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
);
626 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
628 if (scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
629 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
630 if (scsi_try_target_reset(scmd
) != SUCCESS
)
631 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
632 scsi_try_host_reset(scmd
);
636 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
637 * @scmd: SCSI command structure to hijack
638 * @ses: structure to save restore information
639 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
640 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
641 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
643 * This function is used to save a scsi command information before re-execution
644 * as part of the error recovery process. If @sense_bytes is 0 the command
645 * sent must be one that does not transfer any data. If @sense_bytes != 0
646 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
647 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
649 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
650 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
652 struct scsi_device
*sdev
= scmd
->device
;
655 * We need saved copies of a number of fields - this is because
656 * error handling may need to overwrite these with different values
657 * to run different commands, and once error handling is complete,
658 * we will need to restore these values prior to running the actual
661 ses
->cmd_len
= scmd
->cmd_len
;
662 ses
->cmnd
= scmd
->cmnd
;
663 ses
->data_direction
= scmd
->sc_data_direction
;
664 ses
->sdb
= scmd
->sdb
;
665 ses
->next_rq
= scmd
->request
->next_rq
;
666 ses
->result
= scmd
->result
;
667 ses
->underflow
= scmd
->underflow
;
668 ses
->prot_op
= scmd
->prot_op
;
670 scmd
->prot_op
= SCSI_PROT_NORMAL
;
671 scmd
->cmnd
= ses
->eh_cmnd
;
672 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
673 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
674 scmd
->request
->next_rq
= NULL
;
677 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
679 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
681 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
682 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
683 scmd
->sdb
.table
.nents
= 1;
684 scmd
->cmnd
[0] = REQUEST_SENSE
;
685 scmd
->cmnd
[4] = scmd
->sdb
.length
;
686 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
688 scmd
->sc_data_direction
= DMA_NONE
;
690 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
691 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
692 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
698 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
699 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
700 (sdev
->lun
<< 5 & 0xe0);
703 * Zero the sense buffer. The scsi spec mandates that any
704 * untransferred sense data should be interpreted as being zero.
706 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
708 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
711 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
712 * @scmd: SCSI command structure to restore
713 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
715 * Undo any damage done by above scsi_eh_prep_cmnd().
717 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
720 * Restore original data
722 scmd
->cmd_len
= ses
->cmd_len
;
723 scmd
->cmnd
= ses
->cmnd
;
724 scmd
->sc_data_direction
= ses
->data_direction
;
725 scmd
->sdb
= ses
->sdb
;
726 scmd
->request
->next_rq
= ses
->next_rq
;
727 scmd
->result
= ses
->result
;
728 scmd
->underflow
= ses
->underflow
;
729 scmd
->prot_op
= ses
->prot_op
;
731 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
734 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
735 * @scmd: SCSI command structure to hijack
737 * @cmnd_size: size in bytes of @cmnd
738 * @timeout: timeout for this request
739 * @sense_bytes: size of sense data to copy or 0
741 * This function is used to send a scsi command down to a target device
742 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
745 * SUCCESS or FAILED or NEEDS_RETRY
747 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
748 int cmnd_size
, int timeout
, unsigned sense_bytes
)
750 struct scsi_device
*sdev
= scmd
->device
;
751 struct Scsi_Host
*shost
= sdev
->host
;
752 DECLARE_COMPLETION_ONSTACK(done
);
753 unsigned long timeleft
;
754 struct scsi_eh_save ses
;
757 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
758 shost
->eh_action
= &done
;
761 scmd
->scsi_done
= scsi_eh_done
;
762 shost
->hostt
->queuecommand(shost
, scmd
);
764 timeleft
= wait_for_completion_timeout(&done
, timeout
);
766 shost
->eh_action
= NULL
;
768 scsi_log_completion(scmd
, SUCCESS
);
770 SCSI_LOG_ERROR_RECOVERY(3,
771 printk("%s: scmd: %p, timeleft: %ld\n",
772 __func__
, scmd
, timeleft
));
775 * If there is time left scsi_eh_done got called, and we will
776 * examine the actual status codes to see whether the command
777 * actually did complete normally, else tell the host to forget
778 * about this command.
781 rtn
= scsi_eh_completed_normally(scmd
);
782 SCSI_LOG_ERROR_RECOVERY(3,
783 printk("%s: scsi_eh_completed_normally %x\n",
799 scsi_abort_eh_cmnd(scmd
);
803 scsi_eh_restore_cmnd(scmd
, &ses
);
808 * scsi_request_sense - Request sense data from a particular target.
809 * @scmd: SCSI cmd for request sense.
812 * Some hosts automatically obtain this information, others require
813 * that we obtain it on our own. This function will *not* return until
814 * the command either times out, or it completes.
816 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
818 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
822 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
823 * @scmd: Original SCSI cmd that eh has finished.
824 * @done_q: Queue for processed commands.
827 * We don't want to use the normal command completion while we are are
828 * still handling errors - it may cause other commands to be queued,
829 * and that would disturb what we are doing. Thus we really want to
830 * keep a list of pending commands for final completion, and once we
831 * are ready to leave error handling we handle completion for real.
833 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
835 scmd
->device
->host
->host_failed
--;
837 list_move_tail(&scmd
->eh_entry
, done_q
);
839 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
842 * scsi_eh_get_sense - Get device sense data.
843 * @work_q: Queue of commands to process.
844 * @done_q: Queue of processed commands.
847 * See if we need to request sense information. if so, then get it
848 * now, so we have a better idea of what to do.
851 * This has the unfortunate side effect that if a shost adapter does
852 * not automatically request sense information, we end up shutting
853 * it down before we request it.
855 * All drivers should request sense information internally these days,
856 * so for now all I have to say is tough noogies if you end up in here.
858 * XXX: Long term this code should go away, but that needs an audit of
861 int scsi_eh_get_sense(struct list_head
*work_q
,
862 struct list_head
*done_q
)
864 struct scsi_cmnd
*scmd
, *next
;
867 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
868 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
869 SCSI_SENSE_VALID(scmd
))
872 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
873 "%s: requesting sense\n",
875 rtn
= scsi_request_sense(scmd
);
879 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
880 " result %x\n", scmd
,
882 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
884 rtn
= scsi_decide_disposition(scmd
);
887 * if the result was normal, then just pass it along to the
891 /* we don't want this command reissued, just
892 * finished with the sense data, so set
893 * retries to the max allowed to ensure it
894 * won't get reissued */
895 scmd
->retries
= scmd
->allowed
;
896 else if (rtn
!= NEEDS_RETRY
)
899 scsi_eh_finish_cmd(scmd
, done_q
);
902 return list_empty(work_q
);
904 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
907 * scsi_eh_tur - Send TUR to device.
908 * @scmd: &scsi_cmnd to send TUR
911 * 0 - Device is ready. 1 - Device NOT ready.
913 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
915 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
916 int retry_cnt
= 1, rtn
;
919 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
921 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
922 __func__
, scmd
, rtn
));
937 * scsi_eh_abort_cmds - abort pending commands.
938 * @work_q: &list_head for pending commands.
939 * @done_q: &list_head for processed commands.
942 * Try and see whether or not it makes sense to try and abort the
943 * running command. This only works out to be the case if we have one
944 * command that has timed out. If the command simply failed, it makes
945 * no sense to try and abort the command, since as far as the shost
946 * adapter is concerned, it isn't running.
948 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
949 struct list_head
*done_q
)
951 struct scsi_cmnd
*scmd
, *next
;
954 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
955 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
957 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
958 "0x%p\n", current
->comm
,
960 rtn
= scsi_try_to_abort_cmd(scmd
);
961 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
962 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
963 if (!scsi_device_online(scmd
->device
) ||
964 rtn
== FAST_IO_FAIL
||
965 !scsi_eh_tur(scmd
)) {
966 scsi_eh_finish_cmd(scmd
, done_q
);
970 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
977 return list_empty(work_q
);
981 * scsi_eh_try_stu - Send START_UNIT to device.
982 * @scmd: &scsi_cmnd to send START_UNIT
985 * 0 - Device is ready. 1 - Device NOT ready.
987 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
989 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
991 if (scmd
->device
->allow_restart
) {
992 int i
, rtn
= NEEDS_RETRY
;
994 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
995 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1005 * scsi_eh_stu - send START_UNIT if needed
1006 * @shost: &scsi host being recovered.
1007 * @work_q: &list_head for pending commands.
1008 * @done_q: &list_head for processed commands.
1011 * If commands are failing due to not ready, initializing command required,
1012 * try revalidating the device, which will end up sending a start unit.
1014 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1015 struct list_head
*work_q
,
1016 struct list_head
*done_q
)
1018 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1019 struct scsi_device
*sdev
;
1021 shost_for_each_device(sdev
, shost
) {
1023 list_for_each_entry(scmd
, work_q
, eh_entry
)
1024 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1025 scsi_check_sense(scmd
) == FAILED
) {
1033 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1034 " 0x%p\n", current
->comm
, sdev
));
1036 if (!scsi_eh_try_stu(stu_scmd
)) {
1037 if (!scsi_device_online(sdev
) ||
1038 !scsi_eh_tur(stu_scmd
)) {
1039 list_for_each_entry_safe(scmd
, next
,
1041 if (scmd
->device
== sdev
)
1042 scsi_eh_finish_cmd(scmd
, done_q
);
1046 SCSI_LOG_ERROR_RECOVERY(3,
1047 printk("%s: START_UNIT failed to sdev:"
1048 " 0x%p\n", current
->comm
, sdev
));
1052 return list_empty(work_q
);
1057 * scsi_eh_bus_device_reset - send bdr if needed
1058 * @shost: scsi host being recovered.
1059 * @work_q: &list_head for pending commands.
1060 * @done_q: &list_head for processed commands.
1063 * Try a bus device reset. Still, look to see whether we have multiple
1064 * devices that are jammed or not - if we have multiple devices, it
1065 * makes no sense to try bus_device_reset - we really would need to try
1066 * a bus_reset instead.
1068 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1069 struct list_head
*work_q
,
1070 struct list_head
*done_q
)
1072 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1073 struct scsi_device
*sdev
;
1076 shost_for_each_device(sdev
, shost
) {
1078 list_for_each_entry(scmd
, work_q
, eh_entry
)
1079 if (scmd
->device
== sdev
) {
1087 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1088 " 0x%p\n", current
->comm
,
1090 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1091 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1092 if (!scsi_device_online(sdev
) ||
1093 rtn
== FAST_IO_FAIL
||
1094 !scsi_eh_tur(bdr_scmd
)) {
1095 list_for_each_entry_safe(scmd
, next
,
1097 if (scmd
->device
== sdev
)
1098 scsi_eh_finish_cmd(scmd
,
1103 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1111 return list_empty(work_q
);
1115 * scsi_eh_target_reset - send target reset if needed
1116 * @shost: scsi host being recovered.
1117 * @work_q: &list_head for pending commands.
1118 * @done_q: &list_head for processed commands.
1121 * Try a target reset.
1123 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1124 struct list_head
*work_q
,
1125 struct list_head
*done_q
)
1127 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1128 unsigned int id
= 0;
1133 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1134 if (id
== scmd_id(scmd
)) {
1140 /* not one exactly equal; find the next highest */
1141 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1142 if (scmd_id(scmd
) > id
&&
1144 scmd_id(tgtr_scmd
) > scmd_id(scmd
)))
1149 /* no more commands, that's it */
1152 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1154 current
->comm
, id
));
1155 rtn
= scsi_try_target_reset(tgtr_scmd
);
1156 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1157 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1158 if (id
== scmd_id(scmd
))
1159 if (!scsi_device_online(scmd
->device
) ||
1160 rtn
== FAST_IO_FAIL
||
1161 !scsi_eh_tur(tgtr_scmd
))
1162 scsi_eh_finish_cmd(scmd
,
1166 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1169 current
->comm
, id
));
1173 return list_empty(work_q
);
1177 * scsi_eh_bus_reset - send a bus reset
1178 * @shost: &scsi host being recovered.
1179 * @work_q: &list_head for pending commands.
1180 * @done_q: &list_head for processed commands.
1182 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1183 struct list_head
*work_q
,
1184 struct list_head
*done_q
)
1186 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1187 unsigned int channel
;
1191 * we really want to loop over the various channels, and do this on
1192 * a channel by channel basis. we should also check to see if any
1193 * of the failed commands are on soft_reset devices, and if so, skip
1197 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1199 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1200 if (channel
== scmd_channel(scmd
)) {
1204 * FIXME add back in some support for
1205 * soft_reset devices.
1212 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1213 " %d\n", current
->comm
,
1215 rtn
= scsi_try_bus_reset(chan_scmd
);
1216 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1217 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1218 if (channel
== scmd_channel(scmd
))
1219 if (!scsi_device_online(scmd
->device
) ||
1220 rtn
== FAST_IO_FAIL
||
1222 scsi_eh_finish_cmd(scmd
,
1226 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1227 " failed chan: %d\n",
1232 return list_empty(work_q
);
1236 * scsi_eh_host_reset - send a host reset
1237 * @work_q: list_head for processed commands.
1238 * @done_q: list_head for processed commands.
1240 static int scsi_eh_host_reset(struct list_head
*work_q
,
1241 struct list_head
*done_q
)
1243 struct scsi_cmnd
*scmd
, *next
;
1246 if (!list_empty(work_q
)) {
1247 scmd
= list_entry(work_q
->next
,
1248 struct scsi_cmnd
, eh_entry
);
1250 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1253 rtn
= scsi_try_host_reset(scmd
);
1254 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1255 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1256 if (!scsi_device_online(scmd
->device
) ||
1257 rtn
== FAST_IO_FAIL
||
1258 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1260 scsi_eh_finish_cmd(scmd
, done_q
);
1263 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1268 return list_empty(work_q
);
1272 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1273 * @work_q: list_head for processed commands.
1274 * @done_q: list_head for processed commands.
1276 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1277 struct list_head
*done_q
)
1279 struct scsi_cmnd
*scmd
, *next
;
1281 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1282 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1283 "not ready after error recovery\n");
1284 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1285 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1287 * FIXME: Handle lost cmds.
1290 scsi_eh_finish_cmd(scmd
, done_q
);
1296 * scsi_noretry_cmd - determinte if command should be failed fast
1297 * @scmd: SCSI cmd to examine.
1299 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1301 switch (host_byte(scmd
->result
)) {
1305 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1307 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1309 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1310 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1313 case DID_SOFT_ERROR
:
1314 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1317 switch (status_byte(scmd
->result
)) {
1318 case CHECK_CONDITION
:
1320 * assume caller has checked sense and determinted
1321 * the check condition was retryable.
1323 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1324 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1332 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1333 * @scmd: SCSI cmd to examine.
1336 * This is *only* called when we are examining the status after sending
1337 * out the actual data command. any commands that are queued for error
1338 * recovery (e.g. test_unit_ready) do *not* come through here.
1340 * When this routine returns failed, it means the error handler thread
1341 * is woken. In cases where the error code indicates an error that
1342 * doesn't require the error handler read (i.e. we don't need to
1343 * abort/reset), this function should return SUCCESS.
1345 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1350 * if the device is offline, then we clearly just pass the result back
1351 * up to the top level.
1353 if (!scsi_device_online(scmd
->device
)) {
1354 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1361 * first check the host byte, to see if there is anything in there
1362 * that would indicate what we need to do.
1364 switch (host_byte(scmd
->result
)) {
1365 case DID_PASSTHROUGH
:
1367 * no matter what, pass this through to the upper layer.
1368 * nuke this special code so that it looks like we are saying
1371 scmd
->result
&= 0xff00ffff;
1375 * looks good. drop through, and check the next byte.
1378 case DID_NO_CONNECT
:
1379 case DID_BAD_TARGET
:
1382 * note - this means that we just report the status back
1383 * to the top level driver, not that we actually think
1384 * that it indicates SUCCESS.
1388 * when the low level driver returns did_soft_error,
1389 * it is responsible for keeping an internal retry counter
1390 * in order to avoid endless loops (db)
1392 * actually this is a bug in this function here. we should
1393 * be mindful of the maximum number of retries specified
1394 * and not get stuck in a loop.
1396 case DID_SOFT_ERROR
:
1402 return ADD_TO_MLQUEUE
;
1403 case DID_TRANSPORT_DISRUPTED
:
1405 * LLD/transport was disrupted during processing of the IO.
1406 * The transport class is now blocked/blocking,
1407 * and the transport will decide what to do with the IO
1408 * based on its timers and recovery capablilities if
1409 * there are enough retries.
1412 case DID_TRANSPORT_FAILFAST
:
1414 * The transport decided to failfast the IO (most likely
1415 * the fast io fail tmo fired), so send IO directly upwards.
1419 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1420 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1422 * execute reservation conflict processing code
1433 * when we scan the bus, we get timeout messages for
1434 * these commands if there is no device available.
1435 * other hosts report did_no_connect for the same thing.
1437 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1438 scmd
->cmnd
[0] == INQUIRY
)) {
1450 * next, check the message byte.
1452 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1456 * check the status byte to see if this indicates anything special.
1458 switch (status_byte(scmd
->result
)) {
1460 scsi_handle_queue_full(scmd
->device
);
1462 * the case of trying to send too many commands to a
1463 * tagged queueing device.
1467 * device can't talk to us at the moment. Should only
1468 * occur (SAM-3) when the task queue is empty, so will cause
1469 * the empty queue handling to trigger a stall in the
1472 return ADD_TO_MLQUEUE
;
1474 scsi_handle_queue_ramp_up(scmd
->device
);
1475 case COMMAND_TERMINATED
:
1479 case CHECK_CONDITION
:
1480 rtn
= scsi_check_sense(scmd
);
1481 if (rtn
== NEEDS_RETRY
)
1483 /* if rtn == FAILED, we have no sense information;
1484 * returning FAILED will wake the error handler thread
1485 * to collect the sense and redo the decide
1488 case CONDITION_GOOD
:
1489 case INTERMEDIATE_GOOD
:
1490 case INTERMEDIATE_C_GOOD
:
1493 * who knows? FIXME(eric)
1497 case RESERVATION_CONFLICT
:
1498 sdev_printk(KERN_INFO
, scmd
->device
,
1499 "reservation conflict\n");
1500 return SUCCESS
; /* causes immediate i/o error */
1508 /* we requeue for retry because the error was retryable, and
1509 * the request was not marked fast fail. Note that above,
1510 * even if the request is marked fast fail, we still requeue
1511 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1512 if ((++scmd
->retries
) <= scmd
->allowed
1513 && !scsi_noretry_cmd(scmd
)) {
1517 * no more retries - report this one back to upper level.
1523 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1525 __blk_put_request(req
->q
, req
);
1529 * scsi_eh_lock_door - Prevent medium removal for the specified device
1530 * @sdev: SCSI device to prevent medium removal
1533 * We must be called from process context.
1536 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1537 * head of the devices request queue, and continue.
1539 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1541 struct request
*req
;
1544 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1545 * request becomes available
1547 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1549 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1553 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1556 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1558 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1559 req
->cmd_flags
|= REQ_QUIET
;
1560 req
->timeout
= 10 * HZ
;
1563 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1567 * scsi_restart_operations - restart io operations to the specified host.
1568 * @shost: Host we are restarting.
1571 * When we entered the error handler, we blocked all further i/o to
1572 * this device. we need to 'reverse' this process.
1574 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1576 struct scsi_device
*sdev
;
1577 unsigned long flags
;
1580 * If the door was locked, we need to insert a door lock request
1581 * onto the head of the SCSI request queue for the device. There
1582 * is no point trying to lock the door of an off-line device.
1584 shost_for_each_device(sdev
, shost
) {
1585 if (scsi_device_online(sdev
) && sdev
->locked
)
1586 scsi_eh_lock_door(sdev
);
1590 * next free up anything directly waiting upon the host. this
1591 * will be requests for character device operations, and also for
1592 * ioctls to queued block devices.
1594 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1597 spin_lock_irqsave(shost
->host_lock
, flags
);
1598 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1599 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1600 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1601 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1603 wake_up(&shost
->host_wait
);
1606 * finally we need to re-initiate requests that may be pending. we will
1607 * have had everything blocked while error handling is taking place, and
1608 * now that error recovery is done, we will need to ensure that these
1609 * requests are started.
1611 scsi_run_host_queues(shost
);
1615 * scsi_eh_ready_devs - check device ready state and recover if not.
1616 * @shost: host to be recovered.
1617 * @work_q: &list_head for pending commands.
1618 * @done_q: &list_head for processed commands.
1620 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1621 struct list_head
*work_q
,
1622 struct list_head
*done_q
)
1624 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1625 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1626 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1627 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1628 if (!scsi_eh_host_reset(work_q
, done_q
))
1629 scsi_eh_offline_sdevs(work_q
,
1632 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1635 * scsi_eh_flush_done_q - finish processed commands or retry them.
1636 * @done_q: list_head of processed commands.
1638 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1640 struct scsi_cmnd
*scmd
, *next
;
1642 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1643 list_del_init(&scmd
->eh_entry
);
1644 if (scsi_device_online(scmd
->device
) &&
1645 !scsi_noretry_cmd(scmd
) &&
1646 (++scmd
->retries
<= scmd
->allowed
)) {
1647 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1651 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1654 * If just we got sense for the device (called
1655 * scsi_eh_get_sense), scmd->result is already
1656 * set, do not set DRIVER_TIMEOUT.
1659 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1660 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1662 current
->comm
, scmd
));
1663 scsi_finish_command(scmd
);
1667 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1670 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1671 * @shost: Host to unjam.
1674 * When we come in here, we *know* that all commands on the bus have
1675 * either completed, failed or timed out. we also know that no further
1676 * commands are being sent to the host, so things are relatively quiet
1677 * and we have freedom to fiddle with things as we wish.
1679 * This is only the *default* implementation. it is possible for
1680 * individual drivers to supply their own version of this function, and
1681 * if the maintainer wishes to do this, it is strongly suggested that
1682 * this function be taken as a template and modified. this function
1683 * was designed to correctly handle problems for about 95% of the
1684 * different cases out there, and it should always provide at least a
1685 * reasonable amount of error recovery.
1687 * Any command marked 'failed' or 'timeout' must eventually have
1688 * scsi_finish_cmd() called for it. we do all of the retry stuff
1689 * here, so when we restart the host after we return it should have an
1692 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1694 unsigned long flags
;
1695 LIST_HEAD(eh_work_q
);
1696 LIST_HEAD(eh_done_q
);
1698 spin_lock_irqsave(shost
->host_lock
, flags
);
1699 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1700 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1702 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1704 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1705 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1706 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1708 scsi_eh_flush_done_q(&eh_done_q
);
1712 * scsi_error_handler - SCSI error handler thread
1713 * @data: Host for which we are running.
1716 * This is the main error handling loop. This is run as a kernel thread
1717 * for every SCSI host and handles all error handling activity.
1719 int scsi_error_handler(void *data
)
1721 struct Scsi_Host
*shost
= data
;
1724 * We use TASK_INTERRUPTIBLE so that the thread is not
1725 * counted against the load average as a running process.
1726 * We never actually get interrupted because kthread_run
1727 * disables signal delivery for the created thread.
1729 set_current_state(TASK_INTERRUPTIBLE
);
1730 while (!kthread_should_stop()) {
1731 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1732 shost
->host_failed
!= shost
->host_busy
) {
1733 SCSI_LOG_ERROR_RECOVERY(1,
1734 printk("Error handler scsi_eh_%d sleeping\n",
1737 set_current_state(TASK_INTERRUPTIBLE
);
1741 __set_current_state(TASK_RUNNING
);
1742 SCSI_LOG_ERROR_RECOVERY(1,
1743 printk("Error handler scsi_eh_%d waking up\n",
1747 * We have a host that is failing for some reason. Figure out
1748 * what we need to do to get it up and online again (if we can).
1749 * If we fail, we end up taking the thing offline.
1751 if (scsi_autopm_get_host(shost
) != 0) {
1752 SCSI_LOG_ERROR_RECOVERY(1,
1753 printk(KERN_ERR
"Error handler scsi_eh_%d "
1754 "unable to autoresume\n",
1759 if (shost
->transportt
->eh_strategy_handler
)
1760 shost
->transportt
->eh_strategy_handler(shost
);
1762 scsi_unjam_host(shost
);
1765 * Note - if the above fails completely, the action is to take
1766 * individual devices offline and flush the queue of any
1767 * outstanding requests that may have been pending. When we
1768 * restart, we restart any I/O to any other devices on the bus
1769 * which are still online.
1771 scsi_restart_operations(shost
);
1772 scsi_autopm_put_host(shost
);
1773 set_current_state(TASK_INTERRUPTIBLE
);
1775 __set_current_state(TASK_RUNNING
);
1777 SCSI_LOG_ERROR_RECOVERY(1,
1778 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1779 shost
->ehandler
= NULL
;
1784 * Function: scsi_report_bus_reset()
1786 * Purpose: Utility function used by low-level drivers to report that
1787 * they have observed a bus reset on the bus being handled.
1789 * Arguments: shost - Host in question
1790 * channel - channel on which reset was observed.
1794 * Lock status: Host lock must be held.
1796 * Notes: This only needs to be called if the reset is one which
1797 * originates from an unknown location. Resets originated
1798 * by the mid-level itself don't need to call this, but there
1799 * should be no harm.
1801 * The main purpose of this is to make sure that a CHECK_CONDITION
1802 * is properly treated.
1804 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1806 struct scsi_device
*sdev
;
1808 __shost_for_each_device(sdev
, shost
) {
1809 if (channel
== sdev_channel(sdev
))
1810 __scsi_report_device_reset(sdev
, NULL
);
1813 EXPORT_SYMBOL(scsi_report_bus_reset
);
1816 * Function: scsi_report_device_reset()
1818 * Purpose: Utility function used by low-level drivers to report that
1819 * they have observed a device reset on the device being handled.
1821 * Arguments: shost - Host in question
1822 * channel - channel on which reset was observed
1823 * target - target on which reset was observed
1827 * Lock status: Host lock must be held
1829 * Notes: This only needs to be called if the reset is one which
1830 * originates from an unknown location. Resets originated
1831 * by the mid-level itself don't need to call this, but there
1832 * should be no harm.
1834 * The main purpose of this is to make sure that a CHECK_CONDITION
1835 * is properly treated.
1837 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1839 struct scsi_device
*sdev
;
1841 __shost_for_each_device(sdev
, shost
) {
1842 if (channel
== sdev_channel(sdev
) &&
1843 target
== sdev_id(sdev
))
1844 __scsi_report_device_reset(sdev
, NULL
);
1847 EXPORT_SYMBOL(scsi_report_device_reset
);
1850 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1855 * Function: scsi_reset_provider
1857 * Purpose: Send requested reset to a bus or device at any phase.
1859 * Arguments: device - device to send reset to
1860 * flag - reset type (see scsi.h)
1862 * Returns: SUCCESS/FAILURE.
1864 * Notes: This is used by the SCSI Generic driver to provide
1865 * Bus/Device reset capability.
1868 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1870 struct scsi_cmnd
*scmd
;
1871 struct Scsi_Host
*shost
= dev
->host
;
1873 unsigned long flags
;
1876 if (scsi_autopm_get_host(shost
) < 0)
1879 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1880 blk_rq_init(NULL
, &req
);
1881 scmd
->request
= &req
;
1883 scmd
->cmnd
= req
.cmd
;
1885 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1886 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1890 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1892 spin_lock_irqsave(shost
->host_lock
, flags
);
1893 shost
->tmf_in_progress
= 1;
1894 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1897 case SCSI_TRY_RESET_DEVICE
:
1898 rtn
= scsi_try_bus_device_reset(scmd
);
1902 case SCSI_TRY_RESET_TARGET
:
1903 rtn
= scsi_try_target_reset(scmd
);
1907 case SCSI_TRY_RESET_BUS
:
1908 rtn
= scsi_try_bus_reset(scmd
);
1912 case SCSI_TRY_RESET_HOST
:
1913 rtn
= scsi_try_host_reset(scmd
);
1919 spin_lock_irqsave(shost
->host_lock
, flags
);
1920 shost
->tmf_in_progress
= 0;
1921 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1924 * be sure to wake up anyone who was sleeping or had their queue
1925 * suspended while we performed the TMF.
1927 SCSI_LOG_ERROR_RECOVERY(3,
1928 printk("%s: waking up host to restart after TMF\n",
1931 wake_up(&shost
->host_wait
);
1933 scsi_run_host_queues(shost
);
1935 scsi_next_command(scmd
);
1936 scsi_autopm_put_host(shost
);
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
);