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 (blk_barrier_rq(scmd
->request
))
312 * barrier requests should always retry on UA
313 * otherwise block will get a spurious error
318 * for normal (non barrier) commands, pass the
319 * UA upwards for a determination in the
320 * completion functions
324 /* these three are not supported */
326 case VOLUME_OVERFLOW
:
331 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
332 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
333 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
339 if (scmd
->device
->retry_hwerror
)
340 return ADD_TO_MLQUEUE
;
344 case ILLEGAL_REQUEST
:
352 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
354 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
355 struct scsi_device
*tmp_sdev
;
357 if (!sht
->change_queue_depth
||
358 sdev
->queue_depth
>= sdev
->max_queue_depth
)
361 if (time_before(jiffies
,
362 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
365 if (time_before(jiffies
,
366 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
370 * Walk all devices of a target and do
373 shost_for_each_device(tmp_sdev
, sdev
->host
) {
374 if (tmp_sdev
->channel
!= sdev
->channel
||
375 tmp_sdev
->id
!= sdev
->id
||
376 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
379 * call back into LLD to increase queue_depth by one
380 * with ramp up reason code.
382 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
383 SCSI_QDEPTH_RAMP_UP
);
384 sdev
->last_queue_ramp_up
= jiffies
;
388 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
390 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
391 struct scsi_device
*tmp_sdev
;
393 if (!sht
->change_queue_depth
)
396 shost_for_each_device(tmp_sdev
, sdev
->host
) {
397 if (tmp_sdev
->channel
!= sdev
->channel
||
398 tmp_sdev
->id
!= sdev
->id
)
401 * We do not know the number of commands that were at
402 * the device when we got the queue full so we start
403 * from the highest possible value and work our way down.
405 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
411 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
412 * @scmd: SCSI cmd to examine.
415 * This is *only* called when we are examining the status of commands
416 * queued during error recovery. the main difference here is that we
417 * don't allow for the possibility of retries here, and we are a lot
418 * more restrictive about what we consider acceptable.
420 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
423 * first check the host byte, to see if there is anything in there
424 * that would indicate what we need to do.
426 if (host_byte(scmd
->result
) == DID_RESET
) {
428 * rats. we are already in the error handler, so we now
429 * get to try and figure out what to do next. if the sense
430 * is valid, we have a pretty good idea of what to do.
431 * if not, we mark it as FAILED.
433 return scsi_check_sense(scmd
);
435 if (host_byte(scmd
->result
) != DID_OK
)
439 * next, check the message byte.
441 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
445 * now, check the status byte to see if this indicates
448 switch (status_byte(scmd
->result
)) {
450 scsi_handle_queue_ramp_up(scmd
->device
);
451 case COMMAND_TERMINATED
:
453 case CHECK_CONDITION
:
454 return scsi_check_sense(scmd
);
456 case INTERMEDIATE_GOOD
:
457 case INTERMEDIATE_C_GOOD
:
459 * who knows? FIXME(eric)
462 case RESERVATION_CONFLICT
:
464 * let issuer deal with this, it could be just fine
468 scsi_handle_queue_full(scmd
->device
);
478 * scsi_eh_done - Completion function for error handling.
479 * @scmd: Cmd that is done.
481 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
483 struct completion
*eh_action
;
485 SCSI_LOG_ERROR_RECOVERY(3,
486 printk("%s scmd: %p result: %x\n",
487 __func__
, scmd
, scmd
->result
));
489 eh_action
= scmd
->device
->host
->eh_action
;
495 * scsi_try_host_reset - ask host adapter to reset itself
496 * @scmd: SCSI cmd to send hsot reset.
498 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
503 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
506 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
509 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
511 if (rtn
== SUCCESS
) {
512 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
513 ssleep(HOST_RESET_SETTLE_TIME
);
514 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
515 scsi_report_bus_reset(scmd
->device
->host
,
517 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
524 * scsi_try_bus_reset - ask host to perform a bus reset
525 * @scmd: SCSI cmd to send bus reset.
527 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
532 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
535 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
538 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
540 if (rtn
== SUCCESS
) {
541 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
542 ssleep(BUS_RESET_SETTLE_TIME
);
543 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
544 scsi_report_bus_reset(scmd
->device
->host
,
546 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
552 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
555 sdev
->expecting_cc_ua
= 1;
559 * scsi_try_target_reset - Ask host to perform a target reset
560 * @scmd: SCSI cmd used to send a target reset
563 * There is no timeout for this operation. if this operation is
564 * unreliable for a given host, then the host itself needs to put a
565 * timer on it, and set the host back to a consistent state prior to
568 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
573 if (!scmd
->device
->host
->hostt
->eh_target_reset_handler
)
576 rtn
= scmd
->device
->host
->hostt
->eh_target_reset_handler(scmd
);
577 if (rtn
== SUCCESS
) {
578 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
579 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
580 __scsi_report_device_reset
);
581 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
588 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
589 * @scmd: SCSI cmd used to send BDR
592 * There is no timeout for this operation. if this operation is
593 * unreliable for a given host, then the host itself needs to put a
594 * timer on it, and set the host back to a consistent state prior to
597 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
601 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
604 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
606 __scsi_report_device_reset(scmd
->device
, NULL
);
610 static int __scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
612 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
615 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
619 * scsi_try_to_abort_cmd - Ask host to abort a running command.
620 * @scmd: SCSI cmd to abort from Lower Level.
623 * This function will not return until the user's completion function
624 * has been called. there is no timeout on this operation. if the
625 * author of the low-level driver wishes this operation to be timed,
626 * they can provide this facility themselves. helper functions in
627 * scsi_error.c can be supplied to make this easier to do.
629 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
632 * scsi_done was called just after the command timed out and before
633 * we had a chance to process it. (db)
635 if (scmd
->serial_number
== 0)
637 return __scsi_try_to_abort_cmd(scmd
);
640 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
642 if (__scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
643 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
644 if (scsi_try_target_reset(scmd
) != SUCCESS
)
645 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
646 scsi_try_host_reset(scmd
);
650 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
651 * @scmd: SCSI command structure to hijack
652 * @ses: structure to save restore information
653 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
654 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
655 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
657 * This function is used to save a scsi command information before re-execution
658 * as part of the error recovery process. If @sense_bytes is 0 the command
659 * sent must be one that does not transfer any data. If @sense_bytes != 0
660 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
661 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
663 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
664 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
666 struct scsi_device
*sdev
= scmd
->device
;
669 * We need saved copies of a number of fields - this is because
670 * error handling may need to overwrite these with different values
671 * to run different commands, and once error handling is complete,
672 * we will need to restore these values prior to running the actual
675 ses
->cmd_len
= scmd
->cmd_len
;
676 ses
->cmnd
= scmd
->cmnd
;
677 ses
->data_direction
= scmd
->sc_data_direction
;
678 ses
->sdb
= scmd
->sdb
;
679 ses
->next_rq
= scmd
->request
->next_rq
;
680 ses
->result
= scmd
->result
;
681 ses
->underflow
= scmd
->underflow
;
682 ses
->prot_op
= scmd
->prot_op
;
684 scmd
->prot_op
= SCSI_PROT_NORMAL
;
685 scmd
->cmnd
= ses
->eh_cmnd
;
686 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
687 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
688 scmd
->request
->next_rq
= NULL
;
691 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
693 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
695 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
696 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
697 scmd
->sdb
.table
.nents
= 1;
698 scmd
->cmnd
[0] = REQUEST_SENSE
;
699 scmd
->cmnd
[4] = scmd
->sdb
.length
;
700 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
702 scmd
->sc_data_direction
= DMA_NONE
;
704 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
705 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
706 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
712 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
713 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
714 (sdev
->lun
<< 5 & 0xe0);
717 * Zero the sense buffer. The scsi spec mandates that any
718 * untransferred sense data should be interpreted as being zero.
720 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
722 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
725 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
726 * @scmd: SCSI command structure to restore
727 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
729 * Undo any damage done by above scsi_eh_prep_cmnd().
731 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
734 * Restore original data
736 scmd
->cmd_len
= ses
->cmd_len
;
737 scmd
->cmnd
= ses
->cmnd
;
738 scmd
->sc_data_direction
= ses
->data_direction
;
739 scmd
->sdb
= ses
->sdb
;
740 scmd
->request
->next_rq
= ses
->next_rq
;
741 scmd
->result
= ses
->result
;
742 scmd
->underflow
= ses
->underflow
;
743 scmd
->prot_op
= ses
->prot_op
;
745 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
748 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
749 * @scmd: SCSI command structure to hijack
751 * @cmnd_size: size in bytes of @cmnd
752 * @timeout: timeout for this request
753 * @sense_bytes: size of sense data to copy or 0
755 * This function is used to send a scsi command down to a target device
756 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
759 * SUCCESS or FAILED or NEEDS_RETRY
761 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
762 int cmnd_size
, int timeout
, unsigned sense_bytes
)
764 struct scsi_device
*sdev
= scmd
->device
;
765 struct Scsi_Host
*shost
= sdev
->host
;
766 DECLARE_COMPLETION_ONSTACK(done
);
767 unsigned long timeleft
;
769 struct scsi_eh_save ses
;
772 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
773 shost
->eh_action
= &done
;
775 spin_lock_irqsave(shost
->host_lock
, flags
);
777 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
778 spin_unlock_irqrestore(shost
->host_lock
, flags
);
780 timeleft
= wait_for_completion_timeout(&done
, timeout
);
782 shost
->eh_action
= NULL
;
784 scsi_log_completion(scmd
, SUCCESS
);
786 SCSI_LOG_ERROR_RECOVERY(3,
787 printk("%s: scmd: %p, timeleft: %ld\n",
788 __func__
, scmd
, timeleft
));
791 * If there is time left scsi_eh_done got called, and we will
792 * examine the actual status codes to see whether the command
793 * actually did complete normally, else tell the host to forget
794 * about this command.
797 rtn
= scsi_eh_completed_normally(scmd
);
798 SCSI_LOG_ERROR_RECOVERY(3,
799 printk("%s: scsi_eh_completed_normally %x\n",
815 scsi_abort_eh_cmnd(scmd
);
819 scsi_eh_restore_cmnd(scmd
, &ses
);
824 * scsi_request_sense - Request sense data from a particular target.
825 * @scmd: SCSI cmd for request sense.
828 * Some hosts automatically obtain this information, others require
829 * that we obtain it on our own. This function will *not* return until
830 * the command either times out, or it completes.
832 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
834 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
838 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
839 * @scmd: Original SCSI cmd that eh has finished.
840 * @done_q: Queue for processed commands.
843 * We don't want to use the normal command completion while we are are
844 * still handling errors - it may cause other commands to be queued,
845 * and that would disturb what we are doing. Thus we really want to
846 * keep a list of pending commands for final completion, and once we
847 * are ready to leave error handling we handle completion for real.
849 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
851 scmd
->device
->host
->host_failed
--;
853 list_move_tail(&scmd
->eh_entry
, done_q
);
855 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
858 * scsi_eh_get_sense - Get device sense data.
859 * @work_q: Queue of commands to process.
860 * @done_q: Queue of processed commands.
863 * See if we need to request sense information. if so, then get it
864 * now, so we have a better idea of what to do.
867 * This has the unfortunate side effect that if a shost adapter does
868 * not automatically request sense information, we end up shutting
869 * it down before we request it.
871 * All drivers should request sense information internally these days,
872 * so for now all I have to say is tough noogies if you end up in here.
874 * XXX: Long term this code should go away, but that needs an audit of
877 int scsi_eh_get_sense(struct list_head
*work_q
,
878 struct list_head
*done_q
)
880 struct scsi_cmnd
*scmd
, *next
;
883 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
884 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
885 SCSI_SENSE_VALID(scmd
))
888 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
889 "%s: requesting sense\n",
891 rtn
= scsi_request_sense(scmd
);
895 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
896 " result %x\n", scmd
,
898 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
900 rtn
= scsi_decide_disposition(scmd
);
903 * if the result was normal, then just pass it along to the
907 /* we don't want this command reissued, just
908 * finished with the sense data, so set
909 * retries to the max allowed to ensure it
910 * won't get reissued */
911 scmd
->retries
= scmd
->allowed
;
912 else if (rtn
!= NEEDS_RETRY
)
915 scsi_eh_finish_cmd(scmd
, done_q
);
918 return list_empty(work_q
);
920 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
923 * scsi_eh_tur - Send TUR to device.
924 * @scmd: &scsi_cmnd to send TUR
927 * 0 - Device is ready. 1 - Device NOT ready.
929 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
931 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
932 int retry_cnt
= 1, rtn
;
935 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
937 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
938 __func__
, scmd
, rtn
));
953 * scsi_eh_abort_cmds - abort pending commands.
954 * @work_q: &list_head for pending commands.
955 * @done_q: &list_head for processed commands.
958 * Try and see whether or not it makes sense to try and abort the
959 * running command. This only works out to be the case if we have one
960 * command that has timed out. If the command simply failed, it makes
961 * no sense to try and abort the command, since as far as the shost
962 * adapter is concerned, it isn't running.
964 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
965 struct list_head
*done_q
)
967 struct scsi_cmnd
*scmd
, *next
;
970 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
971 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
973 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
974 "0x%p\n", current
->comm
,
976 rtn
= scsi_try_to_abort_cmd(scmd
);
977 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
978 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
979 if (!scsi_device_online(scmd
->device
) ||
980 rtn
== FAST_IO_FAIL
||
981 !scsi_eh_tur(scmd
)) {
982 scsi_eh_finish_cmd(scmd
, done_q
);
986 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
993 return list_empty(work_q
);
997 * scsi_eh_try_stu - Send START_UNIT to device.
998 * @scmd: &scsi_cmnd to send START_UNIT
1001 * 0 - Device is ready. 1 - Device NOT ready.
1003 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1005 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1007 if (scmd
->device
->allow_restart
) {
1008 int i
, rtn
= NEEDS_RETRY
;
1010 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1011 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1021 * scsi_eh_stu - send START_UNIT if needed
1022 * @shost: &scsi host being recovered.
1023 * @work_q: &list_head for pending commands.
1024 * @done_q: &list_head for processed commands.
1027 * If commands are failing due to not ready, initializing command required,
1028 * try revalidating the device, which will end up sending a start unit.
1030 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1031 struct list_head
*work_q
,
1032 struct list_head
*done_q
)
1034 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1035 struct scsi_device
*sdev
;
1037 shost_for_each_device(sdev
, shost
) {
1039 list_for_each_entry(scmd
, work_q
, eh_entry
)
1040 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1041 scsi_check_sense(scmd
) == FAILED
) {
1049 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1050 " 0x%p\n", current
->comm
, sdev
));
1052 if (!scsi_eh_try_stu(stu_scmd
)) {
1053 if (!scsi_device_online(sdev
) ||
1054 !scsi_eh_tur(stu_scmd
)) {
1055 list_for_each_entry_safe(scmd
, next
,
1057 if (scmd
->device
== sdev
)
1058 scsi_eh_finish_cmd(scmd
, done_q
);
1062 SCSI_LOG_ERROR_RECOVERY(3,
1063 printk("%s: START_UNIT failed to sdev:"
1064 " 0x%p\n", current
->comm
, sdev
));
1068 return list_empty(work_q
);
1073 * scsi_eh_bus_device_reset - send bdr if needed
1074 * @shost: scsi host being recovered.
1075 * @work_q: &list_head for pending commands.
1076 * @done_q: &list_head for processed commands.
1079 * Try a bus device reset. Still, look to see whether we have multiple
1080 * devices that are jammed or not - if we have multiple devices, it
1081 * makes no sense to try bus_device_reset - we really would need to try
1082 * a bus_reset instead.
1084 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1085 struct list_head
*work_q
,
1086 struct list_head
*done_q
)
1088 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1089 struct scsi_device
*sdev
;
1092 shost_for_each_device(sdev
, shost
) {
1094 list_for_each_entry(scmd
, work_q
, eh_entry
)
1095 if (scmd
->device
== sdev
) {
1103 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1104 " 0x%p\n", current
->comm
,
1106 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1107 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1108 if (!scsi_device_online(sdev
) ||
1109 rtn
== FAST_IO_FAIL
||
1110 !scsi_eh_tur(bdr_scmd
)) {
1111 list_for_each_entry_safe(scmd
, next
,
1113 if (scmd
->device
== sdev
)
1114 scsi_eh_finish_cmd(scmd
,
1119 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1127 return list_empty(work_q
);
1131 * scsi_eh_target_reset - send target reset if needed
1132 * @shost: scsi host being recovered.
1133 * @work_q: &list_head for pending commands.
1134 * @done_q: &list_head for processed commands.
1137 * Try a target reset.
1139 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1140 struct list_head
*work_q
,
1141 struct list_head
*done_q
)
1143 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1144 unsigned int id
= 0;
1149 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1150 if (id
== scmd_id(scmd
)) {
1156 /* not one exactly equal; find the next highest */
1157 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1158 if (scmd_id(scmd
) > id
&&
1160 scmd_id(tgtr_scmd
) > scmd_id(scmd
)))
1165 /* no more commands, that's it */
1168 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1170 current
->comm
, id
));
1171 rtn
= scsi_try_target_reset(tgtr_scmd
);
1172 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1173 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1174 if (id
== scmd_id(scmd
))
1175 if (!scsi_device_online(scmd
->device
) ||
1176 rtn
== FAST_IO_FAIL
||
1177 !scsi_eh_tur(tgtr_scmd
))
1178 scsi_eh_finish_cmd(scmd
,
1182 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1185 current
->comm
, id
));
1189 return list_empty(work_q
);
1193 * scsi_eh_bus_reset - send a bus reset
1194 * @shost: &scsi host being recovered.
1195 * @work_q: &list_head for pending commands.
1196 * @done_q: &list_head for processed commands.
1198 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1199 struct list_head
*work_q
,
1200 struct list_head
*done_q
)
1202 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1203 unsigned int channel
;
1207 * we really want to loop over the various channels, and do this on
1208 * a channel by channel basis. we should also check to see if any
1209 * of the failed commands are on soft_reset devices, and if so, skip
1213 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1215 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1216 if (channel
== scmd_channel(scmd
)) {
1220 * FIXME add back in some support for
1221 * soft_reset devices.
1228 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1229 " %d\n", current
->comm
,
1231 rtn
= scsi_try_bus_reset(chan_scmd
);
1232 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1233 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1234 if (channel
== scmd_channel(scmd
))
1235 if (!scsi_device_online(scmd
->device
) ||
1236 rtn
== FAST_IO_FAIL
||
1238 scsi_eh_finish_cmd(scmd
,
1242 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1243 " failed chan: %d\n",
1248 return list_empty(work_q
);
1252 * scsi_eh_host_reset - send a host reset
1253 * @work_q: list_head for processed commands.
1254 * @done_q: list_head for processed commands.
1256 static int scsi_eh_host_reset(struct list_head
*work_q
,
1257 struct list_head
*done_q
)
1259 struct scsi_cmnd
*scmd
, *next
;
1262 if (!list_empty(work_q
)) {
1263 scmd
= list_entry(work_q
->next
,
1264 struct scsi_cmnd
, eh_entry
);
1266 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1269 rtn
= scsi_try_host_reset(scmd
);
1270 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1271 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1272 if (!scsi_device_online(scmd
->device
) ||
1273 rtn
== FAST_IO_FAIL
||
1274 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1276 scsi_eh_finish_cmd(scmd
, done_q
);
1279 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1284 return list_empty(work_q
);
1288 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1289 * @work_q: list_head for processed commands.
1290 * @done_q: list_head for processed commands.
1292 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1293 struct list_head
*done_q
)
1295 struct scsi_cmnd
*scmd
, *next
;
1297 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1298 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1299 "not ready after error recovery\n");
1300 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1301 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1303 * FIXME: Handle lost cmds.
1306 scsi_eh_finish_cmd(scmd
, done_q
);
1312 * scsi_noretry_cmd - determinte if command should be failed fast
1313 * @scmd: SCSI cmd to examine.
1315 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1317 switch (host_byte(scmd
->result
)) {
1321 return blk_failfast_transport(scmd
->request
);
1323 return blk_failfast_dev(scmd
->request
);
1325 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1326 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1329 case DID_SOFT_ERROR
:
1330 return blk_failfast_driver(scmd
->request
);
1333 switch (status_byte(scmd
->result
)) {
1334 case CHECK_CONDITION
:
1336 * assume caller has checked sense and determinted
1337 * the check condition was retryable.
1339 return blk_failfast_dev(scmd
->request
);
1346 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1347 * @scmd: SCSI cmd to examine.
1350 * This is *only* called when we are examining the status after sending
1351 * out the actual data command. any commands that are queued for error
1352 * recovery (e.g. test_unit_ready) do *not* come through here.
1354 * When this routine returns failed, it means the error handler thread
1355 * is woken. In cases where the error code indicates an error that
1356 * doesn't require the error handler read (i.e. we don't need to
1357 * abort/reset), this function should return SUCCESS.
1359 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1364 * if the device is offline, then we clearly just pass the result back
1365 * up to the top level.
1367 if (!scsi_device_online(scmd
->device
)) {
1368 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1375 * first check the host byte, to see if there is anything in there
1376 * that would indicate what we need to do.
1378 switch (host_byte(scmd
->result
)) {
1379 case DID_PASSTHROUGH
:
1381 * no matter what, pass this through to the upper layer.
1382 * nuke this special code so that it looks like we are saying
1385 scmd
->result
&= 0xff00ffff;
1389 * looks good. drop through, and check the next byte.
1392 case DID_NO_CONNECT
:
1393 case DID_BAD_TARGET
:
1396 * note - this means that we just report the status back
1397 * to the top level driver, not that we actually think
1398 * that it indicates SUCCESS.
1402 * when the low level driver returns did_soft_error,
1403 * it is responsible for keeping an internal retry counter
1404 * in order to avoid endless loops (db)
1406 * actually this is a bug in this function here. we should
1407 * be mindful of the maximum number of retries specified
1408 * and not get stuck in a loop.
1410 case DID_SOFT_ERROR
:
1416 return ADD_TO_MLQUEUE
;
1417 case DID_TRANSPORT_DISRUPTED
:
1419 * LLD/transport was disrupted during processing of the IO.
1420 * The transport class is now blocked/blocking,
1421 * and the transport will decide what to do with the IO
1422 * based on its timers and recovery capablilities if
1423 * there are enough retries.
1426 case DID_TRANSPORT_FAILFAST
:
1428 * The transport decided to failfast the IO (most likely
1429 * the fast io fail tmo fired), so send IO directly upwards.
1433 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1434 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1436 * execute reservation conflict processing code
1447 * when we scan the bus, we get timeout messages for
1448 * these commands if there is no device available.
1449 * other hosts report did_no_connect for the same thing.
1451 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1452 scmd
->cmnd
[0] == INQUIRY
)) {
1464 * next, check the message byte.
1466 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1470 * check the status byte to see if this indicates anything special.
1472 switch (status_byte(scmd
->result
)) {
1474 scsi_handle_queue_full(scmd
->device
);
1476 * the case of trying to send too many commands to a
1477 * tagged queueing device.
1481 * device can't talk to us at the moment. Should only
1482 * occur (SAM-3) when the task queue is empty, so will cause
1483 * the empty queue handling to trigger a stall in the
1486 return ADD_TO_MLQUEUE
;
1488 scsi_handle_queue_ramp_up(scmd
->device
);
1489 case COMMAND_TERMINATED
:
1493 case CHECK_CONDITION
:
1494 rtn
= scsi_check_sense(scmd
);
1495 if (rtn
== NEEDS_RETRY
)
1497 /* if rtn == FAILED, we have no sense information;
1498 * returning FAILED will wake the error handler thread
1499 * to collect the sense and redo the decide
1502 case CONDITION_GOOD
:
1503 case INTERMEDIATE_GOOD
:
1504 case INTERMEDIATE_C_GOOD
:
1507 * who knows? FIXME(eric)
1511 case RESERVATION_CONFLICT
:
1512 sdev_printk(KERN_INFO
, scmd
->device
,
1513 "reservation conflict\n");
1514 return SUCCESS
; /* causes immediate i/o error */
1522 /* we requeue for retry because the error was retryable, and
1523 * the request was not marked fast fail. Note that above,
1524 * even if the request is marked fast fail, we still requeue
1525 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1526 if ((++scmd
->retries
) <= scmd
->allowed
1527 && !scsi_noretry_cmd(scmd
)) {
1531 * no more retries - report this one back to upper level.
1537 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1539 __blk_put_request(req
->q
, req
);
1543 * scsi_eh_lock_door - Prevent medium removal for the specified device
1544 * @sdev: SCSI device to prevent medium removal
1547 * We must be called from process context.
1550 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1551 * head of the devices request queue, and continue.
1553 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1555 struct request
*req
;
1558 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1559 * request becomes available
1561 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1563 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1567 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1570 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1572 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1573 req
->cmd_flags
|= REQ_QUIET
;
1574 req
->timeout
= 10 * HZ
;
1577 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1581 * scsi_restart_operations - restart io operations to the specified host.
1582 * @shost: Host we are restarting.
1585 * When we entered the error handler, we blocked all further i/o to
1586 * this device. we need to 'reverse' this process.
1588 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1590 struct scsi_device
*sdev
;
1591 unsigned long flags
;
1594 * If the door was locked, we need to insert a door lock request
1595 * onto the head of the SCSI request queue for the device. There
1596 * is no point trying to lock the door of an off-line device.
1598 shost_for_each_device(sdev
, shost
) {
1599 if (scsi_device_online(sdev
) && sdev
->locked
)
1600 scsi_eh_lock_door(sdev
);
1604 * next free up anything directly waiting upon the host. this
1605 * will be requests for character device operations, and also for
1606 * ioctls to queued block devices.
1608 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1611 spin_lock_irqsave(shost
->host_lock
, flags
);
1612 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1613 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1614 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1615 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1617 wake_up(&shost
->host_wait
);
1620 * finally we need to re-initiate requests that may be pending. we will
1621 * have had everything blocked while error handling is taking place, and
1622 * now that error recovery is done, we will need to ensure that these
1623 * requests are started.
1625 scsi_run_host_queues(shost
);
1629 * scsi_eh_ready_devs - check device ready state and recover if not.
1630 * @shost: host to be recovered.
1631 * @work_q: &list_head for pending commands.
1632 * @done_q: &list_head for processed commands.
1634 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1635 struct list_head
*work_q
,
1636 struct list_head
*done_q
)
1638 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1639 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1640 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1641 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1642 if (!scsi_eh_host_reset(work_q
, done_q
))
1643 scsi_eh_offline_sdevs(work_q
,
1646 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1649 * scsi_eh_flush_done_q - finish processed commands or retry them.
1650 * @done_q: list_head of processed commands.
1652 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1654 struct scsi_cmnd
*scmd
, *next
;
1656 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1657 list_del_init(&scmd
->eh_entry
);
1658 if (scsi_device_online(scmd
->device
) &&
1659 !scsi_noretry_cmd(scmd
) &&
1660 (++scmd
->retries
<= scmd
->allowed
)) {
1661 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1665 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1668 * If just we got sense for the device (called
1669 * scsi_eh_get_sense), scmd->result is already
1670 * set, do not set DRIVER_TIMEOUT.
1673 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1674 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1676 current
->comm
, scmd
));
1677 scsi_finish_command(scmd
);
1681 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1684 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1685 * @shost: Host to unjam.
1688 * When we come in here, we *know* that all commands on the bus have
1689 * either completed, failed or timed out. we also know that no further
1690 * commands are being sent to the host, so things are relatively quiet
1691 * and we have freedom to fiddle with things as we wish.
1693 * This is only the *default* implementation. it is possible for
1694 * individual drivers to supply their own version of this function, and
1695 * if the maintainer wishes to do this, it is strongly suggested that
1696 * this function be taken as a template and modified. this function
1697 * was designed to correctly handle problems for about 95% of the
1698 * different cases out there, and it should always provide at least a
1699 * reasonable amount of error recovery.
1701 * Any command marked 'failed' or 'timeout' must eventually have
1702 * scsi_finish_cmd() called for it. we do all of the retry stuff
1703 * here, so when we restart the host after we return it should have an
1706 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1708 unsigned long flags
;
1709 LIST_HEAD(eh_work_q
);
1710 LIST_HEAD(eh_done_q
);
1712 spin_lock_irqsave(shost
->host_lock
, flags
);
1713 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1714 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1716 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1718 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1719 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1720 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1722 scsi_eh_flush_done_q(&eh_done_q
);
1726 * scsi_error_handler - SCSI error handler thread
1727 * @data: Host for which we are running.
1730 * This is the main error handling loop. This is run as a kernel thread
1731 * for every SCSI host and handles all error handling activity.
1733 int scsi_error_handler(void *data
)
1735 struct Scsi_Host
*shost
= data
;
1738 * We use TASK_INTERRUPTIBLE so that the thread is not
1739 * counted against the load average as a running process.
1740 * We never actually get interrupted because kthread_run
1741 * disables signal delivery for the created thread.
1743 set_current_state(TASK_INTERRUPTIBLE
);
1744 while (!kthread_should_stop()) {
1745 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1746 shost
->host_failed
!= shost
->host_busy
) {
1747 SCSI_LOG_ERROR_RECOVERY(1,
1748 printk("Error handler scsi_eh_%d sleeping\n",
1751 set_current_state(TASK_INTERRUPTIBLE
);
1755 __set_current_state(TASK_RUNNING
);
1756 SCSI_LOG_ERROR_RECOVERY(1,
1757 printk("Error handler scsi_eh_%d waking up\n",
1761 * We have a host that is failing for some reason. Figure out
1762 * what we need to do to get it up and online again (if we can).
1763 * If we fail, we end up taking the thing offline.
1765 if (shost
->transportt
->eh_strategy_handler
)
1766 shost
->transportt
->eh_strategy_handler(shost
);
1768 scsi_unjam_host(shost
);
1771 * Note - if the above fails completely, the action is to take
1772 * individual devices offline and flush the queue of any
1773 * outstanding requests that may have been pending. When we
1774 * restart, we restart any I/O to any other devices on the bus
1775 * which are still online.
1777 scsi_restart_operations(shost
);
1778 set_current_state(TASK_INTERRUPTIBLE
);
1780 __set_current_state(TASK_RUNNING
);
1782 SCSI_LOG_ERROR_RECOVERY(1,
1783 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1784 shost
->ehandler
= NULL
;
1789 * Function: scsi_report_bus_reset()
1791 * Purpose: Utility function used by low-level drivers to report that
1792 * they have observed a bus reset on the bus being handled.
1794 * Arguments: shost - Host in question
1795 * channel - channel on which reset was observed.
1799 * Lock status: Host lock must be held.
1801 * Notes: This only needs to be called if the reset is one which
1802 * originates from an unknown location. Resets originated
1803 * by the mid-level itself don't need to call this, but there
1804 * should be no harm.
1806 * The main purpose of this is to make sure that a CHECK_CONDITION
1807 * is properly treated.
1809 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1811 struct scsi_device
*sdev
;
1813 __shost_for_each_device(sdev
, shost
) {
1814 if (channel
== sdev_channel(sdev
))
1815 __scsi_report_device_reset(sdev
, NULL
);
1818 EXPORT_SYMBOL(scsi_report_bus_reset
);
1821 * Function: scsi_report_device_reset()
1823 * Purpose: Utility function used by low-level drivers to report that
1824 * they have observed a device reset on the device being handled.
1826 * Arguments: shost - Host in question
1827 * channel - channel on which reset was observed
1828 * target - target on which reset was observed
1832 * Lock status: Host lock must be held
1834 * Notes: This only needs to be called if the reset is one which
1835 * originates from an unknown location. Resets originated
1836 * by the mid-level itself don't need to call this, but there
1837 * should be no harm.
1839 * The main purpose of this is to make sure that a CHECK_CONDITION
1840 * is properly treated.
1842 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1844 struct scsi_device
*sdev
;
1846 __shost_for_each_device(sdev
, shost
) {
1847 if (channel
== sdev_channel(sdev
) &&
1848 target
== sdev_id(sdev
))
1849 __scsi_report_device_reset(sdev
, NULL
);
1852 EXPORT_SYMBOL(scsi_report_device_reset
);
1855 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1860 * Function: scsi_reset_provider
1862 * Purpose: Send requested reset to a bus or device at any phase.
1864 * Arguments: device - device to send reset to
1865 * flag - reset type (see scsi.h)
1867 * Returns: SUCCESS/FAILURE.
1869 * Notes: This is used by the SCSI Generic driver to provide
1870 * Bus/Device reset capability.
1873 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1875 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1876 struct Scsi_Host
*shost
= dev
->host
;
1878 unsigned long flags
;
1881 blk_rq_init(NULL
, &req
);
1882 scmd
->request
= &req
;
1884 scmd
->cmnd
= req
.cmd
;
1886 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1887 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1891 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1893 spin_lock_irqsave(shost
->host_lock
, flags
);
1894 shost
->tmf_in_progress
= 1;
1895 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1898 case SCSI_TRY_RESET_DEVICE
:
1899 rtn
= scsi_try_bus_device_reset(scmd
);
1903 case SCSI_TRY_RESET_TARGET
:
1904 rtn
= scsi_try_target_reset(scmd
);
1908 case SCSI_TRY_RESET_BUS
:
1909 rtn
= scsi_try_bus_reset(scmd
);
1913 case SCSI_TRY_RESET_HOST
:
1914 rtn
= scsi_try_host_reset(scmd
);
1920 spin_lock_irqsave(shost
->host_lock
, flags
);
1921 shost
->tmf_in_progress
= 0;
1922 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1925 * be sure to wake up anyone who was sleeping or had their queue
1926 * suspended while we performed the TMF.
1928 SCSI_LOG_ERROR_RECOVERY(3,
1929 printk("%s: waking up host to restart after TMF\n",
1932 wake_up(&shost
->host_wait
);
1934 scsi_run_host_queues(shost
);
1936 scsi_next_command(scmd
);
1939 EXPORT_SYMBOL(scsi_reset_provider
);
1942 * scsi_normalize_sense - normalize main elements from either fixed or
1943 * descriptor sense data format into a common format.
1945 * @sense_buffer: byte array containing sense data returned by device
1946 * @sb_len: number of valid bytes in sense_buffer
1947 * @sshdr: pointer to instance of structure that common
1948 * elements are written to.
1951 * The "main elements" from sense data are: response_code, sense_key,
1952 * asc, ascq and additional_length (only for descriptor format).
1954 * Typically this function can be called after a device has
1955 * responded to a SCSI command with the CHECK_CONDITION status.
1958 * 1 if valid sense data information found, else 0;
1960 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1961 struct scsi_sense_hdr
*sshdr
)
1963 if (!sense_buffer
|| !sb_len
)
1966 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1968 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1970 if (!scsi_sense_valid(sshdr
))
1973 if (sshdr
->response_code
>= 0x72) {
1978 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1980 sshdr
->asc
= sense_buffer
[2];
1982 sshdr
->ascq
= sense_buffer
[3];
1984 sshdr
->additional_length
= sense_buffer
[7];
1990 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1992 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1993 sb_len
: (sense_buffer
[7] + 8);
1995 sshdr
->asc
= sense_buffer
[12];
1997 sshdr
->ascq
= sense_buffer
[13];
2003 EXPORT_SYMBOL(scsi_normalize_sense
);
2005 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2006 struct scsi_sense_hdr
*sshdr
)
2008 return scsi_normalize_sense(cmd
->sense_buffer
,
2009 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2011 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2014 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2015 * @sense_buffer: byte array of descriptor format sense data
2016 * @sb_len: number of valid bytes in sense_buffer
2017 * @desc_type: value of descriptor type to find
2018 * (e.g. 0 -> information)
2021 * only valid when sense data is in descriptor format
2024 * pointer to start of (first) descriptor if found else NULL
2026 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2029 int add_sen_len
, add_len
, desc_len
, k
;
2032 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2034 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2036 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2037 add_sen_len
: (sb_len
- 8);
2038 descp
= &sense_buffer
[8];
2039 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2041 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2042 desc_len
= add_len
+ 2;
2043 if (descp
[0] == desc_type
)
2045 if (add_len
< 0) // short descriptor ??
2050 EXPORT_SYMBOL(scsi_sense_desc_find
);
2053 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2054 * @sense_buffer: byte array of sense data
2055 * @sb_len: number of valid bytes in sense_buffer
2056 * @info_out: pointer to 64 integer where 8 or 4 byte information
2057 * field will be placed if found.
2060 * 1 if information field found, 0 if not found.
2062 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2071 switch (sense_buffer
[0] & 0x7f) {
2074 if (sense_buffer
[0] & 0x80) {
2075 *info_out
= (sense_buffer
[3] << 24) +
2076 (sense_buffer
[4] << 16) +
2077 (sense_buffer
[5] << 8) + sense_buffer
[6];
2083 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2085 if (ucp
&& (0xa == ucp
[1])) {
2087 for (j
= 0; j
< 8; ++j
) {
2100 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2103 * scsi_build_sense_buffer - build sense data in a buffer
2104 * @desc: Sense format (non zero == descriptor format,
2105 * 0 == fixed format)
2106 * @buf: Where to build sense data
2108 * @asc: Additional sense code
2109 * @ascq: Additional sense code qualifier
2112 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2115 buf
[0] = 0x72; /* descriptor, current */
2121 buf
[0] = 0x70; /* fixed, current */
2128 EXPORT_SYMBOL(scsi_build_sense_buffer
);