1
0
Fork 0
alistair23-linux/drivers/scsi/NCR5380.c

2338 lines
72 KiB
C
Raw Normal View History

/*
* NCR 5380 generic driver routines. These should make it *trivial*
* to implement 5380 SCSI drivers under Linux with a non-trantor
* architecture.
*
* Note that these routines also work with NR53c400 family chips.
*
* Copyright 1993, Drew Eckhardt
* Visionary Computing
* (Unix and Linux consulting and custom programming)
* drew@colorado.edu
* +1 (303) 666-5836
*
* For more information, please consult
*
* NCR 5380 Family
* SCSI Protocol Controller
* Databook
*
* NCR Microelectronics
* 1635 Aeroplaza Drive
* Colorado Springs, CO 80916
* 1+ (719) 578-3400
* 1+ (800) 334-5454
*/
/*
* Revision 1.10 1998/9/2 Alan Cox
* (alan@lxorguk.ukuu.org.uk)
* Fixed up the timer lockups reported so far. Things still suck. Looking
* forward to 2.3 and per device request queues. Then it'll be possible to
* SMP thread this beast and improve life no end.
* Revision 1.9 1997/7/27 Ronald van Cuijlenborg
* (ronald.van.cuijlenborg@tip.nl or nutty@dds.nl)
* (hopefully) fixed and enhanced USLEEP
* added support for DTC3181E card (for Mustek scanner)
*
* Revision 1.8 Ingmar Baumgart
* (ingmar@gonzo.schwaben.de)
* added support for NCR53C400a card
*
* Revision 1.7 1996/3/2 Ray Van Tassle (rayvt@comm.mot.com)
* added proc_info
* added support needed for DTC 3180/3280
* fixed a couple of bugs
*
* Revision 1.5 1994/01/19 09:14:57 drew
* Fixed udelay() hack that was being used on DATAOUT phases
* instead of a proper wait for the final handshake.
*
* Revision 1.4 1994/01/19 06:44:25 drew
* *** empty log message ***
*
* Revision 1.3 1994/01/19 05:24:40 drew
* Added support for TCR LAST_BYTE_SENT bit.
*
* Revision 1.2 1994/01/15 06:14:11 drew
* REAL DMA support, bug fixes.
*
* Revision 1.1 1994/01/15 06:00:54 drew
* Initial revision
*
*/
/*
* Further development / testing that should be done :
* 1. Cleanup the NCR5380_transfer_dma function and DMA operation complete
* code so that everything does the same thing that's done at the
* end of a pseudo-DMA read operation.
*
* 2. Fix REAL_DMA (interrupt driven, polled works fine) -
* basically, transfer size needs to be reduced by one
* and the last byte read as is done with PSEUDO_DMA.
*
* 4. Test SCSI-II tagged queueing (I have no devices which support
* tagged queueing)
*/
#ifndef notyet
#undef REAL_DMA
#endif
#ifdef BOARD_REQUIRES_NO_DELAY
#define io_recovery_delay(x)
#else
#define io_recovery_delay(x) udelay(x)
#endif
/*
* Design
*
* This is a generic 5380 driver. To use it on a different platform,
* one simply writes appropriate system specific macros (ie, data
* transfer - some PC's will use the I/O bus, 68K's must use
* memory mapped) and drops this file in their 'C' wrapper.
*
* (Note from hch: unfortunately it was not enough for the different
* m68k folks and instead of improving this driver they copied it
* and hacked it up for their needs. As a consequence they lost
* most updates to this driver. Maybe someone will fix all these
* drivers to use a common core one day..)
*
* As far as command queueing, two queues are maintained for
* each 5380 in the system - commands that haven't been issued yet,
* and commands that are currently executing. This means that an
* unlimited number of commands may be queued, letting
* more commands propagate from the higher driver levels giving higher
* throughput. Note that both I_T_L and I_T_L_Q nexuses are supported,
* allowing multiple commands to propagate all the way to a SCSI-II device
* while a command is already executing.
*
*
* Issues specific to the NCR5380 :
*
* When used in a PIO or pseudo-dma mode, the NCR5380 is a braindead
* piece of hardware that requires you to sit in a loop polling for
* the REQ signal as long as you are connected. Some devices are
* brain dead (ie, many TEXEL CD ROM drives) and won't disconnect
ncr5380: Rework disconnect versus poll logic The atari_NCR5380.c and NCR5380.c core drivers differ in their handling of target disconnection. This is partly because atari_NCR5380.c had all of the polling and sleeping removed to become entirely interrupt-driven, and it is partly because of damage done to NCR5380.c after atari_NCR5380.c was forked. See commit 37cd23b44929 ("Linux 2.1.105") in history/history.git. The polling changes that were made in v2.1.105 are questionable at best: if REQ is not already asserted when NCR5380_transfer_pio() is invoked, and if the expected phase is DATA IN or DATA OUT, the function will schedule main() to execute after USLEEP_SLEEP jiffies and then return. The problems here are the expected REQ timing and the sleep interval*. Avoid this issue by using NCR5380_poll_politely() instead of scheduling main(). The atari_NCR5380.c core driver requires the use of the chip interrupt and always permits target disconnection. It sets the cmd->device->disconnect flag when a device disconnects, but never tests this flag. The NCR5380.c core driver permits disconnection only when instance->irq != NO_IRQ. It sets the cmd->device->disconnect flag when a device disconnects and it tests this flag in a couple of places: 1. During NCR5380_information_transfer(), following COMMAND OUT phase, if !cmd->device->disconnect, the initiator will take a guess as to whether or not the target will then choose to go to MESSAGE IN phase and disconnect. If the driver guesses "yes", it will schedule main() to execute after USLEEP_SLEEP jiffies and then return there. Unfortunately the driver may guess "yes" even after it has denied the target the disconnection privilege. When the target does not disconnect, the sleep can be beneficial, assuming the sleep interval is appropriate (mostly it is not*). And even if the driver guesses "yes" correctly, and the target would then disconnect, the driver still has to go through the MESSAGE IN phase in order to get to BUS FREE phase. The main loop can do nothing useful until BUS FREE, and sleeping just delays the phase transition. 2. If !cmd->device->disconnect and REQ is not already asserted when NCR5380_information_transfer() is invoked, the function polls for REQ for USLEEP_POLL jiffies. If REQ is not asserted, it then schedules main() to execute after USLEEP_SLEEP jiffies and returns. The idea is apparently to yeild the CPU while waiting for REQ. This is conditional upon !cmd->device->disconnect, but there seems to be no rhyme or reason for that. For example, the flag may be unset because disconnection privilege was denied because the driver has no IRQ. Or the flag may be unset because the device has never needed to disconnect before. Or if the flag is set, disconnection may have no relevance to the present bus phase. Another deficiency of the existing algorithm is as follows. When the driver has no IRQ, it prevents disconnection, and generally polls and sleeps more than it would normally. Now, if the driver is going to poll anyway, why not allow the target to disconnect? That way the driver can do something useful with the bus instead of polling unproductively! Avoid this pointless latency, complexity and guesswork by using NCR5380_poll_politely() instead of scheduling main(). * For g_NCR5380, the time intervals for USLEEP_SLEEP and USLEEP_POLL are 200 ms and 10 ms, respectively. They are 20 ms and 200 ms respectively for the other NCR5380 drivers. There doesn't seem to be any reason for this discrepancy. The timing seems to have no relation to the type of adapter. Bizarrely, the timing in g_NCR5380 seems to relate only to one particular type of target device. This patch attempts to solve the problem for all NCR5380 drivers and all target devices. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:26 -07:00
* while doing long seek operations. [...] These
* broken devices are the exception rather than the rule and I'd rather
* spend my time optimizing for the normal case.
*
* Architecture :
*
* At the heart of the design is a coroutine, NCR5380_main,
* which is started from a workqueue for each NCR5380 host in the
* system. It attempts to establish I_T_L or I_T_L_Q nexuses by
* removing the commands from the issue queue and calling
* NCR5380_select() if a nexus is not established.
*
* Once a nexus is established, the NCR5380_information_transfer()
* phase goes through the various phases as instructed by the target.
* if the target goes into MSG IN and sends a DISCONNECT message,
* the command structure is placed into the per instance disconnected
* queue, and NCR5380_main tries to find more work. If the target is
* idle for too long, the system will try to sleep.
*
* If a command has disconnected, eventually an interrupt will trigger,
* calling NCR5380_intr() which will in turn call NCR5380_reselect
* to reestablish a nexus. This will run main if necessary.
*
* On command termination, the done function will be called as
* appropriate.
*
* SCSI pointers are maintained in the SCp field of SCSI command
* structures, being initialized after the command is connected
* in NCR5380_select, and set as appropriate in NCR5380_information_transfer.
* Note that in violation of the standard, an implicit SAVE POINTERS operation
* is done, since some BROKEN disks fail to issue an explicit SAVE POINTERS.
*/
/*
* Using this file :
* This file a skeleton Linux SCSI driver for the NCR 5380 series
* of chips. To use it, you write an architecture specific functions
* and macros and include this file in your driver.
*
* These macros control options :
* AUTOPROBE_IRQ - if defined, the NCR5380_probe_irq() function will be
* defined.
*
* AUTOSENSE - if defined, REQUEST SENSE will be performed automatically
* for commands that return with a CHECK CONDITION status.
*
* DIFFERENTIAL - if defined, NCR53c81 chips will use external differential
* transceivers.
*
* DONT_USE_INTR - if defined, never use interrupts, even if we probe or
* override-configure an IRQ.
*
* PSEUDO_DMA - if defined, PSEUDO DMA is used during the data transfer phases.
*
* REAL_DMA - if defined, REAL DMA is used during the data transfer phases.
*
* REAL_DMA_POLL - if defined, REAL DMA is used but the driver doesn't
* rely on phase mismatch and EOP interrupts to determine end
* of phase.
*
* Defaults for these will be provided although the user may want to adjust
* these to allocate CPU resources to the SCSI driver or "real" code.
*
* These macros MUST be defined :
*
* NCR5380_read(register) - read from the specified register
*
* NCR5380_write(register, value) - write to the specific register
*
* NCR5380_implementation_fields - additional fields needed for this
* specific implementation of the NCR5380
*
* Either real DMA *or* pseudo DMA may be implemented
* REAL functions :
* NCR5380_REAL_DMA should be defined if real DMA is to be used.
* Note that the DMA setup functions should return the number of bytes
* that they were able to program the controller for.
*
* Also note that generic i386/PC versions of these macros are
* available as NCR5380_i386_dma_write_setup,
* NCR5380_i386_dma_read_setup, and NCR5380_i386_dma_residual.
*
* NCR5380_dma_write_setup(instance, src, count) - initialize
* NCR5380_dma_read_setup(instance, dst, count) - initialize
* NCR5380_dma_residual(instance); - residual count
*
* PSEUDO functions :
* NCR5380_pwrite(instance, src, count)
* NCR5380_pread(instance, dst, count);
*
* The generic driver is initialized by calling NCR5380_init(instance),
* after setting the appropriate host specific fields and ID. If the
* driver wishes to autoprobe for an IRQ line, the NCR5380_probe_irq(instance,
* possible) function may be used.
*/
static int do_abort(struct Scsi_Host *);
static void do_reset(struct Scsi_Host *);
/*
* initialize_SCp - init the scsi pointer field
* @cmd: command block to set up
*
* Set up the internal fields in the SCSI command.
*/
static inline void initialize_SCp(struct scsi_cmnd *cmd)
{
/*
* Initialize the Scsi Pointer field so that all of the commands in the
* various queues are valid.
*/
if (scsi_bufflen(cmd)) {
cmd->SCp.buffer = scsi_sglist(cmd);
cmd->SCp.buffers_residual = scsi_sg_count(cmd) - 1;
cmd->SCp.ptr = sg_virt(cmd->SCp.buffer);
cmd->SCp.this_residual = cmd->SCp.buffer->length;
} else {
cmd->SCp.buffer = NULL;
cmd->SCp.buffers_residual = 0;
cmd->SCp.ptr = NULL;
cmd->SCp.this_residual = 0;
}
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
cmd->SCp.Status = 0;
cmd->SCp.Message = 0;
}
/**
* NCR5380_poll_politely2 - wait for two chip register values
* @instance: controller to poll
* @reg1: 5380 register to poll
* @bit1: Bitmask to check
* @val1: Expected value
* @reg2: Second 5380 register to poll
* @bit2: Second bitmask to check
* @val2: Second expected value
* @wait: Time-out in jiffies
*
* Polls the chip in a reasonably efficient manner waiting for an
* event to occur. After a short quick poll we begin to yield the CPU
* (if possible). In irq contexts the time-out is arbitrarily limited.
* Callers may hold locks as long as they are held in irq mode.
*
* Returns 0 if either or both event(s) occurred otherwise -ETIMEDOUT.
*/
static int NCR5380_poll_politely2(struct Scsi_Host *instance,
int reg1, int bit1, int val1,
int reg2, int bit2, int val2, int wait)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
unsigned long deadline = jiffies + wait;
unsigned long n;
/* Busy-wait for up to 10 ms */
n = min(10000U, jiffies_to_usecs(wait));
n *= hostdata->accesses_per_ms;
n /= 2000;
do {
if ((NCR5380_read(reg1) & bit1) == val1)
return 0;
if ((NCR5380_read(reg2) & bit2) == val2)
return 0;
cpu_relax();
} while (n--);
if (irqs_disabled() || in_interrupt())
return -ETIMEDOUT;
/* Repeatedly sleep for 1 ms until deadline */
while (time_is_after_jiffies(deadline)) {
schedule_timeout_uninterruptible(1);
if ((NCR5380_read(reg1) & bit1) == val1)
return 0;
if ((NCR5380_read(reg2) & bit2) == val2)
return 0;
}
return -ETIMEDOUT;
}
static inline int NCR5380_poll_politely(struct Scsi_Host *instance,
int reg, int bit, int val, int wait)
{
return NCR5380_poll_politely2(instance, reg, bit, val,
reg, bit, val, wait);
}
static struct {
unsigned char value;
const char *name;
} phases[] __maybe_unused = {
{PHASE_DATAOUT, "DATAOUT"},
{PHASE_DATAIN, "DATAIN"},
{PHASE_CMDOUT, "CMDOUT"},
{PHASE_STATIN, "STATIN"},
{PHASE_MSGOUT, "MSGOUT"},
{PHASE_MSGIN, "MSGIN"},
{PHASE_UNKNOWN, "UNKNOWN"}
};
#if NDEBUG
static struct {
unsigned char mask;
const char *name;
} signals[] = {
{SR_DBP, "PARITY"},
{SR_RST, "RST"},
{SR_BSY, "BSY"},
{SR_REQ, "REQ"},
{SR_MSG, "MSG"},
{SR_CD, "CD"},
{SR_IO, "IO"},
{SR_SEL, "SEL"},
{0, NULL}
},
basrs[] = {
{BASR_ATN, "ATN"},
{BASR_ACK, "ACK"},
{0, NULL}
},
icrs[] = {
{ICR_ASSERT_RST, "ASSERT RST"},
{ICR_ASSERT_ACK, "ASSERT ACK"},
{ICR_ASSERT_BSY, "ASSERT BSY"},
{ICR_ASSERT_SEL, "ASSERT SEL"},
{ICR_ASSERT_ATN, "ASSERT ATN"},
{ICR_ASSERT_DATA, "ASSERT DATA"},
{0, NULL}
},
mrs[] = {
{MR_BLOCK_DMA_MODE, "MODE BLOCK DMA"},
{MR_TARGET, "MODE TARGET"},
{MR_ENABLE_PAR_CHECK, "MODE PARITY CHECK"},
{MR_ENABLE_PAR_INTR, "MODE PARITY INTR"},
{MR_MONITOR_BSY, "MODE MONITOR BSY"},
{MR_DMA_MODE, "MODE DMA"},
{MR_ARBITRATE, "MODE ARBITRATION"},
{0, NULL}
};
/**
* NCR5380_print - print scsi bus signals
* @instance: adapter state to dump
*
* Print the SCSI bus signals for debugging purposes
*
* Locks: caller holds hostdata lock (not essential)
*/
static void NCR5380_print(struct Scsi_Host *instance)
{
unsigned char status, data, basr, mr, icr, i;
data = NCR5380_read(CURRENT_SCSI_DATA_REG);
status = NCR5380_read(STATUS_REG);
mr = NCR5380_read(MODE_REG);
icr = NCR5380_read(INITIATOR_COMMAND_REG);
basr = NCR5380_read(BUS_AND_STATUS_REG);
printk("STATUS_REG: %02x ", status);
for (i = 0; signals[i].mask; ++i)
if (status & signals[i].mask)
printk(",%s", signals[i].name);
printk("\nBASR: %02x ", basr);
for (i = 0; basrs[i].mask; ++i)
if (basr & basrs[i].mask)
printk(",%s", basrs[i].name);
printk("\nICR: %02x ", icr);
for (i = 0; icrs[i].mask; ++i)
if (icr & icrs[i].mask)
printk(",%s", icrs[i].name);
printk("\nMODE: %02x ", mr);
for (i = 0; mrs[i].mask; ++i)
if (mr & mrs[i].mask)
printk(",%s", mrs[i].name);
printk("\n");
}
/*
* NCR5380_print_phase - show SCSI phase
* @instance: adapter to dump
*
* Print the current SCSI phase for debugging purposes
*
* Locks: none
*/
static void NCR5380_print_phase(struct Scsi_Host *instance)
{
unsigned char status;
int i;
status = NCR5380_read(STATUS_REG);
if (!(status & SR_REQ))
printk("scsi%d : REQ not asserted, phase unknown.\n", instance->host_no);
else {
for (i = 0; (phases[i].value != PHASE_UNKNOWN) && (phases[i].value != (status & PHASE_MASK)); ++i);
printk("scsi%d : phase %s\n", instance->host_no, phases[i].name);
}
}
#endif
static int probe_irq __initdata;
/**
* probe_intr - helper for IRQ autoprobe
* @irq: interrupt number
* @dev_id: unused
* @regs: unused
*
* Set a flag to indicate the IRQ in question was received. This is
* used by the IRQ probe code.
*/
IRQ: Maintain regs pointer globally rather than passing to IRQ handlers Maintain a per-CPU global "struct pt_regs *" variable which can be used instead of passing regs around manually through all ~1800 interrupt handlers in the Linux kernel. The regs pointer is used in few places, but it potentially costs both stack space and code to pass it around. On the FRV arch, removing the regs parameter from all the genirq function results in a 20% speed up of the IRQ exit path (ie: from leaving timer_interrupt() to leaving do_IRQ()). Where appropriate, an arch may override the generic storage facility and do something different with the variable. On FRV, for instance, the address is maintained in GR28 at all times inside the kernel as part of general exception handling. Having looked over the code, it appears that the parameter may be handed down through up to twenty or so layers of functions. Consider a USB character device attached to a USB hub, attached to a USB controller that posts its interrupts through a cascaded auxiliary interrupt controller. A character device driver may want to pass regs to the sysrq handler through the input layer which adds another few layers of parameter passing. I've build this code with allyesconfig for x86_64 and i386. I've runtested the main part of the code on FRV and i386, though I can't test most of the drivers. I've also done partial conversion for powerpc and MIPS - these at least compile with minimal configurations. This will affect all archs. Mostly the changes should be relatively easy. Take do_IRQ(), store the regs pointer at the beginning, saving the old one: struct pt_regs *old_regs = set_irq_regs(regs); And put the old one back at the end: set_irq_regs(old_regs); Don't pass regs through to generic_handle_irq() or __do_IRQ(). In timer_interrupt(), this sort of change will be necessary: - update_process_times(user_mode(regs)); - profile_tick(CPU_PROFILING, regs); + update_process_times(user_mode(get_irq_regs())); + profile_tick(CPU_PROFILING); I'd like to move update_process_times()'s use of get_irq_regs() into itself, except that i386, alone of the archs, uses something other than user_mode(). Some notes on the interrupt handling in the drivers: (*) input_dev() is now gone entirely. The regs pointer is no longer stored in the input_dev struct. (*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does something different depending on whether it's been supplied with a regs pointer or not. (*) Various IRQ handler function pointers have been moved to type irq_handler_t. Signed-Off-By: David Howells <dhowells@redhat.com> (cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
2006-10-05 07:55:46 -06:00
static irqreturn_t __init probe_intr(int irq, void *dev_id)
{
probe_irq = irq;
return IRQ_HANDLED;
}
/**
* NCR5380_probe_irq - find the IRQ of an NCR5380
* @instance: NCR5380 controller
* @possible: bitmask of ISA IRQ lines
*
* Autoprobe for the IRQ line used by the NCR5380 by triggering an IRQ
* and then looking to see what interrupt actually turned up.
*
* Locks: none, irqs must be enabled on entry
*/
static int __init __maybe_unused NCR5380_probe_irq(struct Scsi_Host *instance,
int possible)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
unsigned long timeout;
int trying_irqs, i, mask;
for (trying_irqs = 0, i = 1, mask = 2; i < 16; ++i, mask <<= 1)
if ((mask & possible) && (request_irq(i, &probe_intr, 0, "NCR-probe", NULL) == 0))
trying_irqs |= mask;
timeout = jiffies + msecs_to_jiffies(250);
probe_irq = NO_IRQ;
/*
* A interrupt is triggered whenever BSY = false, SEL = true
* and a bit set in the SELECT_ENABLE_REG is asserted on the
* SCSI bus.
*
* Note that the bus is only driven when the phase control signals
* (I/O, C/D, and MSG) match those in the TCR, so we must reset that
* to zero.
*/
NCR5380_write(TARGET_COMMAND_REG, 0);
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
NCR5380_write(OUTPUT_DATA_REG, hostdata->id_mask);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_DATA | ICR_ASSERT_SEL);
while (probe_irq == NO_IRQ && time_before(jiffies, timeout))
schedule_timeout_uninterruptible(1);
NCR5380_write(SELECT_ENABLE_REG, 0);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
for (i = 1, mask = 2; i < 16; ++i, mask <<= 1)
if (trying_irqs & mask)
free_irq(i, NULL);
return probe_irq;
}
/**
* NCR58380_info - report driver and host information
* @instance: relevant scsi host instance
*
* For use as the host template info() handler.
*
* Locks: none
*/
static const char *NCR5380_info(struct Scsi_Host *instance)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
return hostdata->info;
}
static void prepare_info(struct Scsi_Host *instance)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
snprintf(hostdata->info, sizeof(hostdata->info),
"%s, io_port 0x%lx, n_io_port %d, "
"base 0x%lx, irq %d, "
"can_queue %d, cmd_per_lun %d, "
"sg_tablesize %d, this_id %d, "
"flags { %s%s%s}, "
"options { %s} ",
instance->hostt->name, instance->io_port, instance->n_io_port,
instance->base, instance->irq,
instance->can_queue, instance->cmd_per_lun,
instance->sg_tablesize, instance->this_id,
hostdata->flags & FLAG_NO_DMA_FIXUP ? "NO_DMA_FIXUP " : "",
hostdata->flags & FLAG_NO_PSEUDO_DMA ? "NO_PSEUDO_DMA " : "",
hostdata->flags & FLAG_TOSHIBA_DELAY ? "TOSHIBA_DELAY " : "",
#ifdef AUTOPROBE_IRQ
"AUTOPROBE_IRQ "
#endif
#ifdef DIFFERENTIAL
"DIFFERENTIAL "
#endif
#ifdef REAL_DMA
"REAL_DMA "
#endif
#ifdef REAL_DMA_POLL
"REAL_DMA_POLL "
#endif
#ifdef PARITY
"PARITY "
#endif
#ifdef PSEUDO_DMA
"PSEUDO_DMA "
#endif
"");
}
#ifdef PSEUDO_DMA
/******************************************/
/*
* /proc/scsi/[dtc pas16 t128 generic]/[0-ASC_NUM_BOARD_SUPPORTED]
*
* *buffer: I/O buffer
* **start: if inout == FALSE pointer into buffer where user read should start
* offset: current offset
* length: length of buffer
* hostno: Scsi_Host host_no
* inout: TRUE - user is writing; FALSE - user is reading
*
* Return the number of bytes read from or written
*/
static int __maybe_unused NCR5380_write_info(struct Scsi_Host *instance,
char *buffer, int length)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
hostdata->spin_max_r = 0;
hostdata->spin_max_w = 0;
return 0;
}
static int __maybe_unused NCR5380_show_info(struct seq_file *m,
struct Scsi_Host *instance)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
seq_printf(m, "Highwater I/O busy spin counts: write %d, read %d\n",
hostdata->spin_max_w, hostdata->spin_max_r);
return 0;
}
#endif
/**
* NCR5380_init - initialise an NCR5380
* @instance: adapter to configure
* @flags: control flags
*
* Initializes *instance and corresponding 5380 chip,
* with flags OR'd into the initial flags value.
*
* Notes : I assume that the host, hostno, and id bits have been
* set correctly. I don't care about the irq and other fields.
*
* Returns 0 for success
*
* Locks: interrupts must be enabled when we are called
*/
static int NCR5380_init(struct Scsi_Host *instance, int flags)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
int i;
unsigned long deadline;
if(in_interrupt())
printk(KERN_ERR "NCR5380_init called with interrupts off!\n");
hostdata->id_mask = 1 << instance->this_id;
for (i = hostdata->id_mask; i <= 0x80; i <<= 1)
if (i > hostdata->id_mask)
hostdata->id_higher_mask |= i;
for (i = 0; i < 8; ++i)
hostdata->busy[i] = 0;
#ifdef REAL_DMA
hostdata->dmalen = 0;
#endif
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_init(&hostdata->lock);
hostdata->connected = NULL;
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
hostdata->sensing = NULL;
INIT_LIST_HEAD(&hostdata->autosense);
INIT_LIST_HEAD(&hostdata->unissued);
INIT_LIST_HEAD(&hostdata->disconnected);
hostdata->flags = flags;
INIT_WORK(&hostdata->main_task, NCR5380_main);
hostdata->work_q = alloc_workqueue("ncr5380_%d",
WQ_UNBOUND | WQ_MEM_RECLAIM,
1, instance->host_no);
if (!hostdata->work_q)
return -ENOMEM;
hostdata->host = instance;
prepare_info(instance);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
NCR5380_write(MODE_REG, MR_BASE);
NCR5380_write(TARGET_COMMAND_REG, 0);
NCR5380_write(SELECT_ENABLE_REG, 0);
/* Calibrate register polling loop */
i = 0;
deadline = jiffies + 1;
do {
cpu_relax();
} while (time_is_after_jiffies(deadline));
deadline += msecs_to_jiffies(256);
do {
NCR5380_read(STATUS_REG);
++i;
cpu_relax();
} while (time_is_after_jiffies(deadline));
hostdata->accesses_per_ms = i / 256;
return 0;
}
/**
* NCR5380_maybe_reset_bus - Detect and correct bus wedge problems.
* @instance: adapter to check
*
* If the system crashed, it may have crashed with a connected target and
* the SCSI bus busy. Check for BUS FREE phase. If not, try to abort the
* currently established nexus, which we know nothing about. Failing that
* do a bus reset.
*
* Note that a bus reset will cause the chip to assert IRQ.
*
* Returns 0 if successful, otherwise -ENXIO.
*/
static int NCR5380_maybe_reset_bus(struct Scsi_Host *instance)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
int pass;
for (pass = 1; (NCR5380_read(STATUS_REG) & SR_BSY) && pass <= 6; ++pass) {
switch (pass) {
case 1:
case 3:
case 5:
shost_printk(KERN_ERR, instance, "SCSI bus busy, waiting up to five seconds\n");
NCR5380_poll_politely(instance,
STATUS_REG, SR_BSY, 0, 5 * HZ);
break;
case 2:
shost_printk(KERN_ERR, instance, "bus busy, attempting abort\n");
do_abort(instance);
break;
case 4:
shost_printk(KERN_ERR, instance, "bus busy, attempting reset\n");
do_reset(instance);
/* Wait after a reset; the SCSI standard calls for
* 250ms, we wait 500ms to be on the safe side.
* But some Toshiba CD-ROMs need ten times that.
*/
if (hostdata->flags & FLAG_TOSHIBA_DELAY)
msleep(2500);
else
msleep(500);
break;
case 6:
shost_printk(KERN_ERR, instance, "bus locked solid\n");
return -ENXIO;
}
}
return 0;
}
/**
* NCR5380_exit - remove an NCR5380
* @instance: adapter to remove
*/
static void NCR5380_exit(struct Scsi_Host *instance)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
cancel_work_sync(&hostdata->main_task);
destroy_workqueue(hostdata->work_q);
}
/**
* complete_cmd - finish processing a command and return it to the SCSI ML
* @instance: the host instance
* @cmd: command to complete
*/
static void complete_cmd(struct Scsi_Host *instance,
struct scsi_cmnd *cmd)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
dsprintk(NDEBUG_QUEUES, instance, "complete_cmd: cmd %p\n", cmd);
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
if (hostdata->sensing == cmd) {
/* Autosense processing ends here */
if ((cmd->result & 0xff) != SAM_STAT_GOOD) {
scsi_eh_restore_cmnd(cmd, &hostdata->ses);
set_host_byte(cmd, DID_ERROR);
} else
scsi_eh_restore_cmnd(cmd, &hostdata->ses);
hostdata->sensing = NULL;
}
hostdata->busy[scmd_id(cmd)] &= ~(1 << cmd->device->lun);
cmd->scsi_done(cmd);
}
/**
* NCR5380_queue_command - queue a command
* @instance: the relevant SCSI adapter
* @cmd: SCSI command
*
* cmd is added to the per-instance issue queue, with minor
* twiddling done to the host specific fields of cmd. If the
* main coroutine is not running, it is restarted.
*/
static int NCR5380_queue_command(struct Scsi_Host *instance,
struct scsi_cmnd *cmd)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
struct NCR5380_cmd *ncmd = scsi_cmd_priv(cmd);
unsigned long flags;
#if (NDEBUG & NDEBUG_NO_WRITE)
switch (cmd->cmnd[0]) {
case WRITE_6:
case WRITE_10:
shost_printk(KERN_DEBUG, instance, "WRITE attempted with NDEBUG_NO_WRITE set\n");
cmd->result = (DID_ERROR << 16);
cmd->scsi_done(cmd);
return 0;
}
#endif /* (NDEBUG & NDEBUG_NO_WRITE) */
cmd->result = 0;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irqsave(&hostdata->lock, flags);
/*
* Insert the cmd into the issue queue. Note that REQUEST SENSE
* commands are added to the head of the queue since any command will
* clear the contingent allegiance condition that exists and the
* sense data is only guaranteed to be valid while the condition exists.
*/
if (cmd->cmnd[0] == REQUEST_SENSE)
list_add(&ncmd->list, &hostdata->unissued);
else
list_add_tail(&ncmd->list, &hostdata->unissued);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irqrestore(&hostdata->lock, flags);
dsprintk(NDEBUG_QUEUES, instance, "command %p added to %s of queue\n",
cmd, (cmd->cmnd[0] == REQUEST_SENSE) ? "head" : "tail");
/* Kick off command processing */
queue_work(hostdata->work_q, &hostdata->main_task);
return 0;
}
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
/**
* dequeue_next_cmd - dequeue a command for processing
* @instance: the scsi host instance
*
* Priority is given to commands on the autosense queue. These commands
* need autosense because of a CHECK CONDITION result.
*
* Returns a command pointer if a command is found for a target that is
* not already busy. Otherwise returns NULL.
*/
static struct scsi_cmnd *dequeue_next_cmd(struct Scsi_Host *instance)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
struct NCR5380_cmd *ncmd;
struct scsi_cmnd *cmd;
if (list_empty(&hostdata->autosense)) {
list_for_each_entry(ncmd, &hostdata->unissued, list) {
cmd = NCR5380_to_scmd(ncmd);
dsprintk(NDEBUG_QUEUES, instance, "dequeue: cmd=%p target=%d busy=0x%02x lun=%llu\n",
cmd, scmd_id(cmd), hostdata->busy[scmd_id(cmd)], cmd->device->lun);
if (!(hostdata->busy[scmd_id(cmd)] & (1 << cmd->device->lun))) {
list_del(&ncmd->list);
dsprintk(NDEBUG_QUEUES, instance,
"dequeue: removed %p from issue queue\n", cmd);
return cmd;
}
}
} else {
/* Autosense processing begins here */
ncmd = list_first_entry(&hostdata->autosense,
struct NCR5380_cmd, list);
list_del(&ncmd->list);
cmd = NCR5380_to_scmd(ncmd);
dsprintk(NDEBUG_QUEUES, instance,
"dequeue: removed %p from autosense queue\n", cmd);
scsi_eh_prep_cmnd(cmd, &hostdata->ses, NULL, 0, ~0);
hostdata->sensing = cmd;
return cmd;
}
return NULL;
}
static void requeue_cmd(struct Scsi_Host *instance, struct scsi_cmnd *cmd)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
struct NCR5380_cmd *ncmd = scsi_cmd_priv(cmd);
if (hostdata->sensing) {
scsi_eh_restore_cmnd(cmd, &hostdata->ses);
list_add(&ncmd->list, &hostdata->autosense);
hostdata->sensing = NULL;
} else
list_add(&ncmd->list, &hostdata->unissued);
}
/**
* NCR5380_main - NCR state machines
*
* NCR5380_main is a coroutine that runs as long as more work can
* be done on the NCR5380 host adapters in a system. Both
* NCR5380_queue_command() and NCR5380_intr() will try to start it
* in case it is not running.
*
* Locks: called as its own thread with no locks held. Takes the
* host lock and called routines may take the isa dma lock.
*/
static void NCR5380_main(struct work_struct *work)
{
struct NCR5380_hostdata *hostdata =
container_of(work, struct NCR5380_hostdata, main_task);
struct Scsi_Host *instance = hostdata->host;
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
struct scsi_cmnd *cmd;
int done;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
do {
done = 1;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
while (!hostdata->connected &&
(cmd = dequeue_next_cmd(instance))) {
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
dsprintk(NDEBUG_MAIN, instance, "main: dequeued %p\n", cmd);
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
/*
* Attempt to establish an I_T_L nexus here.
* On success, instance->hostdata->connected is set.
* On failure, we must add the command back to the
* issue queue so we can keep trying.
*/
/*
* REQUEST SENSE commands are issued without tagged
* queueing, even on SCSI-II devices because the
* contingent allegiance condition exists for the
* entire unit.
*/
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
if (!NCR5380_select(instance, cmd)) {
dsprintk(NDEBUG_MAIN, instance, "main: selected target %d for command %p\n",
scmd_id(cmd), cmd);
} else {
dsprintk(NDEBUG_MAIN | NDEBUG_QUEUES, instance,
"main: select failed, returning %p to queue\n", cmd);
requeue_cmd(instance, cmd);
}
}
if (hostdata->connected
#ifdef REAL_DMA
&& !hostdata->dmalen
#endif
) {
dprintk(NDEBUG_MAIN, "scsi%d : main() : performing information transfer\n", instance->host_no);
NCR5380_information_transfer(instance);
dprintk(NDEBUG_MAIN, "scsi%d : main() : done set false\n", instance->host_no);
done = 0;
}
} while (!done);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irq(&hostdata->lock);
}
#ifndef DONT_USE_INTR
/**
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
* NCR5380_intr - generic NCR5380 irq handler
* @irq: interrupt number
* @dev_id: device info
*
* Handle interrupts, reestablishing I_T_L or I_T_L_Q nexuses
* from the disconnected queue, and restarting NCR5380_main()
* as required.
*
* The chip can assert IRQ in any of six different conditions. The IRQ flag
* is then cleared by reading the Reset Parity/Interrupt Register (RPIR).
* Three of these six conditions are latched in the Bus and Status Register:
* - End of DMA (cleared by ending DMA Mode)
* - Parity error (cleared by reading RPIR)
* - Loss of BSY (cleared by reading RPIR)
* Two conditions have flag bits that are not latched:
* - Bus phase mismatch (non-maskable in DMA Mode, cleared by ending DMA Mode)
* - Bus reset (non-maskable)
* The remaining condition has no flag bit at all:
* - Selection/reselection
*
* Hence, establishing the cause(s) of any interrupt is partly guesswork.
* In "The DP8490 and DP5380 Comparison Guide", National Semiconductor
* claimed that "the design of the [DP8490] interrupt logic ensures
* interrupts will not be lost (they can be on the DP5380)."
* The L5380/53C80 datasheet from LOGIC Devices has more details.
*
* Checking for bus reset by reading RST is futile because of interrupt
* latency, but a bus reset will reset chip logic. Checking for parity error
* is unnecessary because that interrupt is never enabled. A Loss of BSY
* condition will clear DMA Mode. We can tell when this occurs because the
* the Busy Monitor interrupt is enabled together with DMA Mode.
*/
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
static irqreturn_t NCR5380_intr(int irq, void *dev_id)
{
struct Scsi_Host *instance = dev_id;
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
struct NCR5380_hostdata *hostdata = shost_priv(instance);
int handled = 0;
unsigned char basr;
unsigned long flags;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irqsave(&hostdata->lock, flags);
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
basr = NCR5380_read(BUS_AND_STATUS_REG);
if (basr & BASR_IRQ) {
unsigned char mr = NCR5380_read(MODE_REG);
unsigned char sr = NCR5380_read(STATUS_REG);
dprintk(NDEBUG_INTR, "scsi%d: IRQ %d, BASR 0x%02x, SR 0x%02x, MR 0x%02x\n",
instance->host_no, irq, basr, sr, mr);
#if defined(REAL_DMA)
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
if ((mr & MR_DMA_MODE) || (mr & MR_MONITOR_BSY)) {
/* Probably End of DMA, Phase Mismatch or Loss of BSY.
* We ack IRQ after clearing Mode Register. Workarounds
* for End of DMA errata need to happen in DMA Mode.
*/
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
dprintk(NDEBUG_INTR, "scsi%d: interrupt in DMA mode\n", intance->host_no);
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
int transferred;
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
if (!hostdata->connected)
panic("scsi%d : DMA interrupt with no connected cmd\n",
instance->hostno);
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
transferred = hostdata->dmalen - NCR5380_dma_residual(instance);
hostdata->connected->SCp.this_residual -= transferred;
hostdata->connected->SCp.ptr += transferred;
hostdata->dmalen = 0;
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
/* FIXME: we need to poll briefly then defer a workqueue task ! */
NCR5380_poll_politely(hostdata, BUS_AND_STATUS_REG, BASR_ACK, 0, 2 * HZ);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
NCR5380_write(MODE_REG, MR_BASE);
NCR5380_read(RESET_PARITY_INTERRUPT_REG);
} else
#endif /* REAL_DMA */
if ((NCR5380_read(CURRENT_SCSI_DATA_REG) & hostdata->id_mask) &&
(sr & (SR_SEL | SR_IO | SR_BSY | SR_RST)) == (SR_SEL | SR_IO)) {
/* Probably reselected */
NCR5380_write(SELECT_ENABLE_REG, 0);
NCR5380_read(RESET_PARITY_INTERRUPT_REG);
dprintk(NDEBUG_INTR, "scsi%d: interrupt with SEL and IO\n",
instance->host_no);
if (!hostdata->connected) {
NCR5380_reselect(instance);
queue_work(hostdata->work_q, &hostdata->main_task);
}
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
if (!hostdata->connected)
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
} else {
/* Probably Bus Reset */
NCR5380_read(RESET_PARITY_INTERRUPT_REG);
dprintk(NDEBUG_INTR, "scsi%d: unknown interrupt\n", instance->host_no);
}
handled = 1;
} else {
shost_printk(KERN_NOTICE, instance, "interrupt without IRQ bit\n");
}
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irqrestore(&hostdata->lock, flags);
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
return IRQ_RETVAL(handled);
}
#endif
/*
* Function : int NCR5380_select(struct Scsi_Host *instance,
* struct scsi_cmnd *cmd)
*
* Purpose : establishes I_T_L or I_T_L_Q nexus for new or existing command,
* including ARBITRATION, SELECTION, and initial message out for
* IDENTIFY and queue messages.
*
* Inputs : instance - instantiation of the 5380 driver on which this
* target lives, cmd - SCSI command to execute.
*
* Returns : -1 if selection failed but should be retried.
* 0 if selection failed and should not be retried.
* 0 if selection succeeded completely (hostdata->connected == cmd).
*
* Side effects :
* If bus busy, arbitration failed, etc, NCR5380_select() will exit
* with registers as they should have been on entry - ie
* SELECT_ENABLE will be set appropriately, the NCR5380
* will cease to drive any SCSI bus signals.
*
* If successful : I_T_L or I_T_L_Q nexus will be established,
* instance->connected will be set to cmd.
* SELECT interrupt will be disabled.
*
* If failed (no target) : cmd->scsi_done() will be called, and the
* cmd->result host byte set to DID_BAD_TARGET.
*
* Locks: caller holds hostdata lock in IRQ mode
*/
static int NCR5380_select(struct Scsi_Host *instance, struct scsi_cmnd *cmd)
{
struct NCR5380_hostdata *hostdata = shost_priv(instance);
unsigned char tmp[3], phase;
unsigned char *data;
int len;
int err;
NCR5380_dprint(NDEBUG_ARBITRATION, instance);
dprintk(NDEBUG_ARBITRATION, "scsi%d : starting arbitration, id = %d\n", instance->host_no, instance->this_id);
/*
* Set the phase bits to 0, otherwise the NCR5380 won't drive the
* data bus during SELECTION.
*/
NCR5380_write(TARGET_COMMAND_REG, 0);
/*
* Start arbitration.
*/
NCR5380_write(OUTPUT_DATA_REG, hostdata->id_mask);
NCR5380_write(MODE_REG, MR_ARBITRATE);
/* The chip now waits for BUS FREE phase. Then after the 800 ns
* Bus Free Delay, arbitration will begin.
*/
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irq(&hostdata->lock);
err = NCR5380_poll_politely2(instance, MODE_REG, MR_ARBITRATE, 0,
INITIATOR_COMMAND_REG, ICR_ARBITRATION_PROGRESS,
ICR_ARBITRATION_PROGRESS, HZ);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
if (!(NCR5380_read(MODE_REG) & MR_ARBITRATE)) {
/* Reselection interrupt */
return -1;
}
if (err < 0) {
NCR5380_write(MODE_REG, MR_BASE);
shost_printk(KERN_ERR, instance,
"select: arbitration timeout\n");
return -1;
}
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irq(&hostdata->lock);
/* The SCSI-2 arbitration delay is 2.4 us */
udelay(3);
/* Check for lost arbitration */
if ((NCR5380_read(INITIATOR_COMMAND_REG) & ICR_ARBITRATION_LOST) || (NCR5380_read(CURRENT_SCSI_DATA_REG) & hostdata->id_higher_mask) || (NCR5380_read(INITIATOR_COMMAND_REG) & ICR_ARBITRATION_LOST)) {
NCR5380_write(MODE_REG, MR_BASE);
dprintk(NDEBUG_ARBITRATION, "scsi%d : lost arbitration, deasserting MR_ARBITRATE\n", instance->host_no);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
return -1;
}
/* After/during arbitration, BSY should be asserted.
* IBM DPES-31080 Version S31Q works now
* Tnx to Thomas_Roesch@m2.maus.de for finding this! (Roman)
*/
NCR5380_write(INITIATOR_COMMAND_REG,
ICR_BASE | ICR_ASSERT_SEL | ICR_ASSERT_BSY);
/*
* Again, bus clear + bus settle time is 1.2us, however, this is
* a minimum so we'll udelay ceil(1.2)
*/
if (hostdata->flags & FLAG_TOSHIBA_DELAY)
udelay(15);
else
udelay(2);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
/* NCR5380_reselect() clears MODE_REG after a reselection interrupt */
if (!(NCR5380_read(MODE_REG) & MR_ARBITRATE))
return -1;
dprintk(NDEBUG_ARBITRATION, "scsi%d : won arbitration\n", instance->host_no);
/*
* Now that we have won arbitration, start Selection process, asserting
* the host and target ID's on the SCSI bus.
*/
NCR5380_write(OUTPUT_DATA_REG, (hostdata->id_mask | (1 << scmd_id(cmd))));
/*
* Raise ATN while SEL is true before BSY goes false from arbitration,
* since this is the only way to guarantee that we'll get a MESSAGE OUT
* phase immediately after selection.
*/
NCR5380_write(INITIATOR_COMMAND_REG, (ICR_BASE | ICR_ASSERT_BSY | ICR_ASSERT_DATA | ICR_ASSERT_ATN | ICR_ASSERT_SEL));
NCR5380_write(MODE_REG, MR_BASE);
/*
* Reselect interrupts must be turned off prior to the dropping of BSY,
* otherwise we will trigger an interrupt.
*/
NCR5380_write(SELECT_ENABLE_REG, 0);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irq(&hostdata->lock);
/*
* The initiator shall then wait at least two deskew delays and release
* the BSY signal.
*/
udelay(1); /* wingel -- wait two bus deskew delay >2*45ns */
/* Reset BSY */
NCR5380_write(INITIATOR_COMMAND_REG, (ICR_BASE | ICR_ASSERT_DATA | ICR_ASSERT_ATN | ICR_ASSERT_SEL));
/*
* Something weird happens when we cease to drive BSY - looks
* like the board/chip is letting us do another read before the
* appropriate propagation delay has expired, and we're confusing
* a BSY signal from ourselves as the target's response to SELECTION.
*
* A small delay (the 'C++' frontend breaks the pipeline with an
* unnecessary jump, making it work on my 386-33/Trantor T128, the
* tighter 'C' code breaks and requires this) solves the problem -
* the 1 us delay is arbitrary, and only used because this delay will
* be the same on other platforms and since it works here, it should
* work there.
*
* wingel suggests that this could be due to failing to wait
* one deskew delay.
*/
udelay(1);
dprintk(NDEBUG_SELECTION, "scsi%d : selecting target %d\n", instance->host_no, scmd_id(cmd));
/*
* The SCSI specification calls for a 250 ms timeout for the actual
* selection.
*/
err = NCR5380_poll_politely(instance, STATUS_REG, SR_BSY, SR_BSY,
msecs_to_jiffies(250));
if ((NCR5380_read(STATUS_REG) & (SR_SEL | SR_IO)) == (SR_SEL | SR_IO)) {
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
NCR5380_reselect(instance);
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
if (!hostdata->connected)
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
printk("scsi%d : reselection after won arbitration?\n", instance->host_no);
return -1;
}
if (err < 0) {
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
cmd->result = DID_BAD_TARGET << 16;
complete_cmd(instance, cmd);
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
dprintk(NDEBUG_SELECTION, "scsi%d : target did not respond within 250ms\n",
instance->host_no);
return 0;
}
/*
* No less than two deskew delays after the initiator detects the
* BSY signal is true, it shall release the SEL signal and may
* change the DATA BUS. -wingel
*/
udelay(1);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN);
/*
* Since we followed the SCSI spec, and raised ATN while SEL
* was true but before BSY was false during selection, the information
* transfer phase should be a MESSAGE OUT phase so that we can send the
* IDENTIFY message.
*
* If SCSI-II tagged queuing is enabled, we also send a SIMPLE_QUEUE_TAG
* message (2 bytes) with a tag ID that we increment with every command
* until it wraps back to 0.
*
* XXX - it turns out that there are some broken SCSI-II devices,
* which claim to support tagged queuing but fail when more than
* some number of commands are issued at once.
*/
/* Wait for start of REQ/ACK handshake */
err = NCR5380_poll_politely(instance, STATUS_REG, SR_REQ, SR_REQ, HZ);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
if (err < 0) {
shost_printk(KERN_ERR, instance, "select: REQ timeout\n");
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
return -1;
}
dprintk(NDEBUG_SELECTION, "scsi%d : target %d selected, going into MESSAGE OUT phase.\n", instance->host_no, cmd->device->id);
tmp[0] = IDENTIFY(((instance->irq == NO_IRQ) ? 0 : 1), cmd->device->lun);
len = 1;
cmd->tag = 0;
/* Send message(s) */
data = tmp;
phase = PHASE_MSGOUT;
NCR5380_transfer_pio(instance, &phase, &len, &data);
dprintk(NDEBUG_SELECTION, "scsi%d : nexus established.\n", instance->host_no);
/* XXX need to handle errors here */
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
hostdata->connected = cmd;
hostdata->busy[cmd->device->id] |= (1 << (cmd->device->lun & 0xFF));
initialize_SCp(cmd);
return 0;
}
/*
* Function : int NCR5380_transfer_pio (struct Scsi_Host *instance,
* unsigned char *phase, int *count, unsigned char **data)
*
* Purpose : transfers data in given phase using polled I/O
*
* Inputs : instance - instance of driver, *phase - pointer to
* what phase is expected, *count - pointer to number of
* bytes to transfer, **data - pointer to data pointer.
*
* Returns : -1 when different phase is entered without transferring
* maximum number of bytes, 0 if all bytes or transferred or exit
* is in same phase.
*
* Also, *phase, *count, *data are modified in place.
*
* XXX Note : handling for bus free may be useful.
*/
/*
* Note : this code is not as quick as it could be, however it
* IS 100% reliable, and for the actual data transfer where speed
* counts, we will always do a pseudo DMA or DMA transfer.
*/
static int NCR5380_transfer_pio(struct Scsi_Host *instance, unsigned char *phase, int *count, unsigned char **data) {
unsigned char p = *phase, tmp;
int c = *count;
unsigned char *d = *data;
if (!(p & SR_IO))
dprintk(NDEBUG_PIO, "scsi%d : pio write %d bytes\n", instance->host_no, c);
else
dprintk(NDEBUG_PIO, "scsi%d : pio read %d bytes\n", instance->host_no, c);
/*
* The NCR5380 chip will only drive the SCSI bus when the
* phase specified in the appropriate bits of the TARGET COMMAND
* REGISTER match the STATUS REGISTER
*/
NCR5380_write(TARGET_COMMAND_REG, PHASE_SR_TO_TCR(p));
do {
/*
* Wait for assertion of REQ, after which the phase bits will be
* valid
*/
ncr5380: Rework disconnect versus poll logic The atari_NCR5380.c and NCR5380.c core drivers differ in their handling of target disconnection. This is partly because atari_NCR5380.c had all of the polling and sleeping removed to become entirely interrupt-driven, and it is partly because of damage done to NCR5380.c after atari_NCR5380.c was forked. See commit 37cd23b44929 ("Linux 2.1.105") in history/history.git. The polling changes that were made in v2.1.105 are questionable at best: if REQ is not already asserted when NCR5380_transfer_pio() is invoked, and if the expected phase is DATA IN or DATA OUT, the function will schedule main() to execute after USLEEP_SLEEP jiffies and then return. The problems here are the expected REQ timing and the sleep interval*. Avoid this issue by using NCR5380_poll_politely() instead of scheduling main(). The atari_NCR5380.c core driver requires the use of the chip interrupt and always permits target disconnection. It sets the cmd->device->disconnect flag when a device disconnects, but never tests this flag. The NCR5380.c core driver permits disconnection only when instance->irq != NO_IRQ. It sets the cmd->device->disconnect flag when a device disconnects and it tests this flag in a couple of places: 1. During NCR5380_information_transfer(), following COMMAND OUT phase, if !cmd->device->disconnect, the initiator will take a guess as to whether or not the target will then choose to go to MESSAGE IN phase and disconnect. If the driver guesses "yes", it will schedule main() to execute after USLEEP_SLEEP jiffies and then return there. Unfortunately the driver may guess "yes" even after it has denied the target the disconnection privilege. When the target does not disconnect, the sleep can be beneficial, assuming the sleep interval is appropriate (mostly it is not*). And even if the driver guesses "yes" correctly, and the target would then disconnect, the driver still has to go through the MESSAGE IN phase in order to get to BUS FREE phase. The main loop can do nothing useful until BUS FREE, and sleeping just delays the phase transition. 2. If !cmd->device->disconnect and REQ is not already asserted when NCR5380_information_transfer() is invoked, the function polls for REQ for USLEEP_POLL jiffies. If REQ is not asserted, it then schedules main() to execute after USLEEP_SLEEP jiffies and returns. The idea is apparently to yeild the CPU while waiting for REQ. This is conditional upon !cmd->device->disconnect, but there seems to be no rhyme or reason for that. For example, the flag may be unset because disconnection privilege was denied because the driver has no IRQ. Or the flag may be unset because the device has never needed to disconnect before. Or if the flag is set, disconnection may have no relevance to the present bus phase. Another deficiency of the existing algorithm is as follows. When the driver has no IRQ, it prevents disconnection, and generally polls and sleeps more than it would normally. Now, if the driver is going to poll anyway, why not allow the target to disconnect? That way the driver can do something useful with the bus instead of polling unproductively! Avoid this pointless latency, complexity and guesswork by using NCR5380_poll_politely() instead of scheduling main(). * For g_NCR5380, the time intervals for USLEEP_SLEEP and USLEEP_POLL are 200 ms and 10 ms, respectively. They are 20 ms and 200 ms respectively for the other NCR5380 drivers. There doesn't seem to be any reason for this discrepancy. The timing seems to have no relation to the type of adapter. Bizarrely, the timing in g_NCR5380 seems to relate only to one particular type of target device. This patch attempts to solve the problem for all NCR5380 drivers and all target devices. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:26 -07:00
if (NCR5380_poll_politely(instance, STATUS_REG, SR_REQ, SR_REQ, HZ) < 0)
break;
dprintk(NDEBUG_HANDSHAKE, "scsi%d : REQ detected\n", instance->host_no);
/* Check for phase mismatch */
ncr5380: Rework disconnect versus poll logic The atari_NCR5380.c and NCR5380.c core drivers differ in their handling of target disconnection. This is partly because atari_NCR5380.c had all of the polling and sleeping removed to become entirely interrupt-driven, and it is partly because of damage done to NCR5380.c after atari_NCR5380.c was forked. See commit 37cd23b44929 ("Linux 2.1.105") in history/history.git. The polling changes that were made in v2.1.105 are questionable at best: if REQ is not already asserted when NCR5380_transfer_pio() is invoked, and if the expected phase is DATA IN or DATA OUT, the function will schedule main() to execute after USLEEP_SLEEP jiffies and then return. The problems here are the expected REQ timing and the sleep interval*. Avoid this issue by using NCR5380_poll_politely() instead of scheduling main(). The atari_NCR5380.c core driver requires the use of the chip interrupt and always permits target disconnection. It sets the cmd->device->disconnect flag when a device disconnects, but never tests this flag. The NCR5380.c core driver permits disconnection only when instance->irq != NO_IRQ. It sets the cmd->device->disconnect flag when a device disconnects and it tests this flag in a couple of places: 1. During NCR5380_information_transfer(), following COMMAND OUT phase, if !cmd->device->disconnect, the initiator will take a guess as to whether or not the target will then choose to go to MESSAGE IN phase and disconnect. If the driver guesses "yes", it will schedule main() to execute after USLEEP_SLEEP jiffies and then return there. Unfortunately the driver may guess "yes" even after it has denied the target the disconnection privilege. When the target does not disconnect, the sleep can be beneficial, assuming the sleep interval is appropriate (mostly it is not*). And even if the driver guesses "yes" correctly, and the target would then disconnect, the driver still has to go through the MESSAGE IN phase in order to get to BUS FREE phase. The main loop can do nothing useful until BUS FREE, and sleeping just delays the phase transition. 2. If !cmd->device->disconnect and REQ is not already asserted when NCR5380_information_transfer() is invoked, the function polls for REQ for USLEEP_POLL jiffies. If REQ is not asserted, it then schedules main() to execute after USLEEP_SLEEP jiffies and returns. The idea is apparently to yeild the CPU while waiting for REQ. This is conditional upon !cmd->device->disconnect, but there seems to be no rhyme or reason for that. For example, the flag may be unset because disconnection privilege was denied because the driver has no IRQ. Or the flag may be unset because the device has never needed to disconnect before. Or if the flag is set, disconnection may have no relevance to the present bus phase. Another deficiency of the existing algorithm is as follows. When the driver has no IRQ, it prevents disconnection, and generally polls and sleeps more than it would normally. Now, if the driver is going to poll anyway, why not allow the target to disconnect? That way the driver can do something useful with the bus instead of polling unproductively! Avoid this pointless latency, complexity and guesswork by using NCR5380_poll_politely() instead of scheduling main(). * For g_NCR5380, the time intervals for USLEEP_SLEEP and USLEEP_POLL are 200 ms and 10 ms, respectively. They are 20 ms and 200 ms respectively for the other NCR5380 drivers. There doesn't seem to be any reason for this discrepancy. The timing seems to have no relation to the type of adapter. Bizarrely, the timing in g_NCR5380 seems to relate only to one particular type of target device. This patch attempts to solve the problem for all NCR5380 drivers and all target devices. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:26 -07:00
if ((NCR5380_read(STATUS_REG) & PHASE_MASK) != p) {
dprintk(NDEBUG_HANDSHAKE, "scsi%d : phase mismatch\n", instance->host_no);
NCR5380_dprint_phase(NDEBUG_HANDSHAKE, instance);
break;
}
/* Do actual transfer from SCSI bus to / from memory */
if (!(p & SR_IO))
NCR5380_write(OUTPUT_DATA_REG, *d);
else
*d = NCR5380_read(CURRENT_SCSI_DATA_REG);
++d;
/*
* The SCSI standard suggests that in MSGOUT phase, the initiator
* should drop ATN on the last byte of the message phase
* after REQ has been asserted for the handshake but before
* the initiator raises ACK.
*/
if (!(p & SR_IO)) {
if (!((p & SR_MSG) && c > 1)) {
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_DATA);
NCR5380_dprint(NDEBUG_PIO, instance);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_DATA | ICR_ASSERT_ACK);
} else {
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_DATA | ICR_ASSERT_ATN);
NCR5380_dprint(NDEBUG_PIO, instance);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_DATA | ICR_ASSERT_ATN | ICR_ASSERT_ACK);
}
} else {
NCR5380_dprint(NDEBUG_PIO, instance);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ACK);
}
if (NCR5380_poll_politely(instance,
STATUS_REG, SR_REQ, 0, 5 * HZ) < 0)
break;
dprintk(NDEBUG_HANDSHAKE, "scsi%d : req false, handshake complete\n", instance->host_no);
/*
* We have several special cases to consider during REQ/ACK handshaking :
* 1. We were in MSGOUT phase, and we are on the last byte of the
* message. ATN must be dropped as ACK is dropped.
*
* 2. We are in a MSGIN phase, and we are on the last byte of the
* message. We must exit with ACK asserted, so that the calling
* code may raise ATN before dropping ACK to reject the message.
*
* 3. ACK and ATN are clear and the target may proceed as normal.
*/
if (!(p == PHASE_MSGIN && c == 1)) {
if (p == PHASE_MSGOUT && c > 1)
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN);
else
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
}
} while (--c);
dprintk(NDEBUG_PIO, "scsi%d : residual %d\n", instance->host_no, c);
*count = c;
*data = d;
tmp = NCR5380_read(STATUS_REG);
/* The phase read from the bus is valid if either REQ is (already)
* asserted or if ACK hasn't been released yet. The latter applies if
* we're in MSG IN, DATA IN or STATUS and all bytes have been received.
*/
if ((tmp & SR_REQ) || ((tmp & SR_IO) && c == 0))
*phase = tmp & PHASE_MASK;
else
*phase = PHASE_UNKNOWN;
if (!c || (*phase == p))
return 0;
else
return -1;
}
/**
* do_reset - issue a reset command
* @instance: adapter to reset
*
* Issue a reset sequence to the NCR5380 and try and get the bus
* back into sane shape.
*
* This clears the reset interrupt flag because there may be no handler for
* it. When the driver is initialized, the NCR5380_intr() handler has not yet
* been installed. And when in EH we may have released the ST DMA interrupt.
*/
static void do_reset(struct Scsi_Host *instance)
{
unsigned long flags;
local_irq_save(flags);
NCR5380_write(TARGET_COMMAND_REG,
PHASE_SR_TO_TCR(NCR5380_read(STATUS_REG) & PHASE_MASK));
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_RST);
udelay(50);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
(void)NCR5380_read(RESET_PARITY_INTERRUPT_REG);
local_irq_restore(flags);
}
/**
* do_abort - abort the currently established nexus by going to
* MESSAGE OUT phase and sending an ABORT message.
* @instance: relevant scsi host instance
*
* Returns 0 on success, -1 on failure.
*/
static int do_abort(struct Scsi_Host *instance)
{
unsigned char *msgptr, phase, tmp;
int len;
int rc;
/* Request message out phase */
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN);
/*
* Wait for the target to indicate a valid phase by asserting
* REQ. Once this happens, we'll have either a MSGOUT phase
* and can immediately send the ABORT message, or we'll have some
* other phase and will have to source/sink data.
*
* We really don't care what value was on the bus or what value
* the target sees, so we just handshake.
*/
rc = NCR5380_poll_politely(instance, STATUS_REG, SR_REQ, SR_REQ, 10 * HZ);
if (rc < 0)
goto timeout;
tmp = NCR5380_read(STATUS_REG) & PHASE_MASK;
NCR5380_write(TARGET_COMMAND_REG, PHASE_SR_TO_TCR(tmp));
if (tmp != PHASE_MSGOUT) {
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN | ICR_ASSERT_ACK);
rc = NCR5380_poll_politely(instance, STATUS_REG, SR_REQ, 0, 3 * HZ);
if (rc < 0)
goto timeout;
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN);
}
tmp = ABORT;
msgptr = &tmp;
len = 1;
phase = PHASE_MSGOUT;
NCR5380_transfer_pio(instance, &phase, &len, &msgptr);
/*
* If we got here, and the command completed successfully,
* we're about to go into bus free state.
*/
return len ? -1 : 0;
timeout:
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
return -1;
}
#if defined(REAL_DMA) || defined(PSEUDO_DMA) || defined (REAL_DMA_POLL)
/*
* Function : int NCR5380_transfer_dma (struct Scsi_Host *instance,
* unsigned char *phase, int *count, unsigned char **data)
*
* Purpose : transfers data in given phase using either real
* or pseudo DMA.
*
* Inputs : instance - instance of driver, *phase - pointer to
* what phase is expected, *count - pointer to number of
* bytes to transfer, **data - pointer to data pointer.
*
* Returns : -1 when different phase is entered without transferring
* maximum number of bytes, 0 if all bytes or transferred or exit
* is in same phase.
*
* Also, *phase, *count, *data are modified in place.
*
* Locks: io_request lock held by caller
*/
static int NCR5380_transfer_dma(struct Scsi_Host *instance, unsigned char *phase, int *count, unsigned char **data) {
register int c = *count;
register unsigned char p = *phase;
register unsigned char *d = *data;
unsigned char tmp;
int foo;
#if defined(REAL_DMA_POLL)
int cnt, toPIO;
unsigned char saved_data = 0, overrun = 0, residue;
#endif
struct NCR5380_hostdata *hostdata = shost_priv(instance);
if ((tmp = (NCR5380_read(STATUS_REG) & PHASE_MASK)) != p) {
*phase = tmp;
return -1;
}
#if defined(REAL_DMA) || defined(REAL_DMA_POLL)
if (p & SR_IO) {
if (!(hostdata->flags & FLAG_NO_DMA_FIXUPS))
c -= 2;
}
dprintk(NDEBUG_DMA, "scsi%d : initializing DMA channel %d for %s, %d bytes %s %0x\n", instance->host_no, instance->dma_channel, (p & SR_IO) ? "reading" : "writing", c, (p & SR_IO) ? "to" : "from", (unsigned) d);
hostdata->dma_len = (p & SR_IO) ? NCR5380_dma_read_setup(instance, d, c) : NCR5380_dma_write_setup(instance, d, c);
#endif
NCR5380_write(TARGET_COMMAND_REG, PHASE_SR_TO_TCR(p));
#ifdef REAL_DMA
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
NCR5380_write(MODE_REG, MR_BASE | MR_DMA_MODE | MR_MONITOR_BSY |
MR_ENABLE_EOP_INTR);
#elif defined(REAL_DMA_POLL)
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
NCR5380_write(MODE_REG, MR_BASE | MR_DMA_MODE | MR_MONITOR_BSY);
#else
/*
* Note : on my sample board, watch-dog timeouts occurred when interrupts
* were not disabled for the duration of a single DMA transfer, from
* before the setting of DMA mode to after transfer of the last byte.
*/
if (hostdata->flags & FLAG_NO_DMA_FIXUP)
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
NCR5380_write(MODE_REG, MR_BASE | MR_DMA_MODE | MR_MONITOR_BSY |
MR_ENABLE_EOP_INTR);
else
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
NCR5380_write(MODE_REG, MR_BASE | MR_DMA_MODE | MR_MONITOR_BSY);
#endif /* def REAL_DMA */
dprintk(NDEBUG_DMA, "scsi%d : mode reg = 0x%X\n", instance->host_no, NCR5380_read(MODE_REG));
/*
* On the PAS16 at least I/O recovery delays are not needed here.
* Everyone else seems to want them.
*/
if (p & SR_IO) {
io_recovery_delay(1);
NCR5380_write(START_DMA_INITIATOR_RECEIVE_REG, 0);
} else {
io_recovery_delay(1);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_DATA);
io_recovery_delay(1);
NCR5380_write(START_DMA_SEND_REG, 0);
io_recovery_delay(1);
}
#if defined(REAL_DMA_POLL)
do {
tmp = NCR5380_read(BUS_AND_STATUS_REG);
} while ((tmp & BASR_PHASE_MATCH) && !(tmp & (BASR_BUSY_ERROR | BASR_END_DMA_TRANSFER)));
/*
At this point, either we've completed DMA, or we have a phase mismatch,
or we've unexpectedly lost BUSY (which is a real error).
For write DMAs, we want to wait until the last byte has been
transferred out over the bus before we turn off DMA mode. Alas, there
seems to be no terribly good way of doing this on a 5380 under all
conditions. For non-scatter-gather operations, we can wait until REQ
and ACK both go false, or until a phase mismatch occurs. Gather-writes
are nastier, since the device will be expecting more data than we
are prepared to send it, and REQ will remain asserted. On a 53C8[01] we
could test LAST BIT SENT to assure transfer (I imagine this is precisely
why this signal was added to the newer chips) but on the older 538[01]
this signal does not exist. The workaround for this lack is a watchdog;
we bail out of the wait-loop after a modest amount of wait-time if
the usual exit conditions are not met. Not a terribly clean or
correct solution :-%
Reads are equally tricky due to a nasty characteristic of the NCR5380.
If the chip is in DMA mode for an READ, it will respond to a target's
REQ by latching the SCSI data into the INPUT DATA register and asserting
ACK, even if it has _already_ been notified by the DMA controller that
the current DMA transfer has completed! If the NCR5380 is then taken
out of DMA mode, this already-acknowledged byte is lost.
This is not a problem for "one DMA transfer per command" reads, because
the situation will never arise... either all of the data is DMA'ed
properly, or the target switches to MESSAGE IN phase to signal a
disconnection (either operation bringing the DMA to a clean halt).
However, in order to handle scatter-reads, we must work around the
problem. The chosen fix is to DMA N-2 bytes, then check for the
condition before taking the NCR5380 out of DMA mode. One or two extra
bytes are transferred via PIO as necessary to fill out the original
request.
*/
if (p & SR_IO) {
if (!(hostdata->flags & FLAG_NO_DMA_FIXUPS)) {
udelay(10);
if ((NCR5380_read(BUS_AND_STATUS_REG) & (BASR_PHASE_MATCH | BASR_ACK)) ==
(BASR_PHASE_MATCH | BASR_ACK)) {
saved_data = NCR5380_read(INPUT_DATA_REGISTER);
overrun = 1;
}
}
} else {
int limit = 100;
while (((tmp = NCR5380_read(BUS_AND_STATUS_REG)) & BASR_ACK) || (NCR5380_read(STATUS_REG) & SR_REQ)) {
if (!(tmp & BASR_PHASE_MATCH))
break;
if (--limit < 0)
break;
}
}
dprintk(NDEBUG_DMA, "scsi%d : polled DMA transfer complete, basr 0x%X, sr 0x%X\n", instance->host_no, tmp, NCR5380_read(STATUS_REG));
NCR5380_write(MODE_REG, MR_BASE);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
residue = NCR5380_dma_residual(instance);
c -= residue;
*count -= c;
*data += c;
*phase = NCR5380_read(STATUS_REG) & PHASE_MASK;
if (!(hostdata->flags & FLAG_NO_DMA_FIXUPS) &&
*phase == p && (p & SR_IO) && residue == 0) {
if (overrun) {
dprintk(NDEBUG_DMA, "Got an input overrun, using saved byte\n");
**data = saved_data;
*data += 1;
*count -= 1;
cnt = toPIO = 1;
} else {
printk("No overrun??\n");
cnt = toPIO = 2;
}
dprintk(NDEBUG_DMA, "Doing %d-byte PIO to 0x%X\n", cnt, *data);
NCR5380_transfer_pio(instance, phase, &cnt, data);
*count -= toPIO - cnt;
}
dprintk(NDEBUG_DMA, "Return with data ptr = 0x%X, count %d, last 0x%X, next 0x%X\n", *data, *count, *(*data + *count - 1), *(*data + *count));
return 0;
#elif defined(REAL_DMA)
return 0;
#else /* defined(REAL_DMA_POLL) */
if (p & SR_IO) {
foo = NCR5380_pread(instance, d,
hostdata->flags & FLAG_NO_DMA_FIXUP ? c : c - 1);
if (!foo && !(hostdata->flags & FLAG_NO_DMA_FIXUP)) {
/*
* We can't disable DMA mode after successfully transferring
* what we plan to be the last byte, since that would open up
* a race condition where if the target asserted REQ before
* we got the DMA mode reset, the NCR5380 would have latched
* an additional byte into the INPUT DATA register and we'd
* have dropped it.
*
* The workaround was to transfer one fewer bytes than we
* intended to with the pseudo-DMA read function, wait for
* the chip to latch the last byte, read it, and then disable
* pseudo-DMA mode.
*
* After REQ is asserted, the NCR5380 asserts DRQ and ACK.
* REQ is deasserted when ACK is asserted, and not reasserted
* until ACK goes false. Since the NCR5380 won't lower ACK
* until DACK is asserted, which won't happen unless we twiddle
* the DMA port or we take the NCR5380 out of DMA mode, we
* can guarantee that we won't handshake another extra
* byte.
*/
if (NCR5380_poll_politely(instance, BUS_AND_STATUS_REG,
BASR_DRQ, BASR_DRQ, HZ) < 0) {
foo = -1;
shost_printk(KERN_ERR, instance, "PDMA read: DRQ timeout\n");
}
if (NCR5380_poll_politely(instance, STATUS_REG,
SR_REQ, 0, HZ) < 0) {
foo = -1;
shost_printk(KERN_ERR, instance, "PDMA read: !REQ timeout\n");
}
d[c - 1] = NCR5380_read(INPUT_DATA_REG);
}
} else {
foo = NCR5380_pwrite(instance, d, c);
if (!foo && !(hostdata->flags & FLAG_NO_DMA_FIXUP)) {
/*
* Wait for the last byte to be sent. If REQ is being asserted for
* the byte we're interested, we'll ACK it and it will go false.
*/
if (NCR5380_poll_politely2(instance,
BUS_AND_STATUS_REG, BASR_DRQ, BASR_DRQ,
BUS_AND_STATUS_REG, BASR_PHASE_MATCH, 0, HZ) < 0) {
foo = -1;
shost_printk(KERN_ERR, instance, "PDMA write: DRQ and phase timeout\n");
}
}
}
NCR5380_write(MODE_REG, MR_BASE);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
ncr5380: Standardize interrupt handling Because interrupt handling is crucial to the core driver(s), all wrapper drivers need to agree on this code. This patch removes discrepancies. NCR5380_intr() in NCR5380.c has the following pointless loop that differs from the code in atari_NCR5380.c. done = 1; do { /* ... */ } while (!done); The 'done' flag gets cleared when a reconnected command is to be processed from the work queue. But in NCR5380.c, the flag is also used to cause the interrupt conditions to be re-examined. Perhaps this was because NCR5380_reselect() was expected to cause another interrupt, or perhaps the remaining present interrupt conditions need to be handled after the NCR5380_reselect() call? Actually, both possibilities are bogus, as is the loop itself. It seems have been overlooked in the hit-and-miss removal of scsi host instance list iteration many years ago; see history/history.git commit 491447e1fcff ("[PATCH] next NCR5380 updates") and commit 69e1a9482e57 ("[PATCH] fix up NCR5380 private data"). See also my earlier patch, "Always retry arbitration and selection". The datasheet says, "IRQ can be reset simply by reading the Reset Parity/Interrupt Register". So don't treat the chip IRQ like a level-triggered interrupt. Of the conditions that set the IRQ flag, some are level-triggered and some are edge-triggered, which means IRQ itself must be edge-triggered. Some interrupt conditions are latched and some are not. Before clearing the chip IRQ flag, clear all state that may cause it to be raised. That means clearing the DMA Mode and Busy Monitor bits in the Mode Register and clearing the host ID in the Select Enable register. Also clean up some printk's and some comments. Keep atari_NCR5380.c and NCR5380.c in agreement. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:40 -07:00
NCR5380_read(RESET_PARITY_INTERRUPT_REG);
*data = d + c;
*count = 0;
*phase = NCR5380_read(STATUS_REG) & PHASE_MASK;
return foo;
#endif /* def REAL_DMA */
}
#endif /* defined(REAL_DMA) | defined(PSEUDO_DMA) */
/*
* Function : NCR5380_information_transfer (struct Scsi_Host *instance)
*
* Purpose : run through the various SCSI phases and do as the target
* directs us to. Operates on the currently connected command,
* instance->connected.
*
* Inputs : instance, instance for which we are doing commands
*
* Side effects : SCSI things happen, the disconnected queue will be
* modified if a command disconnects, *instance->connected will
* change.
*
* XXX Note : we need to watch for bus free or a reset condition here
* to recover from an unexpected bus free condition.
*
* Locks: io_request_lock held by caller in IRQ mode
*/
static void NCR5380_information_transfer(struct Scsi_Host *instance) {
struct NCR5380_hostdata *hostdata = shost_priv(instance);
unsigned char msgout = NOP;
int sink = 0;
int len;
#if defined(PSEUDO_DMA) || defined(REAL_DMA_POLL)
int transfersize;
#endif
unsigned char *data;
unsigned char phase, tmp, extended_msg[10], old_phase = 0xff;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
struct scsi_cmnd *cmd;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
while ((cmd = hostdata->connected)) {
struct NCR5380_cmd *ncmd = scsi_cmd_priv(cmd);
tmp = NCR5380_read(STATUS_REG);
/* We only have a valid SCSI phase when REQ is asserted */
if (tmp & SR_REQ) {
phase = (tmp & PHASE_MASK);
if (phase != old_phase) {
old_phase = phase;
NCR5380_dprint_phase(NDEBUG_INFORMATION, instance);
}
if (sink && (phase != PHASE_MSGOUT)) {
NCR5380_write(TARGET_COMMAND_REG, PHASE_SR_TO_TCR(tmp));
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN | ICR_ASSERT_ACK);
while (NCR5380_read(STATUS_REG) & SR_REQ);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN);
sink = 0;
continue;
}
switch (phase) {
case PHASE_DATAOUT:
#if (NDEBUG & NDEBUG_NO_DATAOUT)
printk("scsi%d : NDEBUG_NO_DATAOUT set, attempted DATAOUT aborted\n", instance->host_no);
sink = 1;
do_abort(instance);
cmd->result = DID_ERROR << 16;
complete_cmd(instance, cmd);
return;
#endif
case PHASE_DATAIN:
/*
* If there is no room left in the current buffer in the
* scatter-gather list, move onto the next one.
*/
if (!cmd->SCp.this_residual && cmd->SCp.buffers_residual) {
++cmd->SCp.buffer;
--cmd->SCp.buffers_residual;
cmd->SCp.this_residual = cmd->SCp.buffer->length;
cmd->SCp.ptr = sg_virt(cmd->SCp.buffer);
dprintk(NDEBUG_INFORMATION, "scsi%d : %d bytes and %d buffers left\n", instance->host_no, cmd->SCp.this_residual, cmd->SCp.buffers_residual);
}
/*
* The preferred transfer method is going to be
* PSEUDO-DMA for systems that are strictly PIO,
* since we can let the hardware do the handshaking.
*
* For this to work, we need to know the transfersize
* ahead of time, since the pseudo-DMA code will sit
* in an unconditional loop.
*/
#if defined(PSEUDO_DMA) || defined(REAL_DMA_POLL)
transfersize = 0;
if (!cmd->device->borken &&
!(hostdata->flags & FLAG_NO_PSEUDO_DMA))
transfersize = NCR5380_dma_xfer_len(instance, cmd, phase);
if (transfersize) {
len = transfersize;
if (NCR5380_transfer_dma(instance, &phase, &len, (unsigned char **) &cmd->SCp.ptr)) {
/*
* If the watchdog timer fires, all future accesses to this
* device will use the polled-IO.
*/
scmd_printk(KERN_INFO, cmd,
"switching to slow handshake\n");
cmd->device->borken = 1;
sink = 1;
do_abort(instance);
cmd->result = DID_ERROR << 16;
complete_cmd(instance, cmd);
/* XXX - need to source or sink data here, as appropriate */
} else
cmd->SCp.this_residual -= transfersize - len;
} else
#endif /* defined(PSEUDO_DMA) || defined(REAL_DMA_POLL) */
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
{
spin_unlock_irq(&hostdata->lock);
NCR5380_transfer_pio(instance, &phase, (int *) &cmd->SCp.this_residual, (unsigned char **)
&cmd->SCp.ptr);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
}
break;
case PHASE_MSGIN:
len = 1;
data = &tmp;
NCR5380_transfer_pio(instance, &phase, &len, &data);
cmd->SCp.Message = tmp;
switch (tmp) {
case ABORT:
case COMMAND_COMPLETE:
/* Accept message by clearing ACK */
sink = 1;
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
dsprintk(NDEBUG_QUEUES, instance,
"COMMAND COMPLETE %p target %d lun %llu\n",
cmd, scmd_id(cmd), cmd->device->lun);
hostdata->connected = NULL;
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
cmd->result &= ~0xffff;
cmd->result |= cmd->SCp.Status;
cmd->result |= cmd->SCp.Message << 8;
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
if (cmd->cmnd[0] == REQUEST_SENSE)
complete_cmd(instance, cmd);
ncr5380: Fix autosense bugs NCR5380_information_transfer() may re-queue a command for autosense, after calling scsi_eh_prep_cmnd(). This creates several possibilities: 1. Reselection may intervene before the re-queued command gets processed. If the reconnected command then undergoes autosense, this causes the scsi_eh_save data from the previous command to be overwritten. 2. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), a new REQUEST SENSE command may arrive. This would be queued ahead of any command already undergoing autosense, which means the scsi_eh_save data might be restored to the wrong command. 3. After NCR5380_information_transfer() calls scsi_eh_prep_cmnd(), eh_abort_handler() may abort the command. But the scsi_eh_save data is not discarded, which means the scsi_eh_save data might be incorrectly restored to the next REQUEST SENSE command issued. This patch adds a new autosense list so that commands that are re-queued because of a CHECK CONDITION result can be kept apart from the REQUEST SENSE commands that arrive via queuecommand. This patch also adds a function dedicated to dequeueing and preparing the next command for processing. By refactoring the main loop in this way, scsi_eh_save takes place when an autosense command is dequeued rather than when re-queued. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:06:00 -07:00
else {
if (cmd->SCp.Status == SAM_STAT_CHECK_CONDITION ||
cmd->SCp.Status == SAM_STAT_COMMAND_TERMINATED) {
dsprintk(NDEBUG_QUEUES, instance, "autosense: adding cmd %p to tail of autosense queue\n",
cmd);
list_add_tail(&ncmd->list,
&hostdata->autosense);
} else
complete_cmd(instance, cmd);
}
/*
* Restore phase bits to 0 so an interrupted selection,
* arbitration can resume.
*/
NCR5380_write(TARGET_COMMAND_REG, 0);
/* Enable reselect interrupts */
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
return;
case MESSAGE_REJECT:
/* Accept message by clearing ACK */
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
switch (hostdata->last_message) {
case HEAD_OF_QUEUE_TAG:
case ORDERED_QUEUE_TAG:
case SIMPLE_QUEUE_TAG:
cmd->device->simple_tags = 0;
hostdata->busy[cmd->device->id] |= (1 << (cmd->device->lun & 0xFF));
break;
default:
break;
}
break;
case DISCONNECT:{
/* Accept message by clearing ACK */
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
hostdata->connected = NULL;
list_add(&ncmd->list, &hostdata->disconnected);
dsprintk(NDEBUG_INFORMATION | NDEBUG_QUEUES,
instance, "connected command %p for target %d lun %llu moved to disconnected queue\n",
cmd, scmd_id(cmd), cmd->device->lun);
/*
* Restore phase bits to 0 so an interrupted selection,
* arbitration can resume.
*/
NCR5380_write(TARGET_COMMAND_REG, 0);
/* Enable reselect interrupts */
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
return;
}
/*
* The SCSI data pointer is *IMPLICITLY* saved on a disconnect
* operation, in violation of the SCSI spec so we can safely
* ignore SAVE/RESTORE pointers calls.
*
* Unfortunately, some disks violate the SCSI spec and
* don't issue the required SAVE_POINTERS message before
* disconnecting, and we have to break spec to remain
* compatible.
*/
case SAVE_POINTERS:
case RESTORE_POINTERS:
/* Accept message by clearing ACK */
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
break;
case EXTENDED_MESSAGE:
/*
* Extended messages are sent in the following format :
* Byte
* 0 EXTENDED_MESSAGE == 1
* 1 length (includes one byte for code, doesn't
* include first two bytes)
* 2 code
* 3..length+1 arguments
*
* Start the extended message buffer with the EXTENDED_MESSAGE
* byte, since spi_print_msg() wants the whole thing.
*/
extended_msg[0] = EXTENDED_MESSAGE;
/* Accept first byte by clearing ACK */
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irq(&hostdata->lock);
dprintk(NDEBUG_EXTENDED, "scsi%d : receiving extended message\n", instance->host_no);
len = 2;
data = extended_msg + 1;
phase = PHASE_MSGIN;
NCR5380_transfer_pio(instance, &phase, &len, &data);
dprintk(NDEBUG_EXTENDED, "scsi%d : length=%d, code=0x%02x\n", instance->host_no, (int) extended_msg[1], (int) extended_msg[2]);
if (!len && extended_msg[1] > 0 &&
extended_msg[1] <= sizeof(extended_msg) - 2) {
/* Accept third byte by clearing ACK */
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
len = extended_msg[1] - 1;
data = extended_msg + 3;
phase = PHASE_MSGIN;
NCR5380_transfer_pio(instance, &phase, &len, &data);
dprintk(NDEBUG_EXTENDED, "scsi%d : message received, residual %d\n", instance->host_no, len);
switch (extended_msg[2]) {
case EXTENDED_SDTR:
case EXTENDED_WDTR:
case EXTENDED_MODIFY_DATA_POINTER:
case EXTENDED_EXTENDED_IDENTIFY:
tmp = 0;
}
} else if (len) {
printk("scsi%d: error receiving extended message\n", instance->host_no);
tmp = 0;
} else {
printk("scsi%d: extended message code %02x length %d is too long\n", instance->host_no, extended_msg[2], extended_msg[1]);
tmp = 0;
}
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
if (!hostdata->connected)
return;
/* Fall through to reject message */
/*
* If we get something weird that we aren't expecting,
* reject it.
*/
default:
if (!tmp) {
printk("scsi%d: rejecting message ", instance->host_no);
spi_print_msg(extended_msg);
printk("\n");
} else if (tmp != EXTENDED_MESSAGE)
scmd_printk(KERN_INFO, cmd,
"rejecting unknown message %02x\n",tmp);
else
scmd_printk(KERN_INFO, cmd,
"rejecting unknown extended message code %02x, length %d\n", extended_msg[1], extended_msg[0]);
msgout = MESSAGE_REJECT;
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_ATN);
break;
} /* switch (tmp) */
break;
case PHASE_MSGOUT:
len = 1;
data = &msgout;
hostdata->last_message = msgout;
NCR5380_transfer_pio(instance, &phase, &len, &data);
if (msgout == ABORT) {
hostdata->connected = NULL;
cmd->result = DID_ERROR << 16;
complete_cmd(instance, cmd);
NCR5380_write(SELECT_ENABLE_REG, hostdata->id_mask);
return;
}
msgout = NOP;
break;
case PHASE_CMDOUT:
len = cmd->cmd_len;
data = cmd->cmnd;
/*
* XXX for performance reasons, on machines with a
* PSEUDO-DMA architecture we should probably
* use the dma transfer function.
*/
NCR5380_transfer_pio(instance, &phase, &len, &data);
break;
case PHASE_STATIN:
len = 1;
data = &tmp;
NCR5380_transfer_pio(instance, &phase, &len, &data);
cmd->SCp.Status = tmp;
break;
default:
printk("scsi%d : unknown phase\n", instance->host_no);
NCR5380_dprint(NDEBUG_ANY, instance);
} /* switch(phase) */
ncr5380: Rework disconnect versus poll logic The atari_NCR5380.c and NCR5380.c core drivers differ in their handling of target disconnection. This is partly because atari_NCR5380.c had all of the polling and sleeping removed to become entirely interrupt-driven, and it is partly because of damage done to NCR5380.c after atari_NCR5380.c was forked. See commit 37cd23b44929 ("Linux 2.1.105") in history/history.git. The polling changes that were made in v2.1.105 are questionable at best: if REQ is not already asserted when NCR5380_transfer_pio() is invoked, and if the expected phase is DATA IN or DATA OUT, the function will schedule main() to execute after USLEEP_SLEEP jiffies and then return. The problems here are the expected REQ timing and the sleep interval*. Avoid this issue by using NCR5380_poll_politely() instead of scheduling main(). The atari_NCR5380.c core driver requires the use of the chip interrupt and always permits target disconnection. It sets the cmd->device->disconnect flag when a device disconnects, but never tests this flag. The NCR5380.c core driver permits disconnection only when instance->irq != NO_IRQ. It sets the cmd->device->disconnect flag when a device disconnects and it tests this flag in a couple of places: 1. During NCR5380_information_transfer(), following COMMAND OUT phase, if !cmd->device->disconnect, the initiator will take a guess as to whether or not the target will then choose to go to MESSAGE IN phase and disconnect. If the driver guesses "yes", it will schedule main() to execute after USLEEP_SLEEP jiffies and then return there. Unfortunately the driver may guess "yes" even after it has denied the target the disconnection privilege. When the target does not disconnect, the sleep can be beneficial, assuming the sleep interval is appropriate (mostly it is not*). And even if the driver guesses "yes" correctly, and the target would then disconnect, the driver still has to go through the MESSAGE IN phase in order to get to BUS FREE phase. The main loop can do nothing useful until BUS FREE, and sleeping just delays the phase transition. 2. If !cmd->device->disconnect and REQ is not already asserted when NCR5380_information_transfer() is invoked, the function polls for REQ for USLEEP_POLL jiffies. If REQ is not asserted, it then schedules main() to execute after USLEEP_SLEEP jiffies and returns. The idea is apparently to yeild the CPU while waiting for REQ. This is conditional upon !cmd->device->disconnect, but there seems to be no rhyme or reason for that. For example, the flag may be unset because disconnection privilege was denied because the driver has no IRQ. Or the flag may be unset because the device has never needed to disconnect before. Or if the flag is set, disconnection may have no relevance to the present bus phase. Another deficiency of the existing algorithm is as follows. When the driver has no IRQ, it prevents disconnection, and generally polls and sleeps more than it would normally. Now, if the driver is going to poll anyway, why not allow the target to disconnect? That way the driver can do something useful with the bus instead of polling unproductively! Avoid this pointless latency, complexity and guesswork by using NCR5380_poll_politely() instead of scheduling main(). * For g_NCR5380, the time intervals for USLEEP_SLEEP and USLEEP_POLL are 200 ms and 10 ms, respectively. They are 20 ms and 200 ms respectively for the other NCR5380 drivers. There doesn't seem to be any reason for this discrepancy. The timing seems to have no relation to the type of adapter. Bizarrely, the timing in g_NCR5380 seems to relate only to one particular type of target device. This patch attempts to solve the problem for all NCR5380 drivers and all target devices. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:26 -07:00
} else {
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irq(&hostdata->lock);
ncr5380: Rework disconnect versus poll logic The atari_NCR5380.c and NCR5380.c core drivers differ in their handling of target disconnection. This is partly because atari_NCR5380.c had all of the polling and sleeping removed to become entirely interrupt-driven, and it is partly because of damage done to NCR5380.c after atari_NCR5380.c was forked. See commit 37cd23b44929 ("Linux 2.1.105") in history/history.git. The polling changes that were made in v2.1.105 are questionable at best: if REQ is not already asserted when NCR5380_transfer_pio() is invoked, and if the expected phase is DATA IN or DATA OUT, the function will schedule main() to execute after USLEEP_SLEEP jiffies and then return. The problems here are the expected REQ timing and the sleep interval*. Avoid this issue by using NCR5380_poll_politely() instead of scheduling main(). The atari_NCR5380.c core driver requires the use of the chip interrupt and always permits target disconnection. It sets the cmd->device->disconnect flag when a device disconnects, but never tests this flag. The NCR5380.c core driver permits disconnection only when instance->irq != NO_IRQ. It sets the cmd->device->disconnect flag when a device disconnects and it tests this flag in a couple of places: 1. During NCR5380_information_transfer(), following COMMAND OUT phase, if !cmd->device->disconnect, the initiator will take a guess as to whether or not the target will then choose to go to MESSAGE IN phase and disconnect. If the driver guesses "yes", it will schedule main() to execute after USLEEP_SLEEP jiffies and then return there. Unfortunately the driver may guess "yes" even after it has denied the target the disconnection privilege. When the target does not disconnect, the sleep can be beneficial, assuming the sleep interval is appropriate (mostly it is not*). And even if the driver guesses "yes" correctly, and the target would then disconnect, the driver still has to go through the MESSAGE IN phase in order to get to BUS FREE phase. The main loop can do nothing useful until BUS FREE, and sleeping just delays the phase transition. 2. If !cmd->device->disconnect and REQ is not already asserted when NCR5380_information_transfer() is invoked, the function polls for REQ for USLEEP_POLL jiffies. If REQ is not asserted, it then schedules main() to execute after USLEEP_SLEEP jiffies and returns. The idea is apparently to yeild the CPU while waiting for REQ. This is conditional upon !cmd->device->disconnect, but there seems to be no rhyme or reason for that. For example, the flag may be unset because disconnection privilege was denied because the driver has no IRQ. Or the flag may be unset because the device has never needed to disconnect before. Or if the flag is set, disconnection may have no relevance to the present bus phase. Another deficiency of the existing algorithm is as follows. When the driver has no IRQ, it prevents disconnection, and generally polls and sleeps more than it would normally. Now, if the driver is going to poll anyway, why not allow the target to disconnect? That way the driver can do something useful with the bus instead of polling unproductively! Avoid this pointless latency, complexity and guesswork by using NCR5380_poll_politely() instead of scheduling main(). * For g_NCR5380, the time intervals for USLEEP_SLEEP and USLEEP_POLL are 200 ms and 10 ms, respectively. They are 20 ms and 200 ms respectively for the other NCR5380 drivers. There doesn't seem to be any reason for this discrepancy. The timing seems to have no relation to the type of adapter. Bizarrely, the timing in g_NCR5380 seems to relate only to one particular type of target device. This patch attempts to solve the problem for all NCR5380 drivers and all target devices. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:26 -07:00
NCR5380_poll_politely(instance, STATUS_REG, SR_REQ, SR_REQ, HZ);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irq(&hostdata->lock);
}
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
}
}
/*
* Function : void NCR5380_reselect (struct Scsi_Host *instance)
*
* Purpose : does reselection, initializing the instance->connected
* field to point to the scsi_cmnd for which the I_T_L or I_T_L_Q
* nexus has been reestablished,
*
* Inputs : instance - this instance of the NCR5380.
*
* Locks: io_request_lock held by caller if IRQ driven
*/
static void NCR5380_reselect(struct Scsi_Host *instance) {
struct NCR5380_hostdata *hostdata = shost_priv(instance);
unsigned char target_mask;
unsigned char lun, phase;
int len;
unsigned char msg[3];
unsigned char *data;
struct NCR5380_cmd *ncmd;
struct scsi_cmnd *tmp;
/*
* Disable arbitration, etc. since the host adapter obviously
* lost, and tell an interrupted NCR5380_select() to restart.
*/
NCR5380_write(MODE_REG, MR_BASE);
target_mask = NCR5380_read(CURRENT_SCSI_DATA_REG) & ~(hostdata->id_mask);
dprintk(NDEBUG_RESELECTION, "scsi%d : reselect\n", instance->host_no);
/*
* At this point, we have detected that our SCSI ID is on the bus,
* SEL is true and BSY was false for at least one bus settle delay
* (400 ns).
*
* We must assert BSY ourselves, until the target drops the SEL
* signal.
*/
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE | ICR_ASSERT_BSY);
if (NCR5380_poll_politely(instance,
STATUS_REG, SR_SEL, 0, 2 * HZ) < 0) {
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
return;
}
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
/*
* Wait for target to go into MSGIN.
*/
if (NCR5380_poll_politely(instance,
STATUS_REG, SR_REQ, SR_REQ, 2 * HZ) < 0) {
do_abort(instance);
return;
}
len = 1;
data = msg;
phase = PHASE_MSGIN;
NCR5380_transfer_pio(instance, &phase, &len, &data);
if (len) {
do_abort(instance);
return;
}
if (!(msg[0] & 0x80)) {
shost_printk(KERN_ERR, instance, "expecting IDENTIFY message, got ");
spi_print_msg(msg);
printk("\n");
do_abort(instance);
return;
}
lun = msg[0] & 0x07;
/*
* We need to add code for SCSI-II to track which devices have
* I_T_L_Q nexuses established, and which have simple I_T_L
* nexuses so we can chose to do additional data transfer.
*/
/*
* Find the command corresponding to the I_T_L or I_T_L_Q nexus we
* just reestablished, and remove it from the disconnected queue.
*/
tmp = NULL;
list_for_each_entry(ncmd, &hostdata->disconnected, list) {
struct scsi_cmnd *cmd = NCR5380_to_scmd(ncmd);
if (target_mask == (1 << scmd_id(cmd)) &&
lun == (u8)cmd->device->lun) {
list_del(&ncmd->list);
tmp = cmd;
break;
}
}
if (tmp) {
dsprintk(NDEBUG_RESELECTION | NDEBUG_QUEUES, instance,
"reselect: removed %p from disconnected queue\n", tmp);
} else {
shost_printk(KERN_ERR, instance, "target bitmask 0x%02x lun %d not in disconnected queue.\n",
target_mask, lun);
/*
* Since we have an established nexus that we can't do anything with,
* we must abort it.
*/
do_abort(instance);
return;
}
/* Accept message by clearing ACK */
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
hostdata->connected = tmp;
dprintk(NDEBUG_RESELECTION, "scsi%d : nexus established, target = %d, lun = %llu, tag = %d\n",
instance->host_no, tmp->device->id, tmp->device->lun, tmp->tag);
}
/*
* Function : void NCR5380_dma_complete (struct Scsi_Host *instance)
*
* Purpose : called by interrupt handler when DMA finishes or a phase
* mismatch occurs (which would finish the DMA transfer).
*
* Inputs : instance - this instance of the NCR5380.
*
* Returns : pointer to the scsi_cmnd structure for which the I_T_L
* nexus has been reestablished, on failure NULL is returned.
*/
#ifdef REAL_DMA
static void NCR5380_dma_complete(NCR5380_instance * instance) {
struct NCR5380_hostdata *hostdata = shost_priv(instance);
int transferred;
/*
* XXX this might not be right.
*
* Wait for final byte to transfer, ie wait for ACK to go false.
*
* We should use the Last Byte Sent bit, unfortunately this is
* not available on the 5380/5381 (only the various CMOS chips)
*
* FIXME: timeout, and need to handle long timeout/irq case
*/
NCR5380_poll_politely(instance, BUS_AND_STATUS_REG, BASR_ACK, 0, 5*HZ);
NCR5380_write(INITIATOR_COMMAND_REG, ICR_BASE);
/*
* The only places we should see a phase mismatch and have to send
* data from the same set of pointers will be the data transfer
* phases. So, residual, requested length are only important here.
*/
if (!(hostdata->connected->SCp.phase & SR_CD)) {
transferred = instance->dmalen - NCR5380_dma_residual();
hostdata->connected->SCp.this_residual -= transferred;
hostdata->connected->SCp.ptr += transferred;
}
}
#endif /* def REAL_DMA */
/*
* Function : int NCR5380_abort (struct scsi_cmnd *cmd)
*
* Purpose : abort a command
*
* Inputs : cmd - the scsi_cmnd to abort, code - code to set the
* host byte of the result field to, if zero DID_ABORTED is
* used.
*
* Returns : SUCCESS - success, FAILED on failure.
*
* XXX - there is no way to abort the command that is currently
* connected, you have to wait for it to complete. If this is
* a problem, we could implement longjmp() / setjmp(), setjmp()
* called where the loop started in NCR5380_main().
*
* Locks: host lock taken by caller
*/
static int NCR5380_abort(struct scsi_cmnd *cmd)
{
struct Scsi_Host *instance = cmd->device->host;
struct NCR5380_hostdata *hostdata = shost_priv(instance);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
unsigned long flags;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irqsave(&hostdata->lock, flags);
#if (NDEBUG & NDEBUG_ANY)
scmd_printk(KERN_INFO, cmd, "aborting command\n");
#endif
NCR5380_dprint(NDEBUG_ANY, instance);
NCR5380_dprint_phase(NDEBUG_ANY, instance);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irqrestore(&hostdata->lock, flags);
return FAILED;
}
/**
* NCR5380_bus_reset - reset the SCSI bus
* @cmd: SCSI command undergoing EH
*
* Returns SUCCESS
*/
static int NCR5380_bus_reset(struct scsi_cmnd *cmd)
{
struct Scsi_Host *instance = cmd->device->host;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
struct NCR5380_hostdata *hostdata = shost_priv(instance);
unsigned long flags;
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_lock_irqsave(&hostdata->lock, flags);
#if (NDEBUG & NDEBUG_ANY)
scmd_printk(KERN_INFO, cmd, "performing bus reset\n");
#endif
NCR5380_dprint(NDEBUG_ANY, instance);
NCR5380_dprint_phase(NDEBUG_ANY, instance);
do_reset(instance);
ncr5380: Change instance->host_lock to hostdata->lock NCR5380.c presently uses the instance->host_lock spin lock. Convert this to a new spin lock that protects the NCR5380_hostdata struct. atari_NCR5380.c previously used local_irq_save/restore() rather than a spin lock. Convert this to hostdata->lock in irq mode. For SMP platforms, the interrupt handler now also acquires the spin lock. This brings all locking in the two core drivers into agreement. Adding this locking also means that a bunch of volatile qualifiers can be removed from the members of the NCR5380_hostdata struct. This is done in a subsequent patch. Proper locking will allow the abort handler to locate a command being aborted. This is presently impossible if the abort handler is invoked when the command has been moved from a queue to a pointer on the stack. (If eh_abort_handler can't determine whether a command has been completed or is still being processed then it can't decide whether to return success or failure.) The hostdata spin lock is now held when calling NCR5380_select() and NCR5380_information_transfer(). Where possible, the lock is dropped for polling and PIO transfers. Clean up the now-redundant SELECT_ENABLE_REG writes, that used to provide limited mutual exclusion between information_transfer() and reselect(). Accessing hostdata->connected without data races means taking the lock; cleanup these accesses. The new spin lock falls away for m68k and other UP builds, so this should have little impact there. In the SMP case the new lock should be uncontested even when the SCSI bus is contested. Signed-off-by: Finn Thain <fthain@telegraphics.com.au> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ondrej Zary <linux@rainbow-software.org> Tested-by: Michael Schmitz <schmitzmic@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2016-01-02 22:05:51 -07:00
spin_unlock_irqrestore(&hostdata->lock, flags);
return SUCCESS;
}