2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/timer.h>
20 #include <linux/string.h>
21 #include <linux/slab.h>
22 #include <linux/kernel.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
40 #define SENSE_TIMEOUT (10*HZ)
41 #define START_UNIT_TIMEOUT (30*HZ)
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
47 #define BUS_RESET_SETTLE_TIME (10)
48 #define HOST_RESET_SETTLE_TIME (10)
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
53 if (shost
->host_busy
== shost
->host_failed
) {
54 wake_up_process(shost
->ehandler
);
55 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
64 * Schedule SCSI EH without scmd.
66 void scsi_schedule_eh(struct Scsi_Host
*shost
)
70 spin_lock_irqsave(shost
->host_lock
, flags
);
72 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
73 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
74 shost
->host_eh_scheduled
++;
75 scsi_eh_wakeup(shost
);
78 spin_unlock_irqrestore(shost
->host_lock
, flags
);
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
90 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
92 struct Scsi_Host
*shost
= scmd
->device
->host
;
99 spin_lock_irqsave(shost
->host_lock
, flags
);
100 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
101 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
105 scmd
->eh_eflags
|= eh_flag
;
106 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
107 shost
->host_failed
++;
108 scsi_eh_wakeup(shost
);
110 spin_unlock_irqrestore(shost
->host_lock
, flags
);
115 * scsi_add_timer - Start timeout timer for a single scsi command.
116 * @scmd: scsi command that is about to start running.
117 * @timeout: amount of time to allow this command to run.
118 * @complete: timeout function to call if timer isn't canceled.
121 * This should be turned into an inline function. Each scsi command
122 * has its own timer, and as it is added to the queue, we set up the
123 * timer. When the command completes, we cancel the timer.
125 void scsi_add_timer(struct scsi_cmnd
*scmd
, int timeout
,
126 void (*complete
)(struct scsi_cmnd
*))
130 * If the clock was already running for this command, then
131 * first delete the timer. The timer handling code gets rather
132 * confused if we don't do this.
134 if (scmd
->eh_timeout
.function
)
135 del_timer(&scmd
->eh_timeout
);
137 scmd
->eh_timeout
.data
= (unsigned long)scmd
;
138 scmd
->eh_timeout
.expires
= jiffies
+ timeout
;
139 scmd
->eh_timeout
.function
= (void (*)(unsigned long)) complete
;
141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
142 " %d, (%p)\n", __FUNCTION__
,
143 scmd
, timeout
, complete
));
145 add_timer(&scmd
->eh_timeout
);
149 * scsi_delete_timer - Delete/cancel timer for a given function.
150 * @scmd: Cmd that we are canceling timer for
153 * This should be turned into an inline function.
156 * 1 if we were able to detach the timer. 0 if we blew it, and the
157 * timer function has already started to run.
159 int scsi_delete_timer(struct scsi_cmnd
*scmd
)
163 rtn
= del_timer(&scmd
->eh_timeout
);
165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
166 " rtn: %d\n", __FUNCTION__
,
169 scmd
->eh_timeout
.data
= (unsigned long)NULL
;
170 scmd
->eh_timeout
.function
= NULL
;
176 * scsi_times_out - Timeout function for normal scsi commands.
177 * @scmd: Cmd that is timing out.
180 * We do not need to lock this. There is the potential for a race
181 * only in that the normal completion handling might run, but if the
182 * normal completion function determines that the timer has already
183 * fired, then it mustn't do anything.
185 void scsi_times_out(struct scsi_cmnd
*scmd
)
187 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
189 if (scmd
->device
->host
->transportt
->eh_timed_out
)
190 switch (scmd
->device
->host
->transportt
->eh_timed_out(scmd
)) {
195 scsi_add_timer(scmd
, scmd
->timeout_per_command
,
202 if (unlikely(!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
203 scmd
->result
|= DID_TIME_OUT
<< 16;
209 * scsi_block_when_processing_errors - Prevent cmds from being queued.
210 * @sdev: Device on which we are performing recovery.
213 * We block until the host is out of error recovery, and then check to
214 * see whether the host or the device is offline.
217 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
219 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
223 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
225 online
= scsi_device_online(sdev
);
227 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__
,
232 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
234 #ifdef CONFIG_SCSI_LOGGING
236 * scsi_eh_prt_fail_stats - Log info on failures.
237 * @shost: scsi host being recovered.
238 * @work_q: Queue of scsi cmds to process.
240 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
241 struct list_head
*work_q
)
243 struct scsi_cmnd
*scmd
;
244 struct scsi_device
*sdev
;
245 int total_failures
= 0;
248 int devices_failed
= 0;
250 shost_for_each_device(sdev
, shost
) {
251 list_for_each_entry(scmd
, work_q
, eh_entry
) {
252 if (scmd
->device
== sdev
) {
254 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
261 if (cmd_cancel
|| cmd_failed
) {
262 SCSI_LOG_ERROR_RECOVERY(3,
263 sdev_printk(KERN_INFO
, sdev
,
264 "%s: cmds failed: %d, cancel: %d\n",
265 __FUNCTION__
, cmd_failed
,
273 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
274 " devices require eh work\n",
275 total_failures
, devices_failed
));
280 * scsi_check_sense - Examine scsi cmd sense
281 * @scmd: Cmd to have sense checked.
284 * SUCCESS or FAILED or NEEDS_RETRY
287 * When a deferred error is detected the current command has
288 * not been executed and needs retrying.
290 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
292 struct scsi_sense_hdr sshdr
;
294 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
295 return FAILED
; /* no valid sense data */
297 if (scsi_sense_is_deferred(&sshdr
))
301 * Previous logic looked for FILEMARK, EOM or ILI which are
302 * mainly associated with tapes and returned SUCCESS.
304 if (sshdr
.response_code
== 0x70) {
306 if (scmd
->sense_buffer
[2] & 0xe0)
310 * descriptor format: look for "stream commands sense data
311 * descriptor" (see SSC-3). Assume single sense data
312 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
314 if ((sshdr
.additional_length
> 3) &&
315 (scmd
->sense_buffer
[8] == 0x4) &&
316 (scmd
->sense_buffer
[11] & 0xe0))
320 switch (sshdr
.sense_key
) {
323 case RECOVERED_ERROR
:
324 return /* soft_error */ SUCCESS
;
326 case ABORTED_COMMAND
:
331 * if we are expecting a cc/ua because of a bus reset that we
332 * performed, treat this just as a retry. otherwise this is
333 * information that we should pass up to the upper-level driver
334 * so that we can deal with it there.
336 if (scmd
->device
->expecting_cc_ua
) {
337 scmd
->device
->expecting_cc_ua
= 0;
341 * if the device is in the process of becoming ready, we
344 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
347 * if the device is not started, we need to wake
348 * the error handler to start the motor
350 if (scmd
->device
->allow_restart
&&
351 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
355 /* these three are not supported */
357 case VOLUME_OVERFLOW
:
365 if (scmd
->device
->retry_hwerror
)
370 case ILLEGAL_REQUEST
:
379 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
380 * @scmd: SCSI cmd to examine.
383 * This is *only* called when we are examining the status of commands
384 * queued during error recovery. the main difference here is that we
385 * don't allow for the possibility of retries here, and we are a lot
386 * more restrictive about what we consider acceptable.
388 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
391 * first check the host byte, to see if there is anything in there
392 * that would indicate what we need to do.
394 if (host_byte(scmd
->result
) == DID_RESET
) {
396 * rats. we are already in the error handler, so we now
397 * get to try and figure out what to do next. if the sense
398 * is valid, we have a pretty good idea of what to do.
399 * if not, we mark it as FAILED.
401 return scsi_check_sense(scmd
);
403 if (host_byte(scmd
->result
) != DID_OK
)
407 * next, check the message byte.
409 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
413 * now, check the status byte to see if this indicates
416 switch (status_byte(scmd
->result
)) {
418 case COMMAND_TERMINATED
:
420 case CHECK_CONDITION
:
421 return scsi_check_sense(scmd
);
423 case INTERMEDIATE_GOOD
:
424 case INTERMEDIATE_C_GOOD
:
426 * who knows? FIXME(eric)
431 case RESERVATION_CONFLICT
:
439 * scsi_eh_done - Completion function for error handling.
440 * @scmd: Cmd that is done.
442 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
444 struct completion
*eh_action
;
446 SCSI_LOG_ERROR_RECOVERY(3,
447 printk("%s scmd: %p result: %x\n",
448 __FUNCTION__
, scmd
, scmd
->result
));
450 eh_action
= scmd
->device
->host
->eh_action
;
456 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery.
457 * @scmd: SCSI Cmd to send.
458 * @timeout: Timeout for cmd.
461 * SUCCESS or FAILED or NEEDS_RETRY
463 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, int timeout
)
465 struct scsi_device
*sdev
= scmd
->device
;
466 struct Scsi_Host
*shost
= sdev
->host
;
467 DECLARE_COMPLETION(done
);
468 unsigned long timeleft
;
472 if (sdev
->scsi_level
<= SCSI_2
)
473 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
474 (sdev
->lun
<< 5 & 0xe0);
476 shost
->eh_action
= &done
;
478 spin_lock_irqsave(shost
->host_lock
, flags
);
480 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
481 spin_unlock_irqrestore(shost
->host_lock
, flags
);
483 timeleft
= wait_for_completion_timeout(&done
, timeout
);
485 shost
->eh_action
= NULL
;
487 scsi_log_completion(scmd
, SUCCESS
);
489 SCSI_LOG_ERROR_RECOVERY(3,
490 printk("%s: scmd: %p, timeleft: %ld\n",
491 __FUNCTION__
, scmd
, timeleft
));
494 * If there is time left scsi_eh_done got called, and we will
495 * examine the actual status codes to see whether the command
496 * actually did complete normally, else tell the host to forget
497 * about this command.
500 rtn
= scsi_eh_completed_normally(scmd
);
501 SCSI_LOG_ERROR_RECOVERY(3,
502 printk("%s: scsi_eh_completed_normally %x\n",
516 * FIXME(eric) - we are not tracking whether we could
517 * abort a timed out command or not. not sure how
518 * we should treat them differently anyways.
520 if (shost
->hostt
->eh_abort_handler
)
521 shost
->hostt
->eh_abort_handler(scmd
);
529 * scsi_request_sense - Request sense data from a particular target.
530 * @scmd: SCSI cmd for request sense.
533 * Some hosts automatically obtain this information, others require
534 * that we obtain it on our own. This function will *not* return until
535 * the command either times out, or it completes.
537 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
539 static unsigned char generic_sense
[6] =
540 {REQUEST_SENSE
, 0, 0, 0, 252, 0};
541 unsigned char *scsi_result
;
545 memcpy(scmd
->cmnd
, generic_sense
, sizeof(generic_sense
));
547 scsi_result
= kmalloc(252, GFP_ATOMIC
| ((scmd
->device
->host
->hostt
->unchecked_isa_dma
) ? __GFP_DMA
: 0));
550 if (unlikely(!scsi_result
)) {
551 printk(KERN_ERR
"%s: cannot allocate scsi_result.\n",
557 * zero the sense buffer. some host adapters automatically always
558 * request sense, so it is not a good idea that
559 * scmd->request_buffer and scmd->sense_buffer point to the same
560 * address (db). 0 is not a valid sense code.
562 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
563 memset(scsi_result
, 0, 252);
565 saved_result
= scmd
->result
;
566 scmd
->request_buffer
= scsi_result
;
567 scmd
->request_bufflen
= 252;
569 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
570 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
573 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
575 /* last chance to have valid sense data */
576 if(!SCSI_SENSE_VALID(scmd
)) {
577 memcpy(scmd
->sense_buffer
, scmd
->request_buffer
,
578 sizeof(scmd
->sense_buffer
));
584 * when we eventually call scsi_finish, we really wish to complete
585 * the original request, so let's restore the original data. (db)
587 scsi_setup_cmd_retry(scmd
);
588 scmd
->result
= saved_result
;
593 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
594 * @scmd: Original SCSI cmd that eh has finished.
595 * @done_q: Queue for processed commands.
598 * We don't want to use the normal command completion while we are are
599 * still handling errors - it may cause other commands to be queued,
600 * and that would disturb what we are doing. thus we really want to
601 * keep a list of pending commands for final completion, and once we
602 * are ready to leave error handling we handle completion for real.
604 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
606 scmd
->device
->host
->host_failed
--;
610 * set this back so that the upper level can correctly free up
613 scsi_setup_cmd_retry(scmd
);
614 list_move_tail(&scmd
->eh_entry
, done_q
);
616 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
619 * scsi_eh_get_sense - Get device sense data.
620 * @work_q: Queue of commands to process.
621 * @done_q: Queue of proccessed commands..
624 * See if we need to request sense information. if so, then get it
625 * now, so we have a better idea of what to do.
628 * This has the unfortunate side effect that if a shost adapter does
629 * not automatically request sense information, that we end up shutting
630 * it down before we request it.
632 * All drivers should request sense information internally these days,
633 * so for now all I have to say is tough noogies if you end up in here.
635 * XXX: Long term this code should go away, but that needs an audit of
638 static int scsi_eh_get_sense(struct list_head
*work_q
,
639 struct list_head
*done_q
)
641 struct scsi_cmnd
*scmd
, *next
;
644 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
645 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
646 SCSI_SENSE_VALID(scmd
))
649 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
650 "%s: requesting sense\n",
652 rtn
= scsi_request_sense(scmd
);
656 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
657 " result %x\n", scmd
,
659 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
661 rtn
= scsi_decide_disposition(scmd
);
664 * if the result was normal, then just pass it along to the
668 /* we don't want this command reissued, just
669 * finished with the sense data, so set
670 * retries to the max allowed to ensure it
671 * won't get reissued */
672 scmd
->retries
= scmd
->allowed
;
673 else if (rtn
!= NEEDS_RETRY
)
676 scsi_eh_finish_cmd(scmd
, done_q
);
679 return list_empty(work_q
);
683 * scsi_try_to_abort_cmd - Ask host to abort a running command.
684 * @scmd: SCSI cmd to abort from Lower Level.
687 * This function will not return until the user's completion function
688 * has been called. there is no timeout on this operation. if the
689 * author of the low-level driver wishes this operation to be timed,
690 * they can provide this facility themselves. helper functions in
691 * scsi_error.c can be supplied to make this easier to do.
693 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
695 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
699 * scsi_done was called just after the command timed out and before
700 * we had a chance to process it. (db)
702 if (scmd
->serial_number
== 0)
704 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
708 * scsi_eh_tur - Send TUR to device.
709 * @scmd: Scsi cmd to send TUR
712 * 0 - Device is ready. 1 - Device NOT ready.
714 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
716 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
717 int retry_cnt
= 1, rtn
;
721 memcpy(scmd
->cmnd
, tur_command
, sizeof(tur_command
));
724 * zero the sense buffer. the scsi spec mandates that any
725 * untransferred sense data should be interpreted as being zero.
727 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
729 saved_result
= scmd
->result
;
730 scmd
->request_buffer
= NULL
;
731 scmd
->request_bufflen
= 0;
733 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
735 scmd
->sc_data_direction
= DMA_NONE
;
737 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
740 * when we eventually call scsi_finish, we really wish to complete
741 * the original request, so let's restore the original data. (db)
743 scsi_setup_cmd_retry(scmd
);
744 scmd
->result
= saved_result
;
747 * hey, we are done. let's look to see what happened.
749 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
750 __FUNCTION__
, scmd
, rtn
));
753 else if (rtn
== NEEDS_RETRY
) {
762 * scsi_eh_abort_cmds - abort canceled commands.
763 * @shost: scsi host being recovered.
764 * @eh_done_q: list_head for processed commands.
767 * Try and see whether or not it makes sense to try and abort the
768 * running command. this only works out to be the case if we have one
769 * command that has timed out. if the command simply failed, it makes
770 * no sense to try and abort the command, since as far as the shost
771 * adapter is concerned, it isn't running.
773 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
774 struct list_head
*done_q
)
776 struct scsi_cmnd
*scmd
, *next
;
779 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
780 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
782 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
783 "0x%p\n", current
->comm
,
785 rtn
= scsi_try_to_abort_cmd(scmd
);
786 if (rtn
== SUCCESS
) {
787 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
788 if (!scsi_device_online(scmd
->device
) ||
789 !scsi_eh_tur(scmd
)) {
790 scsi_eh_finish_cmd(scmd
, done_q
);
794 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
801 return list_empty(work_q
);
805 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
806 * @scmd: SCSI cmd used to send BDR
809 * There is no timeout for this operation. if this operation is
810 * unreliable for a given host, then the host itself needs to put a
811 * timer on it, and set the host back to a consistent state prior to
814 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
818 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
821 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
822 if (rtn
== SUCCESS
) {
823 scmd
->device
->was_reset
= 1;
824 scmd
->device
->expecting_cc_ua
= 1;
831 * scsi_eh_try_stu - Send START_UNIT to device.
832 * @scmd: Scsi cmd to send START_UNIT
835 * 0 - Device is ready. 1 - Device NOT ready.
837 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
839 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
843 if (!scmd
->device
->allow_restart
)
846 memcpy(scmd
->cmnd
, stu_command
, sizeof(stu_command
));
849 * zero the sense buffer. the scsi spec mandates that any
850 * untransferred sense data should be interpreted as being zero.
852 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
854 saved_result
= scmd
->result
;
855 scmd
->request_buffer
= NULL
;
856 scmd
->request_bufflen
= 0;
858 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
860 scmd
->sc_data_direction
= DMA_NONE
;
862 rtn
= scsi_send_eh_cmnd(scmd
, START_UNIT_TIMEOUT
);
865 * when we eventually call scsi_finish, we really wish to complete
866 * the original request, so let's restore the original data. (db)
868 scsi_setup_cmd_retry(scmd
);
869 scmd
->result
= saved_result
;
872 * hey, we are done. let's look to see what happened.
874 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
875 __FUNCTION__
, scmd
, rtn
));
882 * scsi_eh_stu - send START_UNIT if needed
883 * @shost: scsi host being recovered.
884 * @eh_done_q: list_head for processed commands.
887 * If commands are failing due to not ready, initializing command required,
888 * try revalidating the device, which will end up sending a start unit.
890 static int scsi_eh_stu(struct Scsi_Host
*shost
,
891 struct list_head
*work_q
,
892 struct list_head
*done_q
)
894 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
895 struct scsi_device
*sdev
;
897 shost_for_each_device(sdev
, shost
) {
899 list_for_each_entry(scmd
, work_q
, eh_entry
)
900 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
901 scsi_check_sense(scmd
) == FAILED
) {
909 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
910 " 0x%p\n", current
->comm
, sdev
));
912 if (!scsi_eh_try_stu(stu_scmd
)) {
913 if (!scsi_device_online(sdev
) ||
914 !scsi_eh_tur(stu_scmd
)) {
915 list_for_each_entry_safe(scmd
, next
,
917 if (scmd
->device
== sdev
)
918 scsi_eh_finish_cmd(scmd
, done_q
);
922 SCSI_LOG_ERROR_RECOVERY(3,
923 printk("%s: START_UNIT failed to sdev:"
924 " 0x%p\n", current
->comm
, sdev
));
928 return list_empty(work_q
);
933 * scsi_eh_bus_device_reset - send bdr if needed
934 * @shost: scsi host being recovered.
935 * @eh_done_q: list_head for processed commands.
938 * Try a bus device reset. still, look to see whether we have multiple
939 * devices that are jammed or not - if we have multiple devices, it
940 * makes no sense to try bus_device_reset - we really would need to try
941 * a bus_reset instead.
943 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
944 struct list_head
*work_q
,
945 struct list_head
*done_q
)
947 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
948 struct scsi_device
*sdev
;
951 shost_for_each_device(sdev
, shost
) {
953 list_for_each_entry(scmd
, work_q
, eh_entry
)
954 if (scmd
->device
== sdev
) {
962 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
963 " 0x%p\n", current
->comm
,
965 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
966 if (rtn
== SUCCESS
) {
967 if (!scsi_device_online(sdev
) ||
968 !scsi_eh_tur(bdr_scmd
)) {
969 list_for_each_entry_safe(scmd
, next
,
971 if (scmd
->device
== sdev
)
972 scsi_eh_finish_cmd(scmd
,
977 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
985 return list_empty(work_q
);
989 * scsi_try_bus_reset - ask host to perform a bus reset
990 * @scmd: SCSI cmd to send bus reset.
992 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
997 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1000 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
1003 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
1005 if (rtn
== SUCCESS
) {
1006 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1007 ssleep(BUS_RESET_SETTLE_TIME
);
1008 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1009 scsi_report_bus_reset(scmd
->device
->host
,
1010 scmd_channel(scmd
));
1011 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1018 * scsi_try_host_reset - ask host adapter to reset itself
1019 * @scmd: SCSI cmd to send hsot reset.
1021 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
1023 unsigned long flags
;
1026 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1029 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
1032 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
1034 if (rtn
== SUCCESS
) {
1035 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1036 ssleep(HOST_RESET_SETTLE_TIME
);
1037 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1038 scsi_report_bus_reset(scmd
->device
->host
,
1039 scmd_channel(scmd
));
1040 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1047 * scsi_eh_bus_reset - send a bus reset
1048 * @shost: scsi host being recovered.
1049 * @eh_done_q: list_head for processed commands.
1051 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1052 struct list_head
*work_q
,
1053 struct list_head
*done_q
)
1055 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1056 unsigned int channel
;
1060 * we really want to loop over the various channels, and do this on
1061 * a channel by channel basis. we should also check to see if any
1062 * of the failed commands are on soft_reset devices, and if so, skip
1066 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1068 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1069 if (channel
== scmd_channel(scmd
)) {
1073 * FIXME add back in some support for
1074 * soft_reset devices.
1081 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1082 " %d\n", current
->comm
,
1084 rtn
= scsi_try_bus_reset(chan_scmd
);
1085 if (rtn
== SUCCESS
) {
1086 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1087 if (channel
== scmd_channel(scmd
))
1088 if (!scsi_device_online(scmd
->device
) ||
1090 scsi_eh_finish_cmd(scmd
,
1094 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1095 " failed chan: %d\n",
1100 return list_empty(work_q
);
1104 * scsi_eh_host_reset - send a host reset
1105 * @work_q: list_head for processed commands.
1106 * @done_q: list_head for processed commands.
1108 static int scsi_eh_host_reset(struct list_head
*work_q
,
1109 struct list_head
*done_q
)
1111 struct scsi_cmnd
*scmd
, *next
;
1114 if (!list_empty(work_q
)) {
1115 scmd
= list_entry(work_q
->next
,
1116 struct scsi_cmnd
, eh_entry
);
1118 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1121 rtn
= scsi_try_host_reset(scmd
);
1122 if (rtn
== SUCCESS
) {
1123 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1124 if (!scsi_device_online(scmd
->device
) ||
1125 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1127 scsi_eh_finish_cmd(scmd
, done_q
);
1130 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1135 return list_empty(work_q
);
1139 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1140 * @work_q: list_head for processed commands.
1141 * @done_q: list_head for processed commands.
1144 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1145 struct list_head
*done_q
)
1147 struct scsi_cmnd
*scmd
, *next
;
1149 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1150 sdev_printk(KERN_INFO
, scmd
->device
,
1151 "scsi: Device offlined - not"
1152 " ready after error recovery\n");
1153 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1154 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1156 * FIXME: Handle lost cmds.
1159 scsi_eh_finish_cmd(scmd
, done_q
);
1165 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1166 * @scmd: SCSI cmd to examine.
1169 * This is *only* called when we are examining the status after sending
1170 * out the actual data command. any commands that are queued for error
1171 * recovery (e.g. test_unit_ready) do *not* come through here.
1173 * When this routine returns failed, it means the error handler thread
1174 * is woken. In cases where the error code indicates an error that
1175 * doesn't require the error handler read (i.e. we don't need to
1176 * abort/reset), this function should return SUCCESS.
1178 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1183 * if the device is offline, then we clearly just pass the result back
1184 * up to the top level.
1186 if (!scsi_device_online(scmd
->device
)) {
1187 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1194 * first check the host byte, to see if there is anything in there
1195 * that would indicate what we need to do.
1197 switch (host_byte(scmd
->result
)) {
1198 case DID_PASSTHROUGH
:
1200 * no matter what, pass this through to the upper layer.
1201 * nuke this special code so that it looks like we are saying
1204 scmd
->result
&= 0xff00ffff;
1208 * looks good. drop through, and check the next byte.
1211 case DID_NO_CONNECT
:
1212 case DID_BAD_TARGET
:
1215 * note - this means that we just report the status back
1216 * to the top level driver, not that we actually think
1217 * that it indicates SUCCESS.
1221 * when the low level driver returns did_soft_error,
1222 * it is responsible for keeping an internal retry counter
1223 * in order to avoid endless loops (db)
1225 * actually this is a bug in this function here. we should
1226 * be mindful of the maximum number of retries specified
1227 * and not get stuck in a loop.
1229 case DID_SOFT_ERROR
:
1235 return ADD_TO_MLQUEUE
;
1238 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1239 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1241 * execute reservation conflict processing code
1252 * when we scan the bus, we get timeout messages for
1253 * these commands if there is no device available.
1254 * other hosts report did_no_connect for the same thing.
1256 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1257 scmd
->cmnd
[0] == INQUIRY
)) {
1269 * next, check the message byte.
1271 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1275 * check the status byte to see if this indicates anything special.
1277 switch (status_byte(scmd
->result
)) {
1280 * the case of trying to send too many commands to a
1281 * tagged queueing device.
1285 * device can't talk to us at the moment. Should only
1286 * occur (SAM-3) when the task queue is empty, so will cause
1287 * the empty queue handling to trigger a stall in the
1290 return ADD_TO_MLQUEUE
;
1292 case COMMAND_TERMINATED
:
1295 case CHECK_CONDITION
:
1296 rtn
= scsi_check_sense(scmd
);
1297 if (rtn
== NEEDS_RETRY
)
1299 /* if rtn == FAILED, we have no sense information;
1300 * returning FAILED will wake the error handler thread
1301 * to collect the sense and redo the decide
1304 case CONDITION_GOOD
:
1305 case INTERMEDIATE_GOOD
:
1306 case INTERMEDIATE_C_GOOD
:
1309 * who knows? FIXME(eric)
1313 case RESERVATION_CONFLICT
:
1314 sdev_printk(KERN_INFO
, scmd
->device
,
1315 "reservation conflict\n");
1316 return SUCCESS
; /* causes immediate i/o error */
1324 /* we requeue for retry because the error was retryable, and
1325 * the request was not marked fast fail. Note that above,
1326 * even if the request is marked fast fail, we still requeue
1327 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1328 if ((++scmd
->retries
) <= scmd
->allowed
1329 && !blk_noretry_request(scmd
->request
)) {
1333 * no more retries - report this one back to upper level.
1340 * scsi_eh_lock_door - Prevent medium removal for the specified device
1341 * @sdev: SCSI device to prevent medium removal
1344 * We must be called from process context; scsi_allocate_request()
1348 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1349 * head of the devices request queue, and continue.
1352 * scsi_allocate_request() may sleep waiting for existing requests to
1353 * be processed. However, since we haven't kicked off any request
1354 * processing for this host, this may deadlock.
1356 * If scsi_allocate_request() fails for what ever reason, we
1357 * completely forget to lock the door.
1359 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1361 unsigned char cmnd
[MAX_COMMAND_SIZE
];
1363 cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
1367 cmnd
[4] = SCSI_REMOVAL_PREVENT
;
1370 scsi_execute_async(sdev
, cmnd
, 6, DMA_NONE
, NULL
, 0, 0, 10 * HZ
,
1371 5, NULL
, NULL
, GFP_KERNEL
);
1376 * scsi_restart_operations - restart io operations to the specified host.
1377 * @shost: Host we are restarting.
1380 * When we entered the error handler, we blocked all further i/o to
1381 * this device. we need to 'reverse' this process.
1383 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1385 struct scsi_device
*sdev
;
1386 unsigned long flags
;
1389 * If the door was locked, we need to insert a door lock request
1390 * onto the head of the SCSI request queue for the device. There
1391 * is no point trying to lock the door of an off-line device.
1393 shost_for_each_device(sdev
, shost
) {
1394 if (scsi_device_online(sdev
) && sdev
->locked
)
1395 scsi_eh_lock_door(sdev
);
1399 * next free up anything directly waiting upon the host. this
1400 * will be requests for character device operations, and also for
1401 * ioctls to queued block devices.
1403 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1406 spin_lock_irqsave(shost
->host_lock
, flags
);
1407 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1408 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1409 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1410 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1412 wake_up(&shost
->host_wait
);
1415 * finally we need to re-initiate requests that may be pending. we will
1416 * have had everything blocked while error handling is taking place, and
1417 * now that error recovery is done, we will need to ensure that these
1418 * requests are started.
1420 scsi_run_host_queues(shost
);
1424 * scsi_eh_ready_devs - check device ready state and recover if not.
1425 * @shost: host to be recovered.
1426 * @eh_done_q: list_head for processed commands.
1429 static void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1430 struct list_head
*work_q
,
1431 struct list_head
*done_q
)
1433 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1434 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1435 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1436 if (!scsi_eh_host_reset(work_q
, done_q
))
1437 scsi_eh_offline_sdevs(work_q
, done_q
);
1441 * scsi_eh_flush_done_q - finish processed commands or retry them.
1442 * @done_q: list_head of processed commands.
1445 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1447 struct scsi_cmnd
*scmd
, *next
;
1449 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1450 list_del_init(&scmd
->eh_entry
);
1451 if (scsi_device_online(scmd
->device
) &&
1452 !blk_noretry_request(scmd
->request
) &&
1453 (++scmd
->retries
<= scmd
->allowed
)) {
1454 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1458 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1461 * If just we got sense for the device (called
1462 * scsi_eh_get_sense), scmd->result is already
1463 * set, do not set DRIVER_TIMEOUT.
1466 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1467 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1469 current
->comm
, scmd
));
1470 scsi_finish_command(scmd
);
1474 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1477 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1478 * @shost: Host to unjam.
1481 * When we come in here, we *know* that all commands on the bus have
1482 * either completed, failed or timed out. we also know that no further
1483 * commands are being sent to the host, so things are relatively quiet
1484 * and we have freedom to fiddle with things as we wish.
1486 * This is only the *default* implementation. it is possible for
1487 * individual drivers to supply their own version of this function, and
1488 * if the maintainer wishes to do this, it is strongly suggested that
1489 * this function be taken as a template and modified. this function
1490 * was designed to correctly handle problems for about 95% of the
1491 * different cases out there, and it should always provide at least a
1492 * reasonable amount of error recovery.
1494 * Any command marked 'failed' or 'timeout' must eventually have
1495 * scsi_finish_cmd() called for it. we do all of the retry stuff
1496 * here, so when we restart the host after we return it should have an
1499 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1501 unsigned long flags
;
1502 LIST_HEAD(eh_work_q
);
1503 LIST_HEAD(eh_done_q
);
1505 spin_lock_irqsave(shost
->host_lock
, flags
);
1506 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1507 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1509 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1511 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1512 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1513 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1515 scsi_eh_flush_done_q(&eh_done_q
);
1519 * scsi_error_handler - SCSI error handler thread
1520 * @data: Host for which we are running.
1523 * This is the main error handling loop. This is run as a kernel thread
1524 * for every SCSI host and handles all error handling activity.
1526 int scsi_error_handler(void *data
)
1528 struct Scsi_Host
*shost
= data
;
1530 current
->flags
|= PF_NOFREEZE
;
1533 * We use TASK_INTERRUPTIBLE so that the thread is not
1534 * counted against the load average as a running process.
1535 * We never actually get interrupted because kthread_run
1536 * disables singal delivery for the created thread.
1538 set_current_state(TASK_INTERRUPTIBLE
);
1539 while (!kthread_should_stop()) {
1540 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1541 shost
->host_failed
!= shost
->host_busy
) {
1542 SCSI_LOG_ERROR_RECOVERY(1,
1543 printk("Error handler scsi_eh_%d sleeping\n",
1546 set_current_state(TASK_INTERRUPTIBLE
);
1550 __set_current_state(TASK_RUNNING
);
1551 SCSI_LOG_ERROR_RECOVERY(1,
1552 printk("Error handler scsi_eh_%d waking up\n",
1556 * We have a host that is failing for some reason. Figure out
1557 * what we need to do to get it up and online again (if we can).
1558 * If we fail, we end up taking the thing offline.
1560 if (shost
->transportt
->eh_strategy_handler
)
1561 shost
->transportt
->eh_strategy_handler(shost
);
1563 scsi_unjam_host(shost
);
1566 * Note - if the above fails completely, the action is to take
1567 * individual devices offline and flush the queue of any
1568 * outstanding requests that may have been pending. When we
1569 * restart, we restart any I/O to any other devices on the bus
1570 * which are still online.
1572 scsi_restart_operations(shost
);
1573 set_current_state(TASK_INTERRUPTIBLE
);
1575 __set_current_state(TASK_RUNNING
);
1577 SCSI_LOG_ERROR_RECOVERY(1,
1578 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1579 shost
->ehandler
= NULL
;
1584 * Function: scsi_report_bus_reset()
1586 * Purpose: Utility function used by low-level drivers to report that
1587 * they have observed a bus reset on the bus being handled.
1589 * Arguments: shost - Host in question
1590 * channel - channel on which reset was observed.
1594 * Lock status: Host lock must be held.
1596 * Notes: This only needs to be called if the reset is one which
1597 * originates from an unknown location. Resets originated
1598 * by the mid-level itself don't need to call this, but there
1599 * should be no harm.
1601 * The main purpose of this is to make sure that a CHECK_CONDITION
1602 * is properly treated.
1604 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1606 struct scsi_device
*sdev
;
1608 __shost_for_each_device(sdev
, shost
) {
1609 if (channel
== sdev_channel(sdev
)) {
1610 sdev
->was_reset
= 1;
1611 sdev
->expecting_cc_ua
= 1;
1615 EXPORT_SYMBOL(scsi_report_bus_reset
);
1618 * Function: scsi_report_device_reset()
1620 * Purpose: Utility function used by low-level drivers to report that
1621 * they have observed a device reset on the device being handled.
1623 * Arguments: shost - Host in question
1624 * channel - channel on which reset was observed
1625 * target - target on which reset was observed
1629 * Lock status: Host lock must be held
1631 * Notes: This only needs to be called if the reset is one which
1632 * originates from an unknown location. Resets originated
1633 * by the mid-level itself don't need to call this, but there
1634 * should be no harm.
1636 * The main purpose of this is to make sure that a CHECK_CONDITION
1637 * is properly treated.
1639 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1641 struct scsi_device
*sdev
;
1643 __shost_for_each_device(sdev
, shost
) {
1644 if (channel
== sdev_channel(sdev
) &&
1645 target
== sdev_id(sdev
)) {
1646 sdev
->was_reset
= 1;
1647 sdev
->expecting_cc_ua
= 1;
1651 EXPORT_SYMBOL(scsi_report_device_reset
);
1654 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1659 * Function: scsi_reset_provider
1661 * Purpose: Send requested reset to a bus or device at any phase.
1663 * Arguments: device - device to send reset to
1664 * flag - reset type (see scsi.h)
1666 * Returns: SUCCESS/FAILURE.
1668 * Notes: This is used by the SCSI Generic driver to provide
1669 * Bus/Device reset capability.
1672 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1674 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1675 struct Scsi_Host
*shost
= dev
->host
;
1677 unsigned long flags
;
1680 scmd
->request
= &req
;
1681 memset(&scmd
->eh_timeout
, 0, sizeof(scmd
->eh_timeout
));
1683 memset(&scmd
->cmnd
, '\0', sizeof(scmd
->cmnd
));
1685 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1687 scmd
->buffer
= NULL
;
1689 scmd
->request_buffer
= NULL
;
1690 scmd
->request_bufflen
= 0;
1694 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1696 init_timer(&scmd
->eh_timeout
);
1699 * Sometimes the command can get back into the timer chain,
1700 * so use the pid as an identifier.
1704 spin_lock_irqsave(shost
->host_lock
, flags
);
1705 shost
->tmf_in_progress
= 1;
1706 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1709 case SCSI_TRY_RESET_DEVICE
:
1710 rtn
= scsi_try_bus_device_reset(scmd
);
1714 case SCSI_TRY_RESET_BUS
:
1715 rtn
= scsi_try_bus_reset(scmd
);
1719 case SCSI_TRY_RESET_HOST
:
1720 rtn
= scsi_try_host_reset(scmd
);
1726 spin_lock_irqsave(shost
->host_lock
, flags
);
1727 shost
->tmf_in_progress
= 0;
1728 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1731 * be sure to wake up anyone who was sleeping or had their queue
1732 * suspended while we performed the TMF.
1734 SCSI_LOG_ERROR_RECOVERY(3,
1735 printk("%s: waking up host to restart after TMF\n",
1738 wake_up(&shost
->host_wait
);
1740 scsi_run_host_queues(shost
);
1742 scsi_next_command(scmd
);
1745 EXPORT_SYMBOL(scsi_reset_provider
);
1748 * scsi_normalize_sense - normalize main elements from either fixed or
1749 * descriptor sense data format into a common format.
1751 * @sense_buffer: byte array containing sense data returned by device
1752 * @sb_len: number of valid bytes in sense_buffer
1753 * @sshdr: pointer to instance of structure that common
1754 * elements are written to.
1757 * The "main elements" from sense data are: response_code, sense_key,
1758 * asc, ascq and additional_length (only for descriptor format).
1760 * Typically this function can be called after a device has
1761 * responded to a SCSI command with the CHECK_CONDITION status.
1764 * 1 if valid sense data information found, else 0;
1766 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1767 struct scsi_sense_hdr
*sshdr
)
1769 if (!sense_buffer
|| !sb_len
)
1772 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1774 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1776 if (!scsi_sense_valid(sshdr
))
1779 if (sshdr
->response_code
>= 0x72) {
1784 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1786 sshdr
->asc
= sense_buffer
[2];
1788 sshdr
->ascq
= sense_buffer
[3];
1790 sshdr
->additional_length
= sense_buffer
[7];
1796 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1798 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1799 sb_len
: (sense_buffer
[7] + 8);
1801 sshdr
->asc
= sense_buffer
[12];
1803 sshdr
->ascq
= sense_buffer
[13];
1809 EXPORT_SYMBOL(scsi_normalize_sense
);
1811 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1812 struct scsi_sense_hdr
*sshdr
)
1814 return scsi_normalize_sense(cmd
->sense_buffer
,
1815 sizeof(cmd
->sense_buffer
), sshdr
);
1817 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1820 * scsi_sense_desc_find - search for a given descriptor type in
1821 * descriptor sense data format.
1823 * @sense_buffer: byte array of descriptor format sense data
1824 * @sb_len: number of valid bytes in sense_buffer
1825 * @desc_type: value of descriptor type to find
1826 * (e.g. 0 -> information)
1829 * only valid when sense data is in descriptor format
1832 * pointer to start of (first) descriptor if found else NULL
1834 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
1837 int add_sen_len
, add_len
, desc_len
, k
;
1840 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
1842 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
1844 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
1845 add_sen_len
: (sb_len
- 8);
1846 descp
= &sense_buffer
[8];
1847 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
1849 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
1850 desc_len
= add_len
+ 2;
1851 if (descp
[0] == desc_type
)
1853 if (add_len
< 0) // short descriptor ??
1858 EXPORT_SYMBOL(scsi_sense_desc_find
);
1861 * scsi_get_sense_info_fld - attempts to get information field from
1862 * sense data (either fixed or descriptor format)
1864 * @sense_buffer: byte array of sense data
1865 * @sb_len: number of valid bytes in sense_buffer
1866 * @info_out: pointer to 64 integer where 8 or 4 byte information
1867 * field will be placed if found.
1870 * 1 if information field found, 0 if not found.
1872 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
1881 switch (sense_buffer
[0] & 0x7f) {
1884 if (sense_buffer
[0] & 0x80) {
1885 *info_out
= (sense_buffer
[3] << 24) +
1886 (sense_buffer
[4] << 16) +
1887 (sense_buffer
[5] << 8) + sense_buffer
[6];
1893 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
1895 if (ucp
&& (0xa == ucp
[1])) {
1897 for (j
= 0; j
< 8; ++j
) {
1910 EXPORT_SYMBOL(scsi_get_sense_info_fld
);