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_transport.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_ioctl.h>
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
42 #include "scsi_transport_api.h"
44 #include <trace/events/scsi.h>
46 static void scsi_eh_done(struct scsi_cmnd
*scmd
);
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
52 #define BUS_RESET_SETTLE_TIME (10)
53 #define HOST_RESET_SETTLE_TIME (10)
55 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
);
56 static int scsi_try_to_abort_cmd(struct scsi_host_template
*,
59 /* called with shost->host_lock held */
60 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
62 if (shost
->host_busy
== shost
->host_failed
) {
63 trace_scsi_eh_wakeup(shost
);
64 wake_up_process(shost
->ehandler
);
65 SCSI_LOG_ERROR_RECOVERY(5,
66 printk("Waking error handler thread\n"));
71 * scsi_schedule_eh - schedule EH for SCSI host
72 * @shost: SCSI host to invoke error handling on.
74 * Schedule SCSI EH without scmd.
76 void scsi_schedule_eh(struct Scsi_Host
*shost
)
80 spin_lock_irqsave(shost
->host_lock
, flags
);
82 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
83 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
84 shost
->host_eh_scheduled
++;
85 scsi_eh_wakeup(shost
);
88 spin_unlock_irqrestore(shost
->host_lock
, flags
);
90 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
92 static int scsi_host_eh_past_deadline(struct Scsi_Host
*shost
)
94 if (!shost
->last_reset
|| shost
->eh_deadline
== -1)
98 * 32bit accesses are guaranteed to be atomic
99 * (on all supported architectures), so instead
100 * of using a spinlock we can as well double check
101 * if eh_deadline has been set to 'off' during the
104 if (time_before(jiffies
, shost
->last_reset
+ shost
->eh_deadline
) &&
105 shost
->eh_deadline
> -1)
112 * scmd_eh_abort_handler - Handle command aborts
113 * @work: command to be aborted.
116 scmd_eh_abort_handler(struct work_struct
*work
)
118 struct scsi_cmnd
*scmd
=
119 container_of(work
, struct scsi_cmnd
, abort_work
.work
);
120 struct scsi_device
*sdev
= scmd
->device
;
123 if (scsi_host_eh_past_deadline(sdev
->host
)) {
124 SCSI_LOG_ERROR_RECOVERY(3,
125 scmd_printk(KERN_INFO
, scmd
,
126 "scmd %p eh timeout, not aborting\n",
129 SCSI_LOG_ERROR_RECOVERY(3,
130 scmd_printk(KERN_INFO
, scmd
,
131 "aborting command %p\n", scmd
));
132 rtn
= scsi_try_to_abort_cmd(sdev
->host
->hostt
, scmd
);
133 if (rtn
== SUCCESS
) {
134 scmd
->result
|= DID_TIME_OUT
<< 16;
135 if (scsi_host_eh_past_deadline(sdev
->host
)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO
, scmd
,
138 "scmd %p eh timeout, "
139 "not retrying aborted "
141 } else if (!scsi_noretry_cmd(scmd
) &&
142 (++scmd
->retries
<= scmd
->allowed
)) {
143 SCSI_LOG_ERROR_RECOVERY(3,
144 scmd_printk(KERN_WARNING
, scmd
,
146 "aborted command\n", scmd
));
147 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_WARNING
, scmd
,
153 "aborted command\n", scmd
));
154 scsi_finish_command(scmd
);
158 SCSI_LOG_ERROR_RECOVERY(3,
159 scmd_printk(KERN_INFO
, scmd
,
160 "scmd %p abort failed, rtn %d\n",
165 if (!scsi_eh_scmd_add(scmd
, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING
, scmd
,
169 "aborted command\n", scmd
));
170 scmd
->result
|= DID_TIME_OUT
<< 16;
171 scsi_finish_command(scmd
);
176 * scsi_abort_command - schedule a command abort
177 * @scmd: scmd to abort.
179 * We only need to abort commands after a command timeout
182 scsi_abort_command(struct scsi_cmnd
*scmd
)
184 struct scsi_device
*sdev
= scmd
->device
;
185 struct Scsi_Host
*shost
= sdev
->host
;
188 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
190 * Retry after abort failed, escalate to next level.
192 scmd
->eh_eflags
&= ~SCSI_EH_ABORT_SCHEDULED
;
193 SCSI_LOG_ERROR_RECOVERY(3,
194 scmd_printk(KERN_INFO
, scmd
,
195 "scmd %p previous abort failed\n", scmd
));
196 cancel_delayed_work(&scmd
->abort_work
);
201 * Do not try a command abort if
202 * SCSI EH has already started.
204 spin_lock_irqsave(shost
->host_lock
, flags
);
205 if (scsi_host_in_recovery(shost
)) {
206 spin_unlock_irqrestore(shost
->host_lock
, flags
);
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO
, scmd
,
209 "scmd %p not aborting, host in recovery\n",
214 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
215 shost
->last_reset
= jiffies
;
216 spin_unlock_irqrestore(shost
->host_lock
, flags
);
218 scmd
->eh_eflags
|= SCSI_EH_ABORT_SCHEDULED
;
219 SCSI_LOG_ERROR_RECOVERY(3,
220 scmd_printk(KERN_INFO
, scmd
,
221 "scmd %p abort scheduled\n", scmd
));
222 queue_delayed_work(shost
->tmf_work_q
, &scmd
->abort_work
, HZ
/ 100);
227 * scsi_eh_scmd_add - add scsi cmd to error handling.
228 * @scmd: scmd to run eh on.
229 * @eh_flag: optional SCSI_EH flag.
234 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
236 struct Scsi_Host
*shost
= scmd
->device
->host
;
240 if (!shost
->ehandler
)
243 spin_lock_irqsave(shost
->host_lock
, flags
);
244 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
245 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
248 if (shost
->eh_deadline
!= -1 && !shost
->last_reset
)
249 shost
->last_reset
= jiffies
;
252 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
)
253 eh_flag
&= ~SCSI_EH_CANCEL_CMD
;
254 scmd
->eh_eflags
|= eh_flag
;
255 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
256 shost
->host_failed
++;
257 scsi_eh_wakeup(shost
);
259 spin_unlock_irqrestore(shost
->host_lock
, flags
);
264 * scsi_times_out - Timeout function for normal scsi commands.
265 * @req: request that is timing out.
268 * We do not need to lock this. There is the potential for a race
269 * only in that the normal completion handling might run, but if the
270 * normal completion function determines that the timer has already
271 * fired, then it mustn't do anything.
273 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
275 struct scsi_cmnd
*scmd
= req
->special
;
276 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
277 struct Scsi_Host
*host
= scmd
->device
->host
;
279 trace_scsi_dispatch_cmd_timeout(scmd
);
280 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
282 if (host
->eh_deadline
!= -1 && !host
->last_reset
)
283 host
->last_reset
= jiffies
;
285 if (host
->transportt
->eh_timed_out
)
286 rtn
= host
->transportt
->eh_timed_out(scmd
);
287 else if (host
->hostt
->eh_timed_out
)
288 rtn
= host
->hostt
->eh_timed_out(scmd
);
290 if (rtn
== BLK_EH_NOT_HANDLED
&& !host
->hostt
->no_async_abort
)
291 if (scsi_abort_command(scmd
) == SUCCESS
)
292 return BLK_EH_NOT_HANDLED
;
294 scmd
->result
|= DID_TIME_OUT
<< 16;
296 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
297 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
)))
298 rtn
= BLK_EH_HANDLED
;
304 * scsi_block_when_processing_errors - Prevent cmds from being queued.
305 * @sdev: Device on which we are performing recovery.
308 * We block until the host is out of error recovery, and then check to
309 * see whether the host or the device is offline.
312 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
314 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
318 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
320 online
= scsi_device_online(sdev
);
322 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
327 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
329 #ifdef CONFIG_SCSI_LOGGING
331 * scsi_eh_prt_fail_stats - Log info on failures.
332 * @shost: scsi host being recovered.
333 * @work_q: Queue of scsi cmds to process.
335 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
336 struct list_head
*work_q
)
338 struct scsi_cmnd
*scmd
;
339 struct scsi_device
*sdev
;
340 int total_failures
= 0;
343 int devices_failed
= 0;
345 shost_for_each_device(sdev
, shost
) {
346 list_for_each_entry(scmd
, work_q
, eh_entry
) {
347 if (scmd
->device
== sdev
) {
349 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
356 if (cmd_cancel
|| cmd_failed
) {
357 SCSI_LOG_ERROR_RECOVERY(3,
358 sdev_printk(KERN_INFO
, sdev
,
359 "%s: cmds failed: %d, cancel: %d\n",
360 __func__
, cmd_failed
,
368 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
369 " devices require eh work\n",
370 total_failures
, devices_failed
));
375 * scsi_report_lun_change - Set flag on all *other* devices on the same target
376 * to indicate that a UNIT ATTENTION is expected.
377 * @sdev: Device reporting the UNIT ATTENTION
379 static void scsi_report_lun_change(struct scsi_device
*sdev
)
381 sdev
->sdev_target
->expecting_lun_change
= 1;
385 * scsi_report_sense - Examine scsi sense information and log messages for
386 * certain conditions, also issue uevents for some of them.
387 * @sdev: Device reporting the sense code
388 * @sshdr: sshdr to be examined
390 static void scsi_report_sense(struct scsi_device
*sdev
,
391 struct scsi_sense_hdr
*sshdr
)
393 enum scsi_device_event evt_type
= SDEV_EVT_MAXBITS
; /* i.e. none */
395 if (sshdr
->sense_key
== UNIT_ATTENTION
) {
396 if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x03) {
397 evt_type
= SDEV_EVT_INQUIRY_CHANGE_REPORTED
;
398 sdev_printk(KERN_WARNING
, sdev
,
399 "Inquiry data has changed");
400 } else if (sshdr
->asc
== 0x3f && sshdr
->ascq
== 0x0e) {
401 evt_type
= SDEV_EVT_LUN_CHANGE_REPORTED
;
402 scsi_report_lun_change(sdev
);
403 sdev_printk(KERN_WARNING
, sdev
,
404 "Warning! Received an indication that the "
405 "LUN assignments on this target have "
406 "changed. The Linux SCSI layer does not "
407 "automatically remap LUN assignments.\n");
408 } else if (sshdr
->asc
== 0x3f)
409 sdev_printk(KERN_WARNING
, sdev
,
410 "Warning! Received an indication that the "
411 "operating parameters on this target have "
412 "changed. The Linux SCSI layer does not "
413 "automatically adjust these parameters.\n");
415 if (sshdr
->asc
== 0x38 && sshdr
->ascq
== 0x07) {
416 evt_type
= SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED
;
417 sdev_printk(KERN_WARNING
, sdev
,
418 "Warning! Received an indication that the "
419 "LUN reached a thin provisioning soft "
423 if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x01) {
424 evt_type
= SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED
;
425 sdev_printk(KERN_WARNING
, sdev
,
426 "Mode parameters changed");
427 } else if (sshdr
->asc
== 0x2a && sshdr
->ascq
== 0x09) {
428 evt_type
= SDEV_EVT_CAPACITY_CHANGE_REPORTED
;
429 sdev_printk(KERN_WARNING
, sdev
,
430 "Capacity data has changed");
431 } else if (sshdr
->asc
== 0x2a)
432 sdev_printk(KERN_WARNING
, sdev
,
433 "Parameters changed");
436 if (evt_type
!= SDEV_EVT_MAXBITS
) {
437 set_bit(evt_type
, sdev
->pending_events
);
438 schedule_work(&sdev
->event_work
);
443 * scsi_check_sense - Examine scsi cmd sense
444 * @scmd: Cmd to have sense checked.
447 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
450 * When a deferred error is detected the current command has
451 * not been executed and needs retrying.
453 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
455 struct scsi_device
*sdev
= scmd
->device
;
456 struct scsi_sense_hdr sshdr
;
458 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
459 return FAILED
; /* no valid sense data */
461 if (scmd
->cmnd
[0] == TEST_UNIT_READY
&& scmd
->scsi_done
!= scsi_eh_done
)
463 * nasty: for mid-layer issued TURs, we need to return the
464 * actual sense data without any recovery attempt. For eh
465 * issued ones, we need to try to recover and interpret
469 scsi_report_sense(sdev
, &sshdr
);
471 if (scsi_sense_is_deferred(&sshdr
))
474 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
475 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
478 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
479 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
481 /* handler does not care. Drop down to default handling */
485 * Previous logic looked for FILEMARK, EOM or ILI which are
486 * mainly associated with tapes and returned SUCCESS.
488 if (sshdr
.response_code
== 0x70) {
490 if (scmd
->sense_buffer
[2] & 0xe0)
494 * descriptor format: look for "stream commands sense data
495 * descriptor" (see SSC-3). Assume single sense data
496 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
498 if ((sshdr
.additional_length
> 3) &&
499 (scmd
->sense_buffer
[8] == 0x4) &&
500 (scmd
->sense_buffer
[11] & 0xe0))
504 switch (sshdr
.sense_key
) {
507 case RECOVERED_ERROR
:
508 return /* soft_error */ SUCCESS
;
510 case ABORTED_COMMAND
:
511 if (sshdr
.asc
== 0x10) /* DIF */
518 * if we are expecting a cc/ua because of a bus reset that we
519 * performed, treat this just as a retry. otherwise this is
520 * information that we should pass up to the upper-level driver
521 * so that we can deal with it there.
523 if (scmd
->device
->expecting_cc_ua
) {
525 * Because some device does not queue unit
526 * attentions correctly, we carefully check
527 * additional sense code and qualifier so as
528 * not to squash media change unit attention.
530 if (sshdr
.asc
!= 0x28 || sshdr
.ascq
!= 0x00) {
531 scmd
->device
->expecting_cc_ua
= 0;
536 * we might also expect a cc/ua if another LUN on the target
537 * reported a UA with an ASC/ASCQ of 3F 0E -
538 * REPORTED LUNS DATA HAS CHANGED.
540 if (scmd
->device
->sdev_target
->expecting_lun_change
&&
541 sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
544 * if the device is in the process of becoming ready, we
547 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
550 * if the device is not started, we need to wake
551 * the error handler to start the motor
553 if (scmd
->device
->allow_restart
&&
554 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
557 * Pass the UA upwards for a determination in the completion
562 /* these are not supported */
564 if (sshdr
.asc
== 0x27 && sshdr
.ascq
== 0x07) {
565 /* Thin provisioning hard threshold reached */
566 set_host_byte(scmd
, DID_ALLOC_FAILURE
);
570 case VOLUME_OVERFLOW
:
573 set_host_byte(scmd
, DID_TARGET_FAILURE
);
577 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
578 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
579 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
580 set_host_byte(scmd
, DID_MEDIUM_ERROR
);
586 if (scmd
->device
->retry_hwerror
)
587 return ADD_TO_MLQUEUE
;
589 set_host_byte(scmd
, DID_TARGET_FAILURE
);
591 case ILLEGAL_REQUEST
:
592 if (sshdr
.asc
== 0x20 || /* Invalid command operation code */
593 sshdr
.asc
== 0x21 || /* Logical block address out of range */
594 sshdr
.asc
== 0x24 || /* Invalid field in cdb */
595 sshdr
.asc
== 0x26) { /* Parameter value invalid */
596 set_host_byte(scmd
, DID_TARGET_FAILURE
);
605 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
607 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
608 struct scsi_device
*tmp_sdev
;
610 if (!sht
->change_queue_depth
||
611 sdev
->queue_depth
>= sdev
->max_queue_depth
)
614 if (time_before(jiffies
,
615 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
618 if (time_before(jiffies
,
619 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
623 * Walk all devices of a target and do
626 shost_for_each_device(tmp_sdev
, sdev
->host
) {
627 if (tmp_sdev
->channel
!= sdev
->channel
||
628 tmp_sdev
->id
!= sdev
->id
||
629 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
632 * call back into LLD to increase queue_depth by one
633 * with ramp up reason code.
635 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
636 SCSI_QDEPTH_RAMP_UP
);
637 sdev
->last_queue_ramp_up
= jiffies
;
641 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
643 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
644 struct scsi_device
*tmp_sdev
;
646 if (!sht
->change_queue_depth
)
649 shost_for_each_device(tmp_sdev
, sdev
->host
) {
650 if (tmp_sdev
->channel
!= sdev
->channel
||
651 tmp_sdev
->id
!= sdev
->id
)
654 * We do not know the number of commands that were at
655 * the device when we got the queue full so we start
656 * from the highest possible value and work our way down.
658 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
664 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
665 * @scmd: SCSI cmd to examine.
668 * This is *only* called when we are examining the status of commands
669 * queued during error recovery. the main difference here is that we
670 * don't allow for the possibility of retries here, and we are a lot
671 * more restrictive about what we consider acceptable.
673 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
676 * first check the host byte, to see if there is anything in there
677 * that would indicate what we need to do.
679 if (host_byte(scmd
->result
) == DID_RESET
) {
681 * rats. we are already in the error handler, so we now
682 * get to try and figure out what to do next. if the sense
683 * is valid, we have a pretty good idea of what to do.
684 * if not, we mark it as FAILED.
686 return scsi_check_sense(scmd
);
688 if (host_byte(scmd
->result
) != DID_OK
)
692 * next, check the message byte.
694 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
698 * now, check the status byte to see if this indicates
701 switch (status_byte(scmd
->result
)) {
703 scsi_handle_queue_ramp_up(scmd
->device
);
704 case COMMAND_TERMINATED
:
706 case CHECK_CONDITION
:
707 return scsi_check_sense(scmd
);
709 case INTERMEDIATE_GOOD
:
710 case INTERMEDIATE_C_GOOD
:
712 * who knows? FIXME(eric)
715 case RESERVATION_CONFLICT
:
716 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
717 /* it is a success, we probed the device and
720 /* otherwise, we failed to send the command */
723 scsi_handle_queue_full(scmd
->device
);
734 * scsi_eh_done - Completion function for error handling.
735 * @scmd: Cmd that is done.
737 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
739 struct completion
*eh_action
;
741 SCSI_LOG_ERROR_RECOVERY(3,
742 printk("%s scmd: %p result: %x\n",
743 __func__
, scmd
, scmd
->result
));
745 eh_action
= scmd
->device
->host
->eh_action
;
751 * scsi_try_host_reset - ask host adapter to reset itself
752 * @scmd: SCSI cmd to send host reset.
754 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
758 struct Scsi_Host
*host
= scmd
->device
->host
;
759 struct scsi_host_template
*hostt
= host
->hostt
;
761 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
764 if (!hostt
->eh_host_reset_handler
)
767 rtn
= hostt
->eh_host_reset_handler(scmd
);
769 if (rtn
== SUCCESS
) {
770 if (!hostt
->skip_settle_delay
)
771 ssleep(HOST_RESET_SETTLE_TIME
);
772 spin_lock_irqsave(host
->host_lock
, flags
);
773 scsi_report_bus_reset(host
, scmd_channel(scmd
));
774 spin_unlock_irqrestore(host
->host_lock
, flags
);
781 * scsi_try_bus_reset - ask host to perform a bus reset
782 * @scmd: SCSI cmd to send bus reset.
784 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
788 struct Scsi_Host
*host
= scmd
->device
->host
;
789 struct scsi_host_template
*hostt
= host
->hostt
;
791 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
794 if (!hostt
->eh_bus_reset_handler
)
797 rtn
= hostt
->eh_bus_reset_handler(scmd
);
799 if (rtn
== SUCCESS
) {
800 if (!hostt
->skip_settle_delay
)
801 ssleep(BUS_RESET_SETTLE_TIME
);
802 spin_lock_irqsave(host
->host_lock
, flags
);
803 scsi_report_bus_reset(host
, scmd_channel(scmd
));
804 spin_unlock_irqrestore(host
->host_lock
, flags
);
810 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
813 sdev
->expecting_cc_ua
= 1;
817 * scsi_try_target_reset - Ask host to perform a target reset
818 * @scmd: SCSI cmd used to send a target reset
821 * There is no timeout for this operation. if this operation is
822 * unreliable for a given host, then the host itself needs to put a
823 * timer on it, and set the host back to a consistent state prior to
826 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
830 struct Scsi_Host
*host
= scmd
->device
->host
;
831 struct scsi_host_template
*hostt
= host
->hostt
;
833 if (!hostt
->eh_target_reset_handler
)
836 rtn
= hostt
->eh_target_reset_handler(scmd
);
837 if (rtn
== SUCCESS
) {
838 spin_lock_irqsave(host
->host_lock
, flags
);
839 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
840 __scsi_report_device_reset
);
841 spin_unlock_irqrestore(host
->host_lock
, flags
);
848 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
849 * @scmd: SCSI cmd used to send BDR
852 * There is no timeout for this operation. if this operation is
853 * unreliable for a given host, then the host itself needs to put a
854 * timer on it, and set the host back to a consistent state prior to
857 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
860 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
862 if (!hostt
->eh_device_reset_handler
)
865 rtn
= hostt
->eh_device_reset_handler(scmd
);
867 __scsi_report_device_reset(scmd
->device
, NULL
);
871 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
, struct scsi_cmnd
*scmd
)
873 if (!hostt
->eh_abort_handler
)
876 return hostt
->eh_abort_handler(scmd
);
879 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
881 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
882 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
883 if (scsi_try_target_reset(scmd
) != SUCCESS
)
884 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
885 scsi_try_host_reset(scmd
);
889 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
890 * @scmd: SCSI command structure to hijack
891 * @ses: structure to save restore information
892 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
893 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
894 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
896 * This function is used to save a scsi command information before re-execution
897 * as part of the error recovery process. If @sense_bytes is 0 the command
898 * sent must be one that does not transfer any data. If @sense_bytes != 0
899 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
900 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
902 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
903 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
905 struct scsi_device
*sdev
= scmd
->device
;
908 * We need saved copies of a number of fields - this is because
909 * error handling may need to overwrite these with different values
910 * to run different commands, and once error handling is complete,
911 * we will need to restore these values prior to running the actual
914 ses
->cmd_len
= scmd
->cmd_len
;
915 ses
->cmnd
= scmd
->cmnd
;
916 ses
->data_direction
= scmd
->sc_data_direction
;
917 ses
->sdb
= scmd
->sdb
;
918 ses
->next_rq
= scmd
->request
->next_rq
;
919 ses
->result
= scmd
->result
;
920 ses
->underflow
= scmd
->underflow
;
921 ses
->prot_op
= scmd
->prot_op
;
923 scmd
->prot_op
= SCSI_PROT_NORMAL
;
925 scmd
->cmnd
= ses
->eh_cmnd
;
926 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
927 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
928 scmd
->request
->next_rq
= NULL
;
932 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
934 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
936 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
937 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
938 scmd
->sdb
.table
.nents
= 1;
939 scmd
->cmnd
[0] = REQUEST_SENSE
;
940 scmd
->cmnd
[4] = scmd
->sdb
.length
;
941 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
943 scmd
->sc_data_direction
= DMA_NONE
;
945 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
946 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
947 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
953 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
954 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
955 (sdev
->lun
<< 5 & 0xe0);
958 * Zero the sense buffer. The scsi spec mandates that any
959 * untransferred sense data should be interpreted as being zero.
961 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
963 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
966 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
967 * @scmd: SCSI command structure to restore
968 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
970 * Undo any damage done by above scsi_eh_prep_cmnd().
972 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
975 * Restore original data
977 scmd
->cmd_len
= ses
->cmd_len
;
978 scmd
->cmnd
= ses
->cmnd
;
979 scmd
->sc_data_direction
= ses
->data_direction
;
980 scmd
->sdb
= ses
->sdb
;
981 scmd
->request
->next_rq
= ses
->next_rq
;
982 scmd
->result
= ses
->result
;
983 scmd
->underflow
= ses
->underflow
;
984 scmd
->prot_op
= ses
->prot_op
;
986 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
989 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
990 * @scmd: SCSI command structure to hijack
992 * @cmnd_size: size in bytes of @cmnd
993 * @timeout: timeout for this request
994 * @sense_bytes: size of sense data to copy or 0
996 * This function is used to send a scsi command down to a target device
997 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1000 * SUCCESS or FAILED or NEEDS_RETRY
1002 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
1003 int cmnd_size
, int timeout
, unsigned sense_bytes
)
1005 struct scsi_device
*sdev
= scmd
->device
;
1006 struct Scsi_Host
*shost
= sdev
->host
;
1007 DECLARE_COMPLETION_ONSTACK(done
);
1008 unsigned long timeleft
= timeout
;
1009 struct scsi_eh_save ses
;
1010 const unsigned long stall_for
= msecs_to_jiffies(100);
1014 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
1015 shost
->eh_action
= &done
;
1017 scsi_log_send(scmd
);
1018 scmd
->scsi_done
= scsi_eh_done
;
1019 rtn
= shost
->hostt
->queuecommand(shost
, scmd
);
1021 if (timeleft
> stall_for
) {
1022 scsi_eh_restore_cmnd(scmd
, &ses
);
1023 timeleft
-= stall_for
;
1024 msleep(jiffies_to_msecs(stall_for
));
1027 /* signal not to enter either branch of the if () below */
1031 timeleft
= wait_for_completion_timeout(&done
, timeout
);
1034 shost
->eh_action
= NULL
;
1036 scsi_log_completion(scmd
, rtn
);
1038 SCSI_LOG_ERROR_RECOVERY(3,
1039 printk("%s: scmd: %p, timeleft: %ld\n",
1040 __func__
, scmd
, timeleft
));
1043 * If there is time left scsi_eh_done got called, and we will examine
1044 * the actual status codes to see whether the command actually did
1045 * complete normally, else if we have a zero return and no time left,
1046 * the command must still be pending, so abort it and return FAILED.
1047 * If we never actually managed to issue the command, because
1048 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1049 * value above (so don't execute either branch of the if)
1052 rtn
= scsi_eh_completed_normally(scmd
);
1053 SCSI_LOG_ERROR_RECOVERY(3,
1054 printk("%s: scsi_eh_completed_normally %x\n",
1062 case ADD_TO_MLQUEUE
:
1070 scsi_abort_eh_cmnd(scmd
);
1074 scsi_eh_restore_cmnd(scmd
, &ses
);
1080 * scsi_request_sense - Request sense data from a particular target.
1081 * @scmd: SCSI cmd for request sense.
1084 * Some hosts automatically obtain this information, others require
1085 * that we obtain it on our own. This function will *not* return until
1086 * the command either times out, or it completes.
1088 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
1090 return scsi_send_eh_cmnd(scmd
, NULL
, 0, scmd
->device
->eh_timeout
, ~0);
1093 static int scsi_eh_action(struct scsi_cmnd
*scmd
, int rtn
)
1095 if (scmd
->request
->cmd_type
!= REQ_TYPE_BLOCK_PC
) {
1096 struct scsi_driver
*sdrv
= scsi_cmd_to_driver(scmd
);
1097 if (sdrv
->eh_action
)
1098 rtn
= sdrv
->eh_action(scmd
, rtn
);
1104 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1105 * @scmd: Original SCSI cmd that eh has finished.
1106 * @done_q: Queue for processed commands.
1109 * We don't want to use the normal command completion while we are are
1110 * still handling errors - it may cause other commands to be queued,
1111 * and that would disturb what we are doing. Thus we really want to
1112 * keep a list of pending commands for final completion, and once we
1113 * are ready to leave error handling we handle completion for real.
1115 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
1117 scmd
->device
->host
->host_failed
--;
1118 scmd
->eh_eflags
= 0;
1119 list_move_tail(&scmd
->eh_entry
, done_q
);
1121 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
1124 * scsi_eh_get_sense - Get device sense data.
1125 * @work_q: Queue of commands to process.
1126 * @done_q: Queue of processed commands.
1129 * See if we need to request sense information. if so, then get it
1130 * now, so we have a better idea of what to do.
1133 * This has the unfortunate side effect that if a shost adapter does
1134 * not automatically request sense information, we end up shutting
1135 * it down before we request it.
1137 * All drivers should request sense information internally these days,
1138 * so for now all I have to say is tough noogies if you end up in here.
1140 * XXX: Long term this code should go away, but that needs an audit of
1143 int scsi_eh_get_sense(struct list_head
*work_q
,
1144 struct list_head
*done_q
)
1146 struct scsi_cmnd
*scmd
, *next
;
1147 struct Scsi_Host
*shost
;
1150 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1151 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
1152 SCSI_SENSE_VALID(scmd
))
1155 shost
= scmd
->device
->host
;
1156 if (scsi_host_eh_past_deadline(shost
)) {
1157 SCSI_LOG_ERROR_RECOVERY(3,
1158 shost_printk(KERN_INFO
, shost
,
1159 "skip %s, past eh deadline\n",
1163 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1165 * don't request sense if there's no check condition
1166 * status because the error we're processing isn't one
1167 * that has a sense code (and some devices get
1168 * confused by sense requests out of the blue)
1172 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
1173 "%s: requesting sense\n",
1175 rtn
= scsi_request_sense(scmd
);
1179 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
1180 " result %x\n", scmd
,
1182 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
1184 rtn
= scsi_decide_disposition(scmd
);
1187 * if the result was normal, then just pass it along to the
1191 /* we don't want this command reissued, just
1192 * finished with the sense data, so set
1193 * retries to the max allowed to ensure it
1194 * won't get reissued */
1195 scmd
->retries
= scmd
->allowed
;
1196 else if (rtn
!= NEEDS_RETRY
)
1199 scsi_eh_finish_cmd(scmd
, done_q
);
1202 return list_empty(work_q
);
1204 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
1207 * scsi_eh_tur - Send TUR to device.
1208 * @scmd: &scsi_cmnd to send TUR
1211 * 0 - Device is ready. 1 - Device NOT ready.
1213 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
1215 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
1216 int retry_cnt
= 1, rtn
;
1219 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6,
1220 scmd
->device
->eh_timeout
, 0);
1222 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
1223 __func__
, scmd
, rtn
));
1238 * scsi_eh_test_devices - check if devices are responding from error recovery.
1239 * @cmd_list: scsi commands in error recovery.
1240 * @work_q: queue for commands which still need more error recovery
1241 * @done_q: queue for commands which are finished
1242 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1245 * Tests if devices are in a working state. Commands to devices now in
1246 * a working state are sent to the done_q while commands to devices which
1247 * are still failing to respond are returned to the work_q for more
1250 static int scsi_eh_test_devices(struct list_head
*cmd_list
,
1251 struct list_head
*work_q
,
1252 struct list_head
*done_q
, int try_stu
)
1254 struct scsi_cmnd
*scmd
, *next
;
1255 struct scsi_device
*sdev
;
1258 while (!list_empty(cmd_list
)) {
1259 scmd
= list_entry(cmd_list
->next
, struct scsi_cmnd
, eh_entry
);
1260 sdev
= scmd
->device
;
1263 if (scsi_host_eh_past_deadline(sdev
->host
)) {
1264 /* Push items back onto work_q */
1265 list_splice_init(cmd_list
, work_q
);
1266 SCSI_LOG_ERROR_RECOVERY(3,
1267 shost_printk(KERN_INFO
, sdev
->host
,
1268 "skip %s, past eh deadline",
1274 finish_cmds
= !scsi_device_online(scmd
->device
) ||
1275 (try_stu
&& !scsi_eh_try_stu(scmd
) &&
1276 !scsi_eh_tur(scmd
)) ||
1279 list_for_each_entry_safe(scmd
, next
, cmd_list
, eh_entry
)
1280 if (scmd
->device
== sdev
) {
1283 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
))
1284 scsi_eh_finish_cmd(scmd
, done_q
);
1286 list_move_tail(&scmd
->eh_entry
, work_q
);
1289 return list_empty(work_q
);
1294 * scsi_eh_abort_cmds - abort pending commands.
1295 * @work_q: &list_head for pending commands.
1296 * @done_q: &list_head for processed commands.
1299 * Try and see whether or not it makes sense to try and abort the
1300 * running command. This only works out to be the case if we have one
1301 * command that has timed out. If the command simply failed, it makes
1302 * no sense to try and abort the command, since as far as the shost
1303 * adapter is concerned, it isn't running.
1305 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
1306 struct list_head
*done_q
)
1308 struct scsi_cmnd
*scmd
, *next
;
1309 LIST_HEAD(check_list
);
1311 struct Scsi_Host
*shost
;
1313 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1314 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
1316 shost
= scmd
->device
->host
;
1317 if (scsi_host_eh_past_deadline(shost
)) {
1318 list_splice_init(&check_list
, work_q
);
1319 SCSI_LOG_ERROR_RECOVERY(3,
1320 shost_printk(KERN_INFO
, shost
,
1321 "skip %s, past eh deadline\n",
1323 return list_empty(work_q
);
1325 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1326 "0x%p\n", current
->comm
,
1328 rtn
= scsi_try_to_abort_cmd(shost
->hostt
, scmd
);
1329 if (rtn
== FAILED
) {
1330 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1335 list_splice_init(&check_list
, work_q
);
1336 return list_empty(work_q
);
1338 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
1339 if (rtn
== FAST_IO_FAIL
)
1340 scsi_eh_finish_cmd(scmd
, done_q
);
1342 list_move_tail(&scmd
->eh_entry
, &check_list
);
1345 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1349 * scsi_eh_try_stu - Send START_UNIT to device.
1350 * @scmd: &scsi_cmnd to send START_UNIT
1353 * 0 - Device is ready. 1 - Device NOT ready.
1355 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1357 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1359 if (scmd
->device
->allow_restart
) {
1360 int i
, rtn
= NEEDS_RETRY
;
1362 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1363 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1373 * scsi_eh_stu - send START_UNIT if needed
1374 * @shost: &scsi host being recovered.
1375 * @work_q: &list_head for pending commands.
1376 * @done_q: &list_head for processed commands.
1379 * If commands are failing due to not ready, initializing command required,
1380 * try revalidating the device, which will end up sending a start unit.
1382 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1383 struct list_head
*work_q
,
1384 struct list_head
*done_q
)
1386 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1387 struct scsi_device
*sdev
;
1389 shost_for_each_device(sdev
, shost
) {
1390 if (scsi_host_eh_past_deadline(shost
)) {
1391 SCSI_LOG_ERROR_RECOVERY(3,
1392 shost_printk(KERN_INFO
, shost
,
1393 "skip %s, past eh deadline\n",
1398 list_for_each_entry(scmd
, work_q
, eh_entry
)
1399 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1400 scsi_check_sense(scmd
) == FAILED
) {
1408 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1409 " 0x%p\n", current
->comm
, sdev
));
1411 if (!scsi_eh_try_stu(stu_scmd
)) {
1412 if (!scsi_device_online(sdev
) ||
1413 !scsi_eh_tur(stu_scmd
)) {
1414 list_for_each_entry_safe(scmd
, next
,
1416 if (scmd
->device
== sdev
&&
1417 scsi_eh_action(scmd
, SUCCESS
) == SUCCESS
)
1418 scsi_eh_finish_cmd(scmd
, done_q
);
1422 SCSI_LOG_ERROR_RECOVERY(3,
1423 printk("%s: START_UNIT failed to sdev:"
1424 " 0x%p\n", current
->comm
, sdev
));
1428 return list_empty(work_q
);
1433 * scsi_eh_bus_device_reset - send bdr if needed
1434 * @shost: scsi host being recovered.
1435 * @work_q: &list_head for pending commands.
1436 * @done_q: &list_head for processed commands.
1439 * Try a bus device reset. Still, look to see whether we have multiple
1440 * devices that are jammed or not - if we have multiple devices, it
1441 * makes no sense to try bus_device_reset - we really would need to try
1442 * a bus_reset instead.
1444 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1445 struct list_head
*work_q
,
1446 struct list_head
*done_q
)
1448 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1449 struct scsi_device
*sdev
;
1452 shost_for_each_device(sdev
, shost
) {
1453 if (scsi_host_eh_past_deadline(shost
)) {
1454 SCSI_LOG_ERROR_RECOVERY(3,
1455 shost_printk(KERN_INFO
, shost
,
1456 "skip %s, past eh deadline\n",
1461 list_for_each_entry(scmd
, work_q
, eh_entry
)
1462 if (scmd
->device
== sdev
) {
1470 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1471 " 0x%p\n", current
->comm
,
1473 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1474 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1475 if (!scsi_device_online(sdev
) ||
1476 rtn
== FAST_IO_FAIL
||
1477 !scsi_eh_tur(bdr_scmd
)) {
1478 list_for_each_entry_safe(scmd
, next
,
1480 if (scmd
->device
== sdev
&&
1481 scsi_eh_action(scmd
, rtn
) != FAILED
)
1482 scsi_eh_finish_cmd(scmd
,
1487 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1495 return list_empty(work_q
);
1499 * scsi_eh_target_reset - send target reset if needed
1500 * @shost: scsi host being recovered.
1501 * @work_q: &list_head for pending commands.
1502 * @done_q: &list_head for processed commands.
1505 * Try a target reset.
1507 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1508 struct list_head
*work_q
,
1509 struct list_head
*done_q
)
1511 LIST_HEAD(tmp_list
);
1512 LIST_HEAD(check_list
);
1514 list_splice_init(work_q
, &tmp_list
);
1516 while (!list_empty(&tmp_list
)) {
1517 struct scsi_cmnd
*next
, *scmd
;
1521 if (scsi_host_eh_past_deadline(shost
)) {
1522 /* push back on work queue for further processing */
1523 list_splice_init(&check_list
, work_q
);
1524 list_splice_init(&tmp_list
, work_q
);
1525 SCSI_LOG_ERROR_RECOVERY(3,
1526 shost_printk(KERN_INFO
, shost
,
1527 "skip %s, past eh deadline\n",
1529 return list_empty(work_q
);
1532 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1535 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1537 current
->comm
, id
));
1538 rtn
= scsi_try_target_reset(scmd
);
1539 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1540 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1543 current
->comm
, id
));
1544 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1545 if (scmd_id(scmd
) != id
)
1549 list_move_tail(&scmd
->eh_entry
, &check_list
);
1550 else if (rtn
== FAST_IO_FAIL
)
1551 scsi_eh_finish_cmd(scmd
, done_q
);
1553 /* push back on work queue for further processing */
1554 list_move(&scmd
->eh_entry
, work_q
);
1558 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1562 * scsi_eh_bus_reset - send a bus reset
1563 * @shost: &scsi host being recovered.
1564 * @work_q: &list_head for pending commands.
1565 * @done_q: &list_head for processed commands.
1567 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1568 struct list_head
*work_q
,
1569 struct list_head
*done_q
)
1571 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1572 LIST_HEAD(check_list
);
1573 unsigned int channel
;
1577 * we really want to loop over the various channels, and do this on
1578 * a channel by channel basis. we should also check to see if any
1579 * of the failed commands are on soft_reset devices, and if so, skip
1583 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1584 if (scsi_host_eh_past_deadline(shost
)) {
1585 list_splice_init(&check_list
, work_q
);
1586 SCSI_LOG_ERROR_RECOVERY(3,
1587 shost_printk(KERN_INFO
, shost
,
1588 "skip %s, past eh deadline\n",
1590 return list_empty(work_q
);
1594 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1595 if (channel
== scmd_channel(scmd
)) {
1599 * FIXME add back in some support for
1600 * soft_reset devices.
1607 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1608 " %d\n", current
->comm
,
1610 rtn
= scsi_try_bus_reset(chan_scmd
);
1611 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1612 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1613 if (channel
== scmd_channel(scmd
)) {
1614 if (rtn
== FAST_IO_FAIL
)
1615 scsi_eh_finish_cmd(scmd
,
1618 list_move_tail(&scmd
->eh_entry
,
1623 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1624 " failed chan: %d\n",
1629 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 0);
1633 * scsi_eh_host_reset - send a host reset
1634 * @work_q: list_head for processed commands.
1635 * @done_q: list_head for processed commands.
1637 static int scsi_eh_host_reset(struct list_head
*work_q
,
1638 struct list_head
*done_q
)
1640 struct scsi_cmnd
*scmd
, *next
;
1641 LIST_HEAD(check_list
);
1644 if (!list_empty(work_q
)) {
1645 scmd
= list_entry(work_q
->next
,
1646 struct scsi_cmnd
, eh_entry
);
1648 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1651 rtn
= scsi_try_host_reset(scmd
);
1652 if (rtn
== SUCCESS
) {
1653 list_splice_init(work_q
, &check_list
);
1654 } else if (rtn
== FAST_IO_FAIL
) {
1655 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1656 scsi_eh_finish_cmd(scmd
, done_q
);
1659 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1664 return scsi_eh_test_devices(&check_list
, work_q
, done_q
, 1);
1668 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1669 * @work_q: list_head for processed commands.
1670 * @done_q: list_head for processed commands.
1672 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1673 struct list_head
*done_q
)
1675 struct scsi_cmnd
*scmd
, *next
;
1677 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1678 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1679 "not ready after error recovery\n");
1680 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1681 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1683 * FIXME: Handle lost cmds.
1686 scsi_eh_finish_cmd(scmd
, done_q
);
1692 * scsi_noretry_cmd - determine if command should be failed fast
1693 * @scmd: SCSI cmd to examine.
1695 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1697 switch (host_byte(scmd
->result
)) {
1703 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1705 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1707 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1708 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1711 case DID_SOFT_ERROR
:
1712 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1715 if (status_byte(scmd
->result
) != CHECK_CONDITION
)
1720 * assume caller has checked sense and determined
1721 * the check condition was retryable.
1723 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1724 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1731 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1732 * @scmd: SCSI cmd to examine.
1735 * This is *only* called when we are examining the status after sending
1736 * out the actual data command. any commands that are queued for error
1737 * recovery (e.g. test_unit_ready) do *not* come through here.
1739 * When this routine returns failed, it means the error handler thread
1740 * is woken. In cases where the error code indicates an error that
1741 * doesn't require the error handler read (i.e. we don't need to
1742 * abort/reset), this function should return SUCCESS.
1744 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1749 * if the device is offline, then we clearly just pass the result back
1750 * up to the top level.
1752 if (!scsi_device_online(scmd
->device
)) {
1753 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1760 * first check the host byte, to see if there is anything in there
1761 * that would indicate what we need to do.
1763 switch (host_byte(scmd
->result
)) {
1764 case DID_PASSTHROUGH
:
1766 * no matter what, pass this through to the upper layer.
1767 * nuke this special code so that it looks like we are saying
1770 scmd
->result
&= 0xff00ffff;
1774 * looks good. drop through, and check the next byte.
1778 if (scmd
->eh_eflags
& SCSI_EH_ABORT_SCHEDULED
) {
1779 scmd
->result
|= DID_TIME_OUT
<< 16;
1782 case DID_NO_CONNECT
:
1783 case DID_BAD_TARGET
:
1785 * note - this means that we just report the status back
1786 * to the top level driver, not that we actually think
1787 * that it indicates SUCCESS.
1791 * when the low level driver returns did_soft_error,
1792 * it is responsible for keeping an internal retry counter
1793 * in order to avoid endless loops (db)
1795 * actually this is a bug in this function here. we should
1796 * be mindful of the maximum number of retries specified
1797 * and not get stuck in a loop.
1799 case DID_SOFT_ERROR
:
1805 return ADD_TO_MLQUEUE
;
1806 case DID_TRANSPORT_DISRUPTED
:
1808 * LLD/transport was disrupted during processing of the IO.
1809 * The transport class is now blocked/blocking,
1810 * and the transport will decide what to do with the IO
1811 * based on its timers and recovery capablilities if
1812 * there are enough retries.
1815 case DID_TRANSPORT_FAILFAST
:
1817 * The transport decided to failfast the IO (most likely
1818 * the fast io fail tmo fired), so send IO directly upwards.
1822 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1823 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1825 * execute reservation conflict processing code
1835 * when we scan the bus, we get timeout messages for
1836 * these commands if there is no device available.
1837 * other hosts report did_no_connect for the same thing.
1839 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1840 scmd
->cmnd
[0] == INQUIRY
)) {
1852 * next, check the message byte.
1854 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1858 * check the status byte to see if this indicates anything special.
1860 switch (status_byte(scmd
->result
)) {
1862 scsi_handle_queue_full(scmd
->device
);
1864 * the case of trying to send too many commands to a
1865 * tagged queueing device.
1869 * device can't talk to us at the moment. Should only
1870 * occur (SAM-3) when the task queue is empty, so will cause
1871 * the empty queue handling to trigger a stall in the
1874 return ADD_TO_MLQUEUE
;
1876 if (scmd
->cmnd
[0] == REPORT_LUNS
)
1877 scmd
->device
->sdev_target
->expecting_lun_change
= 0;
1878 scsi_handle_queue_ramp_up(scmd
->device
);
1879 case COMMAND_TERMINATED
:
1883 case CHECK_CONDITION
:
1884 rtn
= scsi_check_sense(scmd
);
1885 if (rtn
== NEEDS_RETRY
)
1887 /* if rtn == FAILED, we have no sense information;
1888 * returning FAILED will wake the error handler thread
1889 * to collect the sense and redo the decide
1892 case CONDITION_GOOD
:
1893 case INTERMEDIATE_GOOD
:
1894 case INTERMEDIATE_C_GOOD
:
1897 * who knows? FIXME(eric)
1901 case RESERVATION_CONFLICT
:
1902 sdev_printk(KERN_INFO
, scmd
->device
,
1903 "reservation conflict\n");
1904 set_host_byte(scmd
, DID_NEXUS_FAILURE
);
1905 return SUCCESS
; /* causes immediate i/o error */
1913 /* we requeue for retry because the error was retryable, and
1914 * the request was not marked fast fail. Note that above,
1915 * even if the request is marked fast fail, we still requeue
1916 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1917 if ((++scmd
->retries
) <= scmd
->allowed
1918 && !scsi_noretry_cmd(scmd
)) {
1922 * no more retries - report this one back to upper level.
1928 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1930 __blk_put_request(req
->q
, req
);
1934 * scsi_eh_lock_door - Prevent medium removal for the specified device
1935 * @sdev: SCSI device to prevent medium removal
1938 * We must be called from process context.
1941 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1942 * head of the devices request queue, and continue.
1944 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1946 struct request
*req
;
1949 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1950 * request becomes available
1952 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1954 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1958 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1961 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1963 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1964 req
->cmd_flags
|= REQ_QUIET
;
1965 req
->timeout
= 10 * HZ
;
1968 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1972 * scsi_restart_operations - restart io operations to the specified host.
1973 * @shost: Host we are restarting.
1976 * When we entered the error handler, we blocked all further i/o to
1977 * this device. we need to 'reverse' this process.
1979 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1981 struct scsi_device
*sdev
;
1982 unsigned long flags
;
1985 * If the door was locked, we need to insert a door lock request
1986 * onto the head of the SCSI request queue for the device. There
1987 * is no point trying to lock the door of an off-line device.
1989 shost_for_each_device(sdev
, shost
) {
1990 if (scsi_device_online(sdev
) && sdev
->locked
)
1991 scsi_eh_lock_door(sdev
);
1995 * next free up anything directly waiting upon the host. this
1996 * will be requests for character device operations, and also for
1997 * ioctls to queued block devices.
1999 SCSI_LOG_ERROR_RECOVERY(3,
2000 printk("scsi_eh_%d waking up host to restart\n",
2003 spin_lock_irqsave(shost
->host_lock
, flags
);
2004 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
2005 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
2006 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
2007 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2009 wake_up(&shost
->host_wait
);
2012 * finally we need to re-initiate requests that may be pending. we will
2013 * have had everything blocked while error handling is taking place, and
2014 * now that error recovery is done, we will need to ensure that these
2015 * requests are started.
2017 scsi_run_host_queues(shost
);
2020 * if eh is active and host_eh_scheduled is pending we need to re-run
2021 * recovery. we do this check after scsi_run_host_queues() to allow
2022 * everything pent up since the last eh run a chance to make forward
2023 * progress before we sync again. Either we'll immediately re-run
2024 * recovery or scsi_device_unbusy() will wake us again when these
2025 * pending commands complete.
2027 spin_lock_irqsave(shost
->host_lock
, flags
);
2028 if (shost
->host_eh_scheduled
)
2029 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
2030 WARN_ON(scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
));
2031 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2035 * scsi_eh_ready_devs - check device ready state and recover if not.
2036 * @shost: host to be recovered.
2037 * @work_q: &list_head for pending commands.
2038 * @done_q: &list_head for processed commands.
2040 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
2041 struct list_head
*work_q
,
2042 struct list_head
*done_q
)
2044 if (!scsi_eh_stu(shost
, work_q
, done_q
))
2045 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
2046 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
2047 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
2048 if (!scsi_eh_host_reset(work_q
, done_q
))
2049 scsi_eh_offline_sdevs(work_q
,
2052 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
2055 * scsi_eh_flush_done_q - finish processed commands or retry them.
2056 * @done_q: list_head of processed commands.
2058 void scsi_eh_flush_done_q(struct list_head
*done_q
)
2060 struct scsi_cmnd
*scmd
, *next
;
2062 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
2063 list_del_init(&scmd
->eh_entry
);
2064 if (scsi_device_online(scmd
->device
) &&
2065 !scsi_noretry_cmd(scmd
) &&
2066 (++scmd
->retries
<= scmd
->allowed
)) {
2067 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
2071 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
2074 * If just we got sense for the device (called
2075 * scsi_eh_get_sense), scmd->result is already
2076 * set, do not set DRIVER_TIMEOUT.
2079 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
2080 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
2082 current
->comm
, scmd
));
2083 scsi_finish_command(scmd
);
2087 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
2090 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2091 * @shost: Host to unjam.
2094 * When we come in here, we *know* that all commands on the bus have
2095 * either completed, failed or timed out. we also know that no further
2096 * commands are being sent to the host, so things are relatively quiet
2097 * and we have freedom to fiddle with things as we wish.
2099 * This is only the *default* implementation. it is possible for
2100 * individual drivers to supply their own version of this function, and
2101 * if the maintainer wishes to do this, it is strongly suggested that
2102 * this function be taken as a template and modified. this function
2103 * was designed to correctly handle problems for about 95% of the
2104 * different cases out there, and it should always provide at least a
2105 * reasonable amount of error recovery.
2107 * Any command marked 'failed' or 'timeout' must eventually have
2108 * scsi_finish_cmd() called for it. we do all of the retry stuff
2109 * here, so when we restart the host after we return it should have an
2112 static void scsi_unjam_host(struct Scsi_Host
*shost
)
2114 unsigned long flags
;
2115 LIST_HEAD(eh_work_q
);
2116 LIST_HEAD(eh_done_q
);
2118 spin_lock_irqsave(shost
->host_lock
, flags
);
2119 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
2120 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2122 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
2124 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
2125 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
2126 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
2128 spin_lock_irqsave(shost
->host_lock
, flags
);
2129 if (shost
->eh_deadline
!= -1)
2130 shost
->last_reset
= 0;
2131 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2132 scsi_eh_flush_done_q(&eh_done_q
);
2136 * scsi_error_handler - SCSI error handler thread
2137 * @data: Host for which we are running.
2140 * This is the main error handling loop. This is run as a kernel thread
2141 * for every SCSI host and handles all error handling activity.
2143 int scsi_error_handler(void *data
)
2145 struct Scsi_Host
*shost
= data
;
2148 * We use TASK_INTERRUPTIBLE so that the thread is not
2149 * counted against the load average as a running process.
2150 * We never actually get interrupted because kthread_run
2151 * disables signal delivery for the created thread.
2153 while (!kthread_should_stop()) {
2154 set_current_state(TASK_INTERRUPTIBLE
);
2155 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
2156 shost
->host_failed
!= shost
->host_busy
) {
2157 SCSI_LOG_ERROR_RECOVERY(1,
2158 printk("scsi_eh_%d: sleeping\n",
2164 __set_current_state(TASK_RUNNING
);
2165 SCSI_LOG_ERROR_RECOVERY(1,
2166 printk("scsi_eh_%d: waking up %d/%d/%d\n",
2167 shost
->host_no
, shost
->host_eh_scheduled
,
2168 shost
->host_failed
, shost
->host_busy
));
2171 * We have a host that is failing for some reason. Figure out
2172 * what we need to do to get it up and online again (if we can).
2173 * If we fail, we end up taking the thing offline.
2175 if (!shost
->eh_noresume
&& scsi_autopm_get_host(shost
) != 0) {
2176 SCSI_LOG_ERROR_RECOVERY(1,
2177 printk(KERN_ERR
"Error handler scsi_eh_%d "
2178 "unable to autoresume\n",
2183 if (shost
->transportt
->eh_strategy_handler
)
2184 shost
->transportt
->eh_strategy_handler(shost
);
2186 scsi_unjam_host(shost
);
2189 * Note - if the above fails completely, the action is to take
2190 * individual devices offline and flush the queue of any
2191 * outstanding requests that may have been pending. When we
2192 * restart, we restart any I/O to any other devices on the bus
2193 * which are still online.
2195 scsi_restart_operations(shost
);
2196 if (!shost
->eh_noresume
)
2197 scsi_autopm_put_host(shost
);
2199 __set_current_state(TASK_RUNNING
);
2201 SCSI_LOG_ERROR_RECOVERY(1,
2202 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
2203 shost
->ehandler
= NULL
;
2208 * Function: scsi_report_bus_reset()
2210 * Purpose: Utility function used by low-level drivers to report that
2211 * they have observed a bus reset on the bus being handled.
2213 * Arguments: shost - Host in question
2214 * channel - channel on which reset was observed.
2218 * Lock status: Host lock must be held.
2220 * Notes: This only needs to be called if the reset is one which
2221 * originates from an unknown location. Resets originated
2222 * by the mid-level itself don't need to call this, but there
2223 * should be no harm.
2225 * The main purpose of this is to make sure that a CHECK_CONDITION
2226 * is properly treated.
2228 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
2230 struct scsi_device
*sdev
;
2232 __shost_for_each_device(sdev
, shost
) {
2233 if (channel
== sdev_channel(sdev
))
2234 __scsi_report_device_reset(sdev
, NULL
);
2237 EXPORT_SYMBOL(scsi_report_bus_reset
);
2240 * Function: scsi_report_device_reset()
2242 * Purpose: Utility function used by low-level drivers to report that
2243 * they have observed a device reset on the device being handled.
2245 * Arguments: shost - Host in question
2246 * channel - channel on which reset was observed
2247 * target - target on which reset was observed
2251 * Lock status: Host lock must be held
2253 * Notes: This only needs to be called if the reset is one which
2254 * originates from an unknown location. Resets originated
2255 * by the mid-level itself don't need to call this, but there
2256 * should be no harm.
2258 * The main purpose of this is to make sure that a CHECK_CONDITION
2259 * is properly treated.
2261 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
2263 struct scsi_device
*sdev
;
2265 __shost_for_each_device(sdev
, shost
) {
2266 if (channel
== sdev_channel(sdev
) &&
2267 target
== sdev_id(sdev
))
2268 __scsi_report_device_reset(sdev
, NULL
);
2271 EXPORT_SYMBOL(scsi_report_device_reset
);
2274 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
2279 * Function: scsi_reset_provider
2281 * Purpose: Send requested reset to a bus or device at any phase.
2283 * Arguments: device - device to send reset to
2284 * flag - reset type (see scsi.h)
2286 * Returns: SUCCESS/FAILURE.
2288 * Notes: This is used by the SCSI Generic driver to provide
2289 * Bus/Device reset capability.
2292 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
2294 struct scsi_cmnd
*scmd
;
2295 struct Scsi_Host
*shost
= dev
->host
;
2297 unsigned long flags
;
2300 if (scsi_autopm_get_host(shost
) < 0)
2303 if (!get_device(&dev
->sdev_gendev
)) {
2305 goto out_put_autopm_host
;
2308 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
2309 blk_rq_init(NULL
, &req
);
2310 scmd
->request
= &req
;
2312 scmd
->cmnd
= req
.cmd
;
2314 scmd
->scsi_done
= scsi_reset_provider_done_command
;
2315 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
2319 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
2321 spin_lock_irqsave(shost
->host_lock
, flags
);
2322 shost
->tmf_in_progress
= 1;
2323 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2326 case SCSI_TRY_RESET_DEVICE
:
2327 rtn
= scsi_try_bus_device_reset(scmd
);
2331 case SCSI_TRY_RESET_TARGET
:
2332 rtn
= scsi_try_target_reset(scmd
);
2336 case SCSI_TRY_RESET_BUS
:
2337 rtn
= scsi_try_bus_reset(scmd
);
2341 case SCSI_TRY_RESET_HOST
:
2342 rtn
= scsi_try_host_reset(scmd
);
2348 spin_lock_irqsave(shost
->host_lock
, flags
);
2349 shost
->tmf_in_progress
= 0;
2350 spin_unlock_irqrestore(shost
->host_lock
, flags
);
2353 * be sure to wake up anyone who was sleeping or had their queue
2354 * suspended while we performed the TMF.
2356 SCSI_LOG_ERROR_RECOVERY(3,
2357 printk("%s: waking up host to restart after TMF\n",
2360 wake_up(&shost
->host_wait
);
2362 scsi_run_host_queues(shost
);
2364 scsi_next_command(scmd
);
2365 out_put_autopm_host
:
2366 scsi_autopm_put_host(shost
);
2369 EXPORT_SYMBOL(scsi_reset_provider
);
2372 * scsi_normalize_sense - normalize main elements from either fixed or
2373 * descriptor sense data format into a common format.
2375 * @sense_buffer: byte array containing sense data returned by device
2376 * @sb_len: number of valid bytes in sense_buffer
2377 * @sshdr: pointer to instance of structure that common
2378 * elements are written to.
2381 * The "main elements" from sense data are: response_code, sense_key,
2382 * asc, ascq and additional_length (only for descriptor format).
2384 * Typically this function can be called after a device has
2385 * responded to a SCSI command with the CHECK_CONDITION status.
2388 * 1 if valid sense data information found, else 0;
2390 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
2391 struct scsi_sense_hdr
*sshdr
)
2393 if (!sense_buffer
|| !sb_len
)
2396 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
2398 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
2400 if (!scsi_sense_valid(sshdr
))
2403 if (sshdr
->response_code
>= 0x72) {
2408 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
2410 sshdr
->asc
= sense_buffer
[2];
2412 sshdr
->ascq
= sense_buffer
[3];
2414 sshdr
->additional_length
= sense_buffer
[7];
2420 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2422 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2423 sb_len
: (sense_buffer
[7] + 8);
2425 sshdr
->asc
= sense_buffer
[12];
2427 sshdr
->ascq
= sense_buffer
[13];
2433 EXPORT_SYMBOL(scsi_normalize_sense
);
2435 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2436 struct scsi_sense_hdr
*sshdr
)
2438 return scsi_normalize_sense(cmd
->sense_buffer
,
2439 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2441 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2444 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2445 * @sense_buffer: byte array of descriptor format sense data
2446 * @sb_len: number of valid bytes in sense_buffer
2447 * @desc_type: value of descriptor type to find
2448 * (e.g. 0 -> information)
2451 * only valid when sense data is in descriptor format
2454 * pointer to start of (first) descriptor if found else NULL
2456 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2459 int add_sen_len
, add_len
, desc_len
, k
;
2462 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2464 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2466 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2467 add_sen_len
: (sb_len
- 8);
2468 descp
= &sense_buffer
[8];
2469 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2471 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2472 desc_len
= add_len
+ 2;
2473 if (descp
[0] == desc_type
)
2475 if (add_len
< 0) // short descriptor ??
2480 EXPORT_SYMBOL(scsi_sense_desc_find
);
2483 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2484 * @sense_buffer: byte array of sense data
2485 * @sb_len: number of valid bytes in sense_buffer
2486 * @info_out: pointer to 64 integer where 8 or 4 byte information
2487 * field will be placed if found.
2490 * 1 if information field found, 0 if not found.
2492 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2501 switch (sense_buffer
[0] & 0x7f) {
2504 if (sense_buffer
[0] & 0x80) {
2505 *info_out
= (sense_buffer
[3] << 24) +
2506 (sense_buffer
[4] << 16) +
2507 (sense_buffer
[5] << 8) + sense_buffer
[6];
2513 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2515 if (ucp
&& (0xa == ucp
[1])) {
2517 for (j
= 0; j
< 8; ++j
) {
2530 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2533 * scsi_build_sense_buffer - build sense data in a buffer
2534 * @desc: Sense format (non zero == descriptor format,
2535 * 0 == fixed format)
2536 * @buf: Where to build sense data
2538 * @asc: Additional sense code
2539 * @ascq: Additional sense code qualifier
2542 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2545 buf
[0] = 0x72; /* descriptor, current */
2551 buf
[0] = 0x70; /* fixed, current */
2558 EXPORT_SYMBOL(scsi_build_sense_buffer
);