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/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
40 #include "scsi_transport_api.h"
42 #include <trace/events/scsi.h>
44 #define SENSE_TIMEOUT (10*HZ)
47 * These should *probably* be handled by the host itself.
48 * Since it is allowed to sleep, it probably should.
50 #define BUS_RESET_SETTLE_TIME (10)
51 #define HOST_RESET_SETTLE_TIME (10)
53 /* called with shost->host_lock held */
54 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
56 if (shost
->host_busy
== shost
->host_failed
) {
57 trace_scsi_eh_wakeup(shost
);
58 wake_up_process(shost
->ehandler
);
59 SCSI_LOG_ERROR_RECOVERY(5,
60 printk("Waking error handler thread\n"));
65 * scsi_schedule_eh - schedule EH for SCSI host
66 * @shost: SCSI host to invoke error handling on.
68 * Schedule SCSI EH without scmd.
70 void scsi_schedule_eh(struct Scsi_Host
*shost
)
74 spin_lock_irqsave(shost
->host_lock
, flags
);
76 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
77 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
78 shost
->host_eh_scheduled
++;
79 scsi_eh_wakeup(shost
);
82 spin_unlock_irqrestore(shost
->host_lock
, flags
);
84 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
87 * scsi_eh_scmd_add - add scsi cmd to error handling.
88 * @scmd: scmd to run eh on.
89 * @eh_flag: optional SCSI_EH flag.
94 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
96 struct Scsi_Host
*shost
= scmd
->device
->host
;
100 if (!shost
->ehandler
)
103 spin_lock_irqsave(shost
->host_lock
, flags
);
104 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
105 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
109 scmd
->eh_eflags
|= eh_flag
;
110 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
111 shost
->host_failed
++;
112 scsi_eh_wakeup(shost
);
114 spin_unlock_irqrestore(shost
->host_lock
, flags
);
119 * scsi_times_out - Timeout function for normal scsi commands.
120 * @req: request that is timing out.
123 * We do not need to lock this. There is the potential for a race
124 * only in that the normal completion handling might run, but if the
125 * normal completion function determines that the timer has already
126 * fired, then it mustn't do anything.
128 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
130 struct scsi_cmnd
*scmd
= req
->special
;
131 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
133 trace_scsi_dispatch_cmd_timeout(scmd
);
134 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
136 if (scmd
->device
->host
->transportt
->eh_timed_out
)
137 rtn
= scmd
->device
->host
->transportt
->eh_timed_out(scmd
);
138 else if (scmd
->device
->host
->hostt
->eh_timed_out
)
139 rtn
= scmd
->device
->host
->hostt
->eh_timed_out(scmd
);
141 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
142 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
143 scmd
->result
|= DID_TIME_OUT
<< 16;
144 rtn
= BLK_EH_HANDLED
;
151 * scsi_block_when_processing_errors - Prevent cmds from being queued.
152 * @sdev: Device on which we are performing recovery.
155 * We block until the host is out of error recovery, and then check to
156 * see whether the host or the device is offline.
159 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
161 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
165 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
167 online
= scsi_device_online(sdev
);
169 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
174 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
176 #ifdef CONFIG_SCSI_LOGGING
178 * scsi_eh_prt_fail_stats - Log info on failures.
179 * @shost: scsi host being recovered.
180 * @work_q: Queue of scsi cmds to process.
182 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
183 struct list_head
*work_q
)
185 struct scsi_cmnd
*scmd
;
186 struct scsi_device
*sdev
;
187 int total_failures
= 0;
190 int devices_failed
= 0;
192 shost_for_each_device(sdev
, shost
) {
193 list_for_each_entry(scmd
, work_q
, eh_entry
) {
194 if (scmd
->device
== sdev
) {
196 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
203 if (cmd_cancel
|| cmd_failed
) {
204 SCSI_LOG_ERROR_RECOVERY(3,
205 sdev_printk(KERN_INFO
, sdev
,
206 "%s: cmds failed: %d, cancel: %d\n",
207 __func__
, cmd_failed
,
215 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
216 " devices require eh work\n",
217 total_failures
, devices_failed
));
222 * scsi_check_sense - Examine scsi cmd sense
223 * @scmd: Cmd to have sense checked.
226 * SUCCESS or FAILED or NEEDS_RETRY
229 * When a deferred error is detected the current command has
230 * not been executed and needs retrying.
232 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
234 struct scsi_device
*sdev
= scmd
->device
;
235 struct scsi_sense_hdr sshdr
;
237 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
238 return FAILED
; /* no valid sense data */
240 if (scsi_sense_is_deferred(&sshdr
))
243 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
244 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
247 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
248 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
250 /* handler does not care. Drop down to default handling */
254 * Previous logic looked for FILEMARK, EOM or ILI which are
255 * mainly associated with tapes and returned SUCCESS.
257 if (sshdr
.response_code
== 0x70) {
259 if (scmd
->sense_buffer
[2] & 0xe0)
263 * descriptor format: look for "stream commands sense data
264 * descriptor" (see SSC-3). Assume single sense data
265 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
267 if ((sshdr
.additional_length
> 3) &&
268 (scmd
->sense_buffer
[8] == 0x4) &&
269 (scmd
->sense_buffer
[11] & 0xe0))
273 switch (sshdr
.sense_key
) {
276 case RECOVERED_ERROR
:
277 return /* soft_error */ SUCCESS
;
279 case ABORTED_COMMAND
:
280 if (sshdr
.asc
== 0x10) /* DIF */
287 * if we are expecting a cc/ua because of a bus reset that we
288 * performed, treat this just as a retry. otherwise this is
289 * information that we should pass up to the upper-level driver
290 * so that we can deal with it there.
292 if (scmd
->device
->expecting_cc_ua
) {
293 scmd
->device
->expecting_cc_ua
= 0;
297 * if the device is in the process of becoming ready, we
300 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
303 * if the device is not started, we need to wake
304 * the error handler to start the motor
306 if (scmd
->device
->allow_restart
&&
307 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
310 if (sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
311 scmd_printk(KERN_WARNING
, scmd
,
312 "Warning! Received an indication that the "
313 "LUN assignments on this target have "
314 "changed. The Linux SCSI layer does not "
315 "automatically remap LUN assignments.\n");
316 else if (sshdr
.asc
== 0x3f)
317 scmd_printk(KERN_WARNING
, scmd
,
318 "Warning! Received an indication that the "
319 "operating parameters on this target have "
320 "changed. The Linux SCSI layer does not "
321 "automatically adjust these parameters.\n");
324 * Pass the UA upwards for a determination in the completion
329 /* these three are not supported */
331 case VOLUME_OVERFLOW
:
336 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
337 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
338 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
344 if (scmd
->device
->retry_hwerror
)
345 return ADD_TO_MLQUEUE
;
349 case ILLEGAL_REQUEST
:
357 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
359 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
360 struct scsi_device
*tmp_sdev
;
362 if (!sht
->change_queue_depth
||
363 sdev
->queue_depth
>= sdev
->max_queue_depth
)
366 if (time_before(jiffies
,
367 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
370 if (time_before(jiffies
,
371 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
375 * Walk all devices of a target and do
378 shost_for_each_device(tmp_sdev
, sdev
->host
) {
379 if (tmp_sdev
->channel
!= sdev
->channel
||
380 tmp_sdev
->id
!= sdev
->id
||
381 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
384 * call back into LLD to increase queue_depth by one
385 * with ramp up reason code.
387 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
388 SCSI_QDEPTH_RAMP_UP
);
389 sdev
->last_queue_ramp_up
= jiffies
;
393 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
395 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
396 struct scsi_device
*tmp_sdev
;
398 if (!sht
->change_queue_depth
)
401 shost_for_each_device(tmp_sdev
, sdev
->host
) {
402 if (tmp_sdev
->channel
!= sdev
->channel
||
403 tmp_sdev
->id
!= sdev
->id
)
406 * We do not know the number of commands that were at
407 * the device when we got the queue full so we start
408 * from the highest possible value and work our way down.
410 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
416 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
417 * @scmd: SCSI cmd to examine.
420 * This is *only* called when we are examining the status of commands
421 * queued during error recovery. the main difference here is that we
422 * don't allow for the possibility of retries here, and we are a lot
423 * more restrictive about what we consider acceptable.
425 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
428 * first check the host byte, to see if there is anything in there
429 * that would indicate what we need to do.
431 if (host_byte(scmd
->result
) == DID_RESET
) {
433 * rats. we are already in the error handler, so we now
434 * get to try and figure out what to do next. if the sense
435 * is valid, we have a pretty good idea of what to do.
436 * if not, we mark it as FAILED.
438 return scsi_check_sense(scmd
);
440 if (host_byte(scmd
->result
) != DID_OK
)
444 * next, check the message byte.
446 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
450 * now, check the status byte to see if this indicates
453 switch (status_byte(scmd
->result
)) {
455 scsi_handle_queue_ramp_up(scmd
->device
);
456 case COMMAND_TERMINATED
:
458 case CHECK_CONDITION
:
459 return scsi_check_sense(scmd
);
461 case INTERMEDIATE_GOOD
:
462 case INTERMEDIATE_C_GOOD
:
464 * who knows? FIXME(eric)
467 case RESERVATION_CONFLICT
:
468 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
469 /* it is a success, we probed the device and
472 /* otherwise, we failed to send the command */
475 scsi_handle_queue_full(scmd
->device
);
486 * scsi_eh_done - Completion function for error handling.
487 * @scmd: Cmd that is done.
489 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
491 struct completion
*eh_action
;
493 SCSI_LOG_ERROR_RECOVERY(3,
494 printk("%s scmd: %p result: %x\n",
495 __func__
, scmd
, scmd
->result
));
497 eh_action
= scmd
->device
->host
->eh_action
;
503 * scsi_try_host_reset - ask host adapter to reset itself
504 * @scmd: SCSI cmd to send hsot reset.
506 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
511 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
514 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
517 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
519 if (rtn
== SUCCESS
) {
520 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
521 ssleep(HOST_RESET_SETTLE_TIME
);
522 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
523 scsi_report_bus_reset(scmd
->device
->host
,
525 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
532 * scsi_try_bus_reset - ask host to perform a bus reset
533 * @scmd: SCSI cmd to send bus reset.
535 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
540 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
543 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
546 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
548 if (rtn
== SUCCESS
) {
549 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
550 ssleep(BUS_RESET_SETTLE_TIME
);
551 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
552 scsi_report_bus_reset(scmd
->device
->host
,
554 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
560 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
563 sdev
->expecting_cc_ua
= 1;
567 * scsi_try_target_reset - Ask host to perform a target reset
568 * @scmd: SCSI cmd used to send a target reset
571 * There is no timeout for this operation. if this operation is
572 * unreliable for a given host, then the host itself needs to put a
573 * timer on it, and set the host back to a consistent state prior to
576 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
581 if (!scmd
->device
->host
->hostt
->eh_target_reset_handler
)
584 rtn
= scmd
->device
->host
->hostt
->eh_target_reset_handler(scmd
);
585 if (rtn
== SUCCESS
) {
586 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
587 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
588 __scsi_report_device_reset
);
589 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
596 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
597 * @scmd: SCSI cmd used to send BDR
600 * There is no timeout for this operation. if this operation is
601 * unreliable for a given host, then the host itself needs to put a
602 * timer on it, and set the host back to a consistent state prior to
605 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
609 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
612 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
614 __scsi_report_device_reset(scmd
->device
, NULL
);
618 static int __scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
620 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
623 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
627 * scsi_try_to_abort_cmd - Ask host to abort a running command.
628 * @scmd: SCSI cmd to abort from Lower Level.
631 * This function will not return until the user's completion function
632 * has been called. there is no timeout on this operation. if the
633 * author of the low-level driver wishes this operation to be timed,
634 * they can provide this facility themselves. helper functions in
635 * scsi_error.c can be supplied to make this easier to do.
637 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
640 * scsi_done was called just after the command timed out and before
641 * we had a chance to process it. (db)
643 if (scmd
->serial_number
== 0)
645 return __scsi_try_to_abort_cmd(scmd
);
648 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
650 if (__scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
651 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
652 if (scsi_try_target_reset(scmd
) != SUCCESS
)
653 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
654 scsi_try_host_reset(scmd
);
658 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
659 * @scmd: SCSI command structure to hijack
660 * @ses: structure to save restore information
661 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
662 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
663 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
665 * This function is used to save a scsi command information before re-execution
666 * as part of the error recovery process. If @sense_bytes is 0 the command
667 * sent must be one that does not transfer any data. If @sense_bytes != 0
668 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
669 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
671 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
672 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
674 struct scsi_device
*sdev
= scmd
->device
;
677 * We need saved copies of a number of fields - this is because
678 * error handling may need to overwrite these with different values
679 * to run different commands, and once error handling is complete,
680 * we will need to restore these values prior to running the actual
683 ses
->cmd_len
= scmd
->cmd_len
;
684 ses
->cmnd
= scmd
->cmnd
;
685 ses
->data_direction
= scmd
->sc_data_direction
;
686 ses
->sdb
= scmd
->sdb
;
687 ses
->next_rq
= scmd
->request
->next_rq
;
688 ses
->result
= scmd
->result
;
689 ses
->underflow
= scmd
->underflow
;
690 ses
->prot_op
= scmd
->prot_op
;
692 scmd
->prot_op
= SCSI_PROT_NORMAL
;
693 scmd
->cmnd
= ses
->eh_cmnd
;
694 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
695 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
696 scmd
->request
->next_rq
= NULL
;
699 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
701 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
703 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
704 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
705 scmd
->sdb
.table
.nents
= 1;
706 scmd
->cmnd
[0] = REQUEST_SENSE
;
707 scmd
->cmnd
[4] = scmd
->sdb
.length
;
708 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
710 scmd
->sc_data_direction
= DMA_NONE
;
712 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
713 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
714 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
720 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
721 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
722 (sdev
->lun
<< 5 & 0xe0);
725 * Zero the sense buffer. The scsi spec mandates that any
726 * untransferred sense data should be interpreted as being zero.
728 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
730 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
733 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
734 * @scmd: SCSI command structure to restore
735 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
737 * Undo any damage done by above scsi_eh_prep_cmnd().
739 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
742 * Restore original data
744 scmd
->cmd_len
= ses
->cmd_len
;
745 scmd
->cmnd
= ses
->cmnd
;
746 scmd
->sc_data_direction
= ses
->data_direction
;
747 scmd
->sdb
= ses
->sdb
;
748 scmd
->request
->next_rq
= ses
->next_rq
;
749 scmd
->result
= ses
->result
;
750 scmd
->underflow
= ses
->underflow
;
751 scmd
->prot_op
= ses
->prot_op
;
753 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
756 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
757 * @scmd: SCSI command structure to hijack
759 * @cmnd_size: size in bytes of @cmnd
760 * @timeout: timeout for this request
761 * @sense_bytes: size of sense data to copy or 0
763 * This function is used to send a scsi command down to a target device
764 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
767 * SUCCESS or FAILED or NEEDS_RETRY
769 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
770 int cmnd_size
, int timeout
, unsigned sense_bytes
)
772 struct scsi_device
*sdev
= scmd
->device
;
773 struct Scsi_Host
*shost
= sdev
->host
;
774 DECLARE_COMPLETION_ONSTACK(done
);
775 unsigned long timeleft
;
776 struct scsi_eh_save ses
;
779 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
780 shost
->eh_action
= &done
;
783 scmd
->scsi_done
= scsi_eh_done
;
784 shost
->hostt
->queuecommand(shost
, scmd
);
786 timeleft
= wait_for_completion_timeout(&done
, timeout
);
788 shost
->eh_action
= NULL
;
790 scsi_log_completion(scmd
, SUCCESS
);
792 SCSI_LOG_ERROR_RECOVERY(3,
793 printk("%s: scmd: %p, timeleft: %ld\n",
794 __func__
, scmd
, timeleft
));
797 * If there is time left scsi_eh_done got called, and we will
798 * examine the actual status codes to see whether the command
799 * actually did complete normally, else tell the host to forget
800 * about this command.
803 rtn
= scsi_eh_completed_normally(scmd
);
804 SCSI_LOG_ERROR_RECOVERY(3,
805 printk("%s: scsi_eh_completed_normally %x\n",
821 scsi_abort_eh_cmnd(scmd
);
825 scsi_eh_restore_cmnd(scmd
, &ses
);
830 * scsi_request_sense - Request sense data from a particular target.
831 * @scmd: SCSI cmd for request sense.
834 * Some hosts automatically obtain this information, others require
835 * that we obtain it on our own. This function will *not* return until
836 * the command either times out, or it completes.
838 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
840 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
844 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
845 * @scmd: Original SCSI cmd that eh has finished.
846 * @done_q: Queue for processed commands.
849 * We don't want to use the normal command completion while we are are
850 * still handling errors - it may cause other commands to be queued,
851 * and that would disturb what we are doing. Thus we really want to
852 * keep a list of pending commands for final completion, and once we
853 * are ready to leave error handling we handle completion for real.
855 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
857 scmd
->device
->host
->host_failed
--;
859 list_move_tail(&scmd
->eh_entry
, done_q
);
861 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
864 * scsi_eh_get_sense - Get device sense data.
865 * @work_q: Queue of commands to process.
866 * @done_q: Queue of processed commands.
869 * See if we need to request sense information. if so, then get it
870 * now, so we have a better idea of what to do.
873 * This has the unfortunate side effect that if a shost adapter does
874 * not automatically request sense information, we end up shutting
875 * it down before we request it.
877 * All drivers should request sense information internally these days,
878 * so for now all I have to say is tough noogies if you end up in here.
880 * XXX: Long term this code should go away, but that needs an audit of
883 int scsi_eh_get_sense(struct list_head
*work_q
,
884 struct list_head
*done_q
)
886 struct scsi_cmnd
*scmd
, *next
;
889 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
890 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
891 SCSI_SENSE_VALID(scmd
))
894 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
895 "%s: requesting sense\n",
897 rtn
= scsi_request_sense(scmd
);
901 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
902 " result %x\n", scmd
,
904 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
906 rtn
= scsi_decide_disposition(scmd
);
909 * if the result was normal, then just pass it along to the
913 /* we don't want this command reissued, just
914 * finished with the sense data, so set
915 * retries to the max allowed to ensure it
916 * won't get reissued */
917 scmd
->retries
= scmd
->allowed
;
918 else if (rtn
!= NEEDS_RETRY
)
921 scsi_eh_finish_cmd(scmd
, done_q
);
924 return list_empty(work_q
);
926 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
929 * scsi_eh_tur - Send TUR to device.
930 * @scmd: &scsi_cmnd to send TUR
933 * 0 - Device is ready. 1 - Device NOT ready.
935 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
937 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
938 int retry_cnt
= 1, rtn
;
941 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
943 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
944 __func__
, scmd
, rtn
));
959 * scsi_eh_abort_cmds - abort pending commands.
960 * @work_q: &list_head for pending commands.
961 * @done_q: &list_head for processed commands.
964 * Try and see whether or not it makes sense to try and abort the
965 * running command. This only works out to be the case if we have one
966 * command that has timed out. If the command simply failed, it makes
967 * no sense to try and abort the command, since as far as the shost
968 * adapter is concerned, it isn't running.
970 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
971 struct list_head
*done_q
)
973 struct scsi_cmnd
*scmd
, *next
;
976 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
977 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
979 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
980 "0x%p\n", current
->comm
,
982 rtn
= scsi_try_to_abort_cmd(scmd
);
983 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
984 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
985 if (!scsi_device_online(scmd
->device
) ||
986 rtn
== FAST_IO_FAIL
||
987 !scsi_eh_tur(scmd
)) {
988 scsi_eh_finish_cmd(scmd
, done_q
);
992 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
999 return list_empty(work_q
);
1003 * scsi_eh_try_stu - Send START_UNIT to device.
1004 * @scmd: &scsi_cmnd to send START_UNIT
1007 * 0 - Device is ready. 1 - Device NOT ready.
1009 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1011 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1013 if (scmd
->device
->allow_restart
) {
1014 int i
, rtn
= NEEDS_RETRY
;
1016 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1017 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1027 * scsi_eh_stu - send START_UNIT if needed
1028 * @shost: &scsi host being recovered.
1029 * @work_q: &list_head for pending commands.
1030 * @done_q: &list_head for processed commands.
1033 * If commands are failing due to not ready, initializing command required,
1034 * try revalidating the device, which will end up sending a start unit.
1036 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1037 struct list_head
*work_q
,
1038 struct list_head
*done_q
)
1040 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1041 struct scsi_device
*sdev
;
1043 shost_for_each_device(sdev
, shost
) {
1045 list_for_each_entry(scmd
, work_q
, eh_entry
)
1046 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1047 scsi_check_sense(scmd
) == FAILED
) {
1055 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1056 " 0x%p\n", current
->comm
, sdev
));
1058 if (!scsi_eh_try_stu(stu_scmd
)) {
1059 if (!scsi_device_online(sdev
) ||
1060 !scsi_eh_tur(stu_scmd
)) {
1061 list_for_each_entry_safe(scmd
, next
,
1063 if (scmd
->device
== sdev
)
1064 scsi_eh_finish_cmd(scmd
, done_q
);
1068 SCSI_LOG_ERROR_RECOVERY(3,
1069 printk("%s: START_UNIT failed to sdev:"
1070 " 0x%p\n", current
->comm
, sdev
));
1074 return list_empty(work_q
);
1079 * scsi_eh_bus_device_reset - send bdr if needed
1080 * @shost: scsi host being recovered.
1081 * @work_q: &list_head for pending commands.
1082 * @done_q: &list_head for processed commands.
1085 * Try a bus device reset. Still, look to see whether we have multiple
1086 * devices that are jammed or not - if we have multiple devices, it
1087 * makes no sense to try bus_device_reset - we really would need to try
1088 * a bus_reset instead.
1090 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1091 struct list_head
*work_q
,
1092 struct list_head
*done_q
)
1094 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1095 struct scsi_device
*sdev
;
1098 shost_for_each_device(sdev
, shost
) {
1100 list_for_each_entry(scmd
, work_q
, eh_entry
)
1101 if (scmd
->device
== sdev
) {
1109 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1110 " 0x%p\n", current
->comm
,
1112 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1113 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1114 if (!scsi_device_online(sdev
) ||
1115 rtn
== FAST_IO_FAIL
||
1116 !scsi_eh_tur(bdr_scmd
)) {
1117 list_for_each_entry_safe(scmd
, next
,
1119 if (scmd
->device
== sdev
)
1120 scsi_eh_finish_cmd(scmd
,
1125 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1133 return list_empty(work_q
);
1137 * scsi_eh_target_reset - send target reset if needed
1138 * @shost: scsi host being recovered.
1139 * @work_q: &list_head for pending commands.
1140 * @done_q: &list_head for processed commands.
1143 * Try a target reset.
1145 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1146 struct list_head
*work_q
,
1147 struct list_head
*done_q
)
1149 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1150 unsigned int id
= 0;
1155 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1156 if (id
== scmd_id(scmd
)) {
1162 /* not one exactly equal; find the next highest */
1163 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1164 if (scmd_id(scmd
) > id
&&
1166 scmd_id(tgtr_scmd
) > scmd_id(scmd
)))
1171 /* no more commands, that's it */
1174 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1176 current
->comm
, id
));
1177 rtn
= scsi_try_target_reset(tgtr_scmd
);
1178 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1179 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1180 if (id
== scmd_id(scmd
))
1181 if (!scsi_device_online(scmd
->device
) ||
1182 rtn
== FAST_IO_FAIL
||
1183 !scsi_eh_tur(tgtr_scmd
))
1184 scsi_eh_finish_cmd(scmd
,
1188 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1191 current
->comm
, id
));
1195 return list_empty(work_q
);
1199 * scsi_eh_bus_reset - send a bus reset
1200 * @shost: &scsi host being recovered.
1201 * @work_q: &list_head for pending commands.
1202 * @done_q: &list_head for processed commands.
1204 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1205 struct list_head
*work_q
,
1206 struct list_head
*done_q
)
1208 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1209 unsigned int channel
;
1213 * we really want to loop over the various channels, and do this on
1214 * a channel by channel basis. we should also check to see if any
1215 * of the failed commands are on soft_reset devices, and if so, skip
1219 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1221 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1222 if (channel
== scmd_channel(scmd
)) {
1226 * FIXME add back in some support for
1227 * soft_reset devices.
1234 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1235 " %d\n", current
->comm
,
1237 rtn
= scsi_try_bus_reset(chan_scmd
);
1238 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1239 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1240 if (channel
== scmd_channel(scmd
))
1241 if (!scsi_device_online(scmd
->device
) ||
1242 rtn
== FAST_IO_FAIL
||
1244 scsi_eh_finish_cmd(scmd
,
1248 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1249 " failed chan: %d\n",
1254 return list_empty(work_q
);
1258 * scsi_eh_host_reset - send a host reset
1259 * @work_q: list_head for processed commands.
1260 * @done_q: list_head for processed commands.
1262 static int scsi_eh_host_reset(struct list_head
*work_q
,
1263 struct list_head
*done_q
)
1265 struct scsi_cmnd
*scmd
, *next
;
1268 if (!list_empty(work_q
)) {
1269 scmd
= list_entry(work_q
->next
,
1270 struct scsi_cmnd
, eh_entry
);
1272 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1275 rtn
= scsi_try_host_reset(scmd
);
1276 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1277 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1278 if (!scsi_device_online(scmd
->device
) ||
1279 rtn
== FAST_IO_FAIL
||
1280 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1282 scsi_eh_finish_cmd(scmd
, done_q
);
1285 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1290 return list_empty(work_q
);
1294 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1295 * @work_q: list_head for processed commands.
1296 * @done_q: list_head for processed commands.
1298 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1299 struct list_head
*done_q
)
1301 struct scsi_cmnd
*scmd
, *next
;
1303 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1304 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1305 "not ready after error recovery\n");
1306 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1307 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1309 * FIXME: Handle lost cmds.
1312 scsi_eh_finish_cmd(scmd
, done_q
);
1318 * scsi_noretry_cmd - determinte if command should be failed fast
1319 * @scmd: SCSI cmd to examine.
1321 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1323 switch (host_byte(scmd
->result
)) {
1327 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1329 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1331 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1332 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1335 case DID_SOFT_ERROR
:
1336 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1339 switch (status_byte(scmd
->result
)) {
1340 case CHECK_CONDITION
:
1342 * assume caller has checked sense and determinted
1343 * the check condition was retryable.
1345 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1346 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1354 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1355 * @scmd: SCSI cmd to examine.
1358 * This is *only* called when we are examining the status after sending
1359 * out the actual data command. any commands that are queued for error
1360 * recovery (e.g. test_unit_ready) do *not* come through here.
1362 * When this routine returns failed, it means the error handler thread
1363 * is woken. In cases where the error code indicates an error that
1364 * doesn't require the error handler read (i.e. we don't need to
1365 * abort/reset), this function should return SUCCESS.
1367 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1372 * if the device is offline, then we clearly just pass the result back
1373 * up to the top level.
1375 if (!scsi_device_online(scmd
->device
)) {
1376 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1383 * first check the host byte, to see if there is anything in there
1384 * that would indicate what we need to do.
1386 switch (host_byte(scmd
->result
)) {
1387 case DID_PASSTHROUGH
:
1389 * no matter what, pass this through to the upper layer.
1390 * nuke this special code so that it looks like we are saying
1393 scmd
->result
&= 0xff00ffff;
1397 * looks good. drop through, and check the next byte.
1400 case DID_NO_CONNECT
:
1401 case DID_BAD_TARGET
:
1404 * note - this means that we just report the status back
1405 * to the top level driver, not that we actually think
1406 * that it indicates SUCCESS.
1410 * when the low level driver returns did_soft_error,
1411 * it is responsible for keeping an internal retry counter
1412 * in order to avoid endless loops (db)
1414 * actually this is a bug in this function here. we should
1415 * be mindful of the maximum number of retries specified
1416 * and not get stuck in a loop.
1418 case DID_SOFT_ERROR
:
1424 return ADD_TO_MLQUEUE
;
1425 case DID_TRANSPORT_DISRUPTED
:
1427 * LLD/transport was disrupted during processing of the IO.
1428 * The transport class is now blocked/blocking,
1429 * and the transport will decide what to do with the IO
1430 * based on its timers and recovery capablilities if
1431 * there are enough retries.
1434 case DID_TRANSPORT_FAILFAST
:
1436 * The transport decided to failfast the IO (most likely
1437 * the fast io fail tmo fired), so send IO directly upwards.
1441 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1442 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1444 * execute reservation conflict processing code
1455 * when we scan the bus, we get timeout messages for
1456 * these commands if there is no device available.
1457 * other hosts report did_no_connect for the same thing.
1459 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1460 scmd
->cmnd
[0] == INQUIRY
)) {
1472 * next, check the message byte.
1474 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1478 * check the status byte to see if this indicates anything special.
1480 switch (status_byte(scmd
->result
)) {
1482 scsi_handle_queue_full(scmd
->device
);
1484 * the case of trying to send too many commands to a
1485 * tagged queueing device.
1489 * device can't talk to us at the moment. Should only
1490 * occur (SAM-3) when the task queue is empty, so will cause
1491 * the empty queue handling to trigger a stall in the
1494 return ADD_TO_MLQUEUE
;
1496 scsi_handle_queue_ramp_up(scmd
->device
);
1497 case COMMAND_TERMINATED
:
1501 case CHECK_CONDITION
:
1502 rtn
= scsi_check_sense(scmd
);
1503 if (rtn
== NEEDS_RETRY
)
1505 /* if rtn == FAILED, we have no sense information;
1506 * returning FAILED will wake the error handler thread
1507 * to collect the sense and redo the decide
1510 case CONDITION_GOOD
:
1511 case INTERMEDIATE_GOOD
:
1512 case INTERMEDIATE_C_GOOD
:
1515 * who knows? FIXME(eric)
1519 case RESERVATION_CONFLICT
:
1520 sdev_printk(KERN_INFO
, scmd
->device
,
1521 "reservation conflict\n");
1522 return SUCCESS
; /* causes immediate i/o error */
1530 /* we requeue for retry because the error was retryable, and
1531 * the request was not marked fast fail. Note that above,
1532 * even if the request is marked fast fail, we still requeue
1533 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1534 if ((++scmd
->retries
) <= scmd
->allowed
1535 && !scsi_noretry_cmd(scmd
)) {
1539 * no more retries - report this one back to upper level.
1545 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1547 __blk_put_request(req
->q
, req
);
1551 * scsi_eh_lock_door - Prevent medium removal for the specified device
1552 * @sdev: SCSI device to prevent medium removal
1555 * We must be called from process context.
1558 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1559 * head of the devices request queue, and continue.
1561 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1563 struct request
*req
;
1566 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1567 * request becomes available
1569 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1571 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1575 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1578 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1580 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1581 req
->cmd_flags
|= REQ_QUIET
;
1582 req
->timeout
= 10 * HZ
;
1585 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1589 * scsi_restart_operations - restart io operations to the specified host.
1590 * @shost: Host we are restarting.
1593 * When we entered the error handler, we blocked all further i/o to
1594 * this device. we need to 'reverse' this process.
1596 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1598 struct scsi_device
*sdev
;
1599 unsigned long flags
;
1602 * If the door was locked, we need to insert a door lock request
1603 * onto the head of the SCSI request queue for the device. There
1604 * is no point trying to lock the door of an off-line device.
1606 shost_for_each_device(sdev
, shost
) {
1607 if (scsi_device_online(sdev
) && sdev
->locked
)
1608 scsi_eh_lock_door(sdev
);
1612 * next free up anything directly waiting upon the host. this
1613 * will be requests for character device operations, and also for
1614 * ioctls to queued block devices.
1616 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1619 spin_lock_irqsave(shost
->host_lock
, flags
);
1620 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1621 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1622 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1623 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1625 wake_up(&shost
->host_wait
);
1628 * finally we need to re-initiate requests that may be pending. we will
1629 * have had everything blocked while error handling is taking place, and
1630 * now that error recovery is done, we will need to ensure that these
1631 * requests are started.
1633 scsi_run_host_queues(shost
);
1637 * scsi_eh_ready_devs - check device ready state and recover if not.
1638 * @shost: host to be recovered.
1639 * @work_q: &list_head for pending commands.
1640 * @done_q: &list_head for processed commands.
1642 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1643 struct list_head
*work_q
,
1644 struct list_head
*done_q
)
1646 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1647 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1648 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1649 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1650 if (!scsi_eh_host_reset(work_q
, done_q
))
1651 scsi_eh_offline_sdevs(work_q
,
1654 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1657 * scsi_eh_flush_done_q - finish processed commands or retry them.
1658 * @done_q: list_head of processed commands.
1660 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1662 struct scsi_cmnd
*scmd
, *next
;
1664 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1665 list_del_init(&scmd
->eh_entry
);
1666 if (scsi_device_online(scmd
->device
) &&
1667 !scsi_noretry_cmd(scmd
) &&
1668 (++scmd
->retries
<= scmd
->allowed
)) {
1669 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1673 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1676 * If just we got sense for the device (called
1677 * scsi_eh_get_sense), scmd->result is already
1678 * set, do not set DRIVER_TIMEOUT.
1681 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1682 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1684 current
->comm
, scmd
));
1685 scsi_finish_command(scmd
);
1689 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1692 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1693 * @shost: Host to unjam.
1696 * When we come in here, we *know* that all commands on the bus have
1697 * either completed, failed or timed out. we also know that no further
1698 * commands are being sent to the host, so things are relatively quiet
1699 * and we have freedom to fiddle with things as we wish.
1701 * This is only the *default* implementation. it is possible for
1702 * individual drivers to supply their own version of this function, and
1703 * if the maintainer wishes to do this, it is strongly suggested that
1704 * this function be taken as a template and modified. this function
1705 * was designed to correctly handle problems for about 95% of the
1706 * different cases out there, and it should always provide at least a
1707 * reasonable amount of error recovery.
1709 * Any command marked 'failed' or 'timeout' must eventually have
1710 * scsi_finish_cmd() called for it. we do all of the retry stuff
1711 * here, so when we restart the host after we return it should have an
1714 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1716 unsigned long flags
;
1717 LIST_HEAD(eh_work_q
);
1718 LIST_HEAD(eh_done_q
);
1720 spin_lock_irqsave(shost
->host_lock
, flags
);
1721 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1722 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1724 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1726 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1727 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1728 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1730 scsi_eh_flush_done_q(&eh_done_q
);
1734 * scsi_error_handler - SCSI error handler thread
1735 * @data: Host for which we are running.
1738 * This is the main error handling loop. This is run as a kernel thread
1739 * for every SCSI host and handles all error handling activity.
1741 int scsi_error_handler(void *data
)
1743 struct Scsi_Host
*shost
= data
;
1746 * We use TASK_INTERRUPTIBLE so that the thread is not
1747 * counted against the load average as a running process.
1748 * We never actually get interrupted because kthread_run
1749 * disables signal delivery for the created thread.
1751 set_current_state(TASK_INTERRUPTIBLE
);
1752 while (!kthread_should_stop()) {
1753 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1754 shost
->host_failed
!= shost
->host_busy
) {
1755 SCSI_LOG_ERROR_RECOVERY(1,
1756 printk("Error handler scsi_eh_%d sleeping\n",
1759 set_current_state(TASK_INTERRUPTIBLE
);
1763 __set_current_state(TASK_RUNNING
);
1764 SCSI_LOG_ERROR_RECOVERY(1,
1765 printk("Error handler scsi_eh_%d waking up\n",
1769 * We have a host that is failing for some reason. Figure out
1770 * what we need to do to get it up and online again (if we can).
1771 * If we fail, we end up taking the thing offline.
1773 if (scsi_autopm_get_host(shost
) != 0) {
1774 SCSI_LOG_ERROR_RECOVERY(1,
1775 printk(KERN_ERR
"Error handler scsi_eh_%d "
1776 "unable to autoresume\n",
1781 if (shost
->transportt
->eh_strategy_handler
)
1782 shost
->transportt
->eh_strategy_handler(shost
);
1784 scsi_unjam_host(shost
);
1787 * Note - if the above fails completely, the action is to take
1788 * individual devices offline and flush the queue of any
1789 * outstanding requests that may have been pending. When we
1790 * restart, we restart any I/O to any other devices on the bus
1791 * which are still online.
1793 scsi_restart_operations(shost
);
1794 scsi_autopm_put_host(shost
);
1795 set_current_state(TASK_INTERRUPTIBLE
);
1797 __set_current_state(TASK_RUNNING
);
1799 SCSI_LOG_ERROR_RECOVERY(1,
1800 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1801 shost
->ehandler
= NULL
;
1806 * Function: scsi_report_bus_reset()
1808 * Purpose: Utility function used by low-level drivers to report that
1809 * they have observed a bus reset on the bus being handled.
1811 * Arguments: shost - Host in question
1812 * channel - channel on which reset was observed.
1816 * Lock status: Host lock must be held.
1818 * Notes: This only needs to be called if the reset is one which
1819 * originates from an unknown location. Resets originated
1820 * by the mid-level itself don't need to call this, but there
1821 * should be no harm.
1823 * The main purpose of this is to make sure that a CHECK_CONDITION
1824 * is properly treated.
1826 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1828 struct scsi_device
*sdev
;
1830 __shost_for_each_device(sdev
, shost
) {
1831 if (channel
== sdev_channel(sdev
))
1832 __scsi_report_device_reset(sdev
, NULL
);
1835 EXPORT_SYMBOL(scsi_report_bus_reset
);
1838 * Function: scsi_report_device_reset()
1840 * Purpose: Utility function used by low-level drivers to report that
1841 * they have observed a device reset on the device being handled.
1843 * Arguments: shost - Host in question
1844 * channel - channel on which reset was observed
1845 * target - target on which reset was observed
1849 * Lock status: Host lock must be held
1851 * Notes: This only needs to be called if the reset is one which
1852 * originates from an unknown location. Resets originated
1853 * by the mid-level itself don't need to call this, but there
1854 * should be no harm.
1856 * The main purpose of this is to make sure that a CHECK_CONDITION
1857 * is properly treated.
1859 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1861 struct scsi_device
*sdev
;
1863 __shost_for_each_device(sdev
, shost
) {
1864 if (channel
== sdev_channel(sdev
) &&
1865 target
== sdev_id(sdev
))
1866 __scsi_report_device_reset(sdev
, NULL
);
1869 EXPORT_SYMBOL(scsi_report_device_reset
);
1872 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1877 * Function: scsi_reset_provider
1879 * Purpose: Send requested reset to a bus or device at any phase.
1881 * Arguments: device - device to send reset to
1882 * flag - reset type (see scsi.h)
1884 * Returns: SUCCESS/FAILURE.
1886 * Notes: This is used by the SCSI Generic driver to provide
1887 * Bus/Device reset capability.
1890 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1892 struct scsi_cmnd
*scmd
;
1893 struct Scsi_Host
*shost
= dev
->host
;
1895 unsigned long flags
;
1898 if (scsi_autopm_get_host(shost
) < 0)
1901 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1902 blk_rq_init(NULL
, &req
);
1903 scmd
->request
= &req
;
1905 scmd
->cmnd
= req
.cmd
;
1907 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1908 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1912 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1914 spin_lock_irqsave(shost
->host_lock
, flags
);
1915 shost
->tmf_in_progress
= 1;
1916 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1919 case SCSI_TRY_RESET_DEVICE
:
1920 rtn
= scsi_try_bus_device_reset(scmd
);
1924 case SCSI_TRY_RESET_TARGET
:
1925 rtn
= scsi_try_target_reset(scmd
);
1929 case SCSI_TRY_RESET_BUS
:
1930 rtn
= scsi_try_bus_reset(scmd
);
1934 case SCSI_TRY_RESET_HOST
:
1935 rtn
= scsi_try_host_reset(scmd
);
1941 spin_lock_irqsave(shost
->host_lock
, flags
);
1942 shost
->tmf_in_progress
= 0;
1943 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1946 * be sure to wake up anyone who was sleeping or had their queue
1947 * suspended while we performed the TMF.
1949 SCSI_LOG_ERROR_RECOVERY(3,
1950 printk("%s: waking up host to restart after TMF\n",
1953 wake_up(&shost
->host_wait
);
1955 scsi_run_host_queues(shost
);
1957 scsi_next_command(scmd
);
1958 scsi_autopm_put_host(shost
);
1961 EXPORT_SYMBOL(scsi_reset_provider
);
1964 * scsi_normalize_sense - normalize main elements from either fixed or
1965 * descriptor sense data format into a common format.
1967 * @sense_buffer: byte array containing sense data returned by device
1968 * @sb_len: number of valid bytes in sense_buffer
1969 * @sshdr: pointer to instance of structure that common
1970 * elements are written to.
1973 * The "main elements" from sense data are: response_code, sense_key,
1974 * asc, ascq and additional_length (only for descriptor format).
1976 * Typically this function can be called after a device has
1977 * responded to a SCSI command with the CHECK_CONDITION status.
1980 * 1 if valid sense data information found, else 0;
1982 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1983 struct scsi_sense_hdr
*sshdr
)
1985 if (!sense_buffer
|| !sb_len
)
1988 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1990 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1992 if (!scsi_sense_valid(sshdr
))
1995 if (sshdr
->response_code
>= 0x72) {
2000 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
2002 sshdr
->asc
= sense_buffer
[2];
2004 sshdr
->ascq
= sense_buffer
[3];
2006 sshdr
->additional_length
= sense_buffer
[7];
2012 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2014 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2015 sb_len
: (sense_buffer
[7] + 8);
2017 sshdr
->asc
= sense_buffer
[12];
2019 sshdr
->ascq
= sense_buffer
[13];
2025 EXPORT_SYMBOL(scsi_normalize_sense
);
2027 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2028 struct scsi_sense_hdr
*sshdr
)
2030 return scsi_normalize_sense(cmd
->sense_buffer
,
2031 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2033 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2036 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2037 * @sense_buffer: byte array of descriptor format sense data
2038 * @sb_len: number of valid bytes in sense_buffer
2039 * @desc_type: value of descriptor type to find
2040 * (e.g. 0 -> information)
2043 * only valid when sense data is in descriptor format
2046 * pointer to start of (first) descriptor if found else NULL
2048 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2051 int add_sen_len
, add_len
, desc_len
, k
;
2054 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2056 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2058 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2059 add_sen_len
: (sb_len
- 8);
2060 descp
= &sense_buffer
[8];
2061 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2063 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2064 desc_len
= add_len
+ 2;
2065 if (descp
[0] == desc_type
)
2067 if (add_len
< 0) // short descriptor ??
2072 EXPORT_SYMBOL(scsi_sense_desc_find
);
2075 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2076 * @sense_buffer: byte array of sense data
2077 * @sb_len: number of valid bytes in sense_buffer
2078 * @info_out: pointer to 64 integer where 8 or 4 byte information
2079 * field will be placed if found.
2082 * 1 if information field found, 0 if not found.
2084 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2093 switch (sense_buffer
[0] & 0x7f) {
2096 if (sense_buffer
[0] & 0x80) {
2097 *info_out
= (sense_buffer
[3] << 24) +
2098 (sense_buffer
[4] << 16) +
2099 (sense_buffer
[5] << 8) + sense_buffer
[6];
2105 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2107 if (ucp
&& (0xa == ucp
[1])) {
2109 for (j
= 0; j
< 8; ++j
) {
2122 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2125 * scsi_build_sense_buffer - build sense data in a buffer
2126 * @desc: Sense format (non zero == descriptor format,
2127 * 0 == fixed format)
2128 * @buf: Where to build sense data
2130 * @asc: Additional sense code
2131 * @ascq: Additional sense code qualifier
2134 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2137 buf
[0] = 0x72; /* descriptor, current */
2143 buf
[0] = 0x70; /* fixed, current */
2150 EXPORT_SYMBOL(scsi_build_sense_buffer
);