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>
28 #include <linux/jiffies.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_common.h>
37 #include <scsi/scsi_transport.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_ioctl.h>
40 #include <scsi/scsi_dh.h>
43 #include "scsi_priv.h"
44 #include "scsi_logging.h"
45 #include "scsi_transport_api.h"
47 #include <trace/events/scsi.h>
49 static void scsi_eh_done(struct scsi_cmnd
*scmd
);
52 * These should *probably* be handled by the host itself.
53 * Since it is allowed to sleep, it probably should.
55 #define BUS_RESET_SETTLE_TIME (10)
56 #define HOST_RESET_SETTLE_TIME (10)
58 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
59 static int scsi_try_to_abort_cmd(struct scsi_host_template
*,
62 /* called with shost->host_lock held */
63 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
65 if (atomic_read(&shost
->host_busy
) == shost
->host_failed
) {
66 trace_scsi_eh_wakeup(shost
);
67 wake_up_process(shost
->ehandler
);
68 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO
, shost
,
69 "Waking error handler thread\n"));
74 * scsi_schedule_eh - schedule EH for SCSI host
75 * @shost: SCSI host to invoke error handling on.
77 * Schedule SCSI EH without scmd.
79 void scsi_schedule_eh(struct Scsi_Host
*shost
)
83 spin_lock_irqsave(shost
->host_lock
, flags
);
85 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
86 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
87 shost
->host_eh_scheduled
++;
88 scsi_eh_wakeup(shost
);
91 spin_unlock_irqrestore(shost
->host_lock
, flags
);
93 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
95 static int scsi_host_eh_past_deadline(struct Scsi_Host
*shost
)
97 if (!shost
->last_reset
|| shost
->eh_deadline
== -1)
101 * 32bit accesses are guaranteed to be atomic
102 * (on all supported architectures), so instead
103 * of using a spinlock we can as well double check
104 * if eh_deadline has been set to 'off' during the
107 if (time_before(jiffies
, shost
->last_reset
+ shost
->eh_deadline
) &&
108 shost
->eh_deadline
> -1)
115 * scmd_eh_abort_handler - Handle command aborts
116 * @work: command to be aborted.
119 scmd_eh_abort_handler(struct work_struct
*work
)
121 struct scsi_cmnd
*scmd
=
122 container_of(work
, struct scsi_cmnd
, abort_work
.work
);
123 struct scsi_device
*sdev
= scmd
->device
;
126 if (scsi_host_eh_past_deadline(sdev
->host
)) {
127 SCSI_LOG_ERROR_RECOVERY(3,
128 scmd_printk(KERN_INFO
, scmd
,
129 "eh timeout, not aborting\n"));
131 SCSI_LOG_ERROR_RECOVERY(3,
132 scmd_printk(KERN_INFO
, scmd
,
133 "aborting command\n"));
134 rtn
= scsi_try_to_abort_cmd(sdev
->host
->hostt
, scmd
);
135 if (rtn
== SUCCESS
) {
136 set_host_byte(scmd
, DID_TIME_OUT
);
137 if (scsi_host_eh_past_deadline(sdev
->host
)) {
138 SCSI_LOG_ERROR_RECOVERY(3,
139 scmd_printk(KERN_INFO
, scmd
,
140 "eh timeout, not retrying "
141 "aborted command\n"));
142 } else if (!scsi_noretry_cmd(scmd
) &&
143 (++scmd
->retries
<= scmd
->allowed
)) {
144 SCSI_LOG_ERROR_RECOVERY(3,
145 scmd_printk(KERN_WARNING
, scmd
,
146 "retry aborted command\n"));
147 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_WARNING
, scmd
,
152 "finish aborted command\n"));
153 scsi_finish_command(scmd
);
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO
, scmd
,
160 (rtn
== FAST_IO_FAIL
) ?
161 "not send" : "failed"));
165 if (!scsi_eh_scmd_add(scmd
, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING
, scmd
,
168 "terminate aborted command\n"));
169 set_host_byte(scmd
, DID_TIME_OUT
);
170 scsi_finish_command(scmd
);
175 * scsi_abort_command - schedule a command abort
176 * @scmd: scmd to abort.
178 * We only need to abort commands after a command timeout
181 scsi_abort_command(struct scsi_cmnd
*scmd
)
183 struct scsi_device
*sdev
= scmd
->device
;
184 struct Scsi_Host
*shost
= sdev
->host
;
187 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
189 * Retry after abort failed, escalate to next level.
191 scmd
->eh_eflags
&= ~SCSI_EH_ABORT_SCHEDULED
;
192 SCSI_LOG_ERROR_RECOVERY(3,
193 scmd_printk(KERN_INFO
, scmd
,
194 "previous abort failed\n"));
195 BUG_ON(delayed_work_pending(&scmd
->abort_work
));
200 * Do not try a command abort if
201 * SCSI EH has already started.
203 spin_lock_irqsave(shost
->host_lock
, flags
);
204 if (scsi_host_in_recovery(shost
)) {
205 spin_unlock_irqrestore(shost
->host_lock
, flags
);
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_INFO
, scmd
,
208 "not aborting, host in recovery\n"));
212 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
213 shost
->last_reset
= jiffies
;
214 spin_unlock_irqrestore(shost
->host_lock
, flags
);
216 scmd
->eh_eflags
|= SCSI_EH_ABORT_SCHEDULED
;
217 SCSI_LOG_ERROR_RECOVERY(3,
218 scmd_printk(KERN_INFO
, scmd
, "abort scheduled\n"));
219 queue_delayed_work(shost
->tmf_work_q
, &scmd
->abort_work
, HZ
/ 100);
224 * scsi_eh_scmd_add - add scsi cmd to error handling.
225 * @scmd: scmd to run eh on.
226 * @eh_flag: optional SCSI_EH flag.
231 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
233 struct Scsi_Host
*shost
= scmd
->device
->host
;
237 if (!shost
->ehandler
)
240 spin_lock_irqsave(shost
->host_lock
, flags
);
241 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
242 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
245 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
246 shost
->last_reset
= jiffies
;
249 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
)
250 eh_flag
&= ~SCSI_EH_CANCEL_CMD
;
251 scmd
->eh_eflags
|= eh_flag
;
252 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
253 shost
->host_failed
++;
254 scsi_eh_wakeup(shost
);
256 spin_unlock_irqrestore(shost
->host_lock
, flags
);
261 * scsi_times_out - Timeout function for normal scsi commands.
262 * @req: request that is timing out.
265 * We do not need to lock this. There is the potential for a race
266 * only in that the normal completion handling might run, but if the
267 * normal completion function determines that the timer has already
268 * fired, then it mustn't do anything.
270 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
272 struct scsi_cmnd
*scmd
= req
->special
;
273 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
274 struct Scsi_Host
*host
= scmd
->device
->host
;
276 trace_scsi_dispatch_cmd_timeout(scmd
);
277 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
279 if (host
->eh_deadline
!= -1 && !host
->last_reset
)
280 host
->last_reset
= jiffies
;
282 if (host
->transportt
->eh_timed_out
)
283 rtn
= host
->transportt
->eh_timed_out(scmd
);
284 else if (host
->hostt
->eh_timed_out
)
285 rtn
= host
->hostt
->eh_timed_out(scmd
);
287 if (rtn
== BLK_EH_NOT_HANDLED
) {
288 if (!host
->hostt
->no_async_abort
&&
289 scsi_abort_command(scmd
) == SUCCESS
)
290 return BLK_EH_NOT_HANDLED
;
292 set_host_byte(scmd
, DID_TIME_OUT
);
293 if (!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))
294 rtn
= BLK_EH_HANDLED
;
301 * scsi_block_when_processing_errors - Prevent cmds from being queued.
302 * @sdev: Device on which we are performing recovery.
305 * We block until the host is out of error recovery, and then check to
306 * see whether the host or the device is offline.
309 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
311 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
315 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
317 online
= scsi_device_online(sdev
);
319 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO
, sdev
,
320 "%s: rtn: %d\n", __func__
, online
));
324 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
326 #ifdef CONFIG_SCSI_LOGGING
328 * scsi_eh_prt_fail_stats - Log info on failures.
329 * @shost: scsi host being recovered.
330 * @work_q: Queue of scsi cmds to process.
332 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
333 struct list_head
*work_q
)
335 struct scsi_cmnd
*scmd
;
336 struct scsi_device
*sdev
;
337 int total_failures
= 0;
340 int devices_failed
= 0;
342 shost_for_each_device(sdev
, shost
) {
343 list_for_each_entry(scmd
, work_q
, eh_entry
) {
344 if (scmd
->device
== sdev
) {
346 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
353 if (cmd_cancel
|| cmd_failed
) {
354 SCSI_LOG_ERROR_RECOVERY(3,
355 shost_printk(KERN_INFO
, shost
,
356 "%s: cmds failed: %d, cancel: %d\n",
357 __func__
, cmd_failed
,
365 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO
, shost
,
366 "Total of %d commands on %d"
367 " devices require eh work\n",
368 total_failures
, devices_failed
));
373 * scsi_report_lun_change - Set flag on all *other* devices on the same target
374 * to indicate that a UNIT ATTENTION is expected.
375 * @sdev: Device reporting the UNIT ATTENTION
377 static void scsi_report_lun_change(struct scsi_device
*sdev
)
379 sdev
->sdev_target
->expecting_lun_change
= 1;
383 * scsi_report_sense - Examine scsi sense information and log messages for
384 * certain conditions, also issue uevents for some of them.
385 * @sdev: Device reporting the sense code
386 * @sshdr: sshdr to be examined
388 static void scsi_report_sense(struct scsi_device
*sdev
,
389 struct scsi_sense_hdr
*sshdr
)
391 enum scsi_device_event evt_type
= SDEV_EVT_MAXBITS
; /* i.e. none */
393 if (sshdr
->sense_key
== UNIT_ATTENTION
) {
394 if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x03) {
395 evt_type
= SDEV_EVT_INQUIRY_CHANGE_REPORTED
;
396 sdev_printk(KERN_WARNING
, sdev
,
397 "Inquiry data has changed");
398 } else if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x0e) {
399 evt_type
= SDEV_EVT_LUN_CHANGE_REPORTED
;
400 scsi_report_lun_change(sdev
);
401 sdev_printk(KERN_WARNING
, sdev
,
402 "Warning! Received an indication that the "
403 "LUN assignments on this target have "
404 "changed. The Linux SCSI layer does not "
405 "automatically remap LUN assignments.\n");
406 } else if (sshdr
->asc
== 0x3f)
407 sdev_printk(KERN_WARNING
, sdev
,
408 "Warning! Received an indication that the "
409 "operating parameters on this target have "
410 "changed. The Linux SCSI layer does not "
411 "automatically adjust these parameters.\n");
413 if (sshdr
->asc
== 0x38 && sshdr
->ascq
== 0x07) {
414 evt_type
= SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
;
415 sdev_printk(KERN_WARNING
, sdev
,
416 "Warning! Received an indication that the "
417 "LUN reached a thin provisioning soft "
421 if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x01) {
422 evt_type
= SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
;
423 sdev_printk(KERN_WARNING
, sdev
,
424 "Mode parameters changed");
425 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x06) {
426 evt_type
= SDEV_EVT_ALUA_STATE_CHANGE_REPORTED
;
427 sdev_printk(KERN_WARNING
, sdev
,
428 "Asymmetric access state changed");
429 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x09) {
430 evt_type
= SDEV_EVT_CAPACITY_CHANGE_REPORTED
;
431 sdev_printk(KERN_WARNING
, sdev
,
432 "Capacity data has changed");
433 } else if (sshdr
->asc
== 0x2a)
434 sdev_printk(KERN_WARNING
, sdev
,
435 "Parameters changed");
438 if (evt_type
!= SDEV_EVT_MAXBITS
) {
439 set_bit(evt_type
, sdev
->pending_events
);
440 schedule_work(&sdev
->event_work
);
445 * scsi_check_sense - Examine scsi cmd sense
446 * @scmd: Cmd to have sense checked.
449 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
452 * When a deferred error is detected the current command has
453 * not been executed and needs retrying.
455 int scsi_check_sense(struct scsi_cmnd
*scmd
)
457 struct scsi_device
*sdev
= scmd
->device
;
458 struct scsi_sense_hdr sshdr
;
460 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
461 return FAILED
; /* no valid sense data */
463 scsi_report_sense(sdev
, &sshdr
);
465 if (scsi_sense_is_deferred(&sshdr
))
468 if (sdev
->handler
&& sdev
->handler
->check_sense
) {
471 rc
= sdev
->handler
->check_sense(sdev
, &sshdr
);
472 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
474 /* handler does not care. Drop down to default handling */
477 if (scmd
->cmnd
[0] == TEST_UNIT_READY
&& scmd
->scsi_done
!= scsi_eh_done
)
479 * nasty: for mid-layer issued TURs, we need to return the
480 * actual sense data without any recovery attempt. For eh
481 * issued ones, we need to try to recover and interpret
486 * Previous logic looked for FILEMARK, EOM or ILI which are
487 * mainly associated with tapes and returned SUCCESS.
489 if (sshdr
.response_code
== 0x70) {
491 if (scmd
->sense_buffer
[2] & 0xe0)
495 * descriptor format: look for "stream commands sense data
496 * descriptor" (see SSC-3). Assume single sense data
497 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
499 if ((sshdr
.additional_length
> 3) &&
500 (scmd
->sense_buffer
[8] == 0x4) &&
501 (scmd
->sense_buffer
[11] & 0xe0))
505 switch (sshdr
.sense_key
) {
508 case RECOVERED_ERROR
:
509 return /* soft_error */ SUCCESS
;
511 case ABORTED_COMMAND
:
512 if (sshdr
.asc
== 0x10) /* DIF */
519 * if we are expecting a cc/ua because of a bus reset that we
520 * performed, treat this just as a retry. otherwise this is
521 * information that we should pass up to the upper-level driver
522 * so that we can deal with it there.
524 if (scmd
->device
->expecting_cc_ua
) {
526 * Because some device does not queue unit
527 * attentions correctly, we carefully check
528 * additional sense code and qualifier so as
529 * not to squash media change unit attention.
531 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
532 scmd
->device
->expecting_cc_ua
= 0;
537 * we might also expect a cc/ua if another LUN on the target
538 * reported a UA with an ASC/ASCQ of 3F 0E -
539 * REPORTED LUNS DATA HAS CHANGED.
541 if (scmd
->device
->sdev_target
->expecting_lun_change
&&
542 sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
545 * if the device is in the process of becoming ready, we
548 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
551 * if the device is not started, we need to wake
552 * the error handler to start the motor
554 if (scmd
->device
->allow_restart
&&
555 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
558 * Pass the UA upwards for a determination in the completion
563 /* these are not supported */
565 if (sshdr
.asc
== 0x27 && sshdr
.ascq
== 0x07) {
566 /* Thin provisioning hard threshold reached */
567 set_host_byte(scmd
, DID_ALLOC_FAILURE
);
571 case VOLUME_OVERFLOW
:
574 set_host_byte(scmd
, DID_TARGET_FAILURE
);
578 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
579 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
580 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
581 set_host_byte(scmd
, DID_MEDIUM_ERROR
);
587 if (scmd
->device
->retry_hwerror
)
588 return ADD_TO_MLQUEUE
;
590 set_host_byte(scmd
, DID_TARGET_FAILURE
);
592 case ILLEGAL_REQUEST
:
593 if (sshdr
.asc
== 0x20 || /* Invalid command operation code */
594 sshdr
.asc
== 0x21 || /* Logical block address out of range */
595 sshdr
.asc
== 0x24 || /* Invalid field in cdb */
596 sshdr
.asc
== 0x26) { /* Parameter value invalid */
597 set_host_byte(scmd
, DID_TARGET_FAILURE
);
605 EXPORT_SYMBOL_GPL(scsi_check_sense
);
607 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
609 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
610 struct scsi_device
*tmp_sdev
;
612 if (!sht
->track_queue_depth
||
613 sdev
->queue_depth
>= sdev
->max_queue_depth
)
616 if (time_before(jiffies
,
617 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
620 if (time_before(jiffies
,
621 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
625 * Walk all devices of a target and do
628 shost_for_each_device(tmp_sdev
, sdev
->host
) {
629 if (tmp_sdev
->channel
!= sdev
->channel
||
630 tmp_sdev
->id
!= sdev
->id
||
631 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
634 scsi_change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1);
635 sdev
->last_queue_ramp_up
= jiffies
;
639 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
641 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
642 struct scsi_device
*tmp_sdev
;
644 if (!sht
->track_queue_depth
)
647 shost_for_each_device(tmp_sdev
, sdev
->host
) {
648 if (tmp_sdev
->channel
!= sdev
->channel
||
649 tmp_sdev
->id
!= sdev
->id
)
652 * We do not know the number of commands that were at
653 * the device when we got the queue full so we start
654 * from the highest possible value and work our way down.
656 scsi_track_queue_full(tmp_sdev
, tmp_sdev
->queue_depth
- 1);
661 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
662 * @scmd: SCSI cmd to examine.
665 * This is *only* called when we are examining the status of commands
666 * queued during error recovery. the main difference here is that we
667 * don't allow for the possibility of retries here, and we are a lot
668 * more restrictive about what we consider acceptable.
670 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
673 * first check the host byte, to see if there is anything in there
674 * that would indicate what we need to do.
676 if (host_byte(scmd
->result
) == DID_RESET
) {
678 * rats. we are already in the error handler, so we now
679 * get to try and figure out what to do next. if the sense
680 * is valid, we have a pretty good idea of what to do.
681 * if not, we mark it as FAILED.
683 return scsi_check_sense(scmd
);
685 if (host_byte(scmd
->result
) != DID_OK
)
689 * next, check the message byte.
691 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
695 * now, check the status byte to see if this indicates
698 switch (status_byte(scmd
->result
)) {
700 scsi_handle_queue_ramp_up(scmd
->device
);
701 case COMMAND_TERMINATED
:
703 case CHECK_CONDITION
:
704 return scsi_check_sense(scmd
);
706 case INTERMEDIATE_GOOD
:
707 case INTERMEDIATE_C_GOOD
:
709 * who knows? FIXME(eric)
712 case RESERVATION_CONFLICT
:
713 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
714 /* it is a success, we probed the device and
717 /* otherwise, we failed to send the command */
720 scsi_handle_queue_full(scmd
->device
);
731 * scsi_eh_done - Completion function for error handling.
732 * @scmd: Cmd that is done.
734 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
736 struct completion
*eh_action
;
738 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
739 "%s result: %x\n", __func__
, scmd
->result
));
741 eh_action
= scmd
->device
->host
->eh_action
;
747 * scsi_try_host_reset - ask host adapter to reset itself
748 * @scmd: SCSI cmd to send host reset.
750 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
754 struct Scsi_Host
*host
= scmd
->device
->host
;
755 struct scsi_host_template
*hostt
= host
->hostt
;
757 SCSI_LOG_ERROR_RECOVERY(3,
758 shost_printk(KERN_INFO
, host
, "Snd Host RST\n"));
760 if (!hostt
->eh_host_reset_handler
)
763 rtn
= hostt
->eh_host_reset_handler(scmd
);
765 if (rtn
== SUCCESS
) {
766 if (!hostt
->skip_settle_delay
)
767 ssleep(HOST_RESET_SETTLE_TIME
);
768 spin_lock_irqsave(host
->host_lock
, flags
);
769 scsi_report_bus_reset(host
, scmd_channel(scmd
));
770 spin_unlock_irqrestore(host
->host_lock
, flags
);
777 * scsi_try_bus_reset - ask host to perform a bus reset
778 * @scmd: SCSI cmd to send bus reset.
780 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
784 struct Scsi_Host
*host
= scmd
->device
->host
;
785 struct scsi_host_template
*hostt
= host
->hostt
;
787 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
788 "%s: Snd Bus RST\n", __func__
));
790 if (!hostt
->eh_bus_reset_handler
)
793 rtn
= hostt
->eh_bus_reset_handler(scmd
);
795 if (rtn
== SUCCESS
) {
796 if (!hostt
->skip_settle_delay
)
797 ssleep(BUS_RESET_SETTLE_TIME
);
798 spin_lock_irqsave(host
->host_lock
, flags
);
799 scsi_report_bus_reset(host
, scmd_channel(scmd
));
800 spin_unlock_irqrestore(host
->host_lock
, flags
);
806 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
809 sdev
->expecting_cc_ua
= 1;
813 * scsi_try_target_reset - Ask host to perform a target reset
814 * @scmd: SCSI cmd used to send a target reset
817 * There is no timeout for this operation. if this operation is
818 * unreliable for a given host, then the host itself needs to put a
819 * timer on it, and set the host back to a consistent state prior to
822 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
826 struct Scsi_Host
*host
= scmd
->device
->host
;
827 struct scsi_host_template
*hostt
= host
->hostt
;
829 if (!hostt
->eh_target_reset_handler
)
832 rtn
= hostt
->eh_target_reset_handler(scmd
);
833 if (rtn
== SUCCESS
) {
834 spin_lock_irqsave(host
->host_lock
, flags
);
835 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
836 __scsi_report_device_reset
);
837 spin_unlock_irqrestore(host
->host_lock
, flags
);
844 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
845 * @scmd: SCSI cmd used to send BDR
848 * There is no timeout for this operation. if this operation is
849 * unreliable for a given host, then the host itself needs to put a
850 * timer on it, and set the host back to a consistent state prior to
853 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
856 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
858 if (!hostt
->eh_device_reset_handler
)
861 rtn
= hostt
->eh_device_reset_handler(scmd
);
863 __scsi_report_device_reset(scmd
->device
, NULL
);
868 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
869 * @hostt: SCSI driver host template
870 * @scmd: SCSI cmd used to send a target reset
873 * SUCCESS, FAILED, or FAST_IO_FAIL
876 * SUCCESS does not necessarily indicate that the command
877 * has been aborted; it only indicates that the LLDDs
878 * has cleared all references to that command.
879 * LLDDs should return FAILED only if an abort was required
880 * but could not be executed. LLDDs should return FAST_IO_FAIL
881 * if the device is temporarily unavailable (eg due to a
882 * link down on FibreChannel)
884 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
,
885 struct scsi_cmnd
*scmd
)
887 if (!hostt
->eh_abort_handler
)
890 return hostt
->eh_abort_handler(scmd
);
893 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
895 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
896 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
897 if (scsi_try_target_reset(scmd
) != SUCCESS
)
898 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
899 scsi_try_host_reset(scmd
);
903 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
904 * @scmd: SCSI command structure to hijack
905 * @ses: structure to save restore information
906 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
907 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
908 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
910 * This function is used to save a scsi command information before re-execution
911 * as part of the error recovery process. If @sense_bytes is 0 the command
912 * sent must be one that does not transfer any data. If @sense_bytes != 0
913 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
914 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
916 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
917 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
919 struct scsi_device
*sdev
= scmd
->device
;
922 * We need saved copies of a number of fields - this is because
923 * error handling may need to overwrite these with different values
924 * to run different commands, and once error handling is complete,
925 * we will need to restore these values prior to running the actual
928 ses
->cmd_len
= scmd
->cmd_len
;
929 ses
->cmnd
= scmd
->cmnd
;
930 ses
->data_direction
= scmd
->sc_data_direction
;
931 ses
->sdb
= scmd
->sdb
;
932 ses
->next_rq
= scmd
->request
->next_rq
;
933 ses
->result
= scmd
->result
;
934 ses
->underflow
= scmd
->underflow
;
935 ses
->prot_op
= scmd
->prot_op
;
937 scmd
->prot_op
= SCSI_PROT_NORMAL
;
939 scmd
->cmnd
= ses
->eh_cmnd
;
940 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
941 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
942 scmd
->request
->next_rq
= NULL
;
946 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
948 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
950 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
951 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
952 scmd
->sdb
.table
.nents
= scmd
->sdb
.table
.orig_nents
= 1;
953 scmd
->cmnd
[0] = REQUEST_SENSE
;
954 scmd
->cmnd
[4] = scmd
->sdb
.length
;
955 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
957 scmd
->sc_data_direction
= DMA_NONE
;
959 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
960 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
961 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
967 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
968 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
969 (sdev
->lun
<< 5 & 0xe0);
972 * Zero the sense buffer. The scsi spec mandates that any
973 * untransferred sense data should be interpreted as being zero.
975 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
977 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
980 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
981 * @scmd: SCSI command structure to restore
982 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
984 * Undo any damage done by above scsi_eh_prep_cmnd().
986 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
989 * Restore original data
991 scmd
->cmd_len
= ses
->cmd_len
;
992 scmd
->cmnd
= ses
->cmnd
;
993 scmd
->sc_data_direction
= ses
->data_direction
;
994 scmd
->sdb
= ses
->sdb
;
995 scmd
->request
->next_rq
= ses
->next_rq
;
996 scmd
->result
= ses
->result
;
997 scmd
->underflow
= ses
->underflow
;
998 scmd
->prot_op
= ses
->prot_op
;
1000 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
1003 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1004 * @scmd: SCSI command structure to hijack
1005 * @cmnd: CDB to send
1006 * @cmnd_size: size in bytes of @cmnd
1007 * @timeout: timeout for this request
1008 * @sense_bytes: size of sense data to copy or 0
1010 * This function is used to send a scsi command down to a target device
1011 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1014 * SUCCESS or FAILED or NEEDS_RETRY
1016 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
1017 int cmnd_size
, int timeout
, unsigned sense_bytes
)
1019 struct scsi_device
*sdev
= scmd
->device
;
1020 struct Scsi_Host
*shost
= sdev
->host
;
1021 DECLARE_COMPLETION_ONSTACK(done
);
1022 unsigned long timeleft
= timeout
;
1023 struct scsi_eh_save ses
;
1024 const unsigned long stall_for
= msecs_to_jiffies(100);
1028 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
1029 shost
->eh_action
= &done
;
1031 scsi_log_send(scmd
);
1032 scmd
->scsi_done
= scsi_eh_done
;
1033 rtn
= shost
->hostt
->queuecommand(shost
, scmd
);
1035 if (timeleft
> stall_for
) {
1036 scsi_eh_restore_cmnd(scmd
, &ses
);
1037 timeleft
-= stall_for
;
1038 msleep(jiffies_to_msecs(stall_for
));
1041 /* signal not to enter either branch of the if () below */
1045 timeleft
= wait_for_completion_timeout(&done
, timeout
);
1049 shost
->eh_action
= NULL
;
1051 scsi_log_completion(scmd
, rtn
);
1053 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1054 "%s timeleft: %ld\n",
1055 __func__
, timeleft
));
1058 * If there is time left scsi_eh_done got called, and we will examine
1059 * the actual status codes to see whether the command actually did
1060 * complete normally, else if we have a zero return and no time left,
1061 * the command must still be pending, so abort it and return FAILED.
1062 * If we never actually managed to issue the command, because
1063 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1064 * value above (so don't execute either branch of the if)
1067 rtn
= scsi_eh_completed_normally(scmd
);
1068 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1069 "%s: scsi_eh_completed_normally %x\n", __func__
, rtn
));
1076 case ADD_TO_MLQUEUE
:
1083 } else if (rtn
!= FAILED
) {
1084 scsi_abort_eh_cmnd(scmd
);
1088 scsi_eh_restore_cmnd(scmd
, &ses
);
1094 * scsi_request_sense - Request sense data from a particular target.
1095 * @scmd: SCSI cmd for request sense.
1098 * Some hosts automatically obtain this information, others require
1099 * that we obtain it on our own. This function will *not* return until
1100 * the command either times out, or it completes.
1102 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
1104 return scsi_send_eh_cmnd(scmd
, NULL
, 0, scmd
->device
->eh_timeout
, ~0);
1107 static int scsi_eh_action(struct scsi_cmnd
*scmd
, int rtn
)
1109 if (scmd
->request
->cmd_type
!= REQ_TYPE_BLOCK_PC
) {
1110 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
1111 if (sdrv
->eh_action
)
1112 rtn
= sdrv
->eh_action(scmd
, rtn
);
1118 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1119 * @scmd: Original SCSI cmd that eh has finished.
1120 * @done_q: Queue for processed commands.
1123 * We don't want to use the normal command completion while we are are
1124 * still handling errors - it may cause other commands to be queued,
1125 * and that would disturb what we are doing. Thus we really want to
1126 * keep a list of pending commands for final completion, and once we
1127 * are ready to leave error handling we handle completion for real.
1129 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
1131 scmd
->device
->host
->host_failed
--;
1132 scmd
->eh_eflags
= 0;
1133 list_move_tail(&scmd
->eh_entry
, done_q
);
1135 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
1138 * scsi_eh_get_sense - Get device sense data.
1139 * @work_q: Queue of commands to process.
1140 * @done_q: Queue of processed commands.
1143 * See if we need to request sense information. if so, then get it
1144 * now, so we have a better idea of what to do.
1147 * This has the unfortunate side effect that if a shost adapter does
1148 * not automatically request sense information, we end up shutting
1149 * it down before we request it.
1151 * All drivers should request sense information internally these days,
1152 * so for now all I have to say is tough noogies if you end up in here.
1154 * XXX: Long term this code should go away, but that needs an audit of
1157 int scsi_eh_get_sense(struct list_head
*work_q
,
1158 struct list_head
*done_q
)
1160 struct scsi_cmnd
*scmd
, *next
;
1161 struct Scsi_Host
*shost
;
1165 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1166 * should not get sense.
1168 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1169 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
1170 (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) ||
1171 SCSI_SENSE_VALID(scmd
))
1174 shost
= scmd
->device
->host
;
1175 if (scsi_host_eh_past_deadline(shost
)) {
1176 SCSI_LOG_ERROR_RECOVERY(3,
1177 scmd_printk(KERN_INFO
, scmd
,
1178 "%s: skip request sense, past eh deadline\n",
1182 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1184 * don't request sense if there's no check condition
1185 * status because the error we're processing isn't one
1186 * that has a sense code (and some devices get
1187 * confused by sense requests out of the blue)
1191 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
1192 "%s: requesting sense\n",
1194 rtn
= scsi_request_sense(scmd
);
1198 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1199 "sense requested, result %x\n", scmd
->result
));
1200 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd
));
1202 rtn
= scsi_decide_disposition(scmd
);
1205 * if the result was normal, then just pass it along to the
1209 /* we don't want this command reissued, just
1210 * finished with the sense data, so set
1211 * retries to the max allowed to ensure it
1212 * won't get reissued */
1213 scmd
->retries
= scmd
->allowed
;
1214 else if (rtn
!= NEEDS_RETRY
)
1217 scsi_eh_finish_cmd(scmd
, done_q
);
1220 return list_empty(work_q
);
1222 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
1225 * scsi_eh_tur - Send TUR to device.
1226 * @scmd: &scsi_cmnd to send TUR
1229 * 0 - Device is ready. 1 - Device NOT ready.
1231 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
1233 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
1234 int retry_cnt
= 1, rtn
;
1237 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6,
1238 scmd
->device
->eh_timeout
, 0);
1240 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO
, scmd
,
1241 "%s return: %x\n", __func__
, rtn
));
1256 * scsi_eh_test_devices - check if devices are responding from error recovery.
1257 * @cmd_list: scsi commands in error recovery.
1258 * @work_q: queue for commands which still need more error recovery
1259 * @done_q: queue for commands which are finished
1260 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1263 * Tests if devices are in a working state. Commands to devices now in
1264 * a working state are sent to the done_q while commands to devices which
1265 * are still failing to respond are returned to the work_q for more
1268 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
1269 struct list_head
*work_q
,
1270 struct list_head
*done_q
, int try_stu
)
1272 struct scsi_cmnd
*scmd
, *next
;
1273 struct scsi_device
*sdev
;
1276 while (!list_empty(cmd_list
)) {
1277 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
1278 sdev
= scmd
->device
;
1281 if (scsi_host_eh_past_deadline(sdev
->host
)) {
1282 /* Push items back onto work_q */
1283 list_splice_init(cmd_list
, work_q
);
1284 SCSI_LOG_ERROR_RECOVERY(3,
1285 sdev_printk(KERN_INFO
, sdev
,
1286 "%s: skip test device, past eh deadline",
1292 finish_cmds
= !scsi_device_online(scmd
->device
) ||
1293 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
1294 !scsi_eh_tur(scmd
)) ||
1297 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
1298 if (scmd
->device
== sdev
) {
1301 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
))
1302 scsi_eh_finish_cmd(scmd
, done_q
);
1304 list_move_tail(&scmd
->eh_entry
, work_q
);
1307 return list_empty(work_q
);
1312 * scsi_eh_abort_cmds - abort pending commands.
1313 * @work_q: &list_head for pending commands.
1314 * @done_q: &list_head for processed commands.
1317 * Try and see whether or not it makes sense to try and abort the
1318 * running command. This only works out to be the case if we have one
1319 * command that has timed out. If the command simply failed, it makes
1320 * no sense to try and abort the command, since as far as the shost
1321 * adapter is concerned, it isn't running.
1323 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
1324 struct list_head
*done_q
)
1326 struct scsi_cmnd
*scmd
, *next
;
1327 LIST_HEAD(check_list
);
1329 struct Scsi_Host
*shost
;
1331 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1332 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
1334 shost
= scmd
->device
->host
;
1335 if (scsi_host_eh_past_deadline(shost
)) {
1336 list_splice_init(&check_list
, work_q
);
1337 SCSI_LOG_ERROR_RECOVERY(3,
1338 scmd_printk(KERN_INFO
, scmd
,
1339 "%s: skip aborting cmd, past eh deadline\n",
1341 return list_empty(work_q
);
1343 SCSI_LOG_ERROR_RECOVERY(3,
1344 scmd_printk(KERN_INFO
, scmd
,
1345 "%s: aborting cmd\n", current
->comm
));
1346 rtn
= scsi_try_to_abort_cmd(shost
->hostt
, scmd
);
1347 if (rtn
== FAILED
) {
1348 SCSI_LOG_ERROR_RECOVERY(3,
1349 scmd_printk(KERN_INFO
, scmd
,
1350 "%s: aborting cmd failed\n",
1352 list_splice_init(&check_list
, work_q
);
1353 return list_empty(work_q
);
1355 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
1356 if (rtn
== FAST_IO_FAIL
)
1357 scsi_eh_finish_cmd(scmd
, done_q
);
1359 list_move_tail(&scmd
->eh_entry
, &check_list
);
1362 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1366 * scsi_eh_try_stu - Send START_UNIT to device.
1367 * @scmd: &scsi_cmnd to send START_UNIT
1370 * 0 - Device is ready. 1 - Device NOT ready.
1372 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1374 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1376 if (scmd
->device
->allow_restart
) {
1377 int i
, rtn
= NEEDS_RETRY
;
1379 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1380 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1390 * scsi_eh_stu - send START_UNIT if needed
1391 * @shost: &scsi host being recovered.
1392 * @work_q: &list_head for pending commands.
1393 * @done_q: &list_head for processed commands.
1396 * If commands are failing due to not ready, initializing command required,
1397 * try revalidating the device, which will end up sending a start unit.
1399 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1400 struct list_head
*work_q
,
1401 struct list_head
*done_q
)
1403 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1404 struct scsi_device
*sdev
;
1406 shost_for_each_device(sdev
, shost
) {
1407 if (scsi_host_eh_past_deadline(shost
)) {
1408 SCSI_LOG_ERROR_RECOVERY(3,
1409 sdev_printk(KERN_INFO
, sdev
,
1410 "%s: skip START_UNIT, past eh deadline\n",
1415 list_for_each_entry(scmd
, work_q
, eh_entry
)
1416 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1417 scsi_check_sense(scmd
) == FAILED
) {
1425 SCSI_LOG_ERROR_RECOVERY(3,
1426 sdev_printk(KERN_INFO
, sdev
,
1427 "%s: Sending START_UNIT\n",
1430 if (!scsi_eh_try_stu(stu_scmd
)) {
1431 if (!scsi_device_online(sdev
) ||
1432 !scsi_eh_tur(stu_scmd
)) {
1433 list_for_each_entry_safe(scmd
, next
,
1435 if (scmd
->device
== sdev
&&
1436 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
)
1437 scsi_eh_finish_cmd(scmd
, done_q
);
1441 SCSI_LOG_ERROR_RECOVERY(3,
1442 sdev_printk(KERN_INFO
, sdev
,
1443 "%s: START_UNIT failed\n",
1448 return list_empty(work_q
);
1453 * scsi_eh_bus_device_reset - send bdr if needed
1454 * @shost: scsi host being recovered.
1455 * @work_q: &list_head for pending commands.
1456 * @done_q: &list_head for processed commands.
1459 * Try a bus device reset. Still, look to see whether we have multiple
1460 * devices that are jammed or not - if we have multiple devices, it
1461 * makes no sense to try bus_device_reset - we really would need to try
1462 * a bus_reset instead.
1464 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1465 struct list_head
*work_q
,
1466 struct list_head
*done_q
)
1468 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1469 struct scsi_device
*sdev
;
1472 shost_for_each_device(sdev
, shost
) {
1473 if (scsi_host_eh_past_deadline(shost
)) {
1474 SCSI_LOG_ERROR_RECOVERY(3,
1475 sdev_printk(KERN_INFO
, sdev
,
1476 "%s: skip BDR, past eh deadline\n",
1481 list_for_each_entry(scmd
, work_q
, eh_entry
)
1482 if (scmd
->device
== sdev
) {
1490 SCSI_LOG_ERROR_RECOVERY(3,
1491 sdev_printk(KERN_INFO
, sdev
,
1492 "%s: Sending BDR\n", current
->comm
));
1493 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1494 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1495 if (!scsi_device_online(sdev
) ||
1496 rtn
== FAST_IO_FAIL
||
1497 !scsi_eh_tur(bdr_scmd
)) {
1498 list_for_each_entry_safe(scmd
, next
,
1500 if (scmd
->device
== sdev
&&
1501 scsi_eh_action(scmd
, rtn
) != FAILED
)
1502 scsi_eh_finish_cmd(scmd
,
1507 SCSI_LOG_ERROR_RECOVERY(3,
1508 sdev_printk(KERN_INFO
, sdev
,
1509 "%s: BDR failed\n", current
->comm
));
1513 return list_empty(work_q
);
1517 * scsi_eh_target_reset - send target reset if needed
1518 * @shost: scsi host being recovered.
1519 * @work_q: &list_head for pending commands.
1520 * @done_q: &list_head for processed commands.
1523 * Try a target reset.
1525 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1526 struct list_head
*work_q
,
1527 struct list_head
*done_q
)
1529 LIST_HEAD(tmp_list
);
1530 LIST_HEAD(check_list
);
1532 list_splice_init(work_q
, &tmp_list
);
1534 while (!list_empty(&tmp_list
)) {
1535 struct scsi_cmnd
*next
, *scmd
;
1539 if (scsi_host_eh_past_deadline(shost
)) {
1540 /* push back on work queue for further processing */
1541 list_splice_init(&check_list
, work_q
);
1542 list_splice_init(&tmp_list
, work_q
);
1543 SCSI_LOG_ERROR_RECOVERY(3,
1544 shost_printk(KERN_INFO
, shost
,
1545 "%s: Skip target reset, past eh deadline\n",
1547 return list_empty(work_q
);
1550 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1553 SCSI_LOG_ERROR_RECOVERY(3,
1554 shost_printk(KERN_INFO
, shost
,
1555 "%s: Sending target reset to target %d\n",
1556 current
->comm
, id
));
1557 rtn
= scsi_try_target_reset(scmd
);
1558 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1559 SCSI_LOG_ERROR_RECOVERY(3,
1560 shost_printk(KERN_INFO
, shost
,
1561 "%s: Target reset failed"
1563 current
->comm
, id
));
1564 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1565 if (scmd_id(scmd
) != id
)
1569 list_move_tail(&scmd
->eh_entry
, &check_list
);
1570 else if (rtn
== FAST_IO_FAIL
)
1571 scsi_eh_finish_cmd(scmd
, done_q
);
1573 /* push back on work queue for further processing */
1574 list_move(&scmd
->eh_entry
, work_q
);
1578 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1582 * scsi_eh_bus_reset - send a bus reset
1583 * @shost: &scsi host being recovered.
1584 * @work_q: &list_head for pending commands.
1585 * @done_q: &list_head for processed commands.
1587 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1588 struct list_head
*work_q
,
1589 struct list_head
*done_q
)
1591 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1592 LIST_HEAD(check_list
);
1593 unsigned int channel
;
1597 * we really want to loop over the various channels, and do this on
1598 * a channel by channel basis. we should also check to see if any
1599 * of the failed commands are on soft_reset devices, and if so, skip
1603 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1604 if (scsi_host_eh_past_deadline(shost
)) {
1605 list_splice_init(&check_list
, work_q
);
1606 SCSI_LOG_ERROR_RECOVERY(3,
1607 shost_printk(KERN_INFO
, shost
,
1608 "%s: skip BRST, past eh deadline\n",
1610 return list_empty(work_q
);
1614 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1615 if (channel
== scmd_channel(scmd
)) {
1619 * FIXME add back in some support for
1620 * soft_reset devices.
1627 SCSI_LOG_ERROR_RECOVERY(3,
1628 shost_printk(KERN_INFO
, shost
,
1629 "%s: Sending BRST chan: %d\n",
1630 current
->comm
, channel
));
1631 rtn
= scsi_try_bus_reset(chan_scmd
);
1632 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1633 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1634 if (channel
== scmd_channel(scmd
)) {
1635 if (rtn
== FAST_IO_FAIL
)
1636 scsi_eh_finish_cmd(scmd
,
1639 list_move_tail(&scmd
->eh_entry
,
1644 SCSI_LOG_ERROR_RECOVERY(3,
1645 shost_printk(KERN_INFO
, shost
,
1646 "%s: BRST failed chan: %d\n",
1647 current
->comm
, channel
));
1650 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1654 * scsi_eh_host_reset - send a host reset
1655 * @shost: host to be reset.
1656 * @work_q: &list_head for pending commands.
1657 * @done_q: &list_head for processed commands.
1659 static int scsi_eh_host_reset(struct Scsi_Host
*shost
,
1660 struct list_head
*work_q
,
1661 struct list_head
*done_q
)
1663 struct scsi_cmnd
*scmd
, *next
;
1664 LIST_HEAD(check_list
);
1667 if (!list_empty(work_q
)) {
1668 scmd
= list_entry(work_q
->next
,
1669 struct scsi_cmnd
, eh_entry
);
1671 SCSI_LOG_ERROR_RECOVERY(3,
1672 shost_printk(KERN_INFO
, shost
,
1673 "%s: Sending HRST\n",
1676 rtn
= scsi_try_host_reset(scmd
);
1677 if (rtn
== SUCCESS
) {
1678 list_splice_init(work_q
, &check_list
);
1679 } else if (rtn
== FAST_IO_FAIL
) {
1680 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1681 scsi_eh_finish_cmd(scmd
, done_q
);
1684 SCSI_LOG_ERROR_RECOVERY(3,
1685 shost_printk(KERN_INFO
, shost
,
1686 "%s: HRST failed\n",
1690 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1694 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1695 * @work_q: &list_head for pending commands.
1696 * @done_q: &list_head for processed commands.
1698 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1699 struct list_head
*done_q
)
1701 struct scsi_cmnd
*scmd
, *next
;
1703 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1704 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1705 "not ready after error recovery\n");
1706 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1707 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1709 * FIXME: Handle lost cmds.
1712 scsi_eh_finish_cmd(scmd
, done_q
);
1718 * scsi_noretry_cmd - determine if command should be failed fast
1719 * @scmd: SCSI cmd to examine.
1721 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1723 switch (host_byte(scmd
->result
)) {
1729 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1731 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1733 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1734 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1737 case DID_SOFT_ERROR
:
1738 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1741 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1746 * assume caller has checked sense and determined
1747 * the check condition was retryable.
1749 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1750 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1757 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1758 * @scmd: SCSI cmd to examine.
1761 * This is *only* called when we are examining the status after sending
1762 * out the actual data command. any commands that are queued for error
1763 * recovery (e.g. test_unit_ready) do *not* come through here.
1765 * When this routine returns failed, it means the error handler thread
1766 * is woken. In cases where the error code indicates an error that
1767 * doesn't require the error handler read (i.e. we don't need to
1768 * abort/reset), this function should return SUCCESS.
1770 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1775 * if the device is offline, then we clearly just pass the result back
1776 * up to the top level.
1778 if (!scsi_device_online(scmd
->device
)) {
1779 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO
, scmd
,
1780 "%s: device offline - report as SUCCESS\n", __func__
));
1785 * first check the host byte, to see if there is anything in there
1786 * that would indicate what we need to do.
1788 switch (host_byte(scmd
->result
)) {
1789 case DID_PASSTHROUGH
:
1791 * no matter what, pass this through to the upper layer.
1792 * nuke this special code so that it looks like we are saying
1795 scmd
->result
&= 0xff00ffff;
1799 * looks good. drop through, and check the next byte.
1803 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
1804 set_host_byte(scmd
, DID_TIME_OUT
);
1807 case DID_NO_CONNECT
:
1808 case DID_BAD_TARGET
:
1810 * note - this means that we just report the status back
1811 * to the top level driver, not that we actually think
1812 * that it indicates SUCCESS.
1816 * when the low level driver returns did_soft_error,
1817 * it is responsible for keeping an internal retry counter
1818 * in order to avoid endless loops (db)
1820 * actually this is a bug in this function here. we should
1821 * be mindful of the maximum number of retries specified
1822 * and not get stuck in a loop.
1824 case DID_SOFT_ERROR
:
1830 return ADD_TO_MLQUEUE
;
1831 case DID_TRANSPORT_DISRUPTED
:
1833 * LLD/transport was disrupted during processing of the IO.
1834 * The transport class is now blocked/blocking,
1835 * and the transport will decide what to do with the IO
1836 * based on its timers and recovery capablilities if
1837 * there are enough retries.
1840 case DID_TRANSPORT_FAILFAST
:
1842 * The transport decided to failfast the IO (most likely
1843 * the fast io fail tmo fired), so send IO directly upwards.
1847 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1848 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1850 * execute reservation conflict processing code
1860 * when we scan the bus, we get timeout messages for
1861 * these commands if there is no device available.
1862 * other hosts report did_no_connect for the same thing.
1864 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1865 scmd
->cmnd
[0] == INQUIRY
)) {
1877 * next, check the message byte.
1879 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1883 * check the status byte to see if this indicates anything special.
1885 switch (status_byte(scmd
->result
)) {
1887 scsi_handle_queue_full(scmd
->device
);
1889 * the case of trying to send too many commands to a
1890 * tagged queueing device.
1894 * device can't talk to us at the moment. Should only
1895 * occur (SAM-3) when the task queue is empty, so will cause
1896 * the empty queue handling to trigger a stall in the
1899 return ADD_TO_MLQUEUE
;
1901 if (scmd
->cmnd
[0] == REPORT_LUNS
)
1902 scmd
->device
->sdev_target
->expecting_lun_change
= 0;
1903 scsi_handle_queue_ramp_up(scmd
->device
);
1904 case COMMAND_TERMINATED
:
1908 case CHECK_CONDITION
:
1909 rtn
= scsi_check_sense(scmd
);
1910 if (rtn
== NEEDS_RETRY
)
1912 /* if rtn == FAILED, we have no sense information;
1913 * returning FAILED will wake the error handler thread
1914 * to collect the sense and redo the decide
1917 case CONDITION_GOOD
:
1918 case INTERMEDIATE_GOOD
:
1919 case INTERMEDIATE_C_GOOD
:
1922 * who knows? FIXME(eric)
1926 case RESERVATION_CONFLICT
:
1927 sdev_printk(KERN_INFO
, scmd
->device
,
1928 "reservation conflict\n");
1929 set_host_byte(scmd
, DID_NEXUS_FAILURE
);
1930 return SUCCESS
; /* causes immediate i/o error */
1938 /* we requeue for retry because the error was retryable, and
1939 * the request was not marked fast fail. Note that above,
1940 * even if the request is marked fast fail, we still requeue
1941 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1942 if ((++scmd
->retries
) <= scmd
->allowed
1943 && !scsi_noretry_cmd(scmd
)) {
1947 * no more retries - report this one back to upper level.
1953 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1955 __blk_put_request(req
->q
, req
);
1959 * scsi_eh_lock_door - Prevent medium removal for the specified device
1960 * @sdev: SCSI device to prevent medium removal
1963 * We must be called from process context.
1966 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1967 * head of the devices request queue, and continue.
1969 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1971 struct request
*req
;
1974 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
1975 * request becomes available
1977 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1981 blk_rq_set_block_pc(req
);
1983 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1987 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1990 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1992 req
->cmd_flags
|= REQ_QUIET
;
1993 req
->timeout
= 10 * HZ
;
1996 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
2000 * scsi_restart_operations - restart io operations to the specified host.
2001 * @shost: Host we are restarting.
2004 * When we entered the error handler, we blocked all further i/o to
2005 * this device. we need to 'reverse' this process.
2007 static void scsi_restart_operations(struct Scsi_Host
*shost
)
2009 struct scsi_device
*sdev
;
2010 unsigned long flags
;
2013 * If the door was locked, we need to insert a door lock request
2014 * onto the head of the SCSI request queue for the device. There
2015 * is no point trying to lock the door of an off-line device.
2017 shost_for_each_device(sdev
, shost
) {
2018 if (scsi_device_online(sdev
) && sdev
->was_reset
&& sdev
->locked
) {
2019 scsi_eh_lock_door(sdev
);
2020 sdev
->was_reset
= 0;
2025 * next free up anything directly waiting upon the host. this
2026 * will be requests for character device operations, and also for
2027 * ioctls to queued block devices.
2029 SCSI_LOG_ERROR_RECOVERY(3,
2030 shost_printk(KERN_INFO
, shost
, "waking up host to restart\n"));
2032 spin_lock_irqsave(shost
->host_lock
, flags
);
2033 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
2034 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
2035 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
2036 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2038 wake_up(&shost
->host_wait
);
2041 * finally we need to re-initiate requests that may be pending. we will
2042 * have had everything blocked while error handling is taking place, and
2043 * now that error recovery is done, we will need to ensure that these
2044 * requests are started.
2046 scsi_run_host_queues(shost
);
2049 * if eh is active and host_eh_scheduled is pending we need to re-run
2050 * recovery. we do this check after scsi_run_host_queues() to allow
2051 * everything pent up since the last eh run a chance to make forward
2052 * progress before we sync again. Either we'll immediately re-run
2053 * recovery or scsi_device_unbusy() will wake us again when these
2054 * pending commands complete.
2056 spin_lock_irqsave(shost
->host_lock
, flags
);
2057 if (shost
->host_eh_scheduled
)
2058 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
2059 WARN_ON(scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
));
2060 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2064 * scsi_eh_ready_devs - check device ready state and recover if not.
2065 * @shost: host to be recovered.
2066 * @work_q: &list_head for pending commands.
2067 * @done_q: &list_head for processed commands.
2069 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
2070 struct list_head
*work_q
,
2071 struct list_head
*done_q
)
2073 if (!scsi_eh_stu(shost
, work_q
, done_q
))
2074 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
2075 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
2076 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
2077 if (!scsi_eh_host_reset(shost
, work_q
, done_q
))
2078 scsi_eh_offline_sdevs(work_q
,
2081 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
2084 * scsi_eh_flush_done_q - finish processed commands or retry them.
2085 * @done_q: list_head of processed commands.
2087 void scsi_eh_flush_done_q(struct list_head
*done_q
)
2089 struct scsi_cmnd
*scmd
, *next
;
2091 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
2092 list_del_init(&scmd
->eh_entry
);
2093 if (scsi_device_online(scmd
->device
) &&
2094 !scsi_noretry_cmd(scmd
) &&
2095 (++scmd
->retries
<= scmd
->allowed
)) {
2096 SCSI_LOG_ERROR_RECOVERY(3,
2097 scmd_printk(KERN_INFO
, scmd
,
2098 "%s: flush retry cmd\n",
2100 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
2103 * If just we got sense for the device (called
2104 * scsi_eh_get_sense), scmd->result is already
2105 * set, do not set DRIVER_TIMEOUT.
2108 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
2109 SCSI_LOG_ERROR_RECOVERY(3,
2110 scmd_printk(KERN_INFO
, scmd
,
2111 "%s: flush finish cmd\n",
2113 scsi_finish_command(scmd
);
2117 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
2120 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2121 * @shost: Host to unjam.
2124 * When we come in here, we *know* that all commands on the bus have
2125 * either completed, failed or timed out. we also know that no further
2126 * commands are being sent to the host, so things are relatively quiet
2127 * and we have freedom to fiddle with things as we wish.
2129 * This is only the *default* implementation. it is possible for
2130 * individual drivers to supply their own version of this function, and
2131 * if the maintainer wishes to do this, it is strongly suggested that
2132 * this function be taken as a template and modified. this function
2133 * was designed to correctly handle problems for about 95% of the
2134 * different cases out there, and it should always provide at least a
2135 * reasonable amount of error recovery.
2137 * Any command marked 'failed' or 'timeout' must eventually have
2138 * scsi_finish_cmd() called for it. we do all of the retry stuff
2139 * here, so when we restart the host after we return it should have an
2142 static void scsi_unjam_host(struct Scsi_Host
*shost
)
2144 unsigned long flags
;
2145 LIST_HEAD(eh_work_q
);
2146 LIST_HEAD(eh_done_q
);
2148 spin_lock_irqsave(shost
->host_lock
, flags
);
2149 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
2150 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2152 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
2154 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
2155 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
2156 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
2158 spin_lock_irqsave(shost
->host_lock
, flags
);
2159 if (shost
->eh_deadline
!= -1)
2160 shost
->last_reset
= 0;
2161 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2162 scsi_eh_flush_done_q(&eh_done_q
);
2166 * scsi_error_handler - SCSI error handler thread
2167 * @data: Host for which we are running.
2170 * This is the main error handling loop. This is run as a kernel thread
2171 * for every SCSI host and handles all error handling activity.
2173 int scsi_error_handler(void *data
)
2175 struct Scsi_Host
*shost
= data
;
2178 * We use TASK_INTERRUPTIBLE so that the thread is not
2179 * counted against the load average as a running process.
2180 * We never actually get interrupted because kthread_run
2181 * disables signal delivery for the created thread.
2185 * The sequence in kthread_stop() sets the stop flag first
2186 * then wakes the process. To avoid missed wakeups, the task
2187 * should always be in a non running state before the stop
2190 set_current_state(TASK_INTERRUPTIBLE
);
2191 if (kthread_should_stop())
2194 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
2195 shost
->host_failed
!= atomic_read(&shost
->host_busy
)) {
2196 SCSI_LOG_ERROR_RECOVERY(1,
2197 shost_printk(KERN_INFO
, shost
,
2198 "scsi_eh_%d: sleeping\n",
2204 __set_current_state(TASK_RUNNING
);
2205 SCSI_LOG_ERROR_RECOVERY(1,
2206 shost_printk(KERN_INFO
, shost
,
2207 "scsi_eh_%d: waking up %d/%d/%d\n",
2208 shost
->host_no
, shost
->host_eh_scheduled
,
2210 atomic_read(&shost
->host_busy
)));
2213 * We have a host that is failing for some reason. Figure out
2214 * what we need to do to get it up and online again (if we can).
2215 * If we fail, we end up taking the thing offline.
2217 if (!shost
->eh_noresume
&& scsi_autopm_get_host(shost
) != 0) {
2218 SCSI_LOG_ERROR_RECOVERY(1,
2219 shost_printk(KERN_ERR
, shost
,
2220 "scsi_eh_%d: unable to autoresume\n",
2225 if (shost
->transportt
->eh_strategy_handler
)
2226 shost
->transportt
->eh_strategy_handler(shost
);
2228 scsi_unjam_host(shost
);
2231 * Note - if the above fails completely, the action is to take
2232 * individual devices offline and flush the queue of any
2233 * outstanding requests that may have been pending. When we
2234 * restart, we restart any I/O to any other devices on the bus
2235 * which are still online.
2237 scsi_restart_operations(shost
);
2238 if (!shost
->eh_noresume
)
2239 scsi_autopm_put_host(shost
);
2241 __set_current_state(TASK_RUNNING
);
2243 SCSI_LOG_ERROR_RECOVERY(1,
2244 shost_printk(KERN_INFO
, shost
,
2245 "Error handler scsi_eh_%d exiting\n",
2247 shost
->ehandler
= NULL
;
2252 * Function: scsi_report_bus_reset()
2254 * Purpose: Utility function used by low-level drivers to report that
2255 * they have observed a bus reset on the bus being handled.
2257 * Arguments: shost - Host in question
2258 * channel - channel on which reset was observed.
2262 * Lock status: Host lock must be held.
2264 * Notes: This only needs to be called if the reset is one which
2265 * originates from an unknown location. Resets originated
2266 * by the mid-level itself don't need to call this, but there
2267 * should be no harm.
2269 * The main purpose of this is to make sure that a CHECK_CONDITION
2270 * is properly treated.
2272 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
2274 struct scsi_device
*sdev
;
2276 __shost_for_each_device(sdev
, shost
) {
2277 if (channel
== sdev_channel(sdev
))
2278 __scsi_report_device_reset(sdev
, NULL
);
2281 EXPORT_SYMBOL(scsi_report_bus_reset
);
2284 * Function: scsi_report_device_reset()
2286 * Purpose: Utility function used by low-level drivers to report that
2287 * they have observed a device reset on the device being handled.
2289 * Arguments: shost - Host in question
2290 * channel - channel on which reset was observed
2291 * target - target on which reset was observed
2295 * Lock status: Host lock must be held
2297 * Notes: This only needs to be called if the reset is one which
2298 * originates from an unknown location. Resets originated
2299 * by the mid-level itself don't need to call this, but there
2300 * should be no harm.
2302 * The main purpose of this is to make sure that a CHECK_CONDITION
2303 * is properly treated.
2305 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
2307 struct scsi_device
*sdev
;
2309 __shost_for_each_device(sdev
, shost
) {
2310 if (channel
== sdev_channel(sdev
) &&
2311 target
== sdev_id(sdev
))
2312 __scsi_report_device_reset(sdev
, NULL
);
2315 EXPORT_SYMBOL(scsi_report_device_reset
);
2318 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
2323 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2324 * @dev: scsi_device to operate on
2325 * @arg: reset type (see sg.h)
2328 scsi_ioctl_reset(struct scsi_device
*dev
, int __user
*arg
)
2330 struct scsi_cmnd
*scmd
;
2331 struct Scsi_Host
*shost
= dev
->host
;
2333 unsigned long flags
;
2334 int error
= 0, rtn
, val
;
2336 if (!capable(CAP_SYS_ADMIN
) || !capable(CAP_SYS_RAWIO
))
2339 error
= get_user(val
, arg
);
2343 if (scsi_autopm_get_host(shost
) < 0)
2347 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
2349 goto out_put_autopm_host
;
2351 blk_rq_init(NULL
, &req
);
2352 scmd
->request
= &req
;
2354 scmd
->cmnd
= req
.cmd
;
2356 scmd
->scsi_done
= scsi_reset_provider_done_command
;
2357 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
2361 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
2363 spin_lock_irqsave(shost
->host_lock
, flags
);
2364 shost
->tmf_in_progress
= 1;
2365 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2367 switch (val
& ~SG_SCSI_RESET_NO_ESCALATE
) {
2368 case SG_SCSI_RESET_NOTHING
:
2371 case SG_SCSI_RESET_DEVICE
:
2372 rtn
= scsi_try_bus_device_reset(scmd
);
2373 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2376 case SG_SCSI_RESET_TARGET
:
2377 rtn
= scsi_try_target_reset(scmd
);
2378 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2381 case SG_SCSI_RESET_BUS
:
2382 rtn
= scsi_try_bus_reset(scmd
);
2383 if (rtn
== SUCCESS
|| (val
& SG_SCSI_RESET_NO_ESCALATE
))
2386 case SG_SCSI_RESET_HOST
:
2387 rtn
= scsi_try_host_reset(scmd
);
2396 error
= (rtn
== SUCCESS
) ? 0 : -EIO
;
2398 spin_lock_irqsave(shost
->host_lock
, flags
);
2399 shost
->tmf_in_progress
= 0;
2400 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2403 * be sure to wake up anyone who was sleeping or had their queue
2404 * suspended while we performed the TMF.
2406 SCSI_LOG_ERROR_RECOVERY(3,
2407 shost_printk(KERN_INFO
, shost
,
2408 "waking up host to restart after TMF\n"));
2410 wake_up(&shost
->host_wait
);
2411 scsi_run_host_queues(shost
);
2413 scsi_put_command(scmd
);
2415 out_put_autopm_host
:
2416 scsi_autopm_put_host(shost
);
2419 EXPORT_SYMBOL(scsi_ioctl_reset
);
2421 bool scsi_command_normalize_sense(const struct scsi_cmnd
*cmd
,
2422 struct scsi_sense_hdr
*sshdr
)
2424 return scsi_normalize_sense(cmd
->sense_buffer
,
2425 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2427 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2430 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2431 * @sense_buffer: byte array of sense data
2432 * @sb_len: number of valid bytes in sense_buffer
2433 * @info_out: pointer to 64 integer where 8 or 4 byte information
2434 * field will be placed if found.
2437 * 1 if information field found, 0 if not found.
2439 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2448 switch (sense_buffer
[0] & 0x7f) {
2451 if (sense_buffer
[0] & 0x80) {
2452 *info_out
= (sense_buffer
[3] << 24) +
2453 (sense_buffer
[4] << 16) +
2454 (sense_buffer
[5] << 8) + sense_buffer
[6];
2460 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2462 if (ucp
&& (0xa == ucp
[1])) {
2464 for (j
= 0; j
< 8; ++j
) {
2477 EXPORT_SYMBOL(scsi_get_sense_info_fld
);