2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
40 #include "scsi_transport_api.h"
42 #include <trace/events/scsi.h>
44 #define SENSE_TIMEOUT (10*HZ)
47 * These should *probably* be handled by the host itself.
48 * Since it is allowed to sleep, it probably should.
50 #define BUS_RESET_SETTLE_TIME (10)
51 #define HOST_RESET_SETTLE_TIME (10)
53 /* called with shost->host_lock held */
54 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
56 if (shost
->host_busy
== shost
->host_failed
) {
57 trace_scsi_eh_wakeup(shost
);
58 wake_up_process(shost
->ehandler
);
59 SCSI_LOG_ERROR_RECOVERY(5,
60 printk("Waking error handler thread\n"));
65 * scsi_schedule_eh - schedule EH for SCSI host
66 * @shost: SCSI host to invoke error handling on.
68 * Schedule SCSI EH without scmd.
70 void scsi_schedule_eh(struct Scsi_Host
*shost
)
74 spin_lock_irqsave(shost
->host_lock
, flags
);
76 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
77 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
78 shost
->host_eh_scheduled
++;
79 scsi_eh_wakeup(shost
);
82 spin_unlock_irqrestore(shost
->host_lock
, flags
);
84 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
87 * scsi_eh_scmd_add - add scsi cmd to error handling.
88 * @scmd: scmd to run eh on.
89 * @eh_flag: optional SCSI_EH flag.
94 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
96 struct Scsi_Host
*shost
= scmd
->device
->host
;
100 if (!shost
->ehandler
)
103 spin_lock_irqsave(shost
->host_lock
, flags
);
104 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
105 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
109 scmd
->eh_eflags
|= eh_flag
;
110 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
111 shost
->host_failed
++;
112 scsi_eh_wakeup(shost
);
114 spin_unlock_irqrestore(shost
->host_lock
, flags
);
119 * scsi_times_out - Timeout function for normal scsi commands.
120 * @req: request that is timing out.
123 * We do not need to lock this. There is the potential for a race
124 * only in that the normal completion handling might run, but if the
125 * normal completion function determines that the timer has already
126 * fired, then it mustn't do anything.
128 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
130 struct scsi_cmnd
*scmd
= req
->special
;
131 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
132 struct Scsi_Host
*host
= scmd
->device
->host
;
134 trace_scsi_dispatch_cmd_timeout(scmd
);
135 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
137 if (host
->transportt
->eh_timed_out
)
138 rtn
= host
->transportt
->eh_timed_out(scmd
);
139 else if (host
->hostt
->eh_timed_out
)
140 rtn
= host
->hostt
->eh_timed_out(scmd
);
142 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
143 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
144 scmd
->result
|= DID_TIME_OUT
<< 16;
145 rtn
= BLK_EH_HANDLED
;
152 * scsi_block_when_processing_errors - Prevent cmds from being queued.
153 * @sdev: Device on which we are performing recovery.
156 * We block until the host is out of error recovery, and then check to
157 * see whether the host or the device is offline.
160 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
162 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
166 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
168 online
= scsi_device_online(sdev
);
170 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
175 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
177 #ifdef CONFIG_SCSI_LOGGING
179 * scsi_eh_prt_fail_stats - Log info on failures.
180 * @shost: scsi host being recovered.
181 * @work_q: Queue of scsi cmds to process.
183 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
184 struct list_head
*work_q
)
186 struct scsi_cmnd
*scmd
;
187 struct scsi_device
*sdev
;
188 int total_failures
= 0;
191 int devices_failed
= 0;
193 shost_for_each_device(sdev
, shost
) {
194 list_for_each_entry(scmd
, work_q
, eh_entry
) {
195 if (scmd
->device
== sdev
) {
197 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
204 if (cmd_cancel
|| cmd_failed
) {
205 SCSI_LOG_ERROR_RECOVERY(3,
206 sdev_printk(KERN_INFO
, sdev
,
207 "%s: cmds failed: %d, cancel: %d\n",
208 __func__
, cmd_failed
,
216 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
217 " devices require eh work\n",
218 total_failures
, devices_failed
));
223 * scsi_check_sense - Examine scsi cmd sense
224 * @scmd: Cmd to have sense checked.
227 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
230 * When a deferred error is detected the current command has
231 * not been executed and needs retrying.
233 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
235 struct scsi_device
*sdev
= scmd
->device
;
236 struct scsi_sense_hdr sshdr
;
238 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
239 return FAILED
; /* no valid sense data */
241 if (scsi_sense_is_deferred(&sshdr
))
244 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
245 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
248 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
249 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
251 /* handler does not care. Drop down to default handling */
255 * Previous logic looked for FILEMARK, EOM or ILI which are
256 * mainly associated with tapes and returned SUCCESS.
258 if (sshdr
.response_code
== 0x70) {
260 if (scmd
->sense_buffer
[2] & 0xe0)
264 * descriptor format: look for "stream commands sense data
265 * descriptor" (see SSC-3). Assume single sense data
266 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
268 if ((sshdr
.additional_length
> 3) &&
269 (scmd
->sense_buffer
[8] == 0x4) &&
270 (scmd
->sense_buffer
[11] & 0xe0))
274 switch (sshdr
.sense_key
) {
277 case RECOVERED_ERROR
:
278 return /* soft_error */ SUCCESS
;
280 case ABORTED_COMMAND
:
281 if (sshdr
.asc
== 0x10) /* DIF */
288 * if we are expecting a cc/ua because of a bus reset that we
289 * performed, treat this just as a retry. otherwise this is
290 * information that we should pass up to the upper-level driver
291 * so that we can deal with it there.
293 if (scmd
->device
->expecting_cc_ua
) {
294 scmd
->device
->expecting_cc_ua
= 0;
298 * if the device is in the process of becoming ready, we
301 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
304 * if the device is not started, we need to wake
305 * the error handler to start the motor
307 if (scmd
->device
->allow_restart
&&
308 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
311 if (sshdr
.asc
== 0x3f && sshdr
.ascq
== 0x0e)
312 scmd_printk(KERN_WARNING
, scmd
,
313 "Warning! Received an indication that the "
314 "LUN assignments on this target have "
315 "changed. The Linux SCSI layer does not "
316 "automatically remap LUN assignments.\n");
317 else if (sshdr
.asc
== 0x3f)
318 scmd_printk(KERN_WARNING
, scmd
,
319 "Warning! Received an indication that the "
320 "operating parameters on this target have "
321 "changed. The Linux SCSI layer does not "
322 "automatically adjust these parameters.\n");
324 if (sshdr
.asc
== 0x38 && sshdr
.ascq
== 0x07)
325 scmd_printk(KERN_WARNING
, scmd
,
326 "Warning! Received an indication that the "
327 "LUN reached a thin provisioning soft "
331 * Pass the UA upwards for a determination in the completion
336 /* these are not supported */
338 case VOLUME_OVERFLOW
:
345 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
346 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
347 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
353 if (scmd
->device
->retry_hwerror
)
354 return ADD_TO_MLQUEUE
;
358 case ILLEGAL_REQUEST
:
364 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
366 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
367 struct scsi_device
*tmp_sdev
;
369 if (!sht
->change_queue_depth
||
370 sdev
->queue_depth
>= sdev
->max_queue_depth
)
373 if (time_before(jiffies
,
374 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
377 if (time_before(jiffies
,
378 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
382 * Walk all devices of a target and do
385 shost_for_each_device(tmp_sdev
, sdev
->host
) {
386 if (tmp_sdev
->channel
!= sdev
->channel
||
387 tmp_sdev
->id
!= sdev
->id
||
388 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
391 * call back into LLD to increase queue_depth by one
392 * with ramp up reason code.
394 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
395 SCSI_QDEPTH_RAMP_UP
);
396 sdev
->last_queue_ramp_up
= jiffies
;
400 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
402 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
403 struct scsi_device
*tmp_sdev
;
405 if (!sht
->change_queue_depth
)
408 shost_for_each_device(tmp_sdev
, sdev
->host
) {
409 if (tmp_sdev
->channel
!= sdev
->channel
||
410 tmp_sdev
->id
!= sdev
->id
)
413 * We do not know the number of commands that were at
414 * the device when we got the queue full so we start
415 * from the highest possible value and work our way down.
417 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
423 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
424 * @scmd: SCSI cmd to examine.
427 * This is *only* called when we are examining the status of commands
428 * queued during error recovery. the main difference here is that we
429 * don't allow for the possibility of retries here, and we are a lot
430 * more restrictive about what we consider acceptable.
432 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
435 * first check the host byte, to see if there is anything in there
436 * that would indicate what we need to do.
438 if (host_byte(scmd
->result
) == DID_RESET
) {
440 * rats. we are already in the error handler, so we now
441 * get to try and figure out what to do next. if the sense
442 * is valid, we have a pretty good idea of what to do.
443 * if not, we mark it as FAILED.
445 return scsi_check_sense(scmd
);
447 if (host_byte(scmd
->result
) != DID_OK
)
451 * next, check the message byte.
453 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
457 * now, check the status byte to see if this indicates
460 switch (status_byte(scmd
->result
)) {
462 scsi_handle_queue_ramp_up(scmd
->device
);
463 case COMMAND_TERMINATED
:
465 case CHECK_CONDITION
:
466 return scsi_check_sense(scmd
);
468 case INTERMEDIATE_GOOD
:
469 case INTERMEDIATE_C_GOOD
:
471 * who knows? FIXME(eric)
474 case RESERVATION_CONFLICT
:
475 if (scmd
->cmnd
[0] == TEST_UNIT_READY
)
476 /* it is a success, we probed the device and
479 /* otherwise, we failed to send the command */
482 scsi_handle_queue_full(scmd
->device
);
493 * scsi_eh_done - Completion function for error handling.
494 * @scmd: Cmd that is done.
496 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
498 struct completion
*eh_action
;
500 SCSI_LOG_ERROR_RECOVERY(3,
501 printk("%s scmd: %p result: %x\n",
502 __func__
, scmd
, scmd
->result
));
504 eh_action
= scmd
->device
->host
->eh_action
;
510 * scsi_try_host_reset - ask host adapter to reset itself
511 * @scmd: SCSI cmd to send hsot reset.
513 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
517 struct Scsi_Host
*host
= scmd
->device
->host
;
518 struct scsi_host_template
*hostt
= host
->hostt
;
520 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
523 if (!hostt
->eh_host_reset_handler
)
526 rtn
= hostt
->eh_host_reset_handler(scmd
);
528 if (rtn
== SUCCESS
) {
529 if (!hostt
->skip_settle_delay
)
530 ssleep(HOST_RESET_SETTLE_TIME
);
531 spin_lock_irqsave(host
->host_lock
, flags
);
532 scsi_report_bus_reset(host
, scmd_channel(scmd
));
533 spin_unlock_irqrestore(host
->host_lock
, flags
);
540 * scsi_try_bus_reset - ask host to perform a bus reset
541 * @scmd: SCSI cmd to send bus reset.
543 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
547 struct Scsi_Host
*host
= scmd
->device
->host
;
548 struct scsi_host_template
*hostt
= host
->hostt
;
550 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
553 if (!hostt
->eh_bus_reset_handler
)
556 rtn
= hostt
->eh_bus_reset_handler(scmd
);
558 if (rtn
== SUCCESS
) {
559 if (!hostt
->skip_settle_delay
)
560 ssleep(BUS_RESET_SETTLE_TIME
);
561 spin_lock_irqsave(host
->host_lock
, flags
);
562 scsi_report_bus_reset(host
, scmd_channel(scmd
));
563 spin_unlock_irqrestore(host
->host_lock
, flags
);
569 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
572 sdev
->expecting_cc_ua
= 1;
576 * scsi_try_target_reset - Ask host to perform a target reset
577 * @scmd: SCSI cmd used to send a target reset
580 * There is no timeout for this operation. if this operation is
581 * unreliable for a given host, then the host itself needs to put a
582 * timer on it, and set the host back to a consistent state prior to
585 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
589 struct Scsi_Host
*host
= scmd
->device
->host
;
590 struct scsi_host_template
*hostt
= host
->hostt
;
592 if (!hostt
->eh_target_reset_handler
)
595 rtn
= hostt
->eh_target_reset_handler(scmd
);
596 if (rtn
== SUCCESS
) {
597 spin_lock_irqsave(host
->host_lock
, flags
);
598 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
599 __scsi_report_device_reset
);
600 spin_unlock_irqrestore(host
->host_lock
, flags
);
607 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
608 * @scmd: SCSI cmd used to send BDR
611 * There is no timeout for this operation. if this operation is
612 * unreliable for a given host, then the host itself needs to put a
613 * timer on it, and set the host back to a consistent state prior to
616 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
619 struct scsi_host_template
*hostt
= scmd
->device
->host
->hostt
;
621 if (!hostt
->eh_device_reset_handler
)
624 rtn
= hostt
->eh_device_reset_handler(scmd
);
626 __scsi_report_device_reset(scmd
->device
, NULL
);
630 static int scsi_try_to_abort_cmd(struct scsi_host_template
*hostt
, struct scsi_cmnd
*scmd
)
632 if (!hostt
->eh_abort_handler
)
635 return hostt
->eh_abort_handler(scmd
);
638 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
640 if (scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
) != SUCCESS
)
641 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
642 if (scsi_try_target_reset(scmd
) != SUCCESS
)
643 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
644 scsi_try_host_reset(scmd
);
648 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
649 * @scmd: SCSI command structure to hijack
650 * @ses: structure to save restore information
651 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
652 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
653 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
655 * This function is used to save a scsi command information before re-execution
656 * as part of the error recovery process. If @sense_bytes is 0 the command
657 * sent must be one that does not transfer any data. If @sense_bytes != 0
658 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
659 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
661 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
662 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
664 struct scsi_device
*sdev
= scmd
->device
;
667 * We need saved copies of a number of fields - this is because
668 * error handling may need to overwrite these with different values
669 * to run different commands, and once error handling is complete,
670 * we will need to restore these values prior to running the actual
673 ses
->cmd_len
= scmd
->cmd_len
;
674 ses
->cmnd
= scmd
->cmnd
;
675 ses
->data_direction
= scmd
->sc_data_direction
;
676 ses
->sdb
= scmd
->sdb
;
677 ses
->next_rq
= scmd
->request
->next_rq
;
678 ses
->result
= scmd
->result
;
679 ses
->underflow
= scmd
->underflow
;
680 ses
->prot_op
= scmd
->prot_op
;
682 scmd
->prot_op
= SCSI_PROT_NORMAL
;
683 scmd
->cmnd
= ses
->eh_cmnd
;
684 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
685 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
686 scmd
->request
->next_rq
= NULL
;
689 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
691 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
693 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
694 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
695 scmd
->sdb
.table
.nents
= 1;
696 scmd
->cmnd
[0] = REQUEST_SENSE
;
697 scmd
->cmnd
[4] = scmd
->sdb
.length
;
698 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
700 scmd
->sc_data_direction
= DMA_NONE
;
702 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
703 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
704 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
710 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
711 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
712 (sdev
->lun
<< 5 & 0xe0);
715 * Zero the sense buffer. The scsi spec mandates that any
716 * untransferred sense data should be interpreted as being zero.
718 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
720 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
723 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
724 * @scmd: SCSI command structure to restore
725 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
727 * Undo any damage done by above scsi_eh_prep_cmnd().
729 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
732 * Restore original data
734 scmd
->cmd_len
= ses
->cmd_len
;
735 scmd
->cmnd
= ses
->cmnd
;
736 scmd
->sc_data_direction
= ses
->data_direction
;
737 scmd
->sdb
= ses
->sdb
;
738 scmd
->request
->next_rq
= ses
->next_rq
;
739 scmd
->result
= ses
->result
;
740 scmd
->underflow
= ses
->underflow
;
741 scmd
->prot_op
= ses
->prot_op
;
743 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
746 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
747 * @scmd: SCSI command structure to hijack
749 * @cmnd_size: size in bytes of @cmnd
750 * @timeout: timeout for this request
751 * @sense_bytes: size of sense data to copy or 0
753 * This function is used to send a scsi command down to a target device
754 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
757 * SUCCESS or FAILED or NEEDS_RETRY
759 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
760 int cmnd_size
, int timeout
, unsigned sense_bytes
)
762 struct scsi_device
*sdev
= scmd
->device
;
763 struct Scsi_Host
*shost
= sdev
->host
;
764 DECLARE_COMPLETION_ONSTACK(done
);
765 unsigned long timeleft
;
766 struct scsi_eh_save ses
;
769 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
770 shost
->eh_action
= &done
;
773 scmd
->scsi_done
= scsi_eh_done
;
774 shost
->hostt
->queuecommand(shost
, scmd
);
776 timeleft
= wait_for_completion_timeout(&done
, timeout
);
778 shost
->eh_action
= NULL
;
780 scsi_log_completion(scmd
, SUCCESS
);
782 SCSI_LOG_ERROR_RECOVERY(3,
783 printk("%s: scmd: %p, timeleft: %ld\n",
784 __func__
, scmd
, timeleft
));
787 * If there is time left scsi_eh_done got called, and we will
788 * examine the actual status codes to see whether the command
789 * actually did complete normally, else tell the host to forget
790 * about this command.
793 rtn
= scsi_eh_completed_normally(scmd
);
794 SCSI_LOG_ERROR_RECOVERY(3,
795 printk("%s: scsi_eh_completed_normally %x\n",
812 scsi_abort_eh_cmnd(scmd
);
816 scsi_eh_restore_cmnd(scmd
, &ses
);
821 * scsi_request_sense - Request sense data from a particular target.
822 * @scmd: SCSI cmd for request sense.
825 * Some hosts automatically obtain this information, others require
826 * that we obtain it on our own. This function will *not* return until
827 * the command either times out, or it completes.
829 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
831 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
835 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
836 * @scmd: Original SCSI cmd that eh has finished.
837 * @done_q: Queue for processed commands.
840 * We don't want to use the normal command completion while we are are
841 * still handling errors - it may cause other commands to be queued,
842 * and that would disturb what we are doing. Thus we really want to
843 * keep a list of pending commands for final completion, and once we
844 * are ready to leave error handling we handle completion for real.
846 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
848 scmd
->device
->host
->host_failed
--;
850 list_move_tail(&scmd
->eh_entry
, done_q
);
852 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
855 * scsi_eh_get_sense - Get device sense data.
856 * @work_q: Queue of commands to process.
857 * @done_q: Queue of processed commands.
860 * See if we need to request sense information. if so, then get it
861 * now, so we have a better idea of what to do.
864 * This has the unfortunate side effect that if a shost adapter does
865 * not automatically request sense information, we end up shutting
866 * it down before we request it.
868 * All drivers should request sense information internally these days,
869 * so for now all I have to say is tough noogies if you end up in here.
871 * XXX: Long term this code should go away, but that needs an audit of
874 int scsi_eh_get_sense(struct list_head
*work_q
,
875 struct list_head
*done_q
)
877 struct scsi_cmnd
*scmd
, *next
;
880 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
881 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
882 SCSI_SENSE_VALID(scmd
))
885 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
886 "%s: requesting sense\n",
888 rtn
= scsi_request_sense(scmd
);
892 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
893 " result %x\n", scmd
,
895 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
897 rtn
= scsi_decide_disposition(scmd
);
900 * if the result was normal, then just pass it along to the
904 /* we don't want this command reissued, just
905 * finished with the sense data, so set
906 * retries to the max allowed to ensure it
907 * won't get reissued */
908 scmd
->retries
= scmd
->allowed
;
909 else if (rtn
!= NEEDS_RETRY
)
912 scsi_eh_finish_cmd(scmd
, done_q
);
915 return list_empty(work_q
);
917 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
920 * scsi_eh_tur - Send TUR to device.
921 * @scmd: &scsi_cmnd to send TUR
924 * 0 - Device is ready. 1 - Device NOT ready.
926 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
928 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
929 int retry_cnt
= 1, rtn
;
932 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
934 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
935 __func__
, scmd
, rtn
));
950 * scsi_eh_abort_cmds - abort pending commands.
951 * @work_q: &list_head for pending commands.
952 * @done_q: &list_head for processed commands.
955 * Try and see whether or not it makes sense to try and abort the
956 * running command. This only works out to be the case if we have one
957 * command that has timed out. If the command simply failed, it makes
958 * no sense to try and abort the command, since as far as the shost
959 * adapter is concerned, it isn't running.
961 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
962 struct list_head
*done_q
)
964 struct scsi_cmnd
*scmd
, *next
;
967 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
968 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
970 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
971 "0x%p\n", current
->comm
,
973 rtn
= scsi_try_to_abort_cmd(scmd
->device
->host
->hostt
, scmd
);
974 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
975 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
976 if (!scsi_device_online(scmd
->device
) ||
977 rtn
== FAST_IO_FAIL
||
978 !scsi_eh_tur(scmd
)) {
979 scsi_eh_finish_cmd(scmd
, done_q
);
982 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
989 return list_empty(work_q
);
993 * scsi_eh_try_stu - Send START_UNIT to device.
994 * @scmd: &scsi_cmnd to send START_UNIT
997 * 0 - Device is ready. 1 - Device NOT ready.
999 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1001 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1003 if (scmd
->device
->allow_restart
) {
1004 int i
, rtn
= NEEDS_RETRY
;
1006 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1007 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1017 * scsi_eh_stu - send START_UNIT if needed
1018 * @shost: &scsi host being recovered.
1019 * @work_q: &list_head for pending commands.
1020 * @done_q: &list_head for processed commands.
1023 * If commands are failing due to not ready, initializing command required,
1024 * try revalidating the device, which will end up sending a start unit.
1026 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1027 struct list_head
*work_q
,
1028 struct list_head
*done_q
)
1030 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1031 struct scsi_device
*sdev
;
1033 shost_for_each_device(sdev
, shost
) {
1035 list_for_each_entry(scmd
, work_q
, eh_entry
)
1036 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1037 scsi_check_sense(scmd
) == FAILED
) {
1045 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1046 " 0x%p\n", current
->comm
, sdev
));
1048 if (!scsi_eh_try_stu(stu_scmd
)) {
1049 if (!scsi_device_online(sdev
) ||
1050 !scsi_eh_tur(stu_scmd
)) {
1051 list_for_each_entry_safe(scmd
, next
,
1053 if (scmd
->device
== sdev
)
1054 scsi_eh_finish_cmd(scmd
, done_q
);
1058 SCSI_LOG_ERROR_RECOVERY(3,
1059 printk("%s: START_UNIT failed to sdev:"
1060 " 0x%p\n", current
->comm
, sdev
));
1064 return list_empty(work_q
);
1069 * scsi_eh_bus_device_reset - send bdr if needed
1070 * @shost: scsi host being recovered.
1071 * @work_q: &list_head for pending commands.
1072 * @done_q: &list_head for processed commands.
1075 * Try a bus device reset. Still, look to see whether we have multiple
1076 * devices that are jammed or not - if we have multiple devices, it
1077 * makes no sense to try bus_device_reset - we really would need to try
1078 * a bus_reset instead.
1080 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1081 struct list_head
*work_q
,
1082 struct list_head
*done_q
)
1084 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1085 struct scsi_device
*sdev
;
1088 shost_for_each_device(sdev
, shost
) {
1090 list_for_each_entry(scmd
, work_q
, eh_entry
)
1091 if (scmd
->device
== sdev
) {
1099 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1100 " 0x%p\n", current
->comm
,
1102 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1103 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1104 if (!scsi_device_online(sdev
) ||
1105 rtn
== FAST_IO_FAIL
||
1106 !scsi_eh_tur(bdr_scmd
)) {
1107 list_for_each_entry_safe(scmd
, next
,
1109 if (scmd
->device
== sdev
)
1110 scsi_eh_finish_cmd(scmd
,
1115 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1123 return list_empty(work_q
);
1127 * scsi_eh_target_reset - send target reset if needed
1128 * @shost: scsi host being recovered.
1129 * @work_q: &list_head for pending commands.
1130 * @done_q: &list_head for processed commands.
1133 * Try a target reset.
1135 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1136 struct list_head
*work_q
,
1137 struct list_head
*done_q
)
1139 LIST_HEAD(tmp_list
);
1141 list_splice_init(work_q
, &tmp_list
);
1143 while (!list_empty(&tmp_list
)) {
1144 struct scsi_cmnd
*next
, *scmd
;
1148 scmd
= list_entry(tmp_list
.next
, struct scsi_cmnd
, eh_entry
);
1151 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1153 current
->comm
, id
));
1154 rtn
= scsi_try_target_reset(scmd
);
1155 if (rtn
!= SUCCESS
&& rtn
!= FAST_IO_FAIL
)
1156 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1159 current
->comm
, id
));
1160 list_for_each_entry_safe(scmd
, next
, &tmp_list
, eh_entry
) {
1161 if (scmd_id(scmd
) != id
)
1164 if ((rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
)
1165 && (!scsi_device_online(scmd
->device
) ||
1166 rtn
== FAST_IO_FAIL
|| !scsi_eh_tur(scmd
)))
1167 scsi_eh_finish_cmd(scmd
, done_q
);
1169 /* push back on work queue for further processing */
1170 list_move(&scmd
->eh_entry
, work_q
);
1174 return list_empty(work_q
);
1178 * scsi_eh_bus_reset - send a bus reset
1179 * @shost: &scsi host being recovered.
1180 * @work_q: &list_head for pending commands.
1181 * @done_q: &list_head for processed commands.
1183 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1184 struct list_head
*work_q
,
1185 struct list_head
*done_q
)
1187 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1188 unsigned int channel
;
1192 * we really want to loop over the various channels, and do this on
1193 * a channel by channel basis. we should also check to see if any
1194 * of the failed commands are on soft_reset devices, and if so, skip
1198 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1200 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1201 if (channel
== scmd_channel(scmd
)) {
1205 * FIXME add back in some support for
1206 * soft_reset devices.
1213 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1214 " %d\n", current
->comm
,
1216 rtn
= scsi_try_bus_reset(chan_scmd
);
1217 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1218 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1219 if (channel
== scmd_channel(scmd
))
1220 if (!scsi_device_online(scmd
->device
) ||
1221 rtn
== FAST_IO_FAIL
||
1223 scsi_eh_finish_cmd(scmd
,
1227 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1228 " failed chan: %d\n",
1233 return list_empty(work_q
);
1237 * scsi_eh_host_reset - send a host reset
1238 * @work_q: list_head for processed commands.
1239 * @done_q: list_head for processed commands.
1241 static int scsi_eh_host_reset(struct list_head
*work_q
,
1242 struct list_head
*done_q
)
1244 struct scsi_cmnd
*scmd
, *next
;
1247 if (!list_empty(work_q
)) {
1248 scmd
= list_entry(work_q
->next
,
1249 struct scsi_cmnd
, eh_entry
);
1251 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1254 rtn
= scsi_try_host_reset(scmd
);
1255 if (rtn
== SUCCESS
|| rtn
== FAST_IO_FAIL
) {
1256 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1257 if (!scsi_device_online(scmd
->device
) ||
1258 rtn
== FAST_IO_FAIL
||
1259 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1261 scsi_eh_finish_cmd(scmd
, done_q
);
1264 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1269 return list_empty(work_q
);
1273 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1274 * @work_q: list_head for processed commands.
1275 * @done_q: list_head for processed commands.
1277 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1278 struct list_head
*done_q
)
1280 struct scsi_cmnd
*scmd
, *next
;
1282 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1283 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1284 "not ready after error recovery\n");
1285 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1286 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1288 * FIXME: Handle lost cmds.
1291 scsi_eh_finish_cmd(scmd
, done_q
);
1297 * scsi_noretry_cmd - determinte if command should be failed fast
1298 * @scmd: SCSI cmd to examine.
1300 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1302 switch (host_byte(scmd
->result
)) {
1306 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_TRANSPORT
);
1308 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
);
1310 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1311 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1314 case DID_SOFT_ERROR
:
1315 return (scmd
->request
->cmd_flags
& REQ_FAILFAST_DRIVER
);
1318 switch (status_byte(scmd
->result
)) {
1319 case CHECK_CONDITION
:
1321 * assume caller has checked sense and determinted
1322 * the check condition was retryable.
1324 if (scmd
->request
->cmd_flags
& REQ_FAILFAST_DEV
||
1325 scmd
->request
->cmd_type
== REQ_TYPE_BLOCK_PC
)
1333 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1334 * @scmd: SCSI cmd to examine.
1337 * This is *only* called when we are examining the status after sending
1338 * out the actual data command. any commands that are queued for error
1339 * recovery (e.g. test_unit_ready) do *not* come through here.
1341 * When this routine returns failed, it means the error handler thread
1342 * is woken. In cases where the error code indicates an error that
1343 * doesn't require the error handler read (i.e. we don't need to
1344 * abort/reset), this function should return SUCCESS.
1346 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1351 * if the device is offline, then we clearly just pass the result back
1352 * up to the top level.
1354 if (!scsi_device_online(scmd
->device
)) {
1355 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1362 * first check the host byte, to see if there is anything in there
1363 * that would indicate what we need to do.
1365 switch (host_byte(scmd
->result
)) {
1366 case DID_PASSTHROUGH
:
1368 * no matter what, pass this through to the upper layer.
1369 * nuke this special code so that it looks like we are saying
1372 scmd
->result
&= 0xff00ffff;
1376 * looks good. drop through, and check the next byte.
1379 case DID_NO_CONNECT
:
1380 case DID_BAD_TARGET
:
1383 * note - this means that we just report the status back
1384 * to the top level driver, not that we actually think
1385 * that it indicates SUCCESS.
1389 * when the low level driver returns did_soft_error,
1390 * it is responsible for keeping an internal retry counter
1391 * in order to avoid endless loops (db)
1393 * actually this is a bug in this function here. we should
1394 * be mindful of the maximum number of retries specified
1395 * and not get stuck in a loop.
1397 case DID_SOFT_ERROR
:
1403 return ADD_TO_MLQUEUE
;
1404 case DID_TRANSPORT_DISRUPTED
:
1406 * LLD/transport was disrupted during processing of the IO.
1407 * The transport class is now blocked/blocking,
1408 * and the transport will decide what to do with the IO
1409 * based on its timers and recovery capablilities if
1410 * there are enough retries.
1413 case DID_TRANSPORT_FAILFAST
:
1415 * The transport decided to failfast the IO (most likely
1416 * the fast io fail tmo fired), so send IO directly upwards.
1420 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1421 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1423 * execute reservation conflict processing code
1433 * when we scan the bus, we get timeout messages for
1434 * these commands if there is no device available.
1435 * other hosts report did_no_connect for the same thing.
1437 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1438 scmd
->cmnd
[0] == INQUIRY
)) {
1450 * next, check the message byte.
1452 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1456 * check the status byte to see if this indicates anything special.
1458 switch (status_byte(scmd
->result
)) {
1460 scsi_handle_queue_full(scmd
->device
);
1462 * the case of trying to send too many commands to a
1463 * tagged queueing device.
1467 * device can't talk to us at the moment. Should only
1468 * occur (SAM-3) when the task queue is empty, so will cause
1469 * the empty queue handling to trigger a stall in the
1472 return ADD_TO_MLQUEUE
;
1474 scsi_handle_queue_ramp_up(scmd
->device
);
1475 case COMMAND_TERMINATED
:
1479 case CHECK_CONDITION
:
1480 rtn
= scsi_check_sense(scmd
);
1481 if (rtn
== NEEDS_RETRY
)
1483 else if (rtn
== TARGET_ERROR
) {
1485 * Need to modify host byte to signal a
1486 * permanent target failure
1488 scmd
->result
|= (DID_TARGET_FAILURE
<< 16);
1491 /* if rtn == FAILED, we have no sense information;
1492 * returning FAILED will wake the error handler thread
1493 * to collect the sense and redo the decide
1496 case CONDITION_GOOD
:
1497 case INTERMEDIATE_GOOD
:
1498 case INTERMEDIATE_C_GOOD
:
1501 * who knows? FIXME(eric)
1505 case RESERVATION_CONFLICT
:
1506 sdev_printk(KERN_INFO
, scmd
->device
,
1507 "reservation conflict\n");
1508 scmd
->result
|= (DID_NEXUS_FAILURE
<< 16);
1509 return SUCCESS
; /* causes immediate i/o error */
1517 /* we requeue for retry because the error was retryable, and
1518 * the request was not marked fast fail. Note that above,
1519 * even if the request is marked fast fail, we still requeue
1520 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1521 if ((++scmd
->retries
) <= scmd
->allowed
1522 && !scsi_noretry_cmd(scmd
)) {
1526 * no more retries - report this one back to upper level.
1532 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1534 __blk_put_request(req
->q
, req
);
1538 * scsi_eh_lock_door - Prevent medium removal for the specified device
1539 * @sdev: SCSI device to prevent medium removal
1542 * We must be called from process context.
1545 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1546 * head of the devices request queue, and continue.
1548 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1550 struct request
*req
;
1553 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1554 * request becomes available
1556 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1558 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1562 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1565 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1567 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1568 req
->cmd_flags
|= REQ_QUIET
;
1569 req
->timeout
= 10 * HZ
;
1572 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1576 * scsi_restart_operations - restart io operations to the specified host.
1577 * @shost: Host we are restarting.
1580 * When we entered the error handler, we blocked all further i/o to
1581 * this device. we need to 'reverse' this process.
1583 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1585 struct scsi_device
*sdev
;
1586 unsigned long flags
;
1589 * If the door was locked, we need to insert a door lock request
1590 * onto the head of the SCSI request queue for the device. There
1591 * is no point trying to lock the door of an off-line device.
1593 shost_for_each_device(sdev
, shost
) {
1594 if (scsi_device_online(sdev
) && sdev
->locked
)
1595 scsi_eh_lock_door(sdev
);
1599 * next free up anything directly waiting upon the host. this
1600 * will be requests for character device operations, and also for
1601 * ioctls to queued block devices.
1603 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1606 spin_lock_irqsave(shost
->host_lock
, flags
);
1607 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1608 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1609 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1610 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1612 wake_up(&shost
->host_wait
);
1615 * finally we need to re-initiate requests that may be pending. we will
1616 * have had everything blocked while error handling is taking place, and
1617 * now that error recovery is done, we will need to ensure that these
1618 * requests are started.
1620 scsi_run_host_queues(shost
);
1624 * scsi_eh_ready_devs - check device ready state and recover if not.
1625 * @shost: host to be recovered.
1626 * @work_q: &list_head for pending commands.
1627 * @done_q: &list_head for processed commands.
1629 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1630 struct list_head
*work_q
,
1631 struct list_head
*done_q
)
1633 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1634 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1635 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1636 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1637 if (!scsi_eh_host_reset(work_q
, done_q
))
1638 scsi_eh_offline_sdevs(work_q
,
1641 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1644 * scsi_eh_flush_done_q - finish processed commands or retry them.
1645 * @done_q: list_head of processed commands.
1647 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1649 struct scsi_cmnd
*scmd
, *next
;
1651 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1652 list_del_init(&scmd
->eh_entry
);
1653 if (scsi_device_online(scmd
->device
) &&
1654 !scsi_noretry_cmd(scmd
) &&
1655 (++scmd
->retries
<= scmd
->allowed
)) {
1656 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1660 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1663 * If just we got sense for the device (called
1664 * scsi_eh_get_sense), scmd->result is already
1665 * set, do not set DRIVER_TIMEOUT.
1668 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1669 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1671 current
->comm
, scmd
));
1672 scsi_finish_command(scmd
);
1676 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1679 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1680 * @shost: Host to unjam.
1683 * When we come in here, we *know* that all commands on the bus have
1684 * either completed, failed or timed out. we also know that no further
1685 * commands are being sent to the host, so things are relatively quiet
1686 * and we have freedom to fiddle with things as we wish.
1688 * This is only the *default* implementation. it is possible for
1689 * individual drivers to supply their own version of this function, and
1690 * if the maintainer wishes to do this, it is strongly suggested that
1691 * this function be taken as a template and modified. this function
1692 * was designed to correctly handle problems for about 95% of the
1693 * different cases out there, and it should always provide at least a
1694 * reasonable amount of error recovery.
1696 * Any command marked 'failed' or 'timeout' must eventually have
1697 * scsi_finish_cmd() called for it. we do all of the retry stuff
1698 * here, so when we restart the host after we return it should have an
1701 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1703 unsigned long flags
;
1704 LIST_HEAD(eh_work_q
);
1705 LIST_HEAD(eh_done_q
);
1707 spin_lock_irqsave(shost
->host_lock
, flags
);
1708 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1709 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1711 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1713 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1714 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1715 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1717 scsi_eh_flush_done_q(&eh_done_q
);
1721 * scsi_error_handler - SCSI error handler thread
1722 * @data: Host for which we are running.
1725 * This is the main error handling loop. This is run as a kernel thread
1726 * for every SCSI host and handles all error handling activity.
1728 int scsi_error_handler(void *data
)
1730 struct Scsi_Host
*shost
= data
;
1733 * We use TASK_INTERRUPTIBLE so that the thread is not
1734 * counted against the load average as a running process.
1735 * We never actually get interrupted because kthread_run
1736 * disables signal delivery for the created thread.
1738 set_current_state(TASK_INTERRUPTIBLE
);
1739 while (!kthread_should_stop()) {
1740 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1741 shost
->host_failed
!= shost
->host_busy
) {
1742 SCSI_LOG_ERROR_RECOVERY(1,
1743 printk("Error handler scsi_eh_%d sleeping\n",
1746 set_current_state(TASK_INTERRUPTIBLE
);
1750 __set_current_state(TASK_RUNNING
);
1751 SCSI_LOG_ERROR_RECOVERY(1,
1752 printk("Error handler scsi_eh_%d waking up\n",
1756 * We have a host that is failing for some reason. Figure out
1757 * what we need to do to get it up and online again (if we can).
1758 * If we fail, we end up taking the thing offline.
1760 if (scsi_autopm_get_host(shost
) != 0) {
1761 SCSI_LOG_ERROR_RECOVERY(1,
1762 printk(KERN_ERR
"Error handler scsi_eh_%d "
1763 "unable to autoresume\n",
1768 if (shost
->transportt
->eh_strategy_handler
)
1769 shost
->transportt
->eh_strategy_handler(shost
);
1771 scsi_unjam_host(shost
);
1774 * Note - if the above fails completely, the action is to take
1775 * individual devices offline and flush the queue of any
1776 * outstanding requests that may have been pending. When we
1777 * restart, we restart any I/O to any other devices on the bus
1778 * which are still online.
1780 scsi_restart_operations(shost
);
1781 scsi_autopm_put_host(shost
);
1782 set_current_state(TASK_INTERRUPTIBLE
);
1784 __set_current_state(TASK_RUNNING
);
1786 SCSI_LOG_ERROR_RECOVERY(1,
1787 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1788 shost
->ehandler
= NULL
;
1793 * Function: scsi_report_bus_reset()
1795 * Purpose: Utility function used by low-level drivers to report that
1796 * they have observed a bus reset on the bus being handled.
1798 * Arguments: shost - Host in question
1799 * channel - channel on which reset was observed.
1803 * Lock status: Host lock must be held.
1805 * Notes: This only needs to be called if the reset is one which
1806 * originates from an unknown location. Resets originated
1807 * by the mid-level itself don't need to call this, but there
1808 * should be no harm.
1810 * The main purpose of this is to make sure that a CHECK_CONDITION
1811 * is properly treated.
1813 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1815 struct scsi_device
*sdev
;
1817 __shost_for_each_device(sdev
, shost
) {
1818 if (channel
== sdev_channel(sdev
))
1819 __scsi_report_device_reset(sdev
, NULL
);
1822 EXPORT_SYMBOL(scsi_report_bus_reset
);
1825 * Function: scsi_report_device_reset()
1827 * Purpose: Utility function used by low-level drivers to report that
1828 * they have observed a device reset on the device being handled.
1830 * Arguments: shost - Host in question
1831 * channel - channel on which reset was observed
1832 * target - target on which reset was observed
1836 * Lock status: Host lock must be held
1838 * Notes: This only needs to be called if the reset is one which
1839 * originates from an unknown location. Resets originated
1840 * by the mid-level itself don't need to call this, but there
1841 * should be no harm.
1843 * The main purpose of this is to make sure that a CHECK_CONDITION
1844 * is properly treated.
1846 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1848 struct scsi_device
*sdev
;
1850 __shost_for_each_device(sdev
, shost
) {
1851 if (channel
== sdev_channel(sdev
) &&
1852 target
== sdev_id(sdev
))
1853 __scsi_report_device_reset(sdev
, NULL
);
1856 EXPORT_SYMBOL(scsi_report_device_reset
);
1859 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1864 * Function: scsi_reset_provider
1866 * Purpose: Send requested reset to a bus or device at any phase.
1868 * Arguments: device - device to send reset to
1869 * flag - reset type (see scsi.h)
1871 * Returns: SUCCESS/FAILURE.
1873 * Notes: This is used by the SCSI Generic driver to provide
1874 * Bus/Device reset capability.
1877 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1879 struct scsi_cmnd
*scmd
;
1880 struct Scsi_Host
*shost
= dev
->host
;
1882 unsigned long flags
;
1885 if (scsi_autopm_get_host(shost
) < 0)
1888 scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1889 blk_rq_init(NULL
, &req
);
1890 scmd
->request
= &req
;
1892 scmd
->cmnd
= req
.cmd
;
1894 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1895 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1899 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1901 spin_lock_irqsave(shost
->host_lock
, flags
);
1902 shost
->tmf_in_progress
= 1;
1903 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1906 case SCSI_TRY_RESET_DEVICE
:
1907 rtn
= scsi_try_bus_device_reset(scmd
);
1911 case SCSI_TRY_RESET_TARGET
:
1912 rtn
= scsi_try_target_reset(scmd
);
1916 case SCSI_TRY_RESET_BUS
:
1917 rtn
= scsi_try_bus_reset(scmd
);
1921 case SCSI_TRY_RESET_HOST
:
1922 rtn
= scsi_try_host_reset(scmd
);
1928 spin_lock_irqsave(shost
->host_lock
, flags
);
1929 shost
->tmf_in_progress
= 0;
1930 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1933 * be sure to wake up anyone who was sleeping or had their queue
1934 * suspended while we performed the TMF.
1936 SCSI_LOG_ERROR_RECOVERY(3,
1937 printk("%s: waking up host to restart after TMF\n",
1940 wake_up(&shost
->host_wait
);
1942 scsi_run_host_queues(shost
);
1944 scsi_next_command(scmd
);
1945 scsi_autopm_put_host(shost
);
1948 EXPORT_SYMBOL(scsi_reset_provider
);
1951 * scsi_normalize_sense - normalize main elements from either fixed or
1952 * descriptor sense data format into a common format.
1954 * @sense_buffer: byte array containing sense data returned by device
1955 * @sb_len: number of valid bytes in sense_buffer
1956 * @sshdr: pointer to instance of structure that common
1957 * elements are written to.
1960 * The "main elements" from sense data are: response_code, sense_key,
1961 * asc, ascq and additional_length (only for descriptor format).
1963 * Typically this function can be called after a device has
1964 * responded to a SCSI command with the CHECK_CONDITION status.
1967 * 1 if valid sense data information found, else 0;
1969 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1970 struct scsi_sense_hdr
*sshdr
)
1972 if (!sense_buffer
|| !sb_len
)
1975 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1977 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1979 if (!scsi_sense_valid(sshdr
))
1982 if (sshdr
->response_code
>= 0x72) {
1987 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1989 sshdr
->asc
= sense_buffer
[2];
1991 sshdr
->ascq
= sense_buffer
[3];
1993 sshdr
->additional_length
= sense_buffer
[7];
1999 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
2001 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
2002 sb_len
: (sense_buffer
[7] + 8);
2004 sshdr
->asc
= sense_buffer
[12];
2006 sshdr
->ascq
= sense_buffer
[13];
2012 EXPORT_SYMBOL(scsi_normalize_sense
);
2014 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
2015 struct scsi_sense_hdr
*sshdr
)
2017 return scsi_normalize_sense(cmd
->sense_buffer
,
2018 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2020 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2023 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2024 * @sense_buffer: byte array of descriptor format sense data
2025 * @sb_len: number of valid bytes in sense_buffer
2026 * @desc_type: value of descriptor type to find
2027 * (e.g. 0 -> information)
2030 * only valid when sense data is in descriptor format
2033 * pointer to start of (first) descriptor if found else NULL
2035 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2038 int add_sen_len
, add_len
, desc_len
, k
;
2041 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2043 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2045 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2046 add_sen_len
: (sb_len
- 8);
2047 descp
= &sense_buffer
[8];
2048 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2050 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2051 desc_len
= add_len
+ 2;
2052 if (descp
[0] == desc_type
)
2054 if (add_len
< 0) // short descriptor ??
2059 EXPORT_SYMBOL(scsi_sense_desc_find
);
2062 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2063 * @sense_buffer: byte array of sense data
2064 * @sb_len: number of valid bytes in sense_buffer
2065 * @info_out: pointer to 64 integer where 8 or 4 byte information
2066 * field will be placed if found.
2069 * 1 if information field found, 0 if not found.
2071 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2080 switch (sense_buffer
[0] & 0x7f) {
2083 if (sense_buffer
[0] & 0x80) {
2084 *info_out
= (sense_buffer
[3] << 24) +
2085 (sense_buffer
[4] << 16) +
2086 (sense_buffer
[5] << 8) + sense_buffer
[6];
2092 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2094 if (ucp
&& (0xa == ucp
[1])) {
2096 for (j
= 0; j
< 8; ++j
) {
2109 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2112 * scsi_build_sense_buffer - build sense data in a buffer
2113 * @desc: Sense format (non zero == descriptor format,
2114 * 0 == fixed format)
2115 * @buf: Where to build sense data
2117 * @asc: Additional sense code
2118 * @ascq: Additional sense code qualifier
2121 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2124 buf
[0] = 0x72; /* descriptor, current */
2130 buf
[0] = 0x70; /* fixed, current */
2137 EXPORT_SYMBOL(scsi_build_sense_buffer
);