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/timer.h>
20 #include <linux/string.h>
21 #include <linux/kernel.h>
22 #include <linux/freezer.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
39 #include "scsi_transport_api.h"
41 #define SENSE_TIMEOUT (10*HZ)
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
47 #define BUS_RESET_SETTLE_TIME (10)
48 #define HOST_RESET_SETTLE_TIME (10)
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
53 if (shost
->host_busy
== shost
->host_failed
) {
54 wake_up_process(shost
->ehandler
);
55 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
64 * Schedule SCSI EH without scmd.
66 void scsi_schedule_eh(struct Scsi_Host
*shost
)
70 spin_lock_irqsave(shost
->host_lock
, flags
);
72 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
73 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
74 shost
->host_eh_scheduled
++;
75 scsi_eh_wakeup(shost
);
78 spin_unlock_irqrestore(shost
->host_lock
, flags
);
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
90 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
92 struct Scsi_Host
*shost
= scmd
->device
->host
;
99 spin_lock_irqsave(shost
->host_lock
, flags
);
100 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
101 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
105 scmd
->eh_eflags
|= eh_flag
;
106 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
107 shost
->host_failed
++;
108 scsi_eh_wakeup(shost
);
110 spin_unlock_irqrestore(shost
->host_lock
, flags
);
115 * scsi_times_out - Timeout function for normal scsi commands.
116 * @req: request that is timing out.
119 * We do not need to lock this. There is the potential for a race
120 * only in that the normal completion handling might run, but if the
121 * normal completion function determines that the timer has already
122 * fired, then it mustn't do anything.
124 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
126 struct scsi_cmnd
*scmd
= req
->special
;
127 enum blk_eh_timer_return (*eh_timed_out
)(struct scsi_cmnd
*);
128 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
130 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
132 if (scmd
->device
->host
->transportt
->eh_timed_out
)
133 eh_timed_out
= scmd
->device
->host
->transportt
->eh_timed_out
;
134 else if (scmd
->device
->host
->hostt
->eh_timed_out
)
135 eh_timed_out
= scmd
->device
->host
->hostt
->eh_timed_out
;
140 rtn
= eh_timed_out(scmd
);
142 case BLK_EH_NOT_HANDLED
:
148 if (unlikely(!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
149 scmd
->result
|= DID_TIME_OUT
<< 16;
150 return BLK_EH_HANDLED
;
153 return BLK_EH_NOT_HANDLED
;
157 * scsi_block_when_processing_errors - Prevent cmds from being queued.
158 * @sdev: Device on which we are performing recovery.
161 * We block until the host is out of error recovery, and then check to
162 * see whether the host or the device is offline.
165 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
167 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
171 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
173 online
= scsi_device_online(sdev
);
175 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
180 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
182 #ifdef CONFIG_SCSI_LOGGING
184 * scsi_eh_prt_fail_stats - Log info on failures.
185 * @shost: scsi host being recovered.
186 * @work_q: Queue of scsi cmds to process.
188 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
189 struct list_head
*work_q
)
191 struct scsi_cmnd
*scmd
;
192 struct scsi_device
*sdev
;
193 int total_failures
= 0;
196 int devices_failed
= 0;
198 shost_for_each_device(sdev
, shost
) {
199 list_for_each_entry(scmd
, work_q
, eh_entry
) {
200 if (scmd
->device
== sdev
) {
202 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
209 if (cmd_cancel
|| cmd_failed
) {
210 SCSI_LOG_ERROR_RECOVERY(3,
211 sdev_printk(KERN_INFO
, sdev
,
212 "%s: cmds failed: %d, cancel: %d\n",
213 __func__
, cmd_failed
,
221 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
222 " devices require eh work\n",
223 total_failures
, devices_failed
));
228 * scsi_check_sense - Examine scsi cmd sense
229 * @scmd: Cmd to have sense checked.
232 * SUCCESS or FAILED or NEEDS_RETRY
235 * When a deferred error is detected the current command has
236 * not been executed and needs retrying.
238 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
240 struct scsi_device
*sdev
= scmd
->device
;
241 struct scsi_sense_hdr sshdr
;
243 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
244 return FAILED
; /* no valid sense data */
246 if (scsi_sense_is_deferred(&sshdr
))
249 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
250 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
253 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
254 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
256 /* handler does not care. Drop down to default handling */
260 * Previous logic looked for FILEMARK, EOM or ILI which are
261 * mainly associated with tapes and returned SUCCESS.
263 if (sshdr
.response_code
== 0x70) {
265 if (scmd
->sense_buffer
[2] & 0xe0)
269 * descriptor format: look for "stream commands sense data
270 * descriptor" (see SSC-3). Assume single sense data
271 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
273 if ((sshdr
.additional_length
> 3) &&
274 (scmd
->sense_buffer
[8] == 0x4) &&
275 (scmd
->sense_buffer
[11] & 0xe0))
279 switch (sshdr
.sense_key
) {
282 case RECOVERED_ERROR
:
283 return /* soft_error */ SUCCESS
;
285 case ABORTED_COMMAND
:
286 if (sshdr
.asc
== 0x10) /* DIF */
293 * if we are expecting a cc/ua because of a bus reset that we
294 * performed, treat this just as a retry. otherwise this is
295 * information that we should pass up to the upper-level driver
296 * so that we can deal with it there.
298 if (scmd
->device
->expecting_cc_ua
) {
299 scmd
->device
->expecting_cc_ua
= 0;
303 * if the device is in the process of becoming ready, we
306 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
309 * if the device is not started, we need to wake
310 * the error handler to start the motor
312 if (scmd
->device
->allow_restart
&&
313 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
317 /* these three are not supported */
319 case VOLUME_OVERFLOW
:
324 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
325 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
326 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
332 if (scmd
->device
->retry_hwerror
)
333 return ADD_TO_MLQUEUE
;
337 case ILLEGAL_REQUEST
:
346 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
347 * @scmd: SCSI cmd to examine.
350 * This is *only* called when we are examining the status of commands
351 * queued during error recovery. the main difference here is that we
352 * don't allow for the possibility of retries here, and we are a lot
353 * more restrictive about what we consider acceptable.
355 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
358 * first check the host byte, to see if there is anything in there
359 * that would indicate what we need to do.
361 if (host_byte(scmd
->result
) == DID_RESET
) {
363 * rats. we are already in the error handler, so we now
364 * get to try and figure out what to do next. if the sense
365 * is valid, we have a pretty good idea of what to do.
366 * if not, we mark it as FAILED.
368 return scsi_check_sense(scmd
);
370 if (host_byte(scmd
->result
) != DID_OK
)
374 * next, check the message byte.
376 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
380 * now, check the status byte to see if this indicates
383 switch (status_byte(scmd
->result
)) {
385 case COMMAND_TERMINATED
:
387 case CHECK_CONDITION
:
388 return scsi_check_sense(scmd
);
390 case INTERMEDIATE_GOOD
:
391 case INTERMEDIATE_C_GOOD
:
393 * who knows? FIXME(eric)
398 case RESERVATION_CONFLICT
:
406 * scsi_eh_done - Completion function for error handling.
407 * @scmd: Cmd that is done.
409 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
411 struct completion
*eh_action
;
413 SCSI_LOG_ERROR_RECOVERY(3,
414 printk("%s scmd: %p result: %x\n",
415 __func__
, scmd
, scmd
->result
));
417 eh_action
= scmd
->device
->host
->eh_action
;
423 * scsi_try_host_reset - ask host adapter to reset itself
424 * @scmd: SCSI cmd to send hsot reset.
426 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
431 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
434 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
437 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
439 if (rtn
== SUCCESS
) {
440 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
441 ssleep(HOST_RESET_SETTLE_TIME
);
442 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
443 scsi_report_bus_reset(scmd
->device
->host
,
445 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
452 * scsi_try_bus_reset - ask host to perform a bus reset
453 * @scmd: SCSI cmd to send bus reset.
455 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
460 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
463 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
466 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
468 if (rtn
== SUCCESS
) {
469 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
470 ssleep(BUS_RESET_SETTLE_TIME
);
471 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
472 scsi_report_bus_reset(scmd
->device
->host
,
474 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
480 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
483 sdev
->expecting_cc_ua
= 1;
487 * scsi_try_target_reset - Ask host to perform a target reset
488 * @scmd: SCSI cmd used to send a target reset
491 * There is no timeout for this operation. if this operation is
492 * unreliable for a given host, then the host itself needs to put a
493 * timer on it, and set the host back to a consistent state prior to
496 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
501 if (!scmd
->device
->host
->hostt
->eh_target_reset_handler
)
504 rtn
= scmd
->device
->host
->hostt
->eh_target_reset_handler(scmd
);
505 if (rtn
== SUCCESS
) {
506 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
507 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
508 __scsi_report_device_reset
);
509 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
516 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
517 * @scmd: SCSI cmd used to send BDR
520 * There is no timeout for this operation. if this operation is
521 * unreliable for a given host, then the host itself needs to put a
522 * timer on it, and set the host back to a consistent state prior to
525 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
529 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
532 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
534 __scsi_report_device_reset(scmd
->device
, NULL
);
538 static int __scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
540 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
543 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
547 * scsi_try_to_abort_cmd - Ask host to abort a running command.
548 * @scmd: SCSI cmd to abort from Lower Level.
551 * This function will not return until the user's completion function
552 * has been called. there is no timeout on this operation. if the
553 * author of the low-level driver wishes this operation to be timed,
554 * they can provide this facility themselves. helper functions in
555 * scsi_error.c can be supplied to make this easier to do.
557 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
560 * scsi_done was called just after the command timed out and before
561 * we had a chance to process it. (db)
563 if (scmd
->serial_number
== 0)
565 return __scsi_try_to_abort_cmd(scmd
);
568 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
570 if (__scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
571 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
572 if (scsi_try_target_reset(scmd
) != SUCCESS
)
573 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
574 scsi_try_host_reset(scmd
);
578 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
579 * @scmd: SCSI command structure to hijack
580 * @ses: structure to save restore information
581 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
582 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
583 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
585 * This function is used to save a scsi command information before re-execution
586 * as part of the error recovery process. If @sense_bytes is 0 the command
587 * sent must be one that does not transfer any data. If @sense_bytes != 0
588 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
589 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
591 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
592 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
594 struct scsi_device
*sdev
= scmd
->device
;
597 * We need saved copies of a number of fields - this is because
598 * error handling may need to overwrite these with different values
599 * to run different commands, and once error handling is complete,
600 * we will need to restore these values prior to running the actual
603 ses
->cmd_len
= scmd
->cmd_len
;
604 ses
->cmnd
= scmd
->cmnd
;
605 ses
->data_direction
= scmd
->sc_data_direction
;
606 ses
->sdb
= scmd
->sdb
;
607 ses
->next_rq
= scmd
->request
->next_rq
;
608 ses
->result
= scmd
->result
;
609 ses
->underflow
= scmd
->underflow
;
610 ses
->prot_op
= scmd
->prot_op
;
612 scmd
->prot_op
= SCSI_PROT_NORMAL
;
613 scmd
->cmnd
= ses
->eh_cmnd
;
614 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
615 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
616 scmd
->request
->next_rq
= NULL
;
619 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
621 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
623 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
624 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
625 scmd
->sdb
.table
.nents
= 1;
626 scmd
->cmnd
[0] = REQUEST_SENSE
;
627 scmd
->cmnd
[4] = scmd
->sdb
.length
;
628 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
630 scmd
->sc_data_direction
= DMA_NONE
;
632 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
633 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
634 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
640 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
641 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
642 (sdev
->lun
<< 5 & 0xe0);
645 * Zero the sense buffer. The scsi spec mandates that any
646 * untransferred sense data should be interpreted as being zero.
648 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
650 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
653 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
654 * @scmd: SCSI command structure to restore
655 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
657 * Undo any damage done by above scsi_prep_eh_cmnd().
659 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
662 * Restore original data
664 scmd
->cmd_len
= ses
->cmd_len
;
665 scmd
->cmnd
= ses
->cmnd
;
666 scmd
->sc_data_direction
= ses
->data_direction
;
667 scmd
->sdb
= ses
->sdb
;
668 scmd
->request
->next_rq
= ses
->next_rq
;
669 scmd
->result
= ses
->result
;
670 scmd
->underflow
= ses
->underflow
;
671 scmd
->prot_op
= ses
->prot_op
;
673 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
676 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
677 * @scmd: SCSI command structure to hijack
679 * @cmnd_size: size in bytes of @cmnd
680 * @timeout: timeout for this request
681 * @sense_bytes: size of sense data to copy or 0
683 * This function is used to send a scsi command down to a target device
684 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
687 * SUCCESS or FAILED or NEEDS_RETRY
689 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
690 int cmnd_size
, int timeout
, unsigned sense_bytes
)
692 struct scsi_device
*sdev
= scmd
->device
;
693 struct Scsi_Host
*shost
= sdev
->host
;
694 DECLARE_COMPLETION_ONSTACK(done
);
695 unsigned long timeleft
;
697 struct scsi_eh_save ses
;
700 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
701 shost
->eh_action
= &done
;
703 spin_lock_irqsave(shost
->host_lock
, flags
);
705 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
706 spin_unlock_irqrestore(shost
->host_lock
, flags
);
708 timeleft
= wait_for_completion_timeout(&done
, timeout
);
710 shost
->eh_action
= NULL
;
712 scsi_log_completion(scmd
, SUCCESS
);
714 SCSI_LOG_ERROR_RECOVERY(3,
715 printk("%s: scmd: %p, timeleft: %ld\n",
716 __func__
, scmd
, timeleft
));
719 * If there is time left scsi_eh_done got called, and we will
720 * examine the actual status codes to see whether the command
721 * actually did complete normally, else tell the host to forget
722 * about this command.
725 rtn
= scsi_eh_completed_normally(scmd
);
726 SCSI_LOG_ERROR_RECOVERY(3,
727 printk("%s: scsi_eh_completed_normally %x\n",
740 scsi_abort_eh_cmnd(scmd
);
744 scsi_eh_restore_cmnd(scmd
, &ses
);
749 * scsi_request_sense - Request sense data from a particular target.
750 * @scmd: SCSI cmd for request sense.
753 * Some hosts automatically obtain this information, others require
754 * that we obtain it on our own. This function will *not* return until
755 * the command either times out, or it completes.
757 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
759 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
763 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
764 * @scmd: Original SCSI cmd that eh has finished.
765 * @done_q: Queue for processed commands.
768 * We don't want to use the normal command completion while we are are
769 * still handling errors - it may cause other commands to be queued,
770 * and that would disturb what we are doing. Thus we really want to
771 * keep a list of pending commands for final completion, and once we
772 * are ready to leave error handling we handle completion for real.
774 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
776 scmd
->device
->host
->host_failed
--;
778 list_move_tail(&scmd
->eh_entry
, done_q
);
780 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
783 * scsi_eh_get_sense - Get device sense data.
784 * @work_q: Queue of commands to process.
785 * @done_q: Queue of processed commands.
788 * See if we need to request sense information. if so, then get it
789 * now, so we have a better idea of what to do.
792 * This has the unfortunate side effect that if a shost adapter does
793 * not automatically request sense information, we end up shutting
794 * it down before we request it.
796 * All drivers should request sense information internally these days,
797 * so for now all I have to say is tough noogies if you end up in here.
799 * XXX: Long term this code should go away, but that needs an audit of
802 int scsi_eh_get_sense(struct list_head
*work_q
,
803 struct list_head
*done_q
)
805 struct scsi_cmnd
*scmd
, *next
;
808 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
809 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
810 SCSI_SENSE_VALID(scmd
))
813 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
814 "%s: requesting sense\n",
816 rtn
= scsi_request_sense(scmd
);
820 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
821 " result %x\n", scmd
,
823 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
825 rtn
= scsi_decide_disposition(scmd
);
828 * if the result was normal, then just pass it along to the
832 /* we don't want this command reissued, just
833 * finished with the sense data, so set
834 * retries to the max allowed to ensure it
835 * won't get reissued */
836 scmd
->retries
= scmd
->allowed
;
837 else if (rtn
!= NEEDS_RETRY
)
840 scsi_eh_finish_cmd(scmd
, done_q
);
843 return list_empty(work_q
);
845 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
848 * scsi_eh_tur - Send TUR to device.
849 * @scmd: &scsi_cmnd to send TUR
852 * 0 - Device is ready. 1 - Device NOT ready.
854 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
856 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
857 int retry_cnt
= 1, rtn
;
860 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
862 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
863 __func__
, scmd
, rtn
));
878 * scsi_eh_abort_cmds - abort pending commands.
879 * @work_q: &list_head for pending commands.
880 * @done_q: &list_head for processed commands.
883 * Try and see whether or not it makes sense to try and abort the
884 * running command. This only works out to be the case if we have one
885 * command that has timed out. If the command simply failed, it makes
886 * no sense to try and abort the command, since as far as the shost
887 * adapter is concerned, it isn't running.
889 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
890 struct list_head
*done_q
)
892 struct scsi_cmnd
*scmd
, *next
;
895 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
896 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
898 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
899 "0x%p\n", current
->comm
,
901 rtn
= scsi_try_to_abort_cmd(scmd
);
902 if (rtn
== SUCCESS
) {
903 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
904 if (!scsi_device_online(scmd
->device
) ||
905 !scsi_eh_tur(scmd
)) {
906 scsi_eh_finish_cmd(scmd
, done_q
);
910 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
917 return list_empty(work_q
);
921 * scsi_eh_try_stu - Send START_UNIT to device.
922 * @scmd: &scsi_cmnd to send START_UNIT
925 * 0 - Device is ready. 1 - Device NOT ready.
927 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
929 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
931 if (scmd
->device
->allow_restart
) {
932 int i
, rtn
= NEEDS_RETRY
;
934 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
935 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6,
936 scmd
->device
->timeout
, 0);
946 * scsi_eh_stu - send START_UNIT if needed
947 * @shost: &scsi host being recovered.
948 * @work_q: &list_head for pending commands.
949 * @done_q: &list_head for processed commands.
952 * If commands are failing due to not ready, initializing command required,
953 * try revalidating the device, which will end up sending a start unit.
955 static int scsi_eh_stu(struct Scsi_Host
*shost
,
956 struct list_head
*work_q
,
957 struct list_head
*done_q
)
959 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
960 struct scsi_device
*sdev
;
962 shost_for_each_device(sdev
, shost
) {
964 list_for_each_entry(scmd
, work_q
, eh_entry
)
965 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
966 scsi_check_sense(scmd
) == FAILED
) {
974 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
975 " 0x%p\n", current
->comm
, sdev
));
977 if (!scsi_eh_try_stu(stu_scmd
)) {
978 if (!scsi_device_online(sdev
) ||
979 !scsi_eh_tur(stu_scmd
)) {
980 list_for_each_entry_safe(scmd
, next
,
982 if (scmd
->device
== sdev
)
983 scsi_eh_finish_cmd(scmd
, done_q
);
987 SCSI_LOG_ERROR_RECOVERY(3,
988 printk("%s: START_UNIT failed to sdev:"
989 " 0x%p\n", current
->comm
, sdev
));
993 return list_empty(work_q
);
998 * scsi_eh_bus_device_reset - send bdr if needed
999 * @shost: scsi host being recovered.
1000 * @work_q: &list_head for pending commands.
1001 * @done_q: &list_head for processed commands.
1004 * Try a bus device reset. Still, look to see whether we have multiple
1005 * devices that are jammed or not - if we have multiple devices, it
1006 * makes no sense to try bus_device_reset - we really would need to try
1007 * a bus_reset instead.
1009 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1010 struct list_head
*work_q
,
1011 struct list_head
*done_q
)
1013 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1014 struct scsi_device
*sdev
;
1017 shost_for_each_device(sdev
, shost
) {
1019 list_for_each_entry(scmd
, work_q
, eh_entry
)
1020 if (scmd
->device
== sdev
) {
1028 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1029 " 0x%p\n", current
->comm
,
1031 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1032 if (rtn
== SUCCESS
) {
1033 if (!scsi_device_online(sdev
) ||
1034 !scsi_eh_tur(bdr_scmd
)) {
1035 list_for_each_entry_safe(scmd
, next
,
1037 if (scmd
->device
== sdev
)
1038 scsi_eh_finish_cmd(scmd
,
1043 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1051 return list_empty(work_q
);
1055 * scsi_eh_target_reset - send target reset if needed
1056 * @shost: scsi host being recovered.
1057 * @work_q: &list_head for pending commands.
1058 * @done_q: &list_head for processed commands.
1061 * Try a target reset.
1063 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1064 struct list_head
*work_q
,
1065 struct list_head
*done_q
)
1067 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1068 unsigned int id
= 0;
1073 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1074 if (id
== scmd_id(scmd
)) {
1080 /* not one exactly equal; find the next highest */
1081 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1082 if (scmd_id(scmd
) > id
&&
1084 scmd_id(tgtr_scmd
) > scmd_id(scmd
)))
1089 /* no more commands, that's it */
1092 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1094 current
->comm
, id
));
1095 rtn
= scsi_try_target_reset(tgtr_scmd
);
1096 if (rtn
== SUCCESS
) {
1097 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1098 if (id
== scmd_id(scmd
))
1099 if (!scsi_device_online(scmd
->device
) ||
1100 !scsi_eh_tur(tgtr_scmd
))
1101 scsi_eh_finish_cmd(scmd
,
1105 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1108 current
->comm
, id
));
1112 return list_empty(work_q
);
1116 * scsi_eh_bus_reset - send a bus reset
1117 * @shost: &scsi host being recovered.
1118 * @work_q: &list_head for pending commands.
1119 * @done_q: &list_head for processed commands.
1121 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1122 struct list_head
*work_q
,
1123 struct list_head
*done_q
)
1125 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1126 unsigned int channel
;
1130 * we really want to loop over the various channels, and do this on
1131 * a channel by channel basis. we should also check to see if any
1132 * of the failed commands are on soft_reset devices, and if so, skip
1136 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1138 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1139 if (channel
== scmd_channel(scmd
)) {
1143 * FIXME add back in some support for
1144 * soft_reset devices.
1151 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1152 " %d\n", current
->comm
,
1154 rtn
= scsi_try_bus_reset(chan_scmd
);
1155 if (rtn
== SUCCESS
) {
1156 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1157 if (channel
== scmd_channel(scmd
))
1158 if (!scsi_device_online(scmd
->device
) ||
1160 scsi_eh_finish_cmd(scmd
,
1164 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1165 " failed chan: %d\n",
1170 return list_empty(work_q
);
1174 * scsi_eh_host_reset - send a host reset
1175 * @work_q: list_head for processed commands.
1176 * @done_q: list_head for processed commands.
1178 static int scsi_eh_host_reset(struct list_head
*work_q
,
1179 struct list_head
*done_q
)
1181 struct scsi_cmnd
*scmd
, *next
;
1184 if (!list_empty(work_q
)) {
1185 scmd
= list_entry(work_q
->next
,
1186 struct scsi_cmnd
, eh_entry
);
1188 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1191 rtn
= scsi_try_host_reset(scmd
);
1192 if (rtn
== SUCCESS
) {
1193 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1194 if (!scsi_device_online(scmd
->device
) ||
1195 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1197 scsi_eh_finish_cmd(scmd
, done_q
);
1200 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1205 return list_empty(work_q
);
1209 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1210 * @work_q: list_head for processed commands.
1211 * @done_q: list_head for processed commands.
1213 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1214 struct list_head
*done_q
)
1216 struct scsi_cmnd
*scmd
, *next
;
1218 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1219 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1220 "not ready after error recovery\n");
1221 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1222 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1224 * FIXME: Handle lost cmds.
1227 scsi_eh_finish_cmd(scmd
, done_q
);
1233 * scsi_noretry_cmd - determinte if command should be failed fast
1234 * @scmd: SCSI cmd to examine.
1236 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1238 switch (host_byte(scmd
->result
)) {
1242 return blk_failfast_transport(scmd
->request
);
1244 return blk_failfast_dev(scmd
->request
);
1246 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1247 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1250 case DID_SOFT_ERROR
:
1251 return blk_failfast_driver(scmd
->request
);
1254 switch (status_byte(scmd
->result
)) {
1255 case CHECK_CONDITION
:
1257 * assume caller has checked sense and determinted
1258 * the check condition was retryable.
1260 return blk_failfast_dev(scmd
->request
);
1267 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1268 * @scmd: SCSI cmd to examine.
1271 * This is *only* called when we are examining the status after sending
1272 * out the actual data command. any commands that are queued for error
1273 * recovery (e.g. test_unit_ready) do *not* come through here.
1275 * When this routine returns failed, it means the error handler thread
1276 * is woken. In cases where the error code indicates an error that
1277 * doesn't require the error handler read (i.e. we don't need to
1278 * abort/reset), this function should return SUCCESS.
1280 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1285 * if the device is offline, then we clearly just pass the result back
1286 * up to the top level.
1288 if (!scsi_device_online(scmd
->device
)) {
1289 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1296 * first check the host byte, to see if there is anything in there
1297 * that would indicate what we need to do.
1299 switch (host_byte(scmd
->result
)) {
1300 case DID_PASSTHROUGH
:
1302 * no matter what, pass this through to the upper layer.
1303 * nuke this special code so that it looks like we are saying
1306 scmd
->result
&= 0xff00ffff;
1310 * looks good. drop through, and check the next byte.
1313 case DID_NO_CONNECT
:
1314 case DID_BAD_TARGET
:
1317 * note - this means that we just report the status back
1318 * to the top level driver, not that we actually think
1319 * that it indicates SUCCESS.
1323 * when the low level driver returns did_soft_error,
1324 * it is responsible for keeping an internal retry counter
1325 * in order to avoid endless loops (db)
1327 * actually this is a bug in this function here. we should
1328 * be mindful of the maximum number of retries specified
1329 * and not get stuck in a loop.
1331 case DID_SOFT_ERROR
:
1337 return ADD_TO_MLQUEUE
;
1338 case DID_TRANSPORT_DISRUPTED
:
1340 * LLD/transport was disrupted during processing of the IO.
1341 * The transport class is now blocked/blocking,
1342 * and the transport will decide what to do with the IO
1343 * based on its timers and recovery capablilities if
1344 * there are enough retries.
1347 case DID_TRANSPORT_FAILFAST
:
1349 * The transport decided to failfast the IO (most likely
1350 * the fast io fail tmo fired), so send IO directly upwards.
1354 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1355 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1357 * execute reservation conflict processing code
1368 * when we scan the bus, we get timeout messages for
1369 * these commands if there is no device available.
1370 * other hosts report did_no_connect for the same thing.
1372 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1373 scmd
->cmnd
[0] == INQUIRY
)) {
1385 * next, check the message byte.
1387 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1391 * check the status byte to see if this indicates anything special.
1393 switch (status_byte(scmd
->result
)) {
1396 * the case of trying to send too many commands to a
1397 * tagged queueing device.
1401 * device can't talk to us at the moment. Should only
1402 * occur (SAM-3) when the task queue is empty, so will cause
1403 * the empty queue handling to trigger a stall in the
1406 return ADD_TO_MLQUEUE
;
1408 case COMMAND_TERMINATED
:
1411 case CHECK_CONDITION
:
1412 rtn
= scsi_check_sense(scmd
);
1413 if (rtn
== NEEDS_RETRY
)
1415 /* if rtn == FAILED, we have no sense information;
1416 * returning FAILED will wake the error handler thread
1417 * to collect the sense and redo the decide
1420 case CONDITION_GOOD
:
1421 case INTERMEDIATE_GOOD
:
1422 case INTERMEDIATE_C_GOOD
:
1425 * who knows? FIXME(eric)
1429 case RESERVATION_CONFLICT
:
1430 sdev_printk(KERN_INFO
, scmd
->device
,
1431 "reservation conflict\n");
1432 return SUCCESS
; /* causes immediate i/o error */
1440 /* we requeue for retry because the error was retryable, and
1441 * the request was not marked fast fail. Note that above,
1442 * even if the request is marked fast fail, we still requeue
1443 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1444 if ((++scmd
->retries
) <= scmd
->allowed
1445 && !scsi_noretry_cmd(scmd
)) {
1449 * no more retries - report this one back to upper level.
1456 * scsi_eh_lock_door - Prevent medium removal for the specified device
1457 * @sdev: SCSI device to prevent medium removal
1460 * We must be called from process context; scsi_allocate_request()
1464 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1465 * head of the devices request queue, and continue.
1468 * scsi_allocate_request() may sleep waiting for existing requests to
1469 * be processed. However, since we haven't kicked off any request
1470 * processing for this host, this may deadlock.
1472 * If scsi_allocate_request() fails for what ever reason, we
1473 * completely forget to lock the door.
1475 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1477 unsigned char cmnd
[MAX_COMMAND_SIZE
];
1479 cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
1483 cmnd
[4] = SCSI_REMOVAL_PREVENT
;
1486 scsi_execute_async(sdev
, cmnd
, 6, DMA_NONE
, NULL
, 0, 0, 10 * HZ
,
1487 5, NULL
, NULL
, GFP_KERNEL
);
1492 * scsi_restart_operations - restart io operations to the specified host.
1493 * @shost: Host we are restarting.
1496 * When we entered the error handler, we blocked all further i/o to
1497 * this device. we need to 'reverse' this process.
1499 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1501 struct scsi_device
*sdev
;
1502 unsigned long flags
;
1505 * If the door was locked, we need to insert a door lock request
1506 * onto the head of the SCSI request queue for the device. There
1507 * is no point trying to lock the door of an off-line device.
1509 shost_for_each_device(sdev
, shost
) {
1510 if (scsi_device_online(sdev
) && sdev
->locked
)
1511 scsi_eh_lock_door(sdev
);
1515 * next free up anything directly waiting upon the host. this
1516 * will be requests for character device operations, and also for
1517 * ioctls to queued block devices.
1519 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1522 spin_lock_irqsave(shost
->host_lock
, flags
);
1523 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1524 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1525 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1526 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1528 wake_up(&shost
->host_wait
);
1531 * finally we need to re-initiate requests that may be pending. we will
1532 * have had everything blocked while error handling is taking place, and
1533 * now that error recovery is done, we will need to ensure that these
1534 * requests are started.
1536 scsi_run_host_queues(shost
);
1540 * scsi_eh_ready_devs - check device ready state and recover if not.
1541 * @shost: host to be recovered.
1542 * @work_q: &list_head for pending commands.
1543 * @done_q: &list_head for processed commands.
1545 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1546 struct list_head
*work_q
,
1547 struct list_head
*done_q
)
1549 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1550 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1551 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1552 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1553 if (!scsi_eh_host_reset(work_q
, done_q
))
1554 scsi_eh_offline_sdevs(work_q
,
1557 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1560 * scsi_eh_flush_done_q - finish processed commands or retry them.
1561 * @done_q: list_head of processed commands.
1563 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1565 struct scsi_cmnd
*scmd
, *next
;
1567 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1568 list_del_init(&scmd
->eh_entry
);
1569 if (scsi_device_online(scmd
->device
) &&
1570 !scsi_noretry_cmd(scmd
) &&
1571 (++scmd
->retries
<= scmd
->allowed
)) {
1572 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1576 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1579 * If just we got sense for the device (called
1580 * scsi_eh_get_sense), scmd->result is already
1581 * set, do not set DRIVER_TIMEOUT.
1584 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1585 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1587 current
->comm
, scmd
));
1588 scsi_finish_command(scmd
);
1592 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1595 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1596 * @shost: Host to unjam.
1599 * When we come in here, we *know* that all commands on the bus have
1600 * either completed, failed or timed out. we also know that no further
1601 * commands are being sent to the host, so things are relatively quiet
1602 * and we have freedom to fiddle with things as we wish.
1604 * This is only the *default* implementation. it is possible for
1605 * individual drivers to supply their own version of this function, and
1606 * if the maintainer wishes to do this, it is strongly suggested that
1607 * this function be taken as a template and modified. this function
1608 * was designed to correctly handle problems for about 95% of the
1609 * different cases out there, and it should always provide at least a
1610 * reasonable amount of error recovery.
1612 * Any command marked 'failed' or 'timeout' must eventually have
1613 * scsi_finish_cmd() called for it. we do all of the retry stuff
1614 * here, so when we restart the host after we return it should have an
1617 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1619 unsigned long flags
;
1620 LIST_HEAD(eh_work_q
);
1621 LIST_HEAD(eh_done_q
);
1623 spin_lock_irqsave(shost
->host_lock
, flags
);
1624 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1625 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1627 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1629 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1630 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1631 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1633 scsi_eh_flush_done_q(&eh_done_q
);
1637 * scsi_error_handler - SCSI error handler thread
1638 * @data: Host for which we are running.
1641 * This is the main error handling loop. This is run as a kernel thread
1642 * for every SCSI host and handles all error handling activity.
1644 int scsi_error_handler(void *data
)
1646 struct Scsi_Host
*shost
= data
;
1649 * We use TASK_INTERRUPTIBLE so that the thread is not
1650 * counted against the load average as a running process.
1651 * We never actually get interrupted because kthread_run
1652 * disables singal delivery for the created thread.
1654 set_current_state(TASK_INTERRUPTIBLE
);
1655 while (!kthread_should_stop()) {
1656 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1657 shost
->host_failed
!= shost
->host_busy
) {
1658 SCSI_LOG_ERROR_RECOVERY(1,
1659 printk("Error handler scsi_eh_%d sleeping\n",
1662 set_current_state(TASK_INTERRUPTIBLE
);
1666 __set_current_state(TASK_RUNNING
);
1667 SCSI_LOG_ERROR_RECOVERY(1,
1668 printk("Error handler scsi_eh_%d waking up\n",
1672 * We have a host that is failing for some reason. Figure out
1673 * what we need to do to get it up and online again (if we can).
1674 * If we fail, we end up taking the thing offline.
1676 if (shost
->transportt
->eh_strategy_handler
)
1677 shost
->transportt
->eh_strategy_handler(shost
);
1679 scsi_unjam_host(shost
);
1682 * Note - if the above fails completely, the action is to take
1683 * individual devices offline and flush the queue of any
1684 * outstanding requests that may have been pending. When we
1685 * restart, we restart any I/O to any other devices on the bus
1686 * which are still online.
1688 scsi_restart_operations(shost
);
1689 set_current_state(TASK_INTERRUPTIBLE
);
1691 __set_current_state(TASK_RUNNING
);
1693 SCSI_LOG_ERROR_RECOVERY(1,
1694 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1695 shost
->ehandler
= NULL
;
1700 * Function: scsi_report_bus_reset()
1702 * Purpose: Utility function used by low-level drivers to report that
1703 * they have observed a bus reset on the bus being handled.
1705 * Arguments: shost - Host in question
1706 * channel - channel on which reset was observed.
1710 * Lock status: Host lock must be held.
1712 * Notes: This only needs to be called if the reset is one which
1713 * originates from an unknown location. Resets originated
1714 * by the mid-level itself don't need to call this, but there
1715 * should be no harm.
1717 * The main purpose of this is to make sure that a CHECK_CONDITION
1718 * is properly treated.
1720 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1722 struct scsi_device
*sdev
;
1724 __shost_for_each_device(sdev
, shost
) {
1725 if (channel
== sdev_channel(sdev
))
1726 __scsi_report_device_reset(sdev
, NULL
);
1729 EXPORT_SYMBOL(scsi_report_bus_reset
);
1732 * Function: scsi_report_device_reset()
1734 * Purpose: Utility function used by low-level drivers to report that
1735 * they have observed a device reset on the device being handled.
1737 * Arguments: shost - Host in question
1738 * channel - channel on which reset was observed
1739 * target - target on which reset was observed
1743 * Lock status: Host lock must be held
1745 * Notes: This only needs to be called if the reset is one which
1746 * originates from an unknown location. Resets originated
1747 * by the mid-level itself don't need to call this, but there
1748 * should be no harm.
1750 * The main purpose of this is to make sure that a CHECK_CONDITION
1751 * is properly treated.
1753 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1755 struct scsi_device
*sdev
;
1757 __shost_for_each_device(sdev
, shost
) {
1758 if (channel
== sdev_channel(sdev
) &&
1759 target
== sdev_id(sdev
))
1760 __scsi_report_device_reset(sdev
, NULL
);
1763 EXPORT_SYMBOL(scsi_report_device_reset
);
1766 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1771 * Function: scsi_reset_provider
1773 * Purpose: Send requested reset to a bus or device at any phase.
1775 * Arguments: device - device to send reset to
1776 * flag - reset type (see scsi.h)
1778 * Returns: SUCCESS/FAILURE.
1780 * Notes: This is used by the SCSI Generic driver to provide
1781 * Bus/Device reset capability.
1784 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1786 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1787 struct Scsi_Host
*shost
= dev
->host
;
1789 unsigned long flags
;
1792 blk_rq_init(NULL
, &req
);
1793 scmd
->request
= &req
;
1795 scmd
->cmnd
= req
.cmd
;
1797 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1798 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1802 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1804 spin_lock_irqsave(shost
->host_lock
, flags
);
1805 shost
->tmf_in_progress
= 1;
1806 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1809 case SCSI_TRY_RESET_DEVICE
:
1810 rtn
= scsi_try_bus_device_reset(scmd
);
1814 case SCSI_TRY_RESET_TARGET
:
1815 rtn
= scsi_try_target_reset(scmd
);
1819 case SCSI_TRY_RESET_BUS
:
1820 rtn
= scsi_try_bus_reset(scmd
);
1824 case SCSI_TRY_RESET_HOST
:
1825 rtn
= scsi_try_host_reset(scmd
);
1831 spin_lock_irqsave(shost
->host_lock
, flags
);
1832 shost
->tmf_in_progress
= 0;
1833 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1836 * be sure to wake up anyone who was sleeping or had their queue
1837 * suspended while we performed the TMF.
1839 SCSI_LOG_ERROR_RECOVERY(3,
1840 printk("%s: waking up host to restart after TMF\n",
1843 wake_up(&shost
->host_wait
);
1845 scsi_run_host_queues(shost
);
1847 scsi_next_command(scmd
);
1850 EXPORT_SYMBOL(scsi_reset_provider
);
1853 * scsi_normalize_sense - normalize main elements from either fixed or
1854 * descriptor sense data format into a common format.
1856 * @sense_buffer: byte array containing sense data returned by device
1857 * @sb_len: number of valid bytes in sense_buffer
1858 * @sshdr: pointer to instance of structure that common
1859 * elements are written to.
1862 * The "main elements" from sense data are: response_code, sense_key,
1863 * asc, ascq and additional_length (only for descriptor format).
1865 * Typically this function can be called after a device has
1866 * responded to a SCSI command with the CHECK_CONDITION status.
1869 * 1 if valid sense data information found, else 0;
1871 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1872 struct scsi_sense_hdr
*sshdr
)
1874 if (!sense_buffer
|| !sb_len
)
1877 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1879 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1881 if (!scsi_sense_valid(sshdr
))
1884 if (sshdr
->response_code
>= 0x72) {
1889 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1891 sshdr
->asc
= sense_buffer
[2];
1893 sshdr
->ascq
= sense_buffer
[3];
1895 sshdr
->additional_length
= sense_buffer
[7];
1901 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1903 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1904 sb_len
: (sense_buffer
[7] + 8);
1906 sshdr
->asc
= sense_buffer
[12];
1908 sshdr
->ascq
= sense_buffer
[13];
1914 EXPORT_SYMBOL(scsi_normalize_sense
);
1916 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1917 struct scsi_sense_hdr
*sshdr
)
1919 return scsi_normalize_sense(cmd
->sense_buffer
,
1920 SCSI_SENSE_BUFFERSIZE
, sshdr
);
1922 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1925 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1926 * @sense_buffer: byte array of descriptor format sense data
1927 * @sb_len: number of valid bytes in sense_buffer
1928 * @desc_type: value of descriptor type to find
1929 * (e.g. 0 -> information)
1932 * only valid when sense data is in descriptor format
1935 * pointer to start of (first) descriptor if found else NULL
1937 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
1940 int add_sen_len
, add_len
, desc_len
, k
;
1943 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
1945 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
1947 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
1948 add_sen_len
: (sb_len
- 8);
1949 descp
= &sense_buffer
[8];
1950 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
1952 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
1953 desc_len
= add_len
+ 2;
1954 if (descp
[0] == desc_type
)
1956 if (add_len
< 0) // short descriptor ??
1961 EXPORT_SYMBOL(scsi_sense_desc_find
);
1964 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1965 * @sense_buffer: byte array of sense data
1966 * @sb_len: number of valid bytes in sense_buffer
1967 * @info_out: pointer to 64 integer where 8 or 4 byte information
1968 * field will be placed if found.
1971 * 1 if information field found, 0 if not found.
1973 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
1982 switch (sense_buffer
[0] & 0x7f) {
1985 if (sense_buffer
[0] & 0x80) {
1986 *info_out
= (sense_buffer
[3] << 24) +
1987 (sense_buffer
[4] << 16) +
1988 (sense_buffer
[5] << 8) + sense_buffer
[6];
1994 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
1996 if (ucp
&& (0xa == ucp
[1])) {
1998 for (j
= 0; j
< 8; ++j
) {
2011 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2014 * scsi_build_sense_buffer - build sense data in a buffer
2015 * @desc: Sense format (non zero == descriptor format,
2016 * 0 == fixed format)
2017 * @buf: Where to build sense data
2019 * @asc: Additional sense code
2020 * @ascq: Additional sense code qualifier
2023 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2026 buf
[0] = 0x72; /* descriptor, current */
2032 buf
[0] = 0x70; /* fixed, current */
2039 EXPORT_SYMBOL(scsi_build_sense_buffer
);