1
0
Fork 0
alistair23-linux/drivers/net/ethernet/broadcom/bnx2x/bnx2x_sp.c

6562 lines
182 KiB
C
Raw Normal View History

/* bnx2x_sp.c: Qlogic Everest network driver.
*
* Copyright 2011-2013 Broadcom Corporation
* Copyright (c) 2014 QLogic Corporation
* All rights reserved
*
* Unless you and Qlogic execute a separate written software license
* agreement governing use of this software, this software is licensed to you
* under the terms of the GNU General Public License version 2, available
* at http://www.gnu.org/licenses/gpl-2.0.html (the "GPL").
*
* Notwithstanding the above, under no circumstances may you combine this
* software in any way with any other Qlogic software provided under a
* license other than the GPL, without Qlogic's express prior written
* consent.
*
* Maintained by: Ariel Elior <ariel.elior@qlogic.com>
* Written by: Vladislav Zolotarov
*
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/module.h>
#include <linux/crc32.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/crc32c.h>
#include "bnx2x.h"
#include "bnx2x_cmn.h"
#include "bnx2x_sp.h"
#define BNX2X_MAX_EMUL_MULTI 16
/**** Exe Queue interfaces ****/
/**
* bnx2x_exe_queue_init - init the Exe Queue object
*
* @o: pointer to the object
* @exe_len: length
* @owner: pointer to the owner
* @validate: validate function pointer
* @optimize: optimize function pointer
* @exec: execute function pointer
* @get: get function pointer
*/
static inline void bnx2x_exe_queue_init(struct bnx2x *bp,
struct bnx2x_exe_queue_obj *o,
int exe_len,
union bnx2x_qable_obj *owner,
exe_q_validate validate,
exe_q_remove remove,
exe_q_optimize optimize,
exe_q_execute exec,
exe_q_get get)
{
memset(o, 0, sizeof(*o));
INIT_LIST_HEAD(&o->exe_queue);
INIT_LIST_HEAD(&o->pending_comp);
spin_lock_init(&o->lock);
o->exe_chunk_len = exe_len;
o->owner = owner;
/* Owner specific callbacks */
o->validate = validate;
o->remove = remove;
o->optimize = optimize;
o->execute = exec;
o->get = get;
DP(BNX2X_MSG_SP, "Setup the execution queue with the chunk length of %d\n",
exe_len);
}
static inline void bnx2x_exe_queue_free_elem(struct bnx2x *bp,
struct bnx2x_exeq_elem *elem)
{
DP(BNX2X_MSG_SP, "Deleting an exe_queue element\n");
kfree(elem);
}
static inline int bnx2x_exe_queue_length(struct bnx2x_exe_queue_obj *o)
{
struct bnx2x_exeq_elem *elem;
int cnt = 0;
spin_lock_bh(&o->lock);
list_for_each_entry(elem, &o->exe_queue, link)
cnt++;
spin_unlock_bh(&o->lock);
return cnt;
}
/**
* bnx2x_exe_queue_add - add a new element to the execution queue
*
* @bp: driver handle
* @o: queue
* @cmd: new command to add
* @restore: true - do not optimize the command
*
* If the element is optimized or is illegal, frees it.
*/
static inline int bnx2x_exe_queue_add(struct bnx2x *bp,
struct bnx2x_exe_queue_obj *o,
struct bnx2x_exeq_elem *elem,
bool restore)
{
int rc;
spin_lock_bh(&o->lock);
if (!restore) {
/* Try to cancel this element queue */
rc = o->optimize(bp, o->owner, elem);
if (rc)
goto free_and_exit;
/* Check if this request is ok */
rc = o->validate(bp, o->owner, elem);
if (rc) {
DP(BNX2X_MSG_SP, "Preamble failed: %d\n", rc);
goto free_and_exit;
}
}
/* If so, add it to the execution queue */
list_add_tail(&elem->link, &o->exe_queue);
spin_unlock_bh(&o->lock);
return 0;
free_and_exit:
bnx2x_exe_queue_free_elem(bp, elem);
spin_unlock_bh(&o->lock);
return rc;
}
static inline void __bnx2x_exe_queue_reset_pending(
struct bnx2x *bp,
struct bnx2x_exe_queue_obj *o)
{
struct bnx2x_exeq_elem *elem;
while (!list_empty(&o->pending_comp)) {
elem = list_first_entry(&o->pending_comp,
struct bnx2x_exeq_elem, link);
list_del(&elem->link);
bnx2x_exe_queue_free_elem(bp, elem);
}
}
/**
* bnx2x_exe_queue_step - execute one execution chunk atomically
*
* @bp: driver handle
* @o: queue
* @ramrod_flags: flags
*
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
* (Should be called while holding the exe_queue->lock).
*/
static inline int bnx2x_exe_queue_step(struct bnx2x *bp,
struct bnx2x_exe_queue_obj *o,
unsigned long *ramrod_flags)
{
struct bnx2x_exeq_elem *elem, spacer;
int cur_len = 0, rc;
memset(&spacer, 0, sizeof(spacer));
/* Next step should not be performed until the current is finished,
* unless a DRV_CLEAR_ONLY bit is set. In this case we just want to
* properly clear object internals without sending any command to the FW
* which also implies there won't be any completion to clear the
* 'pending' list.
*/
if (!list_empty(&o->pending_comp)) {
if (test_bit(RAMROD_DRV_CLR_ONLY, ramrod_flags)) {
DP(BNX2X_MSG_SP, "RAMROD_DRV_CLR_ONLY requested: resetting a pending_comp list\n");
__bnx2x_exe_queue_reset_pending(bp, o);
} else {
return 1;
}
}
/* Run through the pending commands list and create a next
* execution chunk.
*/
while (!list_empty(&o->exe_queue)) {
elem = list_first_entry(&o->exe_queue, struct bnx2x_exeq_elem,
link);
WARN_ON(!elem->cmd_len);
if (cur_len + elem->cmd_len <= o->exe_chunk_len) {
cur_len += elem->cmd_len;
/* Prevent from both lists being empty when moving an
* element. This will allow the call of
* bnx2x_exe_queue_empty() without locking.
*/
list_add_tail(&spacer.link, &o->pending_comp);
mb();
list_move_tail(&elem->link, &o->pending_comp);
list_del(&spacer.link);
} else
break;
}
/* Sanity check */
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
if (!cur_len)
return 0;
rc = o->execute(bp, o->owner, &o->pending_comp, ramrod_flags);
if (rc < 0)
/* In case of an error return the commands back to the queue
* and reset the pending_comp.
*/
list_splice_init(&o->pending_comp, &o->exe_queue);
else if (!rc)
/* If zero is returned, means there are no outstanding pending
* completions and we may dismiss the pending list.
*/
__bnx2x_exe_queue_reset_pending(bp, o);
return rc;
}
static inline bool bnx2x_exe_queue_empty(struct bnx2x_exe_queue_obj *o)
{
bool empty = list_empty(&o->exe_queue);
/* Don't reorder!!! */
mb();
return empty && list_empty(&o->pending_comp);
}
static inline struct bnx2x_exeq_elem *bnx2x_exe_queue_alloc_elem(
struct bnx2x *bp)
{
DP(BNX2X_MSG_SP, "Allocating a new exe_queue element\n");
return kzalloc(sizeof(struct bnx2x_exeq_elem), GFP_ATOMIC);
}
/************************ raw_obj functions ***********************************/
static bool bnx2x_raw_check_pending(struct bnx2x_raw_obj *o)
{
return !!test_bit(o->state, o->pstate);
}
static void bnx2x_raw_clear_pending(struct bnx2x_raw_obj *o)
{
smp_mb__before_atomic();
clear_bit(o->state, o->pstate);
smp_mb__after_atomic();
}
static void bnx2x_raw_set_pending(struct bnx2x_raw_obj *o)
{
smp_mb__before_atomic();
set_bit(o->state, o->pstate);
smp_mb__after_atomic();
}
/**
* bnx2x_state_wait - wait until the given bit(state) is cleared
*
* @bp: device handle
* @state: state which is to be cleared
* @state_p: state buffer
*
*/
static inline int bnx2x_state_wait(struct bnx2x *bp, int state,
unsigned long *pstate)
{
/* can take a while if any port is running */
int cnt = 5000;
if (CHIP_REV_IS_EMUL(bp))
cnt *= 20;
DP(BNX2X_MSG_SP, "waiting for state to become %d\n", state);
might_sleep();
while (cnt--) {
if (!test_bit(state, pstate)) {
#ifdef BNX2X_STOP_ON_ERROR
DP(BNX2X_MSG_SP, "exit (cnt %d)\n", 5000 - cnt);
#endif
return 0;
}
usleep_range(1000, 2000);
if (bp->panic)
return -EIO;
}
/* timeout! */
BNX2X_ERR("timeout waiting for state %d\n", state);
#ifdef BNX2X_STOP_ON_ERROR
bnx2x_panic();
#endif
return -EBUSY;
}
static int bnx2x_raw_wait(struct bnx2x *bp, struct bnx2x_raw_obj *raw)
{
return bnx2x_state_wait(bp, raw->state, raw->pstate);
}
/***************** Classification verbs: Set/Del MAC/VLAN/VLAN-MAC ************/
/* credit handling callbacks */
static bool bnx2x_get_cam_offset_mac(struct bnx2x_vlan_mac_obj *o, int *offset)
{
struct bnx2x_credit_pool_obj *mp = o->macs_pool;
WARN_ON(!mp);
return mp->get_entry(mp, offset);
}
static bool bnx2x_get_credit_mac(struct bnx2x_vlan_mac_obj *o)
{
struct bnx2x_credit_pool_obj *mp = o->macs_pool;
WARN_ON(!mp);
return mp->get(mp, 1);
}
static bool bnx2x_get_cam_offset_vlan(struct bnx2x_vlan_mac_obj *o, int *offset)
{
struct bnx2x_credit_pool_obj *vp = o->vlans_pool;
WARN_ON(!vp);
return vp->get_entry(vp, offset);
}
static bool bnx2x_get_credit_vlan(struct bnx2x_vlan_mac_obj *o)
{
struct bnx2x_credit_pool_obj *vp = o->vlans_pool;
WARN_ON(!vp);
return vp->get(vp, 1);
}
static bool bnx2x_get_credit_vlan_mac(struct bnx2x_vlan_mac_obj *o)
{
struct bnx2x_credit_pool_obj *mp = o->macs_pool;
struct bnx2x_credit_pool_obj *vp = o->vlans_pool;
if (!mp->get(mp, 1))
return false;
if (!vp->get(vp, 1)) {
mp->put(mp, 1);
return false;
}
return true;
}
static bool bnx2x_put_cam_offset_mac(struct bnx2x_vlan_mac_obj *o, int offset)
{
struct bnx2x_credit_pool_obj *mp = o->macs_pool;
return mp->put_entry(mp, offset);
}
static bool bnx2x_put_credit_mac(struct bnx2x_vlan_mac_obj *o)
{
struct bnx2x_credit_pool_obj *mp = o->macs_pool;
return mp->put(mp, 1);
}
static bool bnx2x_put_cam_offset_vlan(struct bnx2x_vlan_mac_obj *o, int offset)
{
struct bnx2x_credit_pool_obj *vp = o->vlans_pool;
return vp->put_entry(vp, offset);
}
static bool bnx2x_put_credit_vlan(struct bnx2x_vlan_mac_obj *o)
{
struct bnx2x_credit_pool_obj *vp = o->vlans_pool;
return vp->put(vp, 1);
}
static bool bnx2x_put_credit_vlan_mac(struct bnx2x_vlan_mac_obj *o)
{
struct bnx2x_credit_pool_obj *mp = o->macs_pool;
struct bnx2x_credit_pool_obj *vp = o->vlans_pool;
if (!mp->put(mp, 1))
return false;
if (!vp->put(vp, 1)) {
mp->get(mp, 1);
return false;
}
return true;
}
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
/**
* __bnx2x_vlan_mac_h_write_trylock - try getting the vlan mac writer lock
*
* @bp: device handle
* @o: vlan_mac object
*
* @details: Non-blocking implementation; should be called under execution
* queue lock.
*/
static int __bnx2x_vlan_mac_h_write_trylock(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
if (o->head_reader) {
DP(BNX2X_MSG_SP, "vlan_mac_lock writer - There are readers; Busy\n");
return -EBUSY;
}
DP(BNX2X_MSG_SP, "vlan_mac_lock writer - Taken\n");
return 0;
}
/**
* __bnx2x_vlan_mac_h_exec_pending - execute step instead of a previous step
*
* @bp: device handle
* @o: vlan_mac object
*
* @details Should be called under execution queue lock; notice it might release
* and reclaim it during its run.
*/
static void __bnx2x_vlan_mac_h_exec_pending(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
int rc;
unsigned long ramrod_flags = o->saved_ramrod_flags;
DP(BNX2X_MSG_SP, "vlan_mac_lock execute pending command with ramrod flags %lu\n",
ramrod_flags);
o->head_exe_request = false;
o->saved_ramrod_flags = 0;
rc = bnx2x_exe_queue_step(bp, &o->exe_queue, &ramrod_flags);
if ((rc != 0) && (rc != 1)) {
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
BNX2X_ERR("execution of pending commands failed with rc %d\n",
rc);
#ifdef BNX2X_STOP_ON_ERROR
bnx2x_panic();
#endif
}
}
/**
* __bnx2x_vlan_mac_h_pend - Pend an execution step which couldn't run
*
* @bp: device handle
* @o: vlan_mac object
* @ramrod_flags: ramrod flags of missed execution
*
* @details Should be called under execution queue lock.
*/
static void __bnx2x_vlan_mac_h_pend(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
unsigned long ramrod_flags)
{
o->head_exe_request = true;
o->saved_ramrod_flags = ramrod_flags;
DP(BNX2X_MSG_SP, "Placing pending execution with ramrod flags %lu\n",
ramrod_flags);
}
/**
* __bnx2x_vlan_mac_h_write_unlock - unlock the vlan mac head list writer lock
*
* @bp: device handle
* @o: vlan_mac object
*
* @details Should be called under execution queue lock. Notice if a pending
* execution exists, it would perform it - possibly releasing and
* reclaiming the execution queue lock.
*/
static void __bnx2x_vlan_mac_h_write_unlock(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
/* It's possible a new pending execution was added since this writer
* executed. If so, execute again. [Ad infinitum]
*/
while (o->head_exe_request) {
DP(BNX2X_MSG_SP, "vlan_mac_lock - writer release encountered a pending request\n");
__bnx2x_vlan_mac_h_exec_pending(bp, o);
}
}
/**
* __bnx2x_vlan_mac_h_read_lock - lock the vlan mac head list reader lock
*
* @bp: device handle
* @o: vlan_mac object
*
* @details Should be called under the execution queue lock. May sleep. May
* release and reclaim execution queue lock during its run.
*/
static int __bnx2x_vlan_mac_h_read_lock(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
/* If we got here, we're holding lock --> no WRITER exists */
o->head_reader++;
DP(BNX2X_MSG_SP, "vlan_mac_lock - locked reader - number %d\n",
o->head_reader);
return 0;
}
/**
* bnx2x_vlan_mac_h_read_lock - lock the vlan mac head list reader lock
*
* @bp: device handle
* @o: vlan_mac object
*
* @details May sleep. Claims and releases execution queue lock during its run.
*/
int bnx2x_vlan_mac_h_read_lock(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
int rc;
spin_lock_bh(&o->exe_queue.lock);
rc = __bnx2x_vlan_mac_h_read_lock(bp, o);
spin_unlock_bh(&o->exe_queue.lock);
return rc;
}
/**
* __bnx2x_vlan_mac_h_read_unlock - unlock the vlan mac head list reader lock
*
* @bp: device handle
* @o: vlan_mac object
*
* @details Should be called under execution queue lock. Notice if a pending
* execution exists, it would be performed if this was the last
* reader. possibly releasing and reclaiming the execution queue lock.
*/
static void __bnx2x_vlan_mac_h_read_unlock(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
if (!o->head_reader) {
BNX2X_ERR("Need to release vlan mac reader lock, but lock isn't taken\n");
#ifdef BNX2X_STOP_ON_ERROR
bnx2x_panic();
#endif
} else {
o->head_reader--;
DP(BNX2X_MSG_SP, "vlan_mac_lock - decreased readers to %d\n",
o->head_reader);
}
/* It's possible a new pending execution was added, and that this reader
* was last - if so we need to execute the command.
*/
if (!o->head_reader && o->head_exe_request) {
DP(BNX2X_MSG_SP, "vlan_mac_lock - reader release encountered a pending request\n");
/* Writer release will do the trick */
__bnx2x_vlan_mac_h_write_unlock(bp, o);
}
}
/**
* bnx2x_vlan_mac_h_read_unlock - unlock the vlan mac head list reader lock
*
* @bp: device handle
* @o: vlan_mac object
*
* @details Notice if a pending execution exists, it would be performed if this
* was the last reader. Claims and releases the execution queue lock
* during its run.
*/
void bnx2x_vlan_mac_h_read_unlock(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
spin_lock_bh(&o->exe_queue.lock);
__bnx2x_vlan_mac_h_read_unlock(bp, o);
spin_unlock_bh(&o->exe_queue.lock);
}
static int bnx2x_get_n_elements(struct bnx2x *bp, struct bnx2x_vlan_mac_obj *o,
int n, u8 *base, u8 stride, u8 size)
{
struct bnx2x_vlan_mac_registry_elem *pos;
u8 *next = base;
int counter = 0;
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
int read_lock;
DP(BNX2X_MSG_SP, "get_n_elements - taking vlan_mac_lock (reader)\n");
read_lock = bnx2x_vlan_mac_h_read_lock(bp, o);
if (read_lock != 0)
BNX2X_ERR("get_n_elements failed to get vlan mac reader lock; Access without lock\n");
/* traverse list */
list_for_each_entry(pos, &o->head, link) {
if (counter < n) {
memcpy(next, &pos->u, size);
counter++;
DP(BNX2X_MSG_SP, "copied element number %d to address %p element was:\n",
counter, next);
next += stride + size;
}
}
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
if (read_lock == 0) {
DP(BNX2X_MSG_SP, "get_n_elements - releasing vlan_mac_lock (reader)\n");
bnx2x_vlan_mac_h_read_unlock(bp, o);
}
return counter * ETH_ALEN;
}
/* check_add() callbacks */
static int bnx2x_check_mac_add(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
union bnx2x_classification_ramrod_data *data)
{
struct bnx2x_vlan_mac_registry_elem *pos;
DP(BNX2X_MSG_SP, "Checking MAC %pM for ADD command\n", data->mac.mac);
if (!is_valid_ether_addr(data->mac.mac))
return -EINVAL;
/* Check if a requested MAC already exists */
list_for_each_entry(pos, &o->head, link)
if (ether_addr_equal(data->mac.mac, pos->u.mac.mac) &&
(data->mac.is_inner_mac == pos->u.mac.is_inner_mac))
return -EEXIST;
return 0;
}
static int bnx2x_check_vlan_add(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
union bnx2x_classification_ramrod_data *data)
{
struct bnx2x_vlan_mac_registry_elem *pos;
DP(BNX2X_MSG_SP, "Checking VLAN %d for ADD command\n", data->vlan.vlan);
list_for_each_entry(pos, &o->head, link)
if (data->vlan.vlan == pos->u.vlan.vlan)
return -EEXIST;
return 0;
}
static int bnx2x_check_vlan_mac_add(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
union bnx2x_classification_ramrod_data *data)
{
struct bnx2x_vlan_mac_registry_elem *pos;
DP(BNX2X_MSG_SP, "Checking VLAN_MAC (%pM, %d) for ADD command\n",
data->vlan_mac.mac, data->vlan_mac.vlan);
list_for_each_entry(pos, &o->head, link)
if ((data->vlan_mac.vlan == pos->u.vlan_mac.vlan) &&
(!memcmp(data->vlan_mac.mac, pos->u.vlan_mac.mac,
ETH_ALEN)) &&
(data->vlan_mac.is_inner_mac ==
pos->u.vlan_mac.is_inner_mac))
return -EEXIST;
return 0;
}
/* check_del() callbacks */
static struct bnx2x_vlan_mac_registry_elem *
bnx2x_check_mac_del(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
union bnx2x_classification_ramrod_data *data)
{
struct bnx2x_vlan_mac_registry_elem *pos;
DP(BNX2X_MSG_SP, "Checking MAC %pM for DEL command\n", data->mac.mac);
list_for_each_entry(pos, &o->head, link)
if (ether_addr_equal(data->mac.mac, pos->u.mac.mac) &&
(data->mac.is_inner_mac == pos->u.mac.is_inner_mac))
return pos;
return NULL;
}
static struct bnx2x_vlan_mac_registry_elem *
bnx2x_check_vlan_del(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
union bnx2x_classification_ramrod_data *data)
{
struct bnx2x_vlan_mac_registry_elem *pos;
DP(BNX2X_MSG_SP, "Checking VLAN %d for DEL command\n", data->vlan.vlan);
list_for_each_entry(pos, &o->head, link)
if (data->vlan.vlan == pos->u.vlan.vlan)
return pos;
return NULL;
}
static struct bnx2x_vlan_mac_registry_elem *
bnx2x_check_vlan_mac_del(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
union bnx2x_classification_ramrod_data *data)
{
struct bnx2x_vlan_mac_registry_elem *pos;
DP(BNX2X_MSG_SP, "Checking VLAN_MAC (%pM, %d) for DEL command\n",
data->vlan_mac.mac, data->vlan_mac.vlan);
list_for_each_entry(pos, &o->head, link)
if ((data->vlan_mac.vlan == pos->u.vlan_mac.vlan) &&
(!memcmp(data->vlan_mac.mac, pos->u.vlan_mac.mac,
ETH_ALEN)) &&
(data->vlan_mac.is_inner_mac ==
pos->u.vlan_mac.is_inner_mac))
return pos;
return NULL;
}
/* check_move() callback */
static bool bnx2x_check_move(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *src_o,
struct bnx2x_vlan_mac_obj *dst_o,
union bnx2x_classification_ramrod_data *data)
{
struct bnx2x_vlan_mac_registry_elem *pos;
int rc;
/* Check if we can delete the requested configuration from the first
* object.
*/
pos = src_o->check_del(bp, src_o, data);
/* check if configuration can be added */
rc = dst_o->check_add(bp, dst_o, data);
/* If this classification can not be added (is already set)
* or can't be deleted - return an error.
*/
if (rc || !pos)
return false;
return true;
}
static bool bnx2x_check_move_always_err(
struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *src_o,
struct bnx2x_vlan_mac_obj *dst_o,
union bnx2x_classification_ramrod_data *data)
{
return false;
}
static inline u8 bnx2x_vlan_mac_get_rx_tx_flag(struct bnx2x_vlan_mac_obj *o)
{
struct bnx2x_raw_obj *raw = &o->raw;
u8 rx_tx_flag = 0;
if ((raw->obj_type == BNX2X_OBJ_TYPE_TX) ||
(raw->obj_type == BNX2X_OBJ_TYPE_RX_TX))
rx_tx_flag |= ETH_CLASSIFY_CMD_HEADER_TX_CMD;
if ((raw->obj_type == BNX2X_OBJ_TYPE_RX) ||
(raw->obj_type == BNX2X_OBJ_TYPE_RX_TX))
rx_tx_flag |= ETH_CLASSIFY_CMD_HEADER_RX_CMD;
return rx_tx_flag;
}
static void bnx2x_set_mac_in_nig(struct bnx2x *bp,
bool add, unsigned char *dev_addr, int index)
{
u32 wb_data[2];
u32 reg_offset = BP_PORT(bp) ? NIG_REG_LLH1_FUNC_MEM :
NIG_REG_LLH0_FUNC_MEM;
if (!IS_MF_SI(bp) && !IS_MF_AFEX(bp))
return;
if (index > BNX2X_LLH_CAM_MAX_PF_LINE)
return;
DP(BNX2X_MSG_SP, "Going to %s LLH configuration at entry %d\n",
(add ? "ADD" : "DELETE"), index);
if (add) {
/* LLH_FUNC_MEM is a u64 WB register */
reg_offset += 8*index;
wb_data[0] = ((dev_addr[2] << 24) | (dev_addr[3] << 16) |
(dev_addr[4] << 8) | dev_addr[5]);
wb_data[1] = ((dev_addr[0] << 8) | dev_addr[1]);
REG_WR_DMAE(bp, reg_offset, wb_data, 2);
}
REG_WR(bp, (BP_PORT(bp) ? NIG_REG_LLH1_FUNC_MEM_ENABLE :
NIG_REG_LLH0_FUNC_MEM_ENABLE) + 4*index, add);
}
/**
* bnx2x_vlan_mac_set_cmd_hdr_e2 - set a header in a single classify ramrod
*
* @bp: device handle
* @o: queue for which we want to configure this rule
* @add: if true the command is an ADD command, DEL otherwise
* @opcode: CLASSIFY_RULE_OPCODE_XXX
* @hdr: pointer to a header to setup
*
*/
static inline void bnx2x_vlan_mac_set_cmd_hdr_e2(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o, bool add, int opcode,
struct eth_classify_cmd_header *hdr)
{
struct bnx2x_raw_obj *raw = &o->raw;
hdr->client_id = raw->cl_id;
hdr->func_id = raw->func_id;
/* Rx or/and Tx (internal switching) configuration ? */
hdr->cmd_general_data |=
bnx2x_vlan_mac_get_rx_tx_flag(o);
if (add)
hdr->cmd_general_data |= ETH_CLASSIFY_CMD_HEADER_IS_ADD;
hdr->cmd_general_data |=
(opcode << ETH_CLASSIFY_CMD_HEADER_OPCODE_SHIFT);
}
/**
* bnx2x_vlan_mac_set_rdata_hdr_e2 - set the classify ramrod data header
*
* @cid: connection id
* @type: BNX2X_FILTER_XXX_PENDING
* @hdr: pointer to header to setup
* @rule_cnt:
*
* currently we always configure one rule and echo field to contain a CID and an
* opcode type.
*/
static inline void bnx2x_vlan_mac_set_rdata_hdr_e2(u32 cid, int type,
struct eth_classify_header *hdr, int rule_cnt)
{
hdr->echo = cpu_to_le32((cid & BNX2X_SWCID_MASK) |
(type << BNX2X_SWCID_SHIFT));
hdr->rule_cnt = (u8)rule_cnt;
}
/* hw_config() callbacks */
static void bnx2x_set_one_mac_e2(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
struct bnx2x_exeq_elem *elem, int rule_idx,
int cam_offset)
{
struct bnx2x_raw_obj *raw = &o->raw;
struct eth_classify_rules_ramrod_data *data =
(struct eth_classify_rules_ramrod_data *)(raw->rdata);
int rule_cnt = rule_idx + 1, cmd = elem->cmd_data.vlan_mac.cmd;
union eth_classify_rule_cmd *rule_entry = &data->rules[rule_idx];
bool add = (cmd == BNX2X_VLAN_MAC_ADD) ? true : false;
unsigned long *vlan_mac_flags = &elem->cmd_data.vlan_mac.vlan_mac_flags;
u8 *mac = elem->cmd_data.vlan_mac.u.mac.mac;
/* Set LLH CAM entry: currently only iSCSI and ETH macs are
* relevant. In addition, current implementation is tuned for a
* single ETH MAC.
*
* When multiple unicast ETH MACs PF configuration in switch
* independent mode is required (NetQ, multiple netdev MACs,
* etc.), consider better utilisation of 8 per function MAC
* entries in the LLH register. There is also
* NIG_REG_P[01]_LLH_FUNC_MEM2 registers that complete the
* total number of CAM entries to 16.
*
* Currently we won't configure NIG for MACs other than a primary ETH
* MAC and iSCSI L2 MAC.
*
* If this MAC is moving from one Queue to another, no need to change
* NIG configuration.
*/
if (cmd != BNX2X_VLAN_MAC_MOVE) {
if (test_bit(BNX2X_ISCSI_ETH_MAC, vlan_mac_flags))
bnx2x_set_mac_in_nig(bp, add, mac,
BNX2X_LLH_CAM_ISCSI_ETH_LINE);
else if (test_bit(BNX2X_ETH_MAC, vlan_mac_flags))
bnx2x_set_mac_in_nig(bp, add, mac,
BNX2X_LLH_CAM_ETH_LINE);
}
/* Reset the ramrod data buffer for the first rule */
if (rule_idx == 0)
memset(data, 0, sizeof(*data));
/* Setup a command header */
bnx2x_vlan_mac_set_cmd_hdr_e2(bp, o, add, CLASSIFY_RULE_OPCODE_MAC,
&rule_entry->mac.header);
DP(BNX2X_MSG_SP, "About to %s MAC %pM for Queue %d\n",
(add ? "add" : "delete"), mac, raw->cl_id);
/* Set a MAC itself */
bnx2x_set_fw_mac_addr(&rule_entry->mac.mac_msb,
&rule_entry->mac.mac_mid,
&rule_entry->mac.mac_lsb, mac);
rule_entry->mac.inner_mac =
cpu_to_le16(elem->cmd_data.vlan_mac.u.mac.is_inner_mac);
/* MOVE: Add a rule that will add this MAC to the target Queue */
if (cmd == BNX2X_VLAN_MAC_MOVE) {
rule_entry++;
rule_cnt++;
/* Setup ramrod data */
bnx2x_vlan_mac_set_cmd_hdr_e2(bp,
elem->cmd_data.vlan_mac.target_obj,
true, CLASSIFY_RULE_OPCODE_MAC,
&rule_entry->mac.header);
/* Set a MAC itself */
bnx2x_set_fw_mac_addr(&rule_entry->mac.mac_msb,
&rule_entry->mac.mac_mid,
&rule_entry->mac.mac_lsb, mac);
rule_entry->mac.inner_mac =
cpu_to_le16(elem->cmd_data.vlan_mac.
u.mac.is_inner_mac);
}
/* Set the ramrod data header */
/* TODO: take this to the higher level in order to prevent multiple
writing */
bnx2x_vlan_mac_set_rdata_hdr_e2(raw->cid, raw->state, &data->header,
rule_cnt);
}
/**
* bnx2x_vlan_mac_set_rdata_hdr_e1x - set a header in a single classify ramrod
*
* @bp: device handle
* @o: queue
* @type:
* @cam_offset: offset in cam memory
* @hdr: pointer to a header to setup
*
* E1/E1H
*/
static inline void bnx2x_vlan_mac_set_rdata_hdr_e1x(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o, int type, int cam_offset,
struct mac_configuration_hdr *hdr)
{
struct bnx2x_raw_obj *r = &o->raw;
hdr->length = 1;
hdr->offset = (u8)cam_offset;
hdr->client_id = cpu_to_le16(0xff);
hdr->echo = cpu_to_le32((r->cid & BNX2X_SWCID_MASK) |
(type << BNX2X_SWCID_SHIFT));
}
static inline void bnx2x_vlan_mac_set_cfg_entry_e1x(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o, bool add, int opcode, u8 *mac,
u16 vlan_id, struct mac_configuration_entry *cfg_entry)
{
struct bnx2x_raw_obj *r = &o->raw;
u32 cl_bit_vec = (1 << r->cl_id);
cfg_entry->clients_bit_vector = cpu_to_le32(cl_bit_vec);
cfg_entry->pf_id = r->func_id;
cfg_entry->vlan_id = cpu_to_le16(vlan_id);
if (add) {
SET_FLAG(cfg_entry->flags, MAC_CONFIGURATION_ENTRY_ACTION_TYPE,
T_ETH_MAC_COMMAND_SET);
SET_FLAG(cfg_entry->flags,
MAC_CONFIGURATION_ENTRY_VLAN_FILTERING_MODE, opcode);
/* Set a MAC in a ramrod data */
bnx2x_set_fw_mac_addr(&cfg_entry->msb_mac_addr,
&cfg_entry->middle_mac_addr,
&cfg_entry->lsb_mac_addr, mac);
} else
SET_FLAG(cfg_entry->flags, MAC_CONFIGURATION_ENTRY_ACTION_TYPE,
T_ETH_MAC_COMMAND_INVALIDATE);
}
static inline void bnx2x_vlan_mac_set_rdata_e1x(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o, int type, int cam_offset, bool add,
u8 *mac, u16 vlan_id, int opcode, struct mac_configuration_cmd *config)
{
struct mac_configuration_entry *cfg_entry = &config->config_table[0];
struct bnx2x_raw_obj *raw = &o->raw;
bnx2x_vlan_mac_set_rdata_hdr_e1x(bp, o, type, cam_offset,
&config->hdr);
bnx2x_vlan_mac_set_cfg_entry_e1x(bp, o, add, opcode, mac, vlan_id,
cfg_entry);
DP(BNX2X_MSG_SP, "%s MAC %pM CLID %d CAM offset %d\n",
(add ? "setting" : "clearing"),
mac, raw->cl_id, cam_offset);
}
/**
* bnx2x_set_one_mac_e1x - fill a single MAC rule ramrod data
*
* @bp: device handle
* @o: bnx2x_vlan_mac_obj
* @elem: bnx2x_exeq_elem
* @rule_idx: rule_idx
* @cam_offset: cam_offset
*/
static void bnx2x_set_one_mac_e1x(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
struct bnx2x_exeq_elem *elem, int rule_idx,
int cam_offset)
{
struct bnx2x_raw_obj *raw = &o->raw;
struct mac_configuration_cmd *config =
(struct mac_configuration_cmd *)(raw->rdata);
/* 57710 and 57711 do not support MOVE command,
* so it's either ADD or DEL
*/
bool add = (elem->cmd_data.vlan_mac.cmd == BNX2X_VLAN_MAC_ADD) ?
true : false;
/* Reset the ramrod data buffer */
memset(config, 0, sizeof(*config));
bnx2x_vlan_mac_set_rdata_e1x(bp, o, raw->state,
cam_offset, add,
elem->cmd_data.vlan_mac.u.mac.mac, 0,
ETH_VLAN_FILTER_ANY_VLAN, config);
}
static void bnx2x_set_one_vlan_e2(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
struct bnx2x_exeq_elem *elem, int rule_idx,
int cam_offset)
{
struct bnx2x_raw_obj *raw = &o->raw;
struct eth_classify_rules_ramrod_data *data =
(struct eth_classify_rules_ramrod_data *)(raw->rdata);
int rule_cnt = rule_idx + 1;
union eth_classify_rule_cmd *rule_entry = &data->rules[rule_idx];
enum bnx2x_vlan_mac_cmd cmd = elem->cmd_data.vlan_mac.cmd;
bool add = (cmd == BNX2X_VLAN_MAC_ADD) ? true : false;
u16 vlan = elem->cmd_data.vlan_mac.u.vlan.vlan;
/* Reset the ramrod data buffer for the first rule */
if (rule_idx == 0)
memset(data, 0, sizeof(*data));
/* Set a rule header */
bnx2x_vlan_mac_set_cmd_hdr_e2(bp, o, add, CLASSIFY_RULE_OPCODE_VLAN,
&rule_entry->vlan.header);
DP(BNX2X_MSG_SP, "About to %s VLAN %d\n", (add ? "add" : "delete"),
vlan);
/* Set a VLAN itself */
rule_entry->vlan.vlan = cpu_to_le16(vlan);
/* MOVE: Add a rule that will add this MAC to the target Queue */
if (cmd == BNX2X_VLAN_MAC_MOVE) {
rule_entry++;
rule_cnt++;
/* Setup ramrod data */
bnx2x_vlan_mac_set_cmd_hdr_e2(bp,
elem->cmd_data.vlan_mac.target_obj,
true, CLASSIFY_RULE_OPCODE_VLAN,
&rule_entry->vlan.header);
/* Set a VLAN itself */
rule_entry->vlan.vlan = cpu_to_le16(vlan);
}
/* Set the ramrod data header */
/* TODO: take this to the higher level in order to prevent multiple
writing */
bnx2x_vlan_mac_set_rdata_hdr_e2(raw->cid, raw->state, &data->header,
rule_cnt);
}
static void bnx2x_set_one_vlan_mac_e2(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
struct bnx2x_exeq_elem *elem,
int rule_idx, int cam_offset)
{
struct bnx2x_raw_obj *raw = &o->raw;
struct eth_classify_rules_ramrod_data *data =
(struct eth_classify_rules_ramrod_data *)(raw->rdata);
int rule_cnt = rule_idx + 1;
union eth_classify_rule_cmd *rule_entry = &data->rules[rule_idx];
enum bnx2x_vlan_mac_cmd cmd = elem->cmd_data.vlan_mac.cmd;
bool add = (cmd == BNX2X_VLAN_MAC_ADD) ? true : false;
u16 vlan = elem->cmd_data.vlan_mac.u.vlan_mac.vlan;
u8 *mac = elem->cmd_data.vlan_mac.u.vlan_mac.mac;
u16 inner_mac;
/* Reset the ramrod data buffer for the first rule */
if (rule_idx == 0)
memset(data, 0, sizeof(*data));
/* Set a rule header */
bnx2x_vlan_mac_set_cmd_hdr_e2(bp, o, add, CLASSIFY_RULE_OPCODE_PAIR,
&rule_entry->pair.header);
/* Set VLAN and MAC themselves */
rule_entry->pair.vlan = cpu_to_le16(vlan);
bnx2x_set_fw_mac_addr(&rule_entry->pair.mac_msb,
&rule_entry->pair.mac_mid,
&rule_entry->pair.mac_lsb, mac);
inner_mac = elem->cmd_data.vlan_mac.u.vlan_mac.is_inner_mac;
rule_entry->pair.inner_mac = cpu_to_le16(inner_mac);
/* MOVE: Add a rule that will add this MAC/VLAN to the target Queue */
if (cmd == BNX2X_VLAN_MAC_MOVE) {
struct bnx2x_vlan_mac_obj *target_obj;
rule_entry++;
rule_cnt++;
/* Setup ramrod data */
target_obj = elem->cmd_data.vlan_mac.target_obj;
bnx2x_vlan_mac_set_cmd_hdr_e2(bp, target_obj,
true, CLASSIFY_RULE_OPCODE_PAIR,
&rule_entry->pair.header);
/* Set a VLAN itself */
rule_entry->pair.vlan = cpu_to_le16(vlan);
bnx2x_set_fw_mac_addr(&rule_entry->pair.mac_msb,
&rule_entry->pair.mac_mid,
&rule_entry->pair.mac_lsb, mac);
rule_entry->pair.inner_mac = cpu_to_le16(inner_mac);
}
/* Set the ramrod data header */
bnx2x_vlan_mac_set_rdata_hdr_e2(raw->cid, raw->state, &data->header,
rule_cnt);
}
/**
* bnx2x_set_one_vlan_mac_e1h -
*
* @bp: device handle
* @o: bnx2x_vlan_mac_obj
* @elem: bnx2x_exeq_elem
* @rule_idx: rule_idx
* @cam_offset: cam_offset
*/
static void bnx2x_set_one_vlan_mac_e1h(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
struct bnx2x_exeq_elem *elem,
int rule_idx, int cam_offset)
{
struct bnx2x_raw_obj *raw = &o->raw;
struct mac_configuration_cmd *config =
(struct mac_configuration_cmd *)(raw->rdata);
/* 57710 and 57711 do not support MOVE command,
* so it's either ADD or DEL
*/
bool add = (elem->cmd_data.vlan_mac.cmd == BNX2X_VLAN_MAC_ADD) ?
true : false;
/* Reset the ramrod data buffer */
memset(config, 0, sizeof(*config));
bnx2x_vlan_mac_set_rdata_e1x(bp, o, BNX2X_FILTER_VLAN_MAC_PENDING,
cam_offset, add,
elem->cmd_data.vlan_mac.u.vlan_mac.mac,
elem->cmd_data.vlan_mac.u.vlan_mac.vlan,
ETH_VLAN_FILTER_CLASSIFY, config);
}
/**
* bnx2x_vlan_mac_restore - reconfigure next MAC/VLAN/VLAN-MAC element
*
* @bp: device handle
* @p: command parameters
* @ppos: pointer to the cookie
*
* reconfigure next MAC/VLAN/VLAN-MAC element from the
* previously configured elements list.
*
* from command parameters only RAMROD_COMP_WAIT bit in ramrod_flags is taken
* into an account
*
* pointer to the cookie - that should be given back in the next call to make
* function handle the next element. If *ppos is set to NULL it will restart the
* iterator. If returned *ppos == NULL this means that the last element has been
* handled.
*
*/
static int bnx2x_vlan_mac_restore(struct bnx2x *bp,
struct bnx2x_vlan_mac_ramrod_params *p,
struct bnx2x_vlan_mac_registry_elem **ppos)
{
struct bnx2x_vlan_mac_registry_elem *pos;
struct bnx2x_vlan_mac_obj *o = p->vlan_mac_obj;
/* If list is empty - there is nothing to do here */
if (list_empty(&o->head)) {
*ppos = NULL;
return 0;
}
/* make a step... */
if (*ppos == NULL)
*ppos = list_first_entry(&o->head,
struct bnx2x_vlan_mac_registry_elem,
link);
else
*ppos = list_next_entry(*ppos, link);
pos = *ppos;
/* If it's the last step - return NULL */
if (list_is_last(&pos->link, &o->head))
*ppos = NULL;
/* Prepare a 'user_req' */
memcpy(&p->user_req.u, &pos->u, sizeof(pos->u));
/* Set the command */
p->user_req.cmd = BNX2X_VLAN_MAC_ADD;
/* Set vlan_mac_flags */
p->user_req.vlan_mac_flags = pos->vlan_mac_flags;
/* Set a restore bit */
__set_bit(RAMROD_RESTORE, &p->ramrod_flags);
return bnx2x_config_vlan_mac(bp, p);
}
/* bnx2x_exeq_get_mac/bnx2x_exeq_get_vlan/bnx2x_exeq_get_vlan_mac return a
* pointer to an element with a specific criteria and NULL if such an element
* hasn't been found.
*/
static struct bnx2x_exeq_elem *bnx2x_exeq_get_mac(
struct bnx2x_exe_queue_obj *o,
struct bnx2x_exeq_elem *elem)
{
struct bnx2x_exeq_elem *pos;
struct bnx2x_mac_ramrod_data *data = &elem->cmd_data.vlan_mac.u.mac;
/* Check pending for execution commands */
list_for_each_entry(pos, &o->exe_queue, link)
if (!memcmp(&pos->cmd_data.vlan_mac.u.mac, data,
sizeof(*data)) &&
(pos->cmd_data.vlan_mac.cmd == elem->cmd_data.vlan_mac.cmd))
return pos;
return NULL;
}
static struct bnx2x_exeq_elem *bnx2x_exeq_get_vlan(
struct bnx2x_exe_queue_obj *o,
struct bnx2x_exeq_elem *elem)
{
struct bnx2x_exeq_elem *pos;
struct bnx2x_vlan_ramrod_data *data = &elem->cmd_data.vlan_mac.u.vlan;
/* Check pending for execution commands */
list_for_each_entry(pos, &o->exe_queue, link)
if (!memcmp(&pos->cmd_data.vlan_mac.u.vlan, data,
sizeof(*data)) &&
(pos->cmd_data.vlan_mac.cmd == elem->cmd_data.vlan_mac.cmd))
return pos;
return NULL;
}
static struct bnx2x_exeq_elem *bnx2x_exeq_get_vlan_mac(
struct bnx2x_exe_queue_obj *o,
struct bnx2x_exeq_elem *elem)
{
struct bnx2x_exeq_elem *pos;
struct bnx2x_vlan_mac_ramrod_data *data =
&elem->cmd_data.vlan_mac.u.vlan_mac;
/* Check pending for execution commands */
list_for_each_entry(pos, &o->exe_queue, link)
if (!memcmp(&pos->cmd_data.vlan_mac.u.vlan_mac, data,
sizeof(*data)) &&
(pos->cmd_data.vlan_mac.cmd ==
elem->cmd_data.vlan_mac.cmd))
return pos;
return NULL;
}
/**
* bnx2x_validate_vlan_mac_add - check if an ADD command can be executed
*
* @bp: device handle
* @qo: bnx2x_qable_obj
* @elem: bnx2x_exeq_elem
*
* Checks that the requested configuration can be added. If yes and if
* requested, consume CAM credit.
*
* The 'validate' is run after the 'optimize'.
*
*/
static inline int bnx2x_validate_vlan_mac_add(struct bnx2x *bp,
union bnx2x_qable_obj *qo,
struct bnx2x_exeq_elem *elem)
{
struct bnx2x_vlan_mac_obj *o = &qo->vlan_mac;
struct bnx2x_exe_queue_obj *exeq = &o->exe_queue;
int rc;
/* Check the registry */
rc = o->check_add(bp, o, &elem->cmd_data.vlan_mac.u);
if (rc) {
DP(BNX2X_MSG_SP, "ADD command is not allowed considering current registry state.\n");
return rc;
}
/* Check if there is a pending ADD command for this
* MAC/VLAN/VLAN-MAC. Return an error if there is.
*/
if (exeq->get(exeq, elem)) {
DP(BNX2X_MSG_SP, "There is a pending ADD command already\n");
return -EEXIST;
}
/* TODO: Check the pending MOVE from other objects where this
* object is a destination object.
*/
/* Consume the credit if not requested not to */
if (!(test_bit(BNX2X_DONT_CONSUME_CAM_CREDIT,
&elem->cmd_data.vlan_mac.vlan_mac_flags) ||
o->get_credit(o)))
return -EINVAL;
return 0;
}
/**
* bnx2x_validate_vlan_mac_del - check if the DEL command can be executed
*
* @bp: device handle
* @qo: quable object to check
* @elem: element that needs to be deleted
*
* Checks that the requested configuration can be deleted. If yes and if
* requested, returns a CAM credit.
*
* The 'validate' is run after the 'optimize'.
*/
static inline int bnx2x_validate_vlan_mac_del(struct bnx2x *bp,
union bnx2x_qable_obj *qo,
struct bnx2x_exeq_elem *elem)
{
struct bnx2x_vlan_mac_obj *o = &qo->vlan_mac;
struct bnx2x_vlan_mac_registry_elem *pos;
struct bnx2x_exe_queue_obj *exeq = &o->exe_queue;
struct bnx2x_exeq_elem query_elem;
/* If this classification can not be deleted (doesn't exist)
* - return a BNX2X_EXIST.
*/
pos = o->check_del(bp, o, &elem->cmd_data.vlan_mac.u);
if (!pos) {
DP(BNX2X_MSG_SP, "DEL command is not allowed considering current registry state\n");
return -EEXIST;
}
/* Check if there are pending DEL or MOVE commands for this
* MAC/VLAN/VLAN-MAC. Return an error if so.
*/
memcpy(&query_elem, elem, sizeof(query_elem));
/* Check for MOVE commands */
query_elem.cmd_data.vlan_mac.cmd = BNX2X_VLAN_MAC_MOVE;
if (exeq->get(exeq, &query_elem)) {
BNX2X_ERR("There is a pending MOVE command already\n");
return -EINVAL;
}
/* Check for DEL commands */
if (exeq->get(exeq, elem)) {
DP(BNX2X_MSG_SP, "There is a pending DEL command already\n");
return -EEXIST;
}
/* Return the credit to the credit pool if not requested not to */
if (!(test_bit(BNX2X_DONT_CONSUME_CAM_CREDIT,
&elem->cmd_data.vlan_mac.vlan_mac_flags) ||
o->put_credit(o))) {
BNX2X_ERR("Failed to return a credit\n");
return -EINVAL;
}
return 0;
}
/**
* bnx2x_validate_vlan_mac_move - check if the MOVE command can be executed
*
* @bp: device handle
* @qo: quable object to check (source)
* @elem: element that needs to be moved
*
* Checks that the requested configuration can be moved. If yes and if
* requested, returns a CAM credit.
*
* The 'validate' is run after the 'optimize'.
*/
static inline int bnx2x_validate_vlan_mac_move(struct bnx2x *bp,
union bnx2x_qable_obj *qo,
struct bnx2x_exeq_elem *elem)
{
struct bnx2x_vlan_mac_obj *src_o = &qo->vlan_mac;
struct bnx2x_vlan_mac_obj *dest_o = elem->cmd_data.vlan_mac.target_obj;
struct bnx2x_exeq_elem query_elem;
struct bnx2x_exe_queue_obj *src_exeq = &src_o->exe_queue;
struct bnx2x_exe_queue_obj *dest_exeq = &dest_o->exe_queue;
/* Check if we can perform this operation based on the current registry
* state.
*/
if (!src_o->check_move(bp, src_o, dest_o,
&elem->cmd_data.vlan_mac.u)) {
DP(BNX2X_MSG_SP, "MOVE command is not allowed considering current registry state\n");
return -EINVAL;
}
/* Check if there is an already pending DEL or MOVE command for the
* source object or ADD command for a destination object. Return an
* error if so.
*/
memcpy(&query_elem, elem, sizeof(query_elem));
/* Check DEL on source */
query_elem.cmd_data.vlan_mac.cmd = BNX2X_VLAN_MAC_DEL;
if (src_exeq->get(src_exeq, &query_elem)) {
BNX2X_ERR("There is a pending DEL command on the source queue already\n");
return -EINVAL;
}
/* Check MOVE on source */
if (src_exeq->get(src_exeq, elem)) {
DP(BNX2X_MSG_SP, "There is a pending MOVE command already\n");
return -EEXIST;
}
/* Check ADD on destination */
query_elem.cmd_data.vlan_mac.cmd = BNX2X_VLAN_MAC_ADD;
if (dest_exeq->get(dest_exeq, &query_elem)) {
BNX2X_ERR("There is a pending ADD command on the destination queue already\n");
return -EINVAL;
}
/* Consume the credit if not requested not to */
if (!(test_bit(BNX2X_DONT_CONSUME_CAM_CREDIT_DEST,
&elem->cmd_data.vlan_mac.vlan_mac_flags) ||
dest_o->get_credit(dest_o)))
return -EINVAL;
if (!(test_bit(BNX2X_DONT_CONSUME_CAM_CREDIT,
&elem->cmd_data.vlan_mac.vlan_mac_flags) ||
src_o->put_credit(src_o))) {
/* return the credit taken from dest... */
dest_o->put_credit(dest_o);
return -EINVAL;
}
return 0;
}
static int bnx2x_validate_vlan_mac(struct bnx2x *bp,
union bnx2x_qable_obj *qo,
struct bnx2x_exeq_elem *elem)
{
switch (elem->cmd_data.vlan_mac.cmd) {
case BNX2X_VLAN_MAC_ADD:
return bnx2x_validate_vlan_mac_add(bp, qo, elem);
case BNX2X_VLAN_MAC_DEL:
return bnx2x_validate_vlan_mac_del(bp, qo, elem);
case BNX2X_VLAN_MAC_MOVE:
return bnx2x_validate_vlan_mac_move(bp, qo, elem);
default:
return -EINVAL;
}
}
static int bnx2x_remove_vlan_mac(struct bnx2x *bp,
union bnx2x_qable_obj *qo,
struct bnx2x_exeq_elem *elem)
{
int rc = 0;
/* If consumption wasn't required, nothing to do */
if (test_bit(BNX2X_DONT_CONSUME_CAM_CREDIT,
&elem->cmd_data.vlan_mac.vlan_mac_flags))
return 0;
switch (elem->cmd_data.vlan_mac.cmd) {
case BNX2X_VLAN_MAC_ADD:
case BNX2X_VLAN_MAC_MOVE:
rc = qo->vlan_mac.put_credit(&qo->vlan_mac);
break;
case BNX2X_VLAN_MAC_DEL:
rc = qo->vlan_mac.get_credit(&qo->vlan_mac);
break;
default:
return -EINVAL;
}
if (rc != true)
return -EINVAL;
return 0;
}
/**
* bnx2x_wait_vlan_mac - passively wait for 5 seconds until all work completes.
*
* @bp: device handle
* @o: bnx2x_vlan_mac_obj
*
*/
static int bnx2x_wait_vlan_mac(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o)
{
int cnt = 5000, rc;
struct bnx2x_exe_queue_obj *exeq = &o->exe_queue;
struct bnx2x_raw_obj *raw = &o->raw;
while (cnt--) {
/* Wait for the current command to complete */
rc = raw->wait_comp(bp, raw);
if (rc)
return rc;
/* Wait until there are no pending commands */
if (!bnx2x_exe_queue_empty(exeq))
usleep_range(1000, 2000);
else
return 0;
}
return -EBUSY;
}
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
static int __bnx2x_vlan_mac_execute_step(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
unsigned long *ramrod_flags)
{
int rc = 0;
spin_lock_bh(&o->exe_queue.lock);
DP(BNX2X_MSG_SP, "vlan_mac_execute_step - trying to take writer lock\n");
rc = __bnx2x_vlan_mac_h_write_trylock(bp, o);
if (rc != 0) {
__bnx2x_vlan_mac_h_pend(bp, o, *ramrod_flags);
/* Calling function should not differentiate between this case
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
* and the case in which there is already a pending ramrod
*/
rc = 1;
} else {
rc = bnx2x_exe_queue_step(bp, &o->exe_queue, ramrod_flags);
}
spin_unlock_bh(&o->exe_queue.lock);
return rc;
}
/**
* bnx2x_complete_vlan_mac - complete one VLAN-MAC ramrod
*
* @bp: device handle
* @o: bnx2x_vlan_mac_obj
* @cqe:
* @cont: if true schedule next execution chunk
*
*/
static int bnx2x_complete_vlan_mac(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
union event_ring_elem *cqe,
unsigned long *ramrod_flags)
{
struct bnx2x_raw_obj *r = &o->raw;
int rc;
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
/* Clearing the pending list & raw state should be made
* atomically (as execution flow assumes they represent the same).
*/
spin_lock_bh(&o->exe_queue.lock);
/* Reset pending list */
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
__bnx2x_exe_queue_reset_pending(bp, &o->exe_queue);
/* Clear pending */
r->clear_pending(r);
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
spin_unlock_bh(&o->exe_queue.lock);
/* If ramrod failed this is most likely a SW bug */
if (cqe->message.error)
return -EINVAL;
/* Run the next bulk of pending commands if requested */
if (test_bit(RAMROD_CONT, ramrod_flags)) {
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
rc = __bnx2x_vlan_mac_execute_step(bp, o, ramrod_flags);
if (rc < 0)
return rc;
}
/* If there is more work to do return PENDING */
if (!bnx2x_exe_queue_empty(&o->exe_queue))
return 1;
return 0;
}
/**
* bnx2x_optimize_vlan_mac - optimize ADD and DEL commands.
*
* @bp: device handle
* @o: bnx2x_qable_obj
* @elem: bnx2x_exeq_elem
*/
static int bnx2x_optimize_vlan_mac(struct bnx2x *bp,
union bnx2x_qable_obj *qo,
struct bnx2x_exeq_elem *elem)
{
struct bnx2x_exeq_elem query, *pos;
struct bnx2x_vlan_mac_obj *o = &qo->vlan_mac;
struct bnx2x_exe_queue_obj *exeq = &o->exe_queue;
memcpy(&query, elem, sizeof(query));
switch (elem->cmd_data.vlan_mac.cmd) {
case BNX2X_VLAN_MAC_ADD:
query.cmd_data.vlan_mac.cmd = BNX2X_VLAN_MAC_DEL;
break;
case BNX2X_VLAN_MAC_DEL:
query.cmd_data.vlan_mac.cmd = BNX2X_VLAN_MAC_ADD;
break;
default:
/* Don't handle anything other than ADD or DEL */
return 0;
}
/* If we found the appropriate element - delete it */
pos = exeq->get(exeq, &query);
if (pos) {
/* Return the credit of the optimized command */
if (!test_bit(BNX2X_DONT_CONSUME_CAM_CREDIT,
&pos->cmd_data.vlan_mac.vlan_mac_flags)) {
if ((query.cmd_data.vlan_mac.cmd ==
BNX2X_VLAN_MAC_ADD) && !o->put_credit(o)) {
BNX2X_ERR("Failed to return the credit for the optimized ADD command\n");
return -EINVAL;
} else if (!o->get_credit(o)) { /* VLAN_MAC_DEL */
BNX2X_ERR("Failed to recover the credit from the optimized DEL command\n");
return -EINVAL;
}
}
DP(BNX2X_MSG_SP, "Optimizing %s command\n",
(elem->cmd_data.vlan_mac.cmd == BNX2X_VLAN_MAC_ADD) ?
"ADD" : "DEL");
list_del(&pos->link);
bnx2x_exe_queue_free_elem(bp, pos);
return 1;
}
return 0;
}
/**
* bnx2x_vlan_mac_get_registry_elem - prepare a registry element
*
* @bp: device handle
* @o:
* @elem:
* @restore:
* @re:
*
* prepare a registry element according to the current command request.
*/
static inline int bnx2x_vlan_mac_get_registry_elem(
struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
struct bnx2x_exeq_elem *elem,
bool restore,
struct bnx2x_vlan_mac_registry_elem **re)
{
enum bnx2x_vlan_mac_cmd cmd = elem->cmd_data.vlan_mac.cmd;
struct bnx2x_vlan_mac_registry_elem *reg_elem;
/* Allocate a new registry element if needed. */
if (!restore &&
((cmd == BNX2X_VLAN_MAC_ADD) || (cmd == BNX2X_VLAN_MAC_MOVE))) {
reg_elem = kzalloc(sizeof(*reg_elem), GFP_ATOMIC);
if (!reg_elem)
return -ENOMEM;
/* Get a new CAM offset */
if (!o->get_cam_offset(o, &reg_elem->cam_offset)) {
/* This shall never happen, because we have checked the
* CAM availability in the 'validate'.
*/
WARN_ON(1);
kfree(reg_elem);
return -EINVAL;
}
DP(BNX2X_MSG_SP, "Got cam offset %d\n", reg_elem->cam_offset);
/* Set a VLAN-MAC data */
memcpy(&reg_elem->u, &elem->cmd_data.vlan_mac.u,
sizeof(reg_elem->u));
/* Copy the flags (needed for DEL and RESTORE flows) */
reg_elem->vlan_mac_flags =
elem->cmd_data.vlan_mac.vlan_mac_flags;
} else /* DEL, RESTORE */
reg_elem = o->check_del(bp, o, &elem->cmd_data.vlan_mac.u);
*re = reg_elem;
return 0;
}
/**
* bnx2x_execute_vlan_mac - execute vlan mac command
*
* @bp: device handle
* @qo:
* @exe_chunk:
* @ramrod_flags:
*
* go and send a ramrod!
*/
static int bnx2x_execute_vlan_mac(struct bnx2x *bp,
union bnx2x_qable_obj *qo,
struct list_head *exe_chunk,
unsigned long *ramrod_flags)
{
struct bnx2x_exeq_elem *elem;
struct bnx2x_vlan_mac_obj *o = &qo->vlan_mac, *cam_obj;
struct bnx2x_raw_obj *r = &o->raw;
int rc, idx = 0;
bool restore = test_bit(RAMROD_RESTORE, ramrod_flags);
bool drv_only = test_bit(RAMROD_DRV_CLR_ONLY, ramrod_flags);
struct bnx2x_vlan_mac_registry_elem *reg_elem;
enum bnx2x_vlan_mac_cmd cmd;
/* If DRIVER_ONLY execution is requested, cleanup a registry
* and exit. Otherwise send a ramrod to FW.
*/
if (!drv_only) {
WARN_ON(r->check_pending(r));
/* Set pending */
r->set_pending(r);
/* Fill the ramrod data */
list_for_each_entry(elem, exe_chunk, link) {
cmd = elem->cmd_data.vlan_mac.cmd;
/* We will add to the target object in MOVE command, so
* change the object for a CAM search.
*/
if (cmd == BNX2X_VLAN_MAC_MOVE)
cam_obj = elem->cmd_data.vlan_mac.target_obj;
else
cam_obj = o;
rc = bnx2x_vlan_mac_get_registry_elem(bp, cam_obj,
elem, restore,
&reg_elem);
if (rc)
goto error_exit;
WARN_ON(!reg_elem);
/* Push a new entry into the registry */
if (!restore &&
((cmd == BNX2X_VLAN_MAC_ADD) ||
(cmd == BNX2X_VLAN_MAC_MOVE)))
list_add(&reg_elem->link, &cam_obj->head);
/* Configure a single command in a ramrod data buffer */
o->set_one_rule(bp, o, elem, idx,
reg_elem->cam_offset);
/* MOVE command consumes 2 entries in the ramrod data */
if (cmd == BNX2X_VLAN_MAC_MOVE)
idx += 2;
else
idx++;
}
/* No need for an explicit memory barrier here as long we would
* need to ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read and we will have to put a full memory barrier there
* (inside bnx2x_sp_post()).
*/
rc = bnx2x_sp_post(bp, o->ramrod_cmd, r->cid,
U64_HI(r->rdata_mapping),
U64_LO(r->rdata_mapping),
ETH_CONNECTION_TYPE);
if (rc)
goto error_exit;
}
/* Now, when we are done with the ramrod - clean up the registry */
list_for_each_entry(elem, exe_chunk, link) {
cmd = elem->cmd_data.vlan_mac.cmd;
if ((cmd == BNX2X_VLAN_MAC_DEL) ||
(cmd == BNX2X_VLAN_MAC_MOVE)) {
reg_elem = o->check_del(bp, o,
&elem->cmd_data.vlan_mac.u);
WARN_ON(!reg_elem);
o->put_cam_offset(o, reg_elem->cam_offset);
list_del(&reg_elem->link);
kfree(reg_elem);
}
}
if (!drv_only)
return 1;
else
return 0;
error_exit:
r->clear_pending(r);
/* Cleanup a registry in case of a failure */
list_for_each_entry(elem, exe_chunk, link) {
cmd = elem->cmd_data.vlan_mac.cmd;
if (cmd == BNX2X_VLAN_MAC_MOVE)
cam_obj = elem->cmd_data.vlan_mac.target_obj;
else
cam_obj = o;
/* Delete all newly added above entries */
if (!restore &&
((cmd == BNX2X_VLAN_MAC_ADD) ||
(cmd == BNX2X_VLAN_MAC_MOVE))) {
reg_elem = o->check_del(bp, cam_obj,
&elem->cmd_data.vlan_mac.u);
if (reg_elem) {
list_del(&reg_elem->link);
kfree(reg_elem);
}
}
}
return rc;
}
static inline int bnx2x_vlan_mac_push_new_cmd(
struct bnx2x *bp,
struct bnx2x_vlan_mac_ramrod_params *p)
{
struct bnx2x_exeq_elem *elem;
struct bnx2x_vlan_mac_obj *o = p->vlan_mac_obj;
bool restore = test_bit(RAMROD_RESTORE, &p->ramrod_flags);
/* Allocate the execution queue element */
elem = bnx2x_exe_queue_alloc_elem(bp);
if (!elem)
return -ENOMEM;
/* Set the command 'length' */
switch (p->user_req.cmd) {
case BNX2X_VLAN_MAC_MOVE:
elem->cmd_len = 2;
break;
default:
elem->cmd_len = 1;
}
/* Fill the object specific info */
memcpy(&elem->cmd_data.vlan_mac, &p->user_req, sizeof(p->user_req));
/* Try to add a new command to the pending list */
return bnx2x_exe_queue_add(bp, &o->exe_queue, elem, restore);
}
/**
* bnx2x_config_vlan_mac - configure VLAN/MAC/VLAN_MAC filtering rules.
*
* @bp: device handle
* @p:
*
*/
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
int bnx2x_config_vlan_mac(struct bnx2x *bp,
struct bnx2x_vlan_mac_ramrod_params *p)
{
int rc = 0;
struct bnx2x_vlan_mac_obj *o = p->vlan_mac_obj;
unsigned long *ramrod_flags = &p->ramrod_flags;
bool cont = test_bit(RAMROD_CONT, ramrod_flags);
struct bnx2x_raw_obj *raw = &o->raw;
/*
* Add new elements to the execution list for commands that require it.
*/
if (!cont) {
rc = bnx2x_vlan_mac_push_new_cmd(bp, p);
if (rc)
return rc;
}
/* If nothing will be executed further in this iteration we want to
* return PENDING if there are pending commands
*/
if (!bnx2x_exe_queue_empty(&o->exe_queue))
rc = 1;
if (test_bit(RAMROD_DRV_CLR_ONLY, ramrod_flags)) {
DP(BNX2X_MSG_SP, "RAMROD_DRV_CLR_ONLY requested: clearing a pending bit.\n");
raw->clear_pending(raw);
}
/* Execute commands if required */
if (cont || test_bit(RAMROD_EXEC, ramrod_flags) ||
test_bit(RAMROD_COMP_WAIT, ramrod_flags)) {
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
rc = __bnx2x_vlan_mac_execute_step(bp, p->vlan_mac_obj,
&p->ramrod_flags);
if (rc < 0)
return rc;
}
/* RAMROD_COMP_WAIT is a superset of RAMROD_EXEC. If it was set
* then user want to wait until the last command is done.
*/
if (test_bit(RAMROD_COMP_WAIT, &p->ramrod_flags)) {
/* Wait maximum for the current exe_queue length iterations plus
* one (for the current pending command).
*/
int max_iterations = bnx2x_exe_queue_length(&o->exe_queue) + 1;
while (!bnx2x_exe_queue_empty(&o->exe_queue) &&
max_iterations--) {
/* Wait for the current command to complete */
rc = raw->wait_comp(bp, raw);
if (rc)
return rc;
/* Make a next step */
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
rc = __bnx2x_vlan_mac_execute_step(bp,
p->vlan_mac_obj,
&p->ramrod_flags);
if (rc < 0)
return rc;
}
return 0;
}
return rc;
}
/**
* bnx2x_vlan_mac_del_all - delete elements with given vlan_mac_flags spec
*
* @bp: device handle
* @o:
* @vlan_mac_flags:
* @ramrod_flags: execution flags to be used for this deletion
*
* if the last operation has completed successfully and there are no
* more elements left, positive value if the last operation has completed
* successfully and there are more previously configured elements, negative
* value is current operation has failed.
*/
static int bnx2x_vlan_mac_del_all(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *o,
unsigned long *vlan_mac_flags,
unsigned long *ramrod_flags)
{
struct bnx2x_vlan_mac_registry_elem *pos = NULL;
struct bnx2x_vlan_mac_ramrod_params p;
struct bnx2x_exe_queue_obj *exeq = &o->exe_queue;
struct bnx2x_exeq_elem *exeq_pos, *exeq_pos_n;
unsigned long flags;
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
int read_lock;
int rc = 0;
/* Clear pending commands first */
spin_lock_bh(&exeq->lock);
list_for_each_entry_safe(exeq_pos, exeq_pos_n, &exeq->exe_queue, link) {
flags = exeq_pos->cmd_data.vlan_mac.vlan_mac_flags;
if (BNX2X_VLAN_MAC_CMP_FLAGS(flags) ==
BNX2X_VLAN_MAC_CMP_FLAGS(*vlan_mac_flags)) {
rc = exeq->remove(bp, exeq->owner, exeq_pos);
if (rc) {
BNX2X_ERR("Failed to remove command\n");
spin_unlock_bh(&exeq->lock);
return rc;
}
list_del(&exeq_pos->link);
bnx2x_exe_queue_free_elem(bp, exeq_pos);
}
}
spin_unlock_bh(&exeq->lock);
/* Prepare a command request */
memset(&p, 0, sizeof(p));
p.vlan_mac_obj = o;
p.ramrod_flags = *ramrod_flags;
p.user_req.cmd = BNX2X_VLAN_MAC_DEL;
/* Add all but the last VLAN-MAC to the execution queue without actually
* execution anything.
*/
__clear_bit(RAMROD_COMP_WAIT, &p.ramrod_flags);
__clear_bit(RAMROD_EXEC, &p.ramrod_flags);
__clear_bit(RAMROD_CONT, &p.ramrod_flags);
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
DP(BNX2X_MSG_SP, "vlan_mac_del_all -- taking vlan_mac_lock (reader)\n");
read_lock = bnx2x_vlan_mac_h_read_lock(bp, o);
if (read_lock != 0)
return read_lock;
list_for_each_entry(pos, &o->head, link) {
flags = pos->vlan_mac_flags;
if (BNX2X_VLAN_MAC_CMP_FLAGS(flags) ==
BNX2X_VLAN_MAC_CMP_FLAGS(*vlan_mac_flags)) {
p.user_req.vlan_mac_flags = pos->vlan_mac_flags;
memcpy(&p.user_req.u, &pos->u, sizeof(pos->u));
rc = bnx2x_config_vlan_mac(bp, &p);
if (rc < 0) {
BNX2X_ERR("Failed to add a new DEL command\n");
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
bnx2x_vlan_mac_h_read_unlock(bp, o);
return rc;
}
}
}
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
DP(BNX2X_MSG_SP, "vlan_mac_del_all -- releasing vlan_mac_lock (reader)\n");
bnx2x_vlan_mac_h_read_unlock(bp, o);
p.ramrod_flags = *ramrod_flags;
__set_bit(RAMROD_CONT, &p.ramrod_flags);
return bnx2x_config_vlan_mac(bp, &p);
}
static inline void bnx2x_init_raw_obj(struct bnx2x_raw_obj *raw, u8 cl_id,
u32 cid, u8 func_id, void *rdata, dma_addr_t rdata_mapping, int state,
unsigned long *pstate, bnx2x_obj_type type)
{
raw->func_id = func_id;
raw->cid = cid;
raw->cl_id = cl_id;
raw->rdata = rdata;
raw->rdata_mapping = rdata_mapping;
raw->state = state;
raw->pstate = pstate;
raw->obj_type = type;
raw->check_pending = bnx2x_raw_check_pending;
raw->clear_pending = bnx2x_raw_clear_pending;
raw->set_pending = bnx2x_raw_set_pending;
raw->wait_comp = bnx2x_raw_wait;
}
static inline void bnx2x_init_vlan_mac_common(struct bnx2x_vlan_mac_obj *o,
u8 cl_id, u32 cid, u8 func_id, void *rdata, dma_addr_t rdata_mapping,
int state, unsigned long *pstate, bnx2x_obj_type type,
struct bnx2x_credit_pool_obj *macs_pool,
struct bnx2x_credit_pool_obj *vlans_pool)
{
INIT_LIST_HEAD(&o->head);
bnx2x: Revising locking scheme for MAC configuration On very rare occasions, repeated load/unload stress test in the presence of our storage driver (bnx2i/bnx2fc) causes a kernel panic in bnx2x code (NULL pointer dereference). Stack traces indicate the issue happens during MAC configuration; thorough code review showed that indeed several races exist in which one thread can iterate over the list of configured MACs while another deletes entries from the same list. This patch adds a varient on the single-writer/Multiple-reader lock mechanism - It utilizes an already exsiting bottom-half lock, using it so that Whenever a writer is unable to continue due to the existence of another writer/reader, it pends its request for future deliverance. The writer / last readers will check for the existence of such requests and perform them instead of the original initiator. This prevents the writer from having to sleep while waiting for the lock to be accessible, which might cause deadlocks given the locks already held by the writer. Another result of this patch is that setting of Rx Mode is now made in sleepable context - Setting of Rx Mode is made under a bottom-half lock, which was always nontrivial for the bnx2x driver, as the HW/FW configuration requires wait for completions. Since sleep was impossible (due to the sleepless-context), various mechanisms were utilized to prevent the calling thread from sleep, but the truth was that when the caller thread (i.e, the one calling ndo_set_rx_mode()) returned, the Rx mode was still not set in HW/FW. bnx2x_set_rx_mode() will now overtly schedule for the Rx changes to be configured by the sp_rtnl_task which hold the RTNL lock and is sleepable context. Signed-off-by: Yuval Mintz <yuvalmin@broadcom.com> Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 08:30:59 -06:00
o->head_reader = 0;
o->head_exe_request = false;
o->saved_ramrod_flags = 0;
o->macs_pool = macs_pool;
o->vlans_pool = vlans_pool;
o->delete_all = bnx2x_vlan_mac_del_all;
o->restore = bnx2x_vlan_mac_restore;
o->complete = bnx2x_complete_vlan_mac;
o->wait = bnx2x_wait_vlan_mac;
bnx2x_init_raw_obj(&o->raw, cl_id, cid, func_id, rdata, rdata_mapping,
state, pstate, type);
}
void bnx2x_init_mac_obj(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *mac_obj,
u8 cl_id, u32 cid, u8 func_id, void *rdata,
dma_addr_t rdata_mapping, int state,
unsigned long *pstate, bnx2x_obj_type type,
struct bnx2x_credit_pool_obj *macs_pool)
{
union bnx2x_qable_obj *qable_obj = (union bnx2x_qable_obj *)mac_obj;
bnx2x_init_vlan_mac_common(mac_obj, cl_id, cid, func_id, rdata,
rdata_mapping, state, pstate, type,
macs_pool, NULL);
/* CAM credit pool handling */
mac_obj->get_credit = bnx2x_get_credit_mac;
mac_obj->put_credit = bnx2x_put_credit_mac;
mac_obj->get_cam_offset = bnx2x_get_cam_offset_mac;
mac_obj->put_cam_offset = bnx2x_put_cam_offset_mac;
if (CHIP_IS_E1x(bp)) {
mac_obj->set_one_rule = bnx2x_set_one_mac_e1x;
mac_obj->check_del = bnx2x_check_mac_del;
mac_obj->check_add = bnx2x_check_mac_add;
mac_obj->check_move = bnx2x_check_move_always_err;
mac_obj->ramrod_cmd = RAMROD_CMD_ID_ETH_SET_MAC;
/* Exe Queue */
bnx2x_exe_queue_init(bp,
&mac_obj->exe_queue, 1, qable_obj,
bnx2x_validate_vlan_mac,
bnx2x_remove_vlan_mac,
bnx2x_optimize_vlan_mac,
bnx2x_execute_vlan_mac,
bnx2x_exeq_get_mac);
} else {
mac_obj->set_one_rule = bnx2x_set_one_mac_e2;
mac_obj->check_del = bnx2x_check_mac_del;
mac_obj->check_add = bnx2x_check_mac_add;
mac_obj->check_move = bnx2x_check_move;
mac_obj->ramrod_cmd =
RAMROD_CMD_ID_ETH_CLASSIFICATION_RULES;
mac_obj->get_n_elements = bnx2x_get_n_elements;
/* Exe Queue */
bnx2x_exe_queue_init(bp,
&mac_obj->exe_queue, CLASSIFY_RULES_COUNT,
qable_obj, bnx2x_validate_vlan_mac,
bnx2x_remove_vlan_mac,
bnx2x_optimize_vlan_mac,
bnx2x_execute_vlan_mac,
bnx2x_exeq_get_mac);
}
}
void bnx2x_init_vlan_obj(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *vlan_obj,
u8 cl_id, u32 cid, u8 func_id, void *rdata,
dma_addr_t rdata_mapping, int state,
unsigned long *pstate, bnx2x_obj_type type,
struct bnx2x_credit_pool_obj *vlans_pool)
{
union bnx2x_qable_obj *qable_obj = (union bnx2x_qable_obj *)vlan_obj;
bnx2x_init_vlan_mac_common(vlan_obj, cl_id, cid, func_id, rdata,
rdata_mapping, state, pstate, type, NULL,
vlans_pool);
vlan_obj->get_credit = bnx2x_get_credit_vlan;
vlan_obj->put_credit = bnx2x_put_credit_vlan;
vlan_obj->get_cam_offset = bnx2x_get_cam_offset_vlan;
vlan_obj->put_cam_offset = bnx2x_put_cam_offset_vlan;
if (CHIP_IS_E1x(bp)) {
BNX2X_ERR("Do not support chips others than E2 and newer\n");
BUG();
} else {
vlan_obj->set_one_rule = bnx2x_set_one_vlan_e2;
vlan_obj->check_del = bnx2x_check_vlan_del;
vlan_obj->check_add = bnx2x_check_vlan_add;
vlan_obj->check_move = bnx2x_check_move;
vlan_obj->ramrod_cmd =
RAMROD_CMD_ID_ETH_CLASSIFICATION_RULES;
vlan_obj->get_n_elements = bnx2x_get_n_elements;
/* Exe Queue */
bnx2x_exe_queue_init(bp,
&vlan_obj->exe_queue, CLASSIFY_RULES_COUNT,
qable_obj, bnx2x_validate_vlan_mac,
bnx2x_remove_vlan_mac,
bnx2x_optimize_vlan_mac,
bnx2x_execute_vlan_mac,
bnx2x_exeq_get_vlan);
}
}
void bnx2x_init_vlan_mac_obj(struct bnx2x *bp,
struct bnx2x_vlan_mac_obj *vlan_mac_obj,
u8 cl_id, u32 cid, u8 func_id, void *rdata,
dma_addr_t rdata_mapping, int state,
unsigned long *pstate, bnx2x_obj_type type,
struct bnx2x_credit_pool_obj *macs_pool,
struct bnx2x_credit_pool_obj *vlans_pool)
{
union bnx2x_qable_obj *qable_obj =
(union bnx2x_qable_obj *)vlan_mac_obj;
bnx2x_init_vlan_mac_common(vlan_mac_obj, cl_id, cid, func_id, rdata,
rdata_mapping, state, pstate, type,
macs_pool, vlans_pool);
/* CAM pool handling */
vlan_mac_obj->get_credit = bnx2x_get_credit_vlan_mac;
vlan_mac_obj->put_credit = bnx2x_put_credit_vlan_mac;
/* CAM offset is relevant for 57710 and 57711 chips only which have a
* single CAM for both MACs and VLAN-MAC pairs. So the offset
* will be taken from MACs' pool object only.
*/
vlan_mac_obj->get_cam_offset = bnx2x_get_cam_offset_mac;
vlan_mac_obj->put_cam_offset = bnx2x_put_cam_offset_mac;
if (CHIP_IS_E1(bp)) {
BNX2X_ERR("Do not support chips others than E2\n");
BUG();
} else if (CHIP_IS_E1H(bp)) {
vlan_mac_obj->set_one_rule = bnx2x_set_one_vlan_mac_e1h;
vlan_mac_obj->check_del = bnx2x_check_vlan_mac_del;
vlan_mac_obj->check_add = bnx2x_check_vlan_mac_add;
vlan_mac_obj->check_move = bnx2x_check_move_always_err;
vlan_mac_obj->ramrod_cmd = RAMROD_CMD_ID_ETH_SET_MAC;
/* Exe Queue */
bnx2x_exe_queue_init(bp,
&vlan_mac_obj->exe_queue, 1, qable_obj,
bnx2x_validate_vlan_mac,
bnx2x_remove_vlan_mac,
bnx2x_optimize_vlan_mac,
bnx2x_execute_vlan_mac,
bnx2x_exeq_get_vlan_mac);
} else {
vlan_mac_obj->set_one_rule = bnx2x_set_one_vlan_mac_e2;
vlan_mac_obj->check_del = bnx2x_check_vlan_mac_del;
vlan_mac_obj->check_add = bnx2x_check_vlan_mac_add;
vlan_mac_obj->check_move = bnx2x_check_move;
vlan_mac_obj->ramrod_cmd =
RAMROD_CMD_ID_ETH_CLASSIFICATION_RULES;
/* Exe Queue */
bnx2x_exe_queue_init(bp,
&vlan_mac_obj->exe_queue,
CLASSIFY_RULES_COUNT,
qable_obj, bnx2x_validate_vlan_mac,
bnx2x_remove_vlan_mac,
bnx2x_optimize_vlan_mac,
bnx2x_execute_vlan_mac,
bnx2x_exeq_get_vlan_mac);
}
}
/* RX_MODE verbs: DROP_ALL/ACCEPT_ALL/ACCEPT_ALL_MULTI/ACCEPT_ALL_VLAN/NORMAL */
static inline void __storm_memset_mac_filters(struct bnx2x *bp,
struct tstorm_eth_mac_filter_config *mac_filters,
u16 pf_id)
{
size_t size = sizeof(struct tstorm_eth_mac_filter_config);
u32 addr = BAR_TSTRORM_INTMEM +
TSTORM_MAC_FILTER_CONFIG_OFFSET(pf_id);
__storm_memset_struct(bp, addr, size, (u32 *)mac_filters);
}
static int bnx2x_set_rx_mode_e1x(struct bnx2x *bp,
struct bnx2x_rx_mode_ramrod_params *p)
{
/* update the bp MAC filter structure */
u32 mask = (1 << p->cl_id);
struct tstorm_eth_mac_filter_config *mac_filters =
(struct tstorm_eth_mac_filter_config *)p->rdata;
/* initial setting is drop-all */
u8 drop_all_ucast = 1, drop_all_mcast = 1;
u8 accp_all_ucast = 0, accp_all_bcast = 0, accp_all_mcast = 0;
u8 unmatched_unicast = 0;
/* In e1x there we only take into account rx accept flag since tx switching
* isn't enabled. */
if (test_bit(BNX2X_ACCEPT_UNICAST, &p->rx_accept_flags))
/* accept matched ucast */
drop_all_ucast = 0;
if (test_bit(BNX2X_ACCEPT_MULTICAST, &p->rx_accept_flags))
/* accept matched mcast */
drop_all_mcast = 0;
if (test_bit(BNX2X_ACCEPT_ALL_UNICAST, &p->rx_accept_flags)) {
/* accept all mcast */
drop_all_ucast = 0;
accp_all_ucast = 1;
}
if (test_bit(BNX2X_ACCEPT_ALL_MULTICAST, &p->rx_accept_flags)) {
/* accept all mcast */
drop_all_mcast = 0;
accp_all_mcast = 1;
}
if (test_bit(BNX2X_ACCEPT_BROADCAST, &p->rx_accept_flags))
/* accept (all) bcast */
accp_all_bcast = 1;
if (test_bit(BNX2X_ACCEPT_UNMATCHED, &p->rx_accept_flags))
/* accept unmatched unicasts */
unmatched_unicast = 1;
mac_filters->ucast_drop_all = drop_all_ucast ?
mac_filters->ucast_drop_all | mask :
mac_filters->ucast_drop_all & ~mask;
mac_filters->mcast_drop_all = drop_all_mcast ?
mac_filters->mcast_drop_all | mask :
mac_filters->mcast_drop_all & ~mask;
mac_filters->ucast_accept_all = accp_all_ucast ?
mac_filters->ucast_accept_all | mask :
mac_filters->ucast_accept_all & ~mask;
mac_filters->mcast_accept_all = accp_all_mcast ?
mac_filters->mcast_accept_all | mask :
mac_filters->mcast_accept_all & ~mask;
mac_filters->bcast_accept_all = accp_all_bcast ?
mac_filters->bcast_accept_all | mask :
mac_filters->bcast_accept_all & ~mask;
mac_filters->unmatched_unicast = unmatched_unicast ?
mac_filters->unmatched_unicast | mask :
mac_filters->unmatched_unicast & ~mask;
DP(BNX2X_MSG_SP, "drop_ucast 0x%x\ndrop_mcast 0x%x\n accp_ucast 0x%x\n"
"accp_mcast 0x%x\naccp_bcast 0x%x\n",
mac_filters->ucast_drop_all, mac_filters->mcast_drop_all,
mac_filters->ucast_accept_all, mac_filters->mcast_accept_all,
mac_filters->bcast_accept_all);
/* write the MAC filter structure*/
__storm_memset_mac_filters(bp, mac_filters, p->func_id);
/* The operation is completed */
clear_bit(p->state, p->pstate);
smp_mb__after_atomic();
return 0;
}
/* Setup ramrod data */
static inline void bnx2x_rx_mode_set_rdata_hdr_e2(u32 cid,
struct eth_classify_header *hdr,
u8 rule_cnt)
{
hdr->echo = cpu_to_le32(cid);
hdr->rule_cnt = rule_cnt;
}
static inline void bnx2x_rx_mode_set_cmd_state_e2(struct bnx2x *bp,
unsigned long *accept_flags,
struct eth_filter_rules_cmd *cmd,
bool clear_accept_all)
{
u16 state;
/* start with 'drop-all' */
state = ETH_FILTER_RULES_CMD_UCAST_DROP_ALL |
ETH_FILTER_RULES_CMD_MCAST_DROP_ALL;
if (test_bit(BNX2X_ACCEPT_UNICAST, accept_flags))
state &= ~ETH_FILTER_RULES_CMD_UCAST_DROP_ALL;
if (test_bit(BNX2X_ACCEPT_MULTICAST, accept_flags))
state &= ~ETH_FILTER_RULES_CMD_MCAST_DROP_ALL;
if (test_bit(BNX2X_ACCEPT_ALL_UNICAST, accept_flags)) {
state &= ~ETH_FILTER_RULES_CMD_UCAST_DROP_ALL;
state |= ETH_FILTER_RULES_CMD_UCAST_ACCEPT_ALL;
}
if (test_bit(BNX2X_ACCEPT_ALL_MULTICAST, accept_flags)) {
state |= ETH_FILTER_RULES_CMD_MCAST_ACCEPT_ALL;
state &= ~ETH_FILTER_RULES_CMD_MCAST_DROP_ALL;
}
if (test_bit(BNX2X_ACCEPT_BROADCAST, accept_flags))
state |= ETH_FILTER_RULES_CMD_BCAST_ACCEPT_ALL;
if (test_bit(BNX2X_ACCEPT_UNMATCHED, accept_flags)) {
state &= ~ETH_FILTER_RULES_CMD_UCAST_DROP_ALL;
state |= ETH_FILTER_RULES_CMD_UCAST_ACCEPT_UNMATCHED;
}
if (test_bit(BNX2X_ACCEPT_ANY_VLAN, accept_flags))
state |= ETH_FILTER_RULES_CMD_ACCEPT_ANY_VLAN;
/* Clear ACCEPT_ALL_XXX flags for FCoE L2 Queue */
if (clear_accept_all) {
state &= ~ETH_FILTER_RULES_CMD_MCAST_ACCEPT_ALL;
state &= ~ETH_FILTER_RULES_CMD_BCAST_ACCEPT_ALL;
state &= ~ETH_FILTER_RULES_CMD_UCAST_ACCEPT_ALL;
state &= ~ETH_FILTER_RULES_CMD_UCAST_ACCEPT_UNMATCHED;
}
cmd->state = cpu_to_le16(state);
}
static int bnx2x_set_rx_mode_e2(struct bnx2x *bp,
struct bnx2x_rx_mode_ramrod_params *p)
{
struct eth_filter_rules_ramrod_data *data = p->rdata;
int rc;
u8 rule_idx = 0;
/* Reset the ramrod data buffer */
memset(data, 0, sizeof(*data));
/* Setup ramrod data */
/* Tx (internal switching) */
if (test_bit(RAMROD_TX, &p->ramrod_flags)) {
data->rules[rule_idx].client_id = p->cl_id;
data->rules[rule_idx].func_id = p->func_id;
data->rules[rule_idx].cmd_general_data =
ETH_FILTER_RULES_CMD_TX_CMD;
bnx2x_rx_mode_set_cmd_state_e2(bp, &p->tx_accept_flags,
&(data->rules[rule_idx++]),
false);
}
/* Rx */
if (test_bit(RAMROD_RX, &p->ramrod_flags)) {
data->rules[rule_idx].client_id = p->cl_id;
data->rules[rule_idx].func_id = p->func_id;
data->rules[rule_idx].cmd_general_data =
ETH_FILTER_RULES_CMD_RX_CMD;
bnx2x_rx_mode_set_cmd_state_e2(bp, &p->rx_accept_flags,
&(data->rules[rule_idx++]),
false);
}
/* If FCoE Queue configuration has been requested configure the Rx and
* internal switching modes for this queue in separate rules.
*
* FCoE queue shell never be set to ACCEPT_ALL packets of any sort:
* MCAST_ALL, UCAST_ALL, BCAST_ALL and UNMATCHED.
*/
if (test_bit(BNX2X_RX_MODE_FCOE_ETH, &p->rx_mode_flags)) {
/* Tx (internal switching) */
if (test_bit(RAMROD_TX, &p->ramrod_flags)) {
data->rules[rule_idx].client_id = bnx2x_fcoe(bp, cl_id);
data->rules[rule_idx].func_id = p->func_id;
data->rules[rule_idx].cmd_general_data =
ETH_FILTER_RULES_CMD_TX_CMD;
bnx2x_rx_mode_set_cmd_state_e2(bp, &p->tx_accept_flags,
&(data->rules[rule_idx]),
true);
rule_idx++;
}
/* Rx */
if (test_bit(RAMROD_RX, &p->ramrod_flags)) {
data->rules[rule_idx].client_id = bnx2x_fcoe(bp, cl_id);
data->rules[rule_idx].func_id = p->func_id;
data->rules[rule_idx].cmd_general_data =
ETH_FILTER_RULES_CMD_RX_CMD;
bnx2x_rx_mode_set_cmd_state_e2(bp, &p->rx_accept_flags,
&(data->rules[rule_idx]),
true);
rule_idx++;
}
}
/* Set the ramrod header (most importantly - number of rules to
* configure).
*/
bnx2x_rx_mode_set_rdata_hdr_e2(p->cid, &data->header, rule_idx);
DP(BNX2X_MSG_SP, "About to configure %d rules, rx_accept_flags 0x%lx, tx_accept_flags 0x%lx\n",
data->header.rule_cnt, p->rx_accept_flags,
p->tx_accept_flags);
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
/* Send a ramrod */
rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_FILTER_RULES, p->cid,
U64_HI(p->rdata_mapping),
U64_LO(p->rdata_mapping),
ETH_CONNECTION_TYPE);
if (rc)
return rc;
/* Ramrod completion is pending */
return 1;
}
static int bnx2x_wait_rx_mode_comp_e2(struct bnx2x *bp,
struct bnx2x_rx_mode_ramrod_params *p)
{
return bnx2x_state_wait(bp, p->state, p->pstate);
}
static int bnx2x_empty_rx_mode_wait(struct bnx2x *bp,
struct bnx2x_rx_mode_ramrod_params *p)
{
/* Do nothing */
return 0;
}
int bnx2x_config_rx_mode(struct bnx2x *bp,
struct bnx2x_rx_mode_ramrod_params *p)
{
int rc;
/* Configure the new classification in the chip */
rc = p->rx_mode_obj->config_rx_mode(bp, p);
if (rc < 0)
return rc;
/* Wait for a ramrod completion if was requested */
if (test_bit(RAMROD_COMP_WAIT, &p->ramrod_flags)) {
rc = p->rx_mode_obj->wait_comp(bp, p);
if (rc)
return rc;
}
return rc;
}
void bnx2x_init_rx_mode_obj(struct bnx2x *bp,
struct bnx2x_rx_mode_obj *o)
{
if (CHIP_IS_E1x(bp)) {
o->wait_comp = bnx2x_empty_rx_mode_wait;
o->config_rx_mode = bnx2x_set_rx_mode_e1x;
} else {
o->wait_comp = bnx2x_wait_rx_mode_comp_e2;
o->config_rx_mode = bnx2x_set_rx_mode_e2;
}
}
/********************* Multicast verbs: SET, CLEAR ****************************/
static inline u8 bnx2x_mcast_bin_from_mac(u8 *mac)
{
return (crc32c_le(0, mac, ETH_ALEN) >> 24) & 0xff;
}
struct bnx2x_mcast_mac_elem {
struct list_head link;
u8 mac[ETH_ALEN];
u8 pad[2]; /* For a natural alignment of the following buffer */
};
struct bnx2x_mcast_bin_elem {
struct list_head link;
int bin;
int type; /* BNX2X_MCAST_CMD_SET_{ADD, DEL} */
};
union bnx2x_mcast_elem {
struct bnx2x_mcast_bin_elem bin_elem;
struct bnx2x_mcast_mac_elem mac_elem;
};
struct bnx2x_mcast_elem_group {
struct list_head mcast_group_link;
union bnx2x_mcast_elem mcast_elems[];
};
#define MCAST_MAC_ELEMS_PER_PG \
((PAGE_SIZE - sizeof(struct bnx2x_mcast_elem_group)) / \
sizeof(union bnx2x_mcast_elem))
struct bnx2x_pending_mcast_cmd {
struct list_head link;
struct list_head group_head;
int type; /* BNX2X_MCAST_CMD_X */
union {
struct list_head macs_head;
u32 macs_num; /* Needed for DEL command */
int next_bin; /* Needed for RESTORE flow with aprox match */
} data;
bool set_convert; /* in case type == BNX2X_MCAST_CMD_SET, this is set
* when macs_head had been converted to a list of
* bnx2x_mcast_bin_elem.
*/
bool done; /* set to true, when the command has been handled,
* practically used in 57712 handling only, where one pending
* command may be handled in a few operations. As long as for
* other chips every operation handling is completed in a
* single ramrod, there is no need to utilize this field.
*/
};
static int bnx2x_mcast_wait(struct bnx2x *bp,
struct bnx2x_mcast_obj *o)
{
if (bnx2x_state_wait(bp, o->sched_state, o->raw.pstate) ||
o->raw.wait_comp(bp, &o->raw))
return -EBUSY;
return 0;
}
static void bnx2x_free_groups(struct list_head *mcast_group_list)
{
struct bnx2x_mcast_elem_group *current_mcast_group;
while (!list_empty(mcast_group_list)) {
current_mcast_group = list_first_entry(mcast_group_list,
struct bnx2x_mcast_elem_group,
mcast_group_link);
list_del(&current_mcast_group->mcast_group_link);
free_page((unsigned long)current_mcast_group);
}
}
static int bnx2x_mcast_enqueue_cmd(struct bnx2x *bp,
struct bnx2x_mcast_obj *o,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_pending_mcast_cmd *new_cmd;
struct bnx2x_mcast_list_elem *pos;
struct bnx2x_mcast_elem_group *elem_group;
struct bnx2x_mcast_mac_elem *mac_elem;
int total_elems = 0, macs_list_len = 0, offset = 0;
/* When adding MACs we'll need to store their values */
if (cmd == BNX2X_MCAST_CMD_ADD || cmd == BNX2X_MCAST_CMD_SET)
macs_list_len = p->mcast_list_len;
/* If the command is empty ("handle pending commands only"), break */
if (!p->mcast_list_len)
return 0;
/* Add mcast is called under spin_lock, thus calling with GFP_ATOMIC */
new_cmd = kzalloc(sizeof(*new_cmd), GFP_ATOMIC);
if (!new_cmd)
return -ENOMEM;
INIT_LIST_HEAD(&new_cmd->data.macs_head);
INIT_LIST_HEAD(&new_cmd->group_head);
new_cmd->type = cmd;
new_cmd->done = false;
DP(BNX2X_MSG_SP, "About to enqueue a new %d command. macs_list_len=%d\n",
cmd, macs_list_len);
switch (cmd) {
case BNX2X_MCAST_CMD_ADD:
case BNX2X_MCAST_CMD_SET:
/* For a set command, we need to allocate sufficient memory for
* all the bins, since we can't analyze at this point how much
* memory would be required.
*/
total_elems = macs_list_len;
if (cmd == BNX2X_MCAST_CMD_SET) {
if (total_elems < BNX2X_MCAST_BINS_NUM)
total_elems = BNX2X_MCAST_BINS_NUM;
}
while (total_elems > 0) {
elem_group = (struct bnx2x_mcast_elem_group *)
__get_free_page(GFP_ATOMIC | __GFP_ZERO);
if (!elem_group) {
bnx2x_free_groups(&new_cmd->group_head);
kfree(new_cmd);
return -ENOMEM;
}
total_elems -= MCAST_MAC_ELEMS_PER_PG;
list_add_tail(&elem_group->mcast_group_link,
&new_cmd->group_head);
}
elem_group = list_first_entry(&new_cmd->group_head,
struct bnx2x_mcast_elem_group,
mcast_group_link);
list_for_each_entry(pos, &p->mcast_list, link) {
mac_elem = &elem_group->mcast_elems[offset].mac_elem;
memcpy(mac_elem->mac, pos->mac, ETH_ALEN);
/* Push the MACs of the current command into the pending
* command MACs list: FIFO
*/
list_add_tail(&mac_elem->link,
&new_cmd->data.macs_head);
offset++;
if (offset == MCAST_MAC_ELEMS_PER_PG) {
offset = 0;
elem_group = list_next_entry(elem_group,
mcast_group_link);
}
}
break;
case BNX2X_MCAST_CMD_DEL:
new_cmd->data.macs_num = p->mcast_list_len;
break;
case BNX2X_MCAST_CMD_RESTORE:
new_cmd->data.next_bin = 0;
break;
default:
kfree(new_cmd);
BNX2X_ERR("Unknown command: %d\n", cmd);
return -EINVAL;
}
/* Push the new pending command to the tail of the pending list: FIFO */
list_add_tail(&new_cmd->link, &o->pending_cmds_head);
o->set_sched(o);
return 1;
}
/**
* bnx2x_mcast_get_next_bin - get the next set bin (index)
*
* @o:
* @last: index to start looking from (including)
*
* returns the next found (set) bin or a negative value if none is found.
*/
static inline int bnx2x_mcast_get_next_bin(struct bnx2x_mcast_obj *o, int last)
{
int i, j, inner_start = last % BIT_VEC64_ELEM_SZ;
for (i = last / BIT_VEC64_ELEM_SZ; i < BNX2X_MCAST_VEC_SZ; i++) {
if (o->registry.aprox_match.vec[i])
for (j = inner_start; j < BIT_VEC64_ELEM_SZ; j++) {
int cur_bit = j + BIT_VEC64_ELEM_SZ * i;
if (BIT_VEC64_TEST_BIT(o->registry.aprox_match.
vec, cur_bit)) {
return cur_bit;
}
}
inner_start = 0;
}
/* None found */
return -1;
}
/**
* bnx2x_mcast_clear_first_bin - find the first set bin and clear it
*
* @o:
*
* returns the index of the found bin or -1 if none is found
*/
static inline int bnx2x_mcast_clear_first_bin(struct bnx2x_mcast_obj *o)
{
int cur_bit = bnx2x_mcast_get_next_bin(o, 0);
if (cur_bit >= 0)
BIT_VEC64_CLEAR_BIT(o->registry.aprox_match.vec, cur_bit);
return cur_bit;
}
static inline u8 bnx2x_mcast_get_rx_tx_flag(struct bnx2x_mcast_obj *o)
{
struct bnx2x_raw_obj *raw = &o->raw;
u8 rx_tx_flag = 0;
if ((raw->obj_type == BNX2X_OBJ_TYPE_TX) ||
(raw->obj_type == BNX2X_OBJ_TYPE_RX_TX))
rx_tx_flag |= ETH_MULTICAST_RULES_CMD_TX_CMD;
if ((raw->obj_type == BNX2X_OBJ_TYPE_RX) ||
(raw->obj_type == BNX2X_OBJ_TYPE_RX_TX))
rx_tx_flag |= ETH_MULTICAST_RULES_CMD_RX_CMD;
return rx_tx_flag;
}
static void bnx2x_mcast_set_one_rule_e2(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, int idx,
union bnx2x_mcast_config_data *cfg_data,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_raw_obj *r = &o->raw;
struct eth_multicast_rules_ramrod_data *data =
(struct eth_multicast_rules_ramrod_data *)(r->rdata);
u8 func_id = r->func_id;
u8 rx_tx_add_flag = bnx2x_mcast_get_rx_tx_flag(o);
int bin;
if ((cmd == BNX2X_MCAST_CMD_ADD) || (cmd == BNX2X_MCAST_CMD_RESTORE) ||
(cmd == BNX2X_MCAST_CMD_SET_ADD))
rx_tx_add_flag |= ETH_MULTICAST_RULES_CMD_IS_ADD;
data->rules[idx].cmd_general_data |= rx_tx_add_flag;
/* Get a bin and update a bins' vector */
switch (cmd) {
case BNX2X_MCAST_CMD_ADD:
bin = bnx2x_mcast_bin_from_mac(cfg_data->mac);
BIT_VEC64_SET_BIT(o->registry.aprox_match.vec, bin);
break;
case BNX2X_MCAST_CMD_DEL:
/* If there were no more bins to clear
* (bnx2x_mcast_clear_first_bin() returns -1) then we would
* clear any (0xff) bin.
* See bnx2x_mcast_validate_e2() for explanation when it may
* happen.
*/
bin = bnx2x_mcast_clear_first_bin(o);
break;
case BNX2X_MCAST_CMD_RESTORE:
bin = cfg_data->bin;
break;
case BNX2X_MCAST_CMD_SET_ADD:
bin = cfg_data->bin;
BIT_VEC64_SET_BIT(o->registry.aprox_match.vec, bin);
break;
case BNX2X_MCAST_CMD_SET_DEL:
bin = cfg_data->bin;
BIT_VEC64_CLEAR_BIT(o->registry.aprox_match.vec, bin);
break;
default:
BNX2X_ERR("Unknown command: %d\n", cmd);
return;
}
DP(BNX2X_MSG_SP, "%s bin %d\n",
((rx_tx_add_flag & ETH_MULTICAST_RULES_CMD_IS_ADD) ?
"Setting" : "Clearing"), bin);
data->rules[idx].bin_id = (u8)bin;
data->rules[idx].func_id = func_id;
data->rules[idx].engine_id = o->engine_id;
}
/**
* bnx2x_mcast_handle_restore_cmd_e2 - restore configuration from the registry
*
* @bp: device handle
* @o:
* @start_bin: index in the registry to start from (including)
* @rdata_idx: index in the ramrod data to start from
*
* returns last handled bin index or -1 if all bins have been handled
*/
static inline int bnx2x_mcast_handle_restore_cmd_e2(
struct bnx2x *bp, struct bnx2x_mcast_obj *o , int start_bin,
int *rdata_idx)
{
int cur_bin, cnt = *rdata_idx;
union bnx2x_mcast_config_data cfg_data = {NULL};
/* go through the registry and configure the bins from it */
for (cur_bin = bnx2x_mcast_get_next_bin(o, start_bin); cur_bin >= 0;
cur_bin = bnx2x_mcast_get_next_bin(o, cur_bin + 1)) {
cfg_data.bin = (u8)cur_bin;
o->set_one_rule(bp, o, cnt, &cfg_data,
BNX2X_MCAST_CMD_RESTORE);
cnt++;
DP(BNX2X_MSG_SP, "About to configure a bin %d\n", cur_bin);
/* Break if we reached the maximum number
* of rules.
*/
if (cnt >= o->max_cmd_len)
break;
}
*rdata_idx = cnt;
return cur_bin;
}
static inline void bnx2x_mcast_hdl_pending_add_e2(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, struct bnx2x_pending_mcast_cmd *cmd_pos,
int *line_idx)
{
struct bnx2x_mcast_mac_elem *pmac_pos, *pmac_pos_n;
int cnt = *line_idx;
union bnx2x_mcast_config_data cfg_data = {NULL};
list_for_each_entry_safe(pmac_pos, pmac_pos_n, &cmd_pos->data.macs_head,
link) {
cfg_data.mac = &pmac_pos->mac[0];
o->set_one_rule(bp, o, cnt, &cfg_data, cmd_pos->type);
cnt++;
DP(BNX2X_MSG_SP, "About to configure %pM mcast MAC\n",
pmac_pos->mac);
list_del(&pmac_pos->link);
/* Break if we reached the maximum number
* of rules.
*/
if (cnt >= o->max_cmd_len)
break;
}
*line_idx = cnt;
/* if no more MACs to configure - we are done */
if (list_empty(&cmd_pos->data.macs_head))
cmd_pos->done = true;
}
static inline void bnx2x_mcast_hdl_pending_del_e2(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, struct bnx2x_pending_mcast_cmd *cmd_pos,
int *line_idx)
{
int cnt = *line_idx;
while (cmd_pos->data.macs_num) {
o->set_one_rule(bp, o, cnt, NULL, cmd_pos->type);
cnt++;
cmd_pos->data.macs_num--;
DP(BNX2X_MSG_SP, "Deleting MAC. %d left,cnt is %d\n",
cmd_pos->data.macs_num, cnt);
/* Break if we reached the maximum
* number of rules.
*/
if (cnt >= o->max_cmd_len)
break;
}
*line_idx = cnt;
/* If we cleared all bins - we are done */
if (!cmd_pos->data.macs_num)
cmd_pos->done = true;
}
static inline void bnx2x_mcast_hdl_pending_restore_e2(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, struct bnx2x_pending_mcast_cmd *cmd_pos,
int *line_idx)
{
cmd_pos->data.next_bin = o->hdl_restore(bp, o, cmd_pos->data.next_bin,
line_idx);
if (cmd_pos->data.next_bin < 0)
/* If o->set_restore returned -1 we are done */
cmd_pos->done = true;
else
/* Start from the next bin next time */
cmd_pos->data.next_bin++;
}
static void
bnx2x_mcast_hdl_pending_set_e2_convert(struct bnx2x *bp,
struct bnx2x_mcast_obj *o,
struct bnx2x_pending_mcast_cmd *cmd_pos)
{
u64 cur[BNX2X_MCAST_VEC_SZ], req[BNX2X_MCAST_VEC_SZ];
struct bnx2x_mcast_mac_elem *pmac_pos, *pmac_pos_n;
struct bnx2x_mcast_bin_elem *p_item;
struct bnx2x_mcast_elem_group *elem_group;
int cnt = 0, mac_cnt = 0, offset = 0, i;
memset(req, 0, sizeof(u64) * BNX2X_MCAST_VEC_SZ);
memcpy(cur, o->registry.aprox_match.vec,
sizeof(u64) * BNX2X_MCAST_VEC_SZ);
/* Fill `current' with the required set of bins to configure */
list_for_each_entry_safe(pmac_pos, pmac_pos_n, &cmd_pos->data.macs_head,
link) {
int bin = bnx2x_mcast_bin_from_mac(pmac_pos->mac);
DP(BNX2X_MSG_SP, "Set contains %pM mcast MAC\n",
pmac_pos->mac);
BIT_VEC64_SET_BIT(req, bin);
list_del(&pmac_pos->link);
mac_cnt++;
}
/* We no longer have use for the MACs; Need to re-use memory for
* a list that will be used to configure bins.
*/
cmd_pos->set_convert = true;
INIT_LIST_HEAD(&cmd_pos->data.macs_head);
elem_group = list_first_entry(&cmd_pos->group_head,
struct bnx2x_mcast_elem_group,
mcast_group_link);
for (i = 0; i < BNX2X_MCAST_BINS_NUM; i++) {
bool b_current = !!BIT_VEC64_TEST_BIT(cur, i);
bool b_required = !!BIT_VEC64_TEST_BIT(req, i);
if (b_current == b_required)
continue;
p_item = &elem_group->mcast_elems[offset].bin_elem;
p_item->bin = i;
p_item->type = b_required ? BNX2X_MCAST_CMD_SET_ADD
: BNX2X_MCAST_CMD_SET_DEL;
list_add_tail(&p_item->link , &cmd_pos->data.macs_head);
cnt++;
offset++;
if (offset == MCAST_MAC_ELEMS_PER_PG) {
offset = 0;
elem_group = list_next_entry(elem_group,
mcast_group_link);
}
}
/* We now definitely know how many commands are hiding here.
* Also need to correct the disruption we've added to guarantee this
* would be enqueued.
*/
o->total_pending_num -= (o->max_cmd_len + mac_cnt);
o->total_pending_num += cnt;
DP(BNX2X_MSG_SP, "o->total_pending_num=%d\n", o->total_pending_num);
}
static void
bnx2x_mcast_hdl_pending_set_e2(struct bnx2x *bp,
struct bnx2x_mcast_obj *o,
struct bnx2x_pending_mcast_cmd *cmd_pos,
int *cnt)
{
union bnx2x_mcast_config_data cfg_data = {NULL};
struct bnx2x_mcast_bin_elem *p_item, *p_item_n;
/* This is actually a 2-part scheme - it starts by converting the MACs
* into a list of bins to be added/removed, and correcting the numbers
* on the object. this is now allowed, as we're now sure that all
* previous configured requests have already applied.
* The second part is actually adding rules for the newly introduced
* entries [like all the rest of the hdl_pending functions].
*/
if (!cmd_pos->set_convert)
bnx2x_mcast_hdl_pending_set_e2_convert(bp, o, cmd_pos);
list_for_each_entry_safe(p_item, p_item_n, &cmd_pos->data.macs_head,
link) {
cfg_data.bin = (u8)p_item->bin;
o->set_one_rule(bp, o, *cnt, &cfg_data, p_item->type);
(*cnt)++;
list_del(&p_item->link);
/* Break if we reached the maximum number of rules. */
if (*cnt >= o->max_cmd_len)
break;
}
/* if no more MACs to configure - we are done */
if (list_empty(&cmd_pos->data.macs_head))
cmd_pos->done = true;
}
static inline int bnx2x_mcast_handle_pending_cmds_e2(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p)
{
struct bnx2x_pending_mcast_cmd *cmd_pos, *cmd_pos_n;
int cnt = 0;
struct bnx2x_mcast_obj *o = p->mcast_obj;
list_for_each_entry_safe(cmd_pos, cmd_pos_n, &o->pending_cmds_head,
link) {
switch (cmd_pos->type) {
case BNX2X_MCAST_CMD_ADD:
bnx2x_mcast_hdl_pending_add_e2(bp, o, cmd_pos, &cnt);
break;
case BNX2X_MCAST_CMD_DEL:
bnx2x_mcast_hdl_pending_del_e2(bp, o, cmd_pos, &cnt);
break;
case BNX2X_MCAST_CMD_RESTORE:
bnx2x_mcast_hdl_pending_restore_e2(bp, o, cmd_pos,
&cnt);
break;
case BNX2X_MCAST_CMD_SET:
bnx2x_mcast_hdl_pending_set_e2(bp, o, cmd_pos, &cnt);
break;
default:
BNX2X_ERR("Unknown command: %d\n", cmd_pos->type);
return -EINVAL;
}
/* If the command has been completed - remove it from the list
* and free the memory
*/
if (cmd_pos->done) {
list_del(&cmd_pos->link);
bnx2x_free_groups(&cmd_pos->group_head);
kfree(cmd_pos);
}
/* Break if we reached the maximum number of rules */
if (cnt >= o->max_cmd_len)
break;
}
return cnt;
}
static inline void bnx2x_mcast_hdl_add(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, struct bnx2x_mcast_ramrod_params *p,
int *line_idx)
{
struct bnx2x_mcast_list_elem *mlist_pos;
union bnx2x_mcast_config_data cfg_data = {NULL};
int cnt = *line_idx;
list_for_each_entry(mlist_pos, &p->mcast_list, link) {
cfg_data.mac = mlist_pos->mac;
o->set_one_rule(bp, o, cnt, &cfg_data, BNX2X_MCAST_CMD_ADD);
cnt++;
DP(BNX2X_MSG_SP, "About to configure %pM mcast MAC\n",
mlist_pos->mac);
}
*line_idx = cnt;
}
static inline void bnx2x_mcast_hdl_del(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, struct bnx2x_mcast_ramrod_params *p,
int *line_idx)
{
int cnt = *line_idx, i;
for (i = 0; i < p->mcast_list_len; i++) {
o->set_one_rule(bp, o, cnt, NULL, BNX2X_MCAST_CMD_DEL);
cnt++;
DP(BNX2X_MSG_SP, "Deleting MAC. %d left\n",
p->mcast_list_len - i - 1);
}
*line_idx = cnt;
}
/**
* bnx2x_mcast_handle_current_cmd -
*
* @bp: device handle
* @p:
* @cmd:
* @start_cnt: first line in the ramrod data that may be used
*
* This function is called iff there is enough place for the current command in
* the ramrod data.
* Returns number of lines filled in the ramrod data in total.
*/
static inline int bnx2x_mcast_handle_current_cmd(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd,
int start_cnt)
{
struct bnx2x_mcast_obj *o = p->mcast_obj;
int cnt = start_cnt;
DP(BNX2X_MSG_SP, "p->mcast_list_len=%d\n", p->mcast_list_len);
switch (cmd) {
case BNX2X_MCAST_CMD_ADD:
bnx2x_mcast_hdl_add(bp, o, p, &cnt);
break;
case BNX2X_MCAST_CMD_DEL:
bnx2x_mcast_hdl_del(bp, o, p, &cnt);
break;
case BNX2X_MCAST_CMD_RESTORE:
o->hdl_restore(bp, o, 0, &cnt);
break;
default:
BNX2X_ERR("Unknown command: %d\n", cmd);
return -EINVAL;
}
/* The current command has been handled */
p->mcast_list_len = 0;
return cnt;
}
static int bnx2x_mcast_validate_e2(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_mcast_obj *o = p->mcast_obj;
int reg_sz = o->get_registry_size(o);
switch (cmd) {
/* DEL command deletes all currently configured MACs */
case BNX2X_MCAST_CMD_DEL:
o->set_registry_size(o, 0);
/* fall through */
/* RESTORE command will restore the entire multicast configuration */
case BNX2X_MCAST_CMD_RESTORE:
/* Here we set the approximate amount of work to do, which in
* fact may be only less as some MACs in postponed ADD
* command(s) scheduled before this command may fall into
* the same bin and the actual number of bins set in the
* registry would be less than we estimated here. See
* bnx2x_mcast_set_one_rule_e2() for further details.
*/
p->mcast_list_len = reg_sz;
break;
case BNX2X_MCAST_CMD_ADD:
case BNX2X_MCAST_CMD_CONT:
/* Here we assume that all new MACs will fall into new bins.
* However we will correct the real registry size after we
* handle all pending commands.
*/
o->set_registry_size(o, reg_sz + p->mcast_list_len);
break;
case BNX2X_MCAST_CMD_SET:
/* We can only learn how many commands would actually be used
* when this is being configured. So for now, simply guarantee
* the command will be enqueued [to refrain from adding logic
* that handles this and THEN learns it needs several ramrods].
* Just like for ADD/Cont, the mcast_list_len might be an over
* estimation; or even more so, since we don't take into
* account the possibility of removal of existing bins.
*/
o->set_registry_size(o, reg_sz + p->mcast_list_len);
o->total_pending_num += o->max_cmd_len;
break;
default:
BNX2X_ERR("Unknown command: %d\n", cmd);
return -EINVAL;
}
/* Increase the total number of MACs pending to be configured */
o->total_pending_num += p->mcast_list_len;
return 0;
}
static void bnx2x_mcast_revert_e2(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
int old_num_bins,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_mcast_obj *o = p->mcast_obj;
o->set_registry_size(o, old_num_bins);
o->total_pending_num -= p->mcast_list_len;
if (cmd == BNX2X_MCAST_CMD_SET)
o->total_pending_num -= o->max_cmd_len;
}
/**
* bnx2x_mcast_set_rdata_hdr_e2 - sets a header values
*
* @bp: device handle
* @p:
* @len: number of rules to handle
*/
static inline void bnx2x_mcast_set_rdata_hdr_e2(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
u8 len)
{
struct bnx2x_raw_obj *r = &p->mcast_obj->raw;
struct eth_multicast_rules_ramrod_data *data =
(struct eth_multicast_rules_ramrod_data *)(r->rdata);
data->header.echo = cpu_to_le32((r->cid & BNX2X_SWCID_MASK) |
(BNX2X_FILTER_MCAST_PENDING <<
BNX2X_SWCID_SHIFT));
data->header.rule_cnt = len;
}
/**
* bnx2x_mcast_refresh_registry_e2 - recalculate the actual number of set bins
*
* @bp: device handle
* @o:
*
* Recalculate the actual number of set bins in the registry using Brian
* Kernighan's algorithm: it's execution complexity is as a number of set bins.
*
* returns 0 for the compliance with bnx2x_mcast_refresh_registry_e1().
*/
static inline int bnx2x_mcast_refresh_registry_e2(struct bnx2x *bp,
struct bnx2x_mcast_obj *o)
{
int i, cnt = 0;
u64 elem;
for (i = 0; i < BNX2X_MCAST_VEC_SZ; i++) {
elem = o->registry.aprox_match.vec[i];
for (; elem; cnt++)
elem &= elem - 1;
}
o->set_registry_size(o, cnt);
return 0;
}
static int bnx2x_mcast_setup_e2(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_raw_obj *raw = &p->mcast_obj->raw;
struct bnx2x_mcast_obj *o = p->mcast_obj;
struct eth_multicast_rules_ramrod_data *data =
(struct eth_multicast_rules_ramrod_data *)(raw->rdata);
int cnt = 0, rc;
/* Reset the ramrod data buffer */
memset(data, 0, sizeof(*data));
cnt = bnx2x_mcast_handle_pending_cmds_e2(bp, p);
/* If there are no more pending commands - clear SCHEDULED state */
if (list_empty(&o->pending_cmds_head))
o->clear_sched(o);
/* The below may be true iff there was enough room in ramrod
* data for all pending commands and for the current
* command. Otherwise the current command would have been added
* to the pending commands and p->mcast_list_len would have been
* zeroed.
*/
if (p->mcast_list_len > 0)
cnt = bnx2x_mcast_handle_current_cmd(bp, p, cmd, cnt);
/* We've pulled out some MACs - update the total number of
* outstanding.
*/
o->total_pending_num -= cnt;
/* send a ramrod */
WARN_ON(o->total_pending_num < 0);
WARN_ON(cnt > o->max_cmd_len);
bnx2x_mcast_set_rdata_hdr_e2(bp, p, (u8)cnt);
/* Update a registry size if there are no more pending operations.
*
* We don't want to change the value of the registry size if there are
* pending operations because we want it to always be equal to the
* exact or the approximate number (see bnx2x_mcast_validate_e2()) of
* set bins after the last requested operation in order to properly
* evaluate the size of the next DEL/RESTORE operation.
*
* Note that we update the registry itself during command(s) handling
* - see bnx2x_mcast_set_one_rule_e2(). That's because for 57712 we
* aggregate multiple commands (ADD/DEL/RESTORE) into one ramrod but
* with a limited amount of update commands (per MAC/bin) and we don't
* know in this scope what the actual state of bins configuration is
* going to be after this ramrod.
*/
if (!o->total_pending_num)
bnx2x_mcast_refresh_registry_e2(bp, o);
/* If CLEAR_ONLY was requested - don't send a ramrod and clear
* RAMROD_PENDING status immediately. due to the SET option, it's also
* possible that after evaluating the differences there's no need for
* a ramrod. In that case, we can skip it as well.
*/
if (test_bit(RAMROD_DRV_CLR_ONLY, &p->ramrod_flags) || !cnt) {
raw->clear_pending(raw);
return 0;
} else {
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
/* Send a ramrod */
rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_MULTICAST_RULES,
raw->cid, U64_HI(raw->rdata_mapping),
U64_LO(raw->rdata_mapping),
ETH_CONNECTION_TYPE);
if (rc)
return rc;
/* Ramrod completion is pending */
return 1;
}
}
static int bnx2x_mcast_validate_e1h(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
if (cmd == BNX2X_MCAST_CMD_SET) {
BNX2X_ERR("Can't use `set' command on e1h!\n");
return -EINVAL;
}
/* Mark, that there is a work to do */
if ((cmd == BNX2X_MCAST_CMD_DEL) || (cmd == BNX2X_MCAST_CMD_RESTORE))
p->mcast_list_len = 1;
return 0;
}
static void bnx2x_mcast_revert_e1h(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
int old_num_bins,
enum bnx2x_mcast_cmd cmd)
{
/* Do nothing */
}
#define BNX2X_57711_SET_MC_FILTER(filter, bit) \
do { \
(filter)[(bit) >> 5] |= (1 << ((bit) & 0x1f)); \
} while (0)
static inline void bnx2x_mcast_hdl_add_e1h(struct bnx2x *bp,
struct bnx2x_mcast_obj *o,
struct bnx2x_mcast_ramrod_params *p,
u32 *mc_filter)
{
struct bnx2x_mcast_list_elem *mlist_pos;
int bit;
list_for_each_entry(mlist_pos, &p->mcast_list, link) {
bit = bnx2x_mcast_bin_from_mac(mlist_pos->mac);
BNX2X_57711_SET_MC_FILTER(mc_filter, bit);
DP(BNX2X_MSG_SP, "About to configure %pM mcast MAC, bin %d\n",
mlist_pos->mac, bit);
/* bookkeeping... */
BIT_VEC64_SET_BIT(o->registry.aprox_match.vec,
bit);
}
}
static inline void bnx2x_mcast_hdl_restore_e1h(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, struct bnx2x_mcast_ramrod_params *p,
u32 *mc_filter)
{
int bit;
for (bit = bnx2x_mcast_get_next_bin(o, 0);
bit >= 0;
bit = bnx2x_mcast_get_next_bin(o, bit + 1)) {
BNX2X_57711_SET_MC_FILTER(mc_filter, bit);
DP(BNX2X_MSG_SP, "About to set bin %d\n", bit);
}
}
/* On 57711 we write the multicast MACs' approximate match
* table by directly into the TSTORM's internal RAM. So we don't
* really need to handle any tricks to make it work.
*/
static int bnx2x_mcast_setup_e1h(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
int i;
struct bnx2x_mcast_obj *o = p->mcast_obj;
struct bnx2x_raw_obj *r = &o->raw;
/* If CLEAR_ONLY has been requested - clear the registry
* and clear a pending bit.
*/
if (!test_bit(RAMROD_DRV_CLR_ONLY, &p->ramrod_flags)) {
u32 mc_filter[MC_HASH_SIZE] = {0};
/* Set the multicast filter bits before writing it into
* the internal memory.
*/
switch (cmd) {
case BNX2X_MCAST_CMD_ADD:
bnx2x_mcast_hdl_add_e1h(bp, o, p, mc_filter);
break;
case BNX2X_MCAST_CMD_DEL:
DP(BNX2X_MSG_SP,
"Invalidating multicast MACs configuration\n");
/* clear the registry */
memset(o->registry.aprox_match.vec, 0,
sizeof(o->registry.aprox_match.vec));
break;
case BNX2X_MCAST_CMD_RESTORE:
bnx2x_mcast_hdl_restore_e1h(bp, o, p, mc_filter);
break;
default:
BNX2X_ERR("Unknown command: %d\n", cmd);
return -EINVAL;
}
/* Set the mcast filter in the internal memory */
for (i = 0; i < MC_HASH_SIZE; i++)
REG_WR(bp, MC_HASH_OFFSET(bp, i), mc_filter[i]);
} else
/* clear the registry */
memset(o->registry.aprox_match.vec, 0,
sizeof(o->registry.aprox_match.vec));
/* We are done */
r->clear_pending(r);
return 0;
}
static int bnx2x_mcast_validate_e1(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_mcast_obj *o = p->mcast_obj;
int reg_sz = o->get_registry_size(o);
if (cmd == BNX2X_MCAST_CMD_SET) {
BNX2X_ERR("Can't use `set' command on e1!\n");
return -EINVAL;
}
switch (cmd) {
/* DEL command deletes all currently configured MACs */
case BNX2X_MCAST_CMD_DEL:
o->set_registry_size(o, 0);
/* fall through */
/* RESTORE command will restore the entire multicast configuration */
case BNX2X_MCAST_CMD_RESTORE:
p->mcast_list_len = reg_sz;
DP(BNX2X_MSG_SP, "Command %d, p->mcast_list_len=%d\n",
cmd, p->mcast_list_len);
break;
case BNX2X_MCAST_CMD_ADD:
case BNX2X_MCAST_CMD_CONT:
/* Multicast MACs on 57710 are configured as unicast MACs and
* there is only a limited number of CAM entries for that
* matter.
*/
if (p->mcast_list_len > o->max_cmd_len) {
BNX2X_ERR("Can't configure more than %d multicast MACs on 57710\n",
o->max_cmd_len);
return -EINVAL;
}
/* Every configured MAC should be cleared if DEL command is
* called. Only the last ADD command is relevant as long as
* every ADD commands overrides the previous configuration.
*/
DP(BNX2X_MSG_SP, "p->mcast_list_len=%d\n", p->mcast_list_len);
if (p->mcast_list_len > 0)
o->set_registry_size(o, p->mcast_list_len);
break;
default:
BNX2X_ERR("Unknown command: %d\n", cmd);
return -EINVAL;
}
/* We want to ensure that commands are executed one by one for 57710.
* Therefore each none-empty command will consume o->max_cmd_len.
*/
if (p->mcast_list_len)
o->total_pending_num += o->max_cmd_len;
return 0;
}
static void bnx2x_mcast_revert_e1(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
int old_num_macs,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_mcast_obj *o = p->mcast_obj;
o->set_registry_size(o, old_num_macs);
/* If current command hasn't been handled yet and we are
* here means that it's meant to be dropped and we have to
* update the number of outstanding MACs accordingly.
*/
if (p->mcast_list_len)
o->total_pending_num -= o->max_cmd_len;
}
static void bnx2x_mcast_set_one_rule_e1(struct bnx2x *bp,
struct bnx2x_mcast_obj *o, int idx,
union bnx2x_mcast_config_data *cfg_data,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_raw_obj *r = &o->raw;
struct mac_configuration_cmd *data =
(struct mac_configuration_cmd *)(r->rdata);
/* copy mac */
if ((cmd == BNX2X_MCAST_CMD_ADD) || (cmd == BNX2X_MCAST_CMD_RESTORE)) {
bnx2x_set_fw_mac_addr(&data->config_table[idx].msb_mac_addr,
&data->config_table[idx].middle_mac_addr,
&data->config_table[idx].lsb_mac_addr,
cfg_data->mac);
data->config_table[idx].vlan_id = 0;
data->config_table[idx].pf_id = r->func_id;
data->config_table[idx].clients_bit_vector =
cpu_to_le32(1 << r->cl_id);
SET_FLAG(data->config_table[idx].flags,
MAC_CONFIGURATION_ENTRY_ACTION_TYPE,
T_ETH_MAC_COMMAND_SET);
}
}
/**
* bnx2x_mcast_set_rdata_hdr_e1 - set header values in mac_configuration_cmd
*
* @bp: device handle
* @p:
* @len: number of rules to handle
*/
static inline void bnx2x_mcast_set_rdata_hdr_e1(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
u8 len)
{
struct bnx2x_raw_obj *r = &p->mcast_obj->raw;
struct mac_configuration_cmd *data =
(struct mac_configuration_cmd *)(r->rdata);
u8 offset = (CHIP_REV_IS_SLOW(bp) ?
BNX2X_MAX_EMUL_MULTI*(1 + r->func_id) :
BNX2X_MAX_MULTICAST*(1 + r->func_id));
data->hdr.offset = offset;
data->hdr.client_id = cpu_to_le16(0xff);
data->hdr.echo = cpu_to_le32((r->cid & BNX2X_SWCID_MASK) |
(BNX2X_FILTER_MCAST_PENDING <<
BNX2X_SWCID_SHIFT));
data->hdr.length = len;
}
/**
* bnx2x_mcast_handle_restore_cmd_e1 - restore command for 57710
*
* @bp: device handle
* @o:
* @start_idx: index in the registry to start from
* @rdata_idx: index in the ramrod data to start from
*
* restore command for 57710 is like all other commands - always a stand alone
* command - start_idx and rdata_idx will always be 0. This function will always
* succeed.
* returns -1 to comply with 57712 variant.
*/
static inline int bnx2x_mcast_handle_restore_cmd_e1(
struct bnx2x *bp, struct bnx2x_mcast_obj *o , int start_idx,
int *rdata_idx)
{
struct bnx2x_mcast_mac_elem *elem;
int i = 0;
union bnx2x_mcast_config_data cfg_data = {NULL};
/* go through the registry and configure the MACs from it. */
list_for_each_entry(elem, &o->registry.exact_match.macs, link) {
cfg_data.mac = &elem->mac[0];
o->set_one_rule(bp, o, i, &cfg_data, BNX2X_MCAST_CMD_RESTORE);
i++;
DP(BNX2X_MSG_SP, "About to configure %pM mcast MAC\n",
cfg_data.mac);
}
*rdata_idx = i;
return -1;
}
static inline int bnx2x_mcast_handle_pending_cmds_e1(
struct bnx2x *bp, struct bnx2x_mcast_ramrod_params *p)
{
struct bnx2x_pending_mcast_cmd *cmd_pos;
struct bnx2x_mcast_mac_elem *pmac_pos;
struct bnx2x_mcast_obj *o = p->mcast_obj;
union bnx2x_mcast_config_data cfg_data = {NULL};
int cnt = 0;
/* If nothing to be done - return */
if (list_empty(&o->pending_cmds_head))
return 0;
/* Handle the first command */
cmd_pos = list_first_entry(&o->pending_cmds_head,
struct bnx2x_pending_mcast_cmd, link);
switch (cmd_pos->type) {
case BNX2X_MCAST_CMD_ADD:
list_for_each_entry(pmac_pos, &cmd_pos->data.macs_head, link) {
cfg_data.mac = &pmac_pos->mac[0];
o->set_one_rule(bp, o, cnt, &cfg_data, cmd_pos->type);
cnt++;
DP(BNX2X_MSG_SP, "About to configure %pM mcast MAC\n",
pmac_pos->mac);
}
break;
case BNX2X_MCAST_CMD_DEL:
cnt = cmd_pos->data.macs_num;
DP(BNX2X_MSG_SP, "About to delete %d multicast MACs\n", cnt);
break;
case BNX2X_MCAST_CMD_RESTORE:
o->hdl_restore(bp, o, 0, &cnt);
break;
default:
BNX2X_ERR("Unknown command: %d\n", cmd_pos->type);
return -EINVAL;
}
list_del(&cmd_pos->link);
bnx2x_free_groups(&cmd_pos->group_head);
kfree(cmd_pos);
return cnt;
}
/**
* bnx2x_get_fw_mac_addr - revert the bnx2x_set_fw_mac_addr().
*
* @fw_hi:
* @fw_mid:
* @fw_lo:
* @mac:
*/
static inline void bnx2x_get_fw_mac_addr(__le16 *fw_hi, __le16 *fw_mid,
__le16 *fw_lo, u8 *mac)
{
mac[1] = ((u8 *)fw_hi)[0];
mac[0] = ((u8 *)fw_hi)[1];
mac[3] = ((u8 *)fw_mid)[0];
mac[2] = ((u8 *)fw_mid)[1];
mac[5] = ((u8 *)fw_lo)[0];
mac[4] = ((u8 *)fw_lo)[1];
}
/**
* bnx2x_mcast_refresh_registry_e1 -
*
* @bp: device handle
* @cnt:
*
* Check the ramrod data first entry flag to see if it's a DELETE or ADD command
* and update the registry correspondingly: if ADD - allocate a memory and add
* the entries to the registry (list), if DELETE - clear the registry and free
* the memory.
*/
static inline int bnx2x_mcast_refresh_registry_e1(struct bnx2x *bp,
struct bnx2x_mcast_obj *o)
{
struct bnx2x_raw_obj *raw = &o->raw;
struct bnx2x_mcast_mac_elem *elem;
struct mac_configuration_cmd *data =
(struct mac_configuration_cmd *)(raw->rdata);
/* If first entry contains a SET bit - the command was ADD,
* otherwise - DEL_ALL
*/
if (GET_FLAG(data->config_table[0].flags,
MAC_CONFIGURATION_ENTRY_ACTION_TYPE)) {
int i, len = data->hdr.length;
/* Break if it was a RESTORE command */
if (!list_empty(&o->registry.exact_match.macs))
return 0;
elem = kcalloc(len, sizeof(*elem), GFP_ATOMIC);
if (!elem) {
BNX2X_ERR("Failed to allocate registry memory\n");
return -ENOMEM;
}
for (i = 0; i < len; i++, elem++) {
bnx2x_get_fw_mac_addr(
&data->config_table[i].msb_mac_addr,
&data->config_table[i].middle_mac_addr,
&data->config_table[i].lsb_mac_addr,
elem->mac);
DP(BNX2X_MSG_SP, "Adding registry entry for [%pM]\n",
elem->mac);
list_add_tail(&elem->link,
&o->registry.exact_match.macs);
}
} else {
elem = list_first_entry(&o->registry.exact_match.macs,
struct bnx2x_mcast_mac_elem, link);
DP(BNX2X_MSG_SP, "Deleting a registry\n");
kfree(elem);
INIT_LIST_HEAD(&o->registry.exact_match.macs);
}
return 0;
}
static int bnx2x_mcast_setup_e1(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_mcast_obj *o = p->mcast_obj;
struct bnx2x_raw_obj *raw = &o->raw;
struct mac_configuration_cmd *data =
(struct mac_configuration_cmd *)(raw->rdata);
int cnt = 0, i, rc;
/* Reset the ramrod data buffer */
memset(data, 0, sizeof(*data));
/* First set all entries as invalid */
for (i = 0; i < o->max_cmd_len ; i++)
SET_FLAG(data->config_table[i].flags,
MAC_CONFIGURATION_ENTRY_ACTION_TYPE,
T_ETH_MAC_COMMAND_INVALIDATE);
/* Handle pending commands first */
cnt = bnx2x_mcast_handle_pending_cmds_e1(bp, p);
/* If there are no more pending commands - clear SCHEDULED state */
if (list_empty(&o->pending_cmds_head))
o->clear_sched(o);
/* The below may be true iff there were no pending commands */
if (!cnt)
cnt = bnx2x_mcast_handle_current_cmd(bp, p, cmd, 0);
/* For 57710 every command has o->max_cmd_len length to ensure that
* commands are done one at a time.
*/
o->total_pending_num -= o->max_cmd_len;
/* send a ramrod */
WARN_ON(cnt > o->max_cmd_len);
/* Set ramrod header (in particular, a number of entries to update) */
bnx2x_mcast_set_rdata_hdr_e1(bp, p, (u8)cnt);
/* update a registry: we need the registry contents to be always up
* to date in order to be able to execute a RESTORE opcode. Here
* we use the fact that for 57710 we sent one command at a time
* hence we may take the registry update out of the command handling
* and do it in a simpler way here.
*/
rc = bnx2x_mcast_refresh_registry_e1(bp, o);
if (rc)
return rc;
/* If CLEAR_ONLY was requested - don't send a ramrod and clear
* RAMROD_PENDING status immediately.
*/
if (test_bit(RAMROD_DRV_CLR_ONLY, &p->ramrod_flags)) {
raw->clear_pending(raw);
return 0;
} else {
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
/* Send a ramrod */
rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_SET_MAC, raw->cid,
U64_HI(raw->rdata_mapping),
U64_LO(raw->rdata_mapping),
ETH_CONNECTION_TYPE);
if (rc)
return rc;
/* Ramrod completion is pending */
return 1;
}
}
static int bnx2x_mcast_get_registry_size_exact(struct bnx2x_mcast_obj *o)
{
return o->registry.exact_match.num_macs_set;
}
static int bnx2x_mcast_get_registry_size_aprox(struct bnx2x_mcast_obj *o)
{
return o->registry.aprox_match.num_bins_set;
}
static void bnx2x_mcast_set_registry_size_exact(struct bnx2x_mcast_obj *o,
int n)
{
o->registry.exact_match.num_macs_set = n;
}
static void bnx2x_mcast_set_registry_size_aprox(struct bnx2x_mcast_obj *o,
int n)
{
o->registry.aprox_match.num_bins_set = n;
}
int bnx2x_config_mcast(struct bnx2x *bp,
struct bnx2x_mcast_ramrod_params *p,
enum bnx2x_mcast_cmd cmd)
{
struct bnx2x_mcast_obj *o = p->mcast_obj;
struct bnx2x_raw_obj *r = &o->raw;
int rc = 0, old_reg_size;
/* This is needed to recover number of currently configured mcast macs
* in case of failure.
*/
old_reg_size = o->get_registry_size(o);
/* Do some calculations and checks */
rc = o->validate(bp, p, cmd);
if (rc)
return rc;
/* Return if there is no work to do */
if ((!p->mcast_list_len) && (!o->check_sched(o)))
return 0;
DP(BNX2X_MSG_SP, "o->total_pending_num=%d p->mcast_list_len=%d o->max_cmd_len=%d\n",
o->total_pending_num, p->mcast_list_len, o->max_cmd_len);
/* Enqueue the current command to the pending list if we can't complete
* it in the current iteration
*/
if (r->check_pending(r) ||
((o->max_cmd_len > 0) && (o->total_pending_num > o->max_cmd_len))) {
rc = o->enqueue_cmd(bp, p->mcast_obj, p, cmd);
if (rc < 0)
goto error_exit1;
/* As long as the current command is in a command list we
* don't need to handle it separately.
*/
p->mcast_list_len = 0;
}
if (!r->check_pending(r)) {
/* Set 'pending' state */
r->set_pending(r);
/* Configure the new classification in the chip */
rc = o->config_mcast(bp, p, cmd);
if (rc < 0)
goto error_exit2;
/* Wait for a ramrod completion if was requested */
if (test_bit(RAMROD_COMP_WAIT, &p->ramrod_flags))
rc = o->wait_comp(bp, o);
}
return rc;
error_exit2:
r->clear_pending(r);
error_exit1:
o->revert(bp, p, old_reg_size, cmd);
return rc;
}
static void bnx2x_mcast_clear_sched(struct bnx2x_mcast_obj *o)
{
smp_mb__before_atomic();
clear_bit(o->sched_state, o->raw.pstate);
smp_mb__after_atomic();
}
static void bnx2x_mcast_set_sched(struct bnx2x_mcast_obj *o)
{
smp_mb__before_atomic();
set_bit(o->sched_state, o->raw.pstate);
smp_mb__after_atomic();
}
static bool bnx2x_mcast_check_sched(struct bnx2x_mcast_obj *o)
{
return !!test_bit(o->sched_state, o->raw.pstate);
}
static bool bnx2x_mcast_check_pending(struct bnx2x_mcast_obj *o)
{
return o->raw.check_pending(&o->raw) || o->check_sched(o);
}
void bnx2x_init_mcast_obj(struct bnx2x *bp,
struct bnx2x_mcast_obj *mcast_obj,
u8 mcast_cl_id, u32 mcast_cid, u8 func_id,
u8 engine_id, void *rdata, dma_addr_t rdata_mapping,
int state, unsigned long *pstate, bnx2x_obj_type type)
{
memset(mcast_obj, 0, sizeof(*mcast_obj));
bnx2x_init_raw_obj(&mcast_obj->raw, mcast_cl_id, mcast_cid, func_id,
rdata, rdata_mapping, state, pstate, type);
mcast_obj->engine_id = engine_id;
INIT_LIST_HEAD(&mcast_obj->pending_cmds_head);
mcast_obj->sched_state = BNX2X_FILTER_MCAST_SCHED;
mcast_obj->check_sched = bnx2x_mcast_check_sched;
mcast_obj->set_sched = bnx2x_mcast_set_sched;
mcast_obj->clear_sched = bnx2x_mcast_clear_sched;
if (CHIP_IS_E1(bp)) {
mcast_obj->config_mcast = bnx2x_mcast_setup_e1;
mcast_obj->enqueue_cmd = bnx2x_mcast_enqueue_cmd;
mcast_obj->hdl_restore =
bnx2x_mcast_handle_restore_cmd_e1;
mcast_obj->check_pending = bnx2x_mcast_check_pending;
if (CHIP_REV_IS_SLOW(bp))
mcast_obj->max_cmd_len = BNX2X_MAX_EMUL_MULTI;
else
mcast_obj->max_cmd_len = BNX2X_MAX_MULTICAST;
mcast_obj->wait_comp = bnx2x_mcast_wait;
mcast_obj->set_one_rule = bnx2x_mcast_set_one_rule_e1;
mcast_obj->validate = bnx2x_mcast_validate_e1;
mcast_obj->revert = bnx2x_mcast_revert_e1;
mcast_obj->get_registry_size =
bnx2x_mcast_get_registry_size_exact;
mcast_obj->set_registry_size =
bnx2x_mcast_set_registry_size_exact;
/* 57710 is the only chip that uses the exact match for mcast
* at the moment.
*/
INIT_LIST_HEAD(&mcast_obj->registry.exact_match.macs);
} else if (CHIP_IS_E1H(bp)) {
mcast_obj->config_mcast = bnx2x_mcast_setup_e1h;
mcast_obj->enqueue_cmd = NULL;
mcast_obj->hdl_restore = NULL;
mcast_obj->check_pending = bnx2x_mcast_check_pending;
/* 57711 doesn't send a ramrod, so it has unlimited credit
* for one command.
*/
mcast_obj->max_cmd_len = -1;
mcast_obj->wait_comp = bnx2x_mcast_wait;
mcast_obj->set_one_rule = NULL;
mcast_obj->validate = bnx2x_mcast_validate_e1h;
mcast_obj->revert = bnx2x_mcast_revert_e1h;
mcast_obj->get_registry_size =
bnx2x_mcast_get_registry_size_aprox;
mcast_obj->set_registry_size =
bnx2x_mcast_set_registry_size_aprox;
} else {
mcast_obj->config_mcast = bnx2x_mcast_setup_e2;
mcast_obj->enqueue_cmd = bnx2x_mcast_enqueue_cmd;
mcast_obj->hdl_restore =
bnx2x_mcast_handle_restore_cmd_e2;
mcast_obj->check_pending = bnx2x_mcast_check_pending;
/* TODO: There should be a proper HSI define for this number!!!
*/
mcast_obj->max_cmd_len = 16;
mcast_obj->wait_comp = bnx2x_mcast_wait;
mcast_obj->set_one_rule = bnx2x_mcast_set_one_rule_e2;
mcast_obj->validate = bnx2x_mcast_validate_e2;
mcast_obj->revert = bnx2x_mcast_revert_e2;
mcast_obj->get_registry_size =
bnx2x_mcast_get_registry_size_aprox;
mcast_obj->set_registry_size =
bnx2x_mcast_set_registry_size_aprox;
}
}
/*************************** Credit handling **********************************/
/**
* atomic_add_ifless - add if the result is less than a given value.
*
* @v: pointer of type atomic_t
* @a: the amount to add to v...
* @u: ...if (v + a) is less than u.
*
* returns true if (v + a) was less than u, and false otherwise.
*
*/
static inline bool __atomic_add_ifless(atomic_t *v, int a, int u)
{
int c, old;
c = atomic_read(v);
for (;;) {
if (unlikely(c + a >= u))
return false;
old = atomic_cmpxchg((v), c, c + a);
if (likely(old == c))
break;
c = old;
}
return true;
}
/**
* atomic_dec_ifmoe - dec if the result is more or equal than a given value.
*
* @v: pointer of type atomic_t
* @a: the amount to dec from v...
* @u: ...if (v - a) is more or equal than u.
*
* returns true if (v - a) was more or equal than u, and false
* otherwise.
*/
static inline bool __atomic_dec_ifmoe(atomic_t *v, int a, int u)
{
int c, old;
c = atomic_read(v);
for (;;) {
if (unlikely(c - a < u))
return false;
old = atomic_cmpxchg((v), c, c - a);
if (likely(old == c))
break;
c = old;
}
return true;
}
static bool bnx2x_credit_pool_get(struct bnx2x_credit_pool_obj *o, int cnt)
{
bool rc;
smp_mb();
rc = __atomic_dec_ifmoe(&o->credit, cnt, 0);
smp_mb();
return rc;
}
static bool bnx2x_credit_pool_put(struct bnx2x_credit_pool_obj *o, int cnt)
{
bool rc;
smp_mb();
/* Don't let to refill if credit + cnt > pool_sz */
rc = __atomic_add_ifless(&o->credit, cnt, o->pool_sz + 1);
smp_mb();
return rc;
}
static int bnx2x_credit_pool_check(struct bnx2x_credit_pool_obj *o)
{
int cur_credit;
smp_mb();
cur_credit = atomic_read(&o->credit);
return cur_credit;
}
static bool bnx2x_credit_pool_always_true(struct bnx2x_credit_pool_obj *o,
int cnt)
{
return true;
}
static bool bnx2x_credit_pool_get_entry(
struct bnx2x_credit_pool_obj *o,
int *offset)
{
int idx, vec, i;
*offset = -1;
/* Find "internal cam-offset" then add to base for this object... */
for (vec = 0; vec < BNX2X_POOL_VEC_SIZE; vec++) {
/* Skip the current vector if there are no free entries in it */
if (!o->pool_mirror[vec])
continue;
/* If we've got here we are going to find a free entry */
for (idx = vec * BIT_VEC64_ELEM_SZ, i = 0;
i < BIT_VEC64_ELEM_SZ; idx++, i++)
if (BIT_VEC64_TEST_BIT(o->pool_mirror, idx)) {
/* Got one!! */
BIT_VEC64_CLEAR_BIT(o->pool_mirror, idx);
*offset = o->base_pool_offset + idx;
return true;
}
}
return false;
}
static bool bnx2x_credit_pool_put_entry(
struct bnx2x_credit_pool_obj *o,
int offset)
{
if (offset < o->base_pool_offset)
return false;
offset -= o->base_pool_offset;
if (offset >= o->pool_sz)
return false;
/* Return the entry to the pool */
BIT_VEC64_SET_BIT(o->pool_mirror, offset);
return true;
}
static bool bnx2x_credit_pool_put_entry_always_true(
struct bnx2x_credit_pool_obj *o,
int offset)
{
return true;
}
static bool bnx2x_credit_pool_get_entry_always_true(
struct bnx2x_credit_pool_obj *o,
int *offset)
{
*offset = -1;
return true;
}
/**
* bnx2x_init_credit_pool - initialize credit pool internals.
*
* @p:
* @base: Base entry in the CAM to use.
* @credit: pool size.
*
* If base is negative no CAM entries handling will be performed.
* If credit is negative pool operations will always succeed (unlimited pool).
*
*/
void bnx2x_init_credit_pool(struct bnx2x_credit_pool_obj *p,
int base, int credit)
{
/* Zero the object first */
memset(p, 0, sizeof(*p));
/* Set the table to all 1s */
memset(&p->pool_mirror, 0xff, sizeof(p->pool_mirror));
/* Init a pool as full */
atomic_set(&p->credit, credit);
/* The total poll size */
p->pool_sz = credit;
p->base_pool_offset = base;
/* Commit the change */
smp_mb();
p->check = bnx2x_credit_pool_check;
/* if pool credit is negative - disable the checks */
if (credit >= 0) {
p->put = bnx2x_credit_pool_put;
p->get = bnx2x_credit_pool_get;
p->put_entry = bnx2x_credit_pool_put_entry;
p->get_entry = bnx2x_credit_pool_get_entry;
} else {
p->put = bnx2x_credit_pool_always_true;
p->get = bnx2x_credit_pool_always_true;
p->put_entry = bnx2x_credit_pool_put_entry_always_true;
p->get_entry = bnx2x_credit_pool_get_entry_always_true;
}
/* If base is negative - disable entries handling */
if (base < 0) {
p->put_entry = bnx2x_credit_pool_put_entry_always_true;
p->get_entry = bnx2x_credit_pool_get_entry_always_true;
}
}
void bnx2x_init_mac_credit_pool(struct bnx2x *bp,
struct bnx2x_credit_pool_obj *p, u8 func_id,
u8 func_num)
{
/* TODO: this will be defined in consts as well... */
#define BNX2X_CAM_SIZE_EMUL 5
int cam_sz;
if (CHIP_IS_E1(bp)) {
/* In E1, Multicast is saved in cam... */
if (!CHIP_REV_IS_SLOW(bp))
cam_sz = (MAX_MAC_CREDIT_E1 / 2) - BNX2X_MAX_MULTICAST;
else
cam_sz = BNX2X_CAM_SIZE_EMUL - BNX2X_MAX_EMUL_MULTI;
bnx2x_init_credit_pool(p, func_id * cam_sz, cam_sz);
} else if (CHIP_IS_E1H(bp)) {
/* CAM credit is equaly divided between all active functions
* on the PORT!.
*/
if ((func_num > 0)) {
if (!CHIP_REV_IS_SLOW(bp))
cam_sz = (MAX_MAC_CREDIT_E1H / (2*func_num));
else
cam_sz = BNX2X_CAM_SIZE_EMUL;
bnx2x_init_credit_pool(p, func_id * cam_sz, cam_sz);
} else {
/* this should never happen! Block MAC operations. */
bnx2x_init_credit_pool(p, 0, 0);
}
} else {
/* CAM credit is equaly divided between all active functions
* on the PATH.
*/
if (func_num > 0) {
if (!CHIP_REV_IS_SLOW(bp))
cam_sz = PF_MAC_CREDIT_E2(bp, func_num);
else
cam_sz = BNX2X_CAM_SIZE_EMUL;
/* No need for CAM entries handling for 57712 and
* newer.
*/
bnx2x_init_credit_pool(p, -1, cam_sz);
} else {
/* this should never happen! Block MAC operations. */
bnx2x_init_credit_pool(p, 0, 0);
}
}
}
void bnx2x_init_vlan_credit_pool(struct bnx2x *bp,
struct bnx2x_credit_pool_obj *p,
u8 func_id,
u8 func_num)
{
if (CHIP_IS_E1x(bp)) {
/* There is no VLAN credit in HW on 57710 and 57711 only
* MAC / MAC-VLAN can be set
*/
bnx2x_init_credit_pool(p, 0, -1);
} else {
/* CAM credit is equally divided between all active functions
* on the PATH.
*/
if (func_num > 0) {
int credit = PF_VLAN_CREDIT_E2(bp, func_num);
bnx2x_init_credit_pool(p, -1/*unused for E2*/, credit);
} else
/* this should never happen! Block VLAN operations. */
bnx2x_init_credit_pool(p, 0, 0);
}
}
/****************** RSS Configuration ******************/
/**
* bnx2x_debug_print_ind_table - prints the indirection table configuration.
*
* @bp: driver handle
* @p: pointer to rss configuration
*
* Prints it when NETIF_MSG_IFUP debug level is configured.
*/
static inline void bnx2x_debug_print_ind_table(struct bnx2x *bp,
struct bnx2x_config_rss_params *p)
{
int i;
DP(BNX2X_MSG_SP, "Setting indirection table to:\n");
DP(BNX2X_MSG_SP, "0x0000: ");
for (i = 0; i < T_ETH_INDIRECTION_TABLE_SIZE; i++) {
DP_CONT(BNX2X_MSG_SP, "0x%02x ", p->ind_table[i]);
/* Print 4 bytes in a line */
if ((i + 1 < T_ETH_INDIRECTION_TABLE_SIZE) &&
(((i + 1) & 0x3) == 0)) {
DP_CONT(BNX2X_MSG_SP, "\n");
DP(BNX2X_MSG_SP, "0x%04x: ", i + 1);
}
}
DP_CONT(BNX2X_MSG_SP, "\n");
}
/**
* bnx2x_setup_rss - configure RSS
*
* @bp: device handle
* @p: rss configuration
*
* sends on UPDATE ramrod for that matter.
*/
static int bnx2x_setup_rss(struct bnx2x *bp,
struct bnx2x_config_rss_params *p)
{
struct bnx2x_rss_config_obj *o = p->rss_obj;
struct bnx2x_raw_obj *r = &o->raw;
struct eth_rss_update_ramrod_data *data =
(struct eth_rss_update_ramrod_data *)(r->rdata);
u16 caps = 0;
u8 rss_mode = 0;
int rc;
memset(data, 0, sizeof(*data));
DP(BNX2X_MSG_SP, "Configuring RSS\n");
/* Set an echo field */
data->echo = cpu_to_le32((r->cid & BNX2X_SWCID_MASK) |
(r->state << BNX2X_SWCID_SHIFT));
/* RSS mode */
if (test_bit(BNX2X_RSS_MODE_DISABLED, &p->rss_flags))
rss_mode = ETH_RSS_MODE_DISABLED;
else if (test_bit(BNX2X_RSS_MODE_REGULAR, &p->rss_flags))
rss_mode = ETH_RSS_MODE_REGULAR;
data->rss_mode = rss_mode;
DP(BNX2X_MSG_SP, "rss_mode=%d\n", rss_mode);
/* RSS capabilities */
if (test_bit(BNX2X_RSS_IPV4, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV4_CAPABILITY;
if (test_bit(BNX2X_RSS_IPV4_TCP, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV4_TCP_CAPABILITY;
if (test_bit(BNX2X_RSS_IPV4_UDP, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV4_UDP_CAPABILITY;
if (test_bit(BNX2X_RSS_IPV6, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV6_CAPABILITY;
if (test_bit(BNX2X_RSS_IPV6_TCP, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV6_TCP_CAPABILITY;
if (test_bit(BNX2X_RSS_IPV6_UDP, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV6_UDP_CAPABILITY;
if (test_bit(BNX2X_RSS_IPV4_VXLAN, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV4_VXLAN_CAPABILITY;
if (test_bit(BNX2X_RSS_IPV6_VXLAN, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_IPV6_VXLAN_CAPABILITY;
if (test_bit(BNX2X_RSS_TUNN_INNER_HDRS, &p->rss_flags))
caps |= ETH_RSS_UPDATE_RAMROD_DATA_TUNN_INNER_HDRS_CAPABILITY;
/* RSS keys */
if (test_bit(BNX2X_RSS_SET_SRCH, &p->rss_flags)) {
u8 *dst = (u8 *)(data->rss_key) + sizeof(data->rss_key);
const u8 *src = (const u8 *)p->rss_key;
int i;
/* Apparently, bnx2x reads this array in reverse order
* We need to byte swap rss_key to comply with Toeplitz specs.
*/
for (i = 0; i < sizeof(data->rss_key); i++)
*--dst = *src++;
caps |= ETH_RSS_UPDATE_RAMROD_DATA_UPDATE_RSS_KEY;
}
data->capabilities = cpu_to_le16(caps);
/* Hashing mask */
data->rss_result_mask = p->rss_result_mask;
/* RSS engine ID */
data->rss_engine_id = o->engine_id;
DP(BNX2X_MSG_SP, "rss_engine_id=%d\n", data->rss_engine_id);
/* Indirection table */
memcpy(data->indirection_table, p->ind_table,
T_ETH_INDIRECTION_TABLE_SIZE);
/* Remember the last configuration */
memcpy(o->ind_table, p->ind_table, T_ETH_INDIRECTION_TABLE_SIZE);
/* Print the indirection table */
if (netif_msg_ifup(bp))
bnx2x_debug_print_ind_table(bp, p);
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
/* Send a ramrod */
rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_RSS_UPDATE, r->cid,
U64_HI(r->rdata_mapping),
U64_LO(r->rdata_mapping),
ETH_CONNECTION_TYPE);
if (rc < 0)
return rc;
return 1;
}
void bnx2x_get_rss_ind_table(struct bnx2x_rss_config_obj *rss_obj,
u8 *ind_table)
{
memcpy(ind_table, rss_obj->ind_table, sizeof(rss_obj->ind_table));
}
int bnx2x_config_rss(struct bnx2x *bp,
struct bnx2x_config_rss_params *p)
{
int rc;
struct bnx2x_rss_config_obj *o = p->rss_obj;
struct bnx2x_raw_obj *r = &o->raw;
/* Do nothing if only driver cleanup was requested */
if (test_bit(RAMROD_DRV_CLR_ONLY, &p->ramrod_flags)) {
DP(BNX2X_MSG_SP, "Not configuring RSS ramrod_flags=%lx\n",
p->ramrod_flags);
return 0;
}
r->set_pending(r);
rc = o->config_rss(bp, p);
if (rc < 0) {
r->clear_pending(r);
return rc;
}
if (test_bit(RAMROD_COMP_WAIT, &p->ramrod_flags))
rc = r->wait_comp(bp, r);
return rc;
}
void bnx2x_init_rss_config_obj(struct bnx2x *bp,
struct bnx2x_rss_config_obj *rss_obj,
u8 cl_id, u32 cid, u8 func_id, u8 engine_id,
void *rdata, dma_addr_t rdata_mapping,
int state, unsigned long *pstate,
bnx2x_obj_type type)
{
bnx2x_init_raw_obj(&rss_obj->raw, cl_id, cid, func_id, rdata,
rdata_mapping, state, pstate, type);
rss_obj->engine_id = engine_id;
rss_obj->config_rss = bnx2x_setup_rss;
}
/********************** Queue state object ***********************************/
/**
* bnx2x_queue_state_change - perform Queue state change transition
*
* @bp: device handle
* @params: parameters to perform the transition
*
* returns 0 in case of successfully completed transition, negative error
* code in case of failure, positive (EBUSY) value if there is a completion
* to that is still pending (possible only if RAMROD_COMP_WAIT is
* not set in params->ramrod_flags for asynchronous commands).
*
*/
int bnx2x_queue_state_change(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
int rc, pending_bit;
unsigned long *pending = &o->pending;
/* Check that the requested transition is legal */
rc = o->check_transition(bp, o, params);
if (rc) {
BNX2X_ERR("check transition returned an error. rc %d\n", rc);
return -EINVAL;
}
/* Set "pending" bit */
DP(BNX2X_MSG_SP, "pending bit was=%lx\n", o->pending);
pending_bit = o->set_pending(o, params);
DP(BNX2X_MSG_SP, "pending bit now=%lx\n", o->pending);
/* Don't send a command if only driver cleanup was requested */
if (test_bit(RAMROD_DRV_CLR_ONLY, &params->ramrod_flags))
o->complete_cmd(bp, o, pending_bit);
else {
/* Send a ramrod */
rc = o->send_cmd(bp, params);
if (rc) {
o->next_state = BNX2X_Q_STATE_MAX;
clear_bit(pending_bit, pending);
smp_mb__after_atomic();
return rc;
}
if (test_bit(RAMROD_COMP_WAIT, &params->ramrod_flags)) {
rc = o->wait_comp(bp, o, pending_bit);
if (rc)
return rc;
return 0;
}
}
return !!test_bit(pending_bit, pending);
}
static int bnx2x_queue_set_pending(struct bnx2x_queue_sp_obj *obj,
struct bnx2x_queue_state_params *params)
{
enum bnx2x_queue_cmd cmd = params->cmd, bit;
/* ACTIVATE and DEACTIVATE commands are implemented on top of
* UPDATE command.
*/
if ((cmd == BNX2X_Q_CMD_ACTIVATE) ||
(cmd == BNX2X_Q_CMD_DEACTIVATE))
bit = BNX2X_Q_CMD_UPDATE;
else
bit = cmd;
set_bit(bit, &obj->pending);
return bit;
}
static int bnx2x_queue_wait_comp(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *o,
enum bnx2x_queue_cmd cmd)
{
return bnx2x_state_wait(bp, cmd, &o->pending);
}
/**
* bnx2x_queue_comp_cmd - complete the state change command.
*
* @bp: device handle
* @o:
* @cmd:
*
* Checks that the arrived completion is expected.
*/
static int bnx2x_queue_comp_cmd(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *o,
enum bnx2x_queue_cmd cmd)
{
unsigned long cur_pending = o->pending;
if (!test_and_clear_bit(cmd, &cur_pending)) {
BNX2X_ERR("Bad MC reply %d for queue %d in state %d pending 0x%lx, next_state %d\n",
cmd, o->cids[BNX2X_PRIMARY_CID_INDEX],
o->state, cur_pending, o->next_state);
return -EINVAL;
}
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
if (o->next_tx_only >= o->max_cos)
/* >= because tx only must always be smaller than cos since the
* primary connection supports COS 0
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
*/
BNX2X_ERR("illegal value for next tx_only: %d. max cos was %d",
o->next_tx_only, o->max_cos);
DP(BNX2X_MSG_SP,
"Completing command %d for queue %d, setting state to %d\n",
cmd, o->cids[BNX2X_PRIMARY_CID_INDEX], o->next_state);
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
if (o->next_tx_only) /* print num tx-only if any exist */
DP(BNX2X_MSG_SP, "primary cid %d: num tx-only cons %d\n",
o->cids[BNX2X_PRIMARY_CID_INDEX], o->next_tx_only);
o->state = o->next_state;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
o->num_tx_only = o->next_tx_only;
o->next_state = BNX2X_Q_STATE_MAX;
/* It's important that o->state and o->next_state are
* updated before o->pending.
*/
wmb();
clear_bit(cmd, &o->pending);
smp_mb__after_atomic();
return 0;
}
static void bnx2x_q_fill_setup_data_e2(struct bnx2x *bp,
struct bnx2x_queue_state_params *cmd_params,
struct client_init_ramrod_data *data)
{
struct bnx2x_queue_setup_params *params = &cmd_params->params.setup;
/* Rx data */
/* IPv6 TPA supported for E2 and above only */
data->rx.tpa_en |= test_bit(BNX2X_Q_FLG_TPA_IPV6, &params->flags) *
CLIENT_INIT_RX_DATA_TPA_EN_IPV6;
}
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
static void bnx2x_q_fill_init_general_data(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *o,
struct bnx2x_general_setup_params *params,
struct client_init_general_data *gen_data,
unsigned long *flags)
{
gen_data->client_id = o->cl_id;
if (test_bit(BNX2X_Q_FLG_STATS, flags)) {
gen_data->statistics_counter_id =
params->stat_id;
gen_data->statistics_en_flg = 1;
gen_data->statistics_zero_flg =
test_bit(BNX2X_Q_FLG_ZERO_STATS, flags);
} else
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
gen_data->statistics_counter_id =
DISABLE_STATISTIC_COUNTER_ID_VALUE;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
gen_data->is_fcoe_flg = test_bit(BNX2X_Q_FLG_FCOE, flags);
gen_data->activate_flg = test_bit(BNX2X_Q_FLG_ACTIVE, flags);
gen_data->sp_client_id = params->spcl_id;
gen_data->mtu = cpu_to_le16(params->mtu);
gen_data->func_id = o->func_id;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
gen_data->cos = params->cos;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
gen_data->traffic_type =
test_bit(BNX2X_Q_FLG_FCOE, flags) ?
LLFC_TRAFFIC_TYPE_FCOE : LLFC_TRAFFIC_TYPE_NW;
gen_data->fp_hsi_ver = params->fp_hsi;
DP(BNX2X_MSG_SP, "flags: active %d, cos %d, stats en %d\n",
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
gen_data->activate_flg, gen_data->cos, gen_data->statistics_en_flg);
}
static void bnx2x_q_fill_init_tx_data(struct bnx2x_queue_sp_obj *o,
struct bnx2x_txq_setup_params *params,
struct client_init_tx_data *tx_data,
unsigned long *flags)
{
tx_data->enforce_security_flg =
test_bit(BNX2X_Q_FLG_TX_SEC, flags);
tx_data->default_vlan =
cpu_to_le16(params->default_vlan);
tx_data->default_vlan_flg =
test_bit(BNX2X_Q_FLG_DEF_VLAN, flags);
tx_data->tx_switching_flg =
test_bit(BNX2X_Q_FLG_TX_SWITCH, flags);
tx_data->anti_spoofing_flg =
test_bit(BNX2X_Q_FLG_ANTI_SPOOF, flags);
tx_data->force_default_pri_flg =
test_bit(BNX2X_Q_FLG_FORCE_DEFAULT_PRI, flags);
tx_data->refuse_outband_vlan_flg =
test_bit(BNX2X_Q_FLG_REFUSE_OUTBAND_VLAN, flags);
tx_data->tunnel_lso_inc_ip_id =
test_bit(BNX2X_Q_FLG_TUN_INC_INNER_IP_ID, flags);
tx_data->tunnel_non_lso_pcsum_location =
test_bit(BNX2X_Q_FLG_PCSUM_ON_PKT, flags) ? CSUM_ON_PKT :
CSUM_ON_BD;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
tx_data->tx_status_block_id = params->fw_sb_id;
tx_data->tx_sb_index_number = params->sb_cq_index;
tx_data->tss_leading_client_id = params->tss_leading_cl_id;
tx_data->tx_bd_page_base.lo =
cpu_to_le32(U64_LO(params->dscr_map));
tx_data->tx_bd_page_base.hi =
cpu_to_le32(U64_HI(params->dscr_map));
/* Don't configure any Tx switching mode during queue SETUP */
tx_data->state = 0;
}
static void bnx2x_q_fill_init_pause_data(struct bnx2x_queue_sp_obj *o,
struct rxq_pause_params *params,
struct client_init_rx_data *rx_data)
{
/* flow control data */
rx_data->cqe_pause_thr_low = cpu_to_le16(params->rcq_th_lo);
rx_data->cqe_pause_thr_high = cpu_to_le16(params->rcq_th_hi);
rx_data->bd_pause_thr_low = cpu_to_le16(params->bd_th_lo);
rx_data->bd_pause_thr_high = cpu_to_le16(params->bd_th_hi);
rx_data->sge_pause_thr_low = cpu_to_le16(params->sge_th_lo);
rx_data->sge_pause_thr_high = cpu_to_le16(params->sge_th_hi);
rx_data->rx_cos_mask = cpu_to_le16(params->pri_map);
}
static void bnx2x_q_fill_init_rx_data(struct bnx2x_queue_sp_obj *o,
struct bnx2x_rxq_setup_params *params,
struct client_init_rx_data *rx_data,
unsigned long *flags)
{
rx_data->tpa_en = test_bit(BNX2X_Q_FLG_TPA, flags) *
CLIENT_INIT_RX_DATA_TPA_EN_IPV4;
use FW 7.2.16 The patch integrates FW 7.2.16 HSI and implements driver part of GRO flow. FW 7.2.16 adds the ability to aggregate packets for GRO (and not just LRO) and also fixes some bugs. 1. Added new aggregation mode: GRO. In this mode packets are aggregated such that the original packets can be reconstructed by the OS. 2. 57712 HW bug workaround - initialized all CAM TM registers to 0x32. 3. Adding the FCoE statistics structures to the BNX2X HSI. 4. Wrong configuration of TX HW input buffer size may cause theoretical performance effect. Performed configuration fix. 5. FCOE - Arrival of packets beyond task IO size can lead to crash. Fix firmware data-in flow. 6. iSCSI - In rare cases of on-chip termination the graceful termination timer hangs, and the termination doesn't complete. Firmware fix to MSL timer tolerance. 7. iSCSI - Chip hangs when target sends FIN out-of-order or with isles open at the initiator side. Firmware implementation corrected to drop FIN received out-of-order or with isles still open. 8. iSCSI - Chip hangs when in case of retransmission not aligned to 4-bytes from the beginning of iSCSI PDU. Firmware implementation corrected to support arbitrary aligned retransmissions. 9. iSCSI - Arrival of target-initiated NOP-IN during intense ISCSI traffic might lead to crash. Firmware fix to relevant flow. Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-02-20 02:59:08 -07:00
rx_data->tpa_en |= test_bit(BNX2X_Q_FLG_TPA_GRO, flags) *
CLIENT_INIT_RX_DATA_TPA_MODE;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rx_data->vmqueue_mode_en_flg = 0;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rx_data->cache_line_alignment_log_size =
params->cache_line_log;
rx_data->enable_dynamic_hc =
test_bit(BNX2X_Q_FLG_DHC, flags);
rx_data->max_sges_for_packet = params->max_sges_pkt;
rx_data->client_qzone_id = params->cl_qzone_id;
rx_data->max_agg_size = cpu_to_le16(params->tpa_agg_sz);
/* Always start in DROP_ALL mode */
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rx_data->state = cpu_to_le16(CLIENT_INIT_RX_DATA_UCAST_DROP_ALL |
CLIENT_INIT_RX_DATA_MCAST_DROP_ALL);
/* We don't set drop flags */
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rx_data->drop_ip_cs_err_flg = 0;
rx_data->drop_tcp_cs_err_flg = 0;
rx_data->drop_ttl0_flg = 0;
rx_data->drop_udp_cs_err_flg = 0;
rx_data->inner_vlan_removal_enable_flg =
test_bit(BNX2X_Q_FLG_VLAN, flags);
rx_data->outer_vlan_removal_enable_flg =
test_bit(BNX2X_Q_FLG_OV, flags);
rx_data->status_block_id = params->fw_sb_id;
rx_data->rx_sb_index_number = params->sb_cq_index;
rx_data->max_tpa_queues = params->max_tpa_queues;
rx_data->max_bytes_on_bd = cpu_to_le16(params->buf_sz);
rx_data->sge_buff_size = cpu_to_le16(params->sge_buf_sz);
rx_data->bd_page_base.lo =
cpu_to_le32(U64_LO(params->dscr_map));
rx_data->bd_page_base.hi =
cpu_to_le32(U64_HI(params->dscr_map));
rx_data->sge_page_base.lo =
cpu_to_le32(U64_LO(params->sge_map));
rx_data->sge_page_base.hi =
cpu_to_le32(U64_HI(params->sge_map));
rx_data->cqe_page_base.lo =
cpu_to_le32(U64_LO(params->rcq_map));
rx_data->cqe_page_base.hi =
cpu_to_le32(U64_HI(params->rcq_map));
rx_data->is_leading_rss = test_bit(BNX2X_Q_FLG_LEADING_RSS, flags);
if (test_bit(BNX2X_Q_FLG_MCAST, flags)) {
rx_data->approx_mcast_engine_id = params->mcast_engine_id;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rx_data->is_approx_mcast = 1;
}
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rx_data->rss_engine_id = params->rss_engine_id;
/* silent vlan removal */
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rx_data->silent_vlan_removal_flg =
test_bit(BNX2X_Q_FLG_SILENT_VLAN_REM, flags);
rx_data->silent_vlan_value =
cpu_to_le16(params->silent_removal_value);
rx_data->silent_vlan_mask =
cpu_to_le16(params->silent_removal_mask);
}
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
/* initialize the general, tx and rx parts of a queue object */
static void bnx2x_q_fill_setup_data_cmn(struct bnx2x *bp,
struct bnx2x_queue_state_params *cmd_params,
struct client_init_ramrod_data *data)
{
bnx2x_q_fill_init_general_data(bp, cmd_params->q_obj,
&cmd_params->params.setup.gen_params,
&data->general,
&cmd_params->params.setup.flags);
bnx2x_q_fill_init_tx_data(cmd_params->q_obj,
&cmd_params->params.setup.txq_params,
&data->tx,
&cmd_params->params.setup.flags);
bnx2x_q_fill_init_rx_data(cmd_params->q_obj,
&cmd_params->params.setup.rxq_params,
&data->rx,
&cmd_params->params.setup.flags);
bnx2x_q_fill_init_pause_data(cmd_params->q_obj,
&cmd_params->params.setup.pause_params,
&data->rx);
}
/* initialize the general and tx parts of a tx-only queue object */
static void bnx2x_q_fill_setup_tx_only(struct bnx2x *bp,
struct bnx2x_queue_state_params *cmd_params,
struct tx_queue_init_ramrod_data *data)
{
bnx2x_q_fill_init_general_data(bp, cmd_params->q_obj,
&cmd_params->params.tx_only.gen_params,
&data->general,
&cmd_params->params.tx_only.flags);
bnx2x_q_fill_init_tx_data(cmd_params->q_obj,
&cmd_params->params.tx_only.txq_params,
&data->tx,
&cmd_params->params.tx_only.flags);
DP(BNX2X_MSG_SP, "cid %d, tx bd page lo %x hi %x",
cmd_params->q_obj->cids[0],
data->tx.tx_bd_page_base.lo,
data->tx.tx_bd_page_base.hi);
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
}
/**
* bnx2x_q_init - init HW/FW queue
*
* @bp: device handle
* @params:
*
* HW/FW initial Queue configuration:
* - HC: Rx and Tx
* - CDU context validation
*
*/
static inline int bnx2x_q_init(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
struct bnx2x_queue_init_params *init = &params->params.init;
u16 hc_usec;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
u8 cos;
/* Tx HC configuration */
if (test_bit(BNX2X_Q_TYPE_HAS_TX, &o->type) &&
test_bit(BNX2X_Q_FLG_HC, &init->tx.flags)) {
hc_usec = init->tx.hc_rate ? 1000000 / init->tx.hc_rate : 0;
bnx2x_update_coalesce_sb_index(bp, init->tx.fw_sb_id,
init->tx.sb_cq_index,
!test_bit(BNX2X_Q_FLG_HC_EN, &init->tx.flags),
hc_usec);
}
/* Rx HC configuration */
if (test_bit(BNX2X_Q_TYPE_HAS_RX, &o->type) &&
test_bit(BNX2X_Q_FLG_HC, &init->rx.flags)) {
hc_usec = init->rx.hc_rate ? 1000000 / init->rx.hc_rate : 0;
bnx2x_update_coalesce_sb_index(bp, init->rx.fw_sb_id,
init->rx.sb_cq_index,
!test_bit(BNX2X_Q_FLG_HC_EN, &init->rx.flags),
hc_usec);
}
/* Set CDU context validation values */
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
for (cos = 0; cos < o->max_cos; cos++) {
DP(BNX2X_MSG_SP, "setting context validation. cid %d, cos %d\n",
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
o->cids[cos], cos);
DP(BNX2X_MSG_SP, "context pointer %p\n", init->cxts[cos]);
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
bnx2x_set_ctx_validation(bp, init->cxts[cos], o->cids[cos]);
}
/* As no ramrod is sent, complete the command immediately */
o->complete_cmd(bp, o, BNX2X_Q_CMD_INIT);
smp_mb();
return 0;
}
static inline int bnx2x_q_send_setup_e1x(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
struct client_init_ramrod_data *rdata =
(struct client_init_ramrod_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
int ramrod = RAMROD_CMD_ID_ETH_CLIENT_SETUP;
/* Clear the ramrod data */
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data */
bnx2x_q_fill_setup_data_cmn(bp, params, rdata);
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
return bnx2x_sp_post(bp, ramrod, o->cids[BNX2X_PRIMARY_CID_INDEX],
U64_HI(data_mapping),
U64_LO(data_mapping), ETH_CONNECTION_TYPE);
}
static inline int bnx2x_q_send_setup_e2(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
struct client_init_ramrod_data *rdata =
(struct client_init_ramrod_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
int ramrod = RAMROD_CMD_ID_ETH_CLIENT_SETUP;
/* Clear the ramrod data */
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data */
bnx2x_q_fill_setup_data_cmn(bp, params, rdata);
bnx2x_q_fill_setup_data_e2(bp, params, rdata);
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
return bnx2x_sp_post(bp, ramrod, o->cids[BNX2X_PRIMARY_CID_INDEX],
U64_HI(data_mapping),
U64_LO(data_mapping), ETH_CONNECTION_TYPE);
}
static inline int bnx2x_q_send_setup_tx_only(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
struct tx_queue_init_ramrod_data *rdata =
(struct tx_queue_init_ramrod_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
int ramrod = RAMROD_CMD_ID_ETH_TX_QUEUE_SETUP;
struct bnx2x_queue_setup_tx_only_params *tx_only_params =
&params->params.tx_only;
u8 cid_index = tx_only_params->cid_index;
if (cid_index >= o->max_cos) {
BNX2X_ERR("queue[%d]: cid_index (%d) is out of range\n",
o->cl_id, cid_index);
return -EINVAL;
}
DP(BNX2X_MSG_SP, "parameters received: cos: %d sp-id: %d\n",
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
tx_only_params->gen_params.cos,
tx_only_params->gen_params.spcl_id);
/* Clear the ramrod data */
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data */
bnx2x_q_fill_setup_tx_only(bp, params, rdata);
DP(BNX2X_MSG_SP, "sending tx-only ramrod: cid %d, client-id %d, sp-client id %d, cos %d\n",
o->cids[cid_index], rdata->general.client_id,
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
rdata->general.sp_client_id, rdata->general.cos);
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
*/
return bnx2x_sp_post(bp, ramrod, o->cids[cid_index],
U64_HI(data_mapping),
U64_LO(data_mapping), ETH_CONNECTION_TYPE);
}
static void bnx2x_q_fill_update_data(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *obj,
struct bnx2x_queue_update_params *params,
struct client_update_ramrod_data *data)
{
/* Client ID of the client to update */
data->client_id = obj->cl_id;
/* Function ID of the client to update */
data->func_id = obj->func_id;
/* Default VLAN value */
data->default_vlan = cpu_to_le16(params->def_vlan);
/* Inner VLAN stripping */
data->inner_vlan_removal_enable_flg =
test_bit(BNX2X_Q_UPDATE_IN_VLAN_REM, &params->update_flags);
data->inner_vlan_removal_change_flg =
test_bit(BNX2X_Q_UPDATE_IN_VLAN_REM_CHNG,
&params->update_flags);
/* Outer VLAN stripping */
data->outer_vlan_removal_enable_flg =
test_bit(BNX2X_Q_UPDATE_OUT_VLAN_REM, &params->update_flags);
data->outer_vlan_removal_change_flg =
test_bit(BNX2X_Q_UPDATE_OUT_VLAN_REM_CHNG,
&params->update_flags);
/* Drop packets that have source MAC that doesn't belong to this
* Queue.
*/
data->anti_spoofing_enable_flg =
test_bit(BNX2X_Q_UPDATE_ANTI_SPOOF, &params->update_flags);
data->anti_spoofing_change_flg =
test_bit(BNX2X_Q_UPDATE_ANTI_SPOOF_CHNG, &params->update_flags);
/* Activate/Deactivate */
data->activate_flg =
test_bit(BNX2X_Q_UPDATE_ACTIVATE, &params->update_flags);
data->activate_change_flg =
test_bit(BNX2X_Q_UPDATE_ACTIVATE_CHNG, &params->update_flags);
/* Enable default VLAN */
data->default_vlan_enable_flg =
test_bit(BNX2X_Q_UPDATE_DEF_VLAN_EN, &params->update_flags);
data->default_vlan_change_flg =
test_bit(BNX2X_Q_UPDATE_DEF_VLAN_EN_CHNG,
&params->update_flags);
/* silent vlan removal */
data->silent_vlan_change_flg =
test_bit(BNX2X_Q_UPDATE_SILENT_VLAN_REM_CHNG,
&params->update_flags);
data->silent_vlan_removal_flg =
test_bit(BNX2X_Q_UPDATE_SILENT_VLAN_REM, &params->update_flags);
data->silent_vlan_value = cpu_to_le16(params->silent_removal_value);
data->silent_vlan_mask = cpu_to_le16(params->silent_removal_mask);
/* tx switching */
data->tx_switching_flg =
test_bit(BNX2X_Q_UPDATE_TX_SWITCHING, &params->update_flags);
data->tx_switching_change_flg =
test_bit(BNX2X_Q_UPDATE_TX_SWITCHING_CHNG,
&params->update_flags);
/* PTP */
data->handle_ptp_pkts_flg =
test_bit(BNX2X_Q_UPDATE_PTP_PKTS, &params->update_flags);
data->handle_ptp_pkts_change_flg =
test_bit(BNX2X_Q_UPDATE_PTP_PKTS_CHNG, &params->update_flags);
}
static inline int bnx2x_q_send_update(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
struct client_update_ramrod_data *rdata =
(struct client_update_ramrod_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
struct bnx2x_queue_update_params *update_params =
&params->params.update;
u8 cid_index = update_params->cid_index;
if (cid_index >= o->max_cos) {
BNX2X_ERR("queue[%d]: cid_index (%d) is out of range\n",
o->cl_id, cid_index);
return -EINVAL;
}
/* Clear the ramrod data */
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data */
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
bnx2x_q_fill_update_data(bp, o, update_params, rdata);
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
return bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_CLIENT_UPDATE,
o->cids[cid_index], U64_HI(data_mapping),
U64_LO(data_mapping), ETH_CONNECTION_TYPE);
}
/**
* bnx2x_q_send_deactivate - send DEACTIVATE command
*
* @bp: device handle
* @params:
*
* implemented using the UPDATE command.
*/
static inline int bnx2x_q_send_deactivate(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_update_params *update = &params->params.update;
memset(update, 0, sizeof(*update));
__set_bit(BNX2X_Q_UPDATE_ACTIVATE_CHNG, &update->update_flags);
return bnx2x_q_send_update(bp, params);
}
/**
* bnx2x_q_send_activate - send ACTIVATE command
*
* @bp: device handle
* @params:
*
* implemented using the UPDATE command.
*/
static inline int bnx2x_q_send_activate(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_update_params *update = &params->params.update;
memset(update, 0, sizeof(*update));
__set_bit(BNX2X_Q_UPDATE_ACTIVATE, &update->update_flags);
__set_bit(BNX2X_Q_UPDATE_ACTIVATE_CHNG, &update->update_flags);
return bnx2x_q_send_update(bp, params);
}
static void bnx2x_q_fill_update_tpa_data(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *obj,
struct bnx2x_queue_update_tpa_params *params,
struct tpa_update_ramrod_data *data)
{
data->client_id = obj->cl_id;
data->complete_on_both_clients = params->complete_on_both_clients;
data->dont_verify_rings_pause_thr_flg =
params->dont_verify_thr;
data->max_agg_size = cpu_to_le16(params->max_agg_sz);
data->max_sges_for_packet = params->max_sges_pkt;
data->max_tpa_queues = params->max_tpa_queues;
data->sge_buff_size = cpu_to_le16(params->sge_buff_sz);
data->sge_page_base_hi = cpu_to_le32(U64_HI(params->sge_map));
data->sge_page_base_lo = cpu_to_le32(U64_LO(params->sge_map));
data->sge_pause_thr_high = cpu_to_le16(params->sge_pause_thr_high);
data->sge_pause_thr_low = cpu_to_le16(params->sge_pause_thr_low);
data->tpa_mode = params->tpa_mode;
data->update_ipv4 = params->update_ipv4;
data->update_ipv6 = params->update_ipv6;
}
static inline int bnx2x_q_send_update_tpa(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
struct tpa_update_ramrod_data *rdata =
(struct tpa_update_ramrod_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
struct bnx2x_queue_update_tpa_params *update_tpa_params =
&params->params.update_tpa;
u16 type;
/* Clear the ramrod data */
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data */
bnx2x_q_fill_update_tpa_data(bp, o, update_tpa_params, rdata);
/* Add the function id inside the type, so that sp post function
* doesn't automatically add the PF func-id, this is required
* for operations done by PFs on behalf of their VFs
*/
type = ETH_CONNECTION_TYPE |
((o->func_id) << SPE_HDR_FUNCTION_ID_SHIFT);
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
return bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_TPA_UPDATE,
o->cids[BNX2X_PRIMARY_CID_INDEX],
U64_HI(data_mapping),
U64_LO(data_mapping), type);
}
static inline int bnx2x_q_send_halt(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
return bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_HALT,
o->cids[BNX2X_PRIMARY_CID_INDEX], 0, o->cl_id,
ETH_CONNECTION_TYPE);
}
static inline int bnx2x_q_send_cfc_del(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
u8 cid_idx = params->params.cfc_del.cid_index;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
if (cid_idx >= o->max_cos) {
BNX2X_ERR("queue[%d]: cid_index (%d) is out of range\n",
o->cl_id, cid_idx);
return -EINVAL;
}
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_CFC_DEL,
o->cids[cid_idx], 0, 0, NONE_CONNECTION_TYPE);
}
static inline int bnx2x_q_send_terminate(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
u8 cid_index = params->params.terminate.cid_index;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
if (cid_index >= o->max_cos) {
BNX2X_ERR("queue[%d]: cid_index (%d) is out of range\n",
o->cl_id, cid_index);
return -EINVAL;
}
return bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_TERMINATE,
o->cids[cid_index], 0, 0, ETH_CONNECTION_TYPE);
}
static inline int bnx2x_q_send_empty(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
struct bnx2x_queue_sp_obj *o = params->q_obj;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
return bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_EMPTY,
o->cids[BNX2X_PRIMARY_CID_INDEX], 0, 0,
ETH_CONNECTION_TYPE);
}
static inline int bnx2x_queue_send_cmd_cmn(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
switch (params->cmd) {
case BNX2X_Q_CMD_INIT:
return bnx2x_q_init(bp, params);
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
case BNX2X_Q_CMD_SETUP_TX_ONLY:
return bnx2x_q_send_setup_tx_only(bp, params);
case BNX2X_Q_CMD_DEACTIVATE:
return bnx2x_q_send_deactivate(bp, params);
case BNX2X_Q_CMD_ACTIVATE:
return bnx2x_q_send_activate(bp, params);
case BNX2X_Q_CMD_UPDATE:
return bnx2x_q_send_update(bp, params);
case BNX2X_Q_CMD_UPDATE_TPA:
return bnx2x_q_send_update_tpa(bp, params);
case BNX2X_Q_CMD_HALT:
return bnx2x_q_send_halt(bp, params);
case BNX2X_Q_CMD_CFC_DEL:
return bnx2x_q_send_cfc_del(bp, params);
case BNX2X_Q_CMD_TERMINATE:
return bnx2x_q_send_terminate(bp, params);
case BNX2X_Q_CMD_EMPTY:
return bnx2x_q_send_empty(bp, params);
default:
BNX2X_ERR("Unknown command: %d\n", params->cmd);
return -EINVAL;
}
}
static int bnx2x_queue_send_cmd_e1x(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
switch (params->cmd) {
case BNX2X_Q_CMD_SETUP:
return bnx2x_q_send_setup_e1x(bp, params);
case BNX2X_Q_CMD_INIT:
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
case BNX2X_Q_CMD_SETUP_TX_ONLY:
case BNX2X_Q_CMD_DEACTIVATE:
case BNX2X_Q_CMD_ACTIVATE:
case BNX2X_Q_CMD_UPDATE:
case BNX2X_Q_CMD_UPDATE_TPA:
case BNX2X_Q_CMD_HALT:
case BNX2X_Q_CMD_CFC_DEL:
case BNX2X_Q_CMD_TERMINATE:
case BNX2X_Q_CMD_EMPTY:
return bnx2x_queue_send_cmd_cmn(bp, params);
default:
BNX2X_ERR("Unknown command: %d\n", params->cmd);
return -EINVAL;
}
}
static int bnx2x_queue_send_cmd_e2(struct bnx2x *bp,
struct bnx2x_queue_state_params *params)
{
switch (params->cmd) {
case BNX2X_Q_CMD_SETUP:
return bnx2x_q_send_setup_e2(bp, params);
case BNX2X_Q_CMD_INIT:
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
case BNX2X_Q_CMD_SETUP_TX_ONLY:
case BNX2X_Q_CMD_DEACTIVATE:
case BNX2X_Q_CMD_ACTIVATE:
case BNX2X_Q_CMD_UPDATE:
case BNX2X_Q_CMD_UPDATE_TPA:
case BNX2X_Q_CMD_HALT:
case BNX2X_Q_CMD_CFC_DEL:
case BNX2X_Q_CMD_TERMINATE:
case BNX2X_Q_CMD_EMPTY:
return bnx2x_queue_send_cmd_cmn(bp, params);
default:
BNX2X_ERR("Unknown command: %d\n", params->cmd);
return -EINVAL;
}
}
/**
* bnx2x_queue_chk_transition - check state machine of a regular Queue
*
* @bp: device handle
* @o:
* @params:
*
* (not Forwarding)
* It both checks if the requested command is legal in a current
* state and, if it's legal, sets a `next_state' in the object
* that will be used in the completion flow to set the `state'
* of the object.
*
* returns 0 if a requested command is a legal transition,
* -EINVAL otherwise.
*/
static int bnx2x_queue_chk_transition(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *o,
struct bnx2x_queue_state_params *params)
{
enum bnx2x_q_state state = o->state, next_state = BNX2X_Q_STATE_MAX;
enum bnx2x_queue_cmd cmd = params->cmd;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
struct bnx2x_queue_update_params *update_params =
&params->params.update;
u8 next_tx_only = o->num_tx_only;
/* Forget all pending for completion commands if a driver only state
* transition has been requested.
*/
if (test_bit(RAMROD_DRV_CLR_ONLY, &params->ramrod_flags)) {
o->pending = 0;
o->next_state = BNX2X_Q_STATE_MAX;
}
/* Don't allow a next state transition if we are in the middle of
* the previous one.
*/
if (o->pending) {
BNX2X_ERR("Blocking transition since pending was %lx\n",
o->pending);
return -EBUSY;
}
switch (state) {
case BNX2X_Q_STATE_RESET:
if (cmd == BNX2X_Q_CMD_INIT)
next_state = BNX2X_Q_STATE_INITIALIZED;
break;
case BNX2X_Q_STATE_INITIALIZED:
if (cmd == BNX2X_Q_CMD_SETUP) {
if (test_bit(BNX2X_Q_FLG_ACTIVE,
&params->params.setup.flags))
next_state = BNX2X_Q_STATE_ACTIVE;
else
next_state = BNX2X_Q_STATE_INACTIVE;
}
break;
case BNX2X_Q_STATE_ACTIVE:
if (cmd == BNX2X_Q_CMD_DEACTIVATE)
next_state = BNX2X_Q_STATE_INACTIVE;
else if ((cmd == BNX2X_Q_CMD_EMPTY) ||
(cmd == BNX2X_Q_CMD_UPDATE_TPA))
next_state = BNX2X_Q_STATE_ACTIVE;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
else if (cmd == BNX2X_Q_CMD_SETUP_TX_ONLY) {
next_state = BNX2X_Q_STATE_MULTI_COS;
next_tx_only = 1;
}
else if (cmd == BNX2X_Q_CMD_HALT)
next_state = BNX2X_Q_STATE_STOPPED;
else if (cmd == BNX2X_Q_CMD_UPDATE) {
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
/* If "active" state change is requested, update the
* state accordingly.
*/
if (test_bit(BNX2X_Q_UPDATE_ACTIVATE_CHNG,
&update_params->update_flags) &&
!test_bit(BNX2X_Q_UPDATE_ACTIVATE,
&update_params->update_flags))
next_state = BNX2X_Q_STATE_INACTIVE;
else
next_state = BNX2X_Q_STATE_ACTIVE;
}
break;
case BNX2X_Q_STATE_MULTI_COS:
if (cmd == BNX2X_Q_CMD_TERMINATE)
next_state = BNX2X_Q_STATE_MCOS_TERMINATED;
else if (cmd == BNX2X_Q_CMD_SETUP_TX_ONLY) {
next_state = BNX2X_Q_STATE_MULTI_COS;
next_tx_only = o->num_tx_only + 1;
}
else if ((cmd == BNX2X_Q_CMD_EMPTY) ||
(cmd == BNX2X_Q_CMD_UPDATE_TPA))
next_state = BNX2X_Q_STATE_MULTI_COS;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
else if (cmd == BNX2X_Q_CMD_UPDATE) {
/* If "active" state change is requested, update the
* state accordingly.
*/
if (test_bit(BNX2X_Q_UPDATE_ACTIVATE_CHNG,
&update_params->update_flags) &&
!test_bit(BNX2X_Q_UPDATE_ACTIVATE,
&update_params->update_flags))
next_state = BNX2X_Q_STATE_INACTIVE;
else
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
next_state = BNX2X_Q_STATE_MULTI_COS;
}
break;
case BNX2X_Q_STATE_MCOS_TERMINATED:
if (cmd == BNX2X_Q_CMD_CFC_DEL) {
next_tx_only = o->num_tx_only - 1;
if (next_tx_only == 0)
next_state = BNX2X_Q_STATE_ACTIVE;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
else
next_state = BNX2X_Q_STATE_MULTI_COS;
}
break;
case BNX2X_Q_STATE_INACTIVE:
if (cmd == BNX2X_Q_CMD_ACTIVATE)
next_state = BNX2X_Q_STATE_ACTIVE;
else if ((cmd == BNX2X_Q_CMD_EMPTY) ||
(cmd == BNX2X_Q_CMD_UPDATE_TPA))
next_state = BNX2X_Q_STATE_INACTIVE;
else if (cmd == BNX2X_Q_CMD_HALT)
next_state = BNX2X_Q_STATE_STOPPED;
else if (cmd == BNX2X_Q_CMD_UPDATE) {
/* If "active" state change is requested, update the
* state accordingly.
*/
if (test_bit(BNX2X_Q_UPDATE_ACTIVATE_CHNG,
&update_params->update_flags) &&
test_bit(BNX2X_Q_UPDATE_ACTIVATE,
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
&update_params->update_flags)){
if (o->num_tx_only == 0)
next_state = BNX2X_Q_STATE_ACTIVE;
else /* tx only queues exist for this queue */
next_state = BNX2X_Q_STATE_MULTI_COS;
} else
next_state = BNX2X_Q_STATE_INACTIVE;
}
break;
case BNX2X_Q_STATE_STOPPED:
if (cmd == BNX2X_Q_CMD_TERMINATE)
next_state = BNX2X_Q_STATE_TERMINATED;
break;
case BNX2X_Q_STATE_TERMINATED:
if (cmd == BNX2X_Q_CMD_CFC_DEL)
next_state = BNX2X_Q_STATE_RESET;
break;
default:
BNX2X_ERR("Illegal state: %d\n", state);
}
/* Transition is assured */
if (next_state != BNX2X_Q_STATE_MAX) {
DP(BNX2X_MSG_SP, "Good state transition: %d(%d)->%d\n",
state, cmd, next_state);
o->next_state = next_state;
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
o->next_tx_only = next_tx_only;
return 0;
}
DP(BNX2X_MSG_SP, "Bad state transition request: %d %d\n", state, cmd);
return -EINVAL;
}
void bnx2x_init_queue_obj(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *obj,
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
u8 cl_id, u32 *cids, u8 cid_cnt, u8 func_id,
void *rdata,
dma_addr_t rdata_mapping, unsigned long type)
{
memset(obj, 0, sizeof(*obj));
bnx2x: Multiple concurrent l2 traffic classes Overview: Support mapping of priorities to traffic classes and traffic classes to transmission queues ranges in the net device. The queue ranges are (count, offset) pairs relating to the txq array. This can be done via DCBX negotiation or by kernel. As a result Enhanced Transmission Selection (ETS) and Priority Flow Control (PFC) are supported between L2 network traffic classes. Mapping: This patch uses the netdev_set_num_tc, netdev_set_prio_tc_map and netdev_set_tc_queue functions to map priorities to traffic classes and traffic classes to transmission queue ranges. This mapping is performed by bnx2x_setup_tc function which is connected to the ndo_setup_tc. This function is always called at nic load where by default it maps all priorities to tc 0, and it may also be called by the kernel or by the bnx2x upon DCBX negotiation to modify the mapping. rtnl lock: When the ndo_setup_tc is called at nic load or by kernel the rtnl lock is already taken. However, when DCBX negotiation takes place the lock is not taken. The work is therefore scheduled to be handled by the sp_rtnl task. Fastpath: The fastpath structure of the bnx2x which was previously used to hold the information of one tx queue and one rx queue was redesigned to represent multiple tx queues, one for each traffic class. The transmission queue supplied in the skb by the kernel can no longer be interpreted as a straightforward index into the fastpath structure array, but it must rather be decoded to the appropriate fastpath index and the tc within that fastpath. Slowpath: The bnx2x's queue object was redesigned to accommodate multiple transmission queues. The queue object's state machine was enhanced to allow opening multiple transmission-only connections on top of the regular tx-rx connection. Firmware: This feature relies on the tx-only queue feature introduced in the bnx2x 7.0.23 firmware and the FW likewise must have the bnx2x multi cos support. Signed-off-by: Ariel Elior <ariele@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-07-14 02:31:57 -06:00
/* We support only BNX2X_MULTI_TX_COS Tx CoS at the moment */
BUG_ON(BNX2X_MULTI_TX_COS < cid_cnt);
memcpy(obj->cids, cids, sizeof(obj->cids[0]) * cid_cnt);
obj->max_cos = cid_cnt;
obj->cl_id = cl_id;
obj->func_id = func_id;
obj->rdata = rdata;
obj->rdata_mapping = rdata_mapping;
obj->type = type;
obj->next_state = BNX2X_Q_STATE_MAX;
if (CHIP_IS_E1x(bp))
obj->send_cmd = bnx2x_queue_send_cmd_e1x;
else
obj->send_cmd = bnx2x_queue_send_cmd_e2;
obj->check_transition = bnx2x_queue_chk_transition;
obj->complete_cmd = bnx2x_queue_comp_cmd;
obj->wait_comp = bnx2x_queue_wait_comp;
obj->set_pending = bnx2x_queue_set_pending;
}
/* return a queue object's logical state*/
int bnx2x_get_q_logical_state(struct bnx2x *bp,
struct bnx2x_queue_sp_obj *obj)
{
switch (obj->state) {
case BNX2X_Q_STATE_ACTIVE:
case BNX2X_Q_STATE_MULTI_COS:
return BNX2X_Q_LOGICAL_STATE_ACTIVE;
case BNX2X_Q_STATE_RESET:
case BNX2X_Q_STATE_INITIALIZED:
case BNX2X_Q_STATE_MCOS_TERMINATED:
case BNX2X_Q_STATE_INACTIVE:
case BNX2X_Q_STATE_STOPPED:
case BNX2X_Q_STATE_TERMINATED:
case BNX2X_Q_STATE_FLRED:
return BNX2X_Q_LOGICAL_STATE_STOPPED;
default:
return -EINVAL;
}
}
/********************** Function state object *********************************/
enum bnx2x_func_state bnx2x_func_get_state(struct bnx2x *bp,
struct bnx2x_func_sp_obj *o)
{
/* in the middle of transaction - return INVALID state */
if (o->pending)
return BNX2X_F_STATE_MAX;
/* unsure the order of reading of o->pending and o->state
* o->pending should be read first
*/
rmb();
return o->state;
}
static int bnx2x_func_wait_comp(struct bnx2x *bp,
struct bnx2x_func_sp_obj *o,
enum bnx2x_func_cmd cmd)
{
return bnx2x_state_wait(bp, cmd, &o->pending);
}
/**
* bnx2x_func_state_change_comp - complete the state machine transition
*
* @bp: device handle
* @o:
* @cmd:
*
* Called on state change transition. Completes the state
* machine transition only - no HW interaction.
*/
static inline int bnx2x_func_state_change_comp(struct bnx2x *bp,
struct bnx2x_func_sp_obj *o,
enum bnx2x_func_cmd cmd)
{
unsigned long cur_pending = o->pending;
if (!test_and_clear_bit(cmd, &cur_pending)) {
BNX2X_ERR("Bad MC reply %d for func %d in state %d pending 0x%lx, next_state %d\n",
cmd, BP_FUNC(bp), o->state,
cur_pending, o->next_state);
return -EINVAL;
}
DP(BNX2X_MSG_SP,
"Completing command %d for func %d, setting state to %d\n",
cmd, BP_FUNC(bp), o->next_state);
o->state = o->next_state;
o->next_state = BNX2X_F_STATE_MAX;
/* It's important that o->state and o->next_state are
* updated before o->pending.
*/
wmb();
clear_bit(cmd, &o->pending);
smp_mb__after_atomic();
return 0;
}
/**
* bnx2x_func_comp_cmd - complete the state change command
*
* @bp: device handle
* @o:
* @cmd:
*
* Checks that the arrived completion is expected.
*/
static int bnx2x_func_comp_cmd(struct bnx2x *bp,
struct bnx2x_func_sp_obj *o,
enum bnx2x_func_cmd cmd)
{
/* Complete the state machine part first, check if it's a
* legal completion.
*/
int rc = bnx2x_func_state_change_comp(bp, o, cmd);
return rc;
}
/**
* bnx2x_func_chk_transition - perform function state machine transition
*
* @bp: device handle
* @o:
* @params:
*
* It both checks if the requested command is legal in a current
* state and, if it's legal, sets a `next_state' in the object
* that will be used in the completion flow to set the `state'
* of the object.
*
* returns 0 if a requested command is a legal transition,
* -EINVAL otherwise.
*/
static int bnx2x_func_chk_transition(struct bnx2x *bp,
struct bnx2x_func_sp_obj *o,
struct bnx2x_func_state_params *params)
{
enum bnx2x_func_state state = o->state, next_state = BNX2X_F_STATE_MAX;
enum bnx2x_func_cmd cmd = params->cmd;
/* Forget all pending for completion commands if a driver only state
* transition has been requested.
*/
if (test_bit(RAMROD_DRV_CLR_ONLY, &params->ramrod_flags)) {
o->pending = 0;
o->next_state = BNX2X_F_STATE_MAX;
}
/* Don't allow a next state transition if we are in the middle of
* the previous one.
*/
if (o->pending)
return -EBUSY;
switch (state) {
case BNX2X_F_STATE_RESET:
if (cmd == BNX2X_F_CMD_HW_INIT)
next_state = BNX2X_F_STATE_INITIALIZED;
break;
case BNX2X_F_STATE_INITIALIZED:
if (cmd == BNX2X_F_CMD_START)
next_state = BNX2X_F_STATE_STARTED;
else if (cmd == BNX2X_F_CMD_HW_RESET)
next_state = BNX2X_F_STATE_RESET;
break;
case BNX2X_F_STATE_STARTED:
if (cmd == BNX2X_F_CMD_STOP)
next_state = BNX2X_F_STATE_INITIALIZED;
/* afex ramrods can be sent only in started mode, and only
* if not pending for function_stop ramrod completion
* for these events - next state remained STARTED.
*/
else if ((cmd == BNX2X_F_CMD_AFEX_UPDATE) &&
(!test_bit(BNX2X_F_CMD_STOP, &o->pending)))
next_state = BNX2X_F_STATE_STARTED;
else if ((cmd == BNX2X_F_CMD_AFEX_VIFLISTS) &&
(!test_bit(BNX2X_F_CMD_STOP, &o->pending)))
next_state = BNX2X_F_STATE_STARTED;
/* Switch_update ramrod can be sent in either started or
* tx_stopped state, and it doesn't change the state.
*/
else if ((cmd == BNX2X_F_CMD_SWITCH_UPDATE) &&
(!test_bit(BNX2X_F_CMD_STOP, &o->pending)))
next_state = BNX2X_F_STATE_STARTED;
else if ((cmd == BNX2X_F_CMD_SET_TIMESYNC) &&
(!test_bit(BNX2X_F_CMD_STOP, &o->pending)))
next_state = BNX2X_F_STATE_STARTED;
else if (cmd == BNX2X_F_CMD_TX_STOP)
next_state = BNX2X_F_STATE_TX_STOPPED;
break;
case BNX2X_F_STATE_TX_STOPPED:
if ((cmd == BNX2X_F_CMD_SWITCH_UPDATE) &&
(!test_bit(BNX2X_F_CMD_STOP, &o->pending)))
next_state = BNX2X_F_STATE_TX_STOPPED;
else if ((cmd == BNX2X_F_CMD_SET_TIMESYNC) &&
(!test_bit(BNX2X_F_CMD_STOP, &o->pending)))
next_state = BNX2X_F_STATE_TX_STOPPED;
else if (cmd == BNX2X_F_CMD_TX_START)
next_state = BNX2X_F_STATE_STARTED;
break;
default:
BNX2X_ERR("Unknown state: %d\n", state);
}
/* Transition is assured */
if (next_state != BNX2X_F_STATE_MAX) {
DP(BNX2X_MSG_SP, "Good function state transition: %d(%d)->%d\n",
state, cmd, next_state);
o->next_state = next_state;
return 0;
}
DP(BNX2X_MSG_SP, "Bad function state transition request: %d %d\n",
state, cmd);
return -EINVAL;
}
/**
* bnx2x_func_init_func - performs HW init at function stage
*
* @bp: device handle
* @drv:
*
* Init HW when the current phase is
* FW_MSG_CODE_DRV_LOAD_FUNCTION: initialize only FUNCTION-only
* HW blocks.
*/
static inline int bnx2x_func_init_func(struct bnx2x *bp,
const struct bnx2x_func_sp_drv_ops *drv)
{
return drv->init_hw_func(bp);
}
/**
* bnx2x_func_init_port - performs HW init at port stage
*
* @bp: device handle
* @drv:
*
* Init HW when the current phase is
* FW_MSG_CODE_DRV_LOAD_PORT: initialize PORT-only and
* FUNCTION-only HW blocks.
*
*/
static inline int bnx2x_func_init_port(struct bnx2x *bp,
const struct bnx2x_func_sp_drv_ops *drv)
{
int rc = drv->init_hw_port(bp);
if (rc)
return rc;
return bnx2x_func_init_func(bp, drv);
}
/**
* bnx2x_func_init_cmn_chip - performs HW init at chip-common stage
*
* @bp: device handle
* @drv:
*
* Init HW when the current phase is
* FW_MSG_CODE_DRV_LOAD_COMMON_CHIP: initialize COMMON_CHIP,
* PORT-only and FUNCTION-only HW blocks.
*/
static inline int bnx2x_func_init_cmn_chip(struct bnx2x *bp,
const struct bnx2x_func_sp_drv_ops *drv)
{
int rc = drv->init_hw_cmn_chip(bp);
if (rc)
return rc;
return bnx2x_func_init_port(bp, drv);
}
/**
* bnx2x_func_init_cmn - performs HW init at common stage
*
* @bp: device handle
* @drv:
*
* Init HW when the current phase is
* FW_MSG_CODE_DRV_LOAD_COMMON_CHIP: initialize COMMON,
* PORT-only and FUNCTION-only HW blocks.
*/
static inline int bnx2x_func_init_cmn(struct bnx2x *bp,
const struct bnx2x_func_sp_drv_ops *drv)
{
int rc = drv->init_hw_cmn(bp);
if (rc)
return rc;
return bnx2x_func_init_port(bp, drv);
}
static int bnx2x_func_hw_init(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
u32 load_code = params->params.hw_init.load_phase;
struct bnx2x_func_sp_obj *o = params->f_obj;
const struct bnx2x_func_sp_drv_ops *drv = o->drv;
int rc = 0;
DP(BNX2X_MSG_SP, "function %d load_code %x\n",
BP_ABS_FUNC(bp), load_code);
/* Prepare buffers for unzipping the FW */
rc = drv->gunzip_init(bp);
if (rc)
return rc;
/* Prepare FW */
rc = drv->init_fw(bp);
if (rc) {
BNX2X_ERR("Error loading firmware\n");
goto init_err;
}
/* Handle the beginning of COMMON_XXX pases separately... */
switch (load_code) {
case FW_MSG_CODE_DRV_LOAD_COMMON_CHIP:
rc = bnx2x_func_init_cmn_chip(bp, drv);
if (rc)
goto init_err;
break;
case FW_MSG_CODE_DRV_LOAD_COMMON:
rc = bnx2x_func_init_cmn(bp, drv);
if (rc)
goto init_err;
break;
case FW_MSG_CODE_DRV_LOAD_PORT:
rc = bnx2x_func_init_port(bp, drv);
if (rc)
goto init_err;
break;
case FW_MSG_CODE_DRV_LOAD_FUNCTION:
rc = bnx2x_func_init_func(bp, drv);
if (rc)
goto init_err;
break;
default:
BNX2X_ERR("Unknown load_code (0x%x) from MCP\n", load_code);
rc = -EINVAL;
}
init_err:
drv->gunzip_end(bp);
/* In case of success, complete the command immediately: no ramrods
* have been sent.
*/
if (!rc)
o->complete_cmd(bp, o, BNX2X_F_CMD_HW_INIT);
return rc;
}
/**
* bnx2x_func_reset_func - reset HW at function stage
*
* @bp: device handle
* @drv:
*
* Reset HW at FW_MSG_CODE_DRV_UNLOAD_FUNCTION stage: reset only
* FUNCTION-only HW blocks.
*/
static inline void bnx2x_func_reset_func(struct bnx2x *bp,
const struct bnx2x_func_sp_drv_ops *drv)
{
drv->reset_hw_func(bp);
}
/**
* bnx2x_func_reset_port - reset HW at port stage
*
* @bp: device handle
* @drv:
*
* Reset HW at FW_MSG_CODE_DRV_UNLOAD_PORT stage: reset
* FUNCTION-only and PORT-only HW blocks.
*
* !!!IMPORTANT!!!
*
* It's important to call reset_port before reset_func() as the last thing
* reset_func does is pf_disable() thus disabling PGLUE_B, which
* makes impossible any DMAE transactions.
*/
static inline void bnx2x_func_reset_port(struct bnx2x *bp,
const struct bnx2x_func_sp_drv_ops *drv)
{
drv->reset_hw_port(bp);
bnx2x_func_reset_func(bp, drv);
}
/**
* bnx2x_func_reset_cmn - reset HW at common stage
*
* @bp: device handle
* @drv:
*
* Reset HW at FW_MSG_CODE_DRV_UNLOAD_COMMON and
* FW_MSG_CODE_DRV_UNLOAD_COMMON_CHIP stages: reset COMMON,
* COMMON_CHIP, FUNCTION-only and PORT-only HW blocks.
*/
static inline void bnx2x_func_reset_cmn(struct bnx2x *bp,
const struct bnx2x_func_sp_drv_ops *drv)
{
bnx2x_func_reset_port(bp, drv);
drv->reset_hw_cmn(bp);
}
static inline int bnx2x_func_hw_reset(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
u32 reset_phase = params->params.hw_reset.reset_phase;
struct bnx2x_func_sp_obj *o = params->f_obj;
const struct bnx2x_func_sp_drv_ops *drv = o->drv;
DP(BNX2X_MSG_SP, "function %d reset_phase %x\n", BP_ABS_FUNC(bp),
reset_phase);
switch (reset_phase) {
case FW_MSG_CODE_DRV_UNLOAD_COMMON:
bnx2x_func_reset_cmn(bp, drv);
break;
case FW_MSG_CODE_DRV_UNLOAD_PORT:
bnx2x_func_reset_port(bp, drv);
break;
case FW_MSG_CODE_DRV_UNLOAD_FUNCTION:
bnx2x_func_reset_func(bp, drv);
break;
default:
BNX2X_ERR("Unknown reset_phase (0x%x) from MCP\n",
reset_phase);
break;
}
/* Complete the command immediately: no ramrods have been sent. */
o->complete_cmd(bp, o, BNX2X_F_CMD_HW_RESET);
return 0;
}
static inline int bnx2x_func_send_start(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
struct bnx2x_func_sp_obj *o = params->f_obj;
struct function_start_data *rdata =
(struct function_start_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
struct bnx2x_func_start_params *start_params = &params->params.start;
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data with provided parameters */
rdata->function_mode = (u8)start_params->mf_mode;
rdata->sd_vlan_tag = cpu_to_le16(start_params->sd_vlan_tag);
rdata->path_id = BP_PATH(bp);
rdata->network_cos_mode = start_params->network_cos_mode;
rdata->dmae_cmd_id = BNX2X_FW_DMAE_C;
rdata->vxlan_dst_port = cpu_to_le16(start_params->vxlan_dst_port);
rdata->geneve_dst_port = cpu_to_le16(start_params->geneve_dst_port);
rdata->inner_clss_l2gre = start_params->inner_clss_l2gre;
rdata->inner_clss_l2geneve = start_params->inner_clss_l2geneve;
rdata->inner_clss_vxlan = start_params->inner_clss_vxlan;
rdata->inner_rss = start_params->inner_rss;
rdata->sd_accept_mf_clss_fail = start_params->class_fail;
if (start_params->class_fail_ethtype) {
rdata->sd_accept_mf_clss_fail_match_ethtype = 1;
rdata->sd_accept_mf_clss_fail_ethtype =
cpu_to_le16(start_params->class_fail_ethtype);
}
rdata->sd_vlan_force_pri_flg = start_params->sd_vlan_force_pri;
rdata->sd_vlan_force_pri_val = start_params->sd_vlan_force_pri_val;
if (start_params->sd_vlan_eth_type)
rdata->sd_vlan_eth_type =
cpu_to_le16(start_params->sd_vlan_eth_type);
else
rdata->sd_vlan_eth_type =
cpu_to_le16(0x8100);
rdata->no_added_tags = start_params->no_added_tags;
rdata->c2s_pri_tt_valid = start_params->c2s_pri_valid;
if (rdata->c2s_pri_tt_valid) {
memcpy(rdata->c2s_pri_trans_table.val,
start_params->c2s_pri,
MAX_VLAN_PRIORITIES);
rdata->c2s_pri_default = start_params->c2s_pri_default;
}
/* No need for an explicit memory barrier here as long we would
* need to ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read and we will have to put a full memory barrier there
* (inside bnx2x_sp_post()).
*/
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_FUNCTION_START, 0,
U64_HI(data_mapping),
U64_LO(data_mapping), NONE_CONNECTION_TYPE);
}
static inline int bnx2x_func_send_switch_update(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
struct bnx2x_func_sp_obj *o = params->f_obj;
struct function_update_data *rdata =
(struct function_update_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
struct bnx2x_func_switch_update_params *switch_update_params =
&params->params.switch_update;
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data with provided parameters */
if (test_bit(BNX2X_F_UPDATE_TX_SWITCH_SUSPEND_CHNG,
&switch_update_params->changes)) {
rdata->tx_switch_suspend_change_flg = 1;
rdata->tx_switch_suspend =
test_bit(BNX2X_F_UPDATE_TX_SWITCH_SUSPEND,
&switch_update_params->changes);
}
if (test_bit(BNX2X_F_UPDATE_SD_VLAN_TAG_CHNG,
&switch_update_params->changes)) {
rdata->sd_vlan_tag_change_flg = 1;
rdata->sd_vlan_tag =
cpu_to_le16(switch_update_params->vlan);
}
if (test_bit(BNX2X_F_UPDATE_SD_VLAN_ETH_TYPE_CHNG,
&switch_update_params->changes)) {
rdata->sd_vlan_eth_type_change_flg = 1;
rdata->sd_vlan_eth_type =
cpu_to_le16(switch_update_params->vlan_eth_type);
}
if (test_bit(BNX2X_F_UPDATE_VLAN_FORCE_PRIO_CHNG,
&switch_update_params->changes)) {
rdata->sd_vlan_force_pri_change_flg = 1;
if (test_bit(BNX2X_F_UPDATE_VLAN_FORCE_PRIO_FLAG,
&switch_update_params->changes))
rdata->sd_vlan_force_pri_flg = 1;
rdata->sd_vlan_force_pri_flg =
switch_update_params->vlan_force_prio;
}
if (test_bit(BNX2X_F_UPDATE_TUNNEL_CFG_CHNG,
&switch_update_params->changes)) {
rdata->update_tunn_cfg_flg = 1;
if (test_bit(BNX2X_F_UPDATE_TUNNEL_INNER_CLSS_L2GRE,
&switch_update_params->changes))
rdata->inner_clss_l2gre = 1;
if (test_bit(BNX2X_F_UPDATE_TUNNEL_INNER_CLSS_VXLAN,
&switch_update_params->changes))
rdata->inner_clss_vxlan = 1;
if (test_bit(BNX2X_F_UPDATE_TUNNEL_INNER_CLSS_L2GENEVE,
&switch_update_params->changes))
rdata->inner_clss_l2geneve = 1;
if (test_bit(BNX2X_F_UPDATE_TUNNEL_INNER_RSS,
&switch_update_params->changes))
rdata->inner_rss = 1;
rdata->vxlan_dst_port =
cpu_to_le16(switch_update_params->vxlan_dst_port);
rdata->geneve_dst_port =
cpu_to_le16(switch_update_params->geneve_dst_port);
}
rdata->echo = SWITCH_UPDATE;
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_FUNCTION_UPDATE, 0,
U64_HI(data_mapping),
U64_LO(data_mapping), NONE_CONNECTION_TYPE);
}
static inline int bnx2x_func_send_afex_update(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
struct bnx2x_func_sp_obj *o = params->f_obj;
struct function_update_data *rdata =
(struct function_update_data *)o->afex_rdata;
dma_addr_t data_mapping = o->afex_rdata_mapping;
struct bnx2x_func_afex_update_params *afex_update_params =
&params->params.afex_update;
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data with provided parameters */
rdata->vif_id_change_flg = 1;
rdata->vif_id = cpu_to_le16(afex_update_params->vif_id);
rdata->afex_default_vlan_change_flg = 1;
rdata->afex_default_vlan =
cpu_to_le16(afex_update_params->afex_default_vlan);
rdata->allowed_priorities_change_flg = 1;
rdata->allowed_priorities = afex_update_params->allowed_priorities;
rdata->echo = AFEX_UPDATE;
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
DP(BNX2X_MSG_SP,
"afex: sending func_update vif_id 0x%x dvlan 0x%x prio 0x%x\n",
rdata->vif_id,
rdata->afex_default_vlan, rdata->allowed_priorities);
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_FUNCTION_UPDATE, 0,
U64_HI(data_mapping),
U64_LO(data_mapping), NONE_CONNECTION_TYPE);
}
static
inline int bnx2x_func_send_afex_viflists(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
struct bnx2x_func_sp_obj *o = params->f_obj;
struct afex_vif_list_ramrod_data *rdata =
(struct afex_vif_list_ramrod_data *)o->afex_rdata;
struct bnx2x_func_afex_viflists_params *afex_vif_params =
&params->params.afex_viflists;
u64 *p_rdata = (u64 *)rdata;
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data with provided parameters */
rdata->vif_list_index = cpu_to_le16(afex_vif_params->vif_list_index);
rdata->func_bit_map = afex_vif_params->func_bit_map;
rdata->afex_vif_list_command = afex_vif_params->afex_vif_list_command;
rdata->func_to_clear = afex_vif_params->func_to_clear;
/* send in echo type of sub command */
rdata->echo = afex_vif_params->afex_vif_list_command;
/* No need for an explicit memory barrier here as long we would
* need to ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read and we will have to put a full memory barrier there
* (inside bnx2x_sp_post()).
*/
DP(BNX2X_MSG_SP, "afex: ramrod lists, cmd 0x%x index 0x%x func_bit_map 0x%x func_to_clr 0x%x\n",
rdata->afex_vif_list_command, rdata->vif_list_index,
rdata->func_bit_map, rdata->func_to_clear);
/* this ramrod sends data directly and not through DMA mapping */
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_AFEX_VIF_LISTS, 0,
U64_HI(*p_rdata), U64_LO(*p_rdata),
NONE_CONNECTION_TYPE);
}
static inline int bnx2x_func_send_stop(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_FUNCTION_STOP, 0, 0, 0,
NONE_CONNECTION_TYPE);
}
static inline int bnx2x_func_send_tx_stop(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_STOP_TRAFFIC, 0, 0, 0,
NONE_CONNECTION_TYPE);
}
static inline int bnx2x_func_send_tx_start(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
struct bnx2x_func_sp_obj *o = params->f_obj;
struct flow_control_configuration *rdata =
(struct flow_control_configuration *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
struct bnx2x_func_tx_start_params *tx_start_params =
&params->params.tx_start;
int i;
memset(rdata, 0, sizeof(*rdata));
rdata->dcb_enabled = tx_start_params->dcb_enabled;
rdata->dcb_version = tx_start_params->dcb_version;
rdata->dont_add_pri_0_en = tx_start_params->dont_add_pri_0_en;
for (i = 0; i < ARRAY_SIZE(rdata->traffic_type_to_priority_cos); i++)
rdata->traffic_type_to_priority_cos[i] =
tx_start_params->traffic_type_to_priority_cos[i];
for (i = 0; i < MAX_TRAFFIC_TYPES; i++)
rdata->dcb_outer_pri[i] = tx_start_params->dcb_outer_pri[i];
/* No need for an explicit memory barrier here as long as we
* ensure the ordering of writing to the SPQ element
* and updating of the SPQ producer which involves a memory
* read. If the memory read is removed we will have to put a
* full memory barrier there (inside bnx2x_sp_post()).
*/
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_START_TRAFFIC, 0,
U64_HI(data_mapping),
U64_LO(data_mapping), NONE_CONNECTION_TYPE);
}
static inline
int bnx2x_func_send_set_timesync(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
struct bnx2x_func_sp_obj *o = params->f_obj;
struct set_timesync_ramrod_data *rdata =
(struct set_timesync_ramrod_data *)o->rdata;
dma_addr_t data_mapping = o->rdata_mapping;
struct bnx2x_func_set_timesync_params *set_timesync_params =
&params->params.set_timesync;
memset(rdata, 0, sizeof(*rdata));
/* Fill the ramrod data with provided parameters */
rdata->drift_adjust_cmd = set_timesync_params->drift_adjust_cmd;
rdata->offset_cmd = set_timesync_params->offset_cmd;
rdata->add_sub_drift_adjust_value =
set_timesync_params->add_sub_drift_adjust_value;
rdata->drift_adjust_value = set_timesync_params->drift_adjust_value;
rdata->drift_adjust_period = set_timesync_params->drift_adjust_period;
rdata->offset_delta.lo =
cpu_to_le32(U64_LO(set_timesync_params->offset_delta));
rdata->offset_delta.hi =
cpu_to_le32(U64_HI(set_timesync_params->offset_delta));
DP(BNX2X_MSG_SP, "Set timesync command params: drift_cmd = %d, offset_cmd = %d, add_sub_drift = %d, drift_val = %d, drift_period = %d, offset_lo = %d, offset_hi = %d\n",
rdata->drift_adjust_cmd, rdata->offset_cmd,
rdata->add_sub_drift_adjust_value, rdata->drift_adjust_value,
rdata->drift_adjust_period, rdata->offset_delta.lo,
rdata->offset_delta.hi);
return bnx2x_sp_post(bp, RAMROD_CMD_ID_COMMON_SET_TIMESYNC, 0,
U64_HI(data_mapping),
U64_LO(data_mapping), NONE_CONNECTION_TYPE);
}
static int bnx2x_func_send_cmd(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
switch (params->cmd) {
case BNX2X_F_CMD_HW_INIT:
return bnx2x_func_hw_init(bp, params);
case BNX2X_F_CMD_START:
return bnx2x_func_send_start(bp, params);
case BNX2X_F_CMD_STOP:
return bnx2x_func_send_stop(bp, params);
case BNX2X_F_CMD_HW_RESET:
return bnx2x_func_hw_reset(bp, params);
case BNX2X_F_CMD_AFEX_UPDATE:
return bnx2x_func_send_afex_update(bp, params);
case BNX2X_F_CMD_AFEX_VIFLISTS:
return bnx2x_func_send_afex_viflists(bp, params);
case BNX2X_F_CMD_TX_STOP:
return bnx2x_func_send_tx_stop(bp, params);
case BNX2X_F_CMD_TX_START:
return bnx2x_func_send_tx_start(bp, params);
case BNX2X_F_CMD_SWITCH_UPDATE:
return bnx2x_func_send_switch_update(bp, params);
case BNX2X_F_CMD_SET_TIMESYNC:
return bnx2x_func_send_set_timesync(bp, params);
default:
BNX2X_ERR("Unknown command: %d\n", params->cmd);
return -EINVAL;
}
}
void bnx2x_init_func_obj(struct bnx2x *bp,
struct bnx2x_func_sp_obj *obj,
void *rdata, dma_addr_t rdata_mapping,
void *afex_rdata, dma_addr_t afex_rdata_mapping,
struct bnx2x_func_sp_drv_ops *drv_iface)
{
memset(obj, 0, sizeof(*obj));
mutex_init(&obj->one_pending_mutex);
obj->rdata = rdata;
obj->rdata_mapping = rdata_mapping;
obj->afex_rdata = afex_rdata;
obj->afex_rdata_mapping = afex_rdata_mapping;
obj->send_cmd = bnx2x_func_send_cmd;
obj->check_transition = bnx2x_func_chk_transition;
obj->complete_cmd = bnx2x_func_comp_cmd;
obj->wait_comp = bnx2x_func_wait_comp;
obj->drv = drv_iface;
}
/**
* bnx2x_func_state_change - perform Function state change transition
*
* @bp: device handle
* @params: parameters to perform the transaction
*
* returns 0 in case of successfully completed transition,
* negative error code in case of failure, positive
* (EBUSY) value if there is a completion to that is
* still pending (possible only if RAMROD_COMP_WAIT is
* not set in params->ramrod_flags for asynchronous
* commands).
*/
int bnx2x_func_state_change(struct bnx2x *bp,
struct bnx2x_func_state_params *params)
{
struct bnx2x_func_sp_obj *o = params->f_obj;
int rc, cnt = 300;
enum bnx2x_func_cmd cmd = params->cmd;
unsigned long *pending = &o->pending;
mutex_lock(&o->one_pending_mutex);
/* Check that the requested transition is legal */
rc = o->check_transition(bp, o, params);
if ((rc == -EBUSY) &&
(test_bit(RAMROD_RETRY, &params->ramrod_flags))) {
while ((rc == -EBUSY) && (--cnt > 0)) {
mutex_unlock(&o->one_pending_mutex);
msleep(10);
mutex_lock(&o->one_pending_mutex);
rc = o->check_transition(bp, o, params);
}
if (rc == -EBUSY) {
mutex_unlock(&o->one_pending_mutex);
BNX2X_ERR("timeout waiting for previous ramrod completion\n");
return rc;
}
} else if (rc) {
mutex_unlock(&o->one_pending_mutex);
return rc;
}
/* Set "pending" bit */
set_bit(cmd, pending);
/* Don't send a command if only driver cleanup was requested */
if (test_bit(RAMROD_DRV_CLR_ONLY, &params->ramrod_flags)) {
bnx2x_func_state_change_comp(bp, o, cmd);
mutex_unlock(&o->one_pending_mutex);
} else {
/* Send a ramrod */
rc = o->send_cmd(bp, params);
mutex_unlock(&o->one_pending_mutex);
if (rc) {
o->next_state = BNX2X_F_STATE_MAX;
clear_bit(cmd, pending);
smp_mb__after_atomic();
return rc;
}
if (test_bit(RAMROD_COMP_WAIT, &params->ramrod_flags)) {
rc = o->wait_comp(bp, o, cmd);
if (rc)
return rc;
return 0;
}
}
return !!test_bit(cmd, pending);
}