4_hba.lpfc_xri_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_VFI:
rsrc_blk_list = &phba->sli4_hba.lpfc_vfi_blk_list;
break;
default:
break;
}
list_for_each_entry(rsrc_entry, rsrc_blk_list, list) {
curr_ext_cnt++;
if (rsrc_entry->rsrc_size != rsrc_ext_size)
size_diff++;
}
if (curr_ext_cnt != rsrc_ext_cnt || size_diff != 0)
rc = 1;
return rc;
}
/**
* lpfc_sli4_cfg_post_extnts -
* @phba: Pointer to HBA context object.
* @extnt_cnt - number of available extents.
* @type - the extent type (rpi, xri, vfi, vpi).
* @emb - buffer to hold either MBX_EMBED or MBX_NEMBED operation.
* @mbox - pointer to the caller's allocated mailbox structure.
*
* This function executes the extents allocation request. It also
* takes care of the amount of memory needed to allocate or get the
* allocated extents. It is the caller's responsibility to evaluate
* the response.
*
* Returns:
* -Error: Error value describes the condition found.
* 0: if successful
**/
static int
lpfc_sli4_cfg_post_extnts(struct lpfc_hba *phba, uint16_t extnt_cnt,
uint16_t type, bool *emb, LPFC_MBOXQ_t *mbox)
{
int rc = 0;
uint32_t req_len;
uint32_t emb_len;
uint32_t alloc_len, mbox_tmo;
/* Calculate the total requested length of the dma memory */
req_len = extnt_cnt * sizeof(uint16_t);
/*
* Calculate the size of an embedded mailbox. The uint32_t
* accounts for extents-specific word.
*/
emb_len = sizeof(MAILBOX_t) - sizeof(struct mbox_header) -
sizeof(uint32_t);
/*
* Presume the allocation and response will fit into an embedded
* mailbox. If not true, reconfigure to a non-embedded mailbox.
*/
*emb = LPFC_SLI4_MBX_EMBED;
if (req_len > emb_len) {
req_len = extnt_cnt * sizeof(uint16_t) +
sizeof(union lpfc_sli4_cfg_shdr) +
sizeof(uint32_t);
*emb = LPFC_SLI4_MBX_NEMBED;
}
alloc_len = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_ALLOC_RSRC_EXTENT,
req_len, *emb);
if (alloc_len < req_len) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2982 Allocated DMA memory size (x%x) is "
"less than the requested DMA memory "
"size (x%x)\n", alloc_len, req_len);
return -ENOMEM;
}
rc = lpfc_sli4_mbox_rsrc_extent(phba, mbox, extnt_cnt, type, *emb);
if (unlikely(rc))
return -EIO;
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
if (unlikely(rc))
rc = -EIO;
return rc;
}
/**
* lpfc_sli4_alloc_extent - Allocate an SLI4 resource extent.
* @phba: Pointer to HBA context object.
* @type: The resource extent type to allocate.
*
* This function allocates the number of elements for the specified
* resource type.
**/
static int
lpfc_sli4_alloc_extent(struct lpfc_hba *phba, uint16_t type)
{
bool emb = false;
uint16_t rsrc_id_cnt, rsrc_cnt, rsrc_size;
uint16_t rsrc_id, rsrc_start, j, k;
uint16_t *ids;
int i, rc;
unsigned long longs;
unsigned long *bmask;
struct lpfc_rsrc_blks *rsrc_blks;
LPFC_MBOXQ_t *mbox;
uint32_t length;
struct lpfc_id_range *id_array = NULL;
void *virtaddr = NULL;
struct lpfc_mbx_nembed_rsrc_extent *n_rsrc;
struct lpfc_mbx_alloc_rsrc_extents *rsrc_ext;
struct list_head *ext_blk_list;
rc = lpfc_sli4_get_avail_extnt_rsrc(phba, type,
&rsrc_cnt,
&rsrc_size);
if (unlikely(rc))
return -EIO;
if ((rsrc_cnt == 0) || (rsrc_size == 0)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_INIT,
"3009 No available Resource Extents "
"for resource type 0x%x: Count: 0x%x, "
"Size 0x%x\n", type, rsrc_cnt,
rsrc_size);
return -ENOMEM;
}
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_INIT | LOG_SLI,
"2903 Post resource extents type-0x%x: "
"count:%d, size %d\n", type, rsrc_cnt, rsrc_size);
mbox = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
rc = lpfc_sli4_cfg_post_extnts(phba, rsrc_cnt, type, &emb, mbox);
if (unlikely(rc)) {
rc = -EIO;
goto err_exit;
}
/*
* Figure out where the response is located. Then get local pointers
* to the response data. The port does not guarantee to respond to
* all extents counts request so update the local variable with the
* allocated count from the port.
*/
if (emb == LPFC_SLI4_MBX_EMBED) {
rsrc_ext = &mbox->u.mqe.un.alloc_rsrc_extents;
id_array = &rsrc_ext->u.rsp.id[0];
rsrc_cnt = bf_get(lpfc_mbx_rsrc_cnt, &rsrc_ext->u.rsp);
} else {
virtaddr = mbox->sge_array->addr[0];
n_rsrc = (struct lpfc_mbx_nembed_rsrc_extent *) virtaddr;
rsrc_cnt = bf_get(lpfc_mbx_rsrc_cnt, n_rsrc);
id_array = &n_rsrc->id;
}
longs = ((rsrc_cnt * rsrc_size) + BITS_PER_LONG - 1) / BITS_PER_LONG;
rsrc_id_cnt = rsrc_cnt * rsrc_size;
/*
* Based on the resource size and count, correct the base and max
* resource values.
*/
length = sizeof(struct lpfc_rsrc_blks);
switch (type) {
case LPFC_RSC_TYPE_FCOE_RPI:
phba->sli4_hba.rpi_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.rpi_bmask)) {
rc = -ENOMEM;
goto err_exit;
}
phba->sli4_hba.rpi_ids = kzalloc(rsrc_id_cnt *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.rpi_ids)) {
kfree(phba->sli4_hba.rpi_bmask);
rc = -ENOMEM;
goto err_exit;
}
/*
* The next_rpi was initialized with the maximum available
* count but the port may allocate a smaller number. Catch
* that case and update the next_rpi.
*/
phba->sli4_hba.next_rpi = rsrc_id_cnt;
/* Initialize local ptrs for common extent processing later. */
bmask = phba->sli4_hba.rpi_bmask;
ids = phba->sli4_hba.rpi_ids;
ext_blk_list = &phba->sli4_hba.lpfc_rpi_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_VPI:
phba->vpi_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->vpi_bmask)) {
rc = -ENOMEM;
goto err_exit;
}
phba->vpi_ids = kzalloc(rsrc_id_cnt *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->vpi_ids)) {
kfree(phba->vpi_bmask);
rc = -ENOMEM;
goto err_exit;
}
/* Initialize local ptrs for common extent processing later. */
bmask = phba->vpi_bmask;
ids = phba->vpi_ids;
ext_blk_list = &phba->lpfc_vpi_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_XRI:
phba->sli4_hba.xri_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.xri_bmask)) {
rc = -ENOMEM;
goto err_exit;
}
phba->sli4_hba.max_cfg_param.xri_used = 0;
phba->sli4_hba.xri_ids = kzalloc(rsrc_id_cnt *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.xri_ids)) {
kfree(phba->sli4_hba.xri_bmask);
rc = -ENOMEM;
goto err_exit;
}
/* Initialize local ptrs for common extent processing later. */
bmask = phba->sli4_hba.xri_bmask;
ids = phba->sli4_hba.xri_ids;
ext_blk_list = &phba->sli4_hba.lpfc_xri_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_VFI:
phba->sli4_hba.vfi_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.vfi_bmask)) {
rc = -ENOMEM;
goto err_exit;
}
phba->sli4_hba.vfi_ids = kzalloc(rsrc_id_cnt *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.vfi_ids)) {
kfree(phba->sli4_hba.vfi_bmask);
rc = -ENOMEM;
goto err_exit;
}
/* Initialize local ptrs for common extent processing later. */
bmask = phba->sli4_hba.vfi_bmask;
ids = phba->sli4_hba.vfi_ids;
ext_blk_list = &phba->sli4_hba.lpfc_vfi_blk_list;
break;
default:
/* Unsupported Opcode. Fail call. */
id_array = NULL;
bmask = NULL;
ids = NULL;
ext_blk_list = NULL;
goto err_exit;
}
/*
* Complete initializing the extent configuration with the
* allocated ids assigned to this function. The bitmask serves
* as an index into the array and manages the available ids. The
* array just stores the ids communicated to the port via the wqes.
*/
for (i = 0, j = 0, k = 0; i < rsrc_cnt; i++) {
if ((i % 2) == 0)
rsrc_id = bf_get(lpfc_mbx_rsrc_id_word4_0,
&id_array[k]);
else
rsrc_id = bf_get(lpfc_mbx_rsrc_id_word4_1,
&id_array[k]);
rsrc_blks = kzalloc(length, GFP_KERNEL);
if (unlikely(!rsrc_blks)) {
rc = -ENOMEM;
kfree(bmask);
kfree(ids);
goto err_exit;
}
rsrc_blks->rsrc_start = rsrc_id;
rsrc_blks->rsrc_size = rsrc_size;
list_add_tail(&rsrc_blks->list, ext_blk_list);
rsrc_start = rsrc_id;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if ((type == LPFC_RSC_TYPE_FCOE_XRI) && (j == 0)) {
phba->sli4_hba.scsi_xri_start = rsrc_start +
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli4_get_iocb_cnt(phba);
phba->sli4_hba.nvme_xri_start =
phba->sli4_hba.scsi_xri_start +
phba->sli4_hba.scsi_xri_max;
}
while (rsrc_id < (rsrc_start + rsrc_size)) {
ids[j] = rsrc_id;
rsrc_id++;
j++;
}
/* Entire word processed. Get next word.*/
if ((i % 2) == 1)
k++;
}
err_exit:
lpfc_sli4_mbox_cmd_free(phba, mbox);
return rc;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/**
* lpfc_sli4_dealloc_extent - Deallocate an SLI4 resource extent.
* @phba: Pointer to HBA context object.
* @type: the extent's type.
*
* This function deallocates all extents of a particular resource type.
* SLI4 does not allow for deallocating a particular extent range. It
* is the caller's responsibility to release all kernel memory resources.
**/
static int
lpfc_sli4_dealloc_extent(struct lpfc_hba *phba, uint16_t type)
{
int rc;
uint32_t length, mbox_tmo = 0;
LPFC_MBOXQ_t *mbox;
struct lpfc_mbx_dealloc_rsrc_extents *dealloc_rsrc;
struct lpfc_rsrc_blks *rsrc_blk, *rsrc_blk_next;
mbox = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
/*
* This function sends an embedded mailbox because it only sends the
* the resource type. All extents of this type are released by the
* port.
*/
length = (sizeof(struct lpfc_mbx_dealloc_rsrc_extents) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_DEALLOC_RSRC_EXTENT,
length, LPFC_SLI4_MBX_EMBED);
/* Send an extents count of 0 - the dealloc doesn't use it. */
rc = lpfc_sli4_mbox_rsrc_extent(phba, mbox, 0, type,
LPFC_SLI4_MBX_EMBED);
if (unlikely(rc)) {
rc = -EIO;
goto out_free_mbox;
}
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
if (unlikely(rc)) {
rc = -EIO;
goto out_free_mbox;
}
dealloc_rsrc = &mbox->u.mqe.un.dealloc_rsrc_extents;
if (bf_get(lpfc_mbox_hdr_status,
&dealloc_rsrc->header.cfg_shdr.response)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_INIT,
"2919 Failed to release resource extents "
"for type %d - Status 0x%x Add'l Status 0x%x. "
"Resource memory not released.\n",
type,
bf_get(lpfc_mbox_hdr_status,
&dealloc_rsrc->header.cfg_shdr.response),
bf_get(lpfc_mbox_hdr_add_status,
&dealloc_rsrc->header.cfg_shdr.response));
rc = -EIO;
goto out_free_mbox;
}
/* Release kernel memory resources for the specific type. */
switch (type) {
case LPFC_RSC_TYPE_FCOE_VPI:
kfree(phba->vpi_bmask);
kfree(phba->vpi_ids);
bf_set(lpfc_vpi_rsrc_rdy, &phba->sli4_hba.sli4_flags, 0);
list_for_each_entry_safe(rsrc_blk, rsrc_blk_next,
&phba->lpfc_vpi_blk_list, list) {
list_del_init(&rsrc_blk->list);
kfree(rsrc_blk);
}
phba->sli4_hba.max_cfg_param.vpi_used = 0;
break;
case LPFC_RSC_TYPE_FCOE_XRI:
kfree(phba->sli4_hba.xri_bmask);
kfree(phba->sli4_hba.xri_ids);
list_for_each_entry_safe(rsrc_blk, rsrc_blk_next,
&phba->sli4_hba.lpfc_xri_blk_list, list) {
list_del_init(&rsrc_blk->list);
kfree(rsrc_blk);
}
break;
case LPFC_RSC_TYPE_FCOE_VFI:
kfree(phba->sli4_hba.vfi_bmask);
kfree(phba->sli4_hba.vfi_ids);
bf_set(lpfc_vfi_rsrc_rdy, &phba->sli4_hba.sli4_flags, 0);
list_for_each_entry_safe(rsrc_blk, rsrc_blk_next,
&phba->sli4_hba.lpfc_vfi_blk_list, list) {
list_del_init(&rsrc_blk->list);
kfree(rsrc_blk);
}
break;
case LPFC_RSC_TYPE_FCOE_RPI:
/* RPI bitmask and physical id array are cleaned up earlier. */
list_for_each_entry_safe(rsrc_blk, rsrc_blk_next,
&phba->sli4_hba.lpfc_rpi_blk_list, list) {
list_del_init(&rsrc_blk->list);
kfree(rsrc_blk);
}
break;
default:
break;
}
bf_set(lpfc_idx_rsrc_rdy, &phba->sli4_hba.sli4_flags, 0);
out_free_mbox:
mempool_free(mbox, phba->mbox_mem_pool);
return rc;
}
scsi: lpfc: Mark symbols static where possible We get a few warnings when building kernel with W=1: drivers/scsi/lpfc/lpfc_sli.c:5693:1: warning: no previous prototype for 'lpfc_set_features' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_sli.c:8972:1: warning: no previous prototype for 'lpfc_sli_calc_ring' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4621:1: warning: no previous prototype for 'lpfc_rdp_res_link_service' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4633:1: warning: no previous prototype for 'lpfc_rdp_res_sfp_desc' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4698:1: warning: no previous prototype for 'lpfc_rdp_res_link_error' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4727:1: warning: no previous prototype for 'lpfc_rdp_res_bbc_desc' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4752:1: warning: no previous prototype for 'lpfc_rdp_res_oed_temp_desc' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4780:1: warning: no previous prototype for 'lpfc_rdp_res_oed_voltage_desc' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4809:1: warning: no previous prototype for 'lpfc_rdp_res_oed_txbias_desc' [-Wmissing-prototypes] drivers/scsi/lpfc/lpfc_els.c:4838:1: warning: no previous prototype for 'lpfc_rdp_res_oed_txpower_desc' [-Wmissing-prototypes] .... In fact, these functions are only used in the file in which they are declared and don't need a declaration, but can be made static. So this patch marks these functions with 'static'. Signed-off-by: Baoyou Xie <baoyou.xie@linaro.org> Acked-by: Arnd Bergmann <arnd@arndb.de> Acked-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-09-24 23:44:55 -06:00
static void
lpfc_set_features(struct lpfc_hba *phba, LPFC_MBOXQ_t *mbox,
uint32_t feature)
{
uint32_t len;
len = sizeof(struct lpfc_mbx_set_feature) -
sizeof(struct lpfc_sli4_cfg_mhdr);
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_SET_FEATURES, len,
LPFC_SLI4_MBX_EMBED);
switch (feature) {
case LPFC_SET_UE_RECOVERY:
bf_set(lpfc_mbx_set_feature_UER,
&mbox->u.mqe.un.set_feature, 1);
mbox->u.mqe.un.set_feature.feature = LPFC_SET_UE_RECOVERY;
mbox->u.mqe.un.set_feature.param_len = 8;
break;
case LPFC_SET_MDS_DIAGS:
bf_set(lpfc_mbx_set_feature_mds,
&mbox->u.mqe.un.set_feature, 1);
bf_set(lpfc_mbx_set_feature_mds_deep_loopbk,
&mbox->u.mqe.un.set_feature, 1);
mbox->u.mqe.un.set_feature.feature = LPFC_SET_MDS_DIAGS;
mbox->u.mqe.un.set_feature.param_len = 8;
break;
}
return;
}
/**
* lpfc_sli4_alloc_resource_identifiers - Allocate all SLI4 resource extents.
* @phba: Pointer to HBA context object.
*
* This function allocates all SLI4 resource identifiers.
**/
int
lpfc_sli4_alloc_resource_identifiers(struct lpfc_hba *phba)
{
int i, rc, error = 0;
uint16_t count, base;
unsigned long longs;
if (!phba->sli4_hba.rpi_hdrs_in_use)
phba->sli4_hba.next_rpi = phba->sli4_hba.max_cfg_param.max_rpi;
if (phba->sli4_hba.extents_in_use) {
/*
* The port supports resource extents. The XRI, VPI, VFI, RPI
* resource extent count must be read and allocated before
* provisioning the resource id arrays.
*/
if (bf_get(lpfc_idx_rsrc_rdy, &phba->sli4_hba.sli4_flags) ==
LPFC_IDX_RSRC_RDY) {
/*
* Extent-based resources are set - the driver could
* be in a port reset. Figure out if any corrective
* actions need to be taken.
*/
rc = lpfc_sli4_chk_avail_extnt_rsrc(phba,
LPFC_RSC_TYPE_FCOE_VFI);
if (rc != 0)
error++;
rc = lpfc_sli4_chk_avail_extnt_rsrc(phba,
LPFC_RSC_TYPE_FCOE_VPI);
if (rc != 0)
error++;
rc = lpfc_sli4_chk_avail_extnt_rsrc(phba,
LPFC_RSC_TYPE_FCOE_XRI);
if (rc != 0)
error++;
rc = lpfc_sli4_chk_avail_extnt_rsrc(phba,
LPFC_RSC_TYPE_FCOE_RPI);
if (rc != 0)
error++;
/*
* It's possible that the number of resources
* provided to this port instance changed between
* resets. Detect this condition and reallocate
* resources. Otherwise, there is no action.
*/
if (error) {
lpfc_printf_log(phba, KERN_INFO,
LOG_MBOX | LOG_INIT,
"2931 Detected extent resource "
"change. Reallocating all "
"extents.\n");
rc = lpfc_sli4_dealloc_extent(phba,
LPFC_RSC_TYPE_FCOE_VFI);
rc = lpfc_sli4_dealloc_extent(phba,
LPFC_RSC_TYPE_FCOE_VPI);
rc = lpfc_sli4_dealloc_extent(phba,
LPFC_RSC_TYPE_FCOE_XRI);
rc = lpfc_sli4_dealloc_extent(phba,
LPFC_RSC_TYPE_FCOE_RPI);
} else
return 0;
}
rc = lpfc_sli4_alloc_extent(phba, LPFC_RSC_TYPE_FCOE_VFI);
if (unlikely(rc))
goto err_exit;
rc = lpfc_sli4_alloc_extent(phba, LPFC_RSC_TYPE_FCOE_VPI);
if (unlikely(rc))
goto err_exit;
rc = lpfc_sli4_alloc_extent(phba, LPFC_RSC_TYPE_FCOE_RPI);
if (unlikely(rc))
goto err_exit;
rc = lpfc_sli4_alloc_extent(phba, LPFC_RSC_TYPE_FCOE_XRI);
if (unlikely(rc))
goto err_exit;
bf_set(lpfc_idx_rsrc_rdy, &phba->sli4_hba.sli4_flags,
LPFC_IDX_RSRC_RDY);
return rc;
} else {
/*
* The port does not support resource extents. The XRI, VPI,
* VFI, RPI resource ids were determined from READ_CONFIG.
* Just allocate the bitmasks and provision the resource id
* arrays. If a port reset is active, the resources don't
* need any action - just exit.
*/
if (bf_get(lpfc_idx_rsrc_rdy, &phba->sli4_hba.sli4_flags) ==
LPFC_IDX_RSRC_RDY) {
lpfc_sli4_dealloc_resource_identifiers(phba);
lpfc_sli4_remove_rpis(phba);
}
/* RPIs. */
count = phba->sli4_hba.max_cfg_param.max_rpi;
if (count <= 0) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3279 Invalid provisioning of "
"rpi:%d\n", count);
rc = -EINVAL;
goto err_exit;
}
base = phba->sli4_hba.max_cfg_param.rpi_base;
longs = (count + BITS_PER_LONG - 1) / BITS_PER_LONG;
phba->sli4_hba.rpi_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.rpi_bmask)) {
rc = -ENOMEM;
goto err_exit;
}
phba->sli4_hba.rpi_ids = kzalloc(count *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.rpi_ids)) {
rc = -ENOMEM;
goto free_rpi_bmask;
}
for (i = 0; i < count; i++)
phba->sli4_hba.rpi_ids[i] = base + i;
/* VPIs. */
count = phba->sli4_hba.max_cfg_param.max_vpi;
if (count <= 0) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3280 Invalid provisioning of "
"vpi:%d\n", count);
rc = -EINVAL;
goto free_rpi_ids;
}
base = phba->sli4_hba.max_cfg_param.vpi_base;
longs = (count + BITS_PER_LONG - 1) / BITS_PER_LONG;
phba->vpi_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->vpi_bmask)) {
rc = -ENOMEM;
goto free_rpi_ids;
}
phba->vpi_ids = kzalloc(count *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->vpi_ids)) {
rc = -ENOMEM;
goto free_vpi_bmask;
}
for (i = 0; i < count; i++)
phba->vpi_ids[i] = base + i;
/* XRIs. */
count = phba->sli4_hba.max_cfg_param.max_xri;
if (count <= 0) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3281 Invalid provisioning of "
"xri:%d\n", count);
rc = -EINVAL;
goto free_vpi_ids;
}
base = phba->sli4_hba.max_cfg_param.xri_base;
longs = (count + BITS_PER_LONG - 1) / BITS_PER_LONG;
phba->sli4_hba.xri_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.xri_bmask)) {
rc = -ENOMEM;
goto free_vpi_ids;
}
phba->sli4_hba.max_cfg_param.xri_used = 0;
phba->sli4_hba.xri_ids = kzalloc(count *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.xri_ids)) {
rc = -ENOMEM;
goto free_xri_bmask;
}
for (i = 0; i < count; i++)
phba->sli4_hba.xri_ids[i] = base + i;
/* VFIs. */
count = phba->sli4_hba.max_cfg_param.max_vfi;
if (count <= 0) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3282 Invalid provisioning of "
"vfi:%d\n", count);
rc = -EINVAL;
goto free_xri_ids;
}
base = phba->sli4_hba.max_cfg_param.vfi_base;
longs = (count + BITS_PER_LONG - 1) / BITS_PER_LONG;
phba->sli4_hba.vfi_bmask = kzalloc(longs *
sizeof(unsigned long),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.vfi_bmask)) {
rc = -ENOMEM;
goto free_xri_ids;
}
phba->sli4_hba.vfi_ids = kzalloc(count *
sizeof(uint16_t),
GFP_KERNEL);
if (unlikely(!phba->sli4_hba.vfi_ids)) {
rc = -ENOMEM;
goto free_vfi_bmask;
}
for (i = 0; i < count; i++)
phba->sli4_hba.vfi_ids[i] = base + i;
/*
* Mark all resources ready. An HBA reset doesn't need
* to reset the initialization.
*/
bf_set(lpfc_idx_rsrc_rdy, &phba->sli4_hba.sli4_flags,
LPFC_IDX_RSRC_RDY);
return 0;
}
free_vfi_bmask:
kfree(phba->sli4_hba.vfi_bmask);
phba->sli4_hba.vfi_bmask = NULL;
free_xri_ids:
kfree(phba->sli4_hba.xri_ids);
phba->sli4_hba.xri_ids = NULL;
free_xri_bmask:
kfree(phba->sli4_hba.xri_bmask);
phba->sli4_hba.xri_bmask = NULL;
free_vpi_ids:
kfree(phba->vpi_ids);
phba->vpi_ids = NULL;
free_vpi_bmask:
kfree(phba->vpi_bmask);
phba->vpi_bmask = NULL;
free_rpi_ids:
kfree(phba->sli4_hba.rpi_ids);
phba->sli4_hba.rpi_ids = NULL;
free_rpi_bmask:
kfree(phba->sli4_hba.rpi_bmask);
phba->sli4_hba.rpi_bmask = NULL;
err_exit:
return rc;
}
/**
* lpfc_sli4_dealloc_resource_identifiers - Deallocate all SLI4 resource extents.
* @phba: Pointer to HBA context object.
*
* This function allocates the number of elements for the specified
* resource type.
**/
int
lpfc_sli4_dealloc_resource_identifiers(struct lpfc_hba *phba)
{
if (phba->sli4_hba.extents_in_use) {
lpfc_sli4_dealloc_extent(phba, LPFC_RSC_TYPE_FCOE_VPI);
lpfc_sli4_dealloc_extent(phba, LPFC_RSC_TYPE_FCOE_RPI);
lpfc_sli4_dealloc_extent(phba, LPFC_RSC_TYPE_FCOE_XRI);
lpfc_sli4_dealloc_extent(phba, LPFC_RSC_TYPE_FCOE_VFI);
} else {
kfree(phba->vpi_bmask);
phba->sli4_hba.max_cfg_param.vpi_used = 0;
kfree(phba->vpi_ids);
bf_set(lpfc_vpi_rsrc_rdy, &phba->sli4_hba.sli4_flags, 0);
kfree(phba->sli4_hba.xri_bmask);
kfree(phba->sli4_hba.xri_ids);
kfree(phba->sli4_hba.vfi_bmask);
kfree(phba->sli4_hba.vfi_ids);
bf_set(lpfc_vfi_rsrc_rdy, &phba->sli4_hba.sli4_flags, 0);
bf_set(lpfc_idx_rsrc_rdy, &phba->sli4_hba.sli4_flags, 0);
}
return 0;
}
/**
* lpfc_sli4_get_allocated_extnts - Get the port's allocated extents.
* @phba: Pointer to HBA context object.
* @type: The resource extent type.
* @extnt_count: buffer to hold port extent count response
* @extnt_size: buffer to hold port extent size response.
*
* This function calls the port to read the host allocated extents
* for a particular type.
**/
int
lpfc_sli4_get_allocated_extnts(struct lpfc_hba *phba, uint16_t type,
uint16_t *extnt_cnt, uint16_t *extnt_size)
{
bool emb;
int rc = 0;
uint16_t curr_blks = 0;
uint32_t req_len, emb_len;
uint32_t alloc_len, mbox_tmo;
struct list_head *blk_list_head;
struct lpfc_rsrc_blks *rsrc_blk;
LPFC_MBOXQ_t *mbox;
void *virtaddr = NULL;
struct lpfc_mbx_nembed_rsrc_extent *n_rsrc;
struct lpfc_mbx_alloc_rsrc_extents *rsrc_ext;
union lpfc_sli4_cfg_shdr *shdr;
switch (type) {
case LPFC_RSC_TYPE_FCOE_VPI:
blk_list_head = &phba->lpfc_vpi_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_XRI:
blk_list_head = &phba->sli4_hba.lpfc_xri_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_VFI:
blk_list_head = &phba->sli4_hba.lpfc_vfi_blk_list;
break;
case LPFC_RSC_TYPE_FCOE_RPI:
blk_list_head = &phba->sli4_hba.lpfc_rpi_blk_list;
break;
default:
return -EIO;
}
/* Count the number of extents currently allocatd for this type. */
list_for_each_entry(rsrc_blk, blk_list_head, list) {
if (curr_blks == 0) {
/*
* The GET_ALLOCATED mailbox does not return the size,
* just the count. The size should be just the size
* stored in the current allocated block and all sizes
* for an extent type are the same so set the return
* value now.
*/
*extnt_size = rsrc_blk->rsrc_size;
}
curr_blks++;
}
/*
* Calculate the size of an embedded mailbox. The uint32_t
* accounts for extents-specific word.
*/
emb_len = sizeof(MAILBOX_t) - sizeof(struct mbox_header) -
sizeof(uint32_t);
/*
* Presume the allocation and response will fit into an embedded
* mailbox. If not true, reconfigure to a non-embedded mailbox.
*/
emb = LPFC_SLI4_MBX_EMBED;
req_len = emb_len;
if (req_len > emb_len) {
req_len = curr_blks * sizeof(uint16_t) +
sizeof(union lpfc_sli4_cfg_shdr) +
sizeof(uint32_t);
emb = LPFC_SLI4_MBX_NEMBED;
}
mbox = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
memset(mbox, 0, sizeof(LPFC_MBOXQ_t));
alloc_len = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_GET_ALLOC_RSRC_EXTENT,
req_len, emb);
if (alloc_len < req_len) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2983 Allocated DMA memory size (x%x) is "
"less than the requested DMA memory "
"size (x%x)\n", alloc_len, req_len);
rc = -ENOMEM;
goto err_exit;
}
rc = lpfc_sli4_mbox_rsrc_extent(phba, mbox, curr_blks, type, emb);
if (unlikely(rc)) {
rc = -EIO;
goto err_exit;
}
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
if (unlikely(rc)) {
rc = -EIO;
goto err_exit;
}
/*
* Figure out where the response is located. Then get local pointers
* to the response data. The port does not guarantee to respond to
* all extents counts request so update the local variable with the
* allocated count from the port.
*/
if (emb == LPFC_SLI4_MBX_EMBED) {
rsrc_ext = &mbox->u.mqe.un.alloc_rsrc_extents;
shdr = &rsrc_ext->header.cfg_shdr;
*extnt_cnt = bf_get(lpfc_mbx_rsrc_cnt, &rsrc_ext->u.rsp);
} else {
virtaddr = mbox->sge_array->addr[0];
n_rsrc = (struct lpfc_mbx_nembed_rsrc_extent *) virtaddr;
shdr = &n_rsrc->cfg_shdr;
*extnt_cnt = bf_get(lpfc_mbx_rsrc_cnt, n_rsrc);
}
if (bf_get(lpfc_mbox_hdr_status, &shdr->response)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_INIT,
"2984 Failed to read allocated resources "
"for type %d - Status 0x%x Add'l Status 0x%x.\n",
type,
bf_get(lpfc_mbox_hdr_status, &shdr->response),
bf_get(lpfc_mbox_hdr_add_status, &shdr->response));
rc = -EIO;
goto err_exit;
}
err_exit:
lpfc_sli4_mbox_cmd_free(phba, mbox);
return rc;
}
/**
* lpfc_sli4_repost_sgl_list - Repost the buffers sgl pages as block
* @phba: pointer to lpfc hba data structure.
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* @pring: Pointer to driver SLI ring object.
* @sgl_list: linked link of sgl buffers to post
* @cnt: number of linked list buffers
*
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* This routine walks the list of buffers that have been allocated and
* repost them to the port by using SGL block post. This is needed after a
* pci_function_reset/warm_start or start. It attempts to construct blocks
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* of buffer sgls which contains contiguous xris and uses the non-embedded
* SGL block post mailbox commands to post them to the port. For single
* buffer sgl with non-contiguous xri, if any, it shall use embedded SGL post
* mailbox command for posting.
*
* Returns: 0 = success, non-zero failure.
**/
static int
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli4_repost_sgl_list(struct lpfc_hba *phba,
struct list_head *sgl_list, int cnt)
{
struct lpfc_sglq *sglq_entry = NULL;
struct lpfc_sglq *sglq_entry_next = NULL;
struct lpfc_sglq *sglq_entry_first = NULL;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int status, total_cnt;
int post_cnt = 0, num_posted = 0, block_cnt = 0;
int last_xritag = NO_XRI;
LIST_HEAD(prep_sgl_list);
LIST_HEAD(blck_sgl_list);
LIST_HEAD(allc_sgl_list);
LIST_HEAD(post_sgl_list);
LIST_HEAD(free_sgl_list);
spin_lock_irq(&phba->hbalock);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_lock(&phba->sli4_hba.sgl_list_lock);
list_splice_init(sgl_list, &allc_sgl_list);
spin_unlock(&phba->sli4_hba.sgl_list_lock);
spin_unlock_irq(&phba->hbalock);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
total_cnt = cnt;
list_for_each_entry_safe(sglq_entry, sglq_entry_next,
&allc_sgl_list, list) {
list_del_init(&sglq_entry->list);
block_cnt++;
if ((last_xritag != NO_XRI) &&
(sglq_entry->sli4_xritag != last_xritag + 1)) {
/* a hole in xri block, form a sgl posting block */
list_splice_init(&prep_sgl_list, &blck_sgl_list);
post_cnt = block_cnt - 1;
/* prepare list for next posting block */
list_add_tail(&sglq_entry->list, &prep_sgl_list);
block_cnt = 1;
} else {
/* prepare list for next posting block */
list_add_tail(&sglq_entry->list, &prep_sgl_list);
/* enough sgls for non-embed sgl mbox command */
if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) {
list_splice_init(&prep_sgl_list,
&blck_sgl_list);
post_cnt = block_cnt;
block_cnt = 0;
}
}
num_posted++;
/* keep track of last sgl's xritag */
last_xritag = sglq_entry->sli4_xritag;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* end of repost sgl list condition for buffers */
if (num_posted == total_cnt) {
if (post_cnt == 0) {
list_splice_init(&prep_sgl_list,
&blck_sgl_list);
post_cnt = block_cnt;
} else if (block_cnt == 1) {
status = lpfc_sli4_post_sgl(phba,
sglq_entry->phys, 0,
sglq_entry->sli4_xritag);
if (!status) {
/* successful, put sgl to posted list */
list_add_tail(&sglq_entry->list,
&post_sgl_list);
} else {
/* Failure, put sgl to free list */
lpfc_printf_log(phba, KERN_WARNING,
LOG_SLI,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
"3159 Failed to post "
"sgl, xritag:x%x\n",
sglq_entry->sli4_xritag);
list_add_tail(&sglq_entry->list,
&free_sgl_list);
total_cnt--;
}
}
}
/* continue until a nembed page worth of sgls */
if (post_cnt == 0)
continue;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* post the buffer list sgls as a block */
status = lpfc_sli4_post_sgl_list(phba, &blck_sgl_list,
post_cnt);
if (!status) {
/* success, put sgl list to posted sgl list */
list_splice_init(&blck_sgl_list, &post_sgl_list);
} else {
/* Failure, put sgl list to free sgl list */
sglq_entry_first = list_first_entry(&blck_sgl_list,
struct lpfc_sglq,
list);
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
"3160 Failed to post sgl-list, "
"xritag:x%x-x%x\n",
sglq_entry_first->sli4_xritag,
(sglq_entry_first->sli4_xritag +
post_cnt - 1));
list_splice_init(&blck_sgl_list, &free_sgl_list);
total_cnt -= post_cnt;
}
/* don't reset xirtag due to hole in xri block */
if (block_cnt == 0)
last_xritag = NO_XRI;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* reset sgl post count for next round of posting */
post_cnt = 0;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* free the sgls failed to post */
lpfc_free_sgl_list(phba, &free_sgl_list);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* push sgls posted to the available list */
if (!list_empty(&post_sgl_list)) {
spin_lock_irq(&phba->hbalock);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_lock(&phba->sli4_hba.sgl_list_lock);
list_splice_init(&post_sgl_list, sgl_list);
spin_unlock(&phba->sli4_hba.sgl_list_lock);
spin_unlock_irq(&phba->hbalock);
} else {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
"3161 Failure to post sgl to port.\n");
return -EIO;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* return the number of XRIs actually posted */
return total_cnt;
}
void
lpfc_set_host_data(struct lpfc_hba *phba, LPFC_MBOXQ_t *mbox)
{
uint32_t len;
len = sizeof(struct lpfc_mbx_set_host_data) -
sizeof(struct lpfc_sli4_cfg_mhdr);
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_SET_HOST_DATA, len,
LPFC_SLI4_MBX_EMBED);
mbox->u.mqe.un.set_host_data.param_id = LPFC_SET_HOST_OS_DRIVER_VERSION;
mbox->u.mqe.un.set_host_data.param_len =
LPFC_HOST_OS_DRIVER_VERSION_SIZE;
snprintf(mbox->u.mqe.un.set_host_data.data,
LPFC_HOST_OS_DRIVER_VERSION_SIZE,
"Linux %s v"LPFC_DRIVER_VERSION,
(phba->hba_flag & HBA_FCOE_MODE) ? "FCoE" : "FC");
}
int
lpfc_post_rq_buffer(struct lpfc_hba *phba, struct lpfc_queue *hrq,
struct lpfc_queue *drq, int count, int idx)
{
int rc, i;
struct lpfc_rqe hrqe;
struct lpfc_rqe drqe;
struct lpfc_rqb *rqbp;
unsigned long flags;
struct rqb_dmabuf *rqb_buffer;
LIST_HEAD(rqb_buf_list);
spin_lock_irqsave(&phba->hbalock, flags);
rqbp = hrq->rqbp;
for (i = 0; i < count; i++) {
/* IF RQ is already full, don't bother */
if (rqbp->buffer_count + i >= rqbp->entry_count - 1)
break;
rqb_buffer = rqbp->rqb_alloc_buffer(phba);
if (!rqb_buffer)
break;
rqb_buffer->hrq = hrq;
rqb_buffer->drq = drq;
rqb_buffer->idx = idx;
list_add_tail(&rqb_buffer->hbuf.list, &rqb_buf_list);
}
while (!list_empty(&rqb_buf_list)) {
list_remove_head(&rqb_buf_list, rqb_buffer, struct rqb_dmabuf,
hbuf.list);
hrqe.address_lo = putPaddrLow(rqb_buffer->hbuf.phys);
hrqe.address_hi = putPaddrHigh(rqb_buffer->hbuf.phys);
drqe.address_lo = putPaddrLow(rqb_buffer->dbuf.phys);
drqe.address_hi = putPaddrHigh(rqb_buffer->dbuf.phys);
rc = lpfc_sli4_rq_put(hrq, drq, &hrqe, &drqe);
if (rc < 0) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"6421 Cannot post to HRQ %d: %x %x %x "
"DRQ %x %x\n",
hrq->queue_id,
hrq->host_index,
hrq->hba_index,
hrq->entry_count,
drq->host_index,
drq->hba_index);
rqbp->rqb_free_buffer(phba, rqb_buffer);
} else {
list_add_tail(&rqb_buffer->hbuf.list,
&rqbp->rqb_buffer_list);
rqbp->buffer_count++;
}
}
spin_unlock_irqrestore(&phba->hbalock, flags);
return 1;
}
/**
* lpfc_sli4_hba_setup - SLI4 device initialization PCI function
* @phba: Pointer to HBA context object.
*
* This function is the main SLI4 device initialization PCI function. This
* function is called by the HBA initialization code, HBA reset code and
* HBA error attention handler code. Caller is not required to hold any
* locks.
**/
int
lpfc_sli4_hba_setup(struct lpfc_hba *phba)
{
int rc, i, cnt;
LPFC_MBOXQ_t *mboxq;
struct lpfc_mqe *mqe;
uint8_t *vpd;
uint32_t vpd_size;
uint32_t ftr_rsp = 0;
struct Scsi_Host *shost = lpfc_shost_from_vport(phba->pport);
struct lpfc_vport *vport = phba->pport;
struct lpfc_dmabuf *mp;
struct lpfc_rqb *rqbp;
/* Perform a PCI function reset to start from clean */
rc = lpfc_pci_function_reset(phba);
if (unlikely(rc))
return -ENODEV;
/* Check the HBA Host Status Register for readyness */
rc = lpfc_sli4_post_status_check(phba);
if (unlikely(rc))
return -ENODEV;
else {
spin_lock_irq(&phba->hbalock);
phba->sli.sli_flag |= LPFC_SLI_ACTIVE;
spin_unlock_irq(&phba->hbalock);
}
/*
* Allocate a single mailbox container for initializing the
* port.
*/
mboxq = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq)
return -ENOMEM;
/* Issue READ_REV to collect vpd and FW information. */
vpd_size = SLI4_PAGE_SIZE;
vpd = kzalloc(vpd_size, GFP_KERNEL);
if (!vpd) {
rc = -ENOMEM;
goto out_free_mbox;
}
rc = lpfc_sli4_read_rev(phba, mboxq, vpd, &vpd_size);
2010-11-20 21:11:48 -07:00
if (unlikely(rc)) {
kfree(vpd);
goto out_free_mbox;
}
mqe = &mboxq->u.mqe;
phba->sli_rev = bf_get(lpfc_mbx_rd_rev_sli_lvl, &mqe->un.read_rev);
if (bf_get(lpfc_mbx_rd_rev_fcoe, &mqe->un.read_rev)) {
2010-11-20 21:11:48 -07:00
phba->hba_flag |= HBA_FCOE_MODE;
phba->fcp_embed_io = 0; /* SLI4 FC support only */
} else {
2010-11-20 21:11:48 -07:00
phba->hba_flag &= ~HBA_FCOE_MODE;
}
if (bf_get(lpfc_mbx_rd_rev_cee_ver, &mqe->un.read_rev) ==
LPFC_DCBX_CEE_MODE)
phba->hba_flag |= HBA_FIP_SUPPORT;
else
phba->hba_flag &= ~HBA_FIP_SUPPORT;
phba->hba_flag &= ~HBA_FCP_IOQ_FLUSH;
if (phba->sli_rev != LPFC_SLI_REV4) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0376 READ_REV Error. SLI Level %d "
"FCoE enabled %d\n",
2010-11-20 21:11:48 -07:00
phba->sli_rev, phba->hba_flag & HBA_FCOE_MODE);
rc = -EIO;
2010-11-20 21:11:48 -07:00
kfree(vpd);
goto out_free_mbox;
}
/*
* Continue initialization with default values even if driver failed
* to read FCoE param config regions, only read parameters if the
* board is FCoE
*/
if (phba->hba_flag & HBA_FCOE_MODE &&
lpfc_sli4_read_fcoe_params(phba))
lpfc_printf_log(phba, KERN_WARNING, LOG_MBOX | LOG_INIT,
"2570 Failed to read FCoE parameters\n");
/*
* Retrieve sli4 device physical port name, failure of doing it
* is considered as non-fatal.
*/
rc = lpfc_sli4_retrieve_pport_name(phba);
if (!rc)
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"3080 Successful retrieving SLI4 device "
"physical port name: %s.\n", phba->Port);
/*
* Evaluate the read rev and vpd data. Populate the driver
* state with the results. If this routine fails, the failure
* is not fatal as the driver will use generic values.
*/
rc = lpfc_parse_vpd(phba, vpd, vpd_size);
if (unlikely(!rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0377 Error %d parsing vpd. "
"Using defaults.\n", rc);
rc = 0;
}
2010-11-20 21:11:48 -07:00
kfree(vpd);
/* Save information as VPD data */
phba->vpd.rev.biuRev = mqe->un.read_rev.first_hw_rev;
phba->vpd.rev.smRev = mqe->un.read_rev.second_hw_rev;
/*
* This is because first G7 ASIC doesn't support the standard
* 0x5a NVME cmd descriptor type/subtype
*/
if ((bf_get(lpfc_sli_intf_if_type, &phba->sli4_hba.sli_intf) ==
LPFC_SLI_INTF_IF_TYPE_6) &&
(phba->vpd.rev.biuRev == LPFC_G7_ASIC_1) &&
(phba->vpd.rev.smRev == 0) &&
(phba->cfg_nvme_embed_cmd == 1))
phba->cfg_nvme_embed_cmd = 0;
phba->vpd.rev.endecRev = mqe->un.read_rev.third_hw_rev;
phba->vpd.rev.fcphHigh = bf_get(lpfc_mbx_rd_rev_fcph_high,
&mqe->un.read_rev);
phba->vpd.rev.fcphLow = bf_get(lpfc_mbx_rd_rev_fcph_low,
&mqe->un.read_rev);
phba->vpd.rev.feaLevelHigh = bf_get(lpfc_mbx_rd_rev_ftr_lvl_high,
&mqe->un.read_rev);
phba->vpd.rev.feaLevelLow = bf_get(lpfc_mbx_rd_rev_ftr_lvl_low,
&mqe->un.read_rev);
phba->vpd.rev.sli1FwRev = mqe->un.read_rev.fw_id_rev;
memcpy(phba->vpd.rev.sli1FwName, mqe->un.read_rev.fw_name, 16);
phba->vpd.rev.sli2FwRev = mqe->un.read_rev.ulp_fw_id_rev;
memcpy(phba->vpd.rev.sli2FwName, mqe->un.read_rev.ulp_fw_name, 16);
phba->vpd.rev.opFwRev = mqe->un.read_rev.fw_id_rev;
memcpy(phba->vpd.rev.opFwName, mqe->un.read_rev.fw_name, 16);
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):0380 READ_REV Status x%x "
"fw_rev:%s fcphHi:%x fcphLo:%x flHi:%x flLo:%x\n",
mboxq->vport ? mboxq->vport->vpi : 0,
bf_get(lpfc_mqe_status, mqe),
phba->vpd.rev.opFwName,
phba->vpd.rev.fcphHigh, phba->vpd.rev.fcphLow,
phba->vpd.rev.feaLevelHigh, phba->vpd.rev.feaLevelLow);
/* Reset the DFT_LUN_Q_DEPTH to (max xri >> 3) */
rc = (phba->sli4_hba.max_cfg_param.max_xri >> 3);
if (phba->pport->cfg_lun_queue_depth > rc) {
lpfc_printf_log(phba, KERN_WARNING, LOG_INIT,
"3362 LUN queue depth changed from %d to %d\n",
phba->pport->cfg_lun_queue_depth, rc);
phba->pport->cfg_lun_queue_depth = rc;
}
if (bf_get(lpfc_sli_intf_if_type, &phba->sli4_hba.sli_intf) ==
LPFC_SLI_INTF_IF_TYPE_0) {
lpfc_set_features(phba, mboxq, LPFC_SET_UE_RECOVERY);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc == MBX_SUCCESS) {
phba->hba_flag |= HBA_RECOVERABLE_UE;
/* Set 1Sec interval to detect UE */
phba->eratt_poll_interval = 1;
phba->sli4_hba.ue_to_sr = bf_get(
lpfc_mbx_set_feature_UESR,
&mboxq->u.mqe.un.set_feature);
phba->sli4_hba.ue_to_rp = bf_get(
lpfc_mbx_set_feature_UERP,
&mboxq->u.mqe.un.set_feature);
}
}
if (phba->cfg_enable_mds_diags && phba->mds_diags_support) {
/* Enable MDS Diagnostics only if the SLI Port supports it */
lpfc_set_features(phba, mboxq, LPFC_SET_MDS_DIAGS);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc != MBX_SUCCESS)
phba->mds_diags_support = 0;
}
/*
* Discover the port's supported feature set and match it against the
* hosts requests.
*/
lpfc_request_features(phba, mboxq);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (unlikely(rc)) {
rc = -EIO;
2010-11-20 21:11:48 -07:00
goto out_free_mbox;
}
/*
* The port must support FCP initiator mode as this is the
* only mode running in the host.
*/
if (!(bf_get(lpfc_mbx_rq_ftr_rsp_fcpi, &mqe->un.req_ftrs))) {
lpfc_printf_log(phba, KERN_WARNING, LOG_MBOX | LOG_SLI,
"0378 No support for fcpi mode.\n");
ftr_rsp++;
}
/* Performance Hints are ONLY for FCoE */
if (phba->hba_flag & HBA_FCOE_MODE) {
if (bf_get(lpfc_mbx_rq_ftr_rsp_perfh, &mqe->un.req_ftrs))
phba->sli3_options |= LPFC_SLI4_PERFH_ENABLED;
else
phba->sli3_options &= ~LPFC_SLI4_PERFH_ENABLED;
}
/*
* If the port cannot support the host's requested features
* then turn off the global config parameters to disable the
* feature in the driver. This is not a fatal error.
*/
if (phba->sli3_options & LPFC_SLI3_BG_ENABLED) {
if (!(bf_get(lpfc_mbx_rq_ftr_rsp_dif, &mqe->un.req_ftrs))) {
phba->cfg_enable_bg = 0;
phba->sli3_options &= ~LPFC_SLI3_BG_ENABLED;
ftr_rsp++;
}
}
if (phba->max_vpi && phba->cfg_enable_npiv &&
!(bf_get(lpfc_mbx_rq_ftr_rsp_npiv, &mqe->un.req_ftrs)))
ftr_rsp++;
if (ftr_rsp) {
lpfc_printf_log(phba, KERN_WARNING, LOG_MBOX | LOG_SLI,
"0379 Feature Mismatch Data: x%08x %08x "
"x%x x%x x%x\n", mqe->un.req_ftrs.word2,
mqe->un.req_ftrs.word3, phba->cfg_enable_bg,
phba->cfg_enable_npiv, phba->max_vpi);
if (!(bf_get(lpfc_mbx_rq_ftr_rsp_dif, &mqe->un.req_ftrs)))
phba->cfg_enable_bg = 0;
if (!(bf_get(lpfc_mbx_rq_ftr_rsp_npiv, &mqe->un.req_ftrs)))
phba->cfg_enable_npiv = 0;
}
/* These SLI3 features are assumed in SLI4 */
spin_lock_irq(&phba->hbalock);
phba->sli3_options |= (LPFC_SLI3_NPIV_ENABLED | LPFC_SLI3_HBQ_ENABLED);
spin_unlock_irq(&phba->hbalock);
/*
* Allocate all resources (xri,rpi,vpi,vfi) now. Subsequent
* calls depends on these resources to complete port setup.
*/
rc = lpfc_sli4_alloc_resource_identifiers(phba);
if (rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"2920 Failed to alloc Resource IDs "
"rc = x%x\n", rc);
goto out_free_mbox;
}
lpfc_set_host_data(phba, mboxq);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc) {
lpfc_printf_log(phba, KERN_WARNING, LOG_MBOX | LOG_SLI,
"2134 Failed to set host os driver version %x",
rc);
}
/* Read the port's service parameters. */
rc = lpfc_read_sparam(phba, mboxq, vport->vpi);
if (rc) {
phba->link_state = LPFC_HBA_ERROR;
rc = -ENOMEM;
2010-11-20 21:11:48 -07:00
goto out_free_mbox;
}
mboxq->vport = vport;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
mp = (struct lpfc_dmabuf *) mboxq->context1;
if (rc == MBX_SUCCESS) {
memcpy(&vport->fc_sparam, mp->virt, sizeof(struct serv_parm));
rc = 0;
}
/*
* This memory was allocated by the lpfc_read_sparam routine. Release
* it to the mbuf pool.
*/
lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
mboxq->context1 = NULL;
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0382 READ_SPARAM command failed "
"status %d, mbxStatus x%x\n",
rc, bf_get(lpfc_mqe_status, mqe));
phba->link_state = LPFC_HBA_ERROR;
rc = -EIO;
2010-11-20 21:11:48 -07:00
goto out_free_mbox;
}
lpfc_update_vport_wwn(vport);
/* Update the fc_host data structures with new wwn. */
fc_host_node_name(shost) = wwn_to_u64(vport->fc_nodename.u.wwn);
fc_host_port_name(shost) = wwn_to_u64(vport->fc_portname.u.wwn);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* Create all the SLI4 queues */
rc = lpfc_sli4_queue_create(phba);
if (rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3089 Failed to allocate queues\n");
rc = -ENODEV;
goto out_free_mbox;
}
/* Set up all the queues to the device */
rc = lpfc_sli4_queue_setup(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0381 Error %d during queue setup.\n ", rc);
goto out_stop_timers;
}
/* Initialize the driver internal SLI layer lists. */
lpfc_sli4_setup(phba);
lpfc_sli4_queue_init(phba);
/* update host els xri-sgl sizes and mappings */
rc = lpfc_sli4_els_sgl_update(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"1400 Failed to update xri-sgl size and "
"mapping: %d\n", rc);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
goto out_destroy_queue;
}
/* register the els sgl pool to the port */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
rc = lpfc_sli4_repost_sgl_list(phba, &phba->sli4_hba.lpfc_els_sgl_list,
phba->sli4_hba.els_xri_cnt);
if (unlikely(rc < 0)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0582 Error %d during els sgl post "
"operation\n", rc);
rc = -ENODEV;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
goto out_destroy_queue;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
phba->sli4_hba.els_xri_cnt = rc;
if (phba->nvmet_support) {
/* update host nvmet xri-sgl sizes and mappings */
rc = lpfc_sli4_nvmet_sgl_update(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"6308 Failed to update nvmet-sgl size "
"and mapping: %d\n", rc);
goto out_destroy_queue;
}
/* register the nvmet sgl pool to the port */
rc = lpfc_sli4_repost_sgl_list(
phba,
&phba->sli4_hba.lpfc_nvmet_sgl_list,
phba->sli4_hba.nvmet_xri_cnt);
if (unlikely(rc < 0)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"3117 Error %d during nvmet "
"sgl post\n", rc);
rc = -ENODEV;
goto out_destroy_queue;
}
phba->sli4_hba.nvmet_xri_cnt = rc;
cnt = phba->cfg_iocb_cnt * 1024;
/* We need 1 iocbq for every SGL, for IO processing */
cnt += phba->sli4_hba.nvmet_xri_cnt;
} else {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* update host scsi xri-sgl sizes and mappings */
rc = lpfc_sli4_scsi_sgl_update(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"6309 Failed to update scsi-sgl size "
"and mapping: %d\n", rc);
goto out_destroy_queue;
}
/* update host nvme xri-sgl sizes and mappings */
rc = lpfc_sli4_nvme_sgl_update(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"6082 Failed to update nvme-sgl size "
"and mapping: %d\n", rc);
goto out_destroy_queue;
}
cnt = phba->cfg_iocb_cnt * 1024;
}
if (!phba->sli.iocbq_lookup) {
/* Initialize and populate the iocb list per host */
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2821 initialize iocb list %d total %d\n",
phba->cfg_iocb_cnt, cnt);
rc = lpfc_init_iocb_list(phba, cnt);
if (rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"1413 Failed to init iocb list.\n");
goto out_destroy_queue;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
}
if (phba->nvmet_support)
lpfc_nvmet_create_targetport(phba);
if (phba->nvmet_support && phba->cfg_nvmet_mrq) {
/* Post initial buffers to all RQs created */
for (i = 0; i < phba->cfg_nvmet_mrq; i++) {
rqbp = phba->sli4_hba.nvmet_mrq_hdr[i]->rqbp;
INIT_LIST_HEAD(&rqbp->rqb_buffer_list);
rqbp->rqb_alloc_buffer = lpfc_sli4_nvmet_alloc;
rqbp->rqb_free_buffer = lpfc_sli4_nvmet_free;
rqbp->entry_count = LPFC_NVMET_RQE_DEF_COUNT;
rqbp->buffer_count = 0;
lpfc_post_rq_buffer(
phba, phba->sli4_hba.nvmet_mrq_hdr[i],
phba->sli4_hba.nvmet_mrq_data[i],
LPFC_NVMET_RQE_DEF_COUNT, i);
}
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->cfg_enable_fc4_type & LPFC_ENABLE_FCP) {
/* register the allocated scsi sgl pool to the port */
rc = lpfc_sli4_repost_scsi_sgl_list(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0383 Error %d during scsi sgl post "
"operation\n", rc);
/* Some Scsi buffers were moved to abort scsi list */
/* A pci function reset will repost them */
rc = -ENODEV;
goto out_destroy_queue;
}
}
if ((phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME) &&
(phba->nvmet_support == 0)) {
/* register the allocated nvme sgl pool to the port */
rc = lpfc_repost_nvme_sgl_list(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"6116 Error %d during nvme sgl post "
"operation\n", rc);
/* Some NVME buffers were moved to abort nvme list */
/* A pci function reset will repost them */
rc = -ENODEV;
goto out_destroy_queue;
}
}
/* Post the rpi header region to the device. */
rc = lpfc_sli4_post_all_rpi_hdrs(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0393 Error %d during rpi post operation\n",
rc);
rc = -ENODEV;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
goto out_destroy_queue;
}
lpfc_sli4_node_prep(phba);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (!(phba->hba_flag & HBA_FCOE_MODE)) {
if ((phba->nvmet_support == 0) || (phba->cfg_nvmet_mrq == 1)) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/*
* The FC Port needs to register FCFI (index 0)
*/
lpfc_reg_fcfi(phba, mboxq);
mboxq->vport = phba->pport;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc != MBX_SUCCESS)
goto out_unset_queue;
rc = 0;
phba->fcf.fcfi = bf_get(lpfc_reg_fcfi_fcfi,
&mboxq->u.mqe.un.reg_fcfi);
} else {
/* We are a NVME Target mode with MRQ > 1 */
/* First register the FCFI */
lpfc_reg_fcfi_mrq(phba, mboxq, 0);
mboxq->vport = phba->pport;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc != MBX_SUCCESS)
goto out_unset_queue;
rc = 0;
phba->fcf.fcfi = bf_get(lpfc_reg_fcfi_mrq_fcfi,
&mboxq->u.mqe.un.reg_fcfi_mrq);
/* Next register the MRQs */
lpfc_reg_fcfi_mrq(phba, mboxq, 1);
mboxq->vport = phba->pport;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc != MBX_SUCCESS)
goto out_unset_queue;
rc = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
}
/* Check if the port is configured to be disabled */
lpfc_sli_read_link_ste(phba);
}
/* Arm the CQs and then EQs on device */
lpfc_sli4_arm_cqeq_intr(phba);
/* Indicate device interrupt mode */
phba->sli4_hba.intr_enable = 1;
/* Allow asynchronous mailbox command to go through */
spin_lock_irq(&phba->hbalock);
phba->sli.sli_flag &= ~LPFC_SLI_ASYNC_MBX_BLK;
spin_unlock_irq(&phba->hbalock);
/* Post receive buffers to the device */
lpfc_sli4_rb_setup(phba);
/* Reset HBA FCF states after HBA reset */
phba->fcf.fcf_flag = 0;
phba->fcf.current_rec.flag = 0;
/* Start the ELS watchdog timer */
mod_timer(&vport->els_tmofunc,
jiffies + msecs_to_jiffies(1000 * (phba->fc_ratov * 2)));
/* Start heart beat timer */
mod_timer(&phba->hb_tmofunc,
jiffies + msecs_to_jiffies(1000 * LPFC_HB_MBOX_INTERVAL));
phba->hb_outstanding = 0;
phba->last_completion_time = jiffies;
/* Start error attention (ERATT) polling timer */
mod_timer(&phba->eratt_poll,
jiffies + msecs_to_jiffies(1000 * phba->eratt_poll_interval));
/* Enable PCIe device Advanced Error Reporting (AER) if configured */
if (phba->cfg_aer_support == 1 && !(phba->hba_flag & HBA_AER_ENABLED)) {
rc = pci_enable_pcie_error_reporting(phba->pcidev);
if (!rc) {
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2829 This device supports "
"Advanced Error Reporting (AER)\n");
spin_lock_irq(&phba->hbalock);
phba->hba_flag |= HBA_AER_ENABLED;
spin_unlock_irq(&phba->hbalock);
} else {
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2830 This device does not support "
"Advanced Error Reporting (AER)\n");
phba->cfg_aer_support = 0;
}
rc = 0;
}
/*
* The port is ready, set the host's link state to LINK_DOWN
* in preparation for link interrupts.
*/
spin_lock_irq(&phba->hbalock);
phba->link_state = LPFC_LINK_DOWN;
spin_unlock_irq(&phba->hbalock);
if (!(phba->hba_flag & HBA_FCOE_MODE) &&
(phba->hba_flag & LINK_DISABLED)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT | LOG_SLI,
"3103 Adapter Link is disabled.\n");
lpfc_down_link(phba, mboxq);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc != MBX_SUCCESS) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT | LOG_SLI,
"3104 Adapter failed to issue "
"DOWN_LINK mbox cmd, rc:x%x\n", rc);
goto out_unset_queue;
}
} else if (phba->cfg_suppress_link_up == LPFC_INITIALIZE_LINK) {
/* don't perform init_link on SLI4 FC port loopback test */
if (!(phba->link_flag & LS_LOOPBACK_MODE)) {
rc = phba->lpfc_hba_init_link(phba, MBX_NOWAIT);
if (rc)
goto out_unset_queue;
}
}
mempool_free(mboxq, phba->mbox_mem_pool);
return rc;
2010-11-20 21:11:48 -07:00
out_unset_queue:
/* Unset all the queues set up in this routine when error out */
lpfc_sli4_queue_unset(phba);
out_destroy_queue:
lpfc_free_iocb_list(phba);
lpfc_sli4_queue_destroy(phba);
out_stop_timers:
lpfc_stop_hba_timers(phba);
out_free_mbox:
mempool_free(mboxq, phba->mbox_mem_pool);
return rc;
}
/**
* lpfc_mbox_timeout - Timeout call back function for mbox timer
* @ptr: context object - pointer to hba structure.
*
* This is the callback function for mailbox timer. The mailbox
* timer is armed when a new mailbox command is issued and the timer
* is deleted when the mailbox complete. The function is called by
* the kernel timer code when a mailbox does not complete within
* expected time. This function wakes up the worker thread to
* process the mailbox timeout and returns. All the processing is
* done by the worker thread function lpfc_mbox_timeout_handler.
**/
void
lpfc_mbox_timeout(struct timer_list *t)
{
struct lpfc_hba *phba = from_timer(phba, t, sli.mbox_tmo);
unsigned long iflag;
uint32_t tmo_posted;
spin_lock_irqsave(&phba->pport->work_port_lock, iflag);
tmo_posted = phba->pport->work_port_events & WORKER_MBOX_TMO;
if (!tmo_posted)
phba->pport->work_port_events |= WORKER_MBOX_TMO;
spin_unlock_irqrestore(&phba->pport->work_port_lock, iflag);
if (!tmo_posted)
lpfc_worker_wake_up(phba);
return;
}
/**
* lpfc_sli4_mbox_completions_pending - check to see if any mailbox completions
* are pending
* @phba: Pointer to HBA context object.
*
* This function checks if any mailbox completions are present on the mailbox
* completion queue.
**/
static bool
lpfc_sli4_mbox_completions_pending(struct lpfc_hba *phba)
{
uint32_t idx;
struct lpfc_queue *mcq;
struct lpfc_mcqe *mcqe;
bool pending_completions = false;
uint8_t qe_valid;
if (unlikely(!phba) || (phba->sli_rev != LPFC_SLI_REV4))
return false;
/* Check for completions on mailbox completion queue */
mcq = phba->sli4_hba.mbx_cq;
idx = mcq->hba_index;
qe_valid = mcq->qe_valid;
while (bf_get_le32(lpfc_cqe_valid, mcq->qe[idx].cqe) == qe_valid) {
mcqe = (struct lpfc_mcqe *)mcq->qe[idx].cqe;
if (bf_get_le32(lpfc_trailer_completed, mcqe) &&
(!bf_get_le32(lpfc_trailer_async, mcqe))) {
pending_completions = true;
break;
}
idx = (idx + 1) % mcq->entry_count;
if (mcq->hba_index == idx)
break;
/* if the index wrapped around, toggle the valid bit */
if (phba->sli4_hba.pc_sli4_params.cqav && !idx)
qe_valid = (qe_valid) ? 0 : 1;
}
return pending_completions;
}
/**
* lpfc_sli4_process_missed_mbox_completions - process mbox completions
* that were missed.
* @phba: Pointer to HBA context object.
*
* For sli4, it is possible to miss an interrupt. As such mbox completions
* maybe missed causing erroneous mailbox timeouts to occur. This function
* checks to see if mbox completions are on the mailbox completion queue
* and will process all the completions associated with the eq for the
* mailbox completion queue.
**/
bool
lpfc_sli4_process_missed_mbox_completions(struct lpfc_hba *phba)
{
struct lpfc_sli4_hba *sli4_hba = &phba->sli4_hba;
uint32_t eqidx;
struct lpfc_queue *fpeq = NULL;
struct lpfc_eqe *eqe;
bool mbox_pending;
if (unlikely(!phba) || (phba->sli_rev != LPFC_SLI_REV4))
return false;
/* Find the eq associated with the mcq */
if (sli4_hba->hba_eq)
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (eqidx = 0; eqidx < phba->io_channel_irqs; eqidx++)
if (sli4_hba->hba_eq[eqidx]->queue_id ==
sli4_hba->mbx_cq->assoc_qid) {
fpeq = sli4_hba->hba_eq[eqidx];
break;
}
if (!fpeq)
return false;
/* Turn off interrupts from this EQ */
sli4_hba->sli4_eq_clr_intr(fpeq);
/* Check to see if a mbox completion is pending */
mbox_pending = lpfc_sli4_mbox_completions_pending(phba);
/*
* If a mbox completion is pending, process all the events on EQ
* associated with the mbox completion queue (this could include
* mailbox commands, async events, els commands, receive queue data
* and fcp commands)
*/
if (mbox_pending)
while ((eqe = lpfc_sli4_eq_get(fpeq))) {
lpfc_sli4_hba_handle_eqe(phba, eqe, eqidx);
fpeq->EQ_processed++;
}
/* Always clear and re-arm the EQ */
sli4_hba->sli4_eq_release(fpeq, LPFC_QUEUE_REARM);
return mbox_pending;
}
/**
* lpfc_mbox_timeout_handler - Worker thread function to handle mailbox timeout
* @phba: Pointer to HBA context object.
*
* This function is called from worker thread when a mailbox command times out.
* The caller is not required to hold any locks. This function will reset the
* HBA and recover all the pending commands.
**/
void
lpfc_mbox_timeout_handler(struct lpfc_hba *phba)
{
LPFC_MBOXQ_t *pmbox = phba->sli.mbox_active;
MAILBOX_t *mb = NULL;
struct lpfc_sli *psli = &phba->sli;
/* If the mailbox completed, process the completion and return */
if (lpfc_sli4_process_missed_mbox_completions(phba))
return;
if (pmbox != NULL)
mb = &pmbox->u.mb;
/* Check the pmbox pointer first. There is a race condition
* between the mbox timeout handler getting executed in the
* worklist and the mailbox actually completing. When this
* race condition occurs, the mbox_active will be NULL.
*/
spin_lock_irq(&phba->hbalock);
if (pmbox == NULL) {
lpfc_printf_log(phba, KERN_WARNING,
LOG_MBOX | LOG_SLI,
"0353 Active Mailbox cleared - mailbox timeout "
"exiting\n");
spin_unlock_irq(&phba->hbalock);
return;
}
/* Mbox cmd <mbxCommand> timeout */
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0310 Mailbox command x%x timeout Data: x%x x%x x%p\n",
mb->mbxCommand,
phba->pport->port_state,
phba->sli.sli_flag,
phba->sli.mbox_active);
spin_unlock_irq(&phba->hbalock);
/* Setting state unknown so lpfc_sli_abort_iocb_ring
* would get IOCB_ERROR from lpfc_sli_issue_iocb, allowing
* it to fail all outstanding SCSI IO.
*/
spin_lock_irq(&phba->pport->work_port_lock);
phba->pport->work_port_events &= ~WORKER_MBOX_TMO;
spin_unlock_irq(&phba->pport->work_port_lock);
spin_lock_irq(&phba->hbalock);
phba->link_state = LPFC_LINK_UNKNOWN;
psli->sli_flag &= ~LPFC_SLI_ACTIVE;
spin_unlock_irq(&phba->hbalock);
lpfc_sli_abort_fcp_rings(phba);
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0345 Resetting board due to mailbox timeout\n");
/* Reset the HBA device */
lpfc_reset_hba(phba);
}
/**
* lpfc_sli_issue_mbox_s3 - Issue an SLI3 mailbox command to firmware
* @phba: Pointer to HBA context object.
* @pmbox: Pointer to mailbox object.
* @flag: Flag indicating how the mailbox need to be processed.
*
* This function is called by discovery code and HBA management code
* to submit a mailbox command to firmware with SLI-3 interface spec. This
* function gets the hbalock to protect the data structures.
* The mailbox command can be submitted in polling mode, in which case
* this function will wait in a polling loop for the completion of the
* mailbox.
* If the mailbox is submitted in no_wait mode (not polling) the
* function will submit the command and returns immediately without waiting
* for the mailbox completion. The no_wait is supported only when HBA
* is in SLI2/SLI3 mode - interrupts are enabled.
* The SLI interface allows only one mailbox pending at a time. If the
* mailbox is issued in polling mode and there is already a mailbox
* pending, then the function will return an error. If the mailbox is issued
* in NO_WAIT mode and there is a mailbox pending already, the function
* will return MBX_BUSY after queuing the mailbox into mailbox queue.
* The sli layer owns the mailbox object until the completion of mailbox
* command if this function return MBX_BUSY or MBX_SUCCESS. For all other
* return codes the caller owns the mailbox command after the return of
* the function.
**/
static int
lpfc_sli_issue_mbox_s3(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmbox,
uint32_t flag)
{
MAILBOX_t *mbx;
struct lpfc_sli *psli = &phba->sli;
uint32_t status, evtctr;
uint32_t ha_copy, hc_copy;
int i;
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
unsigned long timeout;
unsigned long drvr_flag = 0;
uint32_t word0, ldata;
void __iomem *to_slim;
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
int processing_queue = 0;
spin_lock_irqsave(&phba->hbalock, drvr_flag);
if (!pmbox) {
phba->sli.sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
/* processing mbox queue from intr_handler */
if (unlikely(psli->sli_flag & LPFC_SLI_ASYNC_MBX_BLK)) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
return MBX_SUCCESS;
}
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
processing_queue = 1;
pmbox = lpfc_mbox_get(phba);
if (!pmbox) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
return MBX_SUCCESS;
}
}
if (pmbox->mbox_cmpl && pmbox->mbox_cmpl != lpfc_sli_def_mbox_cmpl &&
pmbox->mbox_cmpl != lpfc_sli_wake_mbox_wait) {
if(!pmbox->vport) {
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
lpfc_printf_log(phba, KERN_ERR,
LOG_MBOX | LOG_VPORT,
"1806 Mbox x%x failed. No vport\n",
pmbox->u.mb.mbxCommand);
dump_stack();
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
goto out_not_finished;
}
}
/* If the PCI channel is in offline state, do not post mbox. */
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
if (unlikely(pci_channel_offline(phba->pcidev))) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
goto out_not_finished;
}
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/* If HBA has a deferred error attention, fail the iocb. */
if (unlikely(phba->hba_flag & DEFER_ERATT)) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
goto out_not_finished;
}
psli = &phba->sli;
mbx = &pmbox->u.mb;
status = MBX_SUCCESS;
if (phba->link_state == LPFC_HBA_ERROR) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
/* Mbox command <mbxCommand> cannot issue */
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):0311 Mailbox command x%x cannot "
"issue Data: x%x x%x\n",
pmbox->vport ? pmbox->vport->vpi : 0,
pmbox->u.mb.mbxCommand, psli->sli_flag, flag);
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
goto out_not_finished;
}
if (mbx->mbxCommand != MBX_KILL_BOARD && flag & MBX_NOWAIT) {
if (lpfc_readl(phba->HCregaddr, &hc_copy) ||
!(hc_copy & HC_MBINT_ENA)) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2528 Mailbox command x%x cannot "
"issue Data: x%x x%x\n",
pmbox->vport ? pmbox->vport->vpi : 0,
pmbox->u.mb.mbxCommand, psli->sli_flag, flag);
goto out_not_finished;
}
}
if (psli->sli_flag & LPFC_SLI_MBOX_ACTIVE) {
/* Polling for a mbox command when another one is already active
* is not allowed in SLI. Also, the driver must have established
* SLI2 mode to queue and process multiple mbox commands.
*/
if (flag & MBX_POLL) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
/* Mbox command <mbxCommand> cannot issue */
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2529 Mailbox command x%x "
"cannot issue Data: x%x x%x\n",
pmbox->vport ? pmbox->vport->vpi : 0,
pmbox->u.mb.mbxCommand,
psli->sli_flag, flag);
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
goto out_not_finished;
}
if (!(psli->sli_flag & LPFC_SLI_ACTIVE)) {
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
/* Mbox command <mbxCommand> cannot issue */
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2530 Mailbox command x%x "
"cannot issue Data: x%x x%x\n",
pmbox->vport ? pmbox->vport->vpi : 0,
pmbox->u.mb.mbxCommand,
psli->sli_flag, flag);
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
goto out_not_finished;
}
/* Another mailbox command is still being processed, queue this
* command to be processed later.
*/
lpfc_mbox_put(phba, pmbox);
/* Mbox cmd issue - BUSY */
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):0308 Mbox cmd issue - BUSY Data: "
"x%x x%x x%x x%x\n",
pmbox->vport ? pmbox->vport->vpi : 0xffffff,
mbx->mbxCommand,
phba->pport ? phba->pport->port_state : 0xff,
psli->sli_flag, flag);
psli->slistat.mbox_busy++;
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
if (pmbox->vport) {
lpfc_debugfs_disc_trc(pmbox->vport,
LPFC_DISC_TRC_MBOX_VPORT,
"MBOX Bsy vport: cmd:x%x mb:x%x x%x",
(uint32_t)mbx->mbxCommand,
mbx->un.varWords[0], mbx->un.varWords[1]);
}
else {
lpfc_debugfs_disc_trc(phba->pport,
LPFC_DISC_TRC_MBOX,
"MBOX Bsy: cmd:x%x mb:x%x x%x",
(uint32_t)mbx->mbxCommand,
mbx->un.varWords[0], mbx->un.varWords[1]);
}
return MBX_BUSY;
}
psli->sli_flag |= LPFC_SLI_MBOX_ACTIVE;
/* If we are not polling, we MUST be in SLI2 mode */
if (flag != MBX_POLL) {
if (!(psli->sli_flag & LPFC_SLI_ACTIVE) &&
(mbx->mbxCommand != MBX_KILL_BOARD)) {
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
/* Mbox command <mbxCommand> cannot issue */
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2531 Mailbox command x%x "
"cannot issue Data: x%x x%x\n",
pmbox->vport ? pmbox->vport->vpi : 0,
pmbox->u.mb.mbxCommand,
psli->sli_flag, flag);
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
goto out_not_finished;
}
/* timeout active mbox command */
timeout = msecs_to_jiffies(lpfc_mbox_tmo_val(phba, pmbox) *
1000);
mod_timer(&psli->mbox_tmo, jiffies + timeout);
}
/* Mailbox cmd <cmd> issue */
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):0309 Mailbox cmd x%x issue Data: x%x x%x "
"x%x\n",
pmbox->vport ? pmbox->vport->vpi : 0,
mbx->mbxCommand,
phba->pport ? phba->pport->port_state : 0xff,
psli->sli_flag, flag);
if (mbx->mbxCommand != MBX_HEARTBEAT) {
if (pmbox->vport) {
lpfc_debugfs_disc_trc(pmbox->vport,
LPFC_DISC_TRC_MBOX_VPORT,
"MBOX Send vport: cmd:x%x mb:x%x x%x",
(uint32_t)mbx->mbxCommand,
mbx->un.varWords[0], mbx->un.varWords[1]);
}
else {
lpfc_debugfs_disc_trc(phba->pport,
LPFC_DISC_TRC_MBOX,
"MBOX Send: cmd:x%x mb:x%x x%x",
(uint32_t)mbx->mbxCommand,
mbx->un.varWords[0], mbx->un.varWords[1]);
}
}
psli->slistat.mbox_cmd++;
evtctr = psli->slistat.mbox_event;
/* next set own bit for the adapter and copy over command word */
mbx->mbxOwner = OWN_CHIP;
if (psli->sli_flag & LPFC_SLI_ACTIVE) {
/* Populate mbox extension offset word. */
if (pmbox->in_ext_byte_len || pmbox->out_ext_byte_len) {
*(((uint32_t *)mbx) + pmbox->mbox_offset_word)
= (uint8_t *)phba->mbox_ext
- (uint8_t *)phba->mbox;
}
/* Copy the mailbox extension data */
if (pmbox->in_ext_byte_len && pmbox->context2) {
lpfc_sli_pcimem_bcopy(pmbox->context2,
(uint8_t *)phba->mbox_ext,
pmbox->in_ext_byte_len);
}
/* Copy command data to host SLIM area */
lpfc_sli_pcimem_bcopy(mbx, phba->mbox, MAILBOX_CMD_SIZE);
} else {
/* Populate mbox extension offset word. */
if (pmbox->in_ext_byte_len || pmbox->out_ext_byte_len)
*(((uint32_t *)mbx) + pmbox->mbox_offset_word)
= MAILBOX_HBA_EXT_OFFSET;
/* Copy the mailbox extension data */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (pmbox->in_ext_byte_len && pmbox->context2)
lpfc_memcpy_to_slim(phba->MBslimaddr +
MAILBOX_HBA_EXT_OFFSET,
pmbox->context2, pmbox->in_ext_byte_len);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (mbx->mbxCommand == MBX_CONFIG_PORT)
/* copy command data into host mbox for cmpl */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli_pcimem_bcopy(mbx, phba->mbox,
MAILBOX_CMD_SIZE);
/* First copy mbox command data to HBA SLIM, skip past first
word */
to_slim = phba->MBslimaddr + sizeof (uint32_t);
lpfc_memcpy_to_slim(to_slim, &mbx->un.varWords[0],
MAILBOX_CMD_SIZE - sizeof (uint32_t));
/* Next copy over first word, with mbxOwner set */
ldata = *((uint32_t *)mbx);
to_slim = phba->MBslimaddr;
writel(ldata, to_slim);
readl(to_slim); /* flush */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (mbx->mbxCommand == MBX_CONFIG_PORT)
/* switch over to host mailbox */
psli->sli_flag |= LPFC_SLI_ACTIVE;
}
wmb();
switch (flag) {
case MBX_NOWAIT:
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
/* Set up reference to mailbox command */
psli->mbox_active = pmbox;
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
/* Interrupt board to do it */
writel(CA_MBATT, phba->CAregaddr);
readl(phba->CAregaddr); /* flush */
/* Don't wait for it to finish, just return */
break;
case MBX_POLL:
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
/* Set up null reference to mailbox command */
psli->mbox_active = NULL;
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
/* Interrupt board to do it */
writel(CA_MBATT, phba->CAregaddr);
readl(phba->CAregaddr); /* flush */
if (psli->sli_flag & LPFC_SLI_ACTIVE) {
/* First read mbox status word */
word0 = *((uint32_t *)phba->mbox);
word0 = le32_to_cpu(word0);
} else {
/* First read mbox status word */
if (lpfc_readl(phba->MBslimaddr, &word0)) {
spin_unlock_irqrestore(&phba->hbalock,
drvr_flag);
goto out_not_finished;
}
}
/* Read the HBA Host Attention Register */
if (lpfc_readl(phba->HAregaddr, &ha_copy)) {
spin_unlock_irqrestore(&phba->hbalock,
drvr_flag);
goto out_not_finished;
}
timeout = msecs_to_jiffies(lpfc_mbox_tmo_val(phba, pmbox) *
1000) + jiffies;
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
i = 0;
/* Wait for command to complete */
while (((word0 & OWN_CHIP) == OWN_CHIP) ||
(!(ha_copy & HA_MBATT) &&
(phba->link_state > LPFC_WARM_START))) {
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
if (time_after(jiffies, timeout)) {
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
spin_unlock_irqrestore(&phba->hbalock,
drvr_flag);
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
goto out_not_finished;
}
/* Check if we took a mbox interrupt while we were
polling */
if (((word0 & OWN_CHIP) != OWN_CHIP)
&& (evtctr != psli->slistat.mbox_event))
break;
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
if (i++ > 10) {
spin_unlock_irqrestore(&phba->hbalock,
drvr_flag);
msleep(1);
spin_lock_irqsave(&phba->hbalock, drvr_flag);
}
if (psli->sli_flag & LPFC_SLI_ACTIVE) {
/* First copy command data */
word0 = *((uint32_t *)phba->mbox);
word0 = le32_to_cpu(word0);
if (mbx->mbxCommand == MBX_CONFIG_PORT) {
MAILBOX_t *slimmb;
uint32_t slimword0;
/* Check real SLIM for any errors */
slimword0 = readl(phba->MBslimaddr);
slimmb = (MAILBOX_t *) & slimword0;
if (((slimword0 & OWN_CHIP) != OWN_CHIP)
&& slimmb->mbxStatus) {
psli->sli_flag &=
~LPFC_SLI_ACTIVE;
word0 = slimword0;
}
}
} else {
/* First copy command data */
word0 = readl(phba->MBslimaddr);
}
/* Read the HBA Host Attention Register */
if (lpfc_readl(phba->HAregaddr, &ha_copy)) {
spin_unlock_irqrestore(&phba->hbalock,
drvr_flag);
goto out_not_finished;
}
}
if (psli->sli_flag & LPFC_SLI_ACTIVE) {
/* copy results back to user */
lpfc_sli_pcimem_bcopy(phba->mbox, mbx,
MAILBOX_CMD_SIZE);
/* Copy the mailbox extension data */
if (pmbox->out_ext_byte_len && pmbox->context2) {
lpfc_sli_pcimem_bcopy(phba->mbox_ext,
pmbox->context2,
pmbox->out_ext_byte_len);
}
} else {
/* First copy command data */
lpfc_memcpy_from_slim(mbx, phba->MBslimaddr,
MAILBOX_CMD_SIZE);
/* Copy the mailbox extension data */
if (pmbox->out_ext_byte_len && pmbox->context2) {
lpfc_memcpy_from_slim(pmbox->context2,
phba->MBslimaddr +
MAILBOX_HBA_EXT_OFFSET,
pmbox->out_ext_byte_len);
}
}
writel(HA_MBATT, phba->HAregaddr);
readl(phba->HAregaddr); /* flush */
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
status = mbx->mbxStatus;
}
spin_unlock_irqrestore(&phba->hbalock, drvr_flag);
return status;
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
out_not_finished:
if (processing_queue) {
pmbox->u.mb.mbxStatus = MBX_NOT_FINISHED;
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
lpfc_mbox_cmpl_put(phba, pmbox);
}
return MBX_NOT_FINISHED;
}
/**
* lpfc_sli4_async_mbox_block - Block posting SLI4 asynchronous mailbox command
* @phba: Pointer to HBA context object.
*
* The function blocks the posting of SLI4 asynchronous mailbox commands from
* the driver internal pending mailbox queue. It will then try to wait out the
* possible outstanding mailbox command before return.
*
* Returns:
* 0 - the outstanding mailbox command completed; otherwise, the wait for
* the outstanding mailbox command timed out.
**/
static int
lpfc_sli4_async_mbox_block(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
int rc = 0;
unsigned long timeout = 0;
/* Mark the asynchronous mailbox command posting as blocked */
spin_lock_irq(&phba->hbalock);
psli->sli_flag |= LPFC_SLI_ASYNC_MBX_BLK;
/* Determine how long we might wait for the active mailbox
* command to be gracefully completed by firmware.
*/
if (phba->sli.mbox_active)
timeout = msecs_to_jiffies(lpfc_mbox_tmo_val(phba,
phba->sli.mbox_active) *
1000) + jiffies;
spin_unlock_irq(&phba->hbalock);
/* Make sure the mailbox is really active */
if (timeout)
lpfc_sli4_process_missed_mbox_completions(phba);
/* Wait for the outstnading mailbox command to complete */
while (phba->sli.mbox_active) {
/* Check active mailbox complete status every 2ms */
msleep(2);
if (time_after(jiffies, timeout)) {
/* Timeout, marked the outstanding cmd not complete */
rc = 1;
break;
}
}
/* Can not cleanly block async mailbox command, fails it */
if (rc) {
spin_lock_irq(&phba->hbalock);
psli->sli_flag &= ~LPFC_SLI_ASYNC_MBX_BLK;
spin_unlock_irq(&phba->hbalock);
}
return rc;
}
/**
* lpfc_sli4_async_mbox_unblock - Block posting SLI4 async mailbox command
* @phba: Pointer to HBA context object.
*
* The function unblocks and resume posting of SLI4 asynchronous mailbox
* commands from the driver internal pending mailbox queue. It makes sure
* that there is no outstanding mailbox command before resuming posting
* asynchronous mailbox commands. If, for any reason, there is outstanding
* mailbox command, it will try to wait it out before resuming asynchronous
* mailbox command posting.
**/
static void
lpfc_sli4_async_mbox_unblock(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
spin_lock_irq(&phba->hbalock);
if (!(psli->sli_flag & LPFC_SLI_ASYNC_MBX_BLK)) {
/* Asynchronous mailbox posting is not blocked, do nothing */
spin_unlock_irq(&phba->hbalock);
return;
}
/* Outstanding synchronous mailbox command is guaranteed to be done,
* successful or timeout, after timing-out the outstanding mailbox
* command shall always be removed, so just unblock posting async
* mailbox command and resume
*/
psli->sli_flag &= ~LPFC_SLI_ASYNC_MBX_BLK;
spin_unlock_irq(&phba->hbalock);
/* wake up worker thread to post asynchronlous mailbox command */
lpfc_worker_wake_up(phba);
}
/**
* lpfc_sli4_wait_bmbx_ready - Wait for bootstrap mailbox register ready
* @phba: Pointer to HBA context object.
* @mboxq: Pointer to mailbox object.
*
* The function waits for the bootstrap mailbox register ready bit from
* port for twice the regular mailbox command timeout value.
*
* 0 - no timeout on waiting for bootstrap mailbox register ready.
* MBXERR_ERROR - wait for bootstrap mailbox register timed out.
**/
static int
lpfc_sli4_wait_bmbx_ready(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
{
uint32_t db_ready;
unsigned long timeout;
struct lpfc_register bmbx_reg;
timeout = msecs_to_jiffies(lpfc_mbox_tmo_val(phba, mboxq)
* 1000) + jiffies;
do {
bmbx_reg.word0 = readl(phba->sli4_hba.BMBXregaddr);
db_ready = bf_get(lpfc_bmbx_rdy, &bmbx_reg);
if (!db_ready)
msleep(2);
if (time_after(jiffies, timeout))
return MBXERR_ERROR;
} while (!db_ready);
return 0;
}
/**
* lpfc_sli4_post_sync_mbox - Post an SLI4 mailbox to the bootstrap mailbox
* @phba: Pointer to HBA context object.
* @mboxq: Pointer to mailbox object.
*
* The function posts a mailbox to the port. The mailbox is expected
* to be comletely filled in and ready for the port to operate on it.
* This routine executes a synchronous completion operation on the
* mailbox by polling for its completion.
*
* The caller must not be holding any locks when calling this routine.
*
* Returns:
* MBX_SUCCESS - mailbox posted successfully
* Any of the MBX error values.
**/
static int
lpfc_sli4_post_sync_mbox(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
{
int rc = MBX_SUCCESS;
unsigned long iflag;
uint32_t mcqe_status;
uint32_t mbx_cmnd;
struct lpfc_sli *psli = &phba->sli;
struct lpfc_mqe *mb = &mboxq->u.mqe;
struct lpfc_bmbx_create *mbox_rgn;
struct dma_address *dma_address;
/*
* Only one mailbox can be active to the bootstrap mailbox region
* at a time and there is no queueing provided.
*/
spin_lock_irqsave(&phba->hbalock, iflag);
if (psli->sli_flag & LPFC_SLI_MBOX_ACTIVE) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2532 Mailbox command x%x (x%x/x%x) "
"cannot issue Data: x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0,
mboxq->u.mb.mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
psli->sli_flag, MBX_POLL);
return MBXERR_ERROR;
}
/* The server grabs the token and owns it until release */
psli->sli_flag |= LPFC_SLI_MBOX_ACTIVE;
phba->sli.mbox_active = mboxq;
spin_unlock_irqrestore(&phba->hbalock, iflag);
/* wait for bootstrap mbox register for readyness */
rc = lpfc_sli4_wait_bmbx_ready(phba, mboxq);
if (rc)
goto exit;
/*
* Initialize the bootstrap memory region to avoid stale data areas
* in the mailbox post. Then copy the caller's mailbox contents to
* the bmbx mailbox region.
*/
mbx_cmnd = bf_get(lpfc_mqe_command, mb);
memset(phba->sli4_hba.bmbx.avirt, 0, sizeof(struct lpfc_bmbx_create));
lpfc_sli_pcimem_bcopy(mb, phba->sli4_hba.bmbx.avirt,
sizeof(struct lpfc_mqe));
/* Post the high mailbox dma address to the port and wait for ready. */
dma_address = &phba->sli4_hba.bmbx.dma_address;
writel(dma_address->addr_hi, phba->sli4_hba.BMBXregaddr);
/* wait for bootstrap mbox register for hi-address write done */
rc = lpfc_sli4_wait_bmbx_ready(phba, mboxq);
if (rc)
goto exit;
/* Post the low mailbox dma address to the port. */
writel(dma_address->addr_lo, phba->sli4_hba.BMBXregaddr);
/* wait for bootstrap mbox register for low address write done */
rc = lpfc_sli4_wait_bmbx_ready(phba, mboxq);
if (rc)
goto exit;
/*
* Read the CQ to ensure the mailbox has completed.
* If so, update the mailbox status so that the upper layers
* can complete the request normally.
*/
lpfc_sli_pcimem_bcopy(phba->sli4_hba.bmbx.avirt, mb,
sizeof(struct lpfc_mqe));
mbox_rgn = (struct lpfc_bmbx_create *) phba->sli4_hba.bmbx.avirt;
lpfc_sli_pcimem_bcopy(&mbox_rgn->mcqe, &mboxq->mcqe,
sizeof(struct lpfc_mcqe));
mcqe_status = bf_get(lpfc_mcqe_status, &mbox_rgn->mcqe);
/*
* When the CQE status indicates a failure and the mailbox status
* indicates success then copy the CQE status into the mailbox status
* (and prefix it with x4000).
*/
if (mcqe_status != MB_CQE_STATUS_SUCCESS) {
if (bf_get(lpfc_mqe_status, mb) == MBX_SUCCESS)
bf_set(lpfc_mqe_status, mb,
(LPFC_MBX_ERROR_RANGE | mcqe_status));
rc = MBXERR_ERROR;
} else
lpfc_sli4_swap_str(phba, mboxq);
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):0356 Mailbox cmd x%x (x%x/x%x) Status x%x "
"Data: x%x x%x x%x x%x x%x x%x x%x x%x x%x x%x x%x"
" x%x x%x CQ: x%x x%x x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0, mbx_cmnd,
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
bf_get(lpfc_mqe_status, mb),
mb->un.mb_words[0], mb->un.mb_words[1],
mb->un.mb_words[2], mb->un.mb_words[3],
mb->un.mb_words[4], mb->un.mb_words[5],
mb->un.mb_words[6], mb->un.mb_words[7],
mb->un.mb_words[8], mb->un.mb_words[9],
mb->un.mb_words[10], mb->un.mb_words[11],
mb->un.mb_words[12], mboxq->mcqe.word0,
mboxq->mcqe.mcqe_tag0, mboxq->mcqe.mcqe_tag1,
mboxq->mcqe.trailer);
exit:
/* We are holding the token, no needed for lock when release */
spin_lock_irqsave(&phba->hbalock, iflag);
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
phba->sli.mbox_active = NULL;
spin_unlock_irqrestore(&phba->hbalock, iflag);
return rc;
}
/**
* lpfc_sli_issue_mbox_s4 - Issue an SLI4 mailbox command to firmware
* @phba: Pointer to HBA context object.
* @pmbox: Pointer to mailbox object.
* @flag: Flag indicating how the mailbox need to be processed.
*
* This function is called by discovery code and HBA management code to submit
* a mailbox command to firmware with SLI-4 interface spec.
*
* Return codes the caller owns the mailbox command after the return of the
* function.
**/
static int
lpfc_sli_issue_mbox_s4(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq,
uint32_t flag)
{
struct lpfc_sli *psli = &phba->sli;
unsigned long iflags;
int rc;
/* dump from issue mailbox command if setup */
lpfc_idiag_mbxacc_dump_issue_mbox(phba, &mboxq->u.mb);
rc = lpfc_mbox_dev_check(phba);
if (unlikely(rc)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2544 Mailbox command x%x (x%x/x%x) "
"cannot issue Data: x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0,
mboxq->u.mb.mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
psli->sli_flag, flag);
goto out_not_finished;
}
/* Detect polling mode and jump to a handler */
if (!phba->sli4_hba.intr_enable) {
if (flag == MBX_POLL)
rc = lpfc_sli4_post_sync_mbox(phba, mboxq);
else
rc = -EIO;
if (rc != MBX_SUCCESS)
lpfc_printf_log(phba, KERN_WARNING, LOG_MBOX | LOG_SLI,
"(%d):2541 Mailbox command x%x "
"(x%x/x%x) failure: "
"mqe_sta: x%x mcqe_sta: x%x/x%x "
"Data: x%x x%x\n,",
mboxq->vport ? mboxq->vport->vpi : 0,
mboxq->u.mb.mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba,
mboxq),
lpfc_sli_config_mbox_opcode_get(phba,
mboxq),
bf_get(lpfc_mqe_status, &mboxq->u.mqe),
bf_get(lpfc_mcqe_status, &mboxq->mcqe),
bf_get(lpfc_mcqe_ext_status,
&mboxq->mcqe),
psli->sli_flag, flag);
return rc;
} else if (flag == MBX_POLL) {
lpfc_printf_log(phba, KERN_WARNING, LOG_MBOX | LOG_SLI,
"(%d):2542 Try to issue mailbox command "
"x%x (x%x/x%x) synchronously ahead of async "
"mailbox command queue: x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0,
mboxq->u.mb.mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
psli->sli_flag, flag);
/* Try to block the asynchronous mailbox posting */
rc = lpfc_sli4_async_mbox_block(phba);
if (!rc) {
/* Successfully blocked, now issue sync mbox cmd */
rc = lpfc_sli4_post_sync_mbox(phba, mboxq);
if (rc != MBX_SUCCESS)
lpfc_printf_log(phba, KERN_WARNING,
LOG_MBOX | LOG_SLI,
"(%d):2597 Sync Mailbox command "
"x%x (x%x/x%x) failure: "
"mqe_sta: x%x mcqe_sta: x%x/x%x "
"Data: x%x x%x\n,",
mboxq->vport ? mboxq->vport->vpi : 0,
mboxq->u.mb.mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba,
mboxq),
lpfc_sli_config_mbox_opcode_get(phba,
mboxq),
bf_get(lpfc_mqe_status, &mboxq->u.mqe),
bf_get(lpfc_mcqe_status, &mboxq->mcqe),
bf_get(lpfc_mcqe_ext_status,
&mboxq->mcqe),
psli->sli_flag, flag);
/* Unblock the async mailbox posting afterward */
lpfc_sli4_async_mbox_unblock(phba);
}
return rc;
}
/* Now, interrupt mode asynchrous mailbox command */
rc = lpfc_mbox_cmd_check(phba, mboxq);
if (rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2543 Mailbox command x%x (x%x/x%x) "
"cannot issue Data: x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0,
mboxq->u.mb.mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
psli->sli_flag, flag);
goto out_not_finished;
}
/* Put the mailbox command to the driver internal FIFO */
psli->slistat.mbox_busy++;
spin_lock_irqsave(&phba->hbalock, iflags);
lpfc_mbox_put(phba, mboxq);
spin_unlock_irqrestore(&phba->hbalock, iflags);
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):0354 Mbox cmd issue - Enqueue Data: "
"x%x (x%x/x%x) x%x x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0xffffff,
bf_get(lpfc_mqe_command, &mboxq->u.mqe),
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
phba->pport->port_state,
psli->sli_flag, MBX_NOWAIT);
/* Wake up worker thread to transport mailbox command from head */
lpfc_worker_wake_up(phba);
return MBX_BUSY;
out_not_finished:
return MBX_NOT_FINISHED;
}
/**
* lpfc_sli4_post_async_mbox - Post an SLI4 mailbox command to device
* @phba: Pointer to HBA context object.
*
* This function is called by worker thread to send a mailbox command to
* SLI4 HBA firmware.
*
**/
int
lpfc_sli4_post_async_mbox(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
LPFC_MBOXQ_t *mboxq;
int rc = MBX_SUCCESS;
unsigned long iflags;
struct lpfc_mqe *mqe;
uint32_t mbx_cmnd;
/* Check interrupt mode before post async mailbox command */
if (unlikely(!phba->sli4_hba.intr_enable))
return MBX_NOT_FINISHED;
/* Check for mailbox command service token */
spin_lock_irqsave(&phba->hbalock, iflags);
if (unlikely(psli->sli_flag & LPFC_SLI_ASYNC_MBX_BLK)) {
spin_unlock_irqrestore(&phba->hbalock, iflags);
return MBX_NOT_FINISHED;
}
if (psli->sli_flag & LPFC_SLI_MBOX_ACTIVE) {
spin_unlock_irqrestore(&phba->hbalock, iflags);
return MBX_NOT_FINISHED;
}
if (unlikely(phba->sli.mbox_active)) {
spin_unlock_irqrestore(&phba->hbalock, iflags);
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"0384 There is pending active mailbox cmd\n");
return MBX_NOT_FINISHED;
}
/* Take the mailbox command service token */
psli->sli_flag |= LPFC_SLI_MBOX_ACTIVE;
/* Get the next mailbox command from head of queue */
mboxq = lpfc_mbox_get(phba);
/* If no more mailbox command waiting for post, we're done */
if (!mboxq) {
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
spin_unlock_irqrestore(&phba->hbalock, iflags);
return MBX_SUCCESS;
}
phba->sli.mbox_active = mboxq;
spin_unlock_irqrestore(&phba->hbalock, iflags);
/* Check device readiness for posting mailbox command */
rc = lpfc_mbox_dev_check(phba);
if (unlikely(rc))
/* Driver clean routine will clean up pending mailbox */
goto out_not_finished;
/* Prepare the mbox command to be posted */
mqe = &mboxq->u.mqe;
mbx_cmnd = bf_get(lpfc_mqe_command, mqe);
/* Start timer for the mbox_tmo and log some mailbox post messages */
mod_timer(&psli->mbox_tmo, (jiffies +
msecs_to_jiffies(1000 * lpfc_mbox_tmo_val(phba, mboxq))));
lpfc_printf_log(phba, KERN_INFO, LOG_MBOX | LOG_SLI,
"(%d):0355 Mailbox cmd x%x (x%x/x%x) issue Data: "
"x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0, mbx_cmnd,
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
phba->pport->port_state, psli->sli_flag);
if (mbx_cmnd != MBX_HEARTBEAT) {
if (mboxq->vport) {
lpfc_debugfs_disc_trc(mboxq->vport,
LPFC_DISC_TRC_MBOX_VPORT,
"MBOX Send vport: cmd:x%x mb:x%x x%x",
mbx_cmnd, mqe->un.mb_words[0],
mqe->un.mb_words[1]);
} else {
lpfc_debugfs_disc_trc(phba->pport,
LPFC_DISC_TRC_MBOX,
"MBOX Send: cmd:x%x mb:x%x x%x",
mbx_cmnd, mqe->un.mb_words[0],
mqe->un.mb_words[1]);
}
}
psli->slistat.mbox_cmd++;
/* Post the mailbox command to the port */
rc = lpfc_sli4_mq_put(phba->sli4_hba.mbx_wq, mqe);
if (rc != MBX_SUCCESS) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
"(%d):2533 Mailbox command x%x (x%x/x%x) "
"cannot issue Data: x%x x%x\n",
mboxq->vport ? mboxq->vport->vpi : 0,
mboxq->u.mb.mbxCommand,
lpfc_sli_config_mbox_subsys_get(phba, mboxq),
lpfc_sli_config_mbox_opcode_get(phba, mboxq),
psli->sli_flag, MBX_NOWAIT);
goto out_not_finished;
}
return rc;
out_not_finished:
spin_lock_irqsave(&phba->hbalock, iflags);
if (phba->sli.mbox_active) {
mboxq->u.mb.mbxStatus = MBX_NOT_FINISHED;
__lpfc_mbox_cmpl_put(phba, mboxq);
/* Release the token */
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
phba->sli.mbox_active = NULL;
}
spin_unlock_irqrestore(&phba->hbalock, iflags);
return MBX_NOT_FINISHED;
}
/**
* lpfc_sli_issue_mbox - Wrapper func for issuing mailbox command
* @phba: Pointer to HBA context object.
* @pmbox: Pointer to mailbox object.
* @flag: Flag indicating how the mailbox need to be processed.
*
* This routine wraps the actual SLI3 or SLI4 mailbox issuing routine from
* the API jump table function pointer from the lpfc_hba struct.
*
* Return codes the caller owns the mailbox command after the return of the
* function.
**/
int
lpfc_sli_issue_mbox(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmbox, uint32_t flag)
{
return phba->lpfc_sli_issue_mbox(phba, pmbox, flag);
}
/**
* lpfc_mbox_api_table_setup - Set up mbox api function jump table
* @phba: The hba struct for which this call is being executed.
* @dev_grp: The HBA PCI-Device group number.
*
* This routine sets up the mbox interface API function jump table in @phba
* struct.
* Returns: 0 - success, -ENODEV - failure.
**/
int
lpfc_mbox_api_table_setup(struct lpfc_hba *phba, uint8_t dev_grp)
{
switch (dev_grp) {
case LPFC_PCI_DEV_LP:
phba->lpfc_sli_issue_mbox = lpfc_sli_issue_mbox_s3;
phba->lpfc_sli_handle_slow_ring_event =
lpfc_sli_handle_slow_ring_event_s3;
phba->lpfc_sli_hbq_to_firmware = lpfc_sli_hbq_to_firmware_s3;
phba->lpfc_sli_brdrestart = lpfc_sli_brdrestart_s3;
phba->lpfc_sli_brdready = lpfc_sli_brdready_s3;
break;
case LPFC_PCI_DEV_OC:
phba->lpfc_sli_issue_mbox = lpfc_sli_issue_mbox_s4;
phba->lpfc_sli_handle_slow_ring_event =
lpfc_sli_handle_slow_ring_event_s4;
phba->lpfc_sli_hbq_to_firmware = lpfc_sli_hbq_to_firmware_s4;
phba->lpfc_sli_brdrestart = lpfc_sli_brdrestart_s4;
phba->lpfc_sli_brdready = lpfc_sli_brdready_s4;
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"1420 Invalid HBA PCI-device group: 0x%x\n",
dev_grp);
return -ENODEV;
break;
}
return 0;
}
/**
* __lpfc_sli_ringtx_put - Add an iocb to the txq
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @piocb: Pointer to address of newly added command iocb.
*
* This function is called with hbalock held to add a command
* iocb to the txq when SLI layer cannot submit the command iocb
* to the ring.
**/
void
__lpfc_sli_ringtx_put(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *piocb)
{
lockdep_assert_held(&phba->hbalock);
/* Insert the caller's iocb in the txq tail for later processing. */
list_add_tail(&piocb->list, &pring->txq);
}
/**
* lpfc_sli_next_iocb - Get the next iocb in the txq
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @piocb: Pointer to address of newly added command iocb.
*
* This function is called with hbalock held before a new
* iocb is submitted to the firmware. This function checks
* txq to flush the iocbs in txq to Firmware before
* submitting new iocbs to the Firmware.
* If there are iocbs in the txq which need to be submitted
* to firmware, lpfc_sli_next_iocb returns the first element
* of the txq after dequeuing it from txq.
* If there is no iocb in the txq then the function will return
* *piocb and *piocb is set to NULL. Caller needs to check
* *piocb to find if there are more commands in the txq.
**/
static struct lpfc_iocbq *
lpfc_sli_next_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq **piocb)
{
struct lpfc_iocbq * nextiocb;
lockdep_assert_held(&phba->hbalock);
nextiocb = lpfc_sli_ringtx_get(phba, pring);
if (!nextiocb) {
nextiocb = *piocb;
*piocb = NULL;
}
return nextiocb;
}
/**
* __lpfc_sli_issue_iocb_s3 - SLI3 device lockless ver of lpfc_sli_issue_iocb
* @phba: Pointer to HBA context object.
* @ring_number: SLI ring number to issue iocb on.
* @piocb: Pointer to command iocb.
* @flag: Flag indicating if this command can be put into txq.
*
* __lpfc_sli_issue_iocb_s3 is used by other functions in the driver to issue
* an iocb command to an HBA with SLI-3 interface spec. If the PCI slot is
* recovering from error state, if HBA is resetting or if LPFC_STOP_IOCB_EVENT
* flag is turned on, the function returns IOCB_ERROR. When the link is down,
* this function allows only iocbs for posting buffers. This function finds
* next available slot in the command ring and posts the command to the
* available slot and writes the port attention register to request HBA start
* processing new iocb. If there is no slot available in the ring and
* flag & SLI_IOCB_RET_IOCB is set, the new iocb is added to the txq, otherwise
* the function returns IOCB_BUSY.
*
* This function is called with hbalock held. The function will return success
* after it successfully submit the iocb to firmware or after adding to the
* txq.
**/
static int
__lpfc_sli_issue_iocb_s3(struct lpfc_hba *phba, uint32_t ring_number,
struct lpfc_iocbq *piocb, uint32_t flag)
{
struct lpfc_iocbq *nextiocb;
IOCB_t *iocb;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *pring = &phba->sli.sli3_ring[ring_number];
lockdep_assert_held(&phba->hbalock);
if (piocb->iocb_cmpl && (!piocb->vport) &&
(piocb->iocb.ulpCommand != CMD_ABORT_XRI_CN) &&
(piocb->iocb.ulpCommand != CMD_CLOSE_XRI_CN)) {
lpfc_printf_log(phba, KERN_ERR,
LOG_SLI | LOG_VPORT,
"1807 IOCB x%x failed. No vport\n",
piocb->iocb.ulpCommand);
dump_stack();
return IOCB_ERROR;
}
/* If the PCI channel is in offline state, do not post iocbs. */
if (unlikely(pci_channel_offline(phba->pcidev)))
return IOCB_ERROR;
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/* If HBA has a deferred error attention, fail the iocb. */
if (unlikely(phba->hba_flag & DEFER_ERATT))
return IOCB_ERROR;
/*
* We should never get an IOCB if we are in a < LINK_DOWN state
*/
if (unlikely(phba->link_state < LPFC_LINK_DOWN))
return IOCB_ERROR;
/*
* Check to see if we are blocking IOCB processing because of a
* outstanding event.
*/
if (unlikely(pring->flag & LPFC_STOP_IOCB_EVENT))
goto iocb_busy;
if (unlikely(phba->link_state == LPFC_LINK_DOWN)) {
/*
* Only CREATE_XRI, CLOSE_XRI, and QUE_RING_BUF
* can be issued if the link is not up.
*/
switch (piocb->iocb.ulpCommand) {
case CMD_GEN_REQUEST64_CR:
case CMD_GEN_REQUEST64_CX:
if (!(phba->sli.sli_flag & LPFC_MENLO_MAINT) ||
(piocb->iocb.un.genreq64.w5.hcsw.Rctl !=
FC_RCTL_DD_UNSOL_CMD) ||
(piocb->iocb.un.genreq64.w5.hcsw.Type !=
MENLO_TRANSPORT_TYPE))
goto iocb_busy;
break;
case CMD_QUE_RING_BUF_CN:
case CMD_QUE_RING_BUF64_CN:
/*
* For IOCBs, like QUE_RING_BUF, that have no rsp ring
* completion, iocb_cmpl MUST be 0.
*/
if (piocb->iocb_cmpl)
piocb->iocb_cmpl = NULL;
/*FALLTHROUGH*/
case CMD_CREATE_XRI_CR:
case CMD_CLOSE_XRI_CN:
case CMD_CLOSE_XRI_CX:
break;
default:
goto iocb_busy;
}
/*
* For FCP commands, we must be in a state where we can process link
* attention events.
*/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
} else if (unlikely(pring->ringno == LPFC_FCP_RING &&
!(phba->sli.sli_flag & LPFC_PROCESS_LA))) {
goto iocb_busy;
}
while ((iocb = lpfc_sli_next_iocb_slot(phba, pring)) &&
(nextiocb = lpfc_sli_next_iocb(phba, pring, &piocb)))
lpfc_sli_submit_iocb(phba, pring, iocb, nextiocb);
if (iocb)
lpfc_sli_update_ring(phba, pring);
else
lpfc_sli_update_full_ring(phba, pring);
if (!piocb)
return IOCB_SUCCESS;
goto out_busy;
iocb_busy:
pring->stats.iocb_cmd_delay++;
out_busy:
if (!(flag & SLI_IOCB_RET_IOCB)) {
__lpfc_sli_ringtx_put(phba, pring, piocb);
return IOCB_SUCCESS;
}
return IOCB_BUSY;
}
/**
* lpfc_sli4_bpl2sgl - Convert the bpl/bde to a sgl.
* @phba: Pointer to HBA context object.
* @piocb: Pointer to command iocb.
* @sglq: Pointer to the scatter gather queue object.
*
* This routine converts the bpl or bde that is in the IOCB
* to a sgl list for the sli4 hardware. The physical address
* of the bpl/bde is converted back to a virtual address.
* If the IOCB contains a BPL then the list of BDE's is
* converted to sli4_sge's. If the IOCB contains a single
* BDE then it is converted to a single sli_sge.
* The IOCB is still in cpu endianess so the contents of
* the bpl can be used without byte swapping.
*
* Returns valid XRI = Success, NO_XRI = Failure.
**/
static uint16_t
lpfc_sli4_bpl2sgl(struct lpfc_hba *phba, struct lpfc_iocbq *piocbq,
struct lpfc_sglq *sglq)
{
uint16_t xritag = NO_XRI;
struct ulp_bde64 *bpl = NULL;
struct ulp_bde64 bde;
struct sli4_sge *sgl = NULL;
struct lpfc_dmabuf *dmabuf;
IOCB_t *icmd;
int numBdes = 0;
int i = 0;
uint32_t offset = 0; /* accumulated offset in the sg request list */
int inbound = 0; /* number of sg reply entries inbound from firmware */
if (!piocbq || !sglq)
return xritag;
sgl = (struct sli4_sge *)sglq->sgl;
icmd = &piocbq->iocb;
if (icmd->ulpCommand == CMD_XMIT_BLS_RSP64_CX)
return sglq->sli4_xritag;
if (icmd->un.genreq64.bdl.bdeFlags == BUFF_TYPE_BLP_64) {
numBdes = icmd->un.genreq64.bdl.bdeSize /
sizeof(struct ulp_bde64);
/* The addrHigh and addrLow fields within the IOCB
* have not been byteswapped yet so there is no
* need to swap them back.
*/
if (piocbq->context3)
dmabuf = (struct lpfc_dmabuf *)piocbq->context3;
else
return xritag;
bpl = (struct ulp_bde64 *)dmabuf->virt;
if (!bpl)
return xritag;
for (i = 0; i < numBdes; i++) {
/* Should already be byte swapped. */
sgl->addr_hi = bpl->addrHigh;
sgl->addr_lo = bpl->addrLow;
sgl->word2 = le32_to_cpu(sgl->word2);
if ((i+1) == numBdes)
bf_set(lpfc_sli4_sge_last, sgl, 1);
else
bf_set(lpfc_sli4_sge_last, sgl, 0);
/* swap the size field back to the cpu so we
* can assign it to the sgl.
*/
bde.tus.w = le32_to_cpu(bpl->tus.w);
sgl->sge_len = cpu_to_le32(bde.tus.f.bdeSize);
/* The offsets in the sgl need to be accumulated
* separately for the request and reply lists.
* The request is always first, the reply follows.
*/
if (piocbq->iocb.ulpCommand == CMD_GEN_REQUEST64_CR) {
/* add up the reply sg entries */
if (bpl->tus.f.bdeFlags == BUFF_TYPE_BDE_64I)
inbound++;
/* first inbound? reset the offset */
if (inbound == 1)
offset = 0;
bf_set(lpfc_sli4_sge_offset, sgl, offset);
bf_set(lpfc_sli4_sge_type, sgl,
LPFC_SGE_TYPE_DATA);
offset += bde.tus.f.bdeSize;
}
sgl->word2 = cpu_to_le32(sgl->word2);
bpl++;
sgl++;
}
} else if (icmd->un.genreq64.bdl.bdeFlags == BUFF_TYPE_BDE_64) {
/* The addrHigh and addrLow fields of the BDE have not
* been byteswapped yet so they need to be swapped
* before putting them in the sgl.
*/
sgl->addr_hi =
cpu_to_le32(icmd->un.genreq64.bdl.addrHigh);
sgl->addr_lo =
cpu_to_le32(icmd->un.genreq64.bdl.addrLow);
sgl->word2 = le32_to_cpu(sgl->word2);
bf_set(lpfc_sli4_sge_last, sgl, 1);
sgl->word2 = cpu_to_le32(sgl->word2);
sgl->sge_len =
cpu_to_le32(icmd->un.genreq64.bdl.bdeSize);
}
return sglq->sli4_xritag;
}
/**
* lpfc_sli_iocb2wqe - Convert the IOCB to a work queue entry.
* @phba: Pointer to HBA context object.
* @piocb: Pointer to command iocb.
* @wqe: Pointer to the work queue entry.
*
* This routine converts the iocb command to its Work Queue Entry
* equivalent. The wqe pointer should not have any fields set when
* this routine is called because it will memcpy over them.
* This routine does not set the CQ_ID or the WQEC bits in the
* wqe.
*
* Returns: 0 = Success, IOCB_ERROR = Failure.
**/
static int
lpfc_sli4_iocb2wqe(struct lpfc_hba *phba, struct lpfc_iocbq *iocbq,
union lpfc_wqe128 *wqe)
{
uint32_t xmit_len = 0, total_len = 0;
uint8_t ct = 0;
uint32_t fip;
uint32_t abort_tag;
uint8_t command_type = ELS_COMMAND_NON_FIP;
uint8_t cmnd;
uint16_t xritag;
uint16_t abrt_iotag;
struct lpfc_iocbq *abrtiocbq;
struct ulp_bde64 *bpl = NULL;
uint32_t els_id = LPFC_ELS_ID_DEFAULT;
int numBdes, i;
struct ulp_bde64 bde;
struct lpfc_nodelist *ndlp;
uint32_t *pcmd;
uint32_t if_type;
fip = phba->hba_flag & HBA_FIP_SUPPORT;
/* The fcp commands will set command type */
if (iocbq->iocb_flag & LPFC_IO_FCP)
command_type = FCP_COMMAND;
else if (fip && (iocbq->iocb_flag & LPFC_FIP_ELS_ID_MASK))
command_type = ELS_COMMAND_FIP;
else
command_type = ELS_COMMAND_NON_FIP;
if (phba->fcp_embed_io)
memset(wqe, 0, sizeof(union lpfc_wqe128));
/* Some of the fields are in the right position already */
memcpy(wqe, &iocbq->iocb, sizeof(union lpfc_wqe));
if (iocbq->iocb.ulpCommand != CMD_SEND_FRAME) {
/* The ct field has moved so reset */
wqe->generic.wqe_com.word7 = 0;
wqe->generic.wqe_com.word10 = 0;
}
abort_tag = (uint32_t) iocbq->iotag;
xritag = iocbq->sli4_xritag;
/* words0-2 bpl convert bde */
if (iocbq->iocb.un.genreq64.bdl.bdeFlags == BUFF_TYPE_BLP_64) {
numBdes = iocbq->iocb.un.genreq64.bdl.bdeSize /
sizeof(struct ulp_bde64);
bpl = (struct ulp_bde64 *)
((struct lpfc_dmabuf *)iocbq->context3)->virt;
if (!bpl)
return IOCB_ERROR;
/* Should already be byte swapped. */
wqe->generic.bde.addrHigh = le32_to_cpu(bpl->addrHigh);
wqe->generic.bde.addrLow = le32_to_cpu(bpl->addrLow);
/* swap the size field back to the cpu so we
* can assign it to the sgl.
*/
wqe->generic.bde.tus.w = le32_to_cpu(bpl->tus.w);
xmit_len = wqe->generic.bde.tus.f.bdeSize;
total_len = 0;
for (i = 0; i < numBdes; i++) {
bde.tus.w = le32_to_cpu(bpl[i].tus.w);
total_len += bde.tus.f.bdeSize;
}
} else
xmit_len = iocbq->iocb.un.fcpi64.bdl.bdeSize;
iocbq->iocb.ulpIoTag = iocbq->iotag;
cmnd = iocbq->iocb.ulpCommand;
switch (iocbq->iocb.ulpCommand) {
case CMD_ELS_REQUEST64_CR:
if (iocbq->iocb_flag & LPFC_IO_LIBDFC)
ndlp = iocbq->context_un.ndlp;
else
ndlp = (struct lpfc_nodelist *)iocbq->context1;
if (!iocbq->iocb.ulpLe) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2007 Only Limited Edition cmd Format"
" supported 0x%x\n",
iocbq->iocb.ulpCommand);
return IOCB_ERROR;
}
wqe->els_req.payload_len = xmit_len;
/* Els_reguest64 has a TMO */
bf_set(wqe_tmo, &wqe->els_req.wqe_com,
iocbq->iocb.ulpTimeout);
/* Need a VF for word 4 set the vf bit*/
bf_set(els_req64_vf, &wqe->els_req, 0);
/* And a VFID for word 12 */
bf_set(els_req64_vfid, &wqe->els_req, 0);
ct = ((iocbq->iocb.ulpCt_h << 1) | iocbq->iocb.ulpCt_l);
bf_set(wqe_ctxt_tag, &wqe->els_req.wqe_com,
iocbq->iocb.ulpContext);
bf_set(wqe_ct, &wqe->els_req.wqe_com, ct);
bf_set(wqe_pu, &wqe->els_req.wqe_com, 0);
/* CCP CCPE PV PRI in word10 were set in the memcpy */
if (command_type == ELS_COMMAND_FIP)
els_id = ((iocbq->iocb_flag & LPFC_FIP_ELS_ID_MASK)
>> LPFC_FIP_ELS_ID_SHIFT);
pcmd = (uint32_t *) (((struct lpfc_dmabuf *)
iocbq->context2)->virt);
if_type = bf_get(lpfc_sli_intf_if_type,
&phba->sli4_hba.sli_intf);
if (if_type >= LPFC_SLI_INTF_IF_TYPE_2) {
if (pcmd && (*pcmd == ELS_CMD_FLOGI ||
*pcmd == ELS_CMD_SCR ||
*pcmd == ELS_CMD_FDISC ||
*pcmd == ELS_CMD_LOGO ||
*pcmd == ELS_CMD_PLOGI)) {
bf_set(els_req64_sp, &wqe->els_req, 1);
bf_set(els_req64_sid, &wqe->els_req,
iocbq->vport->fc_myDID);
if ((*pcmd == ELS_CMD_FLOGI) &&
!(phba->fc_topology ==
LPFC_TOPOLOGY_LOOP))
bf_set(els_req64_sid, &wqe->els_req, 0);
bf_set(wqe_ct, &wqe->els_req.wqe_com, 1);
bf_set(wqe_ctxt_tag, &wqe->els_req.wqe_com,
phba->vpi_ids[iocbq->vport->vpi]);
} else if (pcmd && iocbq->context1) {
bf_set(wqe_ct, &wqe->els_req.wqe_com, 0);
bf_set(wqe_ctxt_tag, &wqe->els_req.wqe_com,
phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
}
}
bf_set(wqe_temp_rpi, &wqe->els_req.wqe_com,
phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
bf_set(wqe_els_id, &wqe->els_req.wqe_com, els_id);
bf_set(wqe_dbde, &wqe->els_req.wqe_com, 1);
bf_set(wqe_iod, &wqe->els_req.wqe_com, LPFC_WQE_IOD_READ);
bf_set(wqe_qosd, &wqe->els_req.wqe_com, 1);
bf_set(wqe_lenloc, &wqe->els_req.wqe_com, LPFC_WQE_LENLOC_NONE);
bf_set(wqe_ebde_cnt, &wqe->els_req.wqe_com, 0);
wqe->els_req.max_response_payload_len = total_len - xmit_len;
break;
case CMD_XMIT_SEQUENCE64_CX:
bf_set(wqe_ctxt_tag, &wqe->xmit_sequence.wqe_com,
iocbq->iocb.un.ulpWord[3]);
bf_set(wqe_rcvoxid, &wqe->xmit_sequence.wqe_com,
iocbq->iocb.unsli3.rcvsli3.ox_id);
/* The entire sequence is transmitted for this IOCB */
xmit_len = total_len;
cmnd = CMD_XMIT_SEQUENCE64_CR;
if (phba->link_flag & LS_LOOPBACK_MODE)
bf_set(wqe_xo, &wqe->xmit_sequence.wge_ctl, 1);
case CMD_XMIT_SEQUENCE64_CR:
/* word3 iocb=io_tag32 wqe=reserved */
wqe->xmit_sequence.rsvd3 = 0;
/* word4 relative_offset memcpy */
/* word5 r_ctl/df_ctl memcpy */
bf_set(wqe_pu, &wqe->xmit_sequence.wqe_com, 0);
bf_set(wqe_dbde, &wqe->xmit_sequence.wqe_com, 1);
bf_set(wqe_iod, &wqe->xmit_sequence.wqe_com,
LPFC_WQE_IOD_WRITE);
bf_set(wqe_lenloc, &wqe->xmit_sequence.wqe_com,
LPFC_WQE_LENLOC_WORD12);
bf_set(wqe_ebde_cnt, &wqe->xmit_sequence.wqe_com, 0);
wqe->xmit_sequence.xmit_len = xmit_len;
command_type = OTHER_COMMAND;
break;
case CMD_XMIT_BCAST64_CN:
/* word3 iocb=iotag32 wqe=seq_payload_len */
wqe->xmit_bcast64.seq_payload_len = xmit_len;
/* word4 iocb=rsvd wqe=rsvd */
/* word5 iocb=rctl/type/df_ctl wqe=rctl/type/df_ctl memcpy */
/* word6 iocb=ctxt_tag/io_tag wqe=ctxt_tag/xri */
bf_set(wqe_ct, &wqe->xmit_bcast64.wqe_com,
((iocbq->iocb.ulpCt_h << 1) | iocbq->iocb.ulpCt_l));
bf_set(wqe_dbde, &wqe->xmit_bcast64.wqe_com, 1);
bf_set(wqe_iod, &wqe->xmit_bcast64.wqe_com, LPFC_WQE_IOD_WRITE);
bf_set(wqe_lenloc, &wqe->xmit_bcast64.wqe_com,
LPFC_WQE_LENLOC_WORD3);
bf_set(wqe_ebde_cnt, &wqe->xmit_bcast64.wqe_com, 0);
break;
case CMD_FCP_IWRITE64_CR:
command_type = FCP_COMMAND_DATA_OUT;
/* word3 iocb=iotag wqe=payload_offset_len */
/* Add the FCP_CMD and FCP_RSP sizes to get the offset */
bf_set(payload_offset_len, &wqe->fcp_iwrite,
xmit_len + sizeof(struct fcp_rsp));
bf_set(cmd_buff_len, &wqe->fcp_iwrite,
0);
/* word4 iocb=parameter wqe=total_xfer_length memcpy */
/* word5 iocb=initial_xfer_len wqe=initial_xfer_len memcpy */
bf_set(wqe_erp, &wqe->fcp_iwrite.wqe_com,
iocbq->iocb.ulpFCP2Rcvy);
bf_set(wqe_lnk, &wqe->fcp_iwrite.wqe_com, iocbq->iocb.ulpXS);
/* Always open the exchange */
bf_set(wqe_iod, &wqe->fcp_iwrite.wqe_com, LPFC_WQE_IOD_WRITE);
bf_set(wqe_lenloc, &wqe->fcp_iwrite.wqe_com,
LPFC_WQE_LENLOC_WORD4);
bf_set(wqe_pu, &wqe->fcp_iwrite.wqe_com, iocbq->iocb.ulpPU);
bf_set(wqe_dbde, &wqe->fcp_iwrite.wqe_com, 1);
if (iocbq->iocb_flag & LPFC_IO_OAS) {
bf_set(wqe_oas, &wqe->fcp_iwrite.wqe_com, 1);
bf_set(wqe_ccpe, &wqe->fcp_iwrite.wqe_com, 1);
if (iocbq->priority) {
bf_set(wqe_ccp, &wqe->fcp_iwrite.wqe_com,
(iocbq->priority << 1));
} else {
bf_set(wqe_ccp, &wqe->fcp_iwrite.wqe_com,
(phba->cfg_XLanePriority << 1));
}
}
/* Note, word 10 is already initialized to 0 */
/* Don't set PBDE for Perf hints, just fcp_embed_pbde */
if (phba->fcp_embed_pbde)
bf_set(wqe_pbde, &wqe->fcp_iwrite.wqe_com, 1);
else
bf_set(wqe_pbde, &wqe->fcp_iwrite.wqe_com, 0);
if (phba->fcp_embed_io) {
struct lpfc_scsi_buf *lpfc_cmd;
struct sli4_sge *sgl;
struct fcp_cmnd *fcp_cmnd;
uint32_t *ptr;
/* 128 byte wqe support here */
lpfc_cmd = iocbq->context1;
sgl = (struct sli4_sge *)lpfc_cmd->fcp_bpl;
fcp_cmnd = lpfc_cmd->fcp_cmnd;
/* Word 0-2 - FCP_CMND */
wqe->generic.bde.tus.f.bdeFlags =
BUFF_TYPE_BDE_IMMED;
wqe->generic.bde.tus.f.bdeSize = sgl->sge_len;
wqe->generic.bde.addrHigh = 0;
wqe->generic.bde.addrLow = 88; /* Word 22 */
bf_set(wqe_wqes, &wqe->fcp_iwrite.wqe_com, 1);
bf_set(wqe_dbde, &wqe->fcp_iwrite.wqe_com, 0);
/* Word 22-29 FCP CMND Payload */
ptr = &wqe->words[22];
memcpy(ptr, fcp_cmnd, sizeof(struct fcp_cmnd));
}
break;
case CMD_FCP_IREAD64_CR:
/* word3 iocb=iotag wqe=payload_offset_len */
/* Add the FCP_CMD and FCP_RSP sizes to get the offset */
bf_set(payload_offset_len, &wqe->fcp_iread,
xmit_len + sizeof(struct fcp_rsp));
bf_set(cmd_buff_len, &wqe->fcp_iread,
0);
/* word4 iocb=parameter wqe=total_xfer_length memcpy */
/* word5 iocb=initial_xfer_len wqe=initial_xfer_len memcpy */
bf_set(wqe_erp, &wqe->fcp_iread.wqe_com,
iocbq->iocb.ulpFCP2Rcvy);
bf_set(wqe_lnk, &wqe->fcp_iread.wqe_com, iocbq->iocb.ulpXS);
/* Always open the exchange */
bf_set(wqe_iod, &wqe->fcp_iread.wqe_com, LPFC_WQE_IOD_READ);
bf_set(wqe_lenloc, &wqe->fcp_iread.wqe_com,
LPFC_WQE_LENLOC_WORD4);
bf_set(wqe_pu, &wqe->fcp_iread.wqe_com, iocbq->iocb.ulpPU);
bf_set(wqe_dbde, &wqe->fcp_iread.wqe_com, 1);
if (iocbq->iocb_flag & LPFC_IO_OAS) {
bf_set(wqe_oas, &wqe->fcp_iread.wqe_com, 1);
bf_set(wqe_ccpe, &wqe->fcp_iread.wqe_com, 1);
if (iocbq->priority) {
bf_set(wqe_ccp, &wqe->fcp_iread.wqe_com,
(iocbq->priority << 1));
} else {
bf_set(wqe_ccp, &wqe->fcp_iread.wqe_com,
(phba->cfg_XLanePriority << 1));
}
}
/* Note, word 10 is already initialized to 0 */
/* Don't set PBDE for Perf hints, just fcp_embed_pbde */
if (phba->fcp_embed_pbde)
bf_set(wqe_pbde, &wqe->fcp_iread.wqe_com, 1);
else
bf_set(wqe_pbde, &wqe->fcp_iread.wqe_com, 0);
if (phba->fcp_embed_io) {
struct lpfc_scsi_buf *lpfc_cmd;
struct sli4_sge *sgl;
struct fcp_cmnd *fcp_cmnd;
uint32_t *ptr;
/* 128 byte wqe support here */
lpfc_cmd = iocbq->context1;
sgl = (struct sli4_sge *)lpfc_cmd->fcp_bpl;
fcp_cmnd = lpfc_cmd->fcp_cmnd;
/* Word 0-2 - FCP_CMND */
wqe->generic.bde.tus.f.bdeFlags =
BUFF_TYPE_BDE_IMMED;
wqe->generic.bde.tus.f.bdeSize = sgl->sge_len;
wqe->generic.bde.addrHigh = 0;
wqe->generic.bde.addrLow = 88; /* Word 22 */
bf_set(wqe_wqes, &wqe->fcp_iread.wqe_com, 1);
bf_set(wqe_dbde, &wqe->fcp_iread.wqe_com, 0);
/* Word 22-29 FCP CMND Payload */
ptr = &wqe->words[22];
memcpy(ptr, fcp_cmnd, sizeof(struct fcp_cmnd));
}
break;
case CMD_FCP_ICMND64_CR:
/* word3 iocb=iotag wqe=payload_offset_len */
/* Add the FCP_CMD and FCP_RSP sizes to get the offset */
bf_set(payload_offset_len, &wqe->fcp_icmd,
xmit_len + sizeof(struct fcp_rsp));
bf_set(cmd_buff_len, &wqe->fcp_icmd,
0);
/* word3 iocb=IO_TAG wqe=reserved */
bf_set(wqe_pu, &wqe->fcp_icmd.wqe_com, 0);
/* Always open the exchange */
bf_set(wqe_dbde, &wqe->fcp_icmd.wqe_com, 1);
bf_set(wqe_iod, &wqe->fcp_icmd.wqe_com, LPFC_WQE_IOD_WRITE);
bf_set(wqe_qosd, &wqe->fcp_icmd.wqe_com, 1);
bf_set(wqe_lenloc, &wqe->fcp_icmd.wqe_com,
LPFC_WQE_LENLOC_NONE);
bf_set(wqe_erp, &wqe->fcp_icmd.wqe_com,
iocbq->iocb.ulpFCP2Rcvy);
if (iocbq->iocb_flag & LPFC_IO_OAS) {
bf_set(wqe_oas, &wqe->fcp_icmd.wqe_com, 1);
bf_set(wqe_ccpe, &wqe->fcp_icmd.wqe_com, 1);
if (iocbq->priority) {
bf_set(wqe_ccp, &wqe->fcp_icmd.wqe_com,
(iocbq->priority << 1));
} else {
bf_set(wqe_ccp, &wqe->fcp_icmd.wqe_com,
(phba->cfg_XLanePriority << 1));
}
}
/* Note, word 10 is already initialized to 0 */
if (phba->fcp_embed_io) {
struct lpfc_scsi_buf *lpfc_cmd;
struct sli4_sge *sgl;
struct fcp_cmnd *fcp_cmnd;
uint32_t *ptr;
/* 128 byte wqe support here */
lpfc_cmd = iocbq->context1;
sgl = (struct sli4_sge *)lpfc_cmd->fcp_bpl;
fcp_cmnd = lpfc_cmd->fcp_cmnd;
/* Word 0-2 - FCP_CMND */
wqe->generic.bde.tus.f.bdeFlags =
BUFF_TYPE_BDE_IMMED;
wqe->generic.bde.tus.f.bdeSize = sgl->sge_len;
wqe->generic.bde.addrHigh = 0;
wqe->generic.bde.addrLow = 88; /* Word 22 */
bf_set(wqe_wqes, &wqe->fcp_icmd.wqe_com, 1);
bf_set(wqe_dbde, &wqe->fcp_icmd.wqe_com, 0);
/* Word 22-29 FCP CMND Payload */
ptr = &wqe->words[22];
memcpy(ptr, fcp_cmnd, sizeof(struct fcp_cmnd));
}
break;
case CMD_GEN_REQUEST64_CR:
/* For this command calculate the xmit length of the
* request bde.
*/
xmit_len = 0;
numBdes = iocbq->iocb.un.genreq64.bdl.bdeSize /
sizeof(struct ulp_bde64);
for (i = 0; i < numBdes; i++) {
bde.tus.w = le32_to_cpu(bpl[i].tus.w);
if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
break;
xmit_len += bde.tus.f.bdeSize;
}
/* word3 iocb=IO_TAG wqe=request_payload_len */
wqe->gen_req.request_payload_len = xmit_len;
/* word4 iocb=parameter wqe=relative_offset memcpy */
/* word5 [rctl, type, df_ctl, la] copied in memcpy */
/* word6 context tag copied in memcpy */
if (iocbq->iocb.ulpCt_h || iocbq->iocb.ulpCt_l) {
ct = ((iocbq->iocb.ulpCt_h << 1) | iocbq->iocb.ulpCt_l);
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2015 Invalid CT %x command 0x%x\n",
ct, iocbq->iocb.ulpCommand);
return IOCB_ERROR;
}
bf_set(wqe_ct, &wqe->gen_req.wqe_com, 0);
bf_set(wqe_tmo, &wqe->gen_req.wqe_com, iocbq->iocb.ulpTimeout);
bf_set(wqe_pu, &wqe->gen_req.wqe_com, iocbq->iocb.ulpPU);
bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
wqe->gen_req.max_response_payload_len = total_len - xmit_len;
command_type = OTHER_COMMAND;
break;
case CMD_XMIT_ELS_RSP64_CX:
ndlp = (struct lpfc_nodelist *)iocbq->context1;
/* words0-2 BDE memcpy */
/* word3 iocb=iotag32 wqe=response_payload_len */
wqe->xmit_els_rsp.response_payload_len = xmit_len;
/* word4 */
wqe->xmit_els_rsp.word4 = 0;
/* word5 iocb=rsvd wge=did */
bf_set(wqe_els_did, &wqe->xmit_els_rsp.wqe_dest,
iocbq->iocb.un.xseq64.xmit_els_remoteID);
if_type = bf_get(lpfc_sli_intf_if_type,
&phba->sli4_hba.sli_intf);
if (if_type >= LPFC_SLI_INTF_IF_TYPE_2) {
if (iocbq->vport->fc_flag & FC_PT2PT) {
bf_set(els_rsp64_sp, &wqe->xmit_els_rsp, 1);
bf_set(els_rsp64_sid, &wqe->xmit_els_rsp,
iocbq->vport->fc_myDID);
if (iocbq->vport->fc_myDID == Fabric_DID) {
bf_set(wqe_els_did,
&wqe->xmit_els_rsp.wqe_dest, 0);
}
}
}
bf_set(wqe_ct, &wqe->xmit_els_rsp.wqe_com,
((iocbq->iocb.ulpCt_h << 1) | iocbq->iocb.ulpCt_l));
bf_set(wqe_pu, &wqe->xmit_els_rsp.wqe_com, iocbq->iocb.ulpPU);
bf_set(wqe_rcvoxid, &wqe->xmit_els_rsp.wqe_com,
iocbq->iocb.unsli3.rcvsli3.ox_id);
if (!iocbq->iocb.ulpCt_h && iocbq->iocb.ulpCt_l)
bf_set(wqe_ctxt_tag, &wqe->xmit_els_rsp.wqe_com,
phba->vpi_ids[iocbq->vport->vpi]);
bf_set(wqe_dbde, &wqe->xmit_els_rsp.wqe_com, 1);
bf_set(wqe_iod, &wqe->xmit_els_rsp.wqe_com, LPFC_WQE_IOD_WRITE);
bf_set(wqe_qosd, &wqe->xmit_els_rsp.wqe_com, 1);
bf_set(wqe_lenloc, &wqe->xmit_els_rsp.wqe_com,
LPFC_WQE_LENLOC_WORD3);
bf_set(wqe_ebde_cnt, &wqe->xmit_els_rsp.wqe_com, 0);
bf_set(wqe_rsp_temp_rpi, &wqe->xmit_els_rsp,
phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
pcmd = (uint32_t *) (((struct lpfc_dmabuf *)
iocbq->context2)->virt);
if (phba->fc_topology == LPFC_TOPOLOGY_LOOP) {
bf_set(els_rsp64_sp, &wqe->xmit_els_rsp, 1);
bf_set(els_rsp64_sid, &wqe->xmit_els_rsp,
iocbq->vport->fc_myDID);
bf_set(wqe_ct, &wqe->xmit_els_rsp.wqe_com, 1);
bf_set(wqe_ctxt_tag, &wqe->xmit_els_rsp.wqe_com,
phba->vpi_ids[phba->pport->vpi]);
}
command_type = OTHER_COMMAND;
break;
case CMD_CLOSE_XRI_CN:
case CMD_ABORT_XRI_CN:
case CMD_ABORT_XRI_CX:
/* words 0-2 memcpy should be 0 rserved */
/* port will send abts */
abrt_iotag = iocbq->iocb.un.acxri.abortContextTag;
if (abrt_iotag != 0 && abrt_iotag <= phba->sli.last_iotag) {
abrtiocbq = phba->sli.iocbq_lookup[abrt_iotag];
fip = abrtiocbq->iocb_flag & LPFC_FIP_ELS_ID_MASK;
} else
fip = 0;
if ((iocbq->iocb.ulpCommand == CMD_CLOSE_XRI_CN) || fip)
/*
* The link is down, or the command was ELS_FIP
* so the fw does not need to send abts
* on the wire.
*/
bf_set(abort_cmd_ia, &wqe->abort_cmd, 1);
else
bf_set(abort_cmd_ia, &wqe->abort_cmd, 0);
bf_set(abort_cmd_criteria, &wqe->abort_cmd, T_XRI_TAG);
/* word5 iocb=CONTEXT_TAG|IO_TAG wqe=reserved */
wqe->abort_cmd.rsrvd5 = 0;
bf_set(wqe_ct, &wqe->abort_cmd.wqe_com,
((iocbq->iocb.ulpCt_h << 1) | iocbq->iocb.ulpCt_l));
abort_tag = iocbq->iocb.un.acxri.abortIoTag;
/*
* The abort handler will send us CMD_ABORT_XRI_CN or
* CMD_CLOSE_XRI_CN and the fw only accepts CMD_ABORT_XRI_CX
*/
bf_set(wqe_cmnd, &wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX);
bf_set(wqe_qosd, &wqe->abort_cmd.wqe_com, 1);
bf_set(wqe_lenloc, &wqe->abort_cmd.wqe_com,
LPFC_WQE_LENLOC_NONE);
cmnd = CMD_ABORT_XRI_CX;
command_type = OTHER_COMMAND;
xritag = 0;
break;
case CMD_XMIT_BLS_RSP64_CX:
ndlp = (struct lpfc_nodelist *)iocbq->context1;
/* As BLS ABTS RSP WQE is very different from other WQEs,
* we re-construct this WQE here based on information in
* iocbq from scratch.
*/
memset(wqe, 0, sizeof(union lpfc_wqe));
/* OX_ID is invariable to who sent ABTS to CT exchange */
bf_set(xmit_bls_rsp64_oxid, &wqe->xmit_bls_rsp,
bf_get(lpfc_abts_oxid, &iocbq->iocb.un.bls_rsp));
if (bf_get(lpfc_abts_orig, &iocbq->iocb.un.bls_rsp) ==
LPFC_ABTS_UNSOL_INT) {
/* ABTS sent by initiator to CT exchange, the
* RX_ID field will be filled with the newly
* allocated responder XRI.
*/
bf_set(xmit_bls_rsp64_rxid, &wqe->xmit_bls_rsp,
iocbq->sli4_xritag);
} else {
/* ABTS sent by responder to CT exchange, the
* RX_ID field will be filled with the responder
* RX_ID from ABTS.
*/
bf_set(xmit_bls_rsp64_rxid, &wqe->xmit_bls_rsp,
bf_get(lpfc_abts_rxid, &iocbq->iocb.un.bls_rsp));
}
bf_set(xmit_bls_rsp64_seqcnthi, &wqe->xmit_bls_rsp, 0xffff);
bf_set(wqe_xmit_bls_pt, &wqe->xmit_bls_rsp.wqe_dest, 0x1);
/* Use CT=VPI */
bf_set(wqe_els_did, &wqe->xmit_bls_rsp.wqe_dest,
ndlp->nlp_DID);
bf_set(xmit_bls_rsp64_temprpi, &wqe->xmit_bls_rsp,
iocbq->iocb.ulpContext);
bf_set(wqe_ct, &wqe->xmit_bls_rsp.wqe_com, 1);
bf_set(wqe_ctxt_tag, &wqe->xmit_bls_rsp.wqe_com,
phba->vpi_ids[phba->pport->vpi]);
bf_set(wqe_qosd, &wqe->xmit_bls_rsp.wqe_com, 1);
bf_set(wqe_lenloc, &wqe->xmit_bls_rsp.wqe_com,
LPFC_WQE_LENLOC_NONE);
/* Overwrite the pre-set comnd type with OTHER_COMMAND */
command_type = OTHER_COMMAND;
if (iocbq->iocb.un.xseq64.w5.hcsw.Rctl == FC_RCTL_BA_RJT) {
bf_set(xmit_bls_rsp64_rjt_vspec, &wqe->xmit_bls_rsp,
bf_get(lpfc_vndr_code, &iocbq->iocb.un.bls_rsp));
bf_set(xmit_bls_rsp64_rjt_expc, &wqe->xmit_bls_rsp,
bf_get(lpfc_rsn_expln, &iocbq->iocb.un.bls_rsp));
bf_set(xmit_bls_rsp64_rjt_rsnc, &wqe->xmit_bls_rsp,
bf_get(lpfc_rsn_code, &iocbq->iocb.un.bls_rsp));
}
break;
case CMD_SEND_FRAME:
bf_set(wqe_xri_tag, &wqe->generic.wqe_com, xritag);
bf_set(wqe_reqtag, &wqe->generic.wqe_com, iocbq->iotag);
return 0;
case CMD_XRI_ABORTED_CX:
case CMD_CREATE_XRI_CR: /* Do we expect to use this? */
case CMD_IOCB_FCP_IBIDIR64_CR: /* bidirectional xfer */
case CMD_FCP_TSEND64_CX: /* Target mode send xfer-ready */
case CMD_FCP_TRSP64_CX: /* Target mode rcv */
case CMD_FCP_AUTO_TRSP_CX: /* Auto target rsp */
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2014 Invalid command 0x%x\n",
iocbq->iocb.ulpCommand);
return IOCB_ERROR;
break;
}
if (iocbq->iocb_flag & LPFC_IO_DIF_PASS)
bf_set(wqe_dif, &wqe->generic.wqe_com, LPFC_WQE_DIF_PASSTHRU);
else if (iocbq->iocb_flag & LPFC_IO_DIF_STRIP)
bf_set(wqe_dif, &wqe->generic.wqe_com, LPFC_WQE_DIF_STRIP);
else if (iocbq->iocb_flag & LPFC_IO_DIF_INSERT)
bf_set(wqe_dif, &wqe->generic.wqe_com, LPFC_WQE_DIF_INSERT);
iocbq->iocb_flag &= ~(LPFC_IO_DIF_PASS | LPFC_IO_DIF_STRIP |
LPFC_IO_DIF_INSERT);
bf_set(wqe_xri_tag, &wqe->generic.wqe_com, xritag);
bf_set(wqe_reqtag, &wqe->generic.wqe_com, iocbq->iotag);
wqe->generic.wqe_com.abort_tag = abort_tag;
bf_set(wqe_cmd_type, &wqe->generic.wqe_com, command_type);
bf_set(wqe_cmnd, &wqe->generic.wqe_com, cmnd);
bf_set(wqe_class, &wqe->generic.wqe_com, iocbq->iocb.ulpClass);
bf_set(wqe_cqid, &wqe->generic.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
return 0;
}
/**
* __lpfc_sli_issue_iocb_s4 - SLI4 device lockless ver of lpfc_sli_issue_iocb
* @phba: Pointer to HBA context object.
* @ring_number: SLI ring number to issue iocb on.
* @piocb: Pointer to command iocb.
* @flag: Flag indicating if this command can be put into txq.
*
* __lpfc_sli_issue_iocb_s4 is used by other functions in the driver to issue
* an iocb command to an HBA with SLI-4 interface spec.
*
* This function is called with hbalock held. The function will return success
* after it successfully submit the iocb to firmware or after adding to the
* txq.
**/
static int
__lpfc_sli_issue_iocb_s4(struct lpfc_hba *phba, uint32_t ring_number,
struct lpfc_iocbq *piocb, uint32_t flag)
{
struct lpfc_sglq *sglq;
union lpfc_wqe128 wqe;
struct lpfc_queue *wq;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *pring;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* Get the WQ */
if ((piocb->iocb_flag & LPFC_IO_FCP) ||
(piocb->iocb_flag & LPFC_USE_FCPWQIDX)) {
if (!phba->cfg_fof || (!(piocb->iocb_flag & LPFC_IO_OAS)))
wq = phba->sli4_hba.fcp_wq[piocb->hba_wqidx];
else
wq = phba->sli4_hba.oas_wq;
} else {
wq = phba->sli4_hba.els_wq;
}
/* Get corresponding ring */
pring = wq->pring;
/*
* The WQE can be either 64 or 128 bytes,
*/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lockdep_assert_held(&phba->hbalock);
if (piocb->sli4_xritag == NO_XRI) {
if (piocb->iocb.ulpCommand == CMD_ABORT_XRI_CN ||
piocb->iocb.ulpCommand == CMD_CLOSE_XRI_CN)
sglq = NULL;
else {
if (!list_empty(&pring->txq)) {
if (!(flag & SLI_IOCB_RET_IOCB)) {
__lpfc_sli_ringtx_put(phba,
pring, piocb);
return IOCB_SUCCESS;
} else {
return IOCB_BUSY;
}
} else {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
sglq = __lpfc_sli_get_els_sglq(phba, piocb);
if (!sglq) {
if (!(flag & SLI_IOCB_RET_IOCB)) {
__lpfc_sli_ringtx_put(phba,
pring,
piocb);
return IOCB_SUCCESS;
} else
return IOCB_BUSY;
}
}
}
} else if (piocb->iocb_flag & LPFC_IO_FCP)
/* These IO's already have an XRI and a mapped sgl. */
sglq = NULL;
else {
/*
* This is a continuation of a commandi,(CX) so this
* sglq is on the active list
*/
sglq = __lpfc_get_active_sglq(phba, piocb->sli4_lxritag);
if (!sglq)
return IOCB_ERROR;
}
if (sglq) {
piocb->sli4_lxritag = sglq->sli4_lxritag;
piocb->sli4_xritag = sglq->sli4_xritag;
if (NO_XRI == lpfc_sli4_bpl2sgl(phba, piocb, sglq))
return IOCB_ERROR;
}
if (lpfc_sli4_iocb2wqe(phba, piocb, &wqe))
return IOCB_ERROR;
if (lpfc_sli4_wq_put(wq, &wqe))
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
return IOCB_ERROR;
lpfc_sli_ringtxcmpl_put(phba, pring, piocb);
return 0;
}
/**
* __lpfc_sli_issue_iocb - Wrapper func of lockless version for issuing iocb
*
* This routine wraps the actual lockless version for issusing IOCB function
* pointer from the lpfc_hba struct.
*
* Return codes:
* IOCB_ERROR - Error
* IOCB_SUCCESS - Success
* IOCB_BUSY - Busy
**/
int
__lpfc_sli_issue_iocb(struct lpfc_hba *phba, uint32_t ring_number,
struct lpfc_iocbq *piocb, uint32_t flag)
{
return phba->__lpfc_sli_issue_iocb(phba, ring_number, piocb, flag);
}
/**
* lpfc_sli_api_table_setup - Set up sli api function jump table
* @phba: The hba struct for which this call is being executed.
* @dev_grp: The HBA PCI-Device group number.
*
* This routine sets up the SLI interface API function jump table in @phba
* struct.
* Returns: 0 - success, -ENODEV - failure.
**/
int
lpfc_sli_api_table_setup(struct lpfc_hba *phba, uint8_t dev_grp)
{
switch (dev_grp) {
case LPFC_PCI_DEV_LP:
phba->__lpfc_sli_issue_iocb = __lpfc_sli_issue_iocb_s3;
phba->__lpfc_sli_release_iocbq = __lpfc_sli_release_iocbq_s3;
break;
case LPFC_PCI_DEV_OC:
phba->__lpfc_sli_issue_iocb = __lpfc_sli_issue_iocb_s4;
phba->__lpfc_sli_release_iocbq = __lpfc_sli_release_iocbq_s4;
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"1419 Invalid HBA PCI-device group: 0x%x\n",
dev_grp);
return -ENODEV;
break;
}
phba->lpfc_get_iocb_from_iocbq = lpfc_get_iocb_from_iocbq;
return 0;
}
/**
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* lpfc_sli4_calc_ring - Calculates which ring to use
* @phba: Pointer to HBA context object.
* @piocb: Pointer to command iocb.
*
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* For SLI4 only, FCP IO can deferred to one fo many WQs, based on
* hba_wqidx, thus we need to calculate the corresponding ring.
* Since ABORTS must go on the same WQ of the command they are
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* aborting, we use command's hba_wqidx.
*/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *
lpfc_sli4_calc_ring(struct lpfc_hba *phba, struct lpfc_iocbq *piocb)
{
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (piocb->iocb_flag & (LPFC_IO_FCP | LPFC_USE_FCPWQIDX)) {
if (!(phba->cfg_fof) ||
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
(!(piocb->iocb_flag & LPFC_IO_FOF))) {
if (unlikely(!phba->sli4_hba.fcp_wq))
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
return NULL;
/*
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* for abort iocb hba_wqidx should already
* be setup based on what work queue we used.
*/
if (!(piocb->iocb_flag & LPFC_USE_FCPWQIDX)) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
piocb->hba_wqidx =
lpfc_sli4_scmd_to_wqidx_distr(phba,
piocb->context1);
piocb->hba_wqidx = piocb->hba_wqidx %
phba->cfg_fcp_io_channel;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
return phba->sli4_hba.fcp_wq[piocb->hba_wqidx]->pring;
} else {
if (unlikely(!phba->sli4_hba.oas_wq))
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
return NULL;
piocb->hba_wqidx = 0;
return phba->sli4_hba.oas_wq->pring;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
} else {
if (unlikely(!phba->sli4_hba.els_wq))
return NULL;
piocb->hba_wqidx = 0;
return phba->sli4_hba.els_wq->pring;
}
}
/**
* lpfc_sli_issue_iocb - Wrapper function for __lpfc_sli_issue_iocb
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @piocb: Pointer to command iocb.
* @flag: Flag indicating if this command can be put into txq.
*
* lpfc_sli_issue_iocb is a wrapper around __lpfc_sli_issue_iocb
* function. This function gets the hbalock and calls
* __lpfc_sli_issue_iocb function and will return the error returned
* by __lpfc_sli_issue_iocb function. This wrapper is used by
* functions which do not hold hbalock.
**/
int
lpfc_sli_issue_iocb(struct lpfc_hba *phba, uint32_t ring_number,
struct lpfc_iocbq *piocb, uint32_t flag)
{
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_hba_eq_hdl *hba_eq_hdl;
struct lpfc_sli_ring *pring;
struct lpfc_queue *fpeq;
struct lpfc_eqe *eqe;
unsigned long iflags;
int rc, idx;
if (phba->sli_rev == LPFC_SLI_REV4) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = lpfc_sli4_calc_ring(phba, piocb);
if (unlikely(pring == NULL))
return IOCB_ERROR;
spin_lock_irqsave(&pring->ring_lock, iflags);
rc = __lpfc_sli_issue_iocb(phba, ring_number, piocb, flag);
spin_unlock_irqrestore(&pring->ring_lock, iflags);
if (lpfc_fcp_look_ahead && (piocb->iocb_flag & LPFC_IO_FCP)) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
idx = piocb->hba_wqidx;
hba_eq_hdl = &phba->sli4_hba.hba_eq_hdl[idx];
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (atomic_dec_and_test(&hba_eq_hdl->hba_eq_in_use)) {
/* Get associated EQ with this index */
fpeq = phba->sli4_hba.hba_eq[idx];
/* Turn off interrupts from this EQ */
phba->sli4_hba.sli4_eq_clr_intr(fpeq);
/*
* Process all the events on FCP EQ
*/
while ((eqe = lpfc_sli4_eq_get(fpeq))) {
lpfc_sli4_hba_handle_eqe(phba,
eqe, idx);
fpeq->EQ_processed++;
}
/* Always clear and re-arm the EQ */
phba->sli4_hba.sli4_eq_release(fpeq,
LPFC_QUEUE_REARM);
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
atomic_inc(&hba_eq_hdl->hba_eq_in_use);
}
} else {
/* For now, SLI2/3 will still use hbalock */
spin_lock_irqsave(&phba->hbalock, iflags);
rc = __lpfc_sli_issue_iocb(phba, ring_number, piocb, flag);
spin_unlock_irqrestore(&phba->hbalock, iflags);
}
return rc;
}
/**
* lpfc_extra_ring_setup - Extra ring setup function
* @phba: Pointer to HBA context object.
*
* This function is called while driver attaches with the
* HBA to setup the extra ring. The extra ring is used
* only when driver needs to support target mode functionality
* or IP over FC functionalities.
*
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* This function is called with no lock held. SLI3 only.
**/
static int
lpfc_extra_ring_setup( struct lpfc_hba *phba)
{
struct lpfc_sli *psli;
struct lpfc_sli_ring *pring;
psli = &phba->sli;
/* Adjust cmd/rsp ring iocb entries more evenly */
/* Take some away from the FCP ring */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &psli->sli3_ring[LPFC_FCP_RING];
pring->sli.sli3.numCiocb -= SLI2_IOCB_CMD_R1XTRA_ENTRIES;
pring->sli.sli3.numRiocb -= SLI2_IOCB_RSP_R1XTRA_ENTRIES;
pring->sli.sli3.numCiocb -= SLI2_IOCB_CMD_R3XTRA_ENTRIES;
pring->sli.sli3.numRiocb -= SLI2_IOCB_RSP_R3XTRA_ENTRIES;
/* and give them to the extra ring */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &psli->sli3_ring[LPFC_EXTRA_RING];
pring->sli.sli3.numCiocb += SLI2_IOCB_CMD_R1XTRA_ENTRIES;
pring->sli.sli3.numRiocb += SLI2_IOCB_RSP_R1XTRA_ENTRIES;
pring->sli.sli3.numCiocb += SLI2_IOCB_CMD_R3XTRA_ENTRIES;
pring->sli.sli3.numRiocb += SLI2_IOCB_RSP_R3XTRA_ENTRIES;
/* Setup default profile for this ring */
pring->iotag_max = 4096;
pring->num_mask = 1;
pring->prt[0].profile = 0; /* Mask 0 */
pring->prt[0].rctl = phba->cfg_multi_ring_rctl;
pring->prt[0].type = phba->cfg_multi_ring_type;
pring->prt[0].lpfc_sli_rcv_unsol_event = NULL;
return 0;
}
/* lpfc_sli_abts_err_handler - handle a failed ABTS request from an SLI3 port.
* @phba: Pointer to HBA context object.
* @iocbq: Pointer to iocb object.
*
* The async_event handler calls this routine when it receives
* an ASYNC_STATUS_CN event from the port. The port generates
* this event when an Abort Sequence request to an rport fails
* twice in succession. The abort could be originated by the
* driver or by the port. The ABTS could have been for an ELS
* or FCP IO. The port only generates this event when an ABTS
* fails to complete after one retry.
*/
static void
lpfc_sli_abts_err_handler(struct lpfc_hba *phba,
struct lpfc_iocbq *iocbq)
{
struct lpfc_nodelist *ndlp = NULL;
uint16_t rpi = 0, vpi = 0;
struct lpfc_vport *vport = NULL;
/* The rpi in the ulpContext is vport-sensitive. */
vpi = iocbq->iocb.un.asyncstat.sub_ctxt_tag;
rpi = iocbq->iocb.ulpContext;
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"3092 Port generated ABTS async event "
"on vpi %d rpi %d status 0x%x\n",
vpi, rpi, iocbq->iocb.ulpStatus);
vport = lpfc_find_vport_by_vpid(phba, vpi);
if (!vport)
goto err_exit;
ndlp = lpfc_findnode_rpi(vport, rpi);
if (!ndlp || !NLP_CHK_NODE_ACT(ndlp))
goto err_exit;
if (iocbq->iocb.ulpStatus == IOSTAT_LOCAL_REJECT)
lpfc_sli_abts_recover_port(vport, ndlp);
return;
err_exit:
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"3095 Event Context not found, no "
"action on vpi %d rpi %d status 0x%x, reason 0x%x\n",
iocbq->iocb.ulpContext, iocbq->iocb.ulpStatus,
vpi, rpi);
}
/* lpfc_sli4_abts_err_handler - handle a failed ABTS request from an SLI4 port.
* @phba: pointer to HBA context object.
* @ndlp: nodelist pointer for the impacted rport.
* @axri: pointer to the wcqe containing the failed exchange.
*
* The driver calls this routine when it receives an ABORT_XRI_FCP CQE from the
* port. The port generates this event when an abort exchange request to an
* rport fails twice in succession with no reply. The abort could be originated
* by the driver or by the port. The ABTS could have been for an ELS or FCP IO.
*/
void
lpfc_sli4_abts_err_handler(struct lpfc_hba *phba,
struct lpfc_nodelist *ndlp,
struct sli4_wcqe_xri_aborted *axri)
{
struct lpfc_vport *vport;
uint32_t ext_status = 0;
if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"3115 Node Context not found, driver "
"ignoring abts err event\n");
return;
}
vport = ndlp->vport;
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"3116 Port generated FCP XRI ABORT event on "
"vpi %d rpi %d xri x%x status 0x%x parameter x%x\n",
ndlp->vport->vpi, phba->sli4_hba.rpi_ids[ndlp->nlp_rpi],
bf_get(lpfc_wcqe_xa_xri, axri),
bf_get(lpfc_wcqe_xa_status, axri),
axri->parameter);
/*
* Catch the ABTS protocol failure case. Older OCe FW releases returned
* LOCAL_REJECT and 0 for a failed ABTS exchange and later OCe and
* LPe FW releases returned LOCAL_REJECT and SEQUENCE_TIMEOUT.
*/
ext_status = axri->parameter & IOERR_PARAM_MASK;
if ((bf_get(lpfc_wcqe_xa_status, axri) == IOSTAT_LOCAL_REJECT) &&
((ext_status == IOERR_SEQUENCE_TIMEOUT) || (ext_status == 0)))
lpfc_sli_abts_recover_port(vport, ndlp);
}
/**
* lpfc_sli_async_event_handler - ASYNC iocb handler function
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @iocbq: Pointer to iocb object.
*
* This function is called by the slow ring event handler
* function when there is an ASYNC event iocb in the ring.
* This function is called with no lock held.
* Currently this function handles only temperature related
* ASYNC events. The function decodes the temperature sensor
* event message and posts events for the management applications.
**/
static void
lpfc_sli_async_event_handler(struct lpfc_hba * phba,
struct lpfc_sli_ring * pring, struct lpfc_iocbq * iocbq)
{
IOCB_t *icmd;
uint16_t evt_code;
struct temp_event temp_event_data;
struct Scsi_Host *shost;
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
uint32_t *iocb_w;
icmd = &iocbq->iocb;
evt_code = icmd->un.asyncstat.evt_code;
switch (evt_code) {
case ASYNC_TEMP_WARN:
case ASYNC_TEMP_SAFE:
temp_event_data.data = (uint32_t) icmd->ulpContext;
temp_event_data.event_type = FC_REG_TEMPERATURE_EVENT;
if (evt_code == ASYNC_TEMP_WARN) {
temp_event_data.event_code = LPFC_THRESHOLD_TEMP;
lpfc_printf_log(phba, KERN_ERR, LOG_TEMP,
"0347 Adapter is very hot, please take "
"corrective action. temperature : %d Celsius\n",
(uint32_t) icmd->ulpContext);
} else {
temp_event_data.event_code = LPFC_NORMAL_TEMP;
lpfc_printf_log(phba, KERN_ERR, LOG_TEMP,
"0340 Adapter temperature is OK now. "
"temperature : %d Celsius\n",
(uint32_t) icmd->ulpContext);
}
/* Send temperature change event to applications */
shost = lpfc_shost_from_vport(phba->pport);
fc_host_post_vendor_event(shost, fc_get_event_number(),
sizeof(temp_event_data), (char *) &temp_event_data,
LPFC_NL_VENDOR_ID);
break;
case ASYNC_STATUS_CN:
lpfc_sli_abts_err_handler(phba, iocbq);
break;
default:
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
iocb_w = (uint32_t *) icmd;
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0346 Ring %d handler: unexpected ASYNC_STATUS"
" evt_code 0x%x\n"
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
"W0 0x%08x W1 0x%08x W2 0x%08x W3 0x%08x\n"
"W4 0x%08x W5 0x%08x W6 0x%08x W7 0x%08x\n"
"W8 0x%08x W9 0x%08x W10 0x%08x W11 0x%08x\n"
"W12 0x%08x W13 0x%08x W14 0x%08x W15 0x%08x\n",
pring->ringno, icmd->un.asyncstat.evt_code,
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
iocb_w[0], iocb_w[1], iocb_w[2], iocb_w[3],
iocb_w[4], iocb_w[5], iocb_w[6], iocb_w[7],
iocb_w[8], iocb_w[9], iocb_w[10], iocb_w[11],
iocb_w[12], iocb_w[13], iocb_w[14], iocb_w[15]);
break;
}
}
/**
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* lpfc_sli4_setup - SLI ring setup function
* @phba: Pointer to HBA context object.
*
* lpfc_sli_setup sets up rings of the SLI interface with
* number of iocbs per ring and iotags. This function is
* called while driver attach to the HBA and before the
* interrupts are enabled. So there is no need for locking.
*
* This function always returns 0.
**/
int
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli4_setup(struct lpfc_hba *phba)
{
struct lpfc_sli_ring *pring;
pring = phba->sli4_hba.els_wq->pring;
pring->num_mask = LPFC_MAX_RING_MASK;
pring->prt[0].profile = 0; /* Mask 0 */
pring->prt[0].rctl = FC_RCTL_ELS_REQ;
pring->prt[0].type = FC_TYPE_ELS;
pring->prt[0].lpfc_sli_rcv_unsol_event =
lpfc_els_unsol_event;
pring->prt[1].profile = 0; /* Mask 1 */
pring->prt[1].rctl = FC_RCTL_ELS_REP;
pring->prt[1].type = FC_TYPE_ELS;
pring->prt[1].lpfc_sli_rcv_unsol_event =
lpfc_els_unsol_event;
pring->prt[2].profile = 0; /* Mask 2 */
/* NameServer Inquiry */
pring->prt[2].rctl = FC_RCTL_DD_UNSOL_CTL;
/* NameServer */
pring->prt[2].type = FC_TYPE_CT;
pring->prt[2].lpfc_sli_rcv_unsol_event =
lpfc_ct_unsol_event;
pring->prt[3].profile = 0; /* Mask 3 */
/* NameServer response */
pring->prt[3].rctl = FC_RCTL_DD_SOL_CTL;
/* NameServer */
pring->prt[3].type = FC_TYPE_CT;
pring->prt[3].lpfc_sli_rcv_unsol_event =
lpfc_ct_unsol_event;
return 0;
}
/**
* lpfc_sli_setup - SLI ring setup function
* @phba: Pointer to HBA context object.
*
* lpfc_sli_setup sets up rings of the SLI interface with
* number of iocbs per ring and iotags. This function is
* called while driver attach to the HBA and before the
* interrupts are enabled. So there is no need for locking.
*
* This function always returns 0. SLI3 only.
**/
int
lpfc_sli_setup(struct lpfc_hba *phba)
{
int i, totiocbsize = 0;
struct lpfc_sli *psli = &phba->sli;
struct lpfc_sli_ring *pring;
psli->num_rings = MAX_SLI3_CONFIGURED_RINGS;
psli->sli_flag = 0;
psli->iocbq_lookup = NULL;
psli->iocbq_lookup_len = 0;
psli->last_iotag = 0;
for (i = 0; i < psli->num_rings; i++) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &psli->sli3_ring[i];
switch (i) {
case LPFC_FCP_RING: /* ring 0 - FCP */
/* numCiocb and numRiocb are used in config_port */
pring->sli.sli3.numCiocb = SLI2_IOCB_CMD_R0_ENTRIES;
pring->sli.sli3.numRiocb = SLI2_IOCB_RSP_R0_ENTRIES;
pring->sli.sli3.numCiocb +=
SLI2_IOCB_CMD_R1XTRA_ENTRIES;
pring->sli.sli3.numRiocb +=
SLI2_IOCB_RSP_R1XTRA_ENTRIES;
pring->sli.sli3.numCiocb +=
SLI2_IOCB_CMD_R3XTRA_ENTRIES;
pring->sli.sli3.numRiocb +=
SLI2_IOCB_RSP_R3XTRA_ENTRIES;
pring->sli.sli3.sizeCiocb = (phba->sli_rev == 3) ?
SLI3_IOCB_CMD_SIZE :
SLI2_IOCB_CMD_SIZE;
pring->sli.sli3.sizeRiocb = (phba->sli_rev == 3) ?
SLI3_IOCB_RSP_SIZE :
SLI2_IOCB_RSP_SIZE;
pring->iotag_ctr = 0;
pring->iotag_max =
(phba->cfg_hba_queue_depth * 2);
pring->fast_iotag = pring->iotag_max;
pring->num_mask = 0;
break;
case LPFC_EXTRA_RING: /* ring 1 - EXTRA */
/* numCiocb and numRiocb are used in config_port */
pring->sli.sli3.numCiocb = SLI2_IOCB_CMD_R1_ENTRIES;
pring->sli.sli3.numRiocb = SLI2_IOCB_RSP_R1_ENTRIES;
pring->sli.sli3.sizeCiocb = (phba->sli_rev == 3) ?
SLI3_IOCB_CMD_SIZE :
SLI2_IOCB_CMD_SIZE;
pring->sli.sli3.sizeRiocb = (phba->sli_rev == 3) ?
SLI3_IOCB_RSP_SIZE :
SLI2_IOCB_RSP_SIZE;
pring->iotag_max = phba->cfg_hba_queue_depth;
pring->num_mask = 0;
break;
case LPFC_ELS_RING: /* ring 2 - ELS / CT */
/* numCiocb and numRiocb are used in config_port */
pring->sli.sli3.numCiocb = SLI2_IOCB_CMD_R2_ENTRIES;
pring->sli.sli3.numRiocb = SLI2_IOCB_RSP_R2_ENTRIES;
pring->sli.sli3.sizeCiocb = (phba->sli_rev == 3) ?
SLI3_IOCB_CMD_SIZE :
SLI2_IOCB_CMD_SIZE;
pring->sli.sli3.sizeRiocb = (phba->sli_rev == 3) ?
SLI3_IOCB_RSP_SIZE :
SLI2_IOCB_RSP_SIZE;
pring->fast_iotag = 0;
pring->iotag_ctr = 0;
pring->iotag_max = 4096;
pring->lpfc_sli_rcv_async_status =
lpfc_sli_async_event_handler;
pring->num_mask = LPFC_MAX_RING_MASK;
pring->prt[0].profile = 0; /* Mask 0 */
pring->prt[0].rctl = FC_RCTL_ELS_REQ;
pring->prt[0].type = FC_TYPE_ELS;
pring->prt[0].lpfc_sli_rcv_unsol_event =
lpfc_els_unsol_event;
pring->prt[1].profile = 0; /* Mask 1 */
pring->prt[1].rctl = FC_RCTL_ELS_REP;
pring->prt[1].type = FC_TYPE_ELS;
pring->prt[1].lpfc_sli_rcv_unsol_event =
lpfc_els_unsol_event;
pring->prt[2].profile = 0; /* Mask 2 */
/* NameServer Inquiry */
pring->prt[2].rctl = FC_RCTL_DD_UNSOL_CTL;
/* NameServer */
pring->prt[2].type = FC_TYPE_CT;
pring->prt[2].lpfc_sli_rcv_unsol_event =
lpfc_ct_unsol_event;
pring->prt[3].profile = 0; /* Mask 3 */
/* NameServer response */
pring->prt[3].rctl = FC_RCTL_DD_SOL_CTL;
/* NameServer */
pring->prt[3].type = FC_TYPE_CT;
pring->prt[3].lpfc_sli_rcv_unsol_event =
lpfc_ct_unsol_event;
break;
}
totiocbsize += (pring->sli.sli3.numCiocb *
pring->sli.sli3.sizeCiocb) +
(pring->sli.sli3.numRiocb * pring->sli.sli3.sizeRiocb);
}
if (totiocbsize > MAX_SLIM_IOCB_SIZE) {
/* Too many cmd / rsp ring entries in SLI2 SLIM */
printk(KERN_ERR "%d:0462 Too many cmd / rsp ring entries in "
"SLI2 SLIM Data: x%x x%lx\n",
phba->brd_no, totiocbsize,
(unsigned long) MAX_SLIM_IOCB_SIZE);
}
if (phba->cfg_multi_ring_support == 2)
lpfc_extra_ring_setup(phba);
return 0;
}
/**
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* lpfc_sli4_queue_init - Queue initialization function
* @phba: Pointer to HBA context object.
*
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* lpfc_sli4_queue_init sets up mailbox queues and iocb queues for each
* ring. This function also initializes ring indices of each ring.
* This function is called during the initialization of the SLI
* interface of an HBA.
* This function is called with no lock held and always returns
* 1.
**/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
void
lpfc_sli4_queue_init(struct lpfc_hba *phba)
{
struct lpfc_sli *psli;
struct lpfc_sli_ring *pring;
int i;
psli = &phba->sli;
spin_lock_irq(&phba->hbalock);
INIT_LIST_HEAD(&psli->mboxq);
INIT_LIST_HEAD(&psli->mboxq_cmpl);
/* Initialize list headers for txq and txcmplq as double linked lists */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (i = 0; i < phba->cfg_fcp_io_channel; i++) {
pring = phba->sli4_hba.fcp_wq[i]->pring;
pring->flag = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring->ringno = LPFC_FCP_RING;
INIT_LIST_HEAD(&pring->txq);
INIT_LIST_HEAD(&pring->txcmplq);
INIT_LIST_HEAD(&pring->iocb_continueq);
spin_lock_init(&pring->ring_lock);
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (i = 0; i < phba->cfg_nvme_io_channel; i++) {
pring = phba->sli4_hba.nvme_wq[i]->pring;
pring->flag = 0;
pring->ringno = LPFC_FCP_RING;
INIT_LIST_HEAD(&pring->txq);
INIT_LIST_HEAD(&pring->txcmplq);
INIT_LIST_HEAD(&pring->iocb_continueq);
spin_lock_init(&pring->ring_lock);
}
pring = phba->sli4_hba.els_wq->pring;
pring->flag = 0;
pring->ringno = LPFC_ELS_RING;
INIT_LIST_HEAD(&pring->txq);
INIT_LIST_HEAD(&pring->txcmplq);
INIT_LIST_HEAD(&pring->iocb_continueq);
spin_lock_init(&pring->ring_lock);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->cfg_nvme_io_channel) {
pring = phba->sli4_hba.nvmels_wq->pring;
pring->flag = 0;
pring->ringno = LPFC_ELS_RING;
INIT_LIST_HEAD(&pring->txq);
INIT_LIST_HEAD(&pring->txcmplq);
INIT_LIST_HEAD(&pring->iocb_continueq);
spin_lock_init(&pring->ring_lock);
}
if (phba->cfg_fof) {
pring = phba->sli4_hba.oas_wq->pring;
pring->flag = 0;
pring->ringno = LPFC_FCP_RING;
INIT_LIST_HEAD(&pring->txq);
INIT_LIST_HEAD(&pring->txcmplq);
INIT_LIST_HEAD(&pring->iocb_continueq);
spin_lock_init(&pring->ring_lock);
}
spin_unlock_irq(&phba->hbalock);
}
/**
* lpfc_sli_queue_init - Queue initialization function
* @phba: Pointer to HBA context object.
*
* lpfc_sli_queue_init sets up mailbox queues and iocb queues for each
* ring. This function also initializes ring indices of each ring.
* This function is called during the initialization of the SLI
* interface of an HBA.
* This function is called with no lock held and always returns
* 1.
**/
void
lpfc_sli_queue_init(struct lpfc_hba *phba)
{
struct lpfc_sli *psli;
struct lpfc_sli_ring *pring;
int i;
psli = &phba->sli;
spin_lock_irq(&phba->hbalock);
INIT_LIST_HEAD(&psli->mboxq);
INIT_LIST_HEAD(&psli->mboxq_cmpl);
/* Initialize list headers for txq and txcmplq as double linked lists */
for (i = 0; i < psli->num_rings; i++) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &psli->sli3_ring[i];
pring->ringno = i;
pring->sli.sli3.next_cmdidx = 0;
pring->sli.sli3.local_getidx = 0;
pring->sli.sli3.cmdidx = 0;
INIT_LIST_HEAD(&pring->iocb_continueq);
INIT_LIST_HEAD(&pring->iocb_continue_saveq);
INIT_LIST_HEAD(&pring->postbufq);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring->flag = 0;
INIT_LIST_HEAD(&pring->txq);
INIT_LIST_HEAD(&pring->txcmplq);
spin_lock_init(&pring->ring_lock);
}
spin_unlock_irq(&phba->hbalock);
}
/**
* lpfc_sli_mbox_sys_flush - Flush mailbox command sub-system
* @phba: Pointer to HBA context object.
*
* This routine flushes the mailbox command subsystem. It will unconditionally
* flush all the mailbox commands in the three possible stages in the mailbox
* command sub-system: pending mailbox command queue; the outstanding mailbox
* command; and completed mailbox command queue. It is caller's responsibility
* to make sure that the driver is in the proper state to flush the mailbox
* command sub-system. Namely, the posting of mailbox commands into the
* pending mailbox command queue from the various clients must be stopped;
* either the HBA is in a state that it will never works on the outstanding
* mailbox command (such as in EEH or ERATT conditions) or the outstanding
* mailbox command has been completed.
**/
static void
lpfc_sli_mbox_sys_flush(struct lpfc_hba *phba)
{
LIST_HEAD(completions);
struct lpfc_sli *psli = &phba->sli;
LPFC_MBOXQ_t *pmb;
unsigned long iflag;
/* Flush all the mailbox commands in the mbox system */
spin_lock_irqsave(&phba->hbalock, iflag);
/* The pending mailbox command queue */
list_splice_init(&phba->sli.mboxq, &completions);
/* The outstanding active mailbox command */
if (psli->mbox_active) {
list_add_tail(&psli->mbox_active->list, &completions);
psli->mbox_active = NULL;
psli->sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
}
/* The completed mailbox command queue */
list_splice_init(&phba->sli.mboxq_cmpl, &completions);
spin_unlock_irqrestore(&phba->hbalock, iflag);
/* Return all flushed mailbox commands with MBX_NOT_FINISHED status */
while (!list_empty(&completions)) {
list_remove_head(&completions, pmb, LPFC_MBOXQ_t, list);
pmb->u.mb.mbxStatus = MBX_NOT_FINISHED;
if (pmb->mbox_cmpl)
pmb->mbox_cmpl(phba, pmb);
}
}
/**
* lpfc_sli_host_down - Vport cleanup function
* @vport: Pointer to virtual port object.
*
* lpfc_sli_host_down is called to clean up the resources
* associated with a vport before destroying virtual
* port data structures.
* This function does following operations:
* - Free discovery resources associated with this virtual
* port.
* - Free iocbs associated with this virtual port in
* the txq.
* - Send abort for all iocb commands associated with this
* vport in txcmplq.
*
* This function is called with no lock held and always returns 1.
**/
int
lpfc_sli_host_down(struct lpfc_vport *vport)
{
LIST_HEAD(completions);
struct lpfc_hba *phba = vport->phba;
struct lpfc_sli *psli = &phba->sli;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_queue *qp = NULL;
struct lpfc_sli_ring *pring;
struct lpfc_iocbq *iocb, *next_iocb;
int i;
unsigned long flags = 0;
uint16_t prev_pring_flag;
lpfc_cleanup_discovery_resources(vport);
spin_lock_irqsave(&phba->hbalock, flags);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/*
* Error everything on the txq since these iocbs
* have not been given to the FW yet.
* Also issue ABTS for everything on the txcmplq
*/
if (phba->sli_rev != LPFC_SLI_REV4) {
for (i = 0; i < psli->num_rings; i++) {
pring = &psli->sli3_ring[i];
prev_pring_flag = pring->flag;
/* Only slow rings */
if (pring->ringno == LPFC_ELS_RING) {
pring->flag |= LPFC_DEFERRED_RING_EVENT;
/* Set the lpfc data pending flag */
set_bit(LPFC_DATA_READY, &phba->data_flags);
}
list_for_each_entry_safe(iocb, next_iocb,
&pring->txq, list) {
if (iocb->vport != vport)
continue;
list_move_tail(&iocb->list, &completions);
}
list_for_each_entry_safe(iocb, next_iocb,
&pring->txcmplq, list) {
if (iocb->vport != vport)
continue;
lpfc_sli_issue_abort_iotag(phba, pring, iocb);
}
pring->flag = prev_pring_flag;
}
} else {
list_for_each_entry(qp, &phba->sli4_hba.lpfc_wq_list, wq_list) {
pring = qp->pring;
if (!pring)
continue;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (pring == phba->sli4_hba.els_wq->pring) {
pring->flag |= LPFC_DEFERRED_RING_EVENT;
/* Set the lpfc data pending flag */
set_bit(LPFC_DATA_READY, &phba->data_flags);
}
prev_pring_flag = pring->flag;
spin_lock_irq(&pring->ring_lock);
list_for_each_entry_safe(iocb, next_iocb,
&pring->txq, list) {
if (iocb->vport != vport)
continue;
list_move_tail(&iocb->list, &completions);
}
spin_unlock_irq(&pring->ring_lock);
list_for_each_entry_safe(iocb, next_iocb,
&pring->txcmplq, list) {
if (iocb->vport != vport)
continue;
lpfc_sli_issue_abort_iotag(phba, pring, iocb);
}
pring->flag = prev_pring_flag;
}
}
spin_unlock_irqrestore(&phba->hbalock, flags);
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/* Cancel all the IOCBs from the completions list */
lpfc_sli_cancel_iocbs(phba, &completions, IOSTAT_LOCAL_REJECT,
IOERR_SLI_DOWN);
return 1;
}
/**
* lpfc_sli_hba_down - Resource cleanup function for the HBA
* @phba: Pointer to HBA context object.
*
* This function cleans up all iocb, buffers, mailbox commands
* while shutting down the HBA. This function is called with no
* lock held and always returns 1.
* This function does the following to cleanup driver resources:
* - Free discovery resources for each virtual port
* - Cleanup any pending fabric iocbs
* - Iterate through the iocb txq and free each entry
* in the list.
* - Free up any buffer posted to the HBA
* - Free mailbox commands in the mailbox queue.
**/
int
lpfc_sli_hba_down(struct lpfc_hba *phba)
{
LIST_HEAD(completions);
struct lpfc_sli *psli = &phba->sli;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_queue *qp = NULL;
struct lpfc_sli_ring *pring;
struct lpfc_dmabuf *buf_ptr;
unsigned long flags = 0;
int i;
/* Shutdown the mailbox command sub-system */
lpfc_sli_mbox_sys_shutdown(phba, LPFC_MBX_WAIT);
lpfc_hba_down_prep(phba);
lpfc_fabric_abort_hba(phba);
spin_lock_irqsave(&phba->hbalock, flags);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/*
* Error everything on the txq since these iocbs
* have not been given to the FW yet.
*/
if (phba->sli_rev != LPFC_SLI_REV4) {
for (i = 0; i < psli->num_rings; i++) {
pring = &psli->sli3_ring[i];
/* Only slow rings */
if (pring->ringno == LPFC_ELS_RING) {
pring->flag |= LPFC_DEFERRED_RING_EVENT;
/* Set the lpfc data pending flag */
set_bit(LPFC_DATA_READY, &phba->data_flags);
}
list_splice_init(&pring->txq, &completions);
}
} else {
list_for_each_entry(qp, &phba->sli4_hba.lpfc_wq_list, wq_list) {
pring = qp->pring;
if (!pring)
continue;
spin_lock_irq(&pring->ring_lock);
list_splice_init(&pring->txq, &completions);
spin_unlock_irq(&pring->ring_lock);
if (pring == phba->sli4_hba.els_wq->pring) {
pring->flag |= LPFC_DEFERRED_RING_EVENT;
/* Set the lpfc data pending flag */
set_bit(LPFC_DATA_READY, &phba->data_flags);
}
}
}
spin_unlock_irqrestore(&phba->hbalock, flags);
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/* Cancel all the IOCBs from the completions list */
lpfc_sli_cancel_iocbs(phba, &completions, IOSTAT_LOCAL_REJECT,
IOERR_SLI_DOWN);
spin_lock_irqsave(&phba->hbalock, flags);
list_splice_init(&phba->elsbuf, &completions);
phba->elsbuf_cnt = 0;
phba->elsbuf_prev_cnt = 0;
spin_unlock_irqrestore(&phba->hbalock, flags);
while (!list_empty(&completions)) {
list_remove_head(&completions, buf_ptr,
struct lpfc_dmabuf, list);
lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
kfree(buf_ptr);
}
/* Return any active mbox cmds */
del_timer_sync(&psli->mbox_tmo);
spin_lock_irqsave(&phba->pport->work_port_lock, flags);
phba->pport->work_port_events &= ~WORKER_MBOX_TMO;
spin_unlock_irqrestore(&phba->pport->work_port_lock, flags);
return 1;
}
/**
* lpfc_sli_pcimem_bcopy - SLI memory copy function
* @srcp: Source memory pointer.
* @destp: Destination memory pointer.
* @cnt: Number of words required to be copied.
*
* This function is used for copying data between driver memory
* and the SLI memory. This function also changes the endianness
* of each word if native endianness is different from SLI
* endianness. This function can be called with or without
* lock.
**/
void
lpfc_sli_pcimem_bcopy(void *srcp, void *destp, uint32_t cnt)
{
uint32_t *src = srcp;
uint32_t *dest = destp;
uint32_t ldata;
int i;
for (i = 0; i < (int)cnt; i += sizeof (uint32_t)) {
ldata = *src;
ldata = le32_to_cpu(ldata);
*dest = ldata;
src++;
dest++;
}
}
/**
* lpfc_sli_bemem_bcopy - SLI memory copy function
* @srcp: Source memory pointer.
* @destp: Destination memory pointer.
* @cnt: Number of words required to be copied.
*
* This function is used for copying data between a data structure
* with big endian representation to local endianness.
* This function can be called with or without lock.
**/
void
lpfc_sli_bemem_bcopy(void *srcp, void *destp, uint32_t cnt)
{
uint32_t *src = srcp;
uint32_t *dest = destp;
uint32_t ldata;
int i;
for (i = 0; i < (int)cnt; i += sizeof(uint32_t)) {
ldata = *src;
ldata = be32_to_cpu(ldata);
*dest = ldata;
src++;
dest++;
}
}
/**
* lpfc_sli_ringpostbuf_put - Function to add a buffer to postbufq
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @mp: Pointer to driver buffer object.
*
* This function is called with no lock held.
* It always return zero after adding the buffer to the postbufq
* buffer list.
**/
int
lpfc_sli_ringpostbuf_put(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_dmabuf *mp)
{
/* Stick struct lpfc_dmabuf at end of postbufq so driver can look it up
later */
spin_lock_irq(&phba->hbalock);
list_add_tail(&mp->list, &pring->postbufq);
pring->postbufq_cnt++;
spin_unlock_irq(&phba->hbalock);
return 0;
}
/**
* lpfc_sli_get_buffer_tag - allocates a tag for a CMD_QUE_XRI64_CX buffer
* @phba: Pointer to HBA context object.
*
* When HBQ is enabled, buffers are searched based on tags. This function
* allocates a tag for buffer posted using CMD_QUE_XRI64_CX iocb. The
* tag is bit wise or-ed with QUE_BUFTAG_BIT to make sure that the tag
* does not conflict with tags of buffer posted for unsolicited events.
* The function returns the allocated tag. The function is called with
* no locks held.
**/
uint32_t
lpfc_sli_get_buffer_tag(struct lpfc_hba *phba)
{
spin_lock_irq(&phba->hbalock);
phba->buffer_tag_count++;
/*
* Always set the QUE_BUFTAG_BIT to distiguish between
* a tag assigned by HBQ.
*/
phba->buffer_tag_count |= QUE_BUFTAG_BIT;
spin_unlock_irq(&phba->hbalock);
return phba->buffer_tag_count;
}
/**
* lpfc_sli_ring_taggedbuf_get - find HBQ buffer associated with given tag
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @tag: Buffer tag.
*
* Buffers posted using CMD_QUE_XRI64_CX iocb are in pring->postbufq
* list. After HBA DMA data to these buffers, CMD_IOCB_RET_XRI64_CX
* iocb is posted to the response ring with the tag of the buffer.
* This function searches the pring->postbufq list using the tag
* to find buffer associated with CMD_IOCB_RET_XRI64_CX
* iocb. If the buffer is found then lpfc_dmabuf object of the
* buffer is returned to the caller else NULL is returned.
* This function is called with no lock held.
**/
struct lpfc_dmabuf *
lpfc_sli_ring_taggedbuf_get(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
uint32_t tag)
{
struct lpfc_dmabuf *mp, *next_mp;
struct list_head *slp = &pring->postbufq;
/* Search postbufq, from the beginning, looking for a match on tag */
spin_lock_irq(&phba->hbalock);
list_for_each_entry_safe(mp, next_mp, &pring->postbufq, list) {
if (mp->buffer_tag == tag) {
list_del_init(&mp->list);
pring->postbufq_cnt--;
spin_unlock_irq(&phba->hbalock);
return mp;
}
}
spin_unlock_irq(&phba->hbalock);
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0402 Cannot find virtual addr for buffer tag on "
"ring %d Data x%lx x%p x%p x%x\n",
pring->ringno, (unsigned long) tag,
slp->next, slp->prev, pring->postbufq_cnt);
return NULL;
}
/**
* lpfc_sli_ringpostbuf_get - search buffers for unsolicited CT and ELS events
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @phys: DMA address of the buffer.
*
* This function searches the buffer list using the dma_address
* of unsolicited event to find the driver's lpfc_dmabuf object
* corresponding to the dma_address. The function returns the
* lpfc_dmabuf object if a buffer is found else it returns NULL.
* This function is called by the ct and els unsolicited event
* handlers to get the buffer associated with the unsolicited
* event.
*
* This function is called with no lock held.
**/
struct lpfc_dmabuf *
lpfc_sli_ringpostbuf_get(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
dma_addr_t phys)
{
struct lpfc_dmabuf *mp, *next_mp;
struct list_head *slp = &pring->postbufq;
/* Search postbufq, from the beginning, looking for a match on phys */
spin_lock_irq(&phba->hbalock);
list_for_each_entry_safe(mp, next_mp, &pring->postbufq, list) {
if (mp->phys == phys) {
list_del_init(&mp->list);
pring->postbufq_cnt--;
spin_unlock_irq(&phba->hbalock);
return mp;
}
}
spin_unlock_irq(&phba->hbalock);
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0410 Cannot find virtual addr for mapped buf on "
"ring %d Data x%llx x%p x%p x%x\n",
pring->ringno, (unsigned long long)phys,
slp->next, slp->prev, pring->postbufq_cnt);
return NULL;
}
/**
* lpfc_sli_abort_els_cmpl - Completion handler for the els abort iocbs
* @phba: Pointer to HBA context object.
* @cmdiocb: Pointer to driver command iocb object.
* @rspiocb: Pointer to driver response iocb object.
*
* This function is the completion handler for the abort iocbs for
* ELS commands. This function is called from the ELS ring event
* handler with no lock held. This function frees memory resources
* associated with the abort iocb.
**/
static void
lpfc_sli_abort_els_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
struct lpfc_iocbq *rspiocb)
{
IOCB_t *irsp = &rspiocb->iocb;
uint16_t abort_iotag, abort_context;
struct lpfc_iocbq *abort_iocb = NULL;
if (irsp->ulpStatus) {
/*
* Assume that the port already completed and returned, or
* will return the iocb. Just Log the message.
*/
abort_context = cmdiocb->iocb.un.acxri.abortContextTag;
abort_iotag = cmdiocb->iocb.un.acxri.abortIoTag;
spin_lock_irq(&phba->hbalock);
if (phba->sli_rev < LPFC_SLI_REV4) {
if (abort_iotag != 0 &&
abort_iotag <= phba->sli.last_iotag)
abort_iocb =
phba->sli.iocbq_lookup[abort_iotag];
} else
/* For sli4 the abort_tag is the XRI,
* so the abort routine puts the iotag of the iocb
* being aborted in the context field of the abort
* IOCB.
*/
abort_iocb = phba->sli.iocbq_lookup[abort_context];
lpfc_printf_log(phba, KERN_WARNING, LOG_ELS | LOG_SLI,
"0327 Cannot abort els iocb %p "
"with tag %x context %x, abort status %x, "
"abort code %x\n",
abort_iocb, abort_iotag, abort_context,
irsp->ulpStatus, irsp->un.ulpWord[4]);
spin_unlock_irq(&phba->hbalock);
}
lpfc_sli_release_iocbq(phba, cmdiocb);
return;
}
/**
* lpfc_ignore_els_cmpl - Completion handler for aborted ELS command
* @phba: Pointer to HBA context object.
* @cmdiocb: Pointer to driver command iocb object.
* @rspiocb: Pointer to driver response iocb object.
*
* The function is called from SLI ring event handler with no
* lock held. This function is the completion handler for ELS commands
* which are aborted. The function frees memory resources used for
* the aborted ELS commands.
**/
static void
lpfc_ignore_els_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
struct lpfc_iocbq *rspiocb)
{
IOCB_t *irsp = &rspiocb->iocb;
/* ELS cmd tag <ulpIoTag> completes */
lpfc_printf_log(phba, KERN_INFO, LOG_ELS,
"0139 Ignoring ELS cmd tag x%x completion Data: "
"x%x x%x x%x\n",
irsp->ulpIoTag, irsp->ulpStatus,
irsp->un.ulpWord[4], irsp->ulpTimeout);
if (cmdiocb->iocb.ulpCommand == CMD_GEN_REQUEST64_CR)
lpfc_ct_free_iocb(phba, cmdiocb);
else
lpfc_els_free_iocb(phba, cmdiocb);
return;
}
/**
* lpfc_sli_abort_iotag_issue - Issue abort for a command iocb
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @cmdiocb: Pointer to driver command iocb object.
*
* This function issues an abort iocb for the provided command iocb down to
* the port. Other than the case the outstanding command iocb is an abort
* request, this function issues abort out unconditionally. This function is
* called with hbalock held. The function returns 0 when it fails due to
* memory allocation failure or when the command iocb is an abort request.
**/
static int
lpfc_sli_abort_iotag_issue(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *cmdiocb)
{
struct lpfc_vport *vport = cmdiocb->vport;
struct lpfc_iocbq *abtsiocbp;
IOCB_t *icmd = NULL;
IOCB_t *iabt = NULL;
int retval;
unsigned long iflags;
lockdep_assert_held(&phba->hbalock);
/*
* There are certain command types we don't want to abort. And we
* don't want to abort commands that are already in the process of
* being aborted.
*/
icmd = &cmdiocb->iocb;
if (icmd->ulpCommand == CMD_ABORT_XRI_CN ||
icmd->ulpCommand == CMD_CLOSE_XRI_CN ||
(cmdiocb->iocb_flag & LPFC_DRIVER_ABORTED) != 0)
return 0;
/* issue ABTS for this IOCB based on iotag */
abtsiocbp = __lpfc_sli_get_iocbq(phba);
if (abtsiocbp == NULL)
return 0;
/* This signals the response to set the correct status
* before calling the completion handler
*/
cmdiocb->iocb_flag |= LPFC_DRIVER_ABORTED;
iabt = &abtsiocbp->iocb;
iabt->un.acxri.abortType = ABORT_TYPE_ABTS;
iabt->un.acxri.abortContextTag = icmd->ulpContext;
if (phba->sli_rev == LPFC_SLI_REV4) {
iabt->un.acxri.abortIoTag = cmdiocb->sli4_xritag;
iabt->un.acxri.abortContextTag = cmdiocb->iotag;
}
else
iabt->un.acxri.abortIoTag = icmd->ulpIoTag;
iabt->ulpLe = 1;
iabt->ulpClass = icmd->ulpClass;
/* ABTS WQE must go to the same WQ as the WQE to be aborted */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
abtsiocbp->hba_wqidx = cmdiocb->hba_wqidx;
if (cmdiocb->iocb_flag & LPFC_IO_FCP)
abtsiocbp->iocb_flag |= LPFC_USE_FCPWQIDX;
if (cmdiocb->iocb_flag & LPFC_IO_FOF)
abtsiocbp->iocb_flag |= LPFC_IO_FOF;
if (phba->link_state >= LPFC_LINK_UP)
iabt->ulpCommand = CMD_ABORT_XRI_CN;
else
iabt->ulpCommand = CMD_CLOSE_XRI_CN;
abtsiocbp->iocb_cmpl = lpfc_sli_abort_els_cmpl;
abtsiocbp->vport = vport;
lpfc_printf_vlog(vport, KERN_INFO, LOG_SLI,
"0339 Abort xri x%x, original iotag x%x, "
"abort cmd iotag x%x\n",
iabt->un.acxri.abortIoTag,
iabt->un.acxri.abortContextTag,
abtsiocbp->iotag);
if (phba->sli_rev == LPFC_SLI_REV4) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = lpfc_sli4_calc_ring(phba, abtsiocbp);
if (unlikely(pring == NULL))
return 0;
/* Note: both hbalock and ring_lock need to be set here */
spin_lock_irqsave(&pring->ring_lock, iflags);
retval = __lpfc_sli_issue_iocb(phba, pring->ringno,
abtsiocbp, 0);
spin_unlock_irqrestore(&pring->ring_lock, iflags);
} else {
retval = __lpfc_sli_issue_iocb(phba, pring->ringno,
abtsiocbp, 0);
}
if (retval)
__lpfc_sli_release_iocbq(phba, abtsiocbp);
/*
* Caller to this routine should check for IOCB_ERROR
* and handle it properly. This routine no longer removes
* iocb off txcmplq and call compl in case of IOCB_ERROR.
*/
return retval;
}
/**
* lpfc_sli_issue_abort_iotag - Abort function for a command iocb
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @cmdiocb: Pointer to driver command iocb object.
*
* This function issues an abort iocb for the provided command iocb. In case
* of unloading, the abort iocb will not be issued to commands on the ELS
* ring. Instead, the callback function shall be changed to those commands
* so that nothing happens when them finishes. This function is called with
* hbalock held. The function returns 0 when the command iocb is an abort
* request.
**/
int
lpfc_sli_issue_abort_iotag(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *cmdiocb)
{
struct lpfc_vport *vport = cmdiocb->vport;
int retval = IOCB_ERROR;
IOCB_t *icmd = NULL;
lockdep_assert_held(&phba->hbalock);
/*
* There are certain command types we don't want to abort. And we
* don't want to abort commands that are already in the process of
* being aborted.
*/
icmd = &cmdiocb->iocb;
if (icmd->ulpCommand == CMD_ABORT_XRI_CN ||
icmd->ulpCommand == CMD_CLOSE_XRI_CN ||
(cmdiocb->iocb_flag & LPFC_DRIVER_ABORTED) != 0)
return 0;
if (!pring) {
if (cmdiocb->iocb_flag & LPFC_IO_FABRIC)
cmdiocb->fabric_iocb_cmpl = lpfc_ignore_els_cmpl;
else
cmdiocb->iocb_cmpl = lpfc_ignore_els_cmpl;
goto abort_iotag_exit;
}
/*
* If we're unloading, don't abort iocb on the ELS ring, but change
* the callback so that nothing happens when it finishes.
*/
if ((vport->load_flag & FC_UNLOADING) &&
(pring->ringno == LPFC_ELS_RING)) {
if (cmdiocb->iocb_flag & LPFC_IO_FABRIC)
cmdiocb->fabric_iocb_cmpl = lpfc_ignore_els_cmpl;
else
cmdiocb->iocb_cmpl = lpfc_ignore_els_cmpl;
goto abort_iotag_exit;
}
/* Now, we try to issue the abort to the cmdiocb out */
retval = lpfc_sli_abort_iotag_issue(phba, pring, cmdiocb);
abort_iotag_exit:
/*
* Caller to this routine should check for IOCB_ERROR
* and handle it properly. This routine no longer removes
* iocb off txcmplq and call compl in case of IOCB_ERROR.
*/
return retval;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/**
* lpfc_sli4_abort_nvme_io - Issue abort for a command iocb
* @phba: Pointer to HBA context object.
* @pring: Pointer to driver SLI ring object.
* @cmdiocb: Pointer to driver command iocb object.
*
* This function issues an abort iocb for the provided command iocb down to
* the port. Other than the case the outstanding command iocb is an abort
* request, this function issues abort out unconditionally. This function is
* called with hbalock held. The function returns 0 when it fails due to
* memory allocation failure or when the command iocb is an abort request.
**/
static int
lpfc_sli4_abort_nvme_io(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
struct lpfc_iocbq *cmdiocb)
{
struct lpfc_vport *vport = cmdiocb->vport;
struct lpfc_iocbq *abtsiocbp;
union lpfc_wqe128 *abts_wqe;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int retval;
/*
* There are certain command types we don't want to abort. And we
* don't want to abort commands that are already in the process of
* being aborted.
*/
if (cmdiocb->iocb.ulpCommand == CMD_ABORT_XRI_CN ||
cmdiocb->iocb.ulpCommand == CMD_CLOSE_XRI_CN ||
(cmdiocb->iocb_flag & LPFC_DRIVER_ABORTED) != 0)
return 0;
/* issue ABTS for this io based on iotag */
abtsiocbp = __lpfc_sli_get_iocbq(phba);
if (abtsiocbp == NULL)
return 0;
/* This signals the response to set the correct status
* before calling the completion handler
*/
cmdiocb->iocb_flag |= LPFC_DRIVER_ABORTED;
/* Complete prepping the abort wqe and issue to the FW. */
abts_wqe = &abtsiocbp->wqe;
bf_set(abort_cmd_ia, &abts_wqe->abort_cmd, 0);
bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG);
/* Explicitly set reserved fields to zero.*/
abts_wqe->abort_cmd.rsrvd4 = 0;
abts_wqe->abort_cmd.rsrvd5 = 0;
/* WQE Common - word 6. Context is XRI tag. Set 0. */
bf_set(wqe_xri_tag, &abts_wqe->abort_cmd.wqe_com, 0);
bf_set(wqe_ctxt_tag, &abts_wqe->abort_cmd.wqe_com, 0);
/* word 7 */
bf_set(wqe_ct, &abts_wqe->abort_cmd.wqe_com, 0);
bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX);
bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com,
cmdiocb->iocb.ulpClass);
/* word 8 - tell the FW to abort the IO associated with this
* outstanding exchange ID.
*/
abts_wqe->abort_cmd.wqe_com.abort_tag = cmdiocb->sli4_xritag;
/* word 9 - this is the iotag for the abts_wqe completion. */
bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com,
abtsiocbp->iotag);
/* word 10 */
bf_set(wqe_wqid, &abts_wqe->abort_cmd.wqe_com, cmdiocb->hba_wqidx);
bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1);
bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE);
/* word 11 */
bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND);
bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1);
bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
/* ABTS WQE must go to the same WQ as the WQE to be aborted */
abtsiocbp->iocb_flag |= LPFC_IO_NVME;
abtsiocbp->vport = vport;
abtsiocbp->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
retval = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abtsiocbp);
if (retval) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
"6147 Failed abts issue_wqe with status x%x "
"for oxid x%x\n",
retval, cmdiocb->sli4_xritag);
lpfc_sli_release_iocbq(phba, abtsiocbp);
return retval;
}
lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
"6148 Drv Abort NVME Request Issued for "
"ox_id x%x on reqtag x%x\n",
cmdiocb->sli4_xritag,
abtsiocbp->iotag);
return retval;
}
/**
* lpfc_sli_hba_iocb_abort - Abort all iocbs to an hba.
* @phba: pointer to lpfc HBA data structure.
*
* This routine will abort all pending and outstanding iocbs to an HBA.
**/
void
lpfc_sli_hba_iocb_abort(struct lpfc_hba *phba)
{
struct lpfc_sli *psli = &phba->sli;
struct lpfc_sli_ring *pring;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_queue *qp = NULL;
int i;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->sli_rev != LPFC_SLI_REV4) {
for (i = 0; i < psli->num_rings; i++) {
pring = &psli->sli3_ring[i];
lpfc_sli_abort_iocb_ring(phba, pring);
}
return;
}
list_for_each_entry(qp, &phba->sli4_hba.lpfc_wq_list, wq_list) {
pring = qp->pring;
if (!pring)
continue;
lpfc_sli_abort_iocb_ring(phba, pring);
}
}
/**
* lpfc_sli_validate_fcp_iocb - find commands associated with a vport or LUN
* @iocbq: Pointer to driver iocb object.
* @vport: Pointer to driver virtual port object.
* @tgt_id: SCSI ID of the target.
* @lun_id: LUN ID of the scsi device.
* @ctx_cmd: LPFC_CTX_LUN/LPFC_CTX_TGT/LPFC_CTX_HOST
*
* This function acts as an iocb filter for functions which abort or count
* all FCP iocbs pending on a lun/SCSI target/SCSI host. It will return
* 0 if the filtering criteria is met for the given iocb and will return
* 1 if the filtering criteria is not met.
* If ctx_cmd == LPFC_CTX_LUN, the function returns 0 only if the
* given iocb is for the SCSI device specified by vport, tgt_id and
* lun_id parameter.
* If ctx_cmd == LPFC_CTX_TGT, the function returns 0 only if the
* given iocb is for the SCSI target specified by vport and tgt_id
* parameters.
* If ctx_cmd == LPFC_CTX_HOST, the function returns 0 only if the
* given iocb is for the SCSI host associated with the given vport.
* This function is called with no locks held.
**/
static int
lpfc_sli_validate_fcp_iocb(struct lpfc_iocbq *iocbq, struct lpfc_vport *vport,
uint16_t tgt_id, uint64_t lun_id,
lpfc_ctx_cmd ctx_cmd)
{
struct lpfc_scsi_buf *lpfc_cmd;
int rc = 1;
if (!(iocbq->iocb_flag & LPFC_IO_FCP))
return rc;
if (iocbq->vport != vport)
return rc;
lpfc_cmd = container_of(iocbq, struct lpfc_scsi_buf, cur_iocbq);
if (lpfc_cmd->pCmd == NULL)
return rc;
switch (ctx_cmd) {
case LPFC_CTX_LUN:
if ((lpfc_cmd->rdata->pnode) &&
(lpfc_cmd->rdata->pnode->nlp_sid == tgt_id) &&
(scsilun_to_int(&lpfc_cmd->fcp_cmnd->fcp_lun) == lun_id))
rc = 0;
break;
case LPFC_CTX_TGT:
if ((lpfc_cmd->rdata->pnode) &&
(lpfc_cmd->rdata->pnode->nlp_sid == tgt_id))
rc = 0;
break;
case LPFC_CTX_HOST:
rc = 0;
break;
default:
printk(KERN_ERR "%s: Unknown context cmd type, value %d\n",
__func__, ctx_cmd);
break;
}
return rc;
}
/**
* lpfc_sli_sum_iocb - Function to count the number of FCP iocbs pending
* @vport: Pointer to virtual port.
* @tgt_id: SCSI ID of the target.
* @lun_id: LUN ID of the scsi device.
* @ctx_cmd: LPFC_CTX_LUN/LPFC_CTX_TGT/LPFC_CTX_HOST.
*
* This function returns number of FCP commands pending for the vport.
* When ctx_cmd == LPFC_CTX_LUN, the function returns number of FCP
* commands pending on the vport associated with SCSI device specified
* by tgt_id and lun_id parameters.
* When ctx_cmd == LPFC_CTX_TGT, the function returns number of FCP
* commands pending on the vport associated with SCSI target specified
* by tgt_id parameter.
* When ctx_cmd == LPFC_CTX_HOST, the function returns number of FCP
* commands pending on the vport.
* This function returns the number of iocbs which satisfy the filter.
* This function is called without any lock held.
**/
int
lpfc_sli_sum_iocb(struct lpfc_vport *vport, uint16_t tgt_id, uint64_t lun_id,
lpfc_ctx_cmd ctx_cmd)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_iocbq *iocbq;
int sum, i;
spin_lock_irq(&phba->hbalock);
for (i = 1, sum = 0; i <= phba->sli.last_iotag; i++) {
iocbq = phba->sli.iocbq_lookup[i];
if (lpfc_sli_validate_fcp_iocb (iocbq, vport, tgt_id, lun_id,
ctx_cmd) == 0)
sum++;
}
spin_unlock_irq(&phba->hbalock);
return sum;
}
/**
* lpfc_sli_abort_fcp_cmpl - Completion handler function for aborted FCP IOCBs
* @phba: Pointer to HBA context object
* @cmdiocb: Pointer to command iocb object.
* @rspiocb: Pointer to response iocb object.
*
* This function is called when an aborted FCP iocb completes. This
* function is called by the ring event handler with no lock held.
* This function frees the iocb.
**/
void
lpfc_sli_abort_fcp_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
struct lpfc_iocbq *rspiocb)
{
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"3096 ABORT_XRI_CN completing on rpi x%x "
"original iotag x%x, abort cmd iotag x%x "
"status 0x%x, reason 0x%x\n",
cmdiocb->iocb.un.acxri.abortContextTag,
cmdiocb->iocb.un.acxri.abortIoTag,
cmdiocb->iotag, rspiocb->iocb.ulpStatus,
rspiocb->iocb.un.ulpWord[4]);
lpfc_sli_release_iocbq(phba, cmdiocb);
return;
}
/**
* lpfc_sli_abort_iocb - issue abort for all commands on a host/target/LUN
* @vport: Pointer to virtual port.
* @pring: Pointer to driver SLI ring object.
* @tgt_id: SCSI ID of the target.
* @lun_id: LUN ID of the scsi device.
* @abort_cmd: LPFC_CTX_LUN/LPFC_CTX_TGT/LPFC_CTX_HOST.
*
* This function sends an abort command for every SCSI command
* associated with the given virtual port pending on the ring
* filtered by lpfc_sli_validate_fcp_iocb function.
* When abort_cmd == LPFC_CTX_LUN, the function sends abort only to the
* FCP iocbs associated with lun specified by tgt_id and lun_id
* parameters
* When abort_cmd == LPFC_CTX_TGT, the function sends abort only to the
* FCP iocbs associated with SCSI target specified by tgt_id parameter.
* When abort_cmd == LPFC_CTX_HOST, the function sends abort to all
* FCP iocbs associated with virtual port.
* This function returns number of iocbs it failed to abort.
* This function is called with no locks held.
**/
int
lpfc_sli_abort_iocb(struct lpfc_vport *vport, struct lpfc_sli_ring *pring,
uint16_t tgt_id, uint64_t lun_id, lpfc_ctx_cmd abort_cmd)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_iocbq *iocbq;
struct lpfc_iocbq *abtsiocb;
struct lpfc_sli_ring *pring_s4;
IOCB_t *cmd = NULL;
int errcnt = 0, ret_val = 0;
int i;
for (i = 1; i <= phba->sli.last_iotag; i++) {
iocbq = phba->sli.iocbq_lookup[i];
if (lpfc_sli_validate_fcp_iocb(iocbq, vport, tgt_id, lun_id,
abort_cmd) != 0)
continue;
/*
* If the iocbq is already being aborted, don't take a second
* action, but do count it.
*/
if (iocbq->iocb_flag & LPFC_DRIVER_ABORTED)
continue;
/* issue ABTS for this IOCB based on iotag */
abtsiocb = lpfc_sli_get_iocbq(phba);
if (abtsiocb == NULL) {
errcnt++;
continue;
}
/* indicate the IO is being aborted by the driver. */
iocbq->iocb_flag |= LPFC_DRIVER_ABORTED;
cmd = &iocbq->iocb;
abtsiocb->iocb.un.acxri.abortType = ABORT_TYPE_ABTS;
abtsiocb->iocb.un.acxri.abortContextTag = cmd->ulpContext;
if (phba->sli_rev == LPFC_SLI_REV4)
abtsiocb->iocb.un.acxri.abortIoTag = iocbq->sli4_xritag;
else
abtsiocb->iocb.un.acxri.abortIoTag = cmd->ulpIoTag;
abtsiocb->iocb.ulpLe = 1;
abtsiocb->iocb.ulpClass = cmd->ulpClass;
abtsiocb->vport = vport;
/* ABTS WQE must go to the same WQ as the WQE to be aborted */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
abtsiocb->hba_wqidx = iocbq->hba_wqidx;
if (iocbq->iocb_flag & LPFC_IO_FCP)
abtsiocb->iocb_flag |= LPFC_USE_FCPWQIDX;
if (iocbq->iocb_flag & LPFC_IO_FOF)
abtsiocb->iocb_flag |= LPFC_IO_FOF;
if (lpfc_is_link_up(phba))
abtsiocb->iocb.ulpCommand = CMD_ABORT_XRI_CN;
else
abtsiocb->iocb.ulpCommand = CMD_CLOSE_XRI_CN;
/* Setup callback routine and issue the command. */
abtsiocb->iocb_cmpl = lpfc_sli_abort_fcp_cmpl;
if (phba->sli_rev == LPFC_SLI_REV4) {
pring_s4 = lpfc_sli4_calc_ring(phba, iocbq);
if (!pring_s4)
continue;
ret_val = lpfc_sli_issue_iocb(phba, pring_s4->ringno,
abtsiocb, 0);
} else
ret_val = lpfc_sli_issue_iocb(phba, pring->ringno,
abtsiocb, 0);
if (ret_val == IOCB_ERROR) {
lpfc_sli_release_iocbq(phba, abtsiocb);
errcnt++;
continue;
}
}
return errcnt;
}
/**
* lpfc_sli_abort_taskmgmt - issue abort for all commands on a host/target/LUN
* @vport: Pointer to virtual port.
* @pring: Pointer to driver SLI ring object.
* @tgt_id: SCSI ID of the target.
* @lun_id: LUN ID of the scsi device.
* @taskmgmt_cmd: LPFC_CTX_LUN/LPFC_CTX_TGT/LPFC_CTX_HOST.
*
* This function sends an abort command for every SCSI command
* associated with the given virtual port pending on the ring
* filtered by lpfc_sli_validate_fcp_iocb function.
* When taskmgmt_cmd == LPFC_CTX_LUN, the function sends abort only to the
* FCP iocbs associated with lun specified by tgt_id and lun_id
* parameters
* When taskmgmt_cmd == LPFC_CTX_TGT, the function sends abort only to the
* FCP iocbs associated with SCSI target specified by tgt_id parameter.
* When taskmgmt_cmd == LPFC_CTX_HOST, the function sends abort to all
* FCP iocbs associated with virtual port.
* This function returns number of iocbs it aborted .
* This function is called with no locks held right after a taskmgmt
* command is sent.
**/
int
lpfc_sli_abort_taskmgmt(struct lpfc_vport *vport, struct lpfc_sli_ring *pring,
uint16_t tgt_id, uint64_t lun_id, lpfc_ctx_cmd cmd)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_scsi_buf *lpfc_cmd;
struct lpfc_iocbq *abtsiocbq;
struct lpfc_nodelist *ndlp;
struct lpfc_iocbq *iocbq;
IOCB_t *icmd;
int sum, i, ret_val;
unsigned long iflags;
struct lpfc_sli_ring *pring_s4;
spin_lock_irq(&phba->hbalock);
/* all I/Os are in process of being flushed */
if (phba->hba_flag & HBA_FCP_IOQ_FLUSH) {
spin_unlock_irq(&phba->hbalock);
return 0;
}
sum = 0;
for (i = 1; i <= phba->sli.last_iotag; i++) {
iocbq = phba->sli.iocbq_lookup[i];
if (lpfc_sli_validate_fcp_iocb(iocbq, vport, tgt_id, lun_id,
cmd) != 0)
continue;
/*
* If the iocbq is already being aborted, don't take a second
* action, but do count it.
*/
if (iocbq->iocb_flag & LPFC_DRIVER_ABORTED)
continue;
/* issue ABTS for this IOCB based on iotag */
abtsiocbq = __lpfc_sli_get_iocbq(phba);
if (abtsiocbq == NULL)
continue;
icmd = &iocbq->iocb;
abtsiocbq->iocb.un.acxri.abortType = ABORT_TYPE_ABTS;
abtsiocbq->iocb.un.acxri.abortContextTag = icmd->ulpContext;
if (phba->sli_rev == LPFC_SLI_REV4)
abtsiocbq->iocb.un.acxri.abortIoTag =
iocbq->sli4_xritag;
else
abtsiocbq->iocb.un.acxri.abortIoTag = icmd->ulpIoTag;
abtsiocbq->iocb.ulpLe = 1;
abtsiocbq->iocb.ulpClass = icmd->ulpClass;
abtsiocbq->vport = vport;
/* ABTS WQE must go to the same WQ as the WQE to be aborted */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
abtsiocbq->hba_wqidx = iocbq->hba_wqidx;
if (iocbq->iocb_flag & LPFC_IO_FCP)
abtsiocbq->iocb_flag |= LPFC_USE_FCPWQIDX;
if (iocbq->iocb_flag & LPFC_IO_FOF)
abtsiocbq->iocb_flag |= LPFC_IO_FOF;
lpfc_cmd = container_of(iocbq, struct lpfc_scsi_buf, cur_iocbq);
ndlp = lpfc_cmd->rdata->pnode;
if (lpfc_is_link_up(phba) &&
(ndlp && ndlp->nlp_state == NLP_STE_MAPPED_NODE))
abtsiocbq->iocb.ulpCommand = CMD_ABORT_XRI_CN;
else
abtsiocbq->iocb.ulpCommand = CMD_CLOSE_XRI_CN;
/* Setup callback routine and issue the command. */
abtsiocbq->iocb_cmpl = lpfc_sli_abort_fcp_cmpl;
/*
* Indicate the IO is being aborted by the driver and set
* the caller's flag into the aborted IO.
*/
iocbq->iocb_flag |= LPFC_DRIVER_ABORTED;
if (phba->sli_rev == LPFC_SLI_REV4) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring_s4 = lpfc_sli4_calc_ring(phba, iocbq);
if (pring_s4 == NULL)
continue;
/* Note: both hbalock and ring_lock must be set here */
spin_lock_irqsave(&pring_s4->ring_lock, iflags);
ret_val = __lpfc_sli_issue_iocb(phba, pring_s4->ringno,
abtsiocbq, 0);
spin_unlock_irqrestore(&pring_s4->ring_lock, iflags);
} else {
ret_val = __lpfc_sli_issue_iocb(phba, pring->ringno,
abtsiocbq, 0);
}
if (ret_val == IOCB_ERROR)
__lpfc_sli_release_iocbq(phba, abtsiocbq);
else
sum++;
}
spin_unlock_irq(&phba->hbalock);
return sum;
}
/**
* lpfc_sli_wake_iocb_wait - lpfc_sli_issue_iocb_wait's completion handler
* @phba: Pointer to HBA context object.
* @cmdiocbq: Pointer to command iocb.
* @rspiocbq: Pointer to response iocb.
*
* This function is the completion handler for iocbs issued using
* lpfc_sli_issue_iocb_wait function. This function is called by the
* ring event handler function without any lock held. This function
* can be called from both worker thread context and interrupt
* context. This function also can be called from other thread which
* cleans up the SLI layer objects.
* This function copy the contents of the response iocb to the
* response iocb memory object provided by the caller of
* lpfc_sli_issue_iocb_wait and then wakes up the thread which
* sleeps for the iocb completion.
**/
static void
lpfc_sli_wake_iocb_wait(struct lpfc_hba *phba,
struct lpfc_iocbq *cmdiocbq,
struct lpfc_iocbq *rspiocbq)
{
wait_queue_head_t *pdone_q;
unsigned long iflags;
struct lpfc_scsi_buf *lpfc_cmd;
spin_lock_irqsave(&phba->hbalock, iflags);
if (cmdiocbq->iocb_flag & LPFC_IO_WAKE_TMO) {
/*
* A time out has occurred for the iocb. If a time out
* completion handler has been supplied, call it. Otherwise,
* just free the iocbq.
*/
spin_unlock_irqrestore(&phba->hbalock, iflags);
cmdiocbq->iocb_cmpl = cmdiocbq->wait_iocb_cmpl;
cmdiocbq->wait_iocb_cmpl = NULL;
if (cmdiocbq->iocb_cmpl)
(cmdiocbq->iocb_cmpl)(phba, cmdiocbq, NULL);
else
lpfc_sli_release_iocbq(phba, cmdiocbq);
return;
}
cmdiocbq->iocb_flag |= LPFC_IO_WAKE;
if (cmdiocbq->context2 && rspiocbq)
memcpy(&((struct lpfc_iocbq *)cmdiocbq->context2)->iocb,
&rspiocbq->iocb, sizeof(IOCB_t));
/* Set the exchange busy flag for task management commands */
if ((cmdiocbq->iocb_flag & LPFC_IO_FCP) &&
!(cmdiocbq->iocb_flag & LPFC_IO_LIBDFC)) {
lpfc_cmd = container_of(cmdiocbq, struct lpfc_scsi_buf,
cur_iocbq);
lpfc_cmd->exch_busy = rspiocbq->iocb_flag & LPFC_EXCHANGE_BUSY;
}
pdone_q = cmdiocbq->context_un.wait_queue;
if (pdone_q)
wake_up(pdone_q);
spin_unlock_irqrestore(&phba->hbalock, iflags);
return;
}
/**
* lpfc_chk_iocb_flg - Test IOCB flag with lock held.
* @phba: Pointer to HBA context object..
* @piocbq: Pointer to command iocb.
* @flag: Flag to test.
*
* This routine grabs the hbalock and then test the iocb_flag to
* see if the passed in flag is set.
* Returns:
* 1 if flag is set.
* 0 if flag is not set.
**/
static int
lpfc_chk_iocb_flg(struct lpfc_hba *phba,
struct lpfc_iocbq *piocbq, uint32_t flag)
{
unsigned long iflags;
int ret;
spin_lock_irqsave(&phba->hbalock, iflags);
ret = piocbq->iocb_flag & flag;
spin_unlock_irqrestore(&phba->hbalock, iflags);
return ret;
}
/**
* lpfc_sli_issue_iocb_wait - Synchronous function to issue iocb commands
* @phba: Pointer to HBA context object..
* @pring: Pointer to sli ring.
* @piocb: Pointer to command iocb.
* @prspiocbq: Pointer to response iocb.
* @timeout: Timeout in number of seconds.
*
* This function issues the iocb to firmware and waits for the
* iocb to complete. The iocb_cmpl field of the shall be used
* to handle iocbs which time out. If the field is NULL, the
* function shall free the iocbq structure. If more clean up is
* needed, the caller is expected to provide a completion function
* that will provide the needed clean up. If the iocb command is
* not completed within timeout seconds, the function will either
* free the iocbq structure (if iocb_cmpl == NULL) or execute the
* completion function set in the iocb_cmpl field and then return
* a status of IOCB_TIMEDOUT. The caller should not free the iocb
* resources if this function returns IOCB_TIMEDOUT.
* The function waits for the iocb completion using an
* non-interruptible wait.
* This function will sleep while waiting for iocb completion.
* So, this function should not be called from any context which
* does not allow sleeping. Due to the same reason, this function
* cannot be called with interrupt disabled.
* This function assumes that the iocb completions occur while
* this function sleep. So, this function cannot be called from
* the thread which process iocb completion for this ring.
* This function clears the iocb_flag of the iocb object before
* issuing the iocb and the iocb completion handler sets this
* flag and wakes this thread when the iocb completes.
* The contents of the response iocb will be copied to prspiocbq
* by the completion handler when the command completes.
* This function returns IOCB_SUCCESS when success.
* This function is called with no lock held.
**/
int
lpfc_sli_issue_iocb_wait(struct lpfc_hba *phba,
uint32_t ring_number,
struct lpfc_iocbq *piocb,
struct lpfc_iocbq *prspiocbq,
uint32_t timeout)
{
DECLARE_WAIT_QUEUE_HEAD_ONSTACK(done_q);
long timeleft, timeout_req = 0;
int retval = IOCB_SUCCESS;
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
uint32_t creg_val;
struct lpfc_iocbq *iocb;
int txq_cnt = 0;
int txcmplq_cnt = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *pring;
unsigned long iflags;
bool iocb_completed = true;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->sli_rev >= LPFC_SLI_REV4)
pring = lpfc_sli4_calc_ring(phba, piocb);
else
pring = &phba->sli.sli3_ring[ring_number];
/*
* If the caller has provided a response iocbq buffer, then context2
* is NULL or its an error.
*/
if (prspiocbq) {
if (piocb->context2)
return IOCB_ERROR;
piocb->context2 = prspiocbq;
}
piocb->wait_iocb_cmpl = piocb->iocb_cmpl;
piocb->iocb_cmpl = lpfc_sli_wake_iocb_wait;
piocb->context_un.wait_queue = &done_q;
piocb->iocb_flag &= ~(LPFC_IO_WAKE | LPFC_IO_WAKE_TMO);
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
if (phba->cfg_poll & DISABLE_FCP_RING_INT) {
if (lpfc_readl(phba->HCregaddr, &creg_val))
return IOCB_ERROR;
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
creg_val |= (HC_R0INT_ENA << LPFC_FCP_RING);
writel(creg_val, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
}
retval = lpfc_sli_issue_iocb(phba, ring_number, piocb,
SLI_IOCB_RET_IOCB);
if (retval == IOCB_SUCCESS) {
timeout_req = msecs_to_jiffies(timeout * 1000);
timeleft = wait_event_timeout(done_q,
lpfc_chk_iocb_flg(phba, piocb, LPFC_IO_WAKE),
timeout_req);
spin_lock_irqsave(&phba->hbalock, iflags);
if (!(piocb->iocb_flag & LPFC_IO_WAKE)) {
/*
* IOCB timed out. Inform the wake iocb wait
* completion function and set local status
*/
iocb_completed = false;
piocb->iocb_flag |= LPFC_IO_WAKE_TMO;
}
spin_unlock_irqrestore(&phba->hbalock, iflags);
if (iocb_completed) {
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0331 IOCB wake signaled\n");
/* Note: we are not indicating if the IOCB has a success
* status or not - that's for the caller to check.
* IOCB_SUCCESS means just that the command was sent and
* completed. Not that it completed successfully.
* */
} else if (timeleft == 0) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0338 IOCB wait timeout error - no "
"wake response Data x%x\n", timeout);
retval = IOCB_TIMEDOUT;
} else {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0330 IOCB wake NOT set, "
"Data x%x x%lx\n",
timeout, (timeleft / jiffies));
retval = IOCB_TIMEDOUT;
}
} else if (retval == IOCB_BUSY) {
if (phba->cfg_log_verbose & LOG_SLI) {
list_for_each_entry(iocb, &pring->txq, list) {
txq_cnt++;
}
list_for_each_entry(iocb, &pring->txcmplq, list) {
txcmplq_cnt++;
}
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"2818 Max IOCBs %d txq cnt %d txcmplq cnt %d\n",
phba->iocb_cnt, txq_cnt, txcmplq_cnt);
}
return retval;
} else {
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0332 IOCB wait issue failed, Data x%x\n",
retval);
retval = IOCB_ERROR;
}
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
if (phba->cfg_poll & DISABLE_FCP_RING_INT) {
if (lpfc_readl(phba->HCregaddr, &creg_val))
return IOCB_ERROR;
[SCSI] lpfc 8.1.1 : Add polled-mode support - Add functionality to run in polled mode only. Includes run time attribute to enable mode. - Enable runtime writable hba settings for coallescing and delay parameters Customers have requested a mode in the driver to run strictly polled. This is generally to support an environment where the server is extremely loaded and is looking to reclaim some cpu cycles from adapter interrupt handling. This patch adds a new "poll" attribute, and the following behavior: if value is 0 (default): The driver uses the normal method for i/o completion. It uses the firmware feature of interrupt coalesing. The firmware allows a minimum number of i/o completions before an interrupt, or a maximum time delay between interrupts. By default, the driver sets these to no delay (disabled) or 1 i/o - meaning coalescing is disabled. Attributes were provided to change the coalescing values, but it was a module-load time only and global across all adapters. This patch allows them to be writable on a per-adapter basis. if value is 1 : Interrupts are left enabled, expecting that the user has tuned the interrupt coalescing values. When this setting is enabled, the driver will attempt to service completed i/o whenever new i/o is submitted to the adapter. If the coalescing values are large, and the i/o generation rate steady, an interrupt will be avoided by servicing completed i/o prior to the coalescing thresholds kicking in. However, if the i/o completion load is high enough or i/o generation slow, the coalescion values will ensure that completed i/o is serviced in a timely fashion. if value is 3 : Turns off FCP i/o interrupts altogether. The coalescing values now have no effect. A new attribute "poll_tmo" (default 10ms) exists to set the polling interval for i/o completion. When this setting is enabled, the driver will attempt to service completed i/o and restart the interval timer whenever new i/o is submitted. This behavior allows for servicing of completed i/o sooner than the interval timer, but ensures that if no i/o is being issued, then the interval timer will kick in to service the outstanding i/o. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2005-11-29 14:32:13 -07:00
creg_val &= ~(HC_R0INT_ENA << LPFC_FCP_RING);
writel(creg_val, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
}
if (prspiocbq)
piocb->context2 = NULL;
piocb->context_un.wait_queue = NULL;
piocb->iocb_cmpl = NULL;
return retval;
}
/**
* lpfc_sli_issue_mbox_wait - Synchronous function to issue mailbox
* @phba: Pointer to HBA context object.
* @pmboxq: Pointer to driver mailbox object.
* @timeout: Timeout in number of seconds.
*
* This function issues the mailbox to firmware and waits for the
* mailbox command to complete. If the mailbox command is not
* completed within timeout seconds, it returns MBX_TIMEOUT.
* The function waits for the mailbox completion using an
* interruptible wait. If the thread is woken up due to a
* signal, MBX_TIMEOUT error is returned to the caller. Caller
* should not free the mailbox resources, if this function returns
* MBX_TIMEOUT.
* This function will sleep while waiting for mailbox completion.
* So, this function should not be called from any context which
* does not allow sleeping. Due to the same reason, this function
* cannot be called with interrupt disabled.
* This function assumes that the mailbox completion occurs while
* this function sleep. So, this function cannot be called from
* the worker thread which processes mailbox completion.
* This function is called in the context of HBA management
* applications.
* This function returns MBX_SUCCESS when successful.
* This function is called with no lock held.
**/
int
lpfc_sli_issue_mbox_wait(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmboxq,
uint32_t timeout)
{
struct completion mbox_done;
int retval;
unsigned long flag;
pmboxq->mbox_flag &= ~LPFC_MBX_WAKE;
/* setup wake call as IOCB callback */
pmboxq->mbox_cmpl = lpfc_sli_wake_mbox_wait;
/* setup context3 field to pass wait_queue pointer to wake function */
init_completion(&mbox_done);
pmboxq->context3 = &mbox_done;
/* now issue the command */
retval = lpfc_sli_issue_mbox(phba, pmboxq, MBX_NOWAIT);
if (retval == MBX_BUSY || retval == MBX_SUCCESS) {
wait_for_completion_timeout(&mbox_done,
msecs_to_jiffies(timeout * 1000));
spin_lock_irqsave(&phba->hbalock, flag);
pmboxq->context3 = NULL;
/*
* if LPFC_MBX_WAKE flag is set the mailbox is completed
* else do not free the resources.
*/
if (pmboxq->mbox_flag & LPFC_MBX_WAKE) {
retval = MBX_SUCCESS;
} else {
retval = MBX_TIMEOUT;
pmboxq->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
}
spin_unlock_irqrestore(&phba->hbalock, flag);
}
return retval;
}
/**
* lpfc_sli_mbox_sys_shutdown - shutdown mailbox command sub-system
* @phba: Pointer to HBA context.
*
* This function is called to shutdown the driver's mailbox sub-system.
* It first marks the mailbox sub-system is in a block state to prevent
* the asynchronous mailbox command from issued off the pending mailbox
* command queue. If the mailbox command sub-system shutdown is due to
* HBA error conditions such as EEH or ERATT, this routine shall invoke
* the mailbox sub-system flush routine to forcefully bring down the
* mailbox sub-system. Otherwise, if it is due to normal condition (such
* as with offline or HBA function reset), this routine will wait for the
* outstanding mailbox command to complete before invoking the mailbox
* sub-system flush routine to gracefully bring down mailbox sub-system.
**/
void
lpfc_sli_mbox_sys_shutdown(struct lpfc_hba *phba, int mbx_action)
{
struct lpfc_sli *psli = &phba->sli;
unsigned long timeout;
if (mbx_action == LPFC_MBX_NO_WAIT) {
/* delay 100ms for port state */
msleep(100);
lpfc_sli_mbox_sys_flush(phba);
return;
}
timeout = msecs_to_jiffies(LPFC_MBOX_TMO * 1000) + jiffies;
spin_lock_irq(&phba->hbalock);
psli->sli_flag |= LPFC_SLI_ASYNC_MBX_BLK;
if (psli->sli_flag & LPFC_SLI_ACTIVE) {
/* Determine how long we might wait for the active mailbox
* command to be gracefully completed by firmware.
*/
if (phba->sli.mbox_active)
timeout = msecs_to_jiffies(lpfc_mbox_tmo_val(phba,
phba->sli.mbox_active) *
1000) + jiffies;
spin_unlock_irq(&phba->hbalock);
while (phba->sli.mbox_active) {
/* Check active mailbox complete status every 2ms */
msleep(2);
if (time_after(jiffies, timeout))
/* Timeout, let the mailbox flush routine to
* forcefully release active mailbox command
*/
break;
}
} else
spin_unlock_irq(&phba->hbalock);
lpfc_sli_mbox_sys_flush(phba);
}
/**
* lpfc_sli_eratt_read - read sli-3 error attention events
* @phba: Pointer to HBA context.
*
* This function is called to read the SLI3 device error attention registers
* for possible error attention events. The caller must hold the hostlock
* with spin_lock_irq().
*
* This function returns 1 when there is Error Attention in the Host Attention
* Register and returns 0 otherwise.
**/
static int
lpfc_sli_eratt_read(struct lpfc_hba *phba)
{
uint32_t ha_copy;
/* Read chip Host Attention (HA) register */
if (lpfc_readl(phba->HAregaddr, &ha_copy))
goto unplug_err;
if (ha_copy & HA_ERATT) {
/* Read host status register to retrieve error event */
if (lpfc_sli_read_hs(phba))
goto unplug_err;
/* Check if there is a deferred error condition is active */
if ((HS_FFER1 & phba->work_hs) &&
((HS_FFER2 | HS_FFER3 | HS_FFER4 | HS_FFER5 |
HS_FFER6 | HS_FFER7 | HS_FFER8) & phba->work_hs)) {
phba->hba_flag |= DEFER_ERATT;
/* Clear all interrupt enable conditions */
writel(0, phba->HCregaddr);
readl(phba->HCregaddr);
}
/* Set the driver HA work bitmap */
phba->work_ha |= HA_ERATT;
/* Indicate polling handles this ERATT */
phba->hba_flag |= HBA_ERATT_HANDLED;
return 1;
}
return 0;
unplug_err:
/* Set the driver HS work bitmap */
phba->work_hs |= UNPLUG_ERR;
/* Set the driver HA work bitmap */
phba->work_ha |= HA_ERATT;
/* Indicate polling handles this ERATT */
phba->hba_flag |= HBA_ERATT_HANDLED;
return 1;
}
/**
* lpfc_sli4_eratt_read - read sli-4 error attention events
* @phba: Pointer to HBA context.
*
* This function is called to read the SLI4 device error attention registers
* for possible error attention events. The caller must hold the hostlock
* with spin_lock_irq().
*
* This function returns 1 when there is Error Attention in the Host Attention
* Register and returns 0 otherwise.
**/
static int
lpfc_sli4_eratt_read(struct lpfc_hba *phba)
{
uint32_t uerr_sta_hi, uerr_sta_lo;
uint32_t if_type, portsmphr;
struct lpfc_register portstat_reg;
/*
* For now, use the SLI4 device internal unrecoverable error
* registers for error attention. This can be changed later.
*/
if_type = bf_get(lpfc_sli_intf_if_type, &phba->sli4_hba.sli_intf);
switch (if_type) {
case LPFC_SLI_INTF_IF_TYPE_0:
if (lpfc_readl(phba->sli4_hba.u.if_type0.UERRLOregaddr,
&uerr_sta_lo) ||
lpfc_readl(phba->sli4_hba.u.if_type0.UERRHIregaddr,
&uerr_sta_hi)) {
phba->work_hs |= UNPLUG_ERR;
phba->work_ha |= HA_ERATT;
phba->hba_flag |= HBA_ERATT_HANDLED;
return 1;
}
if ((~phba->sli4_hba.ue_mask_lo & uerr_sta_lo) ||
(~phba->sli4_hba.ue_mask_hi & uerr_sta_hi)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"1423 HBA Unrecoverable error: "
"uerr_lo_reg=0x%x, uerr_hi_reg=0x%x, "
"ue_mask_lo_reg=0x%x, "
"ue_mask_hi_reg=0x%x\n",
uerr_sta_lo, uerr_sta_hi,
phba->sli4_hba.ue_mask_lo,
phba->sli4_hba.ue_mask_hi);
phba->work_status[0] = uerr_sta_lo;
phba->work_status[1] = uerr_sta_hi;
phba->work_ha |= HA_ERATT;
phba->hba_flag |= HBA_ERATT_HANDLED;
return 1;
}
break;
case LPFC_SLI_INTF_IF_TYPE_2:
case LPFC_SLI_INTF_IF_TYPE_6:
if (lpfc_readl(phba->sli4_hba.u.if_type2.STATUSregaddr,
&portstat_reg.word0) ||
lpfc_readl(phba->sli4_hba.PSMPHRregaddr,
&portsmphr)){
phba->work_hs |= UNPLUG_ERR;
phba->work_ha |= HA_ERATT;
phba->hba_flag |= HBA_ERATT_HANDLED;
return 1;
}
if (bf_get(lpfc_sliport_status_err, &portstat_reg)) {
phba->work_status[0] =
readl(phba->sli4_hba.u.if_type2.ERR1regaddr);
phba->work_status[1] =
readl(phba->sli4_hba.u.if_type2.ERR2regaddr);
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2885 Port Status Event: "
"port status reg 0x%x, "
"port smphr reg 0x%x, "
"error 1=0x%x, error 2=0x%x\n",
portstat_reg.word0,
portsmphr,
phba->work_status[0],
phba->work_status[1]);
phba->work_ha |= HA_ERATT;
phba->hba_flag |= HBA_ERATT_HANDLED;
return 1;
}
break;
case LPFC_SLI_INTF_IF_TYPE_1:
default:
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2886 HBA Error Attention on unsupported "
"if type %d.", if_type);
return 1;
}
return 0;
}
/**
* lpfc_sli_check_eratt - check error attention events
* @phba: Pointer to HBA context.
*
* This function is called from timer soft interrupt context to check HBA's
* error attention register bit for error attention events.
*
* This function returns 1 when there is Error Attention in the Host Attention
* Register and returns 0 otherwise.
**/
int
lpfc_sli_check_eratt(struct lpfc_hba *phba)
{
uint32_t ha_copy;
/* If somebody is waiting to handle an eratt, don't process it
* here. The brdkill function will do this.
*/
if (phba->link_flag & LS_IGNORE_ERATT)
return 0;
/* Check if interrupt handler handles this ERATT */
spin_lock_irq(&phba->hbalock);
if (phba->hba_flag & HBA_ERATT_HANDLED) {
/* Interrupt handler has handled ERATT */
spin_unlock_irq(&phba->hbalock);
return 0;
}
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/*
* If there is deferred error attention, do not check for error
* attention
*/
if (unlikely(phba->hba_flag & DEFER_ERATT)) {
spin_unlock_irq(&phba->hbalock);
return 0;
}
/* If PCI channel is offline, don't process it */
if (unlikely(pci_channel_offline(phba->pcidev))) {
spin_unlock_irq(&phba->hbalock);
return 0;
}
switch (phba->sli_rev) {
case LPFC_SLI_REV2:
case LPFC_SLI_REV3:
/* Read chip Host Attention (HA) register */
ha_copy = lpfc_sli_eratt_read(phba);
break;
case LPFC_SLI_REV4:
/* Read device Uncoverable Error (UERR) registers */
ha_copy = lpfc_sli4_eratt_read(phba);
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0299 Invalid SLI revision (%d)\n",
phba->sli_rev);
ha_copy = 0;
break;
}
spin_unlock_irq(&phba->hbalock);
return ha_copy;
}
/**
* lpfc_intr_state_check - Check device state for interrupt handling
* @phba: Pointer to HBA context.
*
* This inline routine checks whether a device or its PCI slot is in a state
* that the interrupt should be handled.
*
* This function returns 0 if the device or the PCI slot is in a state that
* interrupt should be handled, otherwise -EIO.
*/
static inline int
lpfc_intr_state_check(struct lpfc_hba *phba)
{
/* If the pci channel is offline, ignore all the interrupts */
if (unlikely(pci_channel_offline(phba->pcidev)))
return -EIO;
/* Update device level interrupt statistics */
phba->sli.slistat.sli_intr++;
/* Ignore all interrupts during initialization. */
if (unlikely(phba->link_state < LPFC_LINK_DOWN))
return -EIO;
return 0;
}
/**
* lpfc_sli_sp_intr_handler - Slow-path interrupt handler to SLI-3 device
* @irq: Interrupt number.
* @dev_id: The device context pointer.
*
* This function is directly called from the PCI layer as an interrupt
* service routine when device with SLI-3 interface spec is enabled with
* MSI-X multi-message interrupt mode and there are slow-path events in
* the HBA. However, when the device is enabled with either MSI or Pin-IRQ
* interrupt mode, this function is called as part of the device-level
* interrupt handler. When the PCI slot is in error recovery or the HBA
* is undergoing initialization, the interrupt handler will not process
* the interrupt. The link attention and ELS ring attention events are
* handled by the worker thread. The interrupt handler signals the worker
* thread and returns for these events. This function is called without
* any lock held. It gets the hbalock to access and update SLI data
* structures.
*
* This function returns IRQ_HANDLED when interrupt is handled else it
* returns IRQ_NONE.
**/
irqreturn_t
lpfc_sli_sp_intr_handler(int irq, void *dev_id)
{
struct lpfc_hba *phba;
uint32_t ha_copy, hc_copy;
uint32_t work_ha_copy;
unsigned long status;
unsigned long iflag;
uint32_t control;
MAILBOX_t *mbox, *pmbox;
struct lpfc_vport *vport;
struct lpfc_nodelist *ndlp;
struct lpfc_dmabuf *mp;
LPFC_MBOXQ_t *pmb;
int rc;
/*
* Get the driver's phba structure from the dev_id and
* assume the HBA is not interrupting.
*/
phba = (struct lpfc_hba *)dev_id;
if (unlikely(!phba))
return IRQ_NONE;
/*
* Stuff needs to be attented to when this function is invoked as an
* individual interrupt handler in MSI-X multi-message interrupt mode
*/
if (phba->intr_type == MSIX) {
/* Check device state for handling interrupt */
if (lpfc_intr_state_check(phba))
return IRQ_NONE;
/* Need to read HA REG for slow-path events */
spin_lock_irqsave(&phba->hbalock, iflag);
if (lpfc_readl(phba->HAregaddr, &ha_copy))
goto unplug_error;
/* If somebody is waiting to handle an eratt don't process it
* here. The brdkill function will do this.
*/
if (phba->link_flag & LS_IGNORE_ERATT)
ha_copy &= ~HA_ERATT;
/* Check the need for handling ERATT in interrupt handler */
if (ha_copy & HA_ERATT) {
if (phba->hba_flag & HBA_ERATT_HANDLED)
/* ERATT polling has handled ERATT */
ha_copy &= ~HA_ERATT;
else
/* Indicate interrupt handler handles ERATT */
phba->hba_flag |= HBA_ERATT_HANDLED;
}
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/*
* If there is deferred error attention, do not check for any
* interrupt.
*/
if (unlikely(phba->hba_flag & DEFER_ERATT)) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
return IRQ_NONE;
}
/* Clear up only attention source related to slow-path */
if (lpfc_readl(phba->HCregaddr, &hc_copy))
goto unplug_error;
writel(hc_copy & ~(HC_MBINT_ENA | HC_R2INT_ENA |
HC_LAINT_ENA | HC_ERINT_ENA),
phba->HCregaddr);
writel((ha_copy & (HA_MBATT | HA_R2_CLR_MSK)),
phba->HAregaddr);
writel(hc_copy, phba->HCregaddr);
readl(phba->HAregaddr); /* flush */
spin_unlock_irqrestore(&phba->hbalock, iflag);
} else
ha_copy = phba->ha_copy;
work_ha_copy = ha_copy & phba->work_ha_mask;
if (work_ha_copy) {
if (work_ha_copy & HA_LATT) {
if (phba->sli.sli_flag & LPFC_PROCESS_LA) {
/*
* Turn off Link Attention interrupts
* until CLEAR_LA done
*/
spin_lock_irqsave(&phba->hbalock, iflag);
phba->sli.sli_flag &= ~LPFC_PROCESS_LA;
if (lpfc_readl(phba->HCregaddr, &control))
goto unplug_error;
control &= ~HC_LAINT_ENA;
writel(control, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
spin_unlock_irqrestore(&phba->hbalock, iflag);
}
else
work_ha_copy &= ~HA_LATT;
}
if (work_ha_copy & ~(HA_ERATT | HA_MBATT | HA_LATT)) {
/*
* Turn off Slow Rings interrupts, LPFC_ELS_RING is
* the only slow ring.
*/
status = (work_ha_copy &
(HA_RXMASK << (4*LPFC_ELS_RING)));
status >>= (4*LPFC_ELS_RING);
if (status & HA_RXMASK) {
spin_lock_irqsave(&phba->hbalock, iflag);
if (lpfc_readl(phba->HCregaddr, &control))
goto unplug_error;
lpfc_debugfs_slow_ring_trc(phba,
"ISR slow ring: ctl:x%x stat:x%x isrcnt:x%x",
control, status,
(uint32_t)phba->sli.slistat.sli_intr);
if (control & (HC_R0INT_ENA << LPFC_ELS_RING)) {
lpfc_debugfs_slow_ring_trc(phba,
"ISR Disable ring:"
"pwork:x%x hawork:x%x wait:x%x",
phba->work_ha, work_ha_copy,
(uint32_t)((unsigned long)
&phba->work_waitq));
control &=
~(HC_R0INT_ENA << LPFC_ELS_RING);
writel(control, phba->HCregaddr);
readl(phba->HCregaddr); /* flush */
}
else {
lpfc_debugfs_slow_ring_trc(phba,
"ISR slow ring: pwork:"
"x%x hawork:x%x wait:x%x",
phba->work_ha, work_ha_copy,
(uint32_t)((unsigned long)
&phba->work_waitq));
}
spin_unlock_irqrestore(&phba->hbalock, iflag);
}
}
spin_lock_irqsave(&phba->hbalock, iflag);
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
if (work_ha_copy & HA_ERATT) {
if (lpfc_sli_read_hs(phba))
goto unplug_error;
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/*
* Check if there is a deferred error condition
* is active
*/
if ((HS_FFER1 & phba->work_hs) &&
((HS_FFER2 | HS_FFER3 | HS_FFER4 | HS_FFER5 |
HS_FFER6 | HS_FFER7 | HS_FFER8) &
phba->work_hs)) {
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
phba->hba_flag |= DEFER_ERATT;
/* Clear all interrupt enable conditions */
writel(0, phba->HCregaddr);
readl(phba->HCregaddr);
}
}
if ((work_ha_copy & HA_MBATT) && (phba->sli.mbox_active)) {
pmb = phba->sli.mbox_active;
pmbox = &pmb->u.mb;
mbox = phba->mbox;
vport = pmb->vport;
/* First check out the status word */
lpfc_sli_pcimem_bcopy(mbox, pmbox, sizeof(uint32_t));
if (pmbox->mbxOwner != OWN_HOST) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
/*
* Stray Mailbox Interrupt, mbxCommand <cmd>
* mbxStatus <status>
*/
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX |
LOG_SLI,
"(%d):0304 Stray Mailbox "
"Interrupt mbxCommand x%x "
"mbxStatus x%x\n",
(vport ? vport->vpi : 0),
pmbox->mbxCommand,
pmbox->mbxStatus);
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
/* clear mailbox attention bit */
work_ha_copy &= ~HA_MBATT;
} else {
phba->sli.mbox_active = NULL;
spin_unlock_irqrestore(&phba->hbalock, iflag);
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
phba->last_completion_time = jiffies;
del_timer(&phba->sli.mbox_tmo);
if (pmb->mbox_cmpl) {
lpfc_sli_pcimem_bcopy(mbox, pmbox,
MAILBOX_CMD_SIZE);
if (pmb->out_ext_byte_len &&
pmb->context2)
lpfc_sli_pcimem_bcopy(
phba->mbox_ext,
pmb->context2,
pmb->out_ext_byte_len);
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
}
if (pmb->mbox_flag & LPFC_MBX_IMED_UNREG) {
pmb->mbox_flag &= ~LPFC_MBX_IMED_UNREG;
lpfc_debugfs_disc_trc(vport,
LPFC_DISC_TRC_MBOX_VPORT,
"MBOX dflt rpi: : "
"status:x%x rpi:x%x",
(uint32_t)pmbox->mbxStatus,
pmbox->un.varWords[0], 0);
if (!pmbox->mbxStatus) {
mp = (struct lpfc_dmabuf *)
(pmb->context1);
ndlp = (struct lpfc_nodelist *)
pmb->context2;
/* Reg_LOGIN of dflt RPI was
* successful. new lets get
* rid of the RPI using the
* same mbox buffer.
*/
lpfc_unreg_login(phba,
vport->vpi,
pmbox->un.varWords[0],
pmb);
pmb->mbox_cmpl =
lpfc_mbx_cmpl_dflt_rpi;
pmb->context1 = mp;
pmb->context2 = ndlp;
pmb->vport = vport;
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
rc = lpfc_sli_issue_mbox(phba,
pmb,
MBX_NOWAIT);
if (rc != MBX_BUSY)
lpfc_printf_log(phba,
KERN_ERR,
LOG_MBOX | LOG_SLI,
"0350 rc should have"
"been MBX_BUSY\n");
if (rc != MBX_NOT_FINISHED)
goto send_current_mbox;
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
}
}
spin_lock_irqsave(
&phba->pport->work_port_lock,
iflag);
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
phba->pport->work_port_events &=
~WORKER_MBOX_TMO;
spin_unlock_irqrestore(
&phba->pport->work_port_lock,
iflag);
[SCSI] lpfc 8.2.4 : Miscellaneous Fixes Miscellaneous Fixes: - Fix a couple of sparse complaints - Reset the FCP recovery flag when the node is not a FCP2 device. - Speed up offline prep delays - Fixed a memory leak in lpfc_mem_alloc failure path - Fixed external loopback test. - Fixed error code returned from the driver when HBA is over heated. - Correct Max NPIV vport to limits read from adapter - Add missing locks around fc_flag and FC_NEEDS_REG_VPI - Add missing hba ids for device identification - Added support for SET_VARIABLE and MBX_WRITE_WWN mailbox commands - Changed all temperature event messages from warning to error - Fix reporting of link speed when link is down - Added support for MBX_WRITE_WWN mailbox command - Change del_timer_sync() in ISR to del_timer() in interrupt handler - Correct instances of beXX_to_cpu() that should be cpu_to_beXX() - Perform target flush before releasing node references on module unload - Avoid bogus devloss_tmo messages when driver unloads - Fix panic when HBA generates ERATT interupt - Fix mbox race condition and a workaround on back-to-back mailbox commands - Force NPIV off for pt2pt mode between 2 NPorts - Stop worker thread before removing fc_host. - Fix up discovery timeout error case due to missing clear_la - Tighten mailbox polling code to speed up detection of fast completions - Only allow DUMP_MEMORY if adapter offline due to overtemp errors - Added extended error information to the log messages in chip init. Signed-off-by: James Smart <James.Smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-01-10 23:52:54 -07:00
lpfc_mbox_cmpl_put(phba, pmb);
}
} else
spin_unlock_irqrestore(&phba->hbalock, iflag);
if ((work_ha_copy & HA_MBATT) &&
(phba->sli.mbox_active == NULL)) {
send_current_mbox:
/* Process next mailbox command if there is one */
[SCSI] lpfc 8.2.6 : Multiple discovery fixes Multiple Discovery Fixes: - Fix race on discovery due to link events coinciding with vport_delete. - Use NLP_FABRIC state to filter out switch-based pseudo initiators that reuse the same WWNs. - Correct erroneous setting of DID=0 in lpfc_matchdid() - Correct extra reference count that was in the lookup path for the remoteid from an unsolicited ELS. - Correct double-free bug in els abort path. - Correct FDMI server discovery logic for switch that return a WWN of 0. - Fix bugs in ndlp mgmt when a node changes address - Correct bug that did not delete RSCNs for vports upon link transitions - Fix "0216 Link event during NS query" error which pops up when vports are swapped to different switch ports. - Add sanity checks on ndlp structures - Fix devloss log message to dump WWN correctly - Hold off mgmt commands that were interferring with discovery mailbox cmds - Remove unnecessary FC_ESTABLISH_LINK logic. - Correct some race conditions in the worker thread, resulting in devloss: - Clear the work_port_events field before handling the work port events - Clear the deferred ring event before handling a deferred ring event - Hold the hba lock when waking up the work thread - Send an acc for the rscn even when we aren't going to handle it - Fix locking behavior that was not properly protecting the ACTIVE flag, thus allowing mailbox command order to shift. Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-07 08:15:56 -06:00
do {
rc = lpfc_sli_issue_mbox(phba, NULL,
MBX_NOWAIT);
} while (rc == MBX_NOT_FINISHED);
if (rc != MBX_SUCCESS)
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX |
LOG_SLI, "0349 rc should be "
"MBX_SUCCESS\n");
}
spin_lock_irqsave(&phba->hbalock, iflag);
phba->work_ha |= work_ha_copy;
spin_unlock_irqrestore(&phba->hbalock, iflag);
lpfc_worker_wake_up(phba);
}
return IRQ_HANDLED;
unplug_error:
spin_unlock_irqrestore(&phba->hbalock, iflag);
return IRQ_HANDLED;
} /* lpfc_sli_sp_intr_handler */
/**
* lpfc_sli_fp_intr_handler - Fast-path interrupt handler to SLI-3 device.
* @irq: Interrupt number.
* @dev_id: The device context pointer.
*
* This function is directly called from the PCI layer as an interrupt
* service routine when device with SLI-3 interface spec is enabled with
* MSI-X multi-message interrupt mode and there is a fast-path FCP IOCB
* ring event in the HBA. However, when the device is enabled with either
* MSI or Pin-IRQ interrupt mode, this function is called as part of the
* device-level interrupt handler. When the PCI slot is in error recovery
* or the HBA is undergoing initialization, the interrupt handler will not
* process the interrupt. The SCSI FCP fast-path ring event are handled in
* the intrrupt context. This function is called without any lock held.
* It gets the hbalock to access and update SLI data structures.
*
* This function returns IRQ_HANDLED when interrupt is handled else it
* returns IRQ_NONE.
**/
irqreturn_t
lpfc_sli_fp_intr_handler(int irq, void *dev_id)
{
struct lpfc_hba *phba;
uint32_t ha_copy;
unsigned long status;
unsigned long iflag;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *pring;
/* Get the driver's phba structure from the dev_id and
* assume the HBA is not interrupting.
*/
phba = (struct lpfc_hba *) dev_id;
if (unlikely(!phba))
return IRQ_NONE;
/*
* Stuff needs to be attented to when this function is invoked as an
* individual interrupt handler in MSI-X multi-message interrupt mode
*/
if (phba->intr_type == MSIX) {
/* Check device state for handling interrupt */
if (lpfc_intr_state_check(phba))
return IRQ_NONE;
/* Need to read HA REG for FCP ring and other ring events */
if (lpfc_readl(phba->HAregaddr, &ha_copy))
return IRQ_HANDLED;
/* Clear up only attention source related to fast-path */
spin_lock_irqsave(&phba->hbalock, iflag);
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/*
* If there is deferred error attention, do not check for
* any interrupt.
*/
if (unlikely(phba->hba_flag & DEFER_ERATT)) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
return IRQ_NONE;
}
writel((ha_copy & (HA_R0_CLR_MSK | HA_R1_CLR_MSK)),
phba->HAregaddr);
readl(phba->HAregaddr); /* flush */
spin_unlock_irqrestore(&phba->hbalock, iflag);
} else
ha_copy = phba->ha_copy;
/*
* Process all events on FCP ring. Take the optimized path for FCP IO.
*/
ha_copy &= ~(phba->work_ha_mask);
status = (ha_copy & (HA_RXMASK << (4*LPFC_FCP_RING)));
status >>= (4*LPFC_FCP_RING);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = &phba->sli.sli3_ring[LPFC_FCP_RING];
if (status & HA_RXMASK)
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli_handle_fast_ring_event(phba, pring, status);
if (phba->cfg_multi_ring_support == 2) {
/*
* Process all events on extra ring. Take the optimized path
* for extra ring IO.
*/
status = (ha_copy & (HA_RXMASK << (4*LPFC_EXTRA_RING)));
status >>= (4*LPFC_EXTRA_RING);
if (status & HA_RXMASK) {
lpfc_sli_handle_fast_ring_event(phba,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
&phba->sli.sli3_ring[LPFC_EXTRA_RING],
status);
}
}
return IRQ_HANDLED;
} /* lpfc_sli_fp_intr_handler */
/**
* lpfc_sli_intr_handler - Device-level interrupt handler to SLI-3 device
* @irq: Interrupt number.
* @dev_id: The device context pointer.
*
* This function is the HBA device-level interrupt handler to device with
* SLI-3 interface spec, called from the PCI layer when either MSI or
* Pin-IRQ interrupt mode is enabled and there is an event in the HBA which
* requires driver attention. This function invokes the slow-path interrupt
* attention handling function and fast-path interrupt attention handling
* function in turn to process the relevant HBA attention events. This
* function is called without any lock held. It gets the hbalock to access
* and update SLI data structures.
*
* This function returns IRQ_HANDLED when interrupt is handled, else it
* returns IRQ_NONE.
**/
irqreturn_t
lpfc_sli_intr_handler(int irq, void *dev_id)
{
struct lpfc_hba *phba;
irqreturn_t sp_irq_rc, fp_irq_rc;
unsigned long status1, status2;
uint32_t hc_copy;
/*
* Get the driver's phba structure from the dev_id and
* assume the HBA is not interrupting.
*/
phba = (struct lpfc_hba *) dev_id;
if (unlikely(!phba))
return IRQ_NONE;
/* Check device state for handling interrupt */
if (lpfc_intr_state_check(phba))
return IRQ_NONE;
spin_lock(&phba->hbalock);
if (lpfc_readl(phba->HAregaddr, &phba->ha_copy)) {
spin_unlock(&phba->hbalock);
return IRQ_HANDLED;
}
if (unlikely(!phba->ha_copy)) {
spin_unlock(&phba->hbalock);
return IRQ_NONE;
} else if (phba->ha_copy & HA_ERATT) {
if (phba->hba_flag & HBA_ERATT_HANDLED)
/* ERATT polling has handled ERATT */
phba->ha_copy &= ~HA_ERATT;
else
/* Indicate interrupt handler handles ERATT */
phba->hba_flag |= HBA_ERATT_HANDLED;
}
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
/*
* If there is deferred error attention, do not check for any interrupt.
*/
if (unlikely(phba->hba_flag & DEFER_ERATT)) {
spin_unlock(&phba->hbalock);
[SCSI] lpfc 8.3.1: misc fixes/changes 8.3.1 Fixes/Changes : - Fix incorrect byte-swapping on word 4 of IOCB (data length) which caused LUNs to not be discovered on big-endian (e.g. PPC) - Remove a bad cast of MBslimaddr which loses the __iomem (sparse) - Make lpfc_debugfs_mask_disc_trc static (sparse) - Correct misspelled word BlockGuard in lpfc_logmsg.h comment - Replaced repeated code segment for canceling IOCBs from a list with a function call, lpfc_sli_cancel_iocbs(). - Increased HBQ buffers to support 40KB SSC sequences. - Added sysfs interface to update speed and topology parameter without link bounce. - Fixed bug with sysfs fc_host WWNs not being updated after changing the WWNs. - Check if the active mailbox is NULL in the beginning of the mailbox timeout handler - fixes panic in the mailbox timeout handler while running IO stress test - Fixed system panic in lpfc_pci_remove_one() due to ndlp indirect reference to phba through vport - Removed de-reference of scsi device after call to scsi_done() to fix panic in scsi completion path while accessing scsi device after scsi_done is called. - Fixed "Nodelist not empty" message when unloading the driver after target reboot test - Added LP2105 HBA model description - Added code to print all 16 words of unrecognized ASYNC events - Fixed memory leak in vport create + delete loop - Added support for handling dual error bit from HBA - Fixed a driver NULL pointer dereference in lpfc_sli_process_sol_iocb - Fixed a discovery bug with FC switch reboot in lpfc_setup_disc_node - Take NULL termintator into account when calculating available buffer space Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2009-04-06 16:48:10 -06:00
return IRQ_NONE;
}
/* Clear attention sources except link and error attentions */
if (lpfc_readl(phba->HCregaddr, &hc_copy)) {
spin_unlock(&phba->hbalock);
return IRQ_HANDLED;
}
writel(hc_copy & ~(HC_MBINT_ENA | HC_R0INT_ENA | HC_R1INT_ENA
| HC_R2INT_ENA | HC_LAINT_ENA | HC_ERINT_ENA),
phba->HCregaddr);
writel((phba->ha_copy & ~(HA_LATT | HA_ERATT)), phba->HAregaddr);
writel(hc_copy, phba->HCregaddr);
readl(phba->HAregaddr); /* flush */
spin_unlock(&phba->hbalock);
/*
* Invokes slow-path host attention interrupt handling as appropriate.
*/
/* status of events with mailbox and link attention */
status1 = phba->ha_copy & (HA_MBATT | HA_LATT | HA_ERATT);
/* status of events with ELS ring */
status2 = (phba->ha_copy & (HA_RXMASK << (4*LPFC_ELS_RING)));
status2 >>= (4*LPFC_ELS_RING);
if (status1 || (status2 & HA_RXMASK))
sp_irq_rc = lpfc_sli_sp_intr_handler(irq, dev_id);
else
sp_irq_rc = IRQ_NONE;
/*
* Invoke fast-path host attention interrupt handling as appropriate.
*/
/* status of events with FCP ring */
status1 = (phba->ha_copy & (HA_RXMASK << (4*LPFC_FCP_RING)));
status1 >>= (4*LPFC_FCP_RING);
/* status of events with extra ring */
if (phba->cfg_multi_ring_support == 2) {
status2 = (phba->ha_copy & (HA_RXMASK << (4*LPFC_EXTRA_RING)));
status2 >>= (4*LPFC_EXTRA_RING);
} else
status2 = 0;
if ((status1 & HA_RXMASK) || (status2 & HA_RXMASK))
fp_irq_rc = lpfc_sli_fp_intr_handler(irq, dev_id);
else
fp_irq_rc = IRQ_NONE;
/* Return device-level interrupt handling status */
return (sp_irq_rc == IRQ_HANDLED) ? sp_irq_rc : fp_irq_rc;
} /* lpfc_sli_intr_handler */
/**
* lpfc_sli4_fcp_xri_abort_event_proc - Process fcp xri abort event
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked by the worker thread to process all the pending
* SLI4 FCP abort XRI events.
**/
void lpfc_sli4_fcp_xri_abort_event_proc(struct lpfc_hba *phba)
{
struct lpfc_cq_event *cq_event;
/* First, declare the fcp xri abort event has been handled */
spin_lock_irq(&phba->hbalock);
phba->hba_flag &= ~FCP_XRI_ABORT_EVENT;
spin_unlock_irq(&phba->hbalock);
/* Now, handle all the fcp xri abort events */
while (!list_empty(&phba->sli4_hba.sp_fcp_xri_aborted_work_queue)) {
/* Get the first event from the head of the event queue */
spin_lock_irq(&phba->hbalock);
list_remove_head(&phba->sli4_hba.sp_fcp_xri_aborted_work_queue,
cq_event, struct lpfc_cq_event, list);
spin_unlock_irq(&phba->hbalock);
/* Notify aborted XRI for FCP work queue */
lpfc_sli4_fcp_xri_aborted(phba, &cq_event->cqe.wcqe_axri);
/* Free the event processed back to the free pool */
lpfc_sli4_cq_event_release(phba, cq_event);
}
}
/**
* lpfc_sli4_els_xri_abort_event_proc - Process els xri abort event
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked by the worker thread to process all the pending
* SLI4 els abort xri events.
**/
void lpfc_sli4_els_xri_abort_event_proc(struct lpfc_hba *phba)
{
struct lpfc_cq_event *cq_event;
/* First, declare the els xri abort event has been handled */
spin_lock_irq(&phba->hbalock);
phba->hba_flag &= ~ELS_XRI_ABORT_EVENT;
spin_unlock_irq(&phba->hbalock);
/* Now, handle all the els xri abort events */
while (!list_empty(&phba->sli4_hba.sp_els_xri_aborted_work_queue)) {
/* Get the first event from the head of the event queue */
spin_lock_irq(&phba->hbalock);
list_remove_head(&phba->sli4_hba.sp_els_xri_aborted_work_queue,
cq_event, struct lpfc_cq_event, list);
spin_unlock_irq(&phba->hbalock);
/* Notify aborted XRI for ELS work queue */
lpfc_sli4_els_xri_aborted(phba, &cq_event->cqe.wcqe_axri);
/* Free the event processed back to the free pool */
lpfc_sli4_cq_event_release(phba, cq_event);
}
}
/**
* lpfc_sli4_iocb_param_transfer - Transfer pIocbOut and cmpl status to pIocbIn
* @phba: pointer to lpfc hba data structure
* @pIocbIn: pointer to the rspiocbq
* @pIocbOut: pointer to the cmdiocbq
* @wcqe: pointer to the complete wcqe
*
* This routine transfers the fields of a command iocbq to a response iocbq
* by copying all the IOCB fields from command iocbq and transferring the
* completion status information from the complete wcqe.
**/
static void
lpfc_sli4_iocb_param_transfer(struct lpfc_hba *phba,
struct lpfc_iocbq *pIocbIn,
struct lpfc_iocbq *pIocbOut,
struct lpfc_wcqe_complete *wcqe)
{
int numBdes, i;
unsigned long iflags;
uint32_t status, max_response;
struct lpfc_dmabuf *dmabuf;
struct ulp_bde64 *bpl, bde;
size_t offset = offsetof(struct lpfc_iocbq, iocb);
memcpy((char *)pIocbIn + offset, (char *)pIocbOut + offset,
sizeof(struct lpfc_iocbq) - offset);
/* Map WCQE parameters into irspiocb parameters */
status = bf_get(lpfc_wcqe_c_status, wcqe);
pIocbIn->iocb.ulpStatus = (status & LPFC_IOCB_STATUS_MASK);
if (pIocbOut->iocb_flag & LPFC_IO_FCP)
if (pIocbIn->iocb.ulpStatus == IOSTAT_FCP_RSP_ERROR)
pIocbIn->iocb.un.fcpi.fcpi_parm =
pIocbOut->iocb.un.fcpi.fcpi_parm -
wcqe->total_data_placed;
else
pIocbIn->iocb.un.ulpWord[4] = wcqe->parameter;
else {
pIocbIn->iocb.un.ulpWord[4] = wcqe->parameter;
switch (pIocbOut->iocb.ulpCommand) {
case CMD_ELS_REQUEST64_CR:
dmabuf = (struct lpfc_dmabuf *)pIocbOut->context3;
bpl = (struct ulp_bde64 *)dmabuf->virt;
bde.tus.w = le32_to_cpu(bpl[1].tus.w);
max_response = bde.tus.f.bdeSize;
break;
case CMD_GEN_REQUEST64_CR:
max_response = 0;
if (!pIocbOut->context3)
break;
numBdes = pIocbOut->iocb.un.genreq64.bdl.bdeSize/
sizeof(struct ulp_bde64);
dmabuf = (struct lpfc_dmabuf *)pIocbOut->context3;
bpl = (struct ulp_bde64 *)dmabuf->virt;
for (i = 0; i < numBdes; i++) {
bde.tus.w = le32_to_cpu(bpl[i].tus.w);
if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
max_response += bde.tus.f.bdeSize;
}
break;
default:
max_response = wcqe->total_data_placed;
break;
}
if (max_response < wcqe->total_data_placed)
pIocbIn->iocb.un.genreq64.bdl.bdeSize = max_response;
else
pIocbIn->iocb.un.genreq64.bdl.bdeSize =
wcqe->total_data_placed;
}
/* Convert BG errors for completion status */
if (status == CQE_STATUS_DI_ERROR) {
pIocbIn->iocb.ulpStatus = IOSTAT_LOCAL_REJECT;
if (bf_get(lpfc_wcqe_c_bg_edir, wcqe))
pIocbIn->iocb.un.ulpWord[4] = IOERR_RX_DMA_FAILED;
else
pIocbIn->iocb.un.ulpWord[4] = IOERR_TX_DMA_FAILED;
pIocbIn->iocb.unsli3.sli3_bg.bgstat = 0;
if (bf_get(lpfc_wcqe_c_bg_ge, wcqe)) /* Guard Check failed */
pIocbIn->iocb.unsli3.sli3_bg.bgstat |=
BGS_GUARD_ERR_MASK;
if (bf_get(lpfc_wcqe_c_bg_ae, wcqe)) /* App Tag Check failed */
pIocbIn->iocb.unsli3.sli3_bg.bgstat |=
BGS_APPTAG_ERR_MASK;
if (bf_get(lpfc_wcqe_c_bg_re, wcqe)) /* Ref Tag Check failed */
pIocbIn->iocb.unsli3.sli3_bg.bgstat |=
BGS_REFTAG_ERR_MASK;
/* Check to see if there was any good data before the error */
if (bf_get(lpfc_wcqe_c_bg_tdpv, wcqe)) {
pIocbIn->iocb.unsli3.sli3_bg.bgstat |=
BGS_HI_WATER_MARK_PRESENT_MASK;
pIocbIn->iocb.unsli3.sli3_bg.bghm =
wcqe->total_data_placed;
}
/*
* Set ALL the error bits to indicate we don't know what
* type of error it is.
*/
if (!pIocbIn->iocb.unsli3.sli3_bg.bgstat)
pIocbIn->iocb.unsli3.sli3_bg.bgstat |=
(BGS_REFTAG_ERR_MASK | BGS_APPTAG_ERR_MASK |
BGS_GUARD_ERR_MASK);
}
/* Pick up HBA exchange busy condition */
if (bf_get(lpfc_wcqe_c_xb, wcqe)) {
spin_lock_irqsave(&phba->hbalock, iflags);
pIocbIn->iocb_flag |= LPFC_EXCHANGE_BUSY;
spin_unlock_irqrestore(&phba->hbalock, iflags);
}
}
/**
* lpfc_sli4_els_wcqe_to_rspiocbq - Get response iocbq from els wcqe
* @phba: Pointer to HBA context object.
* @wcqe: Pointer to work-queue completion queue entry.
*
* This routine handles an ELS work-queue completion event and construct
* a pseudo response ELS IODBQ from the SLI4 ELS WCQE for the common
* discovery engine to handle.
*
* Return: Pointer to the receive IOCBQ, NULL otherwise.
**/
static struct lpfc_iocbq *
lpfc_sli4_els_wcqe_to_rspiocbq(struct lpfc_hba *phba,
struct lpfc_iocbq *irspiocbq)
{
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *pring;
struct lpfc_iocbq *cmdiocbq;
struct lpfc_wcqe_complete *wcqe;
unsigned long iflags;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = lpfc_phba_elsring(phba);
if (unlikely(!pring))
return NULL;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
wcqe = &irspiocbq->cq_event.cqe.wcqe_cmpl;
spin_lock_irqsave(&pring->ring_lock, iflags);
pring->stats.iocb_event++;
/* Look up the ELS command IOCB and create pseudo response IOCB */
cmdiocbq = lpfc_sli_iocbq_lookup_by_tag(phba, pring,
bf_get(lpfc_wcqe_c_request_tag, wcqe));
if (unlikely(!cmdiocbq)) {
spin_unlock_irqrestore(&pring->ring_lock, iflags);
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0386 ELS complete with no corresponding "
"cmdiocb: 0x%x 0x%x 0x%x 0x%x\n",
wcqe->word0, wcqe->total_data_placed,
wcqe->parameter, wcqe->word3);
lpfc_sli_release_iocbq(phba, irspiocbq);
return NULL;
}
/* Put the iocb back on the txcmplq */
lpfc_sli_ringtxcmpl_put(phba, pring, cmdiocbq);
spin_unlock_irqrestore(&pring->ring_lock, iflags);
/* Fake the irspiocbq and copy necessary response information */
lpfc_sli4_iocb_param_transfer(phba, irspiocbq, cmdiocbq, wcqe);
return irspiocbq;
}
inline struct lpfc_cq_event *
lpfc_cq_event_setup(struct lpfc_hba *phba, void *entry, int size)
{
struct lpfc_cq_event *cq_event;
/* Allocate a new internal CQ_EVENT entry */
cq_event = lpfc_sli4_cq_event_alloc(phba);
if (!cq_event) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0602 Failed to alloc CQ_EVENT entry\n");
return NULL;
}
/* Move the CQE into the event */
memcpy(&cq_event->cqe, entry, size);
return cq_event;
}
/**
* lpfc_sli4_sp_handle_async_event - Handle an asynchroous event
* @phba: Pointer to HBA context object.
* @cqe: Pointer to mailbox completion queue entry.
*
* This routine process a mailbox completion queue entry with asynchrous
* event.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_sp_handle_async_event(struct lpfc_hba *phba, struct lpfc_mcqe *mcqe)
{
struct lpfc_cq_event *cq_event;
unsigned long iflags;
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0392 Async Event: word0:x%x, word1:x%x, "
"word2:x%x, word3:x%x\n", mcqe->word0,
mcqe->mcqe_tag0, mcqe->mcqe_tag1, mcqe->trailer);
cq_event = lpfc_cq_event_setup(phba, mcqe, sizeof(struct lpfc_mcqe));
if (!cq_event)
return false;
spin_lock_irqsave(&phba->hbalock, iflags);
list_add_tail(&cq_event->list, &phba->sli4_hba.sp_asynce_work_queue);
/* Set the async event flag */
phba->hba_flag |= ASYNC_EVENT;
spin_unlock_irqrestore(&phba->hbalock, iflags);
return true;
}
/**
* lpfc_sli4_sp_handle_mbox_event - Handle a mailbox completion event
* @phba: Pointer to HBA context object.
* @cqe: Pointer to mailbox completion queue entry.
*
* This routine process a mailbox completion queue entry with mailbox
* completion event.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_sp_handle_mbox_event(struct lpfc_hba *phba, struct lpfc_mcqe *mcqe)
{
uint32_t mcqe_status;
MAILBOX_t *mbox, *pmbox;
struct lpfc_mqe *mqe;
struct lpfc_vport *vport;
struct lpfc_nodelist *ndlp;
struct lpfc_dmabuf *mp;
unsigned long iflags;
LPFC_MBOXQ_t *pmb;
bool workposted = false;
int rc;
/* If not a mailbox complete MCQE, out by checking mailbox consume */
if (!bf_get(lpfc_trailer_completed, mcqe))
goto out_no_mqe_complete;
/* Get the reference to the active mbox command */
spin_lock_irqsave(&phba->hbalock, iflags);
pmb = phba->sli.mbox_active;
if (unlikely(!pmb)) {
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX,
"1832 No pending MBOX command to handle\n");
spin_unlock_irqrestore(&phba->hbalock, iflags);
goto out_no_mqe_complete;
}
spin_unlock_irqrestore(&phba->hbalock, iflags);
mqe = &pmb->u.mqe;
pmbox = (MAILBOX_t *)&pmb->u.mqe;
mbox = phba->mbox;
vport = pmb->vport;
/* Reset heartbeat timer */
phba->last_completion_time = jiffies;
del_timer(&phba->sli.mbox_tmo);
/* Move mbox data to caller's mailbox region, do endian swapping */
if (pmb->mbox_cmpl && mbox)
lpfc_sli_pcimem_bcopy(mbox, mqe, sizeof(struct lpfc_mqe));
2011-10-10 19:32:10 -06:00
/*
* For mcqe errors, conditionally move a modified error code to
* the mbox so that the error will not be missed.
*/
mcqe_status = bf_get(lpfc_mcqe_status, mcqe);
if (mcqe_status != MB_CQE_STATUS_SUCCESS) {
if (bf_get(lpfc_mqe_status, mqe) == MBX_SUCCESS)
bf_set(lpfc_mqe_status, mqe,
(LPFC_MBX_ERROR_RANGE | mcqe_status));
}
if (pmb->mbox_flag & LPFC_MBX_IMED_UNREG) {
pmb->mbox_flag &= ~LPFC_MBX_IMED_UNREG;
lpfc_debugfs_disc_trc(vport, LPFC_DISC_TRC_MBOX_VPORT,
"MBOX dflt rpi: status:x%x rpi:x%x",
mcqe_status,
pmbox->un.varWords[0], 0);
if (mcqe_status == MB_CQE_STATUS_SUCCESS) {
mp = (struct lpfc_dmabuf *)(pmb->context1);
ndlp = (struct lpfc_nodelist *)pmb->context2;
/* Reg_LOGIN of dflt RPI was successful. Now lets get
* RID of the PPI using the same mbox buffer.
*/
lpfc_unreg_login(phba, vport->vpi,
pmbox->un.varWords[0], pmb);
pmb->mbox_cmpl = lpfc_mbx_cmpl_dflt_rpi;
pmb->context1 = mp;
pmb->context2 = ndlp;
pmb->vport = vport;
rc = lpfc_sli_issue_mbox(phba, pmb, MBX_NOWAIT);
if (rc != MBX_BUSY)
lpfc_printf_log(phba, KERN_ERR, LOG_MBOX |
LOG_SLI, "0385 rc should "
"have been MBX_BUSY\n");
if (rc != MBX_NOT_FINISHED)
goto send_current_mbox;
}
}
spin_lock_irqsave(&phba->pport->work_port_lock, iflags);
phba->pport->work_port_events &= ~WORKER_MBOX_TMO;
spin_unlock_irqrestore(&phba->pport->work_port_lock, iflags);
/* There is mailbox completion work to do */
spin_lock_irqsave(&phba->hbalock, iflags);
__lpfc_mbox_cmpl_put(phba, pmb);
phba->work_ha |= HA_MBATT;
spin_unlock_irqrestore(&phba->hbalock, iflags);
workposted = true;
send_current_mbox:
spin_lock_irqsave(&phba->hbalock, iflags);
/* Release the mailbox command posting token */
phba->sli.sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
/* Setting active mailbox pointer need to be in sync to flag clear */
phba->sli.mbox_active = NULL;
spin_unlock_irqrestore(&phba->hbalock, iflags);
/* Wake up worker thread to post the next pending mailbox command */
lpfc_worker_wake_up(phba);
out_no_mqe_complete:
if (bf_get(lpfc_trailer_consumed, mcqe))
lpfc_sli4_mq_release(phba->sli4_hba.mbx_wq);
return workposted;
}
/**
* lpfc_sli4_sp_handle_mcqe - Process a mailbox completion queue entry
* @phba: Pointer to HBA context object.
* @cqe: Pointer to mailbox completion queue entry.
*
* This routine process a mailbox completion queue entry, it invokes the
* proper mailbox complete handling or asynchrous event handling routine
* according to the MCQE's async bit.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_sp_handle_mcqe(struct lpfc_hba *phba, struct lpfc_cqe *cqe)
{
struct lpfc_mcqe mcqe;
bool workposted;
/* Copy the mailbox MCQE and convert endian order as needed */
lpfc_sli_pcimem_bcopy(cqe, &mcqe, sizeof(struct lpfc_mcqe));
/* Invoke the proper event handling routine */
if (!bf_get(lpfc_trailer_async, &mcqe))
workposted = lpfc_sli4_sp_handle_mbox_event(phba, &mcqe);
else
workposted = lpfc_sli4_sp_handle_async_event(phba, &mcqe);
return workposted;
}
/**
* lpfc_sli4_sp_handle_els_wcqe - Handle els work-queue completion event
* @phba: Pointer to HBA context object.
* @cq: Pointer to associated CQ
* @wcqe: Pointer to work-queue completion queue entry.
*
* This routine handles an ELS work-queue completion event.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_sp_handle_els_wcqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
struct lpfc_wcqe_complete *wcqe)
{
struct lpfc_iocbq *irspiocbq;
unsigned long iflags;
struct lpfc_sli_ring *pring = cq->pring;
int txq_cnt = 0;
int txcmplq_cnt = 0;
int fcp_txcmplq_cnt = 0;
/* Get an irspiocbq for later ELS response processing use */
irspiocbq = lpfc_sli_get_iocbq(phba);
if (!irspiocbq) {
if (!list_empty(&pring->txq))
txq_cnt++;
if (!list_empty(&pring->txcmplq))
txcmplq_cnt++;
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0387 NO IOCBQ data: txq_cnt=%d iocb_cnt=%d "
"fcp_txcmplq_cnt=%d, els_txcmplq_cnt=%d\n",
txq_cnt, phba->iocb_cnt,
fcp_txcmplq_cnt,
txcmplq_cnt);
return false;
}
/* Save off the slow-path queue event for work thread to process */
memcpy(&irspiocbq->cq_event.cqe.wcqe_cmpl, wcqe, sizeof(*wcqe));
spin_lock_irqsave(&phba->hbalock, iflags);
list_add_tail(&irspiocbq->cq_event.list,
&phba->sli4_hba.sp_queue_event);
phba->hba_flag |= HBA_SP_QUEUE_EVT;
spin_unlock_irqrestore(&phba->hbalock, iflags);
return true;
}
/**
* lpfc_sli4_sp_handle_rel_wcqe - Handle slow-path WQ entry consumed event
* @phba: Pointer to HBA context object.
* @wcqe: Pointer to work-queue completion queue entry.
*
* This routine handles slow-path WQ entry consumed event by invoking the
* proper WQ release routine to the slow-path WQ.
**/
static void
lpfc_sli4_sp_handle_rel_wcqe(struct lpfc_hba *phba,
struct lpfc_wcqe_release *wcqe)
{
/* sanity check on queue memory */
if (unlikely(!phba->sli4_hba.els_wq))
return;
/* Check for the slow-path ELS work queue */
if (bf_get(lpfc_wcqe_r_wq_id, wcqe) == phba->sli4_hba.els_wq->queue_id)
lpfc_sli4_wq_release(phba->sli4_hba.els_wq,
bf_get(lpfc_wcqe_r_wqe_index, wcqe));
else
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"2579 Slow-path wqe consume event carries "
"miss-matched qid: wcqe-qid=x%x, sp-qid=x%x\n",
bf_get(lpfc_wcqe_r_wqe_index, wcqe),
phba->sli4_hba.els_wq->queue_id);
}
/**
* lpfc_sli4_sp_handle_abort_xri_wcqe - Handle a xri abort event
* @phba: Pointer to HBA context object.
* @cq: Pointer to a WQ completion queue.
* @wcqe: Pointer to work-queue completion queue entry.
*
* This routine handles an XRI abort event.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_sp_handle_abort_xri_wcqe(struct lpfc_hba *phba,
struct lpfc_queue *cq,
struct sli4_wcqe_xri_aborted *wcqe)
{
bool workposted = false;
struct lpfc_cq_event *cq_event;
unsigned long iflags;
switch (cq->subtype) {
case LPFC_FCP:
cq_event = lpfc_cq_event_setup(
phba, wcqe, sizeof(struct sli4_wcqe_xri_aborted));
if (!cq_event)
return false;
spin_lock_irqsave(&phba->hbalock, iflags);
list_add_tail(&cq_event->list,
&phba->sli4_hba.sp_fcp_xri_aborted_work_queue);
/* Set the fcp xri abort event flag */
phba->hba_flag |= FCP_XRI_ABORT_EVENT;
spin_unlock_irqrestore(&phba->hbalock, iflags);
workposted = true;
break;
case LPFC_NVME_LS: /* NVME LS uses ELS resources */
case LPFC_ELS:
cq_event = lpfc_cq_event_setup(
phba, wcqe, sizeof(struct sli4_wcqe_xri_aborted));
if (!cq_event)
return false;
spin_lock_irqsave(&phba->hbalock, iflags);
list_add_tail(&cq_event->list,
&phba->sli4_hba.sp_els_xri_aborted_work_queue);
/* Set the els xri abort event flag */
phba->hba_flag |= ELS_XRI_ABORT_EVENT;
spin_unlock_irqrestore(&phba->hbalock, iflags);
workposted = true;
break;
case LPFC_NVME:
/* Notify aborted XRI for NVME work queue */
if (phba->nvmet_support)
lpfc_sli4_nvmet_xri_aborted(phba, wcqe);
else
lpfc_sli4_nvme_xri_aborted(phba, wcqe);
workposted = false;
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0603 Invalid CQ subtype %d: "
"%08x %08x %08x %08x\n",
cq->subtype, wcqe->word0, wcqe->parameter,
wcqe->word2, wcqe->word3);
workposted = false;
break;
}
return workposted;
}
/**
* lpfc_sli4_sp_handle_rcqe - Process a receive-queue completion queue entry
* @phba: Pointer to HBA context object.
* @rcqe: Pointer to receive-queue completion queue entry.
*
* This routine process a receive-queue completion queue entry.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_sp_handle_rcqe(struct lpfc_hba *phba, struct lpfc_rcqe *rcqe)
{
bool workposted = false;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct fc_frame_header *fc_hdr;
struct lpfc_queue *hrq = phba->sli4_hba.hdr_rq;
struct lpfc_queue *drq = phba->sli4_hba.dat_rq;
struct lpfc_nvmet_tgtport *tgtp;
struct hbq_dmabuf *dma_buf;
uint32_t status, rq_id;
unsigned long iflags;
/* sanity check on queue memory */
if (unlikely(!hrq) || unlikely(!drq))
return workposted;
if (bf_get(lpfc_cqe_code, rcqe) == CQE_CODE_RECEIVE_V1)
rq_id = bf_get(lpfc_rcqe_rq_id_v1, rcqe);
else
rq_id = bf_get(lpfc_rcqe_rq_id, rcqe);
if (rq_id != hrq->queue_id)
goto out;
status = bf_get(lpfc_rcqe_status, rcqe);
switch (status) {
case FC_STATUS_RQ_BUF_LEN_EXCEEDED:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2537 Receive Frame Truncated!!\n");
case FC_STATUS_RQ_SUCCESS:
spin_lock_irqsave(&phba->hbalock, iflags);
lpfc_sli4_rq_release(hrq, drq);
dma_buf = lpfc_sli_hbqbuf_get(&phba->hbqs[0].hbq_buffer_list);
if (!dma_buf) {
hrq->RQ_no_buf_found++;
spin_unlock_irqrestore(&phba->hbalock, iflags);
goto out;
}
hrq->RQ_rcv_buf++;
hrq->RQ_buf_posted--;
memcpy(&dma_buf->cq_event.cqe.rcqe_cmpl, rcqe, sizeof(*rcqe));
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* If a NVME LS event (type 0x28), treat it as Fast path */
fc_hdr = (struct fc_frame_header *)dma_buf->hbuf.virt;
/* save off the frame for the word thread to process */
list_add_tail(&dma_buf->cq_event.list,
&phba->sli4_hba.sp_queue_event);
/* Frame received */
phba->hba_flag |= HBA_SP_QUEUE_EVT;
spin_unlock_irqrestore(&phba->hbalock, iflags);
workposted = true;
break;
case FC_STATUS_INSUFF_BUF_FRM_DISC:
if (phba->nvmet_support) {
tgtp = phba->targetport->private;
lpfc_printf_log(phba, KERN_ERR, LOG_SLI | LOG_NVME,
"6402 RQE Error x%x, posted %d err_cnt "
"%d: %x %x %x\n",
status, hrq->RQ_buf_posted,
hrq->RQ_no_posted_buf,
atomic_read(&tgtp->rcv_fcp_cmd_in),
atomic_read(&tgtp->rcv_fcp_cmd_out),
atomic_read(&tgtp->xmt_fcp_release));
}
/* fallthrough */
case FC_STATUS_INSUFF_BUF_NEED_BUF:
hrq->RQ_no_posted_buf++;
/* Post more buffers if possible */
spin_lock_irqsave(&phba->hbalock, iflags);
phba->hba_flag |= HBA_POST_RECEIVE_BUFFER;
spin_unlock_irqrestore(&phba->hbalock, iflags);
workposted = true;
break;
}
out:
return workposted;
}
/**
* lpfc_sli4_sp_handle_cqe - Process a slow path completion queue entry
* @phba: Pointer to HBA context object.
* @cq: Pointer to the completion queue.
* @wcqe: Pointer to a completion queue entry.
*
* This routine process a slow-path work-queue or receive queue completion queue
* entry.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_sp_handle_cqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
struct lpfc_cqe *cqe)
{
struct lpfc_cqe cqevt;
bool workposted = false;
/* Copy the work queue CQE and convert endian order if needed */
lpfc_sli_pcimem_bcopy(cqe, &cqevt, sizeof(struct lpfc_cqe));
/* Check and process for different type of WCQE and dispatch */
switch (bf_get(lpfc_cqe_code, &cqevt)) {
case CQE_CODE_COMPL_WQE:
/* Process the WQ/RQ complete event */
phba->last_completion_time = jiffies;
workposted = lpfc_sli4_sp_handle_els_wcqe(phba, cq,
(struct lpfc_wcqe_complete *)&cqevt);
break;
case CQE_CODE_RELEASE_WQE:
/* Process the WQ release event */
lpfc_sli4_sp_handle_rel_wcqe(phba,
(struct lpfc_wcqe_release *)&cqevt);
break;
case CQE_CODE_XRI_ABORTED:
/* Process the WQ XRI abort event */
phba->last_completion_time = jiffies;
workposted = lpfc_sli4_sp_handle_abort_xri_wcqe(phba, cq,
(struct sli4_wcqe_xri_aborted *)&cqevt);
break;
case CQE_CODE_RECEIVE:
case CQE_CODE_RECEIVE_V1:
/* Process the RQ event */
phba->last_completion_time = jiffies;
workposted = lpfc_sli4_sp_handle_rcqe(phba,
(struct lpfc_rcqe *)&cqevt);
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0388 Not a valid WCQE code: x%x\n",
bf_get(lpfc_cqe_code, &cqevt));
break;
}
return workposted;
}
/**
* lpfc_sli4_sp_handle_eqe - Process a slow-path event queue entry
* @phba: Pointer to HBA context object.
* @eqe: Pointer to fast-path event queue entry.
*
* This routine process a event queue entry from the slow-path event queue.
* It will check the MajorCode and MinorCode to determine this is for a
* completion event on a completion queue, if not, an error shall be logged
* and just return. Otherwise, it will get to the corresponding completion
* queue and process all the entries on that completion queue, rearm the
* completion queue, and then return.
*
**/
static void
lpfc_sli4_sp_handle_eqe(struct lpfc_hba *phba, struct lpfc_eqe *eqe,
struct lpfc_queue *speq)
{
struct lpfc_queue *cq = NULL, *childq;
uint16_t cqid;
/* Get the reference to the corresponding CQ */
cqid = bf_get_le32(lpfc_eqe_resource_id, eqe);
list_for_each_entry(childq, &speq->child_list, list) {
if (childq->queue_id == cqid) {
cq = childq;
break;
}
}
if (unlikely(!cq)) {
if (phba->sli.sli_flag & LPFC_SLI_ACTIVE)
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0365 Slow-path CQ identifier "
"(%d) does not exist\n", cqid);
return;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* Save EQ associated with this CQ */
cq->assoc_qp = speq;
if (!queue_work(phba->wq, &cq->spwork))
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0390 Cannot schedule soft IRQ "
"for CQ eqcqid=%d, cqid=%d on CPU %d\n",
cqid, cq->queue_id, smp_processor_id());
}
/**
* lpfc_sli4_sp_process_cq - Process a slow-path event queue entry
* @phba: Pointer to HBA context object.
*
* This routine process a event queue entry from the slow-path event queue.
* It will check the MajorCode and MinorCode to determine this is for a
* completion event on a completion queue, if not, an error shall be logged
* and just return. Otherwise, it will get to the corresponding completion
* queue and process all the entries on that completion queue, rearm the
* completion queue, and then return.
*
**/
static void
lpfc_sli4_sp_process_cq(struct work_struct *work)
{
struct lpfc_queue *cq =
container_of(work, struct lpfc_queue, spwork);
struct lpfc_hba *phba = cq->phba;
struct lpfc_cqe *cqe;
bool workposted = false;
int ccount = 0;
/* Process all the entries to the CQ */
switch (cq->type) {
case LPFC_MCQ:
while ((cqe = lpfc_sli4_cq_get(cq))) {
workposted |= lpfc_sli4_sp_handle_mcqe(phba, cqe);
if (!(++ccount % cq->entry_repost))
break;
cq->CQ_mbox++;
}
break;
case LPFC_WCQ:
while ((cqe = lpfc_sli4_cq_get(cq))) {
if (cq->subtype == LPFC_FCP ||
cq->subtype == LPFC_NVME) {
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
if (phba->ktime_on)
cq->isr_timestamp = ktime_get_ns();
else
cq->isr_timestamp = 0;
#endif
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
workposted |= lpfc_sli4_fp_handle_cqe(phba, cq,
cqe);
} else {
workposted |= lpfc_sli4_sp_handle_cqe(phba, cq,
cqe);
}
if (!(++ccount % cq->entry_repost))
break;
}
/* Track the max number of CQEs processed in 1 EQ */
if (ccount > cq->CQ_max_cqe)
cq->CQ_max_cqe = ccount;
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0370 Invalid completion queue type (%d)\n",
cq->type);
return;
}
/* Catch the no cq entry condition, log an error */
if (unlikely(ccount == 0))
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0371 No entry from the CQ: identifier "
"(x%x), type (%d)\n", cq->queue_id, cq->type);
/* In any case, flash and re-arm the RCQ */
phba->sli4_hba.sli4_cq_release(cq, LPFC_QUEUE_REARM);
/* wake up worker thread if there are works to be done */
if (workposted)
lpfc_worker_wake_up(phba);
}
/**
* lpfc_sli4_fp_handle_fcp_wcqe - Process fast-path work queue completion entry
* @phba: Pointer to HBA context object.
* @cq: Pointer to associated CQ
* @wcqe: Pointer to work-queue completion queue entry.
*
* This routine process a fast-path work queue completion entry from fast-path
* event queue for FCP command response completion.
**/
static void
lpfc_sli4_fp_handle_fcp_wcqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
struct lpfc_wcqe_complete *wcqe)
{
struct lpfc_sli_ring *pring = cq->pring;
struct lpfc_iocbq *cmdiocbq;
struct lpfc_iocbq irspiocbq;
unsigned long iflags;
/* Check for response status */
if (unlikely(bf_get(lpfc_wcqe_c_status, wcqe))) {
/* If resource errors reported from HBA, reduce queue
* depth of the SCSI device.
*/
if (((bf_get(lpfc_wcqe_c_status, wcqe) ==
IOSTAT_LOCAL_REJECT)) &&
((wcqe->parameter & IOERR_PARAM_MASK) ==
IOERR_NO_RESOURCES))
phba->lpfc_rampdown_queue_depth(phba);
/* Log the error status */
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0373 FCP complete error: status=x%x, "
"hw_status=x%x, total_data_specified=%d, "
"parameter=x%x, word3=x%x\n",
bf_get(lpfc_wcqe_c_status, wcqe),
bf_get(lpfc_wcqe_c_hw_status, wcqe),
wcqe->total_data_placed, wcqe->parameter,
wcqe->word3);
}
/* Look up the FCP command IOCB and create pseudo response IOCB */
spin_lock_irqsave(&pring->ring_lock, iflags);
pring->stats.iocb_event++;
cmdiocbq = lpfc_sli_iocbq_lookup_by_tag(phba, pring,
bf_get(lpfc_wcqe_c_request_tag, wcqe));
spin_unlock_irqrestore(&pring->ring_lock, iflags);
if (unlikely(!cmdiocbq)) {
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0374 FCP complete with no corresponding "
"cmdiocb: iotag (%d)\n",
bf_get(lpfc_wcqe_c_request_tag, wcqe));
return;
}
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
cmdiocbq->isr_timestamp = cq->isr_timestamp;
#endif
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (cmdiocbq->iocb_cmpl == NULL) {
if (cmdiocbq->wqe_cmpl) {
if (cmdiocbq->iocb_flag & LPFC_DRIVER_ABORTED) {
spin_lock_irqsave(&phba->hbalock, iflags);
cmdiocbq->iocb_flag &= ~LPFC_DRIVER_ABORTED;
spin_unlock_irqrestore(&phba->hbalock, iflags);
}
/* Pass the cmd_iocb and the wcqe to the upper layer */
(cmdiocbq->wqe_cmpl)(phba, cmdiocbq, wcqe);
return;
}
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0375 FCP cmdiocb not callback function "
"iotag: (%d)\n",
bf_get(lpfc_wcqe_c_request_tag, wcqe));
return;
}
/* Fake the irspiocb and copy necessary response information */
lpfc_sli4_iocb_param_transfer(phba, &irspiocbq, cmdiocbq, wcqe);
if (cmdiocbq->iocb_flag & LPFC_DRIVER_ABORTED) {
spin_lock_irqsave(&phba->hbalock, iflags);
cmdiocbq->iocb_flag &= ~LPFC_DRIVER_ABORTED;
spin_unlock_irqrestore(&phba->hbalock, iflags);
}
/* Pass the cmd_iocb and the rsp state to the upper layer */
(cmdiocbq->iocb_cmpl)(phba, cmdiocbq, &irspiocbq);
}
/**
* lpfc_sli4_fp_handle_rel_wcqe - Handle fast-path WQ entry consumed event
* @phba: Pointer to HBA context object.
* @cq: Pointer to completion queue.
* @wcqe: Pointer to work-queue completion queue entry.
*
* This routine handles an fast-path WQ entry consumed event by invoking the
* proper WQ release routine to the slow-path WQ.
**/
static void
lpfc_sli4_fp_handle_rel_wcqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
struct lpfc_wcqe_release *wcqe)
{
struct lpfc_queue *childwq;
bool wqid_matched = false;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
uint16_t hba_wqid;
/* Check for fast-path FCP work queue release */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
hba_wqid = bf_get(lpfc_wcqe_r_wq_id, wcqe);
list_for_each_entry(childwq, &cq->child_list, list) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (childwq->queue_id == hba_wqid) {
lpfc_sli4_wq_release(childwq,
bf_get(lpfc_wcqe_r_wqe_index, wcqe));
if (childwq->q_flag & HBA_NVMET_WQFULL)
lpfc_nvmet_wqfull_process(phba, childwq);
wqid_matched = true;
break;
}
}
/* Report warning log message if no match found */
if (wqid_matched != true)
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"2580 Fast-path wqe consume event carries "
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
"miss-matched qid: wcqe-qid=x%x\n", hba_wqid);
}
/**
* lpfc_sli4_nvmet_handle_rcqe - Process a receive-queue completion queue entry
* @phba: Pointer to HBA context object.
* @rcqe: Pointer to receive-queue completion queue entry.
*
* This routine process a receive-queue completion queue entry.
*
* Return: true if work posted to worker thread, otherwise false.
**/
static bool
lpfc_sli4_nvmet_handle_rcqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
struct lpfc_rcqe *rcqe)
{
bool workposted = false;
struct lpfc_queue *hrq;
struct lpfc_queue *drq;
struct rqb_dmabuf *dma_buf;
struct fc_frame_header *fc_hdr;
struct lpfc_nvmet_tgtport *tgtp;
uint32_t status, rq_id;
unsigned long iflags;
uint32_t fctl, idx;
if ((phba->nvmet_support == 0) ||
(phba->sli4_hba.nvmet_cqset == NULL))
return workposted;
idx = cq->queue_id - phba->sli4_hba.nvmet_cqset[0]->queue_id;
hrq = phba->sli4_hba.nvmet_mrq_hdr[idx];
drq = phba->sli4_hba.nvmet_mrq_data[idx];
/* sanity check on queue memory */
if (unlikely(!hrq) || unlikely(!drq))
return workposted;
if (bf_get(lpfc_cqe_code, rcqe) == CQE_CODE_RECEIVE_V1)
rq_id = bf_get(lpfc_rcqe_rq_id_v1, rcqe);
else
rq_id = bf_get(lpfc_rcqe_rq_id, rcqe);
if ((phba->nvmet_support == 0) ||
(rq_id != hrq->queue_id))
return workposted;
status = bf_get(lpfc_rcqe_status, rcqe);
switch (status) {
case FC_STATUS_RQ_BUF_LEN_EXCEEDED:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"6126 Receive Frame Truncated!!\n");
/* Drop thru */
case FC_STATUS_RQ_SUCCESS:
spin_lock_irqsave(&phba->hbalock, iflags);
lpfc_sli4_rq_release(hrq, drq);
dma_buf = lpfc_sli_rqbuf_get(phba, hrq);
if (!dma_buf) {
hrq->RQ_no_buf_found++;
spin_unlock_irqrestore(&phba->hbalock, iflags);
goto out;
}
spin_unlock_irqrestore(&phba->hbalock, iflags);
hrq->RQ_rcv_buf++;
hrq->RQ_buf_posted--;
fc_hdr = (struct fc_frame_header *)dma_buf->hbuf.virt;
/* Just some basic sanity checks on FCP Command frame */
fctl = (fc_hdr->fh_f_ctl[0] << 16 |
fc_hdr->fh_f_ctl[1] << 8 |
fc_hdr->fh_f_ctl[2]);
if (((fctl &
(FC_FC_FIRST_SEQ | FC_FC_END_SEQ | FC_FC_SEQ_INIT)) !=
(FC_FC_FIRST_SEQ | FC_FC_END_SEQ | FC_FC_SEQ_INIT)) ||
(fc_hdr->fh_seq_cnt != 0)) /* 0 byte swapped is still 0 */
goto drop;
if (fc_hdr->fh_type == FC_TYPE_FCP) {
dma_buf->bytes_recv = bf_get(lpfc_rcqe_length, rcqe);
lpfc_nvmet_unsol_fcp_event(
phba, idx, dma_buf,
cq->isr_timestamp);
return false;
}
drop:
lpfc_in_buf_free(phba, &dma_buf->dbuf);
break;
case FC_STATUS_INSUFF_BUF_FRM_DISC:
if (phba->nvmet_support) {
tgtp = phba->targetport->private;
lpfc_printf_log(phba, KERN_ERR, LOG_SLI | LOG_NVME,
"6401 RQE Error x%x, posted %d err_cnt "
"%d: %x %x %x\n",
status, hrq->RQ_buf_posted,
hrq->RQ_no_posted_buf,
atomic_read(&tgtp->rcv_fcp_cmd_in),
atomic_read(&tgtp->rcv_fcp_cmd_out),
atomic_read(&tgtp->xmt_fcp_release));
}
/* fallthrough */
case FC_STATUS_INSUFF_BUF_NEED_BUF:
hrq->RQ_no_posted_buf++;
/* Post more buffers if possible */
break;
}
out:
return workposted;
}
/**
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* lpfc_sli4_fp_handle_cqe - Process fast-path work queue completion entry
* @cq: Pointer to the completion queue.
* @eqe: Pointer to fast-path completion queue entry.
*
* This routine process a fast-path work queue completion entry from fast-path
* event queue for FCP command response completion.
**/
static int
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli4_fp_handle_cqe(struct lpfc_hba *phba, struct lpfc_queue *cq,
struct lpfc_cqe *cqe)
{
struct lpfc_wcqe_release wcqe;
bool workposted = false;
/* Copy the work queue CQE and convert endian order if needed */
lpfc_sli_pcimem_bcopy(cqe, &wcqe, sizeof(struct lpfc_cqe));
/* Check and process for different type of WCQE and dispatch */
switch (bf_get(lpfc_wcqe_c_code, &wcqe)) {
case CQE_CODE_COMPL_WQE:
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
case CQE_CODE_NVME_ERSP:
cq->CQ_wq++;
/* Process the WQ complete event */
phba->last_completion_time = jiffies;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if ((cq->subtype == LPFC_FCP) || (cq->subtype == LPFC_NVME))
lpfc_sli4_fp_handle_fcp_wcqe(phba, cq,
(struct lpfc_wcqe_complete *)&wcqe);
if (cq->subtype == LPFC_NVME_LS)
lpfc_sli4_fp_handle_fcp_wcqe(phba, cq,
(struct lpfc_wcqe_complete *)&wcqe);
break;
case CQE_CODE_RELEASE_WQE:
cq->CQ_release_wqe++;
/* Process the WQ release event */
lpfc_sli4_fp_handle_rel_wcqe(phba, cq,
(struct lpfc_wcqe_release *)&wcqe);
break;
case CQE_CODE_XRI_ABORTED:
cq->CQ_xri_aborted++;
/* Process the WQ XRI abort event */
phba->last_completion_time = jiffies;
workposted = lpfc_sli4_sp_handle_abort_xri_wcqe(phba, cq,
(struct sli4_wcqe_xri_aborted *)&wcqe);
break;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
case CQE_CODE_RECEIVE_V1:
case CQE_CODE_RECEIVE:
phba->last_completion_time = jiffies;
if (cq->subtype == LPFC_NVMET) {
workposted = lpfc_sli4_nvmet_handle_rcqe(
phba, cq, (struct lpfc_rcqe *)&wcqe);
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
break;
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
"0144 Not a valid CQE code: x%x\n",
bf_get(lpfc_wcqe_c_code, &wcqe));
break;
}
return workposted;
}
/**
* lpfc_sli4_hba_handle_eqe - Process a fast-path event queue entry
* @phba: Pointer to HBA context object.
* @eqe: Pointer to fast-path event queue entry.
*
* This routine process a event queue entry from the fast-path event queue.
* It will check the MajorCode and MinorCode to determine this is for a
* completion event on a completion queue, if not, an error shall be logged
* and just return. Otherwise, it will get to the corresponding completion
* queue and process all the entries on the completion queue, rearm the
* completion queue, and then return.
**/
static void
lpfc_sli4_hba_handle_eqe(struct lpfc_hba *phba, struct lpfc_eqe *eqe,
uint32_t qidx)
{
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_queue *cq = NULL;
uint16_t cqid, id;
if (unlikely(bf_get_le32(lpfc_eqe_major_code, eqe) != 0)) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0366 Not a valid completion "
"event: majorcode=x%x, minorcode=x%x\n",
bf_get_le32(lpfc_eqe_major_code, eqe),
bf_get_le32(lpfc_eqe_minor_code, eqe));
return;
}
/* Get the reference to the corresponding CQ */
cqid = bf_get_le32(lpfc_eqe_resource_id, eqe);
if (phba->cfg_nvmet_mrq && phba->sli4_hba.nvmet_cqset) {
id = phba->sli4_hba.nvmet_cqset[0]->queue_id;
if ((cqid >= id) && (cqid < (id + phba->cfg_nvmet_mrq))) {
/* Process NVMET unsol rcv */
cq = phba->sli4_hba.nvmet_cqset[cqid - id];
goto process_cq;
}
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->sli4_hba.nvme_cq_map &&
(cqid == phba->sli4_hba.nvme_cq_map[qidx])) {
/* Process NVME / NVMET command completion */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
cq = phba->sli4_hba.nvme_cq[qidx];
goto process_cq;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->sli4_hba.fcp_cq_map &&
(cqid == phba->sli4_hba.fcp_cq_map[qidx])) {
/* Process FCP command completion */
cq = phba->sli4_hba.fcp_cq[qidx];
goto process_cq;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (phba->sli4_hba.nvmels_cq &&
(cqid == phba->sli4_hba.nvmels_cq->queue_id)) {
/* Process NVME unsol rcv */
cq = phba->sli4_hba.nvmels_cq;
}
/* Otherwise this is a Slow path event */
if (cq == NULL) {
lpfc_sli4_sp_handle_eqe(phba, eqe, phba->sli4_hba.hba_eq[qidx]);
return;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
process_cq:
if (unlikely(cqid != cq->queue_id)) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0368 Miss-matched fast-path completion "
"queue identifier: eqcqid=%d, fcpcqid=%d\n",
cqid, cq->queue_id);
return;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* Save EQ associated with this CQ */
cq->assoc_qp = phba->sli4_hba.hba_eq[qidx];
if (!queue_work(phba->wq, &cq->irqwork))
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0363 Cannot schedule soft IRQ "
"for CQ eqcqid=%d, cqid=%d on CPU %d\n",
cqid, cq->queue_id, smp_processor_id());
}
/**
* lpfc_sli4_hba_process_cq - Process a fast-path event queue entry
* @phba: Pointer to HBA context object.
* @eqe: Pointer to fast-path event queue entry.
*
* This routine process a event queue entry from the fast-path event queue.
* It will check the MajorCode and MinorCode to determine this is for a
* completion event on a completion queue, if not, an error shall be logged
* and just return. Otherwise, it will get to the corresponding completion
* queue and process all the entries on the completion queue, rearm the
* completion queue, and then return.
**/
static void
lpfc_sli4_hba_process_cq(struct work_struct *work)
{
struct lpfc_queue *cq =
container_of(work, struct lpfc_queue, irqwork);
struct lpfc_hba *phba = cq->phba;
struct lpfc_cqe *cqe;
bool workposted = false;
int ccount = 0;
/* Process all the entries to the CQ */
while ((cqe = lpfc_sli4_cq_get(cq))) {
#ifdef CONFIG_SCSI_LPFC_DEBUG_FS
if (phba->ktime_on)
cq->isr_timestamp = ktime_get_ns();
else
cq->isr_timestamp = 0;
#endif
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
workposted |= lpfc_sli4_fp_handle_cqe(phba, cq, cqe);
if (!(++ccount % cq->entry_repost))
break;
}
/* Track the max number of CQEs processed in 1 EQ */
if (ccount > cq->CQ_max_cqe)
cq->CQ_max_cqe = ccount;
cq->assoc_qp->EQ_cqe_cnt += ccount;
/* Catch the no cq entry condition */
if (unlikely(ccount == 0))
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0369 No entry from fast-path completion "
"queue fcpcqid=%d\n", cq->queue_id);
/* In any case, flash and re-arm the CQ */
phba->sli4_hba.sli4_cq_release(cq, LPFC_QUEUE_REARM);
/* wake up worker thread if there are works to be done */
if (workposted)
lpfc_worker_wake_up(phba);
}
static void
lpfc_sli4_eq_flush(struct lpfc_hba *phba, struct lpfc_queue *eq)
{
struct lpfc_eqe *eqe;
/* walk all the EQ entries and drop on the floor */
while ((eqe = lpfc_sli4_eq_get(eq)))
;
/* Clear and re-arm the EQ */
phba->sli4_hba.sli4_eq_release(eq, LPFC_QUEUE_REARM);
}
/**
* lpfc_sli4_fof_handle_eqe - Process a Flash Optimized Fabric event queue
* entry
* @phba: Pointer to HBA context object.
* @eqe: Pointer to fast-path event queue entry.
*
* This routine process a event queue entry from the Flash Optimized Fabric
* event queue. It will check the MajorCode and MinorCode to determine this
* is for a completion event on a completion queue, if not, an error shall be
* logged and just return. Otherwise, it will get to the corresponding
* completion queue and process all the entries on the completion queue, rearm
* the completion queue, and then return.
**/
static void
lpfc_sli4_fof_handle_eqe(struct lpfc_hba *phba, struct lpfc_eqe *eqe)
{
struct lpfc_queue *cq;
uint16_t cqid;
if (unlikely(bf_get_le32(lpfc_eqe_major_code, eqe) != 0)) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"9147 Not a valid completion "
"event: majorcode=x%x, minorcode=x%x\n",
bf_get_le32(lpfc_eqe_major_code, eqe),
bf_get_le32(lpfc_eqe_minor_code, eqe));
return;
}
/* Get the reference to the corresponding CQ */
cqid = bf_get_le32(lpfc_eqe_resource_id, eqe);
/* Next check for OAS */
cq = phba->sli4_hba.oas_cq;
if (unlikely(!cq)) {
if (phba->sli.sli_flag & LPFC_SLI_ACTIVE)
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"9148 OAS completion queue "
"does not exist\n");
return;
}
if (unlikely(cqid != cq->queue_id)) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"9149 Miss-matched fast-path compl "
"queue id: eqcqid=%d, fcpcqid=%d\n",
cqid, cq->queue_id);
return;
}
/* Save EQ associated with this CQ */
cq->assoc_qp = phba->sli4_hba.fof_eq;
/* CQ work will be processed on CPU affinitized to this IRQ */
if (!queue_work(phba->wq, &cq->irqwork))
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0367 Cannot schedule soft IRQ "
"for CQ eqcqid=%d, cqid=%d on CPU %d\n",
cqid, cq->queue_id, smp_processor_id());
}
/**
* lpfc_sli4_fof_intr_handler - HBA interrupt handler to SLI-4 device
* @irq: Interrupt number.
* @dev_id: The device context pointer.
*
* This function is directly called from the PCI layer as an interrupt
* service routine when device with SLI-4 interface spec is enabled with
* MSI-X multi-message interrupt mode and there is a Flash Optimized Fabric
* IOCB ring event in the HBA. However, when the device is enabled with either
* MSI or Pin-IRQ interrupt mode, this function is called as part of the
* device-level interrupt handler. When the PCI slot is in error recovery
* or the HBA is undergoing initialization, the interrupt handler will not
* process the interrupt. The Flash Optimized Fabric ring event are handled in
* the intrrupt context. This function is called without any lock held.
* It gets the hbalock to access and update SLI data structures. Note that,
* the EQ to CQ are one-to-one map such that the EQ index is
* equal to that of CQ index.
*
* This function returns IRQ_HANDLED when interrupt is handled else it
* returns IRQ_NONE.
**/
irqreturn_t
lpfc_sli4_fof_intr_handler(int irq, void *dev_id)
{
struct lpfc_hba *phba;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_hba_eq_hdl *hba_eq_hdl;
struct lpfc_queue *eq;
struct lpfc_eqe *eqe;
unsigned long iflag;
int ecount = 0;
/* Get the driver's phba structure from the dev_id */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
hba_eq_hdl = (struct lpfc_hba_eq_hdl *)dev_id;
phba = hba_eq_hdl->phba;
if (unlikely(!phba))
return IRQ_NONE;
/* Get to the EQ struct associated with this vector */
eq = phba->sli4_hba.fof_eq;
if (unlikely(!eq))
return IRQ_NONE;
/* Check device state for handling interrupt */
if (unlikely(lpfc_intr_state_check(phba))) {
/* Check again for link_state with lock held */
spin_lock_irqsave(&phba->hbalock, iflag);
if (phba->link_state < LPFC_LINK_DOWN)
/* Flush, clear interrupt, and rearm the EQ */
lpfc_sli4_eq_flush(phba, eq);
spin_unlock_irqrestore(&phba->hbalock, iflag);
return IRQ_NONE;
}
/*
* Process all the event on FCP fast-path EQ
*/
while ((eqe = lpfc_sli4_eq_get(eq))) {
lpfc_sli4_fof_handle_eqe(phba, eqe);
if (!(++ecount % eq->entry_repost))
break;
eq->EQ_processed++;
}
/* Track the max number of EQEs processed in 1 intr */
if (ecount > eq->EQ_max_eqe)
eq->EQ_max_eqe = ecount;
if (unlikely(ecount == 0)) {
eq->EQ_no_entry++;
if (phba->intr_type == MSIX)
/* MSI-X treated interrupt served as no EQ share INT */
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"9145 MSI-X interrupt with no EQE\n");
else {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"9146 ISR interrupt with no EQE\n");
/* Non MSI-X treated on interrupt as EQ share INT */
return IRQ_NONE;
}
}
/* Always clear and re-arm the fast-path EQ */
phba->sli4_hba.sli4_eq_release(eq, LPFC_QUEUE_REARM);
return IRQ_HANDLED;
}
/**
* lpfc_sli4_hba_intr_handler - HBA interrupt handler to SLI-4 device
* @irq: Interrupt number.
* @dev_id: The device context pointer.
*
* This function is directly called from the PCI layer as an interrupt
* service routine when device with SLI-4 interface spec is enabled with
* MSI-X multi-message interrupt mode and there is a fast-path FCP IOCB
* ring event in the HBA. However, when the device is enabled with either
* MSI or Pin-IRQ interrupt mode, this function is called as part of the
* device-level interrupt handler. When the PCI slot is in error recovery
* or the HBA is undergoing initialization, the interrupt handler will not
* process the interrupt. The SCSI FCP fast-path ring event are handled in
* the intrrupt context. This function is called without any lock held.
* It gets the hbalock to access and update SLI data structures. Note that,
* the FCP EQ to FCP CQ are one-to-one map such that the FCP EQ index is
* equal to that of FCP CQ index.
*
* The link attention and ELS ring attention events are handled
* by the worker thread. The interrupt handler signals the worker thread
* and returns for these events. This function is called without any lock
* held. It gets the hbalock to access and update SLI data structures.
*
* This function returns IRQ_HANDLED when interrupt is handled else it
* returns IRQ_NONE.
**/
irqreturn_t
lpfc_sli4_hba_intr_handler(int irq, void *dev_id)
{
struct lpfc_hba *phba;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_hba_eq_hdl *hba_eq_hdl;
struct lpfc_queue *fpeq;
struct lpfc_eqe *eqe;
unsigned long iflag;
int ecount = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int hba_eqidx;
/* Get the driver's phba structure from the dev_id */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
hba_eq_hdl = (struct lpfc_hba_eq_hdl *)dev_id;
phba = hba_eq_hdl->phba;
hba_eqidx = hba_eq_hdl->idx;
if (unlikely(!phba))
return IRQ_NONE;
if (unlikely(!phba->sli4_hba.hba_eq))
return IRQ_NONE;
/* Get to the EQ struct associated with this vector */
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
fpeq = phba->sli4_hba.hba_eq[hba_eqidx];
if (unlikely(!fpeq))
return IRQ_NONE;
if (lpfc_fcp_look_ahead) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (atomic_dec_and_test(&hba_eq_hdl->hba_eq_in_use))
phba->sli4_hba.sli4_eq_clr_intr(fpeq);
else {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
atomic_inc(&hba_eq_hdl->hba_eq_in_use);
return IRQ_NONE;
}
}
/* Check device state for handling interrupt */
if (unlikely(lpfc_intr_state_check(phba))) {
/* Check again for link_state with lock held */
spin_lock_irqsave(&phba->hbalock, iflag);
if (phba->link_state < LPFC_LINK_DOWN)
/* Flush, clear interrupt, and rearm the EQ */
lpfc_sli4_eq_flush(phba, fpeq);
spin_unlock_irqrestore(&phba->hbalock, iflag);
if (lpfc_fcp_look_ahead)
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
atomic_inc(&hba_eq_hdl->hba_eq_in_use);
return IRQ_NONE;
}
/*
* Process all the event on FCP fast-path EQ
*/
while ((eqe = lpfc_sli4_eq_get(fpeq))) {
lpfc_sli4_hba_handle_eqe(phba, eqe, hba_eqidx);
if (!(++ecount % fpeq->entry_repost))
break;
fpeq->EQ_processed++;
}
/* Track the max number of EQEs processed in 1 intr */
if (ecount > fpeq->EQ_max_eqe)
fpeq->EQ_max_eqe = ecount;
/* Always clear and re-arm the fast-path EQ */
phba->sli4_hba.sli4_eq_release(fpeq, LPFC_QUEUE_REARM);
if (unlikely(ecount == 0)) {
fpeq->EQ_no_entry++;
if (lpfc_fcp_look_ahead) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
atomic_inc(&hba_eq_hdl->hba_eq_in_use);
return IRQ_NONE;
}
if (phba->intr_type == MSIX)
/* MSI-X treated interrupt served as no EQ share INT */
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"0358 MSI-X interrupt with no EQE\n");
else
/* Non MSI-X treated on interrupt as EQ share INT */
return IRQ_NONE;
}
if (lpfc_fcp_look_ahead)
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
atomic_inc(&hba_eq_hdl->hba_eq_in_use);
return IRQ_HANDLED;
} /* lpfc_sli4_fp_intr_handler */
/**
* lpfc_sli4_intr_handler - Device-level interrupt handler for SLI-4 device
* @irq: Interrupt number.
* @dev_id: The device context pointer.
*
* This function is the device-level interrupt handler to device with SLI-4
* interface spec, called from the PCI layer when either MSI or Pin-IRQ
* interrupt mode is enabled and there is an event in the HBA which requires
* driver attention. This function invokes the slow-path interrupt attention
* handling function and fast-path interrupt attention handling function in
* turn to process the relevant HBA attention events. This function is called
* without any lock held. It gets the hbalock to access and update SLI data
* structures.
*
* This function returns IRQ_HANDLED when interrupt is handled, else it
* returns IRQ_NONE.
**/
irqreturn_t
lpfc_sli4_intr_handler(int irq, void *dev_id)
{
struct lpfc_hba *phba;
irqreturn_t hba_irq_rc;
bool hba_handled = false;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int qidx;
/* Get the driver's phba structure from the dev_id */
phba = (struct lpfc_hba *)dev_id;
if (unlikely(!phba))
return IRQ_NONE;
/*
* Invoke fast-path host attention interrupt handling as appropriate.
*/
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (qidx = 0; qidx < phba->io_channel_irqs; qidx++) {
hba_irq_rc = lpfc_sli4_hba_intr_handler(irq,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
&phba->sli4_hba.hba_eq_hdl[qidx]);
if (hba_irq_rc == IRQ_HANDLED)
hba_handled |= true;
}
if (phba->cfg_fof) {
hba_irq_rc = lpfc_sli4_fof_intr_handler(irq,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
&phba->sli4_hba.hba_eq_hdl[qidx]);
if (hba_irq_rc == IRQ_HANDLED)
hba_handled |= true;
}
return (hba_handled == true) ? IRQ_HANDLED : IRQ_NONE;
} /* lpfc_sli4_intr_handler */
/**
* lpfc_sli4_queue_free - free a queue structure and associated memory
* @queue: The queue structure to free.
*
* This function frees a queue structure and the DMAable memory used for
* the host resident queue. This function must be called after destroying the
* queue on the HBA.
**/
void
lpfc_sli4_queue_free(struct lpfc_queue *queue)
{
struct lpfc_dmabuf *dmabuf;
if (!queue)
return;
while (!list_empty(&queue->page_list)) {
list_remove_head(&queue->page_list, dmabuf, struct lpfc_dmabuf,
list);
dma_free_coherent(&queue->phba->pcidev->dev, queue->page_size,
dmabuf->virt, dmabuf->phys);
kfree(dmabuf);
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (queue->rqbp) {
lpfc_free_rq_buffer(queue->phba, queue);
kfree(queue->rqbp);
}
if (!list_empty(&queue->wq_list))
list_del(&queue->wq_list);
kfree(queue);
return;
}
/**
* lpfc_sli4_queue_alloc - Allocate and initialize a queue structure
* @phba: The HBA that this queue is being created on.
* @page_size: The size of a queue page
* @entry_size: The size of each queue entry for this queue.
* @entry count: The number of entries that this queue will handle.
*
* This function allocates a queue structure and the DMAable memory used for
* the host resident queue. This function must be called before creating the
* queue on the HBA.
**/
struct lpfc_queue *
lpfc_sli4_queue_alloc(struct lpfc_hba *phba, uint32_t page_size,
uint32_t entry_size, uint32_t entry_count)
{
struct lpfc_queue *queue;
struct lpfc_dmabuf *dmabuf;
int x, total_qe_count;
void *dma_pointer;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = page_size;
queue = kzalloc(sizeof(struct lpfc_queue) +
(sizeof(union sli4_qe) * entry_count), GFP_KERNEL);
if (!queue)
return NULL;
queue->page_count = (ALIGN(entry_size * entry_count,
hw_page_size))/hw_page_size;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* If needed, Adjust page count to match the max the adapter supports */
if (queue->page_count > phba->sli4_hba.pc_sli4_params.wqpcnt)
queue->page_count = phba->sli4_hba.pc_sli4_params.wqpcnt;
INIT_LIST_HEAD(&queue->list);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
INIT_LIST_HEAD(&queue->wq_list);
INIT_LIST_HEAD(&queue->wqfull_list);
INIT_LIST_HEAD(&queue->page_list);
INIT_LIST_HEAD(&queue->child_list);
/* Set queue parameters now. If the system cannot provide memory
* resources, the free routine needs to know what was allocated.
*/
queue->entry_size = entry_size;
queue->entry_count = entry_count;
queue->page_size = hw_page_size;
queue->phba = phba;
for (x = 0, total_qe_count = 0; x < queue->page_count; x++) {
dmabuf = kzalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
if (!dmabuf)
goto out_fail;
dmabuf->virt = dma_zalloc_coherent(&phba->pcidev->dev,
hw_page_size, &dmabuf->phys,
GFP_KERNEL);
if (!dmabuf->virt) {
kfree(dmabuf);
goto out_fail;
}
dmabuf->buffer_tag = x;
list_add_tail(&dmabuf->list, &queue->page_list);
/* initialize queue's entry array */
dma_pointer = dmabuf->virt;
for (; total_qe_count < entry_count &&
dma_pointer < (hw_page_size + dmabuf->virt);
total_qe_count++, dma_pointer += entry_size) {
queue->qe[total_qe_count].address = dma_pointer;
}
}
INIT_WORK(&queue->irqwork, lpfc_sli4_hba_process_cq);
INIT_WORK(&queue->spwork, lpfc_sli4_sp_process_cq);
/* entry_repost will be set during q creation */
return queue;
out_fail:
lpfc_sli4_queue_free(queue);
return NULL;
}
/**
* lpfc_dual_chute_pci_bar_map - Map pci base address register to host memory
* @phba: HBA structure that indicates port to create a queue on.
* @pci_barset: PCI BAR set flag.
*
* This function shall perform iomap of the specified PCI BAR address to host
* memory address if not already done so and return it. The returned host
* memory address can be NULL.
*/
static void __iomem *
lpfc_dual_chute_pci_bar_map(struct lpfc_hba *phba, uint16_t pci_barset)
{
if (!phba->pcidev)
return NULL;
switch (pci_barset) {
case WQ_PCI_BAR_0_AND_1:
return phba->pci_bar0_memmap_p;
case WQ_PCI_BAR_2_AND_3:
return phba->pci_bar2_memmap_p;
case WQ_PCI_BAR_4_AND_5:
return phba->pci_bar4_memmap_p;
default:
break;
}
return NULL;
}
/**
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* lpfc_modify_hba_eq_delay - Modify Delay Multiplier on FCP EQs
* @phba: HBA structure that indicates port to create a queue on.
* @startq: The starting FCP EQ to modify
*
* This function sends an MODIFY_EQ_DELAY mailbox command to the HBA.
* The command allows up to LPFC_MAX_EQ_DELAY_EQID_CNT EQ ID's to be
* updated in one mailbox command.
*
* The @phba struct is used to send mailbox command to HBA. The @startq
* is used to get the starting FCP EQ to change.
* This function is asynchronous and will wait for the mailbox
* command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int
lpfc_modify_hba_eq_delay(struct lpfc_hba *phba, uint32_t startq,
uint32_t numq, uint32_t imax)
{
struct lpfc_mbx_modify_eq_delay *eq_delay;
LPFC_MBOXQ_t *mbox;
struct lpfc_queue *eq;
int cnt, rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
uint32_t result, val;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int qidx;
union lpfc_sli4_cfg_shdr *shdr;
uint16_t dmult;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (startq >= phba->io_channel_irqs)
return 0;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_modify_eq_delay) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_MODIFY_EQ_DELAY,
length, LPFC_SLI4_MBX_EMBED);
eq_delay = &mbox->u.mqe.un.eq_delay;
/* Calculate delay multiper from maximum interrupt per second */
result = imax / phba->io_channel_irqs;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (result > LPFC_DMULT_CONST || result == 0)
dmult = 0;
else
dmult = LPFC_DMULT_CONST/result - 1;
if (dmult > LPFC_DMULT_MAX)
dmult = LPFC_DMULT_MAX;
cnt = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
for (qidx = startq; qidx < phba->io_channel_irqs; qidx++) {
eq = phba->sli4_hba.hba_eq[qidx];
if (!eq)
continue;
eq->q_mode = imax;
eq_delay->u.request.eq[cnt].eq_id = eq->queue_id;
eq_delay->u.request.eq[cnt].phase = 0;
eq_delay->u.request.eq[cnt].delay_multi = dmult;
cnt++;
/* q_mode is only used for auto_imax */
if (phba->sli.sli_flag & LPFC_SLI_USE_EQDR) {
/* Use EQ Delay Register method for q_mode */
/* Convert for EQ Delay register */
val = phba->cfg_fcp_imax;
if (val) {
/* First, interrupts per sec per EQ */
val = phba->cfg_fcp_imax /
phba->io_channel_irqs;
/* us delay between each interrupt */
val = LPFC_SEC_TO_USEC / val;
}
eq->q_mode = val;
} else {
eq->q_mode = imax;
}
if (cnt >= numq)
break;
}
eq_delay->u.request.num_eq = cnt;
mbox->vport = phba->pport;
mbox->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
mbox->context1 = NULL;
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
shdr = (union lpfc_sli4_cfg_shdr *) &eq_delay->header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2512 MODIFY_EQ_DELAY mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
}
mempool_free(mbox, phba->mbox_mem_pool);
return status;
}
/**
* lpfc_eq_create - Create an Event Queue on the HBA
* @phba: HBA structure that indicates port to create a queue on.
* @eq: The queue structure to use to create the event queue.
* @imax: The maximum interrupt per second limit.
*
* This function creates an event queue, as detailed in @eq, on a port,
* described by @phba by sending an EQ_CREATE mailbox command to the HBA.
*
* The @phba struct is used to send mailbox command to HBA. The @eq struct
* is used to get the entry count and entry size that are necessary to
* determine the number of pages to allocate and use for this queue. This
* function will send the EQ_CREATE mailbox command to the HBA to setup the
* event queue. This function is asynchronous and will wait for the mailbox
* command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int
lpfc_eq_create(struct lpfc_hba *phba, struct lpfc_queue *eq, uint32_t imax)
{
struct lpfc_mbx_eq_create *eq_create;
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
struct lpfc_dmabuf *dmabuf;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
uint16_t dmult;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
/* sanity check on queue memory */
if (!eq)
return -ENODEV;
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = SLI4_PAGE_SIZE;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_eq_create) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_EQ_CREATE,
length, LPFC_SLI4_MBX_EMBED);
eq_create = &mbox->u.mqe.un.eq_create;
shdr = (union lpfc_sli4_cfg_shdr *) &eq_create->header.cfg_shdr;
bf_set(lpfc_mbx_eq_create_num_pages, &eq_create->u.request,
eq->page_count);
bf_set(lpfc_eq_context_size, &eq_create->u.request.context,
LPFC_EQE_SIZE);
bf_set(lpfc_eq_context_valid, &eq_create->u.request.context, 1);
/* Use version 2 of CREATE_EQ if eqav is set */
if (phba->sli4_hba.pc_sli4_params.eqav) {
bf_set(lpfc_mbox_hdr_version, &shdr->request,
LPFC_Q_CREATE_VERSION_2);
bf_set(lpfc_eq_context_autovalid, &eq_create->u.request.context,
phba->sli4_hba.pc_sli4_params.eqav);
}
/* don't setup delay multiplier using EQ_CREATE */
dmult = 0;
bf_set(lpfc_eq_context_delay_multi, &eq_create->u.request.context,
dmult);
switch (eq->entry_count) {
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0360 Unsupported EQ count. (%d)\n",
eq->entry_count);
if (eq->entry_count < 256)
return -EINVAL;
/* otherwise default to smallest count (drop through) */
case 256:
bf_set(lpfc_eq_context_count, &eq_create->u.request.context,
LPFC_EQ_CNT_256);
break;
case 512:
bf_set(lpfc_eq_context_count, &eq_create->u.request.context,
LPFC_EQ_CNT_512);
break;
case 1024:
bf_set(lpfc_eq_context_count, &eq_create->u.request.context,
LPFC_EQ_CNT_1024);
break;
case 2048:
bf_set(lpfc_eq_context_count, &eq_create->u.request.context,
LPFC_EQ_CNT_2048);
break;
case 4096:
bf_set(lpfc_eq_context_count, &eq_create->u.request.context,
LPFC_EQ_CNT_4096);
break;
}
list_for_each_entry(dmabuf, &eq->page_list, list) {
memset(dmabuf->virt, 0, hw_page_size);
eq_create->u.request.page[dmabuf->buffer_tag].addr_lo =
putPaddrLow(dmabuf->phys);
eq_create->u.request.page[dmabuf->buffer_tag].addr_hi =
putPaddrHigh(dmabuf->phys);
}
mbox->vport = phba->pport;
mbox->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
mbox->context1 = NULL;
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2500 EQ_CREATE mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
}
eq->type = LPFC_EQ;
eq->subtype = LPFC_NONE;
eq->queue_id = bf_get(lpfc_mbx_eq_create_q_id, &eq_create->u.response);
if (eq->queue_id == 0xFFFF)
status = -ENXIO;
eq->host_index = 0;
eq->hba_index = 0;
eq->entry_repost = LPFC_EQ_REPOST;
mempool_free(mbox, phba->mbox_mem_pool);
return status;
}
/**
* lpfc_cq_create - Create a Completion Queue on the HBA
* @phba: HBA structure that indicates port to create a queue on.
* @cq: The queue structure to use to create the completion queue.
* @eq: The event queue to bind this completion queue to.
*
* This function creates a completion queue, as detailed in @wq, on a port,
* described by @phba by sending a CQ_CREATE mailbox command to the HBA.
*
* The @phba struct is used to send mailbox command to HBA. The @cq struct
* is used to get the entry count and entry size that are necessary to
* determine the number of pages to allocate and use for this queue. The @eq
* is used to indicate which event queue to bind this completion queue to. This
* function will send the CQ_CREATE mailbox command to the HBA to setup the
* completion queue. This function is asynchronous and will wait for the mailbox
* command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int
lpfc_cq_create(struct lpfc_hba *phba, struct lpfc_queue *cq,
struct lpfc_queue *eq, uint32_t type, uint32_t subtype)
{
struct lpfc_mbx_cq_create *cq_create;
struct lpfc_dmabuf *dmabuf;
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
/* sanity check on queue memory */
if (!cq || !eq)
return -ENODEV;
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = cq->page_size;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_cq_create) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_CQ_CREATE,
length, LPFC_SLI4_MBX_EMBED);
cq_create = &mbox->u.mqe.un.cq_create;
shdr = (union lpfc_sli4_cfg_shdr *) &cq_create->header.cfg_shdr;
bf_set(lpfc_mbx_cq_create_num_pages, &cq_create->u.request,
cq->page_count);
bf_set(lpfc_cq_context_event, &cq_create->u.request.context, 1);
bf_set(lpfc_cq_context_valid, &cq_create->u.request.context, 1);
bf_set(lpfc_mbox_hdr_version, &shdr->request,
phba->sli4_hba.pc_sli4_params.cqv);
if (phba->sli4_hba.pc_sli4_params.cqv == LPFC_Q_CREATE_VERSION_2) {
bf_set(lpfc_mbx_cq_create_page_size, &cq_create->u.request,
(cq->page_size / SLI4_PAGE_SIZE));
bf_set(lpfc_cq_eq_id_2, &cq_create->u.request.context,
eq->queue_id);
bf_set(lpfc_cq_context_autovalid, &cq_create->u.request.context,
phba->sli4_hba.pc_sli4_params.cqav);
} else {
bf_set(lpfc_cq_eq_id, &cq_create->u.request.context,
eq->queue_id);
}
switch (cq->entry_count) {
case 2048:
case 4096:
if (phba->sli4_hba.pc_sli4_params.cqv ==
LPFC_Q_CREATE_VERSION_2) {
cq_create->u.request.context.lpfc_cq_context_count =
cq->entry_count;
bf_set(lpfc_cq_context_count,
&cq_create->u.request.context,
LPFC_CQ_CNT_WORD7);
break;
}
/* Fall Thru */
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0361 Unsupported CQ count: "
"entry cnt %d sz %d pg cnt %d\n",
cq->entry_count, cq->entry_size,
cq->page_count);
if (cq->entry_count < 256) {
status = -EINVAL;
goto out;
}
/* otherwise default to smallest count (drop through) */
case 256:
bf_set(lpfc_cq_context_count, &cq_create->u.request.context,
LPFC_CQ_CNT_256);
break;
case 512:
bf_set(lpfc_cq_context_count, &cq_create->u.request.context,
LPFC_CQ_CNT_512);
break;
case 1024:
bf_set(lpfc_cq_context_count, &cq_create->u.request.context,
LPFC_CQ_CNT_1024);
break;
}
list_for_each_entry(dmabuf, &cq->page_list, list) {
memset(dmabuf->virt, 0, cq->page_size);
cq_create->u.request.page[dmabuf->buffer_tag].addr_lo =
putPaddrLow(dmabuf->phys);
cq_create->u.request.page[dmabuf->buffer_tag].addr_hi =
putPaddrHigh(dmabuf->phys);
}
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2501 CQ_CREATE mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
goto out;
}
cq->queue_id = bf_get(lpfc_mbx_cq_create_q_id, &cq_create->u.response);
if (cq->queue_id == 0xFFFF) {
status = -ENXIO;
goto out;
}
/* link the cq onto the parent eq child list */
list_add_tail(&cq->list, &eq->child_list);
/* Set up completion queue's type and subtype */
cq->type = type;
cq->subtype = subtype;
cq->queue_id = bf_get(lpfc_mbx_cq_create_q_id, &cq_create->u.response);
cq->assoc_qid = eq->queue_id;
cq->host_index = 0;
cq->hba_index = 0;
cq->entry_repost = LPFC_CQ_REPOST;
out:
mempool_free(mbox, phba->mbox_mem_pool);
return status;
}
/**
* lpfc_cq_create_set - Create a set of Completion Queues on the HBA for MRQ
* @phba: HBA structure that indicates port to create a queue on.
* @cqp: The queue structure array to use to create the completion queues.
* @eqp: The event queue array to bind these completion queues to.
*
* This function creates a set of completion queue, s to support MRQ
* as detailed in @cqp, on a port,
* described by @phba by sending a CREATE_CQ_SET mailbox command to the HBA.
*
* The @phba struct is used to send mailbox command to HBA. The @cq struct
* is used to get the entry count and entry size that are necessary to
* determine the number of pages to allocate and use for this queue. The @eq
* is used to indicate which event queue to bind this completion queue to. This
* function will send the CREATE_CQ_SET mailbox command to the HBA to setup the
* completion queue. This function is asynchronous and will wait for the mailbox
* command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int
lpfc_cq_create_set(struct lpfc_hba *phba, struct lpfc_queue **cqp,
struct lpfc_queue **eqp, uint32_t type, uint32_t subtype)
{
struct lpfc_queue *cq;
struct lpfc_queue *eq;
struct lpfc_mbx_cq_create_set *cq_set;
struct lpfc_dmabuf *dmabuf;
LPFC_MBOXQ_t *mbox;
int rc, length, alloclen, status = 0;
int cnt, idx, numcq, page_idx = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
/* sanity check on queue memory */
numcq = phba->cfg_nvmet_mrq;
if (!cqp || !eqp || !numcq)
return -ENODEV;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = sizeof(struct lpfc_mbx_cq_create_set);
length += ((numcq * cqp[0]->page_count) *
sizeof(struct dma_address));
alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_CQ_CREATE_SET, length,
LPFC_SLI4_MBX_NEMBED);
if (alloclen < length) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3098 Allocated DMA memory size (%d) is "
"less than the requested DMA memory size "
"(%d)\n", alloclen, length);
status = -ENOMEM;
goto out;
}
cq_set = mbox->sge_array->addr[0];
shdr = (union lpfc_sli4_cfg_shdr *)&cq_set->cfg_shdr;
bf_set(lpfc_mbox_hdr_version, &shdr->request, 0);
for (idx = 0; idx < numcq; idx++) {
cq = cqp[idx];
eq = eqp[idx];
if (!cq || !eq) {
status = -ENOMEM;
goto out;
}
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = cq->page_size;
switch (idx) {
case 0:
bf_set(lpfc_mbx_cq_create_set_page_size,
&cq_set->u.request,
(hw_page_size / SLI4_PAGE_SIZE));
bf_set(lpfc_mbx_cq_create_set_num_pages,
&cq_set->u.request, cq->page_count);
bf_set(lpfc_mbx_cq_create_set_evt,
&cq_set->u.request, 1);
bf_set(lpfc_mbx_cq_create_set_valid,
&cq_set->u.request, 1);
bf_set(lpfc_mbx_cq_create_set_cqe_size,
&cq_set->u.request, 0);
bf_set(lpfc_mbx_cq_create_set_num_cq,
&cq_set->u.request, numcq);
bf_set(lpfc_mbx_cq_create_set_autovalid,
&cq_set->u.request,
phba->sli4_hba.pc_sli4_params.cqav);
switch (cq->entry_count) {
case 2048:
case 4096:
if (phba->sli4_hba.pc_sli4_params.cqv ==
LPFC_Q_CREATE_VERSION_2) {
bf_set(lpfc_mbx_cq_create_set_cqe_cnt,
&cq_set->u.request,
cq->entry_count);
bf_set(lpfc_mbx_cq_create_set_cqe_cnt,
&cq_set->u.request,
LPFC_CQ_CNT_WORD7);
break;
}
/* Fall Thru */
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3118 Bad CQ count. (%d)\n",
cq->entry_count);
if (cq->entry_count < 256) {
status = -EINVAL;
goto out;
}
/* otherwise default to smallest (drop thru) */
case 256:
bf_set(lpfc_mbx_cq_create_set_cqe_cnt,
&cq_set->u.request, LPFC_CQ_CNT_256);
break;
case 512:
bf_set(lpfc_mbx_cq_create_set_cqe_cnt,
&cq_set->u.request, LPFC_CQ_CNT_512);
break;
case 1024:
bf_set(lpfc_mbx_cq_create_set_cqe_cnt,
&cq_set->u.request, LPFC_CQ_CNT_1024);
break;
}
bf_set(lpfc_mbx_cq_create_set_eq_id0,
&cq_set->u.request, eq->queue_id);
break;
case 1:
bf_set(lpfc_mbx_cq_create_set_eq_id1,
&cq_set->u.request, eq->queue_id);
break;
case 2:
bf_set(lpfc_mbx_cq_create_set_eq_id2,
&cq_set->u.request, eq->queue_id);
break;
case 3:
bf_set(lpfc_mbx_cq_create_set_eq_id3,
&cq_set->u.request, eq->queue_id);
break;
case 4:
bf_set(lpfc_mbx_cq_create_set_eq_id4,
&cq_set->u.request, eq->queue_id);
break;
case 5:
bf_set(lpfc_mbx_cq_create_set_eq_id5,
&cq_set->u.request, eq->queue_id);
break;
case 6:
bf_set(lpfc_mbx_cq_create_set_eq_id6,
&cq_set->u.request, eq->queue_id);
break;
case 7:
bf_set(lpfc_mbx_cq_create_set_eq_id7,
&cq_set->u.request, eq->queue_id);
break;
case 8:
bf_set(lpfc_mbx_cq_create_set_eq_id8,
&cq_set->u.request, eq->queue_id);
break;
case 9:
bf_set(lpfc_mbx_cq_create_set_eq_id9,
&cq_set->u.request, eq->queue_id);
break;
case 10:
bf_set(lpfc_mbx_cq_create_set_eq_id10,
&cq_set->u.request, eq->queue_id);
break;
case 11:
bf_set(lpfc_mbx_cq_create_set_eq_id11,
&cq_set->u.request, eq->queue_id);
break;
case 12:
bf_set(lpfc_mbx_cq_create_set_eq_id12,
&cq_set->u.request, eq->queue_id);
break;
case 13:
bf_set(lpfc_mbx_cq_create_set_eq_id13,
&cq_set->u.request, eq->queue_id);
break;
case 14:
bf_set(lpfc_mbx_cq_create_set_eq_id14,
&cq_set->u.request, eq->queue_id);
break;
case 15:
bf_set(lpfc_mbx_cq_create_set_eq_id15,
&cq_set->u.request, eq->queue_id);
break;
}
/* link the cq onto the parent eq child list */
list_add_tail(&cq->list, &eq->child_list);
/* Set up completion queue's type and subtype */
cq->type = type;
cq->subtype = subtype;
cq->assoc_qid = eq->queue_id;
cq->host_index = 0;
cq->hba_index = 0;
cq->entry_repost = LPFC_CQ_REPOST;
cq->chann = idx;
rc = 0;
list_for_each_entry(dmabuf, &cq->page_list, list) {
memset(dmabuf->virt, 0, hw_page_size);
cnt = page_idx + dmabuf->buffer_tag;
cq_set->u.request.page[cnt].addr_lo =
putPaddrLow(dmabuf->phys);
cq_set->u.request.page[cnt].addr_hi =
putPaddrHigh(dmabuf->phys);
rc++;
}
page_idx += rc;
}
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3119 CQ_CREATE_SET mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
goto out;
}
rc = bf_get(lpfc_mbx_cq_create_set_base_id, &cq_set->u.response);
if (rc == 0xFFFF) {
status = -ENXIO;
goto out;
}
for (idx = 0; idx < numcq; idx++) {
cq = cqp[idx];
cq->queue_id = rc + idx;
}
out:
lpfc_sli4_mbox_cmd_free(phba, mbox);
return status;
}
/**
* lpfc_mq_create_fb_init - Send MCC_CREATE without async events registration
* @phba: HBA structure that indicates port to create a queue on.
* @mq: The queue structure to use to create the mailbox queue.
* @mbox: An allocated pointer to type LPFC_MBOXQ_t
* @cq: The completion queue to associate with this cq.
*
* This function provides failback (fb) functionality when the
* mq_create_ext fails on older FW generations. It's purpose is identical
* to mq_create_ext otherwise.
*
* This routine cannot fail as all attributes were previously accessed and
* initialized in mq_create_ext.
**/
static void
lpfc_mq_create_fb_init(struct lpfc_hba *phba, struct lpfc_queue *mq,
LPFC_MBOXQ_t *mbox, struct lpfc_queue *cq)
{
struct lpfc_mbx_mq_create *mq_create;
struct lpfc_dmabuf *dmabuf;
int length;
length = (sizeof(struct lpfc_mbx_mq_create) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_MQ_CREATE,
length, LPFC_SLI4_MBX_EMBED);
mq_create = &mbox->u.mqe.un.mq_create;
bf_set(lpfc_mbx_mq_create_num_pages, &mq_create->u.request,
mq->page_count);
bf_set(lpfc_mq_context_cq_id, &mq_create->u.request.context,
cq->queue_id);
bf_set(lpfc_mq_context_valid, &mq_create->u.request.context, 1);
switch (mq->entry_count) {
case 16:
bf_set(lpfc_mq_context_ring_size, &mq_create->u.request.context,
LPFC_MQ_RING_SIZE_16);
break;
case 32:
bf_set(lpfc_mq_context_ring_size, &mq_create->u.request.context,
LPFC_MQ_RING_SIZE_32);
break;
case 64:
bf_set(lpfc_mq_context_ring_size, &mq_create->u.request.context,
LPFC_MQ_RING_SIZE_64);
break;
case 128:
bf_set(lpfc_mq_context_ring_size, &mq_create->u.request.context,
LPFC_MQ_RING_SIZE_128);
break;
}
list_for_each_entry(dmabuf, &mq->page_list, list) {
mq_create->u.request.page[dmabuf->buffer_tag].addr_lo =
putPaddrLow(dmabuf->phys);
mq_create->u.request.page[dmabuf->buffer_tag].addr_hi =
putPaddrHigh(dmabuf->phys);
}
}
/**
* lpfc_mq_create - Create a mailbox Queue on the HBA
* @phba: HBA structure that indicates port to create a queue on.
* @mq: The queue structure to use to create the mailbox queue.
* @cq: The completion queue to associate with this cq.
* @subtype: The queue's subtype.
*
* This function creates a mailbox queue, as detailed in @mq, on a port,
* described by @phba by sending a MQ_CREATE mailbox command to the HBA.
*
* The @phba struct is used to send mailbox command to HBA. The @cq struct
* is used to get the entry count and entry size that are necessary to
* determine the number of pages to allocate and use for this queue. This
* function will send the MQ_CREATE mailbox command to the HBA to setup the
* mailbox queue. This function is asynchronous and will wait for the mailbox
* command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int32_t
lpfc_mq_create(struct lpfc_hba *phba, struct lpfc_queue *mq,
struct lpfc_queue *cq, uint32_t subtype)
{
struct lpfc_mbx_mq_create *mq_create;
struct lpfc_mbx_mq_create_ext *mq_create_ext;
struct lpfc_dmabuf *dmabuf;
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
/* sanity check on queue memory */
if (!mq || !cq)
return -ENODEV;
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = SLI4_PAGE_SIZE;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_mq_create_ext) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_MQ_CREATE_EXT,
length, LPFC_SLI4_MBX_EMBED);
mq_create_ext = &mbox->u.mqe.un.mq_create_ext;
shdr = (union lpfc_sli4_cfg_shdr *) &mq_create_ext->header.cfg_shdr;
bf_set(lpfc_mbx_mq_create_ext_num_pages,
&mq_create_ext->u.request, mq->page_count);
bf_set(lpfc_mbx_mq_create_ext_async_evt_link,
&mq_create_ext->u.request, 1);
bf_set(lpfc_mbx_mq_create_ext_async_evt_fip,
&mq_create_ext->u.request, 1);
bf_set(lpfc_mbx_mq_create_ext_async_evt_group5,
&mq_create_ext->u.request, 1);
bf_set(lpfc_mbx_mq_create_ext_async_evt_fc,
&mq_create_ext->u.request, 1);
bf_set(lpfc_mbx_mq_create_ext_async_evt_sli,
&mq_create_ext->u.request, 1);
bf_set(lpfc_mq_context_valid, &mq_create_ext->u.request.context, 1);
bf_set(lpfc_mbox_hdr_version, &shdr->request,
phba->sli4_hba.pc_sli4_params.mqv);
if (phba->sli4_hba.pc_sli4_params.mqv == LPFC_Q_CREATE_VERSION_1)
bf_set(lpfc_mbx_mq_create_ext_cq_id, &mq_create_ext->u.request,
cq->queue_id);
else
bf_set(lpfc_mq_context_cq_id, &mq_create_ext->u.request.context,
cq->queue_id);
switch (mq->entry_count) {
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0362 Unsupported MQ count. (%d)\n",
mq->entry_count);
if (mq->entry_count < 16) {
status = -EINVAL;
goto out;
}
/* otherwise default to smallest count (drop through) */
case 16:
bf_set(lpfc_mq_context_ring_size,
&mq_create_ext->u.request.context,
LPFC_MQ_RING_SIZE_16);
break;
case 32:
bf_set(lpfc_mq_context_ring_size,
&mq_create_ext->u.request.context,
LPFC_MQ_RING_SIZE_32);
break;
case 64:
bf_set(lpfc_mq_context_ring_size,
&mq_create_ext->u.request.context,
LPFC_MQ_RING_SIZE_64);
break;
case 128:
bf_set(lpfc_mq_context_ring_size,
&mq_create_ext->u.request.context,
LPFC_MQ_RING_SIZE_128);
break;
}
list_for_each_entry(dmabuf, &mq->page_list, list) {
memset(dmabuf->virt, 0, hw_page_size);
mq_create_ext->u.request.page[dmabuf->buffer_tag].addr_lo =
putPaddrLow(dmabuf->phys);
mq_create_ext->u.request.page[dmabuf->buffer_tag].addr_hi =
putPaddrHigh(dmabuf->phys);
}
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
mq->queue_id = bf_get(lpfc_mbx_mq_create_q_id,
&mq_create_ext->u.response);
if (rc != MBX_SUCCESS) {
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2795 MQ_CREATE_EXT failed with "
"status x%x. Failback to MQ_CREATE.\n",
rc);
lpfc_mq_create_fb_init(phba, mq, mbox, cq);
mq_create = &mbox->u.mqe.un.mq_create;
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
shdr = (union lpfc_sli4_cfg_shdr *) &mq_create->header.cfg_shdr;
mq->queue_id = bf_get(lpfc_mbx_mq_create_q_id,
&mq_create->u.response);
}
/* The IOCTL status is embedded in the mailbox subheader. */
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2502 MQ_CREATE mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
goto out;
}
if (mq->queue_id == 0xFFFF) {
status = -ENXIO;
goto out;
}
mq->type = LPFC_MQ;
mq->assoc_qid = cq->queue_id;
mq->subtype = subtype;
mq->host_index = 0;
mq->hba_index = 0;
mq->entry_repost = LPFC_MQ_REPOST;
/* link the mq onto the parent cq child list */
list_add_tail(&mq->list, &cq->child_list);
out:
mempool_free(mbox, phba->mbox_mem_pool);
return status;
}
/**
* lpfc_wq_create - Create a Work Queue on the HBA
* @phba: HBA structure that indicates port to create a queue on.
* @wq: The queue structure to use to create the work queue.
* @cq: The completion queue to bind this work queue to.
* @subtype: The subtype of the work queue indicating its functionality.
*
* This function creates a work queue, as detailed in @wq, on a port, described
* by @phba by sending a WQ_CREATE mailbox command to the HBA.
*
* The @phba struct is used to send mailbox command to HBA. The @wq struct
* is used to get the entry count and entry size that are necessary to
* determine the number of pages to allocate and use for this queue. The @cq
* is used to indicate which completion queue to bind this work queue to. This
* function will send the WQ_CREATE mailbox command to the HBA to setup the
* work queue. This function is asynchronous and will wait for the mailbox
* command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int
lpfc_wq_create(struct lpfc_hba *phba, struct lpfc_queue *wq,
struct lpfc_queue *cq, uint32_t subtype)
{
struct lpfc_mbx_wq_create *wq_create;
struct lpfc_dmabuf *dmabuf;
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
struct dma_address *page;
void __iomem *bar_memmap_p;
uint32_t db_offset;
uint16_t pci_barset;
uint8_t dpp_barset;
uint32_t dpp_offset;
unsigned long pg_addr;
uint8_t wq_create_version;
/* sanity check on queue memory */
if (!wq || !cq)
return -ENODEV;
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = wq->page_size;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_wq_create) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_WQ_CREATE,
length, LPFC_SLI4_MBX_EMBED);
wq_create = &mbox->u.mqe.un.wq_create;
shdr = (union lpfc_sli4_cfg_shdr *) &wq_create->header.cfg_shdr;
bf_set(lpfc_mbx_wq_create_num_pages, &wq_create->u.request,
wq->page_count);
bf_set(lpfc_mbx_wq_create_cq_id, &wq_create->u.request,
cq->queue_id);
/* wqv is the earliest version supported, NOT the latest */
bf_set(lpfc_mbox_hdr_version, &shdr->request,
phba->sli4_hba.pc_sli4_params.wqv);
if ((phba->sli4_hba.pc_sli4_params.wqsize & LPFC_WQ_SZ128_SUPPORT) ||
(wq->page_size > SLI4_PAGE_SIZE))
wq_create_version = LPFC_Q_CREATE_VERSION_1;
else
wq_create_version = LPFC_Q_CREATE_VERSION_0;
if (phba->sli4_hba.pc_sli4_params.wqsize & LPFC_WQ_SZ128_SUPPORT)
wq_create_version = LPFC_Q_CREATE_VERSION_1;
else
wq_create_version = LPFC_Q_CREATE_VERSION_0;
switch (wq_create_version) {
case LPFC_Q_CREATE_VERSION_1:
bf_set(lpfc_mbx_wq_create_wqe_count, &wq_create->u.request_1,
wq->entry_count);
bf_set(lpfc_mbox_hdr_version, &shdr->request,
LPFC_Q_CREATE_VERSION_1);
switch (wq->entry_size) {
default:
case 64:
bf_set(lpfc_mbx_wq_create_wqe_size,
&wq_create->u.request_1,
LPFC_WQ_WQE_SIZE_64);
break;
case 128:
bf_set(lpfc_mbx_wq_create_wqe_size,
&wq_create->u.request_1,
LPFC_WQ_WQE_SIZE_128);
break;
}
/* Request DPP by default */
bf_set(lpfc_mbx_wq_create_dpp_req, &wq_create->u.request_1, 1);
bf_set(lpfc_mbx_wq_create_page_size,
&wq_create->u.request_1,
(wq->page_size / SLI4_PAGE_SIZE));
page = wq_create->u.request_1.page;
break;
default:
page = wq_create->u.request.page;
break;
}
list_for_each_entry(dmabuf, &wq->page_list, list) {
memset(dmabuf->virt, 0, hw_page_size);
page[dmabuf->buffer_tag].addr_lo = putPaddrLow(dmabuf->phys);
page[dmabuf->buffer_tag].addr_hi = putPaddrHigh(dmabuf->phys);
}
if (phba->sli4_hba.fw_func_mode & LPFC_DUA_MODE)
bf_set(lpfc_mbx_wq_create_dua, &wq_create->u.request, 1);
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2503 WQ_CREATE mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
goto out;
}
if (wq_create_version == LPFC_Q_CREATE_VERSION_0)
wq->queue_id = bf_get(lpfc_mbx_wq_create_q_id,
&wq_create->u.response);
else
wq->queue_id = bf_get(lpfc_mbx_wq_create_v1_q_id,
&wq_create->u.response_1);
if (wq->queue_id == 0xFFFF) {
status = -ENXIO;
goto out;
}
wq->db_format = LPFC_DB_LIST_FORMAT;
if (wq_create_version == LPFC_Q_CREATE_VERSION_0) {
if (phba->sli4_hba.fw_func_mode & LPFC_DUA_MODE) {
wq->db_format = bf_get(lpfc_mbx_wq_create_db_format,
&wq_create->u.response);
if ((wq->db_format != LPFC_DB_LIST_FORMAT) &&
(wq->db_format != LPFC_DB_RING_FORMAT)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3265 WQ[%d] doorbell format "
"not supported: x%x\n",
wq->queue_id, wq->db_format);
status = -EINVAL;
goto out;
}
pci_barset = bf_get(lpfc_mbx_wq_create_bar_set,
&wq_create->u.response);
bar_memmap_p = lpfc_dual_chute_pci_bar_map(phba,
pci_barset);
if (!bar_memmap_p) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3263 WQ[%d] failed to memmap "
"pci barset:x%x\n",
wq->queue_id, pci_barset);
status = -ENOMEM;
goto out;
}
db_offset = wq_create->u.response.doorbell_offset;
if ((db_offset != LPFC_ULP0_WQ_DOORBELL) &&
(db_offset != LPFC_ULP1_WQ_DOORBELL)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3252 WQ[%d] doorbell offset "
"not supported: x%x\n",
wq->queue_id, db_offset);
status = -EINVAL;
goto out;
}
wq->db_regaddr = bar_memmap_p + db_offset;
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"3264 WQ[%d]: barset:x%x, offset:x%x, "
"format:x%x\n", wq->queue_id,
pci_barset, db_offset, wq->db_format);
} else
wq->db_regaddr = phba->sli4_hba.WQDBregaddr;
} else {
/* Check if DPP was honored by the firmware */
wq->dpp_enable = bf_get(lpfc_mbx_wq_create_dpp_rsp,
&wq_create->u.response_1);
if (wq->dpp_enable) {
pci_barset = bf_get(lpfc_mbx_wq_create_v1_bar_set,
&wq_create->u.response_1);
bar_memmap_p = lpfc_dual_chute_pci_bar_map(phba,
pci_barset);
if (!bar_memmap_p) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3267 WQ[%d] failed to memmap "
"pci barset:x%x\n",
wq->queue_id, pci_barset);
status = -ENOMEM;
goto out;
}
db_offset = wq_create->u.response_1.doorbell_offset;
wq->db_regaddr = bar_memmap_p + db_offset;
wq->dpp_id = bf_get(lpfc_mbx_wq_create_dpp_id,
&wq_create->u.response_1);
dpp_barset = bf_get(lpfc_mbx_wq_create_dpp_bar,
&wq_create->u.response_1);
bar_memmap_p = lpfc_dual_chute_pci_bar_map(phba,
dpp_barset);
if (!bar_memmap_p) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3268 WQ[%d] failed to memmap "
"pci barset:x%x\n",
wq->queue_id, dpp_barset);
status = -ENOMEM;
goto out;
}
dpp_offset = wq_create->u.response_1.dpp_offset;
wq->dpp_regaddr = bar_memmap_p + dpp_offset;
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"3271 WQ[%d]: barset:x%x, offset:x%x, "
"dpp_id:x%x dpp_barset:x%x "
"dpp_offset:x%x\n",
wq->queue_id, pci_barset, db_offset,
wq->dpp_id, dpp_barset, dpp_offset);
/* Enable combined writes for DPP aperture */
pg_addr = (unsigned long)(wq->dpp_regaddr) & PAGE_MASK;
#ifdef CONFIG_X86
rc = set_memory_wc(pg_addr, 1);
if (rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3272 Cannot setup Combined "
"Write on WQ[%d] - disable DPP\n",
wq->queue_id);
phba->cfg_enable_dpp = 0;
}
#else
phba->cfg_enable_dpp = 0;
#endif
} else
wq->db_regaddr = phba->sli4_hba.WQDBregaddr;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
wq->pring = kzalloc(sizeof(struct lpfc_sli_ring), GFP_KERNEL);
if (wq->pring == NULL) {
status = -ENOMEM;
goto out;
}
wq->type = LPFC_WQ;
wq->assoc_qid = cq->queue_id;
wq->subtype = subtype;
wq->host_index = 0;
wq->hba_index = 0;
wq->entry_repost = LPFC_RELEASE_NOTIFICATION_INTERVAL;
/* link the wq onto the parent cq child list */
list_add_tail(&wq->list, &cq->child_list);
out:
mempool_free(mbox, phba->mbox_mem_pool);
return status;
}
/**
* lpfc_rq_create - Create a Receive Queue on the HBA
* @phba: HBA structure that indicates port to create a queue on.
* @hrq: The queue structure to use to create the header receive queue.
* @drq: The queue structure to use to create the data receive queue.
* @cq: The completion queue to bind this work queue to.
*
* This function creates a receive buffer queue pair , as detailed in @hrq and
* @drq, on a port, described by @phba by sending a RQ_CREATE mailbox command
* to the HBA.
*
* The @phba struct is used to send mailbox command to HBA. The @drq and @hrq
* struct is used to get the entry count that is necessary to determine the
* number of pages to use for this queue. The @cq is used to indicate which
* completion queue to bind received buffers that are posted to these queues to.
* This function will send the RQ_CREATE mailbox command to the HBA to setup the
* receive queue pair. This function is asynchronous and will wait for the
* mailbox command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int
lpfc_rq_create(struct lpfc_hba *phba, struct lpfc_queue *hrq,
struct lpfc_queue *drq, struct lpfc_queue *cq, uint32_t subtype)
{
struct lpfc_mbx_rq_create *rq_create;
struct lpfc_dmabuf *dmabuf;
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
void __iomem *bar_memmap_p;
uint32_t db_offset;
uint16_t pci_barset;
/* sanity check on queue memory */
if (!hrq || !drq || !cq)
return -ENODEV;
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = SLI4_PAGE_SIZE;
if (hrq->entry_count != drq->entry_count)
return -EINVAL;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_rq_create) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_RQ_CREATE,
length, LPFC_SLI4_MBX_EMBED);
rq_create = &mbox->u.mqe.un.rq_create;
shdr = (union lpfc_sli4_cfg_shdr *) &rq_create->header.cfg_shdr;
bf_set(lpfc_mbox_hdr_version, &shdr->request,
phba->sli4_hba.pc_sli4_params.rqv);
if (phba->sli4_hba.pc_sli4_params.rqv == LPFC_Q_CREATE_VERSION_1) {
bf_set(lpfc_rq_context_rqe_count_1,
&rq_create->u.request.context,
hrq->entry_count);
rq_create->u.request.context.buffer_size = LPFC_HDR_BUF_SIZE;
bf_set(lpfc_rq_context_rqe_size,
&rq_create->u.request.context,
LPFC_RQE_SIZE_8);
bf_set(lpfc_rq_context_page_size,
&rq_create->u.request.context,
LPFC_RQ_PAGE_SIZE_4096);
} else {
switch (hrq->entry_count) {
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2535 Unsupported RQ count. (%d)\n",
hrq->entry_count);
if (hrq->entry_count < 512) {
status = -EINVAL;
goto out;
}
/* otherwise default to smallest count (drop through) */
case 512:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_512);
break;
case 1024:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_1024);
break;
case 2048:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_2048);
break;
case 4096:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_4096);
break;
}
bf_set(lpfc_rq_context_buf_size, &rq_create->u.request.context,
LPFC_HDR_BUF_SIZE);
}
bf_set(lpfc_rq_context_cq_id, &rq_create->u.request.context,
cq->queue_id);
bf_set(lpfc_mbx_rq_create_num_pages, &rq_create->u.request,
hrq->page_count);
list_for_each_entry(dmabuf, &hrq->page_list, list) {
memset(dmabuf->virt, 0, hw_page_size);
rq_create->u.request.page[dmabuf->buffer_tag].addr_lo =
putPaddrLow(dmabuf->phys);
rq_create->u.request.page[dmabuf->buffer_tag].addr_hi =
putPaddrHigh(dmabuf->phys);
}
if (phba->sli4_hba.fw_func_mode & LPFC_DUA_MODE)
bf_set(lpfc_mbx_rq_create_dua, &rq_create->u.request, 1);
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2504 RQ_CREATE mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
goto out;
}
hrq->queue_id = bf_get(lpfc_mbx_rq_create_q_id, &rq_create->u.response);
if (hrq->queue_id == 0xFFFF) {
status = -ENXIO;
goto out;
}
if (phba->sli4_hba.fw_func_mode & LPFC_DUA_MODE) {
hrq->db_format = bf_get(lpfc_mbx_rq_create_db_format,
&rq_create->u.response);
if ((hrq->db_format != LPFC_DB_LIST_FORMAT) &&
(hrq->db_format != LPFC_DB_RING_FORMAT)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3262 RQ [%d] doorbell format not "
"supported: x%x\n", hrq->queue_id,
hrq->db_format);
status = -EINVAL;
goto out;
}
pci_barset = bf_get(lpfc_mbx_rq_create_bar_set,
&rq_create->u.response);
bar_memmap_p = lpfc_dual_chute_pci_bar_map(phba, pci_barset);
if (!bar_memmap_p) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3269 RQ[%d] failed to memmap pci "
"barset:x%x\n", hrq->queue_id,
pci_barset);
status = -ENOMEM;
goto out;
}
db_offset = rq_create->u.response.doorbell_offset;
if ((db_offset != LPFC_ULP0_RQ_DOORBELL) &&
(db_offset != LPFC_ULP1_RQ_DOORBELL)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3270 RQ[%d] doorbell offset not "
"supported: x%x\n", hrq->queue_id,
db_offset);
status = -EINVAL;
goto out;
}
hrq->db_regaddr = bar_memmap_p + db_offset;
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"3266 RQ[qid:%d]: barset:x%x, offset:x%x, "
"format:x%x\n", hrq->queue_id, pci_barset,
db_offset, hrq->db_format);
} else {
hrq->db_format = LPFC_DB_RING_FORMAT;
hrq->db_regaddr = phba->sli4_hba.RQDBregaddr;
}
hrq->type = LPFC_HRQ;
hrq->assoc_qid = cq->queue_id;
hrq->subtype = subtype;
hrq->host_index = 0;
hrq->hba_index = 0;
hrq->entry_repost = LPFC_RQ_REPOST;
/* now create the data queue */
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_RQ_CREATE,
length, LPFC_SLI4_MBX_EMBED);
bf_set(lpfc_mbox_hdr_version, &shdr->request,
phba->sli4_hba.pc_sli4_params.rqv);
if (phba->sli4_hba.pc_sli4_params.rqv == LPFC_Q_CREATE_VERSION_1) {
bf_set(lpfc_rq_context_rqe_count_1,
&rq_create->u.request.context, hrq->entry_count);
if (subtype == LPFC_NVMET)
rq_create->u.request.context.buffer_size =
LPFC_NVMET_DATA_BUF_SIZE;
else
rq_create->u.request.context.buffer_size =
LPFC_DATA_BUF_SIZE;
bf_set(lpfc_rq_context_rqe_size, &rq_create->u.request.context,
LPFC_RQE_SIZE_8);
bf_set(lpfc_rq_context_page_size, &rq_create->u.request.context,
(PAGE_SIZE/SLI4_PAGE_SIZE));
} else {
switch (drq->entry_count) {
default:
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2536 Unsupported RQ count. (%d)\n",
drq->entry_count);
if (drq->entry_count < 512) {
status = -EINVAL;
goto out;
}
/* otherwise default to smallest count (drop through) */
case 512:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_512);
break;
case 1024:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_1024);
break;
case 2048:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_2048);
break;
case 4096:
bf_set(lpfc_rq_context_rqe_count,
&rq_create->u.request.context,
LPFC_RQ_RING_SIZE_4096);
break;
}
if (subtype == LPFC_NVMET)
bf_set(lpfc_rq_context_buf_size,
&rq_create->u.request.context,
LPFC_NVMET_DATA_BUF_SIZE);
else
bf_set(lpfc_rq_context_buf_size,
&rq_create->u.request.context,
LPFC_DATA_BUF_SIZE);
}
bf_set(lpfc_rq_context_cq_id, &rq_create->u.request.context,
cq->queue_id);
bf_set(lpfc_mbx_rq_create_num_pages, &rq_create->u.request,
drq->page_count);
list_for_each_entry(dmabuf, &drq->page_list, list) {
rq_create->u.request.page[dmabuf->buffer_tag].addr_lo =
putPaddrLow(dmabuf->phys);
rq_create->u.request.page[dmabuf->buffer_tag].addr_hi =
putPaddrHigh(dmabuf->phys);
}
if (phba->sli4_hba.fw_func_mode & LPFC_DUA_MODE)
bf_set(lpfc_mbx_rq_create_dua, &rq_create->u.request, 1);
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *) &rq_create->header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
status = -ENXIO;
goto out;
}
drq->queue_id = bf_get(lpfc_mbx_rq_create_q_id, &rq_create->u.response);
if (drq->queue_id == 0xFFFF) {
status = -ENXIO;
goto out;
}
drq->type = LPFC_DRQ;
drq->assoc_qid = cq->queue_id;
drq->subtype = subtype;
drq->host_index = 0;
drq->hba_index = 0;
drq->entry_repost = LPFC_RQ_REPOST;
/* link the header and data RQs onto the parent cq child list */
list_add_tail(&hrq->list, &cq->child_list);
list_add_tail(&drq->list, &cq->child_list);
out:
mempool_free(mbox, phba->mbox_mem_pool);
return status;
}
/**
* lpfc_mrq_create - Create MRQ Receive Queues on the HBA
* @phba: HBA structure that indicates port to create a queue on.
* @hrqp: The queue structure array to use to create the header receive queues.
* @drqp: The queue structure array to use to create the data receive queues.
* @cqp: The completion queue array to bind these receive queues to.
*
* This function creates a receive buffer queue pair , as detailed in @hrq and
* @drq, on a port, described by @phba by sending a RQ_CREATE mailbox command
* to the HBA.
*
* The @phba struct is used to send mailbox command to HBA. The @drq and @hrq
* struct is used to get the entry count that is necessary to determine the
* number of pages to use for this queue. The @cq is used to indicate which
* completion queue to bind received buffers that are posted to these queues to.
* This function will send the RQ_CREATE mailbox command to the HBA to setup the
* receive queue pair. This function is asynchronous and will wait for the
* mailbox command to finish before continuing.
*
* On success this function will return a zero. If unable to allocate enough
* memory this function will return -ENOMEM. If the queue create mailbox command
* fails this function will return -ENXIO.
**/
int
lpfc_mrq_create(struct lpfc_hba *phba, struct lpfc_queue **hrqp,
struct lpfc_queue **drqp, struct lpfc_queue **cqp,
uint32_t subtype)
{
struct lpfc_queue *hrq, *drq, *cq;
struct lpfc_mbx_rq_create_v2 *rq_create;
struct lpfc_dmabuf *dmabuf;
LPFC_MBOXQ_t *mbox;
int rc, length, alloclen, status = 0;
int cnt, idx, numrq, page_idx = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
uint32_t hw_page_size = phba->sli4_hba.pc_sli4_params.if_page_sz;
numrq = phba->cfg_nvmet_mrq;
/* sanity check on array memory */
if (!hrqp || !drqp || !cqp || !numrq)
return -ENODEV;
if (!phba->sli4_hba.pc_sli4_params.supported)
hw_page_size = SLI4_PAGE_SIZE;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = sizeof(struct lpfc_mbx_rq_create_v2);
length += ((2 * numrq * hrqp[0]->page_count) *
sizeof(struct dma_address));
alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_RQ_CREATE, length,
LPFC_SLI4_MBX_NEMBED);
if (alloclen < length) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3099 Allocated DMA memory size (%d) is "
"less than the requested DMA memory size "
"(%d)\n", alloclen, length);
status = -ENOMEM;
goto out;
}
rq_create = mbox->sge_array->addr[0];
shdr = (union lpfc_sli4_cfg_shdr *)&rq_create->cfg_shdr;
bf_set(lpfc_mbox_hdr_version, &shdr->request, LPFC_Q_CREATE_VERSION_2);
cnt = 0;
for (idx = 0; idx < numrq; idx++) {
hrq = hrqp[idx];
drq = drqp[idx];
cq = cqp[idx];
/* sanity check on queue memory */
if (!hrq || !drq || !cq) {
status = -ENODEV;
goto out;
}
if (hrq->entry_count != drq->entry_count) {
status = -EINVAL;
goto out;
}
if (idx == 0) {
bf_set(lpfc_mbx_rq_create_num_pages,
&rq_create->u.request,
hrq->page_count);
bf_set(lpfc_mbx_rq_create_rq_cnt,
&rq_create->u.request, (numrq * 2));
bf_set(lpfc_mbx_rq_create_dnb, &rq_create->u.request,
1);
bf_set(lpfc_rq_context_base_cq,
&rq_create->u.request.context,
cq->queue_id);
bf_set(lpfc_rq_context_data_size,
&rq_create->u.request.context,
LPFC_NVMET_DATA_BUF_SIZE);
bf_set(lpfc_rq_context_hdr_size,
&rq_create->u.request.context,
LPFC_HDR_BUF_SIZE);
bf_set(lpfc_rq_context_rqe_count_1,
&rq_create->u.request.context,
hrq->entry_count);
bf_set(lpfc_rq_context_rqe_size,
&rq_create->u.request.context,
LPFC_RQE_SIZE_8);
bf_set(lpfc_rq_context_page_size,
&rq_create->u.request.context,
(PAGE_SIZE/SLI4_PAGE_SIZE));
}
rc = 0;
list_for_each_entry(dmabuf, &hrq->page_list, list) {
memset(dmabuf->virt, 0, hw_page_size);
cnt = page_idx + dmabuf->buffer_tag;
rq_create->u.request.page[cnt].addr_lo =
putPaddrLow(dmabuf->phys);
rq_create->u.request.page[cnt].addr_hi =
putPaddrHigh(dmabuf->phys);
rc++;
}
page_idx += rc;
rc = 0;
list_for_each_entry(dmabuf, &drq->page_list, list) {
memset(dmabuf->virt, 0, hw_page_size);
cnt = page_idx + dmabuf->buffer_tag;
rq_create->u.request.page[cnt].addr_lo =
putPaddrLow(dmabuf->phys);
rq_create->u.request.page[cnt].addr_hi =
putPaddrHigh(dmabuf->phys);
rc++;
}
page_idx += rc;
hrq->db_format = LPFC_DB_RING_FORMAT;
hrq->db_regaddr = phba->sli4_hba.RQDBregaddr;
hrq->type = LPFC_HRQ;
hrq->assoc_qid = cq->queue_id;
hrq->subtype = subtype;
hrq->host_index = 0;
hrq->hba_index = 0;
hrq->entry_repost = LPFC_RQ_REPOST;
drq->db_format = LPFC_DB_RING_FORMAT;
drq->db_regaddr = phba->sli4_hba.RQDBregaddr;
drq->type = LPFC_DRQ;
drq->assoc_qid = cq->queue_id;
drq->subtype = subtype;
drq->host_index = 0;
drq->hba_index = 0;
drq->entry_repost = LPFC_RQ_REPOST;
list_add_tail(&hrq->list, &cq->child_list);
list_add_tail(&drq->list, &cq->child_list);
}
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3120 RQ_CREATE mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
goto out;
}
rc = bf_get(lpfc_mbx_rq_create_q_id, &rq_create->u.response);
if (rc == 0xFFFF) {
status = -ENXIO;
goto out;
}
/* Initialize all RQs with associated queue id */
for (idx = 0; idx < numrq; idx++) {
hrq = hrqp[idx];
hrq->queue_id = rc + (2 * idx);
drq = drqp[idx];
drq->queue_id = rc + (2 * idx) + 1;
}
out:
lpfc_sli4_mbox_cmd_free(phba, mbox);
return status;
}
/**
* lpfc_eq_destroy - Destroy an event Queue on the HBA
* @eq: The queue structure associated with the queue to destroy.
*
* This function destroys a queue, as detailed in @eq by sending an mailbox
* command, specific to the type of queue, to the HBA.
*
* The @eq struct is used to get the queue ID of the queue to destroy.
*
* On success this function will return a zero. If the queue destroy mailbox
* command fails this function will return -ENXIO.
**/
int
lpfc_eq_destroy(struct lpfc_hba *phba, struct lpfc_queue *eq)
{
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
/* sanity check on queue memory */
if (!eq)
return -ENODEV;
mbox = mempool_alloc(eq->phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_eq_destroy) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_EQ_DESTROY,
length, LPFC_SLI4_MBX_EMBED);
bf_set(lpfc_mbx_eq_destroy_q_id, &mbox->u.mqe.un.eq_destroy.u.request,
eq->queue_id);
mbox->vport = eq->phba->pport;
mbox->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
rc = lpfc_sli_issue_mbox(eq->phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *)
&mbox->u.mqe.un.eq_destroy.header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2505 EQ_DESTROY mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
}
/* Remove eq from any list */
list_del_init(&eq->list);
mempool_free(mbox, eq->phba->mbox_mem_pool);
return status;
}
/**
* lpfc_cq_destroy - Destroy a Completion Queue on the HBA
* @cq: The queue structure associated with the queue to destroy.
*
* This function destroys a queue, as detailed in @cq by sending an mailbox
* command, specific to the type of queue, to the HBA.
*
* The @cq struct is used to get the queue ID of the queue to destroy.
*
* On success this function will return a zero. If the queue destroy mailbox
* command fails this function will return -ENXIO.
**/
int
lpfc_cq_destroy(struct lpfc_hba *phba, struct lpfc_queue *cq)
{
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
/* sanity check on queue memory */
if (!cq)
return -ENODEV;
mbox = mempool_alloc(cq->phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_cq_destroy) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_CQ_DESTROY,
length, LPFC_SLI4_MBX_EMBED);
bf_set(lpfc_mbx_cq_destroy_q_id, &mbox->u.mqe.un.cq_destroy.u.request,
cq->queue_id);
mbox->vport = cq->phba->pport;
mbox->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
rc = lpfc_sli_issue_mbox(cq->phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *)
&mbox->u.mqe.un.wq_create.header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2506 CQ_DESTROY mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
}
/* Remove cq from any list */
list_del_init(&cq->list);
mempool_free(mbox, cq->phba->mbox_mem_pool);
return status;
}
/**
* lpfc_mq_destroy - Destroy a Mailbox Queue on the HBA
* @qm: The queue structure associated with the queue to destroy.
*
* This function destroys a queue, as detailed in @mq by sending an mailbox
* command, specific to the type of queue, to the HBA.
*
* The @mq struct is used to get the queue ID of the queue to destroy.
*
* On success this function will return a zero. If the queue destroy mailbox
* command fails this function will return -ENXIO.
**/
int
lpfc_mq_destroy(struct lpfc_hba *phba, struct lpfc_queue *mq)
{
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
/* sanity check on queue memory */
if (!mq)
return -ENODEV;
mbox = mempool_alloc(mq->phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_mq_destroy) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_MQ_DESTROY,
length, LPFC_SLI4_MBX_EMBED);
bf_set(lpfc_mbx_mq_destroy_q_id, &mbox->u.mqe.un.mq_destroy.u.request,
mq->queue_id);
mbox->vport = mq->phba->pport;
mbox->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
rc = lpfc_sli_issue_mbox(mq->phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *)
&mbox->u.mqe.un.mq_destroy.header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2507 MQ_DESTROY mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
}
/* Remove mq from any list */
list_del_init(&mq->list);
mempool_free(mbox, mq->phba->mbox_mem_pool);
return status;
}
/**
* lpfc_wq_destroy - Destroy a Work Queue on the HBA
* @wq: The queue structure associated with the queue to destroy.
*
* This function destroys a queue, as detailed in @wq by sending an mailbox
* command, specific to the type of queue, to the HBA.
*
* The @wq struct is used to get the queue ID of the queue to destroy.
*
* On success this function will return a zero. If the queue destroy mailbox
* command fails this function will return -ENXIO.
**/
int
lpfc_wq_destroy(struct lpfc_hba *phba, struct lpfc_queue *wq)
{
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
/* sanity check on queue memory */
if (!wq)
return -ENODEV;
mbox = mempool_alloc(wq->phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_wq_destroy) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_WQ_DESTROY,
length, LPFC_SLI4_MBX_EMBED);
bf_set(lpfc_mbx_wq_destroy_q_id, &mbox->u.mqe.un.wq_destroy.u.request,
wq->queue_id);
mbox->vport = wq->phba->pport;
mbox->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
rc = lpfc_sli_issue_mbox(wq->phba, mbox, MBX_POLL);
shdr = (union lpfc_sli4_cfg_shdr *)
&mbox->u.mqe.un.wq_destroy.header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2508 WQ_DESTROY mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
}
/* Remove wq from any list */
list_del_init(&wq->list);
kfree(wq->pring);
wq->pring = NULL;
mempool_free(mbox, wq->phba->mbox_mem_pool);
return status;
}
/**
* lpfc_rq_destroy - Destroy a Receive Queue on the HBA
* @rq: The queue structure associated with the queue to destroy.
*
* This function destroys a queue, as detailed in @rq by sending an mailbox
* command, specific to the type of queue, to the HBA.
*
* The @rq struct is used to get the queue ID of the queue to destroy.
*
* On success this function will return a zero. If the queue destroy mailbox
* command fails this function will return -ENXIO.
**/
int
lpfc_rq_destroy(struct lpfc_hba *phba, struct lpfc_queue *hrq,
struct lpfc_queue *drq)
{
LPFC_MBOXQ_t *mbox;
int rc, length, status = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
/* sanity check on queue memory */
if (!hrq || !drq)
return -ENODEV;
mbox = mempool_alloc(hrq->phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
length = (sizeof(struct lpfc_mbx_rq_destroy) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_RQ_DESTROY,
length, LPFC_SLI4_MBX_EMBED);
bf_set(lpfc_mbx_rq_destroy_q_id, &mbox->u.mqe.un.rq_destroy.u.request,
hrq->queue_id);
mbox->vport = hrq->phba->pport;
mbox->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
rc = lpfc_sli_issue_mbox(hrq->phba, mbox, MBX_POLL);
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *)
&mbox->u.mqe.un.rq_destroy.header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2509 RQ_DESTROY mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
if (rc != MBX_TIMEOUT)
mempool_free(mbox, hrq->phba->mbox_mem_pool);
return -ENXIO;
}
bf_set(lpfc_mbx_rq_destroy_q_id, &mbox->u.mqe.un.rq_destroy.u.request,
drq->queue_id);
rc = lpfc_sli_issue_mbox(drq->phba, mbox, MBX_POLL);
shdr = (union lpfc_sli4_cfg_shdr *)
&mbox->u.mqe.un.rq_destroy.header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2510 RQ_DESTROY mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
status = -ENXIO;
}
list_del_init(&hrq->list);
list_del_init(&drq->list);
mempool_free(mbox, hrq->phba->mbox_mem_pool);
return status;
}
/**
* lpfc_sli4_post_sgl - Post scatter gather list for an XRI to HBA
* @phba: The virtual port for which this call being executed.
* @pdma_phys_addr0: Physical address of the 1st SGL page.
* @pdma_phys_addr1: Physical address of the 2nd SGL page.
* @xritag: the xritag that ties this io to the SGL pages.
*
* This routine will post the sgl pages for the IO that has the xritag
* that is in the iocbq structure. The xritag is assigned during iocbq
* creation and persists for as long as the driver is loaded.
* if the caller has fewer than 256 scatter gather segments to map then
* pdma_phys_addr1 should be 0.
* If the caller needs to map more than 256 scatter gather segment then
* pdma_phys_addr1 should be a valid physical address.
* physical address for SGLs must be 64 byte aligned.
* If you are going to map 2 SGL's then the first one must have 256 entries
* the second sgl can have between 1 and 256 entries.
*
* Return codes:
* 0 - Success
* -ENXIO, -ENOMEM - Failure
**/
int
lpfc_sli4_post_sgl(struct lpfc_hba *phba,
dma_addr_t pdma_phys_addr0,
dma_addr_t pdma_phys_addr1,
uint16_t xritag)
{
struct lpfc_mbx_post_sgl_pages *post_sgl_pages;
LPFC_MBOXQ_t *mbox;
int rc;
uint32_t shdr_status, shdr_add_status;
uint32_t mbox_tmo;
union lpfc_sli4_cfg_shdr *shdr;
if (xritag == NO_XRI) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"0364 Invalid param:\n");
return -EINVAL;
}
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES,
sizeof(struct lpfc_mbx_post_sgl_pages) -
sizeof(struct lpfc_sli4_cfg_mhdr), LPFC_SLI4_MBX_EMBED);
post_sgl_pages = (struct lpfc_mbx_post_sgl_pages *)
&mbox->u.mqe.un.post_sgl_pages;
bf_set(lpfc_post_sgl_pages_xri, post_sgl_pages, xritag);
bf_set(lpfc_post_sgl_pages_xricnt, post_sgl_pages, 1);
post_sgl_pages->sgl_pg_pairs[0].sgl_pg0_addr_lo =
cpu_to_le32(putPaddrLow(pdma_phys_addr0));
post_sgl_pages->sgl_pg_pairs[0].sgl_pg0_addr_hi =
cpu_to_le32(putPaddrHigh(pdma_phys_addr0));
post_sgl_pages->sgl_pg_pairs[0].sgl_pg1_addr_lo =
cpu_to_le32(putPaddrLow(pdma_phys_addr1));
post_sgl_pages->sgl_pg_pairs[0].sgl_pg1_addr_hi =
cpu_to_le32(putPaddrHigh(pdma_phys_addr1));
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *) &post_sgl_pages->header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (rc != MBX_TIMEOUT)
mempool_free(mbox, phba->mbox_mem_pool);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2511 POST_SGL mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
}
return 0;
}
/**
* lpfc_sli4_alloc_xri - Get an available rpi in the device's range
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to post rpi header templates to the
* HBA consistent with the SLI-4 interface spec. This routine
* posts a SLI4_PAGE_SIZE memory region to the port to hold up to
* SLI4_PAGE_SIZE modulo 64 rpi context headers.
*
* Returns
* A nonzero rpi defined as rpi_base <= rpi < max_rpi if successful
* LPFC_RPI_ALLOC_ERROR if no rpis are available.
**/
static uint16_t
lpfc_sli4_alloc_xri(struct lpfc_hba *phba)
{
unsigned long xri;
/*
* Fetch the next logical xri. Because this index is logical,
* the driver starts at 0 each time.
*/
spin_lock_irq(&phba->hbalock);
xri = find_next_zero_bit(phba->sli4_hba.xri_bmask,
phba->sli4_hba.max_cfg_param.max_xri, 0);
if (xri >= phba->sli4_hba.max_cfg_param.max_xri) {
spin_unlock_irq(&phba->hbalock);
return NO_XRI;
} else {
set_bit(xri, phba->sli4_hba.xri_bmask);
phba->sli4_hba.max_cfg_param.xri_used++;
}
spin_unlock_irq(&phba->hbalock);
return xri;
}
/**
* lpfc_sli4_free_xri - Release an xri for reuse.
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to release an xri to the pool of
* available rpis maintained by the driver.
**/
static void
__lpfc_sli4_free_xri(struct lpfc_hba *phba, int xri)
{
if (test_and_clear_bit(xri, phba->sli4_hba.xri_bmask)) {
phba->sli4_hba.max_cfg_param.xri_used--;
}
}
/**
* lpfc_sli4_free_xri - Release an xri for reuse.
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to release an xri to the pool of
* available rpis maintained by the driver.
**/
void
lpfc_sli4_free_xri(struct lpfc_hba *phba, int xri)
{
spin_lock_irq(&phba->hbalock);
__lpfc_sli4_free_xri(phba, xri);
spin_unlock_irq(&phba->hbalock);
}
/**
* lpfc_sli4_next_xritag - Get an xritag for the io
* @phba: Pointer to HBA context object.
*
* This function gets an xritag for the iocb. If there is no unused xritag
* it will return 0xffff.
* The function returns the allocated xritag if successful, else returns zero.
* Zero is not a valid xritag.
* The caller is not required to hold any lock.
**/
uint16_t
lpfc_sli4_next_xritag(struct lpfc_hba *phba)
{
uint16_t xri_index;
xri_index = lpfc_sli4_alloc_xri(phba);
if (xri_index == NO_XRI)
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"2004 Failed to allocate XRI.last XRITAG is %d"
" Max XRI is %d, Used XRI is %d\n",
xri_index,
phba->sli4_hba.max_cfg_param.max_xri,
phba->sli4_hba.max_cfg_param.xri_used);
return xri_index;
}
/**
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* lpfc_sli4_post_sgl_list - post a block of ELS sgls to the port.
* @phba: pointer to lpfc hba data structure.
* @post_sgl_list: pointer to els sgl entry list.
* @count: number of els sgl entries on the list.
*
* This routine is invoked to post a block of driver's sgl pages to the
* HBA using non-embedded mailbox command. No Lock is held. This routine
* is only called when the driver is loading and after all IO has been
* stopped.
**/
static int
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli4_post_sgl_list(struct lpfc_hba *phba,
struct list_head *post_sgl_list,
int post_cnt)
{
struct lpfc_sglq *sglq_entry = NULL, *sglq_next = NULL;
struct lpfc_mbx_post_uembed_sgl_page1 *sgl;
struct sgl_page_pairs *sgl_pg_pairs;
void *viraddr;
LPFC_MBOXQ_t *mbox;
uint32_t reqlen, alloclen, pg_pairs;
uint32_t mbox_tmo;
uint16_t xritag_start = 0;
int rc = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
reqlen = post_cnt * sizeof(struct sgl_page_pairs) +
sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t);
if (reqlen > SLI4_PAGE_SIZE) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2559 Block sgl registration required DMA "
"size (%d) great than a page\n", reqlen);
return -ENOMEM;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
/* Allocate DMA memory and set up the non-embedded mailbox command */
alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen,
LPFC_SLI4_MBX_NEMBED);
if (alloclen < reqlen) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0285 Allocated DMA memory size (%d) is "
"less than the requested DMA memory "
"size (%d)\n", alloclen, reqlen);
lpfc_sli4_mbox_cmd_free(phba, mbox);
return -ENOMEM;
}
/* Set up the SGL pages in the non-embedded DMA pages */
viraddr = mbox->sge_array->addr[0];
sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr;
sgl_pg_pairs = &sgl->sgl_pg_pairs;
pg_pairs = 0;
list_for_each_entry_safe(sglq_entry, sglq_next, post_sgl_list, list) {
/* Set up the sge entry */
sgl_pg_pairs->sgl_pg0_addr_lo =
cpu_to_le32(putPaddrLow(sglq_entry->phys));
sgl_pg_pairs->sgl_pg0_addr_hi =
cpu_to_le32(putPaddrHigh(sglq_entry->phys));
sgl_pg_pairs->sgl_pg1_addr_lo =
cpu_to_le32(putPaddrLow(0));
sgl_pg_pairs->sgl_pg1_addr_hi =
cpu_to_le32(putPaddrHigh(0));
/* Keep the first xritag on the list */
if (pg_pairs == 0)
xritag_start = sglq_entry->sli4_xritag;
sgl_pg_pairs++;
pg_pairs++;
}
/* Complete initialization and perform endian conversion. */
bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
bf_set(lpfc_post_sgl_pages_xricnt, sgl, post_cnt);
sgl->word0 = cpu_to_le32(sgl->word0);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
shdr = (union lpfc_sli4_cfg_shdr *) &sgl->cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (rc != MBX_TIMEOUT)
lpfc_sli4_mbox_cmd_free(phba, mbox);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2513 POST_SGL_BLOCK mailbox command failed "
"status x%x add_status x%x mbx status x%x\n",
shdr_status, shdr_add_status, rc);
rc = -ENXIO;
}
return rc;
}
/**
* lpfc_sli4_post_scsi_sgl_block - post a block of scsi sgl list to firmware
* @phba: pointer to lpfc hba data structure.
* @sblist: pointer to scsi buffer list.
* @count: number of scsi buffers on the list.
*
* This routine is invoked to post a block of @count scsi sgl pages from a
* SCSI buffer list @sblist to the HBA using non-embedded mailbox command.
* No Lock is held.
*
**/
int
lpfc_sli4_post_scsi_sgl_block(struct lpfc_hba *phba,
struct list_head *sblist,
int count)
{
struct lpfc_scsi_buf *psb;
struct lpfc_mbx_post_uembed_sgl_page1 *sgl;
struct sgl_page_pairs *sgl_pg_pairs;
void *viraddr;
LPFC_MBOXQ_t *mbox;
uint32_t reqlen, alloclen, pg_pairs;
uint32_t mbox_tmo;
uint16_t xritag_start = 0;
int rc = 0;
uint32_t shdr_status, shdr_add_status;
dma_addr_t pdma_phys_bpl1;
union lpfc_sli4_cfg_shdr *shdr;
/* Calculate the requested length of the dma memory */
reqlen = count * sizeof(struct sgl_page_pairs) +
sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t);
if (reqlen > SLI4_PAGE_SIZE) {
lpfc_printf_log(phba, KERN_WARNING, LOG_INIT,
"0217 Block sgl registration required DMA "
"size (%d) great than a page\n", reqlen);
return -ENOMEM;
}
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"0283 Failed to allocate mbox cmd memory\n");
return -ENOMEM;
}
/* Allocate DMA memory and set up the non-embedded mailbox command */
alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen,
LPFC_SLI4_MBX_NEMBED);
if (alloclen < reqlen) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2561 Allocated DMA memory size (%d) is "
"less than the requested DMA memory "
"size (%d)\n", alloclen, reqlen);
lpfc_sli4_mbox_cmd_free(phba, mbox);
return -ENOMEM;
}
/* Get the first SGE entry from the non-embedded DMA memory */
viraddr = mbox->sge_array->addr[0];
/* Set up the SGL pages in the non-embedded DMA pages */
sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr;
sgl_pg_pairs = &sgl->sgl_pg_pairs;
pg_pairs = 0;
list_for_each_entry(psb, sblist, list) {
/* Set up the sge entry */
sgl_pg_pairs->sgl_pg0_addr_lo =
cpu_to_le32(putPaddrLow(psb->dma_phys_bpl));
sgl_pg_pairs->sgl_pg0_addr_hi =
cpu_to_le32(putPaddrHigh(psb->dma_phys_bpl));
if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE)
pdma_phys_bpl1 = psb->dma_phys_bpl + SGL_PAGE_SIZE;
else
pdma_phys_bpl1 = 0;
sgl_pg_pairs->sgl_pg1_addr_lo =
cpu_to_le32(putPaddrLow(pdma_phys_bpl1));
sgl_pg_pairs->sgl_pg1_addr_hi =
cpu_to_le32(putPaddrHigh(pdma_phys_bpl1));
/* Keep the first xritag on the list */
if (pg_pairs == 0)
xritag_start = psb->cur_iocbq.sli4_xritag;
sgl_pg_pairs++;
pg_pairs++;
}
bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start);
bf_set(lpfc_post_sgl_pages_xricnt, sgl, pg_pairs);
/* Perform endian conversion if necessary */
sgl->word0 = cpu_to_le32(sgl->word0);
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
shdr = (union lpfc_sli4_cfg_shdr *) &sgl->cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (rc != MBX_TIMEOUT)
lpfc_sli4_mbox_cmd_free(phba, mbox);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2564 POST_SGL_BLOCK mailbox command failed "
"status x%x add_status x%x mbx status x%x\n",
shdr_status, shdr_add_status, rc);
rc = -ENXIO;
}
return rc;
}
/**
* lpfc_fc_frame_check - Check that this frame is a valid frame to handle
* @phba: pointer to lpfc_hba struct that the frame was received on
* @fc_hdr: A pointer to the FC Header data (In Big Endian Format)
*
* This function checks the fields in the @fc_hdr to see if the FC frame is a
* valid type of frame that the LPFC driver will handle. This function will
* return a zero if the frame is a valid frame or a non zero value when the
* frame does not pass the check.
**/
static int
lpfc_fc_frame_check(struct lpfc_hba *phba, struct fc_frame_header *fc_hdr)
{
/* make rctl_names static to save stack space */
struct fc_vft_header *fc_vft_hdr;
uint32_t *header = (uint32_t *) fc_hdr;
#define FC_RCTL_MDS_DIAGS 0xF4
switch (fc_hdr->fh_r_ctl) {
case FC_RCTL_DD_UNCAT: /* uncategorized information */
case FC_RCTL_DD_SOL_DATA: /* solicited data */
case FC_RCTL_DD_UNSOL_CTL: /* unsolicited control */
case FC_RCTL_DD_SOL_CTL: /* solicited control or reply */
case FC_RCTL_DD_UNSOL_DATA: /* unsolicited data */
case FC_RCTL_DD_DATA_DESC: /* data descriptor */
case FC_RCTL_DD_UNSOL_CMD: /* unsolicited command */
case FC_RCTL_DD_CMD_STATUS: /* command status */
case FC_RCTL_ELS_REQ: /* extended link services request */
case FC_RCTL_ELS_REP: /* extended link services reply */
case FC_RCTL_ELS4_REQ: /* FC-4 ELS request */
case FC_RCTL_ELS4_REP: /* FC-4 ELS reply */
case FC_RCTL_BA_NOP: /* basic link service NOP */
case FC_RCTL_BA_ABTS: /* basic link service abort */
case FC_RCTL_BA_RMC: /* remove connection */
case FC_RCTL_BA_ACC: /* basic accept */
case FC_RCTL_BA_RJT: /* basic reject */
case FC_RCTL_BA_PRMT:
case FC_RCTL_ACK_1: /* acknowledge_1 */
case FC_RCTL_ACK_0: /* acknowledge_0 */
case FC_RCTL_P_RJT: /* port reject */
case FC_RCTL_F_RJT: /* fabric reject */
case FC_RCTL_P_BSY: /* port busy */
case FC_RCTL_F_BSY: /* fabric busy to data frame */
case FC_RCTL_F_BSYL: /* fabric busy to link control frame */
case FC_RCTL_LCR: /* link credit reset */
case FC_RCTL_MDS_DIAGS: /* MDS Diagnostics */
case FC_RCTL_END: /* end */
break;
case FC_RCTL_VFTH: /* Virtual Fabric tagging Header */
fc_vft_hdr = (struct fc_vft_header *)fc_hdr;
fc_hdr = &((struct fc_frame_header *)fc_vft_hdr)[1];
return lpfc_fc_frame_check(phba, fc_hdr);
default:
goto drop;
}
#define FC_TYPE_VENDOR_UNIQUE 0xFF
switch (fc_hdr->fh_type) {
case FC_TYPE_BLS:
case FC_TYPE_ELS:
case FC_TYPE_FCP:
case FC_TYPE_CT:
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
case FC_TYPE_NVME:
case FC_TYPE_VENDOR_UNIQUE:
break;
case FC_TYPE_IP:
case FC_TYPE_ILS:
default:
goto drop;
}
lpfc_printf_log(phba, KERN_INFO, LOG_ELS,
"2538 Received frame rctl:x%x, type:x%x, "
"frame Data:%08x %08x %08x %08x %08x %08x %08x\n",
fc_hdr->fh_r_ctl, fc_hdr->fh_type,
be32_to_cpu(header[0]), be32_to_cpu(header[1]),
be32_to_cpu(header[2]), be32_to_cpu(header[3]),
be32_to_cpu(header[4]), be32_to_cpu(header[5]),
be32_to_cpu(header[6]));
return 0;
drop:
lpfc_printf_log(phba, KERN_WARNING, LOG_ELS,
"2539 Dropped frame rctl:x%x type:x%x\n",
fc_hdr->fh_r_ctl, fc_hdr->fh_type);
return 1;
}
/**
* lpfc_fc_hdr_get_vfi - Get the VFI from an FC frame
* @fc_hdr: A pointer to the FC Header data (In Big Endian Format)
*
* This function processes the FC header to retrieve the VFI from the VF
* header, if one exists. This function will return the VFI if one exists
* or 0 if no VSAN Header exists.
**/
static uint32_t
lpfc_fc_hdr_get_vfi(struct fc_frame_header *fc_hdr)
{
struct fc_vft_header *fc_vft_hdr = (struct fc_vft_header *)fc_hdr;
if (fc_hdr->fh_r_ctl != FC_RCTL_VFTH)
return 0;
return bf_get(fc_vft_hdr_vf_id, fc_vft_hdr);
}
/**
* lpfc_fc_frame_to_vport - Finds the vport that a frame is destined to
* @phba: Pointer to the HBA structure to search for the vport on
* @fc_hdr: A pointer to the FC Header data (In Big Endian Format)
* @fcfi: The FC Fabric ID that the frame came from
*
* This function searches the @phba for a vport that matches the content of the
* @fc_hdr passed in and the @fcfi. This function uses the @fc_hdr to fetch the
* VFI, if the Virtual Fabric Tagging Header exists, and the DID. This function
* returns the matching vport pointer or NULL if unable to match frame to a
* vport.
**/
static struct lpfc_vport *
lpfc_fc_frame_to_vport(struct lpfc_hba *phba, struct fc_frame_header *fc_hdr,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
uint16_t fcfi, uint32_t did)
{
struct lpfc_vport **vports;
struct lpfc_vport *vport = NULL;
int i;
if (did == Fabric_DID)
return phba->pport;
if ((phba->pport->fc_flag & FC_PT2PT) &&
!(phba->link_state == LPFC_HBA_READY))
return phba->pport;
vports = lpfc_create_vport_work_array(phba);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
if (vports != NULL) {
for (i = 0; i <= phba->max_vpi && vports[i] != NULL; i++) {
if (phba->fcf.fcfi == fcfi &&
vports[i]->vfi == lpfc_fc_hdr_get_vfi(fc_hdr) &&
vports[i]->fc_myDID == did) {
vport = vports[i];
break;
}
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
}
lpfc_destroy_vport_work_array(phba, vports);
return vport;
}
/**
* lpfc_update_rcv_time_stamp - Update vport's rcv seq time stamp
* @vport: The vport to work on.
*
* This function updates the receive sequence time stamp for this vport. The
* receive sequence time stamp indicates the time that the last frame of the
* the sequence that has been idle for the longest amount of time was received.
* the driver uses this time stamp to indicate if any received sequences have
* timed out.
**/
static void
lpfc_update_rcv_time_stamp(struct lpfc_vport *vport)
{
struct lpfc_dmabuf *h_buf;
struct hbq_dmabuf *dmabuf = NULL;
/* get the oldest sequence on the rcv list */
h_buf = list_get_first(&vport->rcv_buffer_list,
struct lpfc_dmabuf, list);
if (!h_buf)
return;
dmabuf = container_of(h_buf, struct hbq_dmabuf, hbuf);
vport->rcv_buffer_time_stamp = dmabuf->time_stamp;
}
/**
* lpfc_cleanup_rcv_buffers - Cleans up all outstanding receive sequences.
* @vport: The vport that the received sequences were sent to.
*
* This function cleans up all outstanding received sequences. This is called
* by the driver when a link event or user action invalidates all the received
* sequences.
**/
void
lpfc_cleanup_rcv_buffers(struct lpfc_vport *vport)
{
struct lpfc_dmabuf *h_buf, *hnext;
struct lpfc_dmabuf *d_buf, *dnext;
struct hbq_dmabuf *dmabuf = NULL;
/* start with the oldest sequence on the rcv list */
list_for_each_entry_safe(h_buf, hnext, &vport->rcv_buffer_list, list) {
dmabuf = container_of(h_buf, struct hbq_dmabuf, hbuf);
list_del_init(&dmabuf->hbuf.list);
list_for_each_entry_safe(d_buf, dnext,
&dmabuf->dbuf.list, list) {
list_del_init(&d_buf->list);
lpfc_in_buf_free(vport->phba, d_buf);
}
lpfc_in_buf_free(vport->phba, &dmabuf->dbuf);
}
}
/**
* lpfc_rcv_seq_check_edtov - Cleans up timed out receive sequences.
* @vport: The vport that the received sequences were sent to.
*
* This function determines whether any received sequences have timed out by
* first checking the vport's rcv_buffer_time_stamp. If this time_stamp
* indicates that there is at least one timed out sequence this routine will
* go through the received sequences one at a time from most inactive to most
* active to determine which ones need to be cleaned up. Once it has determined
* that a sequence needs to be cleaned up it will simply free up the resources
* without sending an abort.
**/
void
lpfc_rcv_seq_check_edtov(struct lpfc_vport *vport)
{
struct lpfc_dmabuf *h_buf, *hnext;
struct lpfc_dmabuf *d_buf, *dnext;
struct hbq_dmabuf *dmabuf = NULL;
unsigned long timeout;
int abort_count = 0;
timeout = (msecs_to_jiffies(vport->phba->fc_edtov) +
vport->rcv_buffer_time_stamp);
if (list_empty(&vport->rcv_buffer_list) ||
time_before(jiffies, timeout))
return;
/* start with the oldest sequence on the rcv list */
list_for_each_entry_safe(h_buf, hnext, &vport->rcv_buffer_list, list) {
dmabuf = container_of(h_buf, struct hbq_dmabuf, hbuf);
timeout = (msecs_to_jiffies(vport->phba->fc_edtov) +
dmabuf->time_stamp);
if (time_before(jiffies, timeout))
break;
abort_count++;
list_del_init(&dmabuf->hbuf.list);
list_for_each_entry_safe(d_buf, dnext,
&dmabuf->dbuf.list, list) {
list_del_init(&d_buf->list);
lpfc_in_buf_free(vport->phba, d_buf);
}
lpfc_in_buf_free(vport->phba, &dmabuf->dbuf);
}
if (abort_count)
lpfc_update_rcv_time_stamp(vport);
}
/**
* lpfc_fc_frame_add - Adds a frame to the vport's list of received sequences
* @dmabuf: pointer to a dmabuf that describes the hdr and data of the FC frame
*
* This function searches through the existing incomplete sequences that have
* been sent to this @vport. If the frame matches one of the incomplete
* sequences then the dbuf in the @dmabuf is added to the list of frames that
* make up that sequence. If no sequence is found that matches this frame then
* the function will add the hbuf in the @dmabuf to the @vport's rcv_buffer_list
* This function returns a pointer to the first dmabuf in the sequence list that
* the frame was linked to.
**/
static struct hbq_dmabuf *
lpfc_fc_frame_add(struct lpfc_vport *vport, struct hbq_dmabuf *dmabuf)
{
struct fc_frame_header *new_hdr;
struct fc_frame_header *temp_hdr;
struct lpfc_dmabuf *d_buf;
struct lpfc_dmabuf *h_buf;
struct hbq_dmabuf *seq_dmabuf = NULL;
struct hbq_dmabuf *temp_dmabuf = NULL;
uint8_t found = 0;
INIT_LIST_HEAD(&dmabuf->dbuf.list);
dmabuf->time_stamp = jiffies;
new_hdr = (struct fc_frame_header *)dmabuf->hbuf.virt;
/* Use the hdr_buf to find the sequence that this frame belongs to */
list_for_each_entry(h_buf, &vport->rcv_buffer_list, list) {
temp_hdr = (struct fc_frame_header *)h_buf->virt;
if ((temp_hdr->fh_seq_id != new_hdr->fh_seq_id) ||
(temp_hdr->fh_ox_id != new_hdr->fh_ox_id) ||
(memcmp(&temp_hdr->fh_s_id, &new_hdr->fh_s_id, 3)))
continue;
/* found a pending sequence that matches this frame */
seq_dmabuf = container_of(h_buf, struct hbq_dmabuf, hbuf);
break;
}
if (!seq_dmabuf) {
/*
* This indicates first frame received for this sequence.
* Queue the buffer on the vport's rcv_buffer_list.
*/
list_add_tail(&dmabuf->hbuf.list, &vport->rcv_buffer_list);
lpfc_update_rcv_time_stamp(vport);
return dmabuf;
}
temp_hdr = seq_dmabuf->hbuf.virt;
if (be16_to_cpu(new_hdr->fh_seq_cnt) <
be16_to_cpu(temp_hdr->fh_seq_cnt)) {
list_del_init(&seq_dmabuf->hbuf.list);
list_add_tail(&dmabuf->hbuf.list, &vport->rcv_buffer_list);
list_add_tail(&dmabuf->dbuf.list, &seq_dmabuf->dbuf.list);
lpfc_update_rcv_time_stamp(vport);
return dmabuf;
}
/* move this sequence to the tail to indicate a young sequence */
list_move_tail(&seq_dmabuf->hbuf.list, &vport->rcv_buffer_list);
seq_dmabuf->time_stamp = jiffies;
lpfc_update_rcv_time_stamp(vport);
if (list_empty(&seq_dmabuf->dbuf.list)) {
temp_hdr = dmabuf->hbuf.virt;
list_add_tail(&dmabuf->dbuf.list, &seq_dmabuf->dbuf.list);
return seq_dmabuf;
}
/* find the correct place in the sequence to insert this frame */
d_buf = list_entry(seq_dmabuf->dbuf.list.prev, typeof(*d_buf), list);
while (!found) {
temp_dmabuf = container_of(d_buf, struct hbq_dmabuf, dbuf);
temp_hdr = (struct fc_frame_header *)temp_dmabuf->hbuf.virt;
/*
* If the frame's sequence count is greater than the frame on
* the list then insert the frame right after this frame
*/
if (be16_to_cpu(new_hdr->fh_seq_cnt) >
be16_to_cpu(temp_hdr->fh_seq_cnt)) {
list_add(&dmabuf->dbuf.list, &temp_dmabuf->dbuf.list);
found = 1;
break;
}
if (&d_buf->list == &seq_dmabuf->dbuf.list)
break;
d_buf = list_entry(d_buf->list.prev, typeof(*d_buf), list);
}
if (found)
return seq_dmabuf;
return NULL;
}
/**
* lpfc_sli4_abort_partial_seq - Abort partially assembled unsol sequence
* @vport: pointer to a vitural port
* @dmabuf: pointer to a dmabuf that describes the FC sequence
*
* This function tries to abort from the partially assembed sequence, described
* by the information from basic abbort @dmabuf. It checks to see whether such
* partially assembled sequence held by the driver. If so, it shall free up all
* the frames from the partially assembled sequence.
*
* Return
* true -- if there is matching partially assembled sequence present and all
* the frames freed with the sequence;
* false -- if there is no matching partially assembled sequence present so
* nothing got aborted in the lower layer driver
**/
static bool
lpfc_sli4_abort_partial_seq(struct lpfc_vport *vport,
struct hbq_dmabuf *dmabuf)
{
struct fc_frame_header *new_hdr;
struct fc_frame_header *temp_hdr;
struct lpfc_dmabuf *d_buf, *n_buf, *h_buf;
struct hbq_dmabuf *seq_dmabuf = NULL;
/* Use the hdr_buf to find the sequence that matches this frame */
INIT_LIST_HEAD(&dmabuf->dbuf.list);
INIT_LIST_HEAD(&dmabuf->hbuf.list);
new_hdr = (struct fc_frame_header *)dmabuf->hbuf.virt;
list_for_each_entry(h_buf, &vport->rcv_buffer_list, list) {
temp_hdr = (struct fc_frame_header *)h_buf->virt;
if ((temp_hdr->fh_seq_id != new_hdr->fh_seq_id) ||
(temp_hdr->fh_ox_id != new_hdr->fh_ox_id) ||
(memcmp(&temp_hdr->fh_s_id, &new_hdr->fh_s_id, 3)))
continue;
/* found a pending sequence that matches this frame */
seq_dmabuf = container_of(h_buf, struct hbq_dmabuf, hbuf);
break;
}
/* Free up all the frames from the partially assembled sequence */
if (seq_dmabuf) {
list_for_each_entry_safe(d_buf, n_buf,
&seq_dmabuf->dbuf.list, list) {
list_del_init(&d_buf->list);
lpfc_in_buf_free(vport->phba, d_buf);
}
return true;
}
return false;
}
/**
* lpfc_sli4_abort_ulp_seq - Abort assembled unsol sequence from ulp
* @vport: pointer to a vitural port
* @dmabuf: pointer to a dmabuf that describes the FC sequence
*
* This function tries to abort from the assembed sequence from upper level
* protocol, described by the information from basic abbort @dmabuf. It
* checks to see whether such pending context exists at upper level protocol.
* If so, it shall clean up the pending context.
*
* Return
* true -- if there is matching pending context of the sequence cleaned
* at ulp;
* false -- if there is no matching pending context of the sequence present
* at ulp.
**/
static bool
lpfc_sli4_abort_ulp_seq(struct lpfc_vport *vport, struct hbq_dmabuf *dmabuf)
{
struct lpfc_hba *phba = vport->phba;
int handled;
/* Accepting abort at ulp with SLI4 only */
if (phba->sli_rev < LPFC_SLI_REV4)
return false;
/* Register all caring upper level protocols to attend abort */
handled = lpfc_ct_handle_unsol_abort(phba, dmabuf);
if (handled)
return true;
return false;
}
/**
* lpfc_sli4_seq_abort_rsp_cmpl - BLS ABORT RSP seq abort iocb complete handler
* @phba: Pointer to HBA context object.
* @cmd_iocbq: pointer to the command iocbq structure.
* @rsp_iocbq: pointer to the response iocbq structure.
*
* This function handles the sequence abort response iocb command complete
* event. It properly releases the memory allocated to the sequence abort
* accept iocb.
**/
static void
lpfc_sli4_seq_abort_rsp_cmpl(struct lpfc_hba *phba,
struct lpfc_iocbq *cmd_iocbq,
struct lpfc_iocbq *rsp_iocbq)
{
struct lpfc_nodelist *ndlp;
if (cmd_iocbq) {
ndlp = (struct lpfc_nodelist *)cmd_iocbq->context1;
lpfc_nlp_put(ndlp);
lpfc_nlp_not_used(ndlp);
lpfc_sli_release_iocbq(phba, cmd_iocbq);
}
/* Failure means BLS ABORT RSP did not get delivered to remote node*/
if (rsp_iocbq && rsp_iocbq->iocb.ulpStatus)
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"3154 BLS ABORT RSP failed, data: x%x/x%x\n",
rsp_iocbq->iocb.ulpStatus,
rsp_iocbq->iocb.un.ulpWord[4]);
}
/**
* lpfc_sli4_xri_inrange - check xri is in range of xris owned by driver.
* @phba: Pointer to HBA context object.
* @xri: xri id in transaction.
*
* This function validates the xri maps to the known range of XRIs allocated an
* used by the driver.
**/
uint16_t
lpfc_sli4_xri_inrange(struct lpfc_hba *phba,
uint16_t xri)
{
uint16_t i;
for (i = 0; i < phba->sli4_hba.max_cfg_param.max_xri; i++) {
if (xri == phba->sli4_hba.xri_ids[i])
return i;
}
return NO_XRI;
}
/**
* lpfc_sli4_seq_abort_rsp - bls rsp to sequence abort
* @phba: Pointer to HBA context object.
* @fc_hdr: pointer to a FC frame header.
*
* This function sends a basic response to a previous unsol sequence abort
* event after aborting the sequence handling.
**/
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-21 17:05:04 -06:00
void
lpfc_sli4_seq_abort_rsp(struct lpfc_vport *vport,
struct fc_frame_header *fc_hdr, bool aborted)
{
struct lpfc_hba *phba = vport->phba;
struct lpfc_iocbq *ctiocb = NULL;
struct lpfc_nodelist *ndlp;
uint16_t oxid, rxid, xri, lxri;
uint32_t sid, fctl;
IOCB_t *icmd;
int rc;
if (!lpfc_is_link_up(phba))
return;
sid = sli4_sid_from_fc_hdr(fc_hdr);
oxid = be16_to_cpu(fc_hdr->fh_ox_id);
rxid = be16_to_cpu(fc_hdr->fh_rx_id);
ndlp = lpfc_findnode_did(vport, sid);
if (!ndlp) {
ndlp = lpfc_nlp_init(vport, sid);
if (!ndlp) {
lpfc_printf_vlog(vport, KERN_WARNING, LOG_ELS,
"1268 Failed to allocate ndlp for "
"oxid:x%x SID:x%x\n", oxid, sid);
return;
}
/* Put ndlp onto pport node list */
lpfc_enqueue_node(vport, ndlp);
} else if (!NLP_CHK_NODE_ACT(ndlp)) {
/* re-setup ndlp without removing from node list */
ndlp = lpfc_enable_node(vport, ndlp, NLP_STE_UNUSED_NODE);
if (!ndlp) {
lpfc_printf_vlog(vport, KERN_WARNING, LOG_ELS,
"3275 Failed to active ndlp found "
"for oxid:x%x SID:x%x\n", oxid, sid);
return;
}
}
/* Allocate buffer for rsp iocb */
ctiocb = lpfc_sli_get_iocbq(phba);
if (!ctiocb)
return;
/* Extract the F_CTL field from FC_HDR */
fctl = sli4_fctl_from_fc_hdr(fc_hdr);
icmd = &ctiocb->iocb;
icmd->un.xseq64.bdl.bdeSize = 0;
icmd->un.xseq64.bdl.ulpIoTag32 = 0;
icmd->un.xseq64.w5.hcsw.Dfctl = 0;
icmd->un.xseq64.w5.hcsw.Rctl = FC_RCTL_BA_ACC;
icmd->un.xseq64.w5.hcsw.Type = FC_TYPE_BLS;
/* Fill in the rest of iocb fields */
icmd->ulpCommand = CMD_XMIT_BLS_RSP64_CX;
icmd->ulpBdeCount = 0;
icmd->ulpLe = 1;
icmd->ulpClass = CLASS3;
icmd->ulpContext = phba->sli4_hba.rpi_ids[ndlp->nlp_rpi];
ctiocb->context1 = lpfc_nlp_get(ndlp);
ctiocb->iocb_cmpl = NULL;
ctiocb->vport = phba->pport;
ctiocb->iocb_cmpl = lpfc_sli4_seq_abort_rsp_cmpl;
ctiocb->sli4_lxritag = NO_XRI;
ctiocb->sli4_xritag = NO_XRI;
if (fctl & FC_FC_EX_CTX)
/* Exchange responder sent the abort so we
* own the oxid.
*/
xri = oxid;
else
xri = rxid;
lxri = lpfc_sli4_xri_inrange(phba, xri);
if (lxri != NO_XRI)
lpfc_set_rrq_active(phba, ndlp, lxri,
(xri == oxid) ? rxid : oxid, 0);
/* For BA_ABTS from exchange responder, if the logical xri with
* the oxid maps to the FCP XRI range, the port no longer has
* that exchange context, send a BLS_RJT. Override the IOCB for
* a BA_RJT.
*/
if ((fctl & FC_FC_EX_CTX) &&
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
(lxri > lpfc_sli4_get_iocb_cnt(phba))) {
icmd->un.xseq64.w5.hcsw.Rctl = FC_RCTL_BA_RJT;
bf_set(lpfc_vndr_code, &icmd->un.bls_rsp, 0);
bf_set(lpfc_rsn_expln, &icmd->un.bls_rsp, FC_BA_RJT_INV_XID);
bf_set(lpfc_rsn_code, &icmd->un.bls_rsp, FC_BA_RJT_UNABLE);
}
/* If BA_ABTS failed to abort a partially assembled receive sequence,
* the driver no longer has that exchange, send a BLS_RJT. Override
* the IOCB for a BA_RJT.
*/
if (aborted == false) {
icmd->un.xseq64.w5.hcsw.Rctl = FC_RCTL_BA_RJT;
bf_set(lpfc_vndr_code, &icmd->un.bls_rsp, 0);
bf_set(lpfc_rsn_expln, &icmd->un.bls_rsp, FC_BA_RJT_INV_XID);
bf_set(lpfc_rsn_code, &icmd->un.bls_rsp, FC_BA_RJT_UNABLE);
}
if (fctl & FC_FC_EX_CTX) {
/* ABTS sent by responder to CT exchange, construction
* of BA_ACC will use OX_ID from ABTS for the XRI_TAG
* field and RX_ID from ABTS for RX_ID field.
*/
bf_set(lpfc_abts_orig, &icmd->un.bls_rsp, LPFC_ABTS_UNSOL_RSP);
} else {
/* ABTS sent by initiator to CT exchange, construction
* of BA_ACC will need to allocate a new XRI as for the
* XRI_TAG field.
*/
bf_set(lpfc_abts_orig, &icmd->un.bls_rsp, LPFC_ABTS_UNSOL_INT);
}
bf_set(lpfc_abts_rxid, &icmd->un.bls_rsp, rxid);
bf_set(lpfc_abts_oxid, &icmd->un.bls_rsp, oxid);
/* Xmit CT abts response on exchange <xid> */
lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS,
"1200 Send BLS cmd x%x on oxid x%x Data: x%x\n",
icmd->un.xseq64.w5.hcsw.Rctl, oxid, phba->link_state);
rc = lpfc_sli_issue_iocb(phba, LPFC_ELS_RING, ctiocb, 0);
if (rc == IOCB_ERROR) {
lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS,
"2925 Failed to issue CT ABTS RSP x%x on "
"xri x%x, Data x%x\n",
icmd->un.xseq64.w5.hcsw.Rctl, oxid,
phba->link_state);
lpfc_nlp_put(ndlp);
ctiocb->context1 = NULL;
lpfc_sli_release_iocbq(phba, ctiocb);
}
}
/**
* lpfc_sli4_handle_unsol_abort - Handle sli-4 unsolicited abort event
* @vport: Pointer to the vport on which this sequence was received
* @dmabuf: pointer to a dmabuf that describes the FC sequence
*
* This function handles an SLI-4 unsolicited abort event. If the unsolicited
* receive sequence is only partially assembed by the driver, it shall abort
* the partially assembled frames for the sequence. Otherwise, if the
* unsolicited receive sequence has been completely assembled and passed to
* the Upper Layer Protocol (UPL), it then mark the per oxid status for the
* unsolicited sequence has been aborted. After that, it will issue a basic
* accept to accept the abort.
**/
static void
lpfc_sli4_handle_unsol_abort(struct lpfc_vport *vport,
struct hbq_dmabuf *dmabuf)
{
struct lpfc_hba *phba = vport->phba;
struct fc_frame_header fc_hdr;
uint32_t fctl;
bool aborted;
/* Make a copy of fc_hdr before the dmabuf being released */
memcpy(&fc_hdr, dmabuf->hbuf.virt, sizeof(struct fc_frame_header));
fctl = sli4_fctl_from_fc_hdr(&fc_hdr);
if (fctl & FC_FC_EX_CTX) {
/* ABTS by responder to exchange, no cleanup needed */
aborted = true;
} else {
/* ABTS by initiator to exchange, need to do cleanup */
aborted = lpfc_sli4_abort_partial_seq(vport, dmabuf);
if (aborted == false)
aborted = lpfc_sli4_abort_ulp_seq(vport, dmabuf);
}
lpfc_in_buf_free(phba, &dmabuf->dbuf);
Update ABORT processing for NVMET. The driver with nvme had this routine stubbed. Right now XRI_ABORTED_CQE is not handled and the FC NVMET Transport has a new API for the driver. Missing code path, new NVME abort API Update ABORT processing for NVMET There are 3 new FC NVMET Transport API/ template routines for NVMET: lpfc_nvmet_xmt_fcp_release This NVMET template callback routine called to release context associated with an IO This routine is ALWAYS called last, even if the IO was aborted or completed in error. lpfc_nvmet_xmt_fcp_abort This NVMET template callback routine called to abort an exchange that has an IO in progress nvmet_fc_rcv_fcp_req When the lpfc driver receives an ABTS, this NVME FC transport layer callback routine is called. For this case there are 2 paths thru the driver: the driver either has an outstanding exchange / context for the XRI to be aborted or not. If not, a BA_RJT is issued otherwise a BA_ACC NVMET Driver abort paths: There are 2 paths for aborting an IO. The first one is we receive an IO and decide not to process it because of lack of resources. An unsolicated ABTS is immediately sent back to the initiator as a response. lpfc_nvmet_unsol_fcp_buffer lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) The second one is we sent the IO up to the NVMET transport layer to process, and for some reason the NVME Transport layer decided to abort the IO before it completes all its phases. For this case there are 2 paths thru the driver: the driver either has an outstanding TSEND/TRECEIVE/TRSP WQE or no outstanding WQEs are present for the exchange / context. lpfc_nvmet_xmt_fcp_abort if (LPFC_NVMET_IO_INP) lpfc_nvmet_sol_fcp_issue_abort (ABORT_WQE) lpfc_nvmet_sol_fcp_abort_cmp else lpfc_nvmet_unsol_fcp_issue_abort lpfc_nvmet_unsol_issue_abort (XMIT_SEQUENCE_WQE) lpfc_nvmet_unsol_fcp_abort_cmp Context flags: LPFC_NVMET_IOP - his flag signifies an IO is in progress on the exchange. LPFC_NVMET_XBUSY - this flag indicates the IO completed but the firmware is still busy with the corresponding exchange. The exchange should not be reused until after a XRI_ABORTED_CQE is received for that exchange. LPFC_NVMET_ABORT_OP - this flag signifies an ABORT_WQE was issued on the exchange. LPFC_NVMET_CTX_RLS - this flag signifies a context free was requested, but we are deferring it due to an XBUSY or ABORT in progress. A ctxlock is added to the context structure that is used whenever these flags are set/read within the context of an IO. The LPFC_NVMET_CTX_RLS flag is only set in the defer_relase routine when the transport has resolved all IO associated with the buffer. The flag is cleared when the CTX is associated with a new IO. An exchange can has both an LPFC_NVMET_XBUSY and a LPFC_NVMET_ABORT_OP condition active simultaneously. Both conditions must complete before the exchange is freed. When the abort callback (lpfc_nvmet_xmt_fcp_abort) is envoked: If there is an outstanding IO, the driver will issue an ABORT_WQE. This should result in 3 completions for the exchange: 1) IO cmpl with XB bit set 2) Abort WQE cmpl 3) XRI_ABORTED_CQE cmpl For this scenerio, after completion #1, the NVMET Transport IO rsp callback is called. After completion #2, no action is taken with respect to the exchange / context. After completion #3, the exchange context is free for re-use on another IO. If there is no outstanding activity on the exchange, the driver will send a ABTS to the Initiator. Upon completion of this WQE, the exchange / context is freed for re-use on another IO. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
2017-04-21 17:05:04 -06:00
if (phba->nvmet_support) {
lpfc_nvmet_rcv_unsol_abort(vport, &fc_hdr);
return;
}
/* Respond with BA_ACC or BA_RJT accordingly */
lpfc_sli4_seq_abort_rsp(vport, &fc_hdr, aborted);
}
/**
* lpfc_seq_complete - Indicates if a sequence is complete
* @dmabuf: pointer to a dmabuf that describes the FC sequence
*
* This function checks the sequence, starting with the frame described by
* @dmabuf, to see if all the frames associated with this sequence are present.
* the frames associated with this sequence are linked to the @dmabuf using the
* dbuf list. This function looks for two major things. 1) That the first frame
* has a sequence count of zero. 2) There is a frame with last frame of sequence
* set. 3) That there are no holes in the sequence count. The function will
* return 1 when the sequence is complete, otherwise it will return 0.
**/
static int
lpfc_seq_complete(struct hbq_dmabuf *dmabuf)
{
struct fc_frame_header *hdr;
struct lpfc_dmabuf *d_buf;
struct hbq_dmabuf *seq_dmabuf;
uint32_t fctl;
int seq_count = 0;
hdr = (struct fc_frame_header *)dmabuf->hbuf.virt;
/* make sure first fame of sequence has a sequence count of zero */
if (hdr->fh_seq_cnt != seq_count)
return 0;
fctl = (hdr->fh_f_ctl[0] << 16 |
hdr->fh_f_ctl[1] << 8 |
hdr->fh_f_ctl[2]);
/* If last frame of sequence we can return success. */
if (fctl & FC_FC_END_SEQ)
return 1;
list_for_each_entry(d_buf, &dmabuf->dbuf.list, list) {
seq_dmabuf = container_of(d_buf, struct hbq_dmabuf, dbuf);
hdr = (struct fc_frame_header *)seq_dmabuf->hbuf.virt;
/* If there is a hole in the sequence count then fail. */
if (++seq_count != be16_to_cpu(hdr->fh_seq_cnt))
return 0;
fctl = (hdr->fh_f_ctl[0] << 16 |
hdr->fh_f_ctl[1] << 8 |
hdr->fh_f_ctl[2]);
/* If last frame of sequence we can return success. */
if (fctl & FC_FC_END_SEQ)
return 1;
}
return 0;
}
/**
* lpfc_prep_seq - Prep sequence for ULP processing
* @vport: Pointer to the vport on which this sequence was received
* @dmabuf: pointer to a dmabuf that describes the FC sequence
*
* This function takes a sequence, described by a list of frames, and creates
* a list of iocbq structures to describe the sequence. This iocbq list will be
* used to issue to the generic unsolicited sequence handler. This routine
* returns a pointer to the first iocbq in the list. If the function is unable
* to allocate an iocbq then it throw out the received frames that were not
* able to be described and return a pointer to the first iocbq. If unable to
* allocate any iocbqs (including the first) this function will return NULL.
**/
static struct lpfc_iocbq *
lpfc_prep_seq(struct lpfc_vport *vport, struct hbq_dmabuf *seq_dmabuf)
{
struct hbq_dmabuf *hbq_buf;
struct lpfc_dmabuf *d_buf, *n_buf;
struct lpfc_iocbq *first_iocbq, *iocbq;
struct fc_frame_header *fc_hdr;
uint32_t sid;
uint32_t len, tot_len;
struct ulp_bde64 *pbde;
fc_hdr = (struct fc_frame_header *)seq_dmabuf->hbuf.virt;
/* remove from receive buffer list */
list_del_init(&seq_dmabuf->hbuf.list);
lpfc_update_rcv_time_stamp(vport);
/* get the Remote Port's SID */
sid = sli4_sid_from_fc_hdr(fc_hdr);
tot_len = 0;
/* Get an iocbq struct to fill in. */
first_iocbq = lpfc_sli_get_iocbq(vport->phba);
if (first_iocbq) {
/* Initialize the first IOCB. */
first_iocbq->iocb.unsli3.rcvsli3.acc_len = 0;
first_iocbq->iocb.ulpStatus = IOSTAT_SUCCESS;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
first_iocbq->vport = vport;
/* Check FC Header to see what TYPE of frame we are rcv'ing */
if (sli4_type_from_fc_hdr(fc_hdr) == FC_TYPE_ELS) {
first_iocbq->iocb.ulpCommand = CMD_IOCB_RCV_ELS64_CX;
first_iocbq->iocb.un.rcvels.parmRo =
sli4_did_from_fc_hdr(fc_hdr);
first_iocbq->iocb.ulpPU = PARM_NPIV_DID;
} else
first_iocbq->iocb.ulpCommand = CMD_IOCB_RCV_SEQ64_CX;
first_iocbq->iocb.ulpContext = NO_XRI;
first_iocbq->iocb.unsli3.rcvsli3.ox_id =
be16_to_cpu(fc_hdr->fh_ox_id);
/* iocbq is prepped for internal consumption. Physical vpi. */
first_iocbq->iocb.unsli3.rcvsli3.vpi =
vport->phba->vpi_ids[vport->vpi];
/* put the first buffer into the first IOCBq */
tot_len = bf_get(lpfc_rcqe_length,
&seq_dmabuf->cq_event.cqe.rcqe_cmpl);
first_iocbq->context2 = &seq_dmabuf->dbuf;
first_iocbq->context3 = NULL;
first_iocbq->iocb.ulpBdeCount = 1;
if (tot_len > LPFC_DATA_BUF_SIZE)
first_iocbq->iocb.un.cont64[0].tus.f.bdeSize =
LPFC_DATA_BUF_SIZE;
else
first_iocbq->iocb.un.cont64[0].tus.f.bdeSize = tot_len;
first_iocbq->iocb.un.rcvels.remoteID = sid;
first_iocbq->iocb.unsli3.rcvsli3.acc_len = tot_len;
}
iocbq = first_iocbq;
/*
* Each IOCBq can have two Buffers assigned, so go through the list
* of buffers for this sequence and save two buffers in each IOCBq
*/
list_for_each_entry_safe(d_buf, n_buf, &seq_dmabuf->dbuf.list, list) {
if (!iocbq) {
lpfc_in_buf_free(vport->phba, d_buf);
continue;
}
if (!iocbq->context3) {
iocbq->context3 = d_buf;
iocbq->iocb.ulpBdeCount++;
/* We need to get the size out of the right CQE */
hbq_buf = container_of(d_buf, struct hbq_dmabuf, dbuf);
len = bf_get(lpfc_rcqe_length,
&hbq_buf->cq_event.cqe.rcqe_cmpl);
pbde = (struct ulp_bde64 *)
&iocbq->iocb.unsli3.sli3Words[4];
if (len > LPFC_DATA_BUF_SIZE)
pbde->tus.f.bdeSize = LPFC_DATA_BUF_SIZE;
else
pbde->tus.f.bdeSize = len;
iocbq->iocb.unsli3.rcvsli3.acc_len += len;
tot_len += len;
} else {
iocbq = lpfc_sli_get_iocbq(vport->phba);
if (!iocbq) {
if (first_iocbq) {
first_iocbq->iocb.ulpStatus =
IOSTAT_FCP_RSP_ERROR;
first_iocbq->iocb.un.ulpWord[4] =
IOERR_NO_RESOURCES;
}
lpfc_in_buf_free(vport->phba, d_buf);
continue;
}
/* We need to get the size out of the right CQE */
hbq_buf = container_of(d_buf, struct hbq_dmabuf, dbuf);
len = bf_get(lpfc_rcqe_length,
&hbq_buf->cq_event.cqe.rcqe_cmpl);
iocbq->context2 = d_buf;
iocbq->context3 = NULL;
iocbq->iocb.ulpBdeCount = 1;
if (len > LPFC_DATA_BUF_SIZE)
iocbq->iocb.un.cont64[0].tus.f.bdeSize =
LPFC_DATA_BUF_SIZE;
else
iocbq->iocb.un.cont64[0].tus.f.bdeSize = len;
tot_len += len;
iocbq->iocb.unsli3.rcvsli3.acc_len = tot_len;
iocbq->iocb.un.rcvels.remoteID = sid;
list_add_tail(&iocbq->list, &first_iocbq->list);
}
}
return first_iocbq;
}
static void
lpfc_sli4_send_seq_to_ulp(struct lpfc_vport *vport,
struct hbq_dmabuf *seq_dmabuf)
{
struct fc_frame_header *fc_hdr;
struct lpfc_iocbq *iocbq, *curr_iocb, *next_iocb;
struct lpfc_hba *phba = vport->phba;
fc_hdr = (struct fc_frame_header *)seq_dmabuf->hbuf.virt;
iocbq = lpfc_prep_seq(vport, seq_dmabuf);
if (!iocbq) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2707 Ring %d handler: Failed to allocate "
"iocb Rctl x%x Type x%x received\n",
LPFC_ELS_RING,
fc_hdr->fh_r_ctl, fc_hdr->fh_type);
return;
}
if (!lpfc_complete_unsol_iocb(phba,
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
phba->sli4_hba.els_wq->pring,
iocbq, fc_hdr->fh_r_ctl,
fc_hdr->fh_type))
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2540 Ring %d handler: unexpected Rctl "
"x%x Type x%x received\n",
LPFC_ELS_RING,
fc_hdr->fh_r_ctl, fc_hdr->fh_type);
/* Free iocb created in lpfc_prep_seq */
list_for_each_entry_safe(curr_iocb, next_iocb,
&iocbq->list, list) {
list_del_init(&curr_iocb->list);
lpfc_sli_release_iocbq(phba, curr_iocb);
}
lpfc_sli_release_iocbq(phba, iocbq);
}
static void
lpfc_sli4_mds_loopback_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
struct lpfc_iocbq *rspiocb)
{
struct lpfc_dmabuf *pcmd = cmdiocb->context2;
if (pcmd && pcmd->virt)
dma_pool_free(phba->lpfc_drb_pool, pcmd->virt, pcmd->phys);
kfree(pcmd);
lpfc_sli_release_iocbq(phba, cmdiocb);
}
static void
lpfc_sli4_handle_mds_loopback(struct lpfc_vport *vport,
struct hbq_dmabuf *dmabuf)
{
struct fc_frame_header *fc_hdr;
struct lpfc_hba *phba = vport->phba;
struct lpfc_iocbq *iocbq = NULL;
union lpfc_wqe *wqe;
struct lpfc_dmabuf *pcmd = NULL;
uint32_t frame_len;
int rc;
fc_hdr = (struct fc_frame_header *)dmabuf->hbuf.virt;
frame_len = bf_get(lpfc_rcqe_length, &dmabuf->cq_event.cqe.rcqe_cmpl);
/* Send the received frame back */
iocbq = lpfc_sli_get_iocbq(phba);
if (!iocbq)
goto exit;
/* Allocate buffer for command payload */
pcmd = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
if (pcmd)
pcmd->virt = dma_pool_alloc(phba->lpfc_drb_pool, GFP_KERNEL,
&pcmd->phys);
if (!pcmd || !pcmd->virt)
goto exit;
INIT_LIST_HEAD(&pcmd->list);
/* copyin the payload */
memcpy(pcmd->virt, dmabuf->dbuf.virt, frame_len);
/* fill in BDE's for command */
iocbq->iocb.un.xseq64.bdl.addrHigh = putPaddrHigh(pcmd->phys);
iocbq->iocb.un.xseq64.bdl.addrLow = putPaddrLow(pcmd->phys);
iocbq->iocb.un.xseq64.bdl.bdeFlags = BUFF_TYPE_BDE_64;
iocbq->iocb.un.xseq64.bdl.bdeSize = frame_len;
iocbq->context2 = pcmd;
iocbq->vport = vport;
iocbq->iocb_flag &= ~LPFC_FIP_ELS_ID_MASK;
iocbq->iocb_flag |= LPFC_USE_FCPWQIDX;
/*
* Setup rest of the iocb as though it were a WQE
* Build the SEND_FRAME WQE
*/
wqe = (union lpfc_wqe *)&iocbq->iocb;
wqe->send_frame.frame_len = frame_len;
wqe->send_frame.fc_hdr_wd0 = be32_to_cpu(*((uint32_t *)fc_hdr));
wqe->send_frame.fc_hdr_wd1 = be32_to_cpu(*((uint32_t *)fc_hdr + 1));
wqe->send_frame.fc_hdr_wd2 = be32_to_cpu(*((uint32_t *)fc_hdr + 2));
wqe->send_frame.fc_hdr_wd3 = be32_to_cpu(*((uint32_t *)fc_hdr + 3));
wqe->send_frame.fc_hdr_wd4 = be32_to_cpu(*((uint32_t *)fc_hdr + 4));
wqe->send_frame.fc_hdr_wd5 = be32_to_cpu(*((uint32_t *)fc_hdr + 5));
iocbq->iocb.ulpCommand = CMD_SEND_FRAME;
iocbq->iocb.ulpLe = 1;
iocbq->iocb_cmpl = lpfc_sli4_mds_loopback_cmpl;
rc = lpfc_sli_issue_iocb(phba, LPFC_ELS_RING, iocbq, 0);
if (rc == IOCB_ERROR)
goto exit;
lpfc_in_buf_free(phba, &dmabuf->dbuf);
return;
exit:
lpfc_printf_log(phba, KERN_WARNING, LOG_SLI,
"2023 Unable to process MDS loopback frame\n");
if (pcmd && pcmd->virt)
dma_pool_free(phba->lpfc_drb_pool, pcmd->virt, pcmd->phys);
kfree(pcmd);
if (iocbq)
lpfc_sli_release_iocbq(phba, iocbq);
lpfc_in_buf_free(phba, &dmabuf->dbuf);
}
/**
* lpfc_sli4_handle_received_buffer - Handle received buffers from firmware
* @phba: Pointer to HBA context object.
*
* This function is called with no lock held. This function processes all
* the received buffers and gives it to upper layers when a received buffer
* indicates that it is the final frame in the sequence. The interrupt
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
* service routine processes received buffers at interrupt contexts.
* Worker thread calls lpfc_sli4_handle_received_buffer, which will call the
* appropriate receive function when the final frame in a sequence is received.
**/
void
lpfc_sli4_handle_received_buffer(struct lpfc_hba *phba,
struct hbq_dmabuf *dmabuf)
{
struct hbq_dmabuf *seq_dmabuf;
struct fc_frame_header *fc_hdr;
struct lpfc_vport *vport;
uint32_t fcfi;
uint32_t did;
/* Process each received buffer */
fc_hdr = (struct fc_frame_header *)dmabuf->hbuf.virt;
/* check to see if this a valid type of frame */
if (lpfc_fc_frame_check(phba, fc_hdr)) {
lpfc_in_buf_free(phba, &dmabuf->dbuf);
return;
}
if ((bf_get(lpfc_cqe_code,
&dmabuf->cq_event.cqe.rcqe_cmpl) == CQE_CODE_RECEIVE_V1))
fcfi = bf_get(lpfc_rcqe_fcf_id_v1,
&dmabuf->cq_event.cqe.rcqe_cmpl);
else
fcfi = bf_get(lpfc_rcqe_fcf_id,
&dmabuf->cq_event.cqe.rcqe_cmpl);
if (fc_hdr->fh_r_ctl == 0xF4 && fc_hdr->fh_type == 0xFF) {
vport = phba->pport;
/* Handle MDS Loopback frames */
lpfc_sli4_handle_mds_loopback(vport, dmabuf);
return;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* d_id this frame is directed to */
did = sli4_did_from_fc_hdr(fc_hdr);
vport = lpfc_fc_frame_to_vport(phba, fc_hdr, fcfi, did);
if (!vport) {
/* throw out the frame */
lpfc_in_buf_free(phba, &dmabuf->dbuf);
return;
}
/* vport is registered unless we rcv a FLOGI directed to Fabric_DID */
if (!(vport->vpi_state & LPFC_VPI_REGISTERED) &&
(did != Fabric_DID)) {
/*
* Throw out the frame if we are not pt2pt.
* The pt2pt protocol allows for discovery frames
* to be received without a registered VPI.
*/
if (!(vport->fc_flag & FC_PT2PT) ||
(phba->link_state == LPFC_HBA_READY)) {
lpfc_in_buf_free(phba, &dmabuf->dbuf);
return;
}
}
/* Handle the basic abort sequence (BA_ABTS) event */
if (fc_hdr->fh_r_ctl == FC_RCTL_BA_ABTS) {
lpfc_sli4_handle_unsol_abort(vport, dmabuf);
return;
}
/* Link this frame */
seq_dmabuf = lpfc_fc_frame_add(vport, dmabuf);
if (!seq_dmabuf) {
/* unable to add frame to vport - throw it out */
lpfc_in_buf_free(phba, &dmabuf->dbuf);
return;
}
/* If not last frame in sequence continue processing frames. */
if (!lpfc_seq_complete(seq_dmabuf))
return;
/* Send the complete sequence to the upper layer protocol */
lpfc_sli4_send_seq_to_ulp(vport, seq_dmabuf);
}
/**
* lpfc_sli4_post_all_rpi_hdrs - Post the rpi header memory region to the port
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to post rpi header templates to the
* HBA consistent with the SLI-4 interface spec. This routine
* posts a SLI4_PAGE_SIZE memory region to the port to hold up to
* SLI4_PAGE_SIZE modulo 64 rpi context headers.
*
* This routine does not require any locks. It's usage is expected
* to be driver load or reset recovery when the driver is
* sequential.
*
* Return codes
* 0 - successful
* -EIO - The mailbox failed to complete successfully.
* When this error occurs, the driver is not guaranteed
* to have any rpi regions posted to the device and
* must either attempt to repost the regions or take a
* fatal error.
**/
int
lpfc_sli4_post_all_rpi_hdrs(struct lpfc_hba *phba)
{
struct lpfc_rpi_hdr *rpi_page;
uint32_t rc = 0;
uint16_t lrpi = 0;
/* SLI4 ports that support extents do not require RPI headers. */
if (!phba->sli4_hba.rpi_hdrs_in_use)
goto exit;
if (phba->sli4_hba.extents_in_use)
return -EIO;
list_for_each_entry(rpi_page, &phba->sli4_hba.lpfc_rpi_hdr_list, list) {
/*
* Assign the rpi headers a physical rpi only if the driver
* has not initialized those resources. A port reset only
* needs the headers posted.
*/
if (bf_get(lpfc_rpi_rsrc_rdy, &phba->sli4_hba.sli4_flags) !=
LPFC_RPI_RSRC_RDY)
rpi_page->start_rpi = phba->sli4_hba.rpi_ids[lrpi];
rc = lpfc_sli4_post_rpi_hdr(phba, rpi_page);
if (rc != MBX_SUCCESS) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2008 Error %d posting all rpi "
"headers\n", rc);
rc = -EIO;
break;
}
}
exit:
bf_set(lpfc_rpi_rsrc_rdy, &phba->sli4_hba.sli4_flags,
LPFC_RPI_RSRC_RDY);
return rc;
}
/**
* lpfc_sli4_post_rpi_hdr - Post an rpi header memory region to the port
* @phba: pointer to lpfc hba data structure.
* @rpi_page: pointer to the rpi memory region.
*
* This routine is invoked to post a single rpi header to the
* HBA consistent with the SLI-4 interface spec. This memory region
* maps up to 64 rpi context regions.
*
* Return codes
* 0 - successful
* -ENOMEM - No available memory
* -EIO - The mailbox failed to complete successfully.
**/
int
lpfc_sli4_post_rpi_hdr(struct lpfc_hba *phba, struct lpfc_rpi_hdr *rpi_page)
{
LPFC_MBOXQ_t *mboxq;
struct lpfc_mbx_post_hdr_tmpl *hdr_tmpl;
uint32_t rc = 0;
uint32_t shdr_status, shdr_add_status;
union lpfc_sli4_cfg_shdr *shdr;
/* SLI4 ports that support extents do not require RPI headers. */
if (!phba->sli4_hba.rpi_hdrs_in_use)
return rc;
if (phba->sli4_hba.extents_in_use)
return -EIO;
/* The port is notified of the header region via a mailbox command. */
mboxq = (LPFC_MBOXQ_t *) mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2001 Unable to allocate memory for issuing "
"SLI_CONFIG_SPECIAL mailbox command\n");
return -ENOMEM;
}
/* Post all rpi memory regions to the port. */
hdr_tmpl = &mboxq->u.mqe.un.hdr_tmpl;
lpfc_sli4_config(phba, mboxq, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_POST_HDR_TEMPLATE,
sizeof(struct lpfc_mbx_post_hdr_tmpl) -
sizeof(struct lpfc_sli4_cfg_mhdr),
LPFC_SLI4_MBX_EMBED);
/* Post the physical rpi to the port for this rpi header. */
bf_set(lpfc_mbx_post_hdr_tmpl_rpi_offset, hdr_tmpl,
rpi_page->start_rpi);
bf_set(lpfc_mbx_post_hdr_tmpl_page_cnt,
hdr_tmpl, rpi_page->page_count);
hdr_tmpl->rpi_paddr_lo = putPaddrLow(rpi_page->dmabuf->phys);
hdr_tmpl->rpi_paddr_hi = putPaddrHigh(rpi_page->dmabuf->phys);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
shdr = (union lpfc_sli4_cfg_shdr *) &hdr_tmpl->header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (rc != MBX_TIMEOUT)
mempool_free(mboxq, phba->mbox_mem_pool);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2514 POST_RPI_HDR mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
rc = -ENXIO;
} else {
/*
* The next_rpi stores the next logical module-64 rpi value used
* to post physical rpis in subsequent rpi postings.
*/
spin_lock_irq(&phba->hbalock);
phba->sli4_hba.next_rpi = rpi_page->next_rpi;
spin_unlock_irq(&phba->hbalock);
}
return rc;
}
/**
* lpfc_sli4_alloc_rpi - Get an available rpi in the device's range
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to post rpi header templates to the
* HBA consistent with the SLI-4 interface spec. This routine
* posts a SLI4_PAGE_SIZE memory region to the port to hold up to
* SLI4_PAGE_SIZE modulo 64 rpi context headers.
*
* Returns
* A nonzero rpi defined as rpi_base <= rpi < max_rpi if successful
* LPFC_RPI_ALLOC_ERROR if no rpis are available.
**/
int
lpfc_sli4_alloc_rpi(struct lpfc_hba *phba)
{
unsigned long rpi;
uint16_t max_rpi, rpi_limit;
uint16_t rpi_remaining, lrpi = 0;
struct lpfc_rpi_hdr *rpi_hdr;
unsigned long iflag;
/*
* Fetch the next logical rpi. Because this index is logical,
* the driver starts at 0 each time.
*/
spin_lock_irqsave(&phba->hbalock, iflag);
max_rpi = phba->sli4_hba.max_cfg_param.max_rpi;
rpi_limit = phba->sli4_hba.next_rpi;
rpi = find_next_zero_bit(phba->sli4_hba.rpi_bmask, rpi_limit, 0);
if (rpi >= rpi_limit)
rpi = LPFC_RPI_ALLOC_ERROR;
else {
set_bit(rpi, phba->sli4_hba.rpi_bmask);
phba->sli4_hba.max_cfg_param.rpi_used++;
phba->sli4_hba.rpi_count++;
}
lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
"0001 rpi:%x max:%x lim:%x\n",
(int) rpi, max_rpi, rpi_limit);
/*
* Don't try to allocate more rpi header regions if the device limit
* has been exhausted.
*/
if ((rpi == LPFC_RPI_ALLOC_ERROR) &&
(phba->sli4_hba.rpi_count >= max_rpi)) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
return rpi;
}
/*
* RPI header postings are not required for SLI4 ports capable of
* extents.
*/
if (!phba->sli4_hba.rpi_hdrs_in_use) {
spin_unlock_irqrestore(&phba->hbalock, iflag);
return rpi;
}
/*
* If the driver is running low on rpi resources, allocate another
* page now. Note that the next_rpi value is used because
* it represents how many are actually in use whereas max_rpi notes
* how many are supported max by the device.
*/
rpi_remaining = phba->sli4_hba.next_rpi - phba->sli4_hba.rpi_count;
spin_unlock_irqrestore(&phba->hbalock, iflag);
if (rpi_remaining < LPFC_RPI_LOW_WATER_MARK) {
rpi_hdr = lpfc_sli4_create_rpi_hdr(phba);
if (!rpi_hdr) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2002 Error Could not grow rpi "
"count\n");
} else {
lrpi = rpi_hdr->start_rpi;
rpi_hdr->start_rpi = phba->sli4_hba.rpi_ids[lrpi];
lpfc_sli4_post_rpi_hdr(phba, rpi_hdr);
}
}
return rpi;
}
/**
* lpfc_sli4_free_rpi - Release an rpi for reuse.
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to release an rpi to the pool of
* available rpis maintained by the driver.
**/
static void
__lpfc_sli4_free_rpi(struct lpfc_hba *phba, int rpi)
{
if (test_and_clear_bit(rpi, phba->sli4_hba.rpi_bmask)) {
phba->sli4_hba.rpi_count--;
phba->sli4_hba.max_cfg_param.rpi_used--;
}
}
/**
* lpfc_sli4_free_rpi - Release an rpi for reuse.
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to release an rpi to the pool of
* available rpis maintained by the driver.
**/
void
lpfc_sli4_free_rpi(struct lpfc_hba *phba, int rpi)
{
spin_lock_irq(&phba->hbalock);
__lpfc_sli4_free_rpi(phba, rpi);
spin_unlock_irq(&phba->hbalock);
}
/**
* lpfc_sli4_remove_rpis - Remove the rpi bitmask region
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to remove the memory region that
* provided rpi via a bitmask.
**/
void
lpfc_sli4_remove_rpis(struct lpfc_hba *phba)
{
kfree(phba->sli4_hba.rpi_bmask);
kfree(phba->sli4_hba.rpi_ids);
bf_set(lpfc_rpi_rsrc_rdy, &phba->sli4_hba.sli4_flags, 0);
}
/**
* lpfc_sli4_resume_rpi - Remove the rpi bitmask region
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to remove the memory region that
* provided rpi via a bitmask.
**/
int
lpfc_sli4_resume_rpi(struct lpfc_nodelist *ndlp,
void (*cmpl)(struct lpfc_hba *, LPFC_MBOXQ_t *), void *arg)
{
LPFC_MBOXQ_t *mboxq;
struct lpfc_hba *phba = ndlp->phba;
int rc;
/* The port is notified of the header region via a mailbox command. */
mboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq)
return -ENOMEM;
/* Post all rpi memory regions to the port. */
lpfc_resume_rpi(mboxq, ndlp);
if (cmpl) {
mboxq->mbox_cmpl = cmpl;
mboxq->context1 = arg;
mboxq->context2 = ndlp;
} else
mboxq->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
mboxq->vport = ndlp->vport;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_NOWAIT);
if (rc == MBX_NOT_FINISHED) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2010 Resume RPI Mailbox failed "
"status %d, mbxStatus x%x\n", rc,
bf_get(lpfc_mqe_status, &mboxq->u.mqe));
mempool_free(mboxq, phba->mbox_mem_pool);
return -EIO;
}
return 0;
}
/**
* lpfc_sli4_init_vpi - Initialize a vpi with the port
2010-11-20 21:11:48 -07:00
* @vport: Pointer to the vport for which the vpi is being initialized
*
2010-11-20 21:11:48 -07:00
* This routine is invoked to activate a vpi with the port.
*
* Returns:
* 0 success
* -Evalue otherwise
**/
int
2010-11-20 21:11:48 -07:00
lpfc_sli4_init_vpi(struct lpfc_vport *vport)
{
LPFC_MBOXQ_t *mboxq;
int rc = 0;
int retval = MBX_SUCCESS;
uint32_t mbox_tmo;
2010-11-20 21:11:48 -07:00
struct lpfc_hba *phba = vport->phba;
mboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq)
return -ENOMEM;
2010-11-20 21:11:48 -07:00
lpfc_init_vpi(phba, mboxq, vport->vpi);
mbox_tmo = lpfc_mbox_tmo_val(phba, mboxq);
rc = lpfc_sli_issue_mbox_wait(phba, mboxq, mbox_tmo);
if (rc != MBX_SUCCESS) {
2010-11-20 21:11:48 -07:00
lpfc_printf_vlog(vport, KERN_ERR, LOG_SLI,
"2022 INIT VPI Mailbox failed "
"status %d, mbxStatus x%x\n", rc,
bf_get(lpfc_mqe_status, &mboxq->u.mqe));
retval = -EIO;
}
if (rc != MBX_TIMEOUT)
2010-11-20 21:11:48 -07:00
mempool_free(mboxq, vport->phba->mbox_mem_pool);
return retval;
}
/**
* lpfc_mbx_cmpl_add_fcf_record - add fcf mbox completion handler.
* @phba: pointer to lpfc hba data structure.
* @mboxq: Pointer to mailbox object.
*
* This routine is invoked to manually add a single FCF record. The caller
* must pass a completely initialized FCF_Record. This routine takes
* care of the nonembedded mailbox operations.
**/
static void
lpfc_mbx_cmpl_add_fcf_record(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
{
void *virt_addr;
union lpfc_sli4_cfg_shdr *shdr;
uint32_t shdr_status, shdr_add_status;
virt_addr = mboxq->sge_array->addr[0];
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *) virt_addr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if ((shdr_status || shdr_add_status) &&
(shdr_status != STATUS_FCF_IN_USE))
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2558 ADD_FCF_RECORD mailbox failed with "
"status x%x add_status x%x\n",
shdr_status, shdr_add_status);
lpfc_sli4_mbox_cmd_free(phba, mboxq);
}
/**
* lpfc_sli4_add_fcf_record - Manually add an FCF Record.
* @phba: pointer to lpfc hba data structure.
* @fcf_record: pointer to the initialized fcf record to add.
*
* This routine is invoked to manually add a single FCF record. The caller
* must pass a completely initialized FCF_Record. This routine takes
* care of the nonembedded mailbox operations.
**/
int
lpfc_sli4_add_fcf_record(struct lpfc_hba *phba, struct fcf_record *fcf_record)
{
int rc = 0;
LPFC_MBOXQ_t *mboxq;
uint8_t *bytep;
void *virt_addr;
struct lpfc_mbx_sge sge;
uint32_t alloc_len, req_len;
uint32_t fcfindex;
mboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2009 Failed to allocate mbox for ADD_FCF cmd\n");
return -ENOMEM;
}
req_len = sizeof(struct fcf_record) + sizeof(union lpfc_sli4_cfg_shdr) +
sizeof(uint32_t);
/* Allocate DMA memory and set up the non-embedded mailbox command */
alloc_len = lpfc_sli4_config(phba, mboxq, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_ADD_FCF,
req_len, LPFC_SLI4_MBX_NEMBED);
if (alloc_len < req_len) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2523 Allocated DMA memory size (x%x) is "
"less than the requested DMA memory "
"size (x%x)\n", alloc_len, req_len);
lpfc_sli4_mbox_cmd_free(phba, mboxq);
return -ENOMEM;
}
/*
* Get the first SGE entry from the non-embedded DMA memory. This
* routine only uses a single SGE.
*/
lpfc_sli4_mbx_sge_get(mboxq, 0, &sge);
virt_addr = mboxq->sge_array->addr[0];
/*
* Configure the FCF record for FCFI 0. This is the driver's
* hardcoded default and gets used in nonFIP mode.
*/
fcfindex = bf_get(lpfc_fcf_record_fcf_index, fcf_record);
bytep = virt_addr + sizeof(union lpfc_sli4_cfg_shdr);
lpfc_sli_pcimem_bcopy(&fcfindex, bytep, sizeof(uint32_t));
/*
* Copy the fcf_index and the FCF Record Data. The data starts after
* the FCoE header plus word10. The data copy needs to be endian
* correct.
*/
bytep += sizeof(uint32_t);
lpfc_sli_pcimem_bcopy(fcf_record, bytep, sizeof(struct fcf_record));
mboxq->vport = phba->pport;
mboxq->mbox_cmpl = lpfc_mbx_cmpl_add_fcf_record;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_NOWAIT);
if (rc == MBX_NOT_FINISHED) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2515 ADD_FCF_RECORD mailbox failed with "
"status 0x%x\n", rc);
lpfc_sli4_mbox_cmd_free(phba, mboxq);
rc = -EIO;
} else
rc = 0;
return rc;
}
/**
* lpfc_sli4_build_dflt_fcf_record - Build the driver's default FCF Record.
* @phba: pointer to lpfc hba data structure.
* @fcf_record: pointer to the fcf record to write the default data.
* @fcf_index: FCF table entry index.
*
* This routine is invoked to build the driver's default FCF record. The
* values used are hardcoded. This routine handles memory initialization.
*
**/
void
lpfc_sli4_build_dflt_fcf_record(struct lpfc_hba *phba,
struct fcf_record *fcf_record,
uint16_t fcf_index)
{
memset(fcf_record, 0, sizeof(struct fcf_record));
fcf_record->max_rcv_size = LPFC_FCOE_MAX_RCV_SIZE;
fcf_record->fka_adv_period = LPFC_FCOE_FKA_ADV_PER;
fcf_record->fip_priority = LPFC_FCOE_FIP_PRIORITY;
bf_set(lpfc_fcf_record_mac_0, fcf_record, phba->fc_map[0]);
bf_set(lpfc_fcf_record_mac_1, fcf_record, phba->fc_map[1]);
bf_set(lpfc_fcf_record_mac_2, fcf_record, phba->fc_map[2]);
bf_set(lpfc_fcf_record_mac_3, fcf_record, LPFC_FCOE_FCF_MAC3);
bf_set(lpfc_fcf_record_mac_4, fcf_record, LPFC_FCOE_FCF_MAC4);
bf_set(lpfc_fcf_record_mac_5, fcf_record, LPFC_FCOE_FCF_MAC5);
bf_set(lpfc_fcf_record_fc_map_0, fcf_record, phba->fc_map[0]);
bf_set(lpfc_fcf_record_fc_map_1, fcf_record, phba->fc_map[1]);
bf_set(lpfc_fcf_record_fc_map_2, fcf_record, phba->fc_map[2]);
bf_set(lpfc_fcf_record_fcf_valid, fcf_record, 1);
bf_set(lpfc_fcf_record_fcf_avail, fcf_record, 1);
bf_set(lpfc_fcf_record_fcf_index, fcf_record, fcf_index);
bf_set(lpfc_fcf_record_mac_addr_prov, fcf_record,
LPFC_FCF_FPMA | LPFC_FCF_SPMA);
/* Set the VLAN bit map */
if (phba->valid_vlan) {
fcf_record->vlan_bitmap[phba->vlan_id / 8]
= 1 << (phba->vlan_id % 8);
}
}
/**
* lpfc_sli4_fcf_scan_read_fcf_rec - Read hba fcf record for fcf scan.
* @phba: pointer to lpfc hba data structure.
* @fcf_index: FCF table entry offset.
*
* This routine is invoked to scan the entire FCF table by reading FCF
* record and processing it one at a time starting from the @fcf_index
* for initial FCF discovery or fast FCF failover rediscovery.
*
* Return 0 if the mailbox command is submitted successfully, none 0
* otherwise.
**/
int
lpfc_sli4_fcf_scan_read_fcf_rec(struct lpfc_hba *phba, uint16_t fcf_index)
{
int rc = 0, error;
LPFC_MBOXQ_t *mboxq;
phba->fcoe_eventtag_at_fcf_scan = phba->fcoe_eventtag;
phba->fcoe_cvl_eventtag_attn = phba->fcoe_cvl_eventtag;
mboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2000 Failed to allocate mbox for "
"READ_FCF cmd\n");
error = -ENOMEM;
goto fail_fcf_scan;
}
/* Construct the read FCF record mailbox command */
rc = lpfc_sli4_mbx_read_fcf_rec(phba, mboxq, fcf_index);
if (rc) {
error = -EINVAL;
goto fail_fcf_scan;
}
/* Issue the mailbox command asynchronously */
mboxq->vport = phba->pport;
mboxq->mbox_cmpl = lpfc_mbx_cmpl_fcf_scan_read_fcf_rec;
spin_lock_irq(&phba->hbalock);
phba->hba_flag |= FCF_TS_INPROG;
spin_unlock_irq(&phba->hbalock);
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_NOWAIT);
if (rc == MBX_NOT_FINISHED)
error = -EIO;
else {
/* Reset eligible FCF count for new scan */
if (fcf_index == LPFC_FCOE_FCF_GET_FIRST)
phba->fcf.eligible_fcf_cnt = 0;
error = 0;
}
fail_fcf_scan:
if (error) {
if (mboxq)
lpfc_sli4_mbox_cmd_free(phba, mboxq);
/* FCF scan failed, clear FCF_TS_INPROG flag */
spin_lock_irq(&phba->hbalock);
phba->hba_flag &= ~FCF_TS_INPROG;
spin_unlock_irq(&phba->hbalock);
}
return error;
}
/**
* lpfc_sli4_fcf_rr_read_fcf_rec - Read hba fcf record for roundrobin fcf.
* @phba: pointer to lpfc hba data structure.
* @fcf_index: FCF table entry offset.
*
* This routine is invoked to read an FCF record indicated by @fcf_index
* and to use it for FLOGI roundrobin FCF failover.
*
* Return 0 if the mailbox command is submitted successfully, none 0
* otherwise.
**/
int
lpfc_sli4_fcf_rr_read_fcf_rec(struct lpfc_hba *phba, uint16_t fcf_index)
{
int rc = 0, error;
LPFC_MBOXQ_t *mboxq;
mboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq) {
lpfc_printf_log(phba, KERN_ERR, LOG_FIP | LOG_INIT,
"2763 Failed to allocate mbox for "
"READ_FCF cmd\n");
error = -ENOMEM;
goto fail_fcf_read;
}
/* Construct the read FCF record mailbox command */
rc = lpfc_sli4_mbx_read_fcf_rec(phba, mboxq, fcf_index);
if (rc) {
error = -EINVAL;
goto fail_fcf_read;
}
/* Issue the mailbox command asynchronously */
mboxq->vport = phba->pport;
mboxq->mbox_cmpl = lpfc_mbx_cmpl_fcf_rr_read_fcf_rec;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_NOWAIT);
if (rc == MBX_NOT_FINISHED)
error = -EIO;
else
error = 0;
fail_fcf_read:
if (error && mboxq)
lpfc_sli4_mbox_cmd_free(phba, mboxq);
return error;
}
/**
* lpfc_sli4_read_fcf_rec - Read hba fcf record for update eligible fcf bmask.
* @phba: pointer to lpfc hba data structure.
* @fcf_index: FCF table entry offset.
*
* This routine is invoked to read an FCF record indicated by @fcf_index to
* determine whether it's eligible for FLOGI roundrobin failover list.
*
* Return 0 if the mailbox command is submitted successfully, none 0
* otherwise.
**/
int
lpfc_sli4_read_fcf_rec(struct lpfc_hba *phba, uint16_t fcf_index)
{
int rc = 0, error;
LPFC_MBOXQ_t *mboxq;
mboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq) {
lpfc_printf_log(phba, KERN_ERR, LOG_FIP | LOG_INIT,
"2758 Failed to allocate mbox for "
"READ_FCF cmd\n");
error = -ENOMEM;
goto fail_fcf_read;
}
/* Construct the read FCF record mailbox command */
rc = lpfc_sli4_mbx_read_fcf_rec(phba, mboxq, fcf_index);
if (rc) {
error = -EINVAL;
goto fail_fcf_read;
}
/* Issue the mailbox command asynchronously */
mboxq->vport = phba->pport;
mboxq->mbox_cmpl = lpfc_mbx_cmpl_read_fcf_rec;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_NOWAIT);
if (rc == MBX_NOT_FINISHED)
error = -EIO;
else
error = 0;
fail_fcf_read:
if (error && mboxq)
lpfc_sli4_mbox_cmd_free(phba, mboxq);
return error;
}
/**
* lpfc_check_next_fcf_pri_level
* phba pointer to the lpfc_hba struct for this port.
* This routine is called from the lpfc_sli4_fcf_rr_next_index_get
* routine when the rr_bmask is empty. The FCF indecies are put into the
* rr_bmask based on their priority level. Starting from the highest priority
* to the lowest. The most likely FCF candidate will be in the highest
* priority group. When this routine is called it searches the fcf_pri list for
* next lowest priority group and repopulates the rr_bmask with only those
* fcf_indexes.
* returns:
* 1=success 0=failure
**/
static int
lpfc_check_next_fcf_pri_level(struct lpfc_hba *phba)
{
uint16_t next_fcf_pri;
uint16_t last_index;
struct lpfc_fcf_pri *fcf_pri;
int rc;
int ret = 0;
last_index = find_first_bit(phba->fcf.fcf_rr_bmask,
LPFC_SLI4_FCF_TBL_INDX_MAX);
lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
"3060 Last IDX %d\n", last_index);
/* Verify the priority list has 2 or more entries */
spin_lock_irq(&phba->hbalock);
if (list_empty(&phba->fcf.fcf_pri_list) ||
list_is_singular(&phba->fcf.fcf_pri_list)) {
spin_unlock_irq(&phba->hbalock);
lpfc_printf_log(phba, KERN_ERR, LOG_FIP,
"3061 Last IDX %d\n", last_index);
return 0; /* Empty rr list */
}
spin_unlock_irq(&phba->hbalock);
next_fcf_pri = 0;
/*
* Clear the rr_bmask and set all of the bits that are at this
* priority.
*/
memset(phba->fcf.fcf_rr_bmask, 0,
sizeof(*phba->fcf.fcf_rr_bmask));
spin_lock_irq(&phba->hbalock);
list_for_each_entry(fcf_pri, &phba->fcf.fcf_pri_list, list) {
if (fcf_pri->fcf_rec.flag & LPFC_FCF_FLOGI_FAILED)
continue;
/*
* the 1st priority that has not FLOGI failed
* will be the highest.
*/
if (!next_fcf_pri)
next_fcf_pri = fcf_pri->fcf_rec.priority;
spin_unlock_irq(&phba->hbalock);
if (fcf_pri->fcf_rec.priority == next_fcf_pri) {
rc = lpfc_sli4_fcf_rr_index_set(phba,
fcf_pri->fcf_rec.fcf_index);
if (rc)
return 0;
}
spin_lock_irq(&phba->hbalock);
}
/*
* if next_fcf_pri was not set above and the list is not empty then
* we have failed flogis on all of them. So reset flogi failed
* and start at the beginning.
*/
if (!next_fcf_pri && !list_empty(&phba->fcf.fcf_pri_list)) {
list_for_each_entry(fcf_pri, &phba->fcf.fcf_pri_list, list) {
fcf_pri->fcf_rec.flag &= ~LPFC_FCF_FLOGI_FAILED;
/*
* the 1st priority that has not FLOGI failed
* will be the highest.
*/
if (!next_fcf_pri)
next_fcf_pri = fcf_pri->fcf_rec.priority;
spin_unlock_irq(&phba->hbalock);
if (fcf_pri->fcf_rec.priority == next_fcf_pri) {
rc = lpfc_sli4_fcf_rr_index_set(phba,
fcf_pri->fcf_rec.fcf_index);
if (rc)
return 0;
}
spin_lock_irq(&phba->hbalock);
}
} else
ret = 1;
spin_unlock_irq(&phba->hbalock);
return ret;
}
/**
* lpfc_sli4_fcf_rr_next_index_get - Get next eligible fcf record index
* @phba: pointer to lpfc hba data structure.
*
* This routine is to get the next eligible FCF record index in a round
* robin fashion. If the next eligible FCF record index equals to the
* initial roundrobin FCF record index, LPFC_FCOE_FCF_NEXT_NONE (0xFFFF)
* shall be returned, otherwise, the next eligible FCF record's index
* shall be returned.
**/
uint16_t
lpfc_sli4_fcf_rr_next_index_get(struct lpfc_hba *phba)
{
uint16_t next_fcf_index;
initial_priority:
/* Search start from next bit of currently registered FCF index */
next_fcf_index = phba->fcf.current_rec.fcf_indx;
next_priority:
/* Determine the next fcf index to check */
next_fcf_index = (next_fcf_index + 1) % LPFC_SLI4_FCF_TBL_INDX_MAX;
next_fcf_index = find_next_bit(phba->fcf.fcf_rr_bmask,
LPFC_SLI4_FCF_TBL_INDX_MAX,
next_fcf_index);
/* Wrap around condition on phba->fcf.fcf_rr_bmask */
if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX) {
/*
* If we have wrapped then we need to clear the bits that
* have been tested so that we can detect when we should
* change the priority level.
*/
next_fcf_index = find_next_bit(phba->fcf.fcf_rr_bmask,
LPFC_SLI4_FCF_TBL_INDX_MAX, 0);
}
/* Check roundrobin failover list empty condition */
if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX ||
next_fcf_index == phba->fcf.current_rec.fcf_indx) {
/*
* If next fcf index is not found check if there are lower
* Priority level fcf's in the fcf_priority list.
* Set up the rr_bmask with all of the avaiable fcf bits
* at that level and continue the selection process.
*/
if (lpfc_check_next_fcf_pri_level(phba))
goto initial_priority;
lpfc_printf_log(phba, KERN_WARNING, LOG_FIP,
"2844 No roundrobin failover FCF available\n");
if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX)
return LPFC_FCOE_FCF_NEXT_NONE;
else {
lpfc_printf_log(phba, KERN_WARNING, LOG_FIP,
"3063 Only FCF available idx %d, flag %x\n",
next_fcf_index,
phba->fcf.fcf_pri[next_fcf_index].fcf_rec.flag);
return next_fcf_index;
}
}
if (next_fcf_index < LPFC_SLI4_FCF_TBL_INDX_MAX &&
phba->fcf.fcf_pri[next_fcf_index].fcf_rec.flag &
LPFC_FCF_FLOGI_FAILED) {
if (list_is_singular(&phba->fcf.fcf_pri_list))
return LPFC_FCOE_FCF_NEXT_NONE;
goto next_priority;
}
lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
"2845 Get next roundrobin failover FCF (x%x)\n",
next_fcf_index);
return next_fcf_index;
}
/**
* lpfc_sli4_fcf_rr_index_set - Set bmask with eligible fcf record index
* @phba: pointer to lpfc hba data structure.
*
* This routine sets the FCF record index in to the eligible bmask for
* roundrobin failover search. It checks to make sure that the index
* does not go beyond the range of the driver allocated bmask dimension
* before setting the bit.
*
* Returns 0 if the index bit successfully set, otherwise, it returns
* -EINVAL.
**/
int
lpfc_sli4_fcf_rr_index_set(struct lpfc_hba *phba, uint16_t fcf_index)
{
if (fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX) {
lpfc_printf_log(phba, KERN_ERR, LOG_FIP,
"2610 FCF (x%x) reached driver's book "
"keeping dimension:x%x\n",
fcf_index, LPFC_SLI4_FCF_TBL_INDX_MAX);
return -EINVAL;
}
/* Set the eligible FCF record index bmask */
set_bit(fcf_index, phba->fcf.fcf_rr_bmask);
lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
"2790 Set FCF (x%x) to roundrobin FCF failover "
"bmask\n", fcf_index);
return 0;
}
/**
* lpfc_sli4_fcf_rr_index_clear - Clear bmask from eligible fcf record index
* @phba: pointer to lpfc hba data structure.
*
* This routine clears the FCF record index from the eligible bmask for
* roundrobin failover search. It checks to make sure that the index
* does not go beyond the range of the driver allocated bmask dimension
* before clearing the bit.
**/
void
lpfc_sli4_fcf_rr_index_clear(struct lpfc_hba *phba, uint16_t fcf_index)
{
struct lpfc_fcf_pri *fcf_pri, *fcf_pri_next;
if (fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX) {
lpfc_printf_log(phba, KERN_ERR, LOG_FIP,
"2762 FCF (x%x) reached driver's book "
"keeping dimension:x%x\n",
fcf_index, LPFC_SLI4_FCF_TBL_INDX_MAX);
return;
}
/* Clear the eligible FCF record index bmask */
spin_lock_irq(&phba->hbalock);
list_for_each_entry_safe(fcf_pri, fcf_pri_next, &phba->fcf.fcf_pri_list,
list) {
if (fcf_pri->fcf_rec.fcf_index == fcf_index) {
list_del_init(&fcf_pri->list);
break;
}
}
spin_unlock_irq(&phba->hbalock);
clear_bit(fcf_index, phba->fcf.fcf_rr_bmask);
lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
"2791 Clear FCF (x%x) from roundrobin failover "
"bmask\n", fcf_index);
}
/**
* lpfc_mbx_cmpl_redisc_fcf_table - completion routine for rediscover FCF table
* @phba: pointer to lpfc hba data structure.
*
* This routine is the completion routine for the rediscover FCF table mailbox
* command. If the mailbox command returned failure, it will try to stop the
* FCF rediscover wait timer.
**/
static void
lpfc_mbx_cmpl_redisc_fcf_table(struct lpfc_hba *phba, LPFC_MBOXQ_t *mbox)
{
struct lpfc_mbx_redisc_fcf_tbl *redisc_fcf;
uint32_t shdr_status, shdr_add_status;
redisc_fcf = &mbox->u.mqe.un.redisc_fcf_tbl;
shdr_status = bf_get(lpfc_mbox_hdr_status,
&redisc_fcf->header.cfg_shdr.response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status,
&redisc_fcf->header.cfg_shdr.response);
if (shdr_status || shdr_add_status) {
lpfc_printf_log(phba, KERN_ERR, LOG_FIP,
"2746 Requesting for FCF rediscovery failed "
"status x%x add_status x%x\n",
shdr_status, shdr_add_status);
if (phba->fcf.fcf_flag & FCF_ACVL_DISC) {
spin_lock_irq(&phba->hbalock);
phba->fcf.fcf_flag &= ~FCF_ACVL_DISC;
spin_unlock_irq(&phba->hbalock);
/*
* CVL event triggered FCF rediscover request failed,
* last resort to re-try current registered FCF entry.
*/
lpfc_retry_pport_discovery(phba);
} else {
spin_lock_irq(&phba->hbalock);
phba->fcf.fcf_flag &= ~FCF_DEAD_DISC;
spin_unlock_irq(&phba->hbalock);
/*
* DEAD FCF event triggered FCF rediscover request
* failed, last resort to fail over as a link down
* to FCF registration.
*/
lpfc_sli4_fcf_dead_failthrough(phba);
}
} else {
lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
"2775 Start FCF rediscover quiescent timer\n");
/*
* Start FCF rediscovery wait timer for pending FCF
* before rescan FCF record table.
*/
lpfc_fcf_redisc_wait_start_timer(phba);
}
mempool_free(mbox, phba->mbox_mem_pool);
}
/**
* lpfc_sli4_redisc_fcf_table - Request to rediscover entire FCF table by port.
* @phba: pointer to lpfc hba data structure.
*
* This routine is invoked to request for rediscovery of the entire FCF table
* by the port.
**/
int
lpfc_sli4_redisc_fcf_table(struct lpfc_hba *phba)
{
LPFC_MBOXQ_t *mbox;
struct lpfc_mbx_redisc_fcf_tbl *redisc_fcf;
int rc, length;
/* Cancel retry delay timers to all vports before FCF rediscover */
lpfc_cancel_all_vport_retry_delay_timer(phba);
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox) {
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2745 Failed to allocate mbox for "
"requesting FCF rediscover.\n");
return -ENOMEM;
}
length = (sizeof(struct lpfc_mbx_redisc_fcf_tbl) -
sizeof(struct lpfc_sli4_cfg_mhdr));
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
LPFC_MBOX_OPCODE_FCOE_REDISCOVER_FCF,
length, LPFC_SLI4_MBX_EMBED);
redisc_fcf = &mbox->u.mqe.un.redisc_fcf_tbl;
/* Set count to 0 for invalidating the entire FCF database */
bf_set(lpfc_mbx_redisc_fcf_count, redisc_fcf, 0);
/* Issue the mailbox command asynchronously */
mbox->vport = phba->pport;
mbox->mbox_cmpl = lpfc_mbx_cmpl_redisc_fcf_table;
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_NOWAIT);
if (rc == MBX_NOT_FINISHED) {
mempool_free(mbox, phba->mbox_mem_pool);
return -EIO;
}
return 0;
}
/**
* lpfc_sli4_fcf_dead_failthrough - Failthrough routine to fcf dead event
* @phba: pointer to lpfc hba data structure.
*
* This function is the failover routine as a last resort to the FCF DEAD
* event when driver failed to perform fast FCF failover.
**/
void
lpfc_sli4_fcf_dead_failthrough(struct lpfc_hba *phba)
{
uint32_t link_state;
/*
* Last resort as FCF DEAD event failover will treat this as
* a link down, but save the link state because we don't want
* it to be changed to Link Down unless it is already down.
*/
link_state = phba->link_state;
lpfc_linkdown(phba);
phba->link_state = link_state;
/* Unregister FCF if no devices connected to it */
lpfc_unregister_unused_fcf(phba);
}
/**
* lpfc_sli_get_config_region23 - Get sli3 port region 23 data.
* @phba: pointer to lpfc hba data structure.
* @rgn23_data: pointer to configure region 23 data.
*
* This function gets SLI3 port configure region 23 data through memory dump
* mailbox command. When it successfully retrieves data, the size of the data
* will be returned, otherwise, 0 will be returned.
**/
static uint32_t
lpfc_sli_get_config_region23(struct lpfc_hba *phba, char *rgn23_data)
{
LPFC_MBOXQ_t *pmb = NULL;
MAILBOX_t *mb;
uint32_t offset = 0;
int rc;
if (!rgn23_data)
return 0;
pmb = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!pmb) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2600 failed to allocate mailbox memory\n");
return 0;
}
mb = &pmb->u.mb;
do {
lpfc_dump_mem(phba, pmb, offset, DMP_REGION_23);
rc = lpfc_sli_issue_mbox(phba, pmb, MBX_POLL);
if (rc != MBX_SUCCESS) {
lpfc_printf_log(phba, KERN_INFO, LOG_INIT,
"2601 failed to read config "
"region 23, rc 0x%x Status 0x%x\n",
rc, mb->mbxStatus);
mb->un.varDmp.word_cnt = 0;
}
/*
* dump mem may return a zero when finished or we got a
* mailbox error, either way we are done.
*/
if (mb->un.varDmp.word_cnt == 0)
break;
if (mb->un.varDmp.word_cnt > DMP_RGN23_SIZE - offset)
mb->un.varDmp.word_cnt = DMP_RGN23_SIZE - offset;
lpfc_sli_pcimem_bcopy(((uint8_t *)mb) + DMP_RSP_OFFSET,
rgn23_data + offset,
mb->un.varDmp.word_cnt);
offset += mb->un.varDmp.word_cnt;
} while (mb->un.varDmp.word_cnt && offset < DMP_RGN23_SIZE);
mempool_free(pmb, phba->mbox_mem_pool);
return offset;
}
/**
* lpfc_sli4_get_config_region23 - Get sli4 port region 23 data.
* @phba: pointer to lpfc hba data structure.
* @rgn23_data: pointer to configure region 23 data.
*
* This function gets SLI4 port configure region 23 data through memory dump
* mailbox command. When it successfully retrieves data, the size of the data
* will be returned, otherwise, 0 will be returned.
**/
static uint32_t
lpfc_sli4_get_config_region23(struct lpfc_hba *phba, char *rgn23_data)
{
LPFC_MBOXQ_t *mboxq = NULL;
struct lpfc_dmabuf *mp = NULL;
struct lpfc_mqe *mqe;
uint32_t data_length = 0;
int rc;
if (!rgn23_data)
return 0;
mboxq = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mboxq) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3105 failed to allocate mailbox memory\n");
return 0;
}
if (lpfc_sli4_dump_cfg_rg23(phba, mboxq))
goto out;
mqe = &mboxq->u.mqe;
mp = (struct lpfc_dmabuf *) mboxq->context1;
rc = lpfc_sli_issue_mbox(phba, mboxq, MBX_POLL);
if (rc)
goto out;
data_length = mqe->un.mb_words[5];
if (data_length == 0)
goto out;
if (data_length > DMP_RGN23_SIZE) {
data_length = 0;
goto out;
}
lpfc_sli_pcimem_bcopy((char *)mp->virt, rgn23_data, data_length);
out:
mempool_free(mboxq, phba->mbox_mem_pool);
if (mp) {
lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
}
return data_length;
}
/**
* lpfc_sli_read_link_ste - Read region 23 to decide if link is disabled.
* @phba: pointer to lpfc hba data structure.
*
* This function read region 23 and parse TLV for port status to
* decide if the user disaled the port. If the TLV indicates the
* port is disabled, the hba_flag is set accordingly.
**/
void
lpfc_sli_read_link_ste(struct lpfc_hba *phba)
{
uint8_t *rgn23_data = NULL;
uint32_t if_type, data_size, sub_tlv_len, tlv_offset;
uint32_t offset = 0;
/* Get adapter Region 23 data */
rgn23_data = kzalloc(DMP_RGN23_SIZE, GFP_KERNEL);
if (!rgn23_data)
goto out;
if (phba->sli_rev < LPFC_SLI_REV4)
data_size = lpfc_sli_get_config_region23(phba, rgn23_data);
else {
if_type = bf_get(lpfc_sli_intf_if_type,
&phba->sli4_hba.sli_intf);
if (if_type == LPFC_SLI_INTF_IF_TYPE_0)
goto out;
data_size = lpfc_sli4_get_config_region23(phba, rgn23_data);
}
if (!data_size)
goto out;
/* Check the region signature first */
if (memcmp(&rgn23_data[offset], LPFC_REGION23_SIGNATURE, 4)) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2619 Config region 23 has bad signature\n");
goto out;
}
offset += 4;
/* Check the data structure version */
if (rgn23_data[offset] != LPFC_REGION23_VERSION) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"2620 Config region 23 has bad version\n");
goto out;
}
offset += 4;
/* Parse TLV entries in the region */
while (offset < data_size) {
if (rgn23_data[offset] == LPFC_REGION23_LAST_REC)
break;
/*
* If the TLV is not driver specific TLV or driver id is
* not linux driver id, skip the record.
*/
if ((rgn23_data[offset] != DRIVER_SPECIFIC_TYPE) ||
(rgn23_data[offset + 2] != LINUX_DRIVER_ID) ||
(rgn23_data[offset + 3] != 0)) {
offset += rgn23_data[offset + 1] * 4 + 4;
continue;
}
/* Driver found a driver specific TLV in the config region */
sub_tlv_len = rgn23_data[offset + 1] * 4;
offset += 4;
tlv_offset = 0;
/*
* Search for configured port state sub-TLV.
*/
while ((offset < data_size) &&
(tlv_offset < sub_tlv_len)) {
if (rgn23_data[offset] == LPFC_REGION23_LAST_REC) {
offset += 4;
tlv_offset += 4;
break;
}
if (rgn23_data[offset] != PORT_STE_TYPE) {
offset += rgn23_data[offset + 1] * 4 + 4;
tlv_offset += rgn23_data[offset + 1] * 4 + 4;
continue;
}
/* This HBA contains PORT_STE configured */
if (!rgn23_data[offset + 2])
phba->hba_flag |= LINK_DISABLED;
goto out;
}
}
out:
kfree(rgn23_data);
return;
}
/**
* lpfc_wr_object - write an object to the firmware
* @phba: HBA structure that indicates port to create a queue on.
* @dmabuf_list: list of dmabufs to write to the port.
* @size: the total byte value of the objects to write to the port.
* @offset: the current offset to be used to start the transfer.
*
* This routine will create a wr_object mailbox command to send to the port.
* the mailbox command will be constructed using the dma buffers described in
* @dmabuf_list to create a list of BDEs. This routine will fill in as many
* BDEs that the imbedded mailbox can support. The @offset variable will be
* used to indicate the starting offset of the transfer and will also return
* the offset after the write object mailbox has completed. @size is used to
* determine the end of the object and whether the eof bit should be set.
*
* Return 0 is successful and offset will contain the the new offset to use
* for the next write.
* Return negative value for error cases.
**/
int
lpfc_wr_object(struct lpfc_hba *phba, struct list_head *dmabuf_list,
uint32_t size, uint32_t *offset)
{
struct lpfc_mbx_wr_object *wr_object;
LPFC_MBOXQ_t *mbox;
int rc = 0, i = 0;
uint32_t shdr_status, shdr_add_status;
uint32_t mbox_tmo;
union lpfc_sli4_cfg_shdr *shdr;
struct lpfc_dmabuf *dmabuf;
uint32_t written = 0;
mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
if (!mbox)
return -ENOMEM;
lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_COMMON,
LPFC_MBOX_OPCODE_WRITE_OBJECT,
sizeof(struct lpfc_mbx_wr_object) -
sizeof(struct lpfc_sli4_cfg_mhdr), LPFC_SLI4_MBX_EMBED);
wr_object = (struct lpfc_mbx_wr_object *)&mbox->u.mqe.un.wr_object;
wr_object->u.request.write_offset = *offset;
sprintf((uint8_t *)wr_object->u.request.object_name, "/");
wr_object->u.request.object_name[0] =
cpu_to_le32(wr_object->u.request.object_name[0]);
bf_set(lpfc_wr_object_eof, &wr_object->u.request, 0);
list_for_each_entry(dmabuf, dmabuf_list, list) {
if (i >= LPFC_MBX_WR_CONFIG_MAX_BDE || written >= size)
break;
wr_object->u.request.bde[i].addrLow = putPaddrLow(dmabuf->phys);
wr_object->u.request.bde[i].addrHigh =
putPaddrHigh(dmabuf->phys);
if (written + SLI4_PAGE_SIZE >= size) {
wr_object->u.request.bde[i].tus.f.bdeSize =
(size - written);
written += (size - written);
bf_set(lpfc_wr_object_eof, &wr_object->u.request, 1);
} else {
wr_object->u.request.bde[i].tus.f.bdeSize =
SLI4_PAGE_SIZE;
written += SLI4_PAGE_SIZE;
}
i++;
}
wr_object->u.request.bde_count = i;
bf_set(lpfc_wr_object_write_length, &wr_object->u.request, written);
if (!phba->sli4_hba.intr_enable)
rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
else {
mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
}
/* The IOCTL status is embedded in the mailbox subheader. */
shdr = (union lpfc_sli4_cfg_shdr *) &wr_object->header.cfg_shdr;
shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
if (rc != MBX_TIMEOUT)
mempool_free(mbox, phba->mbox_mem_pool);
if (shdr_status || shdr_add_status || rc) {
lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
"3025 Write Object mailbox failed with "
"status x%x add_status x%x, mbx status x%x\n",
shdr_status, shdr_add_status, rc);
rc = -ENXIO;
*offset = shdr_add_status;
} else
*offset += wr_object->u.response.actual_write_length;
return rc;
}
/**
* lpfc_cleanup_pending_mbox - Free up vport discovery mailbox commands.
* @vport: pointer to vport data structure.
*
* This function iterate through the mailboxq and clean up all REG_LOGIN
* and REG_VPI mailbox commands associated with the vport. This function
* is called when driver want to restart discovery of the vport due to
* a Clear Virtual Link event.
**/
void
lpfc_cleanup_pending_mbox(struct lpfc_vport *vport)
{
struct lpfc_hba *phba = vport->phba;
LPFC_MBOXQ_t *mb, *nextmb;
struct lpfc_dmabuf *mp;
struct lpfc_nodelist *ndlp;
struct lpfc_nodelist *act_mbx_ndlp = NULL;
struct Scsi_Host *shost = lpfc_shost_from_vport(vport);
LIST_HEAD(mbox_cmd_list);
uint8_t restart_loop;
/* Clean up internally queued mailbox commands with the vport */
spin_lock_irq(&phba->hbalock);
list_for_each_entry_safe(mb, nextmb, &phba->sli.mboxq, list) {
if (mb->vport != vport)
continue;
if ((mb->u.mb.mbxCommand != MBX_REG_LOGIN64) &&
(mb->u.mb.mbxCommand != MBX_REG_VPI))
continue;
list_del(&mb->list);
list_add_tail(&mb->list, &mbox_cmd_list);
}
/* Clean up active mailbox command with the vport */
mb = phba->sli.mbox_active;
if (mb && (mb->vport == vport)) {
if ((mb->u.mb.mbxCommand == MBX_REG_LOGIN64) ||
(mb->u.mb.mbxCommand == MBX_REG_VPI))
mb->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
if (mb->u.mb.mbxCommand == MBX_REG_LOGIN64) {
act_mbx_ndlp = (struct lpfc_nodelist *)mb->context2;
/* Put reference count for delayed processing */
act_mbx_ndlp = lpfc_nlp_get(act_mbx_ndlp);
/* Unregister the RPI when mailbox complete */
mb->mbox_flag |= LPFC_MBX_IMED_UNREG;
}
}
/* Cleanup any mailbox completions which are not yet processed */
do {
restart_loop = 0;
list_for_each_entry(mb, &phba->sli.mboxq_cmpl, list) {
/*
* If this mailox is already processed or it is
* for another vport ignore it.
*/
if ((mb->vport != vport) ||
(mb->mbox_flag & LPFC_MBX_IMED_UNREG))
continue;
if ((mb->u.mb.mbxCommand != MBX_REG_LOGIN64) &&
(mb->u.mb.mbxCommand != MBX_REG_VPI))
continue;
mb->mbox_cmpl = lpfc_sli_def_mbox_cmpl;
if (mb->u.mb.mbxCommand == MBX_REG_LOGIN64) {
ndlp = (struct lpfc_nodelist *)mb->context2;
/* Unregister the RPI when mailbox complete */
mb->mbox_flag |= LPFC_MBX_IMED_UNREG;
restart_loop = 1;
spin_unlock_irq(&phba->hbalock);
spin_lock(shost->host_lock);
ndlp->nlp_flag &= ~NLP_IGNR_REG_CMPL;
spin_unlock(shost->host_lock);
spin_lock_irq(&phba->hbalock);
break;
}
}
} while (restart_loop);
spin_unlock_irq(&phba->hbalock);
/* Release the cleaned-up mailbox commands */
while (!list_empty(&mbox_cmd_list)) {
list_remove_head(&mbox_cmd_list, mb, LPFC_MBOXQ_t, list);
if (mb->u.mb.mbxCommand == MBX_REG_LOGIN64) {
mp = (struct lpfc_dmabuf *) (mb->context1);
if (mp) {
__lpfc_mbuf_free(phba, mp->virt, mp->phys);
kfree(mp);
}
ndlp = (struct lpfc_nodelist *) mb->context2;
mb->context2 = NULL;
if (ndlp) {
spin_lock(shost->host_lock);
ndlp->nlp_flag &= ~NLP_IGNR_REG_CMPL;
spin_unlock(shost->host_lock);
lpfc_nlp_put(ndlp);
}
}
mempool_free(mb, phba->mbox_mem_pool);
}
/* Release the ndlp with the cleaned-up active mailbox command */
if (act_mbx_ndlp) {
spin_lock(shost->host_lock);
act_mbx_ndlp->nlp_flag &= ~NLP_IGNR_REG_CMPL;
spin_unlock(shost->host_lock);
lpfc_nlp_put(act_mbx_ndlp);
}
}
/**
* lpfc_drain_txq - Drain the txq
* @phba: Pointer to HBA context object.
*
* This function attempt to submit IOCBs on the txq
* to the adapter. For SLI4 adapters, the txq contains
* ELS IOCBs that have been deferred because the there
* are no SGLs. This congestion can occur with large
* vport counts during node discovery.
**/
uint32_t
lpfc_drain_txq(struct lpfc_hba *phba)
{
LIST_HEAD(completions);
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_sli_ring *pring;
struct lpfc_iocbq *piocbq = NULL;
unsigned long iflags = 0;
char *fail_msg = NULL;
struct lpfc_sglq *sglq;
union lpfc_wqe128 wqe;
uint32_t txq_cnt = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
pring = lpfc_phba_elsring(phba);
if (unlikely(!pring))
return 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_lock_irqsave(&pring->ring_lock, iflags);
list_for_each_entry(piocbq, &pring->txq, list) {
txq_cnt++;
}
if (txq_cnt > pring->txq_max)
pring->txq_max = txq_cnt;
spin_unlock_irqrestore(&pring->ring_lock, iflags);
while (!list_empty(&pring->txq)) {
spin_lock_irqsave(&pring->ring_lock, iflags);
piocbq = lpfc_sli_ringtx_get(phba, pring);
if (!piocbq) {
spin_unlock_irqrestore(&pring->ring_lock, iflags);
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2823 txq empty and txq_cnt is %d\n ",
txq_cnt);
break;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
sglq = __lpfc_sli_get_els_sglq(phba, piocbq);
if (!sglq) {
__lpfc_sli_ringtx_put(phba, pring, piocbq);
spin_unlock_irqrestore(&pring->ring_lock, iflags);
break;
}
txq_cnt--;
/* The xri and iocb resources secured,
* attempt to issue request
*/
piocbq->sli4_lxritag = sglq->sli4_lxritag;
piocbq->sli4_xritag = sglq->sli4_xritag;
if (NO_XRI == lpfc_sli4_bpl2sgl(phba, piocbq, sglq))
fail_msg = "to convert bpl to sgl";
else if (lpfc_sli4_iocb2wqe(phba, piocbq, &wqe))
fail_msg = "to convert iocb to wqe";
else if (lpfc_sli4_wq_put(phba->sli4_hba.els_wq, &wqe))
fail_msg = " - Wq is full";
else
lpfc_sli_ringtxcmpl_put(phba, pring, piocbq);
if (fail_msg) {
/* Failed means we can't issue and need to cancel */
lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
"2822 IOCB failed %s iotag 0x%x "
"xri 0x%x\n",
fail_msg,
piocbq->iotag, piocbq->sli4_xritag);
list_add_tail(&piocbq->list, &completions);
}
spin_unlock_irqrestore(&pring->ring_lock, iflags);
}
/* Cancel all the IOCBs that cannot be issued */
lpfc_sli_cancel_iocbs(phba, &completions, IOSTAT_LOCAL_REJECT,
IOERR_SLI_ABORTED);
return txq_cnt;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/**
* lpfc_wqe_bpl2sgl - Convert the bpl/bde to a sgl.
* @phba: Pointer to HBA context object.
* @pwqe: Pointer to command WQE.
* @sglq: Pointer to the scatter gather queue object.
*
* This routine converts the bpl or bde that is in the WQE
* to a sgl list for the sli4 hardware. The physical address
* of the bpl/bde is converted back to a virtual address.
* If the WQE contains a BPL then the list of BDE's is
* converted to sli4_sge's. If the WQE contains a single
* BDE then it is converted to a single sli_sge.
* The WQE is still in cpu endianness so the contents of
* the bpl can be used without byte swapping.
*
* Returns valid XRI = Success, NO_XRI = Failure.
*/
static uint16_t
lpfc_wqe_bpl2sgl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeq,
struct lpfc_sglq *sglq)
{
uint16_t xritag = NO_XRI;
struct ulp_bde64 *bpl = NULL;
struct ulp_bde64 bde;
struct sli4_sge *sgl = NULL;
struct lpfc_dmabuf *dmabuf;
union lpfc_wqe128 *wqe;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
int numBdes = 0;
int i = 0;
uint32_t offset = 0; /* accumulated offset in the sg request list */
int inbound = 0; /* number of sg reply entries inbound from firmware */
uint32_t cmd;
if (!pwqeq || !sglq)
return xritag;
sgl = (struct sli4_sge *)sglq->sgl;
wqe = &pwqeq->wqe;
pwqeq->iocb.ulpIoTag = pwqeq->iotag;
cmd = bf_get(wqe_cmnd, &wqe->generic.wqe_com);
if (cmd == CMD_XMIT_BLS_RSP64_WQE)
return sglq->sli4_xritag;
numBdes = pwqeq->rsvd2;
if (numBdes) {
/* The addrHigh and addrLow fields within the WQE
* have not been byteswapped yet so there is no
* need to swap them back.
*/
if (pwqeq->context3)
dmabuf = (struct lpfc_dmabuf *)pwqeq->context3;
else
return xritag;
bpl = (struct ulp_bde64 *)dmabuf->virt;
if (!bpl)
return xritag;
for (i = 0; i < numBdes; i++) {
/* Should already be byte swapped. */
sgl->addr_hi = bpl->addrHigh;
sgl->addr_lo = bpl->addrLow;
sgl->word2 = le32_to_cpu(sgl->word2);
if ((i+1) == numBdes)
bf_set(lpfc_sli4_sge_last, sgl, 1);
else
bf_set(lpfc_sli4_sge_last, sgl, 0);
/* swap the size field back to the cpu so we
* can assign it to the sgl.
*/
bde.tus.w = le32_to_cpu(bpl->tus.w);
sgl->sge_len = cpu_to_le32(bde.tus.f.bdeSize);
/* The offsets in the sgl need to be accumulated
* separately for the request and reply lists.
* The request is always first, the reply follows.
*/
switch (cmd) {
case CMD_GEN_REQUEST64_WQE:
/* add up the reply sg entries */
if (bpl->tus.f.bdeFlags == BUFF_TYPE_BDE_64I)
inbound++;
/* first inbound? reset the offset */
if (inbound == 1)
offset = 0;
bf_set(lpfc_sli4_sge_offset, sgl, offset);
bf_set(lpfc_sli4_sge_type, sgl,
LPFC_SGE_TYPE_DATA);
offset += bde.tus.f.bdeSize;
break;
case CMD_FCP_TRSP64_WQE:
bf_set(lpfc_sli4_sge_offset, sgl, 0);
bf_set(lpfc_sli4_sge_type, sgl,
LPFC_SGE_TYPE_DATA);
break;
case CMD_FCP_TSEND64_WQE:
case CMD_FCP_TRECEIVE64_WQE:
bf_set(lpfc_sli4_sge_type, sgl,
bpl->tus.f.bdeFlags);
if (i < 3)
offset = 0;
else
offset += bde.tus.f.bdeSize;
bf_set(lpfc_sli4_sge_offset, sgl, offset);
break;
}
sgl->word2 = cpu_to_le32(sgl->word2);
bpl++;
sgl++;
}
} else if (wqe->gen_req.bde.tus.f.bdeFlags == BUFF_TYPE_BDE_64) {
/* The addrHigh and addrLow fields of the BDE have not
* been byteswapped yet so they need to be swapped
* before putting them in the sgl.
*/
sgl->addr_hi = cpu_to_le32(wqe->gen_req.bde.addrHigh);
sgl->addr_lo = cpu_to_le32(wqe->gen_req.bde.addrLow);
sgl->word2 = le32_to_cpu(sgl->word2);
bf_set(lpfc_sli4_sge_last, sgl, 1);
sgl->word2 = cpu_to_le32(sgl->word2);
sgl->sge_len = cpu_to_le32(wqe->gen_req.bde.tus.f.bdeSize);
}
return sglq->sli4_xritag;
}
/**
* lpfc_sli4_issue_wqe - Issue an SLI4 Work Queue Entry (WQE)
* @phba: Pointer to HBA context object.
* @ring_number: Base sli ring number
* @pwqe: Pointer to command WQE.
**/
int
lpfc_sli4_issue_wqe(struct lpfc_hba *phba, uint32_t ring_number,
struct lpfc_iocbq *pwqe)
{
union lpfc_wqe128 *wqe = &pwqe->wqe;
struct lpfc_nvmet_rcv_ctx *ctxp;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
struct lpfc_queue *wq;
struct lpfc_sglq *sglq;
struct lpfc_sli_ring *pring;
unsigned long iflags;
uint32_t ret = 0;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
/* NVME_LS and NVME_LS ABTS requests. */
if (pwqe->iocb_flag & LPFC_IO_NVME_LS) {
pring = phba->sli4_hba.nvmels_wq->pring;
spin_lock_irqsave(&pring->ring_lock, iflags);
sglq = __lpfc_sli_get_els_sglq(phba, pwqe);
if (!sglq) {
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return WQE_BUSY;
}
pwqe->sli4_lxritag = sglq->sli4_lxritag;
pwqe->sli4_xritag = sglq->sli4_xritag;
if (lpfc_wqe_bpl2sgl(phba, pwqe, sglq) == NO_XRI) {
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return WQE_ERROR;
}
bf_set(wqe_xri_tag, &pwqe->wqe.xmit_bls_rsp.wqe_com,
pwqe->sli4_xritag);
ret = lpfc_sli4_wq_put(phba->sli4_hba.nvmels_wq, wqe);
if (ret) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return ret;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
lpfc_sli_ringtxcmpl_put(phba, pring, pwqe);
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return 0;
}
/* NVME_FCREQ and NVME_ABTS requests */
if (pwqe->iocb_flag & LPFC_IO_NVME) {
/* Get the IO distribution (hba_wqidx) for WQ assignment. */
pring = phba->sli4_hba.nvme_wq[pwqe->hba_wqidx]->pring;
spin_lock_irqsave(&pring->ring_lock, iflags);
wq = phba->sli4_hba.nvme_wq[pwqe->hba_wqidx];
bf_set(wqe_cqid, &wqe->generic.wqe_com,
phba->sli4_hba.nvme_cq[pwqe->hba_wqidx]->queue_id);
ret = lpfc_sli4_wq_put(wq, wqe);
if (ret) {
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return ret;
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
}
lpfc_sli_ringtxcmpl_put(phba, pring, pwqe);
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return 0;
}
/* NVMET requests */
if (pwqe->iocb_flag & LPFC_IO_NVMET) {
/* Get the IO distribution (hba_wqidx) for WQ assignment. */
pring = phba->sli4_hba.nvme_wq[pwqe->hba_wqidx]->pring;
spin_lock_irqsave(&pring->ring_lock, iflags);
ctxp = pwqe->context2;
sglq = ctxp->ctxbuf->sglq;
if (pwqe->sli4_xritag == NO_XRI) {
pwqe->sli4_lxritag = sglq->sli4_lxritag;
pwqe->sli4_xritag = sglq->sli4_xritag;
}
bf_set(wqe_xri_tag, &pwqe->wqe.xmit_bls_rsp.wqe_com,
pwqe->sli4_xritag);
wq = phba->sli4_hba.nvme_wq[pwqe->hba_wqidx];
bf_set(wqe_cqid, &wqe->generic.wqe_com,
phba->sli4_hba.nvme_cq[pwqe->hba_wqidx]->queue_id);
ret = lpfc_sli4_wq_put(wq, wqe);
if (ret) {
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return ret;
}
lpfc_sli_ringtxcmpl_put(phba, pring, pwqe);
spin_unlock_irqrestore(&pring->ring_lock, iflags);
return 0;
}
scsi: lpfc: NVME Initiator: Base modifications NVME Initiator: Base modifications This patch adds base modifications for NVME initiator support. The base modifications consist of: - Formal split of SLI3 rings from SLI-4 WQs (sometimes referred to as rings as well) as implementation now widely varies between the two. - Addition of configuration modes: SCSI initiator only; NVME initiator only; NVME target only; and SCSI and NVME initiator. The configuration mode drives overall adapter configuration, offloads enabled, and resource splits. NVME support is only available on SLI-4 devices and newer fw. - Implements the following based on configuration mode: - Exchange resources are split by protocol; Obviously, if only 1 mode, then no split occurs. Default is 50/50. module attribute allows tuning. - Pools and config parameters are separated per-protocol - Each protocol has it's own set of queues, but share interrupt vectors. SCSI: SLI3 devices have few queues and the original style of queue allocation remains. SLI4 devices piggy back on an "io-channel" concept that eventually needs to merge with scsi-mq/blk-mq support (it is underway). For now, the paradigm continues as it existed prior. io channel allocates N msix and N WQs (N=4 default) and either round robins or uses cpu # modulo N for scheduling. A bunch of module parameters allow the configuration to be tuned. NVME (initiator): Allocates an msix per cpu (or whatever pci_alloc_irq_vectors gets) Allocates a WQ per cpu, and maps the WQs to msix on a WQ # modulo msix vector count basis. Module parameters exist to cap/control the config if desired. - Each protocol has its own buffer and dma pools. I apologize for the size of the patch. Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <james.smart@broadcom.com> ---- Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2017-02-12 14:52:30 -07:00
return WQE_ERROR;
}