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");
323 if (blk_barrier_rq(scmd
->request
))
325 * barrier requests should always retry on UA
326 * otherwise block will get a spurious error
331 * for normal (non barrier) commands, pass the
332 * UA upwards for a determination in the
333 * completion functions
337 /* these three are not supported */
339 case VOLUME_OVERFLOW
:
344 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
345 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
346 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
352 if (scmd
->device
->retry_hwerror
)
353 return ADD_TO_MLQUEUE
;
357 case ILLEGAL_REQUEST
:
365 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
367 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
368 struct scsi_device
*tmp_sdev
;
370 if (!sht
->change_queue_depth
||
371 sdev
->queue_depth
>= sdev
->max_queue_depth
)
374 if (time_before(jiffies
,
375 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
378 if (time_before(jiffies
,
379 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
383 * Walk all devices of a target and do
386 shost_for_each_device(tmp_sdev
, sdev
->host
) {
387 if (tmp_sdev
->channel
!= sdev
->channel
||
388 tmp_sdev
->id
!= sdev
->id
||
389 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
392 * call back into LLD to increase queue_depth by one
393 * with ramp up reason code.
395 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
396 SCSI_QDEPTH_RAMP_UP
);
397 sdev
->last_queue_ramp_up
= jiffies
;
401 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
403 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
404 struct scsi_device
*tmp_sdev
;
406 if (!sht
->change_queue_depth
)
409 shost_for_each_device(tmp_sdev
, sdev
->host
) {
410 if (tmp_sdev
->channel
!= sdev
->channel
||
411 tmp_sdev
->id
!= sdev
->id
)
414 * We do not know the number of commands that were at
415 * the device when we got the queue full so we start
416 * from the highest possible value and work our way down.
418 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
424 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
425 * @scmd: SCSI cmd to examine.
428 * This is *only* called when we are examining the status of commands
429 * queued during error recovery. the main difference here is that we
430 * don't allow for the possibility of retries here, and we are a lot
431 * more restrictive about what we consider acceptable.
433 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
436 * first check the host byte, to see if there is anything in there
437 * that would indicate what we need to do.
439 if (host_byte(scmd
->result
) == DID_RESET
) {
441 * rats. we are already in the error handler, so we now
442 * get to try and figure out what to do next. if the sense
443 * is valid, we have a pretty good idea of what to do.
444 * if not, we mark it as FAILED.
446 return scsi_check_sense(scmd
);
448 if (host_byte(scmd
->result
) != DID_OK
)
452 * next, check the message byte.
454 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
458 * now, check the status byte to see if this indicates
461 switch (status_byte(scmd
->result
)) {
463 scsi_handle_queue_ramp_up(scmd
->device
);
464 case COMMAND_TERMINATED
:
466 case CHECK_CONDITION
:
467 return scsi_check_sense(scmd
);
469 case INTERMEDIATE_GOOD
:
470 case INTERMEDIATE_C_GOOD
:
472 * who knows? FIXME(eric)
475 case RESERVATION_CONFLICT
:
477 * let issuer deal with this, it could be just fine
481 scsi_handle_queue_full(scmd
->device
);
491 * scsi_eh_done - Completion function for error handling.
492 * @scmd: Cmd that is done.
494 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
496 struct completion
*eh_action
;
498 SCSI_LOG_ERROR_RECOVERY(3,
499 printk("%s scmd: %p result: %x\n",
500 __func__
, scmd
, scmd
->result
));
502 eh_action
= scmd
->device
->host
->eh_action
;
508 * scsi_try_host_reset - ask host adapter to reset itself
509 * @scmd: SCSI cmd to send hsot reset.
511 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
516 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
519 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
522 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
524 if (rtn
== SUCCESS
) {
525 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
526 ssleep(HOST_RESET_SETTLE_TIME
);
527 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
528 scsi_report_bus_reset(scmd
->device
->host
,
530 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
537 * scsi_try_bus_reset - ask host to perform a bus reset
538 * @scmd: SCSI cmd to send bus reset.
540 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
545 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
548 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
551 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
553 if (rtn
== SUCCESS
) {
554 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
555 ssleep(BUS_RESET_SETTLE_TIME
);
556 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
557 scsi_report_bus_reset(scmd
->device
->host
,
559 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
565 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
568 sdev
->expecting_cc_ua
= 1;
572 * scsi_try_target_reset - Ask host to perform a target reset
573 * @scmd: SCSI cmd used to send a target reset
576 * There is no timeout for this operation. if this operation is
577 * unreliable for a given host, then the host itself needs to put a
578 * timer on it, and set the host back to a consistent state prior to
581 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
586 if (!scmd
->device
->host
->hostt
->eh_target_reset_handler
)
589 rtn
= scmd
->device
->host
->hostt
->eh_target_reset_handler(scmd
);
590 if (rtn
== SUCCESS
) {
591 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
592 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
593 __scsi_report_device_reset
);
594 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
601 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
602 * @scmd: SCSI cmd used to send BDR
605 * There is no timeout for this operation. if this operation is
606 * unreliable for a given host, then the host itself needs to put a
607 * timer on it, and set the host back to a consistent state prior to
610 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
614 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
617 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
619 __scsi_report_device_reset(scmd
->device
, NULL
);
623 static int __scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
625 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
628 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
632 * scsi_try_to_abort_cmd - Ask host to abort a running command.
633 * @scmd: SCSI cmd to abort from Lower Level.
636 * This function will not return until the user's completion function
637 * has been called. there is no timeout on this operation. if the
638 * author of the low-level driver wishes this operation to be timed,
639 * they can provide this facility themselves. helper functions in
640 * scsi_error.c can be supplied to make this easier to do.
642 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
645 * scsi_done was called just after the command timed out and before
646 * we had a chance to process it. (db)
648 if (scmd
->serial_number
== 0)
650 return __scsi_try_to_abort_cmd(scmd
);
653 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
655 if (__scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
656 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
657 if (scsi_try_target_reset(scmd
) != SUCCESS
)
658 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
659 scsi_try_host_reset(scmd
);
663 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
664 * @scmd: SCSI command structure to hijack
665 * @ses: structure to save restore information
666 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
667 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
668 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
670 * This function is used to save a scsi command information before re-execution
671 * as part of the error recovery process. If @sense_bytes is 0 the command
672 * sent must be one that does not transfer any data. If @sense_bytes != 0
673 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
674 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
676 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
677 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
679 struct scsi_device
*sdev
= scmd
->device
;
682 * We need saved copies of a number of fields - this is because
683 * error handling may need to overwrite these with different values
684 * to run different commands, and once error handling is complete,
685 * we will need to restore these values prior to running the actual
688 ses
->cmd_len
= scmd
->cmd_len
;
689 ses
->cmnd
= scmd
->cmnd
;
690 ses
->data_direction
= scmd
->sc_data_direction
;
691 ses
->sdb
= scmd
->sdb
;
692 ses
->next_rq
= scmd
->request
->next_rq
;
693 ses
->result
= scmd
->result
;
694 ses
->underflow
= scmd
->underflow
;
695 ses
->prot_op
= scmd
->prot_op
;
697 scmd
->prot_op
= SCSI_PROT_NORMAL
;
698 scmd
->cmnd
= ses
->eh_cmnd
;
699 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
700 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
701 scmd
->request
->next_rq
= NULL
;
704 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
706 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
708 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
709 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
710 scmd
->sdb
.table
.nents
= 1;
711 scmd
->cmnd
[0] = REQUEST_SENSE
;
712 scmd
->cmnd
[4] = scmd
->sdb
.length
;
713 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
715 scmd
->sc_data_direction
= DMA_NONE
;
717 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
718 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
719 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
725 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
726 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
727 (sdev
->lun
<< 5 & 0xe0);
730 * Zero the sense buffer. The scsi spec mandates that any
731 * untransferred sense data should be interpreted as being zero.
733 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
735 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
738 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
739 * @scmd: SCSI command structure to restore
740 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
742 * Undo any damage done by above scsi_eh_prep_cmnd().
744 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
747 * Restore original data
749 scmd
->cmd_len
= ses
->cmd_len
;
750 scmd
->cmnd
= ses
->cmnd
;
751 scmd
->sc_data_direction
= ses
->data_direction
;
752 scmd
->sdb
= ses
->sdb
;
753 scmd
->request
->next_rq
= ses
->next_rq
;
754 scmd
->result
= ses
->result
;
755 scmd
->underflow
= ses
->underflow
;
756 scmd
->prot_op
= ses
->prot_op
;
758 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
761 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
762 * @scmd: SCSI command structure to hijack
764 * @cmnd_size: size in bytes of @cmnd
765 * @timeout: timeout for this request
766 * @sense_bytes: size of sense data to copy or 0
768 * This function is used to send a scsi command down to a target device
769 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
772 * SUCCESS or FAILED or NEEDS_RETRY
774 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
775 int cmnd_size
, int timeout
, unsigned sense_bytes
)
777 struct scsi_device
*sdev
= scmd
->device
;
778 struct Scsi_Host
*shost
= sdev
->host
;
779 DECLARE_COMPLETION_ONSTACK(done
);
780 unsigned long timeleft
;
782 struct scsi_eh_save ses
;
785 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
786 shost
->eh_action
= &done
;
788 spin_lock_irqsave(shost
->host_lock
, flags
);
790 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
791 spin_unlock_irqrestore(shost
->host_lock
, flags
);
793 timeleft
= wait_for_completion_timeout(&done
, timeout
);
795 shost
->eh_action
= NULL
;
797 scsi_log_completion(scmd
, SUCCESS
);
799 SCSI_LOG_ERROR_RECOVERY(3,
800 printk("%s: scmd: %p, timeleft: %ld\n",
801 __func__
, scmd
, timeleft
));
804 * If there is time left scsi_eh_done got called, and we will
805 * examine the actual status codes to see whether the command
806 * actually did complete normally, else tell the host to forget
807 * about this command.
810 rtn
= scsi_eh_completed_normally(scmd
);
811 SCSI_LOG_ERROR_RECOVERY(3,
812 printk("%s: scsi_eh_completed_normally %x\n",
828 scsi_abort_eh_cmnd(scmd
);
832 scsi_eh_restore_cmnd(scmd
, &ses
);
837 * scsi_request_sense - Request sense data from a particular target.
838 * @scmd: SCSI cmd for request sense.
841 * Some hosts automatically obtain this information, others require
842 * that we obtain it on our own. This function will *not* return until
843 * the command either times out, or it completes.
845 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
847 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
851 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
852 * @scmd: Original SCSI cmd that eh has finished.
853 * @done_q: Queue for processed commands.
856 * We don't want to use the normal command completion while we are are
857 * still handling errors - it may cause other commands to be queued,
858 * and that would disturb what we are doing. Thus we really want to
859 * keep a list of pending commands for final completion, and once we
860 * are ready to leave error handling we handle completion for real.
862 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
864 scmd
->device
->host
->host_failed
--;
866 list_move_tail(&scmd
->eh_entry
, done_q
);
868 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
871 * scsi_eh_get_sense - Get device sense data.
872 * @work_q: Queue of commands to process.
873 * @done_q: Queue of processed commands.
876 * See if we need to request sense information. if so, then get it
877 * now, so we have a better idea of what to do.
880 * This has the unfortunate side effect that if a shost adapter does
881 * not automatically request sense information, we end up shutting
882 * it down before we request it.
884 * All drivers should request sense information internally these days,
885 * so for now all I have to say is tough noogies if you end up in here.
887 * XXX: Long term this code should go away, but that needs an audit of
890 int scsi_eh_get_sense(struct list_head
*work_q
,
891 struct list_head
*done_q
)
893 struct scsi_cmnd
*scmd
, *next
;
896 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
897 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
898 SCSI_SENSE_VALID(scmd
))
901 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
902 "%s: requesting sense\n",
904 rtn
= scsi_request_sense(scmd
);
908 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
909 " result %x\n", scmd
,
911 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
913 rtn
= scsi_decide_disposition(scmd
);
916 * if the result was normal, then just pass it along to the
920 /* we don't want this command reissued, just
921 * finished with the sense data, so set
922 * retries to the max allowed to ensure it
923 * won't get reissued */
924 scmd
->retries
= scmd
->allowed
;
925 else if (rtn
!= NEEDS_RETRY
)
928 scsi_eh_finish_cmd(scmd
, done_q
);
931 return list_empty(work_q
);
933 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
936 * scsi_eh_tur - Send TUR to device.
937 * @scmd: &scsi_cmnd to send TUR
940 * 0 - Device is ready. 1 - Device NOT ready.
942 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
944 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
945 int retry_cnt
= 1, rtn
;
948 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
950 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
951 __func__
, scmd
, rtn
));
966 * scsi_eh_abort_cmds - abort pending commands.
967 * @work_q: &list_head for pending commands.
968 * @done_q: &list_head for processed commands.
971 * Try and see whether or not it makes sense to try and abort the
972 * running command. This only works out to be the case if we have one
973 * command that has timed out. If the command simply failed, it makes
974 * no sense to try and abort the command, since as far as the shost
975 * adapter is concerned, it isn't running.
977 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
978 struct list_head
*done_q
)
980 struct scsi_cmnd
*scmd
, *next
;
983 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
984 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
986 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
987 "0x%p\n", current
->comm
,
989 rtn
= scsi_try_to_abort_cmd(scmd
);
990 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
991 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
992 if (!scsi_device_online(scmd
->device
) ||
993 rtn
== FAST_IO_FAIL
||
994 !scsi_eh_tur(scmd
)) {
995 scsi_eh_finish_cmd(scmd
, done_q
);
999 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1006 return list_empty(work_q
);
1010 * scsi_eh_try_stu - Send START_UNIT to device.
1011 * @scmd: &scsi_cmnd to send START_UNIT
1014 * 0 - Device is ready. 1 - Device NOT ready.
1016 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1018 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1020 if (scmd
->device
->allow_restart
) {
1021 int i
, rtn
= NEEDS_RETRY
;
1023 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1024 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1034 * scsi_eh_stu - send START_UNIT if needed
1035 * @shost: &scsi host being recovered.
1036 * @work_q: &list_head for pending commands.
1037 * @done_q: &list_head for processed commands.
1040 * If commands are failing due to not ready, initializing command required,
1041 * try revalidating the device, which will end up sending a start unit.
1043 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1044 struct list_head
*work_q
,
1045 struct list_head
*done_q
)
1047 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1048 struct scsi_device
*sdev
;
1050 shost_for_each_device(sdev
, shost
) {
1052 list_for_each_entry(scmd
, work_q
, eh_entry
)
1053 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1054 scsi_check_sense(scmd
) == FAILED
) {
1062 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1063 " 0x%p\n", current
->comm
, sdev
));
1065 if (!scsi_eh_try_stu(stu_scmd
)) {
1066 if (!scsi_device_online(sdev
) ||
1067 !scsi_eh_tur(stu_scmd
)) {
1068 list_for_each_entry_safe(scmd
, next
,
1070 if (scmd
->device
== sdev
)
1071 scsi_eh_finish_cmd(scmd
, done_q
);
1075 SCSI_LOG_ERROR_RECOVERY(3,
1076 printk("%s: START_UNIT failed to sdev:"
1077 " 0x%p\n", current
->comm
, sdev
));
1081 return list_empty(work_q
);
1086 * scsi_eh_bus_device_reset - send bdr if needed
1087 * @shost: scsi host being recovered.
1088 * @work_q: &list_head for pending commands.
1089 * @done_q: &list_head for processed commands.
1092 * Try a bus device reset. Still, look to see whether we have multiple
1093 * devices that are jammed or not - if we have multiple devices, it
1094 * makes no sense to try bus_device_reset - we really would need to try
1095 * a bus_reset instead.
1097 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1098 struct list_head
*work_q
,
1099 struct list_head
*done_q
)
1101 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1102 struct scsi_device
*sdev
;
1105 shost_for_each_device(sdev
, shost
) {
1107 list_for_each_entry(scmd
, work_q
, eh_entry
)
1108 if (scmd
->device
== sdev
) {
1116 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1117 " 0x%p\n", current
->comm
,
1119 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1120 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1121 if (!scsi_device_online(sdev
) ||
1122 rtn
== FAST_IO_FAIL
||
1123 !scsi_eh_tur(bdr_scmd
)) {
1124 list_for_each_entry_safe(scmd
, next
,
1126 if (scmd
->device
== sdev
)
1127 scsi_eh_finish_cmd(scmd
,
1132 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1140 return list_empty(work_q
);
1144 * scsi_eh_target_reset - send target reset if needed
1145 * @shost: scsi host being recovered.
1146 * @work_q: &list_head for pending commands.
1147 * @done_q: &list_head for processed commands.
1150 * Try a target reset.
1152 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1153 struct list_head
*work_q
,
1154 struct list_head
*done_q
)
1156 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1157 unsigned int id
= 0;
1162 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1163 if (id
== scmd_id(scmd
)) {
1169 /* not one exactly equal; find the next highest */
1170 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1171 if (scmd_id(scmd
) > id
&&
1173 scmd_id(tgtr_scmd
) > scmd_id(scmd
)))
1178 /* no more commands, that's it */
1181 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1183 current
->comm
, id
));
1184 rtn
= scsi_try_target_reset(tgtr_scmd
);
1185 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1186 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1187 if (id
== scmd_id(scmd
))
1188 if (!scsi_device_online(scmd
->device
) ||
1189 rtn
== FAST_IO_FAIL
||
1190 !scsi_eh_tur(tgtr_scmd
))
1191 scsi_eh_finish_cmd(scmd
,
1195 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1198 current
->comm
, id
));
1202 return list_empty(work_q
);
1206 * scsi_eh_bus_reset - send a bus reset
1207 * @shost: &scsi host being recovered.
1208 * @work_q: &list_head for pending commands.
1209 * @done_q: &list_head for processed commands.
1211 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1212 struct list_head
*work_q
,
1213 struct list_head
*done_q
)
1215 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1216 unsigned int channel
;
1220 * we really want to loop over the various channels, and do this on
1221 * a channel by channel basis. we should also check to see if any
1222 * of the failed commands are on soft_reset devices, and if so, skip
1226 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1228 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1229 if (channel
== scmd_channel(scmd
)) {
1233 * FIXME add back in some support for
1234 * soft_reset devices.
1241 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1242 " %d\n", current
->comm
,
1244 rtn
= scsi_try_bus_reset(chan_scmd
);
1245 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1246 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1247 if (channel
== scmd_channel(scmd
))
1248 if (!scsi_device_online(scmd
->device
) ||
1249 rtn
== FAST_IO_FAIL
||
1251 scsi_eh_finish_cmd(scmd
,
1255 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1256 " failed chan: %d\n",
1261 return list_empty(work_q
);
1265 * scsi_eh_host_reset - send a host reset
1266 * @work_q: list_head for processed commands.
1267 * @done_q: list_head for processed commands.
1269 static int scsi_eh_host_reset(struct list_head
*work_q
,
1270 struct list_head
*done_q
)
1272 struct scsi_cmnd
*scmd
, *next
;
1275 if (!list_empty(work_q
)) {
1276 scmd
= list_entry(work_q
->next
,
1277 struct scsi_cmnd
, eh_entry
);
1279 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1282 rtn
= scsi_try_host_reset(scmd
);
1283 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1284 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1285 if (!scsi_device_online(scmd
->device
) ||
1286 rtn
== FAST_IO_FAIL
||
1287 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1289 scsi_eh_finish_cmd(scmd
, done_q
);
1292 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1297 return list_empty(work_q
);
1301 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1302 * @work_q: list_head for processed commands.
1303 * @done_q: list_head for processed commands.
1305 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1306 struct list_head
*done_q
)
1308 struct scsi_cmnd
*scmd
, *next
;
1310 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1311 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1312 "not ready after error recovery\n");
1313 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1314 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1316 * FIXME: Handle lost cmds.
1319 scsi_eh_finish_cmd(scmd
, done_q
);
1325 * scsi_noretry_cmd - determinte if command should be failed fast
1326 * @scmd: SCSI cmd to examine.
1328 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1330 switch (host_byte(scmd
->result
)) {
1334 return blk_failfast_transport(scmd
->request
);
1336 return blk_failfast_dev(scmd
->request
);
1338 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1339 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1342 case DID_SOFT_ERROR
:
1343 return blk_failfast_driver(scmd
->request
);
1346 switch (status_byte(scmd
->result
)) {
1347 case CHECK_CONDITION
:
1349 * assume caller has checked sense and determinted
1350 * the check condition was retryable.
1352 return blk_failfast_dev(scmd
->request
);
1359 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1360 * @scmd: SCSI cmd to examine.
1363 * This is *only* called when we are examining the status after sending
1364 * out the actual data command. any commands that are queued for error
1365 * recovery (e.g. test_unit_ready) do *not* come through here.
1367 * When this routine returns failed, it means the error handler thread
1368 * is woken. In cases where the error code indicates an error that
1369 * doesn't require the error handler read (i.e. we don't need to
1370 * abort/reset), this function should return SUCCESS.
1372 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1377 * if the device is offline, then we clearly just pass the result back
1378 * up to the top level.
1380 if (!scsi_device_online(scmd
->device
)) {
1381 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1388 * first check the host byte, to see if there is anything in there
1389 * that would indicate what we need to do.
1391 switch (host_byte(scmd
->result
)) {
1392 case DID_PASSTHROUGH
:
1394 * no matter what, pass this through to the upper layer.
1395 * nuke this special code so that it looks like we are saying
1398 scmd
->result
&= 0xff00ffff;
1402 * looks good. drop through, and check the next byte.
1405 case DID_NO_CONNECT
:
1406 case DID_BAD_TARGET
:
1409 * note - this means that we just report the status back
1410 * to the top level driver, not that we actually think
1411 * that it indicates SUCCESS.
1415 * when the low level driver returns did_soft_error,
1416 * it is responsible for keeping an internal retry counter
1417 * in order to avoid endless loops (db)
1419 * actually this is a bug in this function here. we should
1420 * be mindful of the maximum number of retries specified
1421 * and not get stuck in a loop.
1423 case DID_SOFT_ERROR
:
1429 return ADD_TO_MLQUEUE
;
1430 case DID_TRANSPORT_DISRUPTED
:
1432 * LLD/transport was disrupted during processing of the IO.
1433 * The transport class is now blocked/blocking,
1434 * and the transport will decide what to do with the IO
1435 * based on its timers and recovery capablilities if
1436 * there are enough retries.
1439 case DID_TRANSPORT_FAILFAST
:
1441 * The transport decided to failfast the IO (most likely
1442 * the fast io fail tmo fired), so send IO directly upwards.
1446 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1447 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1449 * execute reservation conflict processing code
1460 * when we scan the bus, we get timeout messages for
1461 * these commands if there is no device available.
1462 * other hosts report did_no_connect for the same thing.
1464 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1465 scmd
->cmnd
[0] == INQUIRY
)) {
1477 * next, check the message byte.
1479 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1483 * check the status byte to see if this indicates anything special.
1485 switch (status_byte(scmd
->result
)) {
1487 scsi_handle_queue_full(scmd
->device
);
1489 * the case of trying to send too many commands to a
1490 * tagged queueing device.
1494 * device can't talk to us at the moment. Should only
1495 * occur (SAM-3) when the task queue is empty, so will cause
1496 * the empty queue handling to trigger a stall in the
1499 return ADD_TO_MLQUEUE
;
1501 scsi_handle_queue_ramp_up(scmd
->device
);
1502 case COMMAND_TERMINATED
:
1506 case CHECK_CONDITION
:
1507 rtn
= scsi_check_sense(scmd
);
1508 if (rtn
== NEEDS_RETRY
)
1510 /* if rtn == FAILED, we have no sense information;
1511 * returning FAILED will wake the error handler thread
1512 * to collect the sense and redo the decide
1515 case CONDITION_GOOD
:
1516 case INTERMEDIATE_GOOD
:
1517 case INTERMEDIATE_C_GOOD
:
1520 * who knows? FIXME(eric)
1524 case RESERVATION_CONFLICT
:
1525 sdev_printk(KERN_INFO
, scmd
->device
,
1526 "reservation conflict\n");
1527 return SUCCESS
; /* causes immediate i/o error */
1535 /* we requeue for retry because the error was retryable, and
1536 * the request was not marked fast fail. Note that above,
1537 * even if the request is marked fast fail, we still requeue
1538 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1539 if ((++scmd
->retries
) <= scmd
->allowed
1540 && !scsi_noretry_cmd(scmd
)) {
1544 * no more retries - report this one back to upper level.
1550 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1552 __blk_put_request(req
->q
, req
);
1556 * scsi_eh_lock_door - Prevent medium removal for the specified device
1557 * @sdev: SCSI device to prevent medium removal
1560 * We must be called from process context.
1563 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1564 * head of the devices request queue, and continue.
1566 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1568 struct request
*req
;
1571 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1572 * request becomes available
1574 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1576 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1580 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1583 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1585 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1586 req
->cmd_flags
|= REQ_QUIET
;
1587 req
->timeout
= 10 * HZ
;
1590 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1594 * scsi_restart_operations - restart io operations to the specified host.
1595 * @shost: Host we are restarting.
1598 * When we entered the error handler, we blocked all further i/o to
1599 * this device. we need to 'reverse' this process.
1601 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1603 struct scsi_device
*sdev
;
1604 unsigned long flags
;
1607 * If the door was locked, we need to insert a door lock request
1608 * onto the head of the SCSI request queue for the device. There
1609 * is no point trying to lock the door of an off-line device.
1611 shost_for_each_device(sdev
, shost
) {
1612 if (scsi_device_online(sdev
) && sdev
->locked
)
1613 scsi_eh_lock_door(sdev
);
1617 * next free up anything directly waiting upon the host. this
1618 * will be requests for character device operations, and also for
1619 * ioctls to queued block devices.
1621 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1624 spin_lock_irqsave(shost
->host_lock
, flags
);
1625 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1626 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1627 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1628 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1630 wake_up(&shost
->host_wait
);
1633 * finally we need to re-initiate requests that may be pending. we will
1634 * have had everything blocked while error handling is taking place, and
1635 * now that error recovery is done, we will need to ensure that these
1636 * requests are started.
1638 scsi_run_host_queues(shost
);
1642 * scsi_eh_ready_devs - check device ready state and recover if not.
1643 * @shost: host to be recovered.
1644 * @work_q: &list_head for pending commands.
1645 * @done_q: &list_head for processed commands.
1647 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1648 struct list_head
*work_q
,
1649 struct list_head
*done_q
)
1651 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1652 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1653 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1654 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1655 if (!scsi_eh_host_reset(work_q
, done_q
))
1656 scsi_eh_offline_sdevs(work_q
,
1659 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1662 * scsi_eh_flush_done_q - finish processed commands or retry them.
1663 * @done_q: list_head of processed commands.
1665 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1667 struct scsi_cmnd
*scmd
, *next
;
1669 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1670 list_del_init(&scmd
->eh_entry
);
1671 if (scsi_device_online(scmd
->device
) &&
1672 !scsi_noretry_cmd(scmd
) &&
1673 (++scmd
->retries
<= scmd
->allowed
)) {
1674 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1678 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1681 * If just we got sense for the device (called
1682 * scsi_eh_get_sense), scmd->result is already
1683 * set, do not set DRIVER_TIMEOUT.
1686 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1687 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1689 current
->comm
, scmd
));
1690 scsi_finish_command(scmd
);
1694 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1697 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1698 * @shost: Host to unjam.
1701 * When we come in here, we *know* that all commands on the bus have
1702 * either completed, failed or timed out. we also know that no further
1703 * commands are being sent to the host, so things are relatively quiet
1704 * and we have freedom to fiddle with things as we wish.
1706 * This is only the *default* implementation. it is possible for
1707 * individual drivers to supply their own version of this function, and
1708 * if the maintainer wishes to do this, it is strongly suggested that
1709 * this function be taken as a template and modified. this function
1710 * was designed to correctly handle problems for about 95% of the
1711 * different cases out there, and it should always provide at least a
1712 * reasonable amount of error recovery.
1714 * Any command marked 'failed' or 'timeout' must eventually have
1715 * scsi_finish_cmd() called for it. we do all of the retry stuff
1716 * here, so when we restart the host after we return it should have an
1719 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1721 unsigned long flags
;
1722 LIST_HEAD(eh_work_q
);
1723 LIST_HEAD(eh_done_q
);
1725 spin_lock_irqsave(shost
->host_lock
, flags
);
1726 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1727 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1729 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1731 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1732 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1733 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1735 scsi_eh_flush_done_q(&eh_done_q
);
1739 * scsi_error_handler - SCSI error handler thread
1740 * @data: Host for which we are running.
1743 * This is the main error handling loop. This is run as a kernel thread
1744 * for every SCSI host and handles all error handling activity.
1746 int scsi_error_handler(void *data
)
1748 struct Scsi_Host
*shost
= data
;
1751 * We use TASK_INTERRUPTIBLE so that the thread is not
1752 * counted against the load average as a running process.
1753 * We never actually get interrupted because kthread_run
1754 * disables signal delivery for the created thread.
1756 set_current_state(TASK_INTERRUPTIBLE
);
1757 while (!kthread_should_stop()) {
1758 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1759 shost
->host_failed
!= shost
->host_busy
) {
1760 SCSI_LOG_ERROR_RECOVERY(1,
1761 printk("Error handler scsi_eh_%d sleeping\n",
1764 set_current_state(TASK_INTERRUPTIBLE
);
1768 __set_current_state(TASK_RUNNING
);
1769 SCSI_LOG_ERROR_RECOVERY(1,
1770 printk("Error handler scsi_eh_%d waking up\n",
1774 * We have a host that is failing for some reason. Figure out
1775 * what we need to do to get it up and online again (if we can).
1776 * If we fail, we end up taking the thing offline.
1778 if (shost
->transportt
->eh_strategy_handler
)
1779 shost
->transportt
->eh_strategy_handler(shost
);
1781 scsi_unjam_host(shost
);
1784 * Note - if the above fails completely, the action is to take
1785 * individual devices offline and flush the queue of any
1786 * outstanding requests that may have been pending. When we
1787 * restart, we restart any I/O to any other devices on the bus
1788 * which are still online.
1790 scsi_restart_operations(shost
);
1791 set_current_state(TASK_INTERRUPTIBLE
);
1793 __set_current_state(TASK_RUNNING
);
1795 SCSI_LOG_ERROR_RECOVERY(1,
1796 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1797 shost
->ehandler
= NULL
;
1802 * Function: scsi_report_bus_reset()
1804 * Purpose: Utility function used by low-level drivers to report that
1805 * they have observed a bus reset on the bus being handled.
1807 * Arguments: shost - Host in question
1808 * channel - channel on which reset was observed.
1812 * Lock status: Host lock must be held.
1814 * Notes: This only needs to be called if the reset is one which
1815 * originates from an unknown location. Resets originated
1816 * by the mid-level itself don't need to call this, but there
1817 * should be no harm.
1819 * The main purpose of this is to make sure that a CHECK_CONDITION
1820 * is properly treated.
1822 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1824 struct scsi_device
*sdev
;
1826 __shost_for_each_device(sdev
, shost
) {
1827 if (channel
== sdev_channel(sdev
))
1828 __scsi_report_device_reset(sdev
, NULL
);
1831 EXPORT_SYMBOL(scsi_report_bus_reset
);
1834 * Function: scsi_report_device_reset()
1836 * Purpose: Utility function used by low-level drivers to report that
1837 * they have observed a device reset on the device being handled.
1839 * Arguments: shost - Host in question
1840 * channel - channel on which reset was observed
1841 * target - target on which reset was observed
1845 * Lock status: Host lock must be held
1847 * Notes: This only needs to be called if the reset is one which
1848 * originates from an unknown location. Resets originated
1849 * by the mid-level itself don't need to call this, but there
1850 * should be no harm.
1852 * The main purpose of this is to make sure that a CHECK_CONDITION
1853 * is properly treated.
1855 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1857 struct scsi_device
*sdev
;
1859 __shost_for_each_device(sdev
, shost
) {
1860 if (channel
== sdev_channel(sdev
) &&
1861 target
== sdev_id(sdev
))
1862 __scsi_report_device_reset(sdev
, NULL
);
1865 EXPORT_SYMBOL(scsi_report_device_reset
);
1868 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1873 * Function: scsi_reset_provider
1875 * Purpose: Send requested reset to a bus or device at any phase.
1877 * Arguments: device - device to send reset to
1878 * flag - reset type (see scsi.h)
1880 * Returns: SUCCESS/FAILURE.
1882 * Notes: This is used by the SCSI Generic driver to provide
1883 * Bus/Device reset capability.
1886 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1888 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1889 struct Scsi_Host
*shost
= dev
->host
;
1891 unsigned long flags
;
1894 blk_rq_init(NULL
, &req
);
1895 scmd
->request
= &req
;
1897 scmd
->cmnd
= req
.cmd
;
1899 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1900 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1904 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1906 spin_lock_irqsave(shost
->host_lock
, flags
);
1907 shost
->tmf_in_progress
= 1;
1908 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1911 case SCSI_TRY_RESET_DEVICE
:
1912 rtn
= scsi_try_bus_device_reset(scmd
);
1916 case SCSI_TRY_RESET_TARGET
:
1917 rtn
= scsi_try_target_reset(scmd
);
1921 case SCSI_TRY_RESET_BUS
:
1922 rtn
= scsi_try_bus_reset(scmd
);
1926 case SCSI_TRY_RESET_HOST
:
1927 rtn
= scsi_try_host_reset(scmd
);
1933 spin_lock_irqsave(shost
->host_lock
, flags
);
1934 shost
->tmf_in_progress
= 0;
1935 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1938 * be sure to wake up anyone who was sleeping or had their queue
1939 * suspended while we performed the TMF.
1941 SCSI_LOG_ERROR_RECOVERY(3,
1942 printk("%s: waking up host to restart after TMF\n",
1945 wake_up(&shost
->host_wait
);
1947 scsi_run_host_queues(shost
);
1949 scsi_next_command(scmd
);
1952 EXPORT_SYMBOL(scsi_reset_provider
);
1955 * scsi_normalize_sense - normalize main elements from either fixed or
1956 * descriptor sense data format into a common format.
1958 * @sense_buffer: byte array containing sense data returned by device
1959 * @sb_len: number of valid bytes in sense_buffer
1960 * @sshdr: pointer to instance of structure that common
1961 * elements are written to.
1964 * The "main elements" from sense data are: response_code, sense_key,
1965 * asc, ascq and additional_length (only for descriptor format).
1967 * Typically this function can be called after a device has
1968 * responded to a SCSI command with the CHECK_CONDITION status.
1971 * 1 if valid sense data information found, else 0;
1973 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1974 struct scsi_sense_hdr
*sshdr
)
1976 if (!sense_buffer
|| !sb_len
)
1979 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1981 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1983 if (!scsi_sense_valid(sshdr
))
1986 if (sshdr
->response_code
>= 0x72) {
1991 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1993 sshdr
->asc
= sense_buffer
[2];
1995 sshdr
->ascq
= sense_buffer
[3];
1997 sshdr
->additional_length
= sense_buffer
[7];
2003 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2005 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2006 sb_len
: (sense_buffer
[7] + 8);
2008 sshdr
->asc
= sense_buffer
[12];
2010 sshdr
->ascq
= sense_buffer
[13];
2016 EXPORT_SYMBOL(scsi_normalize_sense
);
2018 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2019 struct scsi_sense_hdr
*sshdr
)
2021 return scsi_normalize_sense(cmd
->sense_buffer
,
2022 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2024 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2027 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2028 * @sense_buffer: byte array of descriptor format sense data
2029 * @sb_len: number of valid bytes in sense_buffer
2030 * @desc_type: value of descriptor type to find
2031 * (e.g. 0 -> information)
2034 * only valid when sense data is in descriptor format
2037 * pointer to start of (first) descriptor if found else NULL
2039 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2042 int add_sen_len
, add_len
, desc_len
, k
;
2045 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2047 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2049 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2050 add_sen_len
: (sb_len
- 8);
2051 descp
= &sense_buffer
[8];
2052 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2054 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2055 desc_len
= add_len
+ 2;
2056 if (descp
[0] == desc_type
)
2058 if (add_len
< 0) // short descriptor ??
2063 EXPORT_SYMBOL(scsi_sense_desc_find
);
2066 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2067 * @sense_buffer: byte array of sense data
2068 * @sb_len: number of valid bytes in sense_buffer
2069 * @info_out: pointer to 64 integer where 8 or 4 byte information
2070 * field will be placed if found.
2073 * 1 if information field found, 0 if not found.
2075 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2084 switch (sense_buffer
[0] & 0x7f) {
2087 if (sense_buffer
[0] & 0x80) {
2088 *info_out
= (sense_buffer
[3] << 24) +
2089 (sense_buffer
[4] << 16) +
2090 (sense_buffer
[5] << 8) + sense_buffer
[6];
2096 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2098 if (ucp
&& (0xa == ucp
[1])) {
2100 for (j
= 0; j
< 8; ++j
) {
2113 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2116 * scsi_build_sense_buffer - build sense data in a buffer
2117 * @desc: Sense format (non zero == descriptor format,
2118 * 0 == fixed format)
2119 * @buf: Where to build sense data
2121 * @asc: Additional sense code
2122 * @ascq: Additional sense code qualifier
2125 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2128 buf
[0] = 0x72; /* descriptor, current */
2134 buf
[0] = 0x70; /* fixed, current */
2141 EXPORT_SYMBOL(scsi_build_sense_buffer
);