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/slab.h>
22 #include <linux/kernel.h>
23 #include <linux/interrupt.h>
24 #include <linux/blkdev.h>
25 #include <linux/delay.h>
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_eh.h>
31 #include <scsi/scsi_host.h>
32 #include <scsi/scsi_ioctl.h>
33 #include <scsi/scsi_request.h>
35 #include "scsi_priv.h"
36 #include "scsi_logging.h"
38 #define SENSE_TIMEOUT (10*HZ)
39 #define START_UNIT_TIMEOUT (30*HZ)
42 * These should *probably* be handled by the host itself.
43 * Since it is allowed to sleep, it probably should.
45 #define BUS_RESET_SETTLE_TIME (10)
46 #define HOST_RESET_SETTLE_TIME (10)
48 /* called with shost->host_lock held */
49 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
51 if (shost
->host_busy
== shost
->host_failed
) {
53 SCSI_LOG_ERROR_RECOVERY(5,
54 printk("Waking error handler thread\n"));
59 * scsi_eh_scmd_add - add scsi cmd to error handling.
60 * @scmd: scmd to run eh on.
61 * @eh_flag: optional SCSI_EH flag.
66 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
68 struct Scsi_Host
*shost
= scmd
->device
->host
;
71 if (shost
->eh_wait
== NULL
)
74 spin_lock_irqsave(shost
->host_lock
, flags
);
76 scmd
->eh_eflags
|= eh_flag
;
77 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
78 set_bit(SHOST_RECOVERY
, &shost
->shost_state
);
80 scsi_eh_wakeup(shost
);
81 spin_unlock_irqrestore(shost
->host_lock
, flags
);
86 * scsi_add_timer - Start timeout timer for a single scsi command.
87 * @scmd: scsi command that is about to start running.
88 * @timeout: amount of time to allow this command to run.
89 * @complete: timeout function to call if timer isn't canceled.
92 * This should be turned into an inline function. Each scsi command
93 * has its own timer, and as it is added to the queue, we set up the
94 * timer. When the command completes, we cancel the timer.
96 void scsi_add_timer(struct scsi_cmnd
*scmd
, int timeout
,
97 void (*complete
)(struct scsi_cmnd
*))
101 * If the clock was already running for this command, then
102 * first delete the timer. The timer handling code gets rather
103 * confused if we don't do this.
105 if (scmd
->eh_timeout
.function
)
106 del_timer(&scmd
->eh_timeout
);
108 scmd
->eh_timeout
.data
= (unsigned long)scmd
;
109 scmd
->eh_timeout
.expires
= jiffies
+ timeout
;
110 scmd
->eh_timeout
.function
= (void (*)(unsigned long)) complete
;
112 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
113 " %d, (%p)\n", __FUNCTION__
,
114 scmd
, timeout
, complete
));
116 add_timer(&scmd
->eh_timeout
);
118 EXPORT_SYMBOL(scsi_add_timer
);
121 * scsi_delete_timer - Delete/cancel timer for a given function.
122 * @scmd: Cmd that we are canceling timer for
125 * This should be turned into an inline function.
128 * 1 if we were able to detach the timer. 0 if we blew it, and the
129 * timer function has already started to run.
131 int scsi_delete_timer(struct scsi_cmnd
*scmd
)
135 rtn
= del_timer(&scmd
->eh_timeout
);
137 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
138 " rtn: %d\n", __FUNCTION__
,
141 scmd
->eh_timeout
.data
= (unsigned long)NULL
;
142 scmd
->eh_timeout
.function
= NULL
;
146 EXPORT_SYMBOL(scsi_delete_timer
);
149 * scsi_times_out - Timeout function for normal scsi commands.
150 * @scmd: Cmd that is timing out.
153 * We do not need to lock this. There is the potential for a race
154 * only in that the normal completion handling might run, but if the
155 * normal completion function determines that the timer has already
156 * fired, then it mustn't do anything.
158 void scsi_times_out(struct scsi_cmnd
*scmd
)
160 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
162 if (scmd
->device
->host
->hostt
->eh_timed_out
)
163 switch (scmd
->device
->host
->hostt
->eh_timed_out(scmd
)) {
168 /* This allows a single retry even of a command
169 * with allowed == 0 */
170 if (scmd
->retries
++ > scmd
->allowed
)
172 scsi_add_timer(scmd
, scmd
->timeout_per_command
,
179 if (unlikely(!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
180 panic("Error handler thread not present at %p %p %s %d",
181 scmd
, scmd
->device
->host
, __FILE__
, __LINE__
);
186 * scsi_block_when_processing_errors - Prevent cmds from being queued.
187 * @sdev: Device on which we are performing recovery.
190 * We block until the host is out of error recovery, and then check to
191 * see whether the host or the device is offline.
194 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
196 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
200 wait_event(sdev
->host
->host_wait
, (!test_bit(SHOST_RECOVERY
, &sdev
->host
->shost_state
)));
202 online
= scsi_device_online(sdev
);
204 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__
,
209 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
211 #ifdef CONFIG_SCSI_LOGGING
213 * scsi_eh_prt_fail_stats - Log info on failures.
214 * @shost: scsi host being recovered.
215 * @work_q: Queue of scsi cmds to process.
217 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
218 struct list_head
*work_q
)
220 struct scsi_cmnd
*scmd
;
221 struct scsi_device
*sdev
;
222 int total_failures
= 0;
225 int devices_failed
= 0;
227 shost_for_each_device(sdev
, shost
) {
228 list_for_each_entry(scmd
, work_q
, eh_entry
) {
229 if (scmd
->device
== sdev
) {
231 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
238 if (cmd_cancel
|| cmd_failed
) {
239 SCSI_LOG_ERROR_RECOVERY(3,
240 printk("%s: %d:%d:%d:%d cmds failed: %d,"
242 __FUNCTION__
, shost
->host_no
,
243 sdev
->channel
, sdev
->id
, sdev
->lun
,
244 cmd_failed
, cmd_cancel
));
251 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
252 " devices require eh work\n",
253 total_failures
, devices_failed
));
258 * scsi_check_sense - Examine scsi cmd sense
259 * @scmd: Cmd to have sense checked.
262 * SUCCESS or FAILED or NEEDS_RETRY
265 * When a deferred error is detected the current command has
266 * not been executed and needs retrying.
268 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
270 struct scsi_sense_hdr sshdr
;
272 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
273 return FAILED
; /* no valid sense data */
275 if (scsi_sense_is_deferred(&sshdr
))
279 * Previous logic looked for FILEMARK, EOM or ILI which are
280 * mainly associated with tapes and returned SUCCESS.
282 if (sshdr
.response_code
== 0x70) {
284 if (scmd
->sense_buffer
[2] & 0xe0)
288 * descriptor format: look for "stream commands sense data
289 * descriptor" (see SSC-3). Assume single sense data
290 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
292 if ((sshdr
.additional_length
> 3) &&
293 (scmd
->sense_buffer
[8] == 0x4) &&
294 (scmd
->sense_buffer
[11] & 0xe0))
298 switch (sshdr
.sense_key
) {
301 case RECOVERED_ERROR
:
302 return /* soft_error */ SUCCESS
;
304 case ABORTED_COMMAND
:
309 * if we are expecting a cc/ua because of a bus reset that we
310 * performed, treat this just as a retry. otherwise this is
311 * information that we should pass up to the upper-level driver
312 * so that we can deal with it there.
314 if (scmd
->device
->expecting_cc_ua
) {
315 scmd
->device
->expecting_cc_ua
= 0;
319 * if the device is in the process of becoming ready, we
322 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
325 * if the device is not started, we need to wake
326 * the error handler to start the motor
328 if (scmd
->device
->allow_restart
&&
329 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
333 /* these three are not supported */
335 case VOLUME_OVERFLOW
:
343 if (scmd
->device
->retry_hwerror
)
348 case ILLEGAL_REQUEST
:
357 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
358 * @scmd: SCSI cmd to examine.
361 * This is *only* called when we are examining the status of commands
362 * queued during error recovery. the main difference here is that we
363 * don't allow for the possibility of retries here, and we are a lot
364 * more restrictive about what we consider acceptable.
366 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
369 * first check the host byte, to see if there is anything in there
370 * that would indicate what we need to do.
372 if (host_byte(scmd
->result
) == DID_RESET
) {
374 * rats. we are already in the error handler, so we now
375 * get to try and figure out what to do next. if the sense
376 * is valid, we have a pretty good idea of what to do.
377 * if not, we mark it as FAILED.
379 return scsi_check_sense(scmd
);
381 if (host_byte(scmd
->result
) != DID_OK
)
385 * next, check the message byte.
387 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
391 * now, check the status byte to see if this indicates
394 switch (status_byte(scmd
->result
)) {
396 case COMMAND_TERMINATED
:
398 case CHECK_CONDITION
:
399 return scsi_check_sense(scmd
);
401 case INTERMEDIATE_GOOD
:
402 case INTERMEDIATE_C_GOOD
:
404 * who knows? FIXME(eric)
409 case RESERVATION_CONFLICT
:
417 * scsi_eh_times_out - timeout function for error handling.
418 * @scmd: Cmd that is timing out.
421 * During error handling, the kernel thread will be sleeping waiting
422 * for some action to complete on the device. our only job is to
423 * record that it timed out, and to wake up the thread.
425 static void scsi_eh_times_out(struct scsi_cmnd
*scmd
)
427 scmd
->eh_eflags
|= SCSI_EH_REC_TIMEOUT
;
428 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__
,
431 up(scmd
->device
->host
->eh_action
);
435 * scsi_eh_done - Completion function for error handling.
436 * @scmd: Cmd that is done.
438 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
441 * if the timeout handler is already running, then just set the
442 * flag which says we finished late, and return. we have no
443 * way of stopping the timeout handler from running, so we must
444 * always defer to it.
446 if (del_timer(&scmd
->eh_timeout
)) {
447 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
449 SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
450 __FUNCTION__
, scmd
, scmd
->result
));
452 up(scmd
->device
->host
->eh_action
);
457 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery.
458 * @scmd: SCSI Cmd to send.
459 * @timeout: Timeout for cmd.
462 * The initialization of the structures is quite a bit different in
463 * this case, and furthermore, there is a different completion handler
464 * vs scsi_dispatch_cmd.
466 * SUCCESS or FAILED or NEEDS_RETRY
468 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, int timeout
)
470 struct scsi_device
*sdev
= scmd
->device
;
471 struct Scsi_Host
*shost
= sdev
->host
;
472 DECLARE_MUTEX_LOCKED(sem
);
477 * we will use a queued command if possible, otherwise we will
478 * emulate the queuing and calling of completion function ourselves.
480 if (sdev
->scsi_level
<= SCSI_2
)
481 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
482 (sdev
->lun
<< 5 & 0xe0);
484 scsi_add_timer(scmd
, timeout
, scsi_eh_times_out
);
487 * set up the semaphore so we wait for the command to complete.
489 shost
->eh_action
= &sem
;
490 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
492 spin_lock_irqsave(shost
->host_lock
, flags
);
494 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
495 spin_unlock_irqrestore(shost
->host_lock
, flags
);
498 scsi_log_completion(scmd
, SUCCESS
);
500 shost
->eh_action
= NULL
;
503 * see if timeout. if so, tell the host to forget about it.
504 * in other words, we don't want a callback any more.
506 if (scmd
->eh_eflags
& SCSI_EH_REC_TIMEOUT
) {
507 scmd
->eh_eflags
&= ~SCSI_EH_REC_TIMEOUT
;
510 * as far as the low level driver is
511 * concerned, this command is still active, so
512 * we must give the low level driver a chance
515 * FIXME(eric) - we are not tracking whether we could
516 * abort a timed out command or not. not sure how
517 * we should treat them differently anyways.
519 if (shost
->hostt
->eh_abort_handler
)
520 shost
->hostt
->eh_abort_handler(scmd
);
522 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
526 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n",
527 __FUNCTION__
, scmd
, rtn
));
530 * now examine the actual status codes to see whether the command
531 * actually did complete normally.
533 if (rtn
== SUCCESS
) {
534 rtn
= scsi_eh_completed_normally(scmd
);
535 SCSI_LOG_ERROR_RECOVERY(3,
536 printk("%s: scsi_eh_completed_normally %x\n",
553 * scsi_request_sense - Request sense data from a particular target.
554 * @scmd: SCSI cmd for request sense.
557 * Some hosts automatically obtain this information, others require
558 * that we obtain it on our own. This function will *not* return until
559 * the command either times out, or it completes.
561 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
563 static unsigned char generic_sense
[6] =
564 {REQUEST_SENSE
, 0, 0, 0, 252, 0};
565 unsigned char *scsi_result
;
569 memcpy(scmd
->cmnd
, generic_sense
, sizeof(generic_sense
));
571 scsi_result
= kmalloc(252, GFP_ATOMIC
| ((scmd
->device
->host
->hostt
->unchecked_isa_dma
) ? __GFP_DMA
: 0));
574 if (unlikely(!scsi_result
)) {
575 printk(KERN_ERR
"%s: cannot allocate scsi_result.\n",
581 * zero the sense buffer. some host adapters automatically always
582 * request sense, so it is not a good idea that
583 * scmd->request_buffer and scmd->sense_buffer point to the same
584 * address (db). 0 is not a valid sense code.
586 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
587 memset(scsi_result
, 0, 252);
589 saved_result
= scmd
->result
;
590 scmd
->request_buffer
= scsi_result
;
591 scmd
->request_bufflen
= 252;
593 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
594 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
597 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
599 /* last chance to have valid sense data */
600 if(!SCSI_SENSE_VALID(scmd
)) {
601 memcpy(scmd
->sense_buffer
, scmd
->request_buffer
,
602 sizeof(scmd
->sense_buffer
));
608 * when we eventually call scsi_finish, we really wish to complete
609 * the original request, so let's restore the original data. (db)
611 scsi_setup_cmd_retry(scmd
);
612 scmd
->result
= saved_result
;
617 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
618 * @scmd: Original SCSI cmd that eh has finished.
619 * @done_q: Queue for processed commands.
622 * We don't want to use the normal command completion while we are are
623 * still handling errors - it may cause other commands to be queued,
624 * and that would disturb what we are doing. thus we really want to
625 * keep a list of pending commands for final completion, and once we
626 * are ready to leave error handling we handle completion for real.
628 static void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
,
629 struct list_head
*done_q
)
631 scmd
->device
->host
->host_failed
--;
635 * set this back so that the upper level can correctly free up
638 scsi_setup_cmd_retry(scmd
);
639 list_move_tail(&scmd
->eh_entry
, done_q
);
643 * scsi_eh_get_sense - Get device sense data.
644 * @work_q: Queue of commands to process.
645 * @done_q: Queue of proccessed commands..
648 * See if we need to request sense information. if so, then get it
649 * now, so we have a better idea of what to do.
652 * This has the unfortunate side effect that if a shost adapter does
653 * not automatically request sense information, that we end up shutting
654 * it down before we request it.
656 * All drivers should request sense information internally these days,
657 * so for now all I have to say is tough noogies if you end up in here.
659 * XXX: Long term this code should go away, but that needs an audit of
662 static int scsi_eh_get_sense(struct list_head
*work_q
,
663 struct list_head
*done_q
)
665 struct scsi_cmnd
*scmd
, *next
;
668 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
669 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
670 SCSI_SENSE_VALID(scmd
))
673 SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
677 rtn
= scsi_request_sense(scmd
);
681 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
682 " result %x\n", scmd
,
684 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
686 rtn
= scsi_decide_disposition(scmd
);
689 * if the result was normal, then just pass it along to the
693 /* we don't want this command reissued, just
694 * finished with the sense data, so set
695 * retries to the max allowed to ensure it
696 * won't get reissued */
697 scmd
->retries
= scmd
->allowed
;
698 else if (rtn
!= NEEDS_RETRY
)
701 scsi_eh_finish_cmd(scmd
, done_q
);
704 return list_empty(work_q
);
708 * scsi_try_to_abort_cmd - Ask host to abort a running command.
709 * @scmd: SCSI cmd to abort from Lower Level.
712 * This function will not return until the user's completion function
713 * has been called. there is no timeout on this operation. if the
714 * author of the low-level driver wishes this operation to be timed,
715 * they can provide this facility themselves. helper functions in
716 * scsi_error.c can be supplied to make this easier to do.
718 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
720 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
724 * scsi_done was called just after the command timed out and before
725 * we had a chance to process it. (db)
727 if (scmd
->serial_number
== 0)
729 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
733 * scsi_eh_tur - Send TUR to device.
734 * @scmd: Scsi cmd to send TUR
737 * 0 - Device is ready. 1 - Device NOT ready.
739 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
741 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
742 int retry_cnt
= 1, rtn
;
746 memcpy(scmd
->cmnd
, tur_command
, sizeof(tur_command
));
749 * zero the sense buffer. the scsi spec mandates that any
750 * untransferred sense data should be interpreted as being zero.
752 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
754 saved_result
= scmd
->result
;
755 scmd
->request_buffer
= NULL
;
756 scmd
->request_bufflen
= 0;
758 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
760 scmd
->sc_data_direction
= DMA_NONE
;
762 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
765 * when we eventually call scsi_finish, we really wish to complete
766 * the original request, so let's restore the original data. (db)
768 scsi_setup_cmd_retry(scmd
);
769 scmd
->result
= saved_result
;
772 * hey, we are done. let's look to see what happened.
774 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
775 __FUNCTION__
, scmd
, rtn
));
778 else if (rtn
== NEEDS_RETRY
)
785 * scsi_eh_abort_cmds - abort canceled commands.
786 * @shost: scsi host being recovered.
787 * @eh_done_q: list_head for processed commands.
790 * Try and see whether or not it makes sense to try and abort the
791 * running command. this only works out to be the case if we have one
792 * command that has timed out. if the command simply failed, it makes
793 * no sense to try and abort the command, since as far as the shost
794 * adapter is concerned, it isn't running.
796 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
797 struct list_head
*done_q
)
799 struct scsi_cmnd
*scmd
, *next
;
802 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
803 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
805 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
806 "0x%p\n", current
->comm
,
808 rtn
= scsi_try_to_abort_cmd(scmd
);
809 if (rtn
== SUCCESS
) {
810 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
811 if (!scsi_device_online(scmd
->device
) ||
812 !scsi_eh_tur(scmd
)) {
813 scsi_eh_finish_cmd(scmd
, done_q
);
817 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
824 return list_empty(work_q
);
828 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
829 * @scmd: SCSI cmd used to send BDR
832 * There is no timeout for this operation. if this operation is
833 * unreliable for a given host, then the host itself needs to put a
834 * timer on it, and set the host back to a consistent state prior to
837 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
841 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
844 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
845 if (rtn
== SUCCESS
) {
846 scmd
->device
->was_reset
= 1;
847 scmd
->device
->expecting_cc_ua
= 1;
854 * scsi_eh_try_stu - Send START_UNIT to device.
855 * @scmd: Scsi cmd to send START_UNIT
858 * 0 - Device is ready. 1 - Device NOT ready.
860 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
862 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
866 if (!scmd
->device
->allow_restart
)
869 memcpy(scmd
->cmnd
, stu_command
, sizeof(stu_command
));
872 * zero the sense buffer. the scsi spec mandates that any
873 * untransferred sense data should be interpreted as being zero.
875 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
877 saved_result
= scmd
->result
;
878 scmd
->request_buffer
= NULL
;
879 scmd
->request_bufflen
= 0;
881 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
883 scmd
->sc_data_direction
= DMA_NONE
;
885 rtn
= scsi_send_eh_cmnd(scmd
, START_UNIT_TIMEOUT
);
888 * when we eventually call scsi_finish, we really wish to complete
889 * the original request, so let's restore the original data. (db)
891 scsi_setup_cmd_retry(scmd
);
892 scmd
->result
= saved_result
;
895 * hey, we are done. let's look to see what happened.
897 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
898 __FUNCTION__
, scmd
, rtn
));
905 * scsi_eh_stu - send START_UNIT if needed
906 * @shost: scsi host being recovered.
907 * @eh_done_q: list_head for processed commands.
910 * If commands are failing due to not ready, initializing command required,
911 * try revalidating the device, which will end up sending a start unit.
913 static int scsi_eh_stu(struct Scsi_Host
*shost
,
914 struct list_head
*work_q
,
915 struct list_head
*done_q
)
917 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
918 struct scsi_device
*sdev
;
920 shost_for_each_device(sdev
, shost
) {
922 list_for_each_entry(scmd
, work_q
, eh_entry
)
923 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
924 scsi_check_sense(scmd
) == FAILED
) {
932 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
933 " 0x%p\n", current
->comm
, sdev
));
935 if (!scsi_eh_try_stu(stu_scmd
)) {
936 if (!scsi_device_online(sdev
) ||
937 !scsi_eh_tur(stu_scmd
)) {
938 list_for_each_entry_safe(scmd
, next
,
940 if (scmd
->device
== sdev
)
941 scsi_eh_finish_cmd(scmd
, done_q
);
945 SCSI_LOG_ERROR_RECOVERY(3,
946 printk("%s: START_UNIT failed to sdev:"
947 " 0x%p\n", current
->comm
, sdev
));
951 return list_empty(work_q
);
956 * scsi_eh_bus_device_reset - send bdr if needed
957 * @shost: scsi host being recovered.
958 * @eh_done_q: list_head for processed commands.
961 * Try a bus device reset. still, look to see whether we have multiple
962 * devices that are jammed or not - if we have multiple devices, it
963 * makes no sense to try bus_device_reset - we really would need to try
964 * a bus_reset instead.
966 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
967 struct list_head
*work_q
,
968 struct list_head
*done_q
)
970 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
971 struct scsi_device
*sdev
;
974 shost_for_each_device(sdev
, shost
) {
976 list_for_each_entry(scmd
, work_q
, eh_entry
)
977 if (scmd
->device
== sdev
) {
985 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
986 " 0x%p\n", current
->comm
,
988 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
989 if (rtn
== SUCCESS
) {
990 if (!scsi_device_online(sdev
) ||
991 !scsi_eh_tur(bdr_scmd
)) {
992 list_for_each_entry_safe(scmd
, next
,
994 if (scmd
->device
== sdev
)
995 scsi_eh_finish_cmd(scmd
,
1000 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1008 return list_empty(work_q
);
1012 * scsi_try_bus_reset - ask host to perform a bus reset
1013 * @scmd: SCSI cmd to send bus reset.
1015 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
1017 unsigned long flags
;
1020 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1023 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
1026 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
1028 if (rtn
== SUCCESS
) {
1029 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1030 ssleep(BUS_RESET_SETTLE_TIME
);
1031 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1032 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1033 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1040 * scsi_try_host_reset - ask host adapter to reset itself
1041 * @scmd: SCSI cmd to send hsot reset.
1043 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
1045 unsigned long flags
;
1048 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1051 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
1054 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
1056 if (rtn
== SUCCESS
) {
1057 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1058 ssleep(HOST_RESET_SETTLE_TIME
);
1059 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1060 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1061 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1068 * scsi_eh_bus_reset - send a bus reset
1069 * @shost: scsi host being recovered.
1070 * @eh_done_q: list_head for processed commands.
1072 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1073 struct list_head
*work_q
,
1074 struct list_head
*done_q
)
1076 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1077 unsigned int channel
;
1081 * we really want to loop over the various channels, and do this on
1082 * a channel by channel basis. we should also check to see if any
1083 * of the failed commands are on soft_reset devices, and if so, skip
1087 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1089 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1090 if (channel
== scmd
->device
->channel
) {
1094 * FIXME add back in some support for
1095 * soft_reset devices.
1102 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1103 " %d\n", current
->comm
,
1105 rtn
= scsi_try_bus_reset(chan_scmd
);
1106 if (rtn
== SUCCESS
) {
1107 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1108 if (channel
== scmd
->device
->channel
)
1109 if (!scsi_device_online(scmd
->device
) ||
1111 scsi_eh_finish_cmd(scmd
,
1115 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1116 " failed chan: %d\n",
1121 return list_empty(work_q
);
1125 * scsi_eh_host_reset - send a host reset
1126 * @work_q: list_head for processed commands.
1127 * @done_q: list_head for processed commands.
1129 static int scsi_eh_host_reset(struct list_head
*work_q
,
1130 struct list_head
*done_q
)
1132 struct scsi_cmnd
*scmd
, *next
;
1135 if (!list_empty(work_q
)) {
1136 scmd
= list_entry(work_q
->next
,
1137 struct scsi_cmnd
, eh_entry
);
1139 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1142 rtn
= scsi_try_host_reset(scmd
);
1143 if (rtn
== SUCCESS
) {
1144 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1145 if (!scsi_device_online(scmd
->device
) ||
1146 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1148 scsi_eh_finish_cmd(scmd
, done_q
);
1151 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1156 return list_empty(work_q
);
1160 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1161 * @work_q: list_head for processed commands.
1162 * @done_q: list_head for processed commands.
1165 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1166 struct list_head
*done_q
)
1168 struct scsi_cmnd
*scmd
, *next
;
1170 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1171 printk(KERN_INFO
"scsi: Device offlined - not"
1172 " ready after error recovery: host"
1173 " %d channel %d id %d lun %d\n",
1174 scmd
->device
->host
->host_no
,
1175 scmd
->device
->channel
,
1178 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1179 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1181 * FIXME: Handle lost cmds.
1184 scsi_eh_finish_cmd(scmd
, done_q
);
1190 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1191 * @scmd: SCSI cmd to examine.
1194 * This is *only* called when we are examining the status after sending
1195 * out the actual data command. any commands that are queued for error
1196 * recovery (e.g. test_unit_ready) do *not* come through here.
1198 * When this routine returns failed, it means the error handler thread
1199 * is woken. In cases where the error code indicates an error that
1200 * doesn't require the error handler read (i.e. we don't need to
1201 * abort/reset), this function should return SUCCESS.
1203 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1208 * if the device is offline, then we clearly just pass the result back
1209 * up to the top level.
1211 if (!scsi_device_online(scmd
->device
)) {
1212 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1219 * first check the host byte, to see if there is anything in there
1220 * that would indicate what we need to do.
1222 switch (host_byte(scmd
->result
)) {
1223 case DID_PASSTHROUGH
:
1225 * no matter what, pass this through to the upper layer.
1226 * nuke this special code so that it looks like we are saying
1229 scmd
->result
&= 0xff00ffff;
1233 * looks good. drop through, and check the next byte.
1236 case DID_NO_CONNECT
:
1237 case DID_BAD_TARGET
:
1240 * note - this means that we just report the status back
1241 * to the top level driver, not that we actually think
1242 * that it indicates SUCCESS.
1246 * when the low level driver returns did_soft_error,
1247 * it is responsible for keeping an internal retry counter
1248 * in order to avoid endless loops (db)
1250 * actually this is a bug in this function here. we should
1251 * be mindful of the maximum number of retries specified
1252 * and not get stuck in a loop.
1254 case DID_SOFT_ERROR
:
1260 return ADD_TO_MLQUEUE
;
1263 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1264 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1266 * execute reservation conflict processing code
1277 * when we scan the bus, we get timeout messages for
1278 * these commands if there is no device available.
1279 * other hosts report did_no_connect for the same thing.
1281 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1282 scmd
->cmnd
[0] == INQUIRY
)) {
1294 * next, check the message byte.
1296 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1300 * check the status byte to see if this indicates anything special.
1302 switch (status_byte(scmd
->result
)) {
1305 * the case of trying to send too many commands to a
1306 * tagged queueing device.
1310 * device can't talk to us at the moment. Should only
1311 * occur (SAM-3) when the task queue is empty, so will cause
1312 * the empty queue handling to trigger a stall in the
1315 return ADD_TO_MLQUEUE
;
1317 case COMMAND_TERMINATED
:
1320 case CHECK_CONDITION
:
1321 rtn
= scsi_check_sense(scmd
);
1322 if (rtn
== NEEDS_RETRY
)
1324 /* if rtn == FAILED, we have no sense information;
1325 * returning FAILED will wake the error handler thread
1326 * to collect the sense and redo the decide
1329 case CONDITION_GOOD
:
1330 case INTERMEDIATE_GOOD
:
1331 case INTERMEDIATE_C_GOOD
:
1334 * who knows? FIXME(eric)
1338 case RESERVATION_CONFLICT
:
1339 printk(KERN_INFO
"scsi: reservation conflict: host"
1340 " %d channel %d id %d lun %d\n",
1341 scmd
->device
->host
->host_no
, scmd
->device
->channel
,
1342 scmd
->device
->id
, scmd
->device
->lun
);
1343 return SUCCESS
; /* causes immediate i/o error */
1351 /* we requeue for retry because the error was retryable, and
1352 * the request was not marked fast fail. Note that above,
1353 * even if the request is marked fast fail, we still requeue
1354 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1355 if ((++scmd
->retries
) < scmd
->allowed
1356 && !blk_noretry_request(scmd
->request
)) {
1360 * no more retries - report this one back to upper level.
1367 * scsi_eh_lock_done - done function for eh door lock request
1368 * @scmd: SCSI command block for the door lock request
1371 * We completed the asynchronous door lock request, and it has either
1372 * locked the door or failed. We must free the command structures
1373 * associated with this request.
1375 static void scsi_eh_lock_done(struct scsi_cmnd
*scmd
)
1377 struct scsi_request
*sreq
= scmd
->sc_request
;
1379 scsi_release_request(sreq
);
1384 * scsi_eh_lock_door - Prevent medium removal for the specified device
1385 * @sdev: SCSI device to prevent medium removal
1388 * We must be called from process context; scsi_allocate_request()
1392 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1393 * head of the devices request queue, and continue.
1396 * scsi_allocate_request() may sleep waiting for existing requests to
1397 * be processed. However, since we haven't kicked off any request
1398 * processing for this host, this may deadlock.
1400 * If scsi_allocate_request() fails for what ever reason, we
1401 * completely forget to lock the door.
1403 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1405 struct scsi_request
*sreq
= scsi_allocate_request(sdev
, GFP_KERNEL
);
1407 if (unlikely(!sreq
)) {
1408 printk(KERN_ERR
"%s: request allocate failed,"
1409 "prevent media removal cmd not sent\n", __FUNCTION__
);
1413 sreq
->sr_cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
1414 sreq
->sr_cmnd
[1] = 0;
1415 sreq
->sr_cmnd
[2] = 0;
1416 sreq
->sr_cmnd
[3] = 0;
1417 sreq
->sr_cmnd
[4] = SCSI_REMOVAL_PREVENT
;
1418 sreq
->sr_cmnd
[5] = 0;
1419 sreq
->sr_data_direction
= DMA_NONE
;
1420 sreq
->sr_bufflen
= 0;
1421 sreq
->sr_buffer
= NULL
;
1422 sreq
->sr_allowed
= 5;
1423 sreq
->sr_done
= scsi_eh_lock_done
;
1424 sreq
->sr_timeout_per_command
= 10 * HZ
;
1425 sreq
->sr_cmd_len
= COMMAND_SIZE(sreq
->sr_cmnd
[0]);
1427 scsi_insert_special_req(sreq
, 1);
1432 * scsi_restart_operations - restart io operations to the specified host.
1433 * @shost: Host we are restarting.
1436 * When we entered the error handler, we blocked all further i/o to
1437 * this device. we need to 'reverse' this process.
1439 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1441 struct scsi_device
*sdev
;
1444 * If the door was locked, we need to insert a door lock request
1445 * onto the head of the SCSI request queue for the device. There
1446 * is no point trying to lock the door of an off-line device.
1448 shost_for_each_device(sdev
, shost
) {
1449 if (scsi_device_online(sdev
) && sdev
->locked
)
1450 scsi_eh_lock_door(sdev
);
1454 * next free up anything directly waiting upon the host. this
1455 * will be requests for character device operations, and also for
1456 * ioctls to queued block devices.
1458 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1461 clear_bit(SHOST_RECOVERY
, &shost
->shost_state
);
1463 wake_up(&shost
->host_wait
);
1466 * finally we need to re-initiate requests that may be pending. we will
1467 * have had everything blocked while error handling is taking place, and
1468 * now that error recovery is done, we will need to ensure that these
1469 * requests are started.
1471 scsi_run_host_queues(shost
);
1475 * scsi_eh_ready_devs - check device ready state and recover if not.
1476 * @shost: host to be recovered.
1477 * @eh_done_q: list_head for processed commands.
1480 static void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1481 struct list_head
*work_q
,
1482 struct list_head
*done_q
)
1484 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1485 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1486 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1487 if (!scsi_eh_host_reset(work_q
, done_q
))
1488 scsi_eh_offline_sdevs(work_q
, done_q
);
1492 * scsi_eh_flush_done_q - finish processed commands or retry them.
1493 * @done_q: list_head of processed commands.
1496 static void scsi_eh_flush_done_q(struct list_head
*done_q
)
1498 struct scsi_cmnd
*scmd
, *next
;
1500 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1501 list_del_init(&scmd
->eh_entry
);
1502 if (scsi_device_online(scmd
->device
) &&
1503 !blk_noretry_request(scmd
->request
) &&
1504 (++scmd
->retries
< scmd
->allowed
)) {
1505 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1509 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1512 * If just we got sense for the device (called
1513 * scsi_eh_get_sense), scmd->result is already
1514 * set, do not set DRIVER_TIMEOUT.
1517 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1518 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1520 current
->comm
, scmd
));
1521 scsi_finish_command(scmd
);
1527 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1528 * @shost: Host to unjam.
1531 * When we come in here, we *know* that all commands on the bus have
1532 * either completed, failed or timed out. we also know that no further
1533 * commands are being sent to the host, so things are relatively quiet
1534 * and we have freedom to fiddle with things as we wish.
1536 * This is only the *default* implementation. it is possible for
1537 * individual drivers to supply their own version of this function, and
1538 * if the maintainer wishes to do this, it is strongly suggested that
1539 * this function be taken as a template and modified. this function
1540 * was designed to correctly handle problems for about 95% of the
1541 * different cases out there, and it should always provide at least a
1542 * reasonable amount of error recovery.
1544 * Any command marked 'failed' or 'timeout' must eventually have
1545 * scsi_finish_cmd() called for it. we do all of the retry stuff
1546 * here, so when we restart the host after we return it should have an
1549 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1551 unsigned long flags
;
1552 LIST_HEAD(eh_work_q
);
1553 LIST_HEAD(eh_done_q
);
1555 spin_lock_irqsave(shost
->host_lock
, flags
);
1556 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1557 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1559 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1561 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1562 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1563 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1565 scsi_eh_flush_done_q(&eh_done_q
);
1569 * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
1570 * @data: Host for which we are running.
1573 * This is always run in the context of a kernel thread. The idea is
1574 * that we start this thing up when the kernel starts up (one per host
1575 * that we detect), and it immediately goes to sleep and waits for some
1576 * event (i.e. failure). When this takes place, we have the job of
1577 * trying to unjam the bus and restarting things.
1579 int scsi_error_handler(void *data
)
1581 struct Scsi_Host
*shost
= (struct Scsi_Host
*) data
;
1583 DECLARE_MUTEX_LOCKED(sem
);
1589 daemonize("scsi_eh_%d", shost
->host_no
);
1591 current
->flags
|= PF_NOFREEZE
;
1593 shost
->eh_wait
= &sem
;
1594 shost
->ehandler
= current
;
1597 * Wake up the thread that created us.
1599 SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
1600 " scsi_eh_%d\n",shost
->host_no
));
1602 complete(shost
->eh_notify
);
1606 * If we get a signal, it means we are supposed to go
1607 * away and die. This typically happens if the user is
1608 * trying to unload a module.
1610 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1612 " sleeping\n",shost
->host_no
));
1615 * Note - we always use down_interruptible with the semaphore
1616 * even if the module was loaded as part of the kernel. The
1617 * reason is that down() will cause this thread to be counted
1618 * in the load average as a running process, and down
1619 * interruptible doesn't. Given that we need to allow this
1620 * thread to die if the driver was loaded as a module, using
1621 * semaphores isn't unreasonable.
1623 down_interruptible(&sem
);
1627 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1628 " scsi_eh_%d waking"
1629 " up\n",shost
->host_no
));
1631 shost
->eh_active
= 1;
1634 * We have a host that is failing for some reason. Figure out
1635 * what we need to do to get it up and online again (if we can).
1636 * If we fail, we end up taking the thing offline.
1638 if (shost
->hostt
->eh_strategy_handler
)
1639 rtn
= shost
->hostt
->eh_strategy_handler(shost
);
1641 scsi_unjam_host(shost
);
1643 shost
->eh_active
= 0;
1646 * Note - if the above fails completely, the action is to take
1647 * individual devices offline and flush the queue of any
1648 * outstanding requests that may have been pending. When we
1649 * restart, we restart any I/O to any other devices on the bus
1650 * which are still online.
1652 scsi_restart_operations(shost
);
1656 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
1657 " exiting\n",shost
->host_no
));
1660 * Make sure that nobody tries to wake us up again.
1662 shost
->eh_wait
= NULL
;
1665 * Knock this down too. From this point on, the host is flying
1666 * without a pilot. If this is because the module is being unloaded,
1667 * that's fine. If the user sent a signal to this thing, we are
1668 * potentially in real danger.
1670 shost
->eh_active
= 0;
1671 shost
->ehandler
= NULL
;
1674 * If anyone is waiting for us to exit (i.e. someone trying to unload
1675 * a driver), then wake up that process to let them know we are on
1676 * the way out the door.
1678 complete_and_exit(shost
->eh_notify
, 0);
1683 * Function: scsi_report_bus_reset()
1685 * Purpose: Utility function used by low-level drivers to report that
1686 * they have observed a bus reset on the bus being handled.
1688 * Arguments: shost - Host in question
1689 * channel - channel on which reset was observed.
1693 * Lock status: Host lock must be held.
1695 * Notes: This only needs to be called if the reset is one which
1696 * originates from an unknown location. Resets originated
1697 * by the mid-level itself don't need to call this, but there
1698 * should be no harm.
1700 * The main purpose of this is to make sure that a CHECK_CONDITION
1701 * is properly treated.
1703 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1705 struct scsi_device
*sdev
;
1707 __shost_for_each_device(sdev
, shost
) {
1708 if (channel
== sdev
->channel
) {
1709 sdev
->was_reset
= 1;
1710 sdev
->expecting_cc_ua
= 1;
1714 EXPORT_SYMBOL(scsi_report_bus_reset
);
1717 * Function: scsi_report_device_reset()
1719 * Purpose: Utility function used by low-level drivers to report that
1720 * they have observed a device reset on the device being handled.
1722 * Arguments: shost - Host in question
1723 * channel - channel on which reset was observed
1724 * target - target on which reset was observed
1728 * Lock status: Host lock must be held
1730 * Notes: This only needs to be called if the reset is one which
1731 * originates from an unknown location. Resets originated
1732 * by the mid-level itself don't need to call this, but there
1733 * should be no harm.
1735 * The main purpose of this is to make sure that a CHECK_CONDITION
1736 * is properly treated.
1738 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1740 struct scsi_device
*sdev
;
1742 __shost_for_each_device(sdev
, shost
) {
1743 if (channel
== sdev
->channel
&&
1744 target
== sdev
->id
) {
1745 sdev
->was_reset
= 1;
1746 sdev
->expecting_cc_ua
= 1;
1750 EXPORT_SYMBOL(scsi_report_device_reset
);
1753 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1758 * Function: scsi_reset_provider
1760 * Purpose: Send requested reset to a bus or device at any phase.
1762 * Arguments: device - device to send reset to
1763 * flag - reset type (see scsi.h)
1765 * Returns: SUCCESS/FAILURE.
1767 * Notes: This is used by the SCSI Generic driver to provide
1768 * Bus/Device reset capability.
1771 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1773 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1777 scmd
->request
= &req
;
1778 memset(&scmd
->eh_timeout
, 0, sizeof(scmd
->eh_timeout
));
1779 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
1781 memset(&scmd
->cmnd
, '\0', sizeof(scmd
->cmnd
));
1783 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1785 scmd
->buffer
= NULL
;
1787 scmd
->request_buffer
= NULL
;
1788 scmd
->request_bufflen
= 0;
1792 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1793 scmd
->sc_request
= NULL
;
1794 scmd
->sc_magic
= SCSI_CMND_MAGIC
;
1796 init_timer(&scmd
->eh_timeout
);
1799 * Sometimes the command can get back into the timer chain,
1800 * so use the pid as an identifier.
1805 case SCSI_TRY_RESET_DEVICE
:
1806 rtn
= scsi_try_bus_device_reset(scmd
);
1810 case SCSI_TRY_RESET_BUS
:
1811 rtn
= scsi_try_bus_reset(scmd
);
1815 case SCSI_TRY_RESET_HOST
:
1816 rtn
= scsi_try_host_reset(scmd
);
1822 scsi_next_command(scmd
);
1825 EXPORT_SYMBOL(scsi_reset_provider
);
1828 * scsi_normalize_sense - normalize main elements from either fixed or
1829 * descriptor sense data format into a common format.
1831 * @sense_buffer: byte array containing sense data returned by device
1832 * @sb_len: number of valid bytes in sense_buffer
1833 * @sshdr: pointer to instance of structure that common
1834 * elements are written to.
1837 * The "main elements" from sense data are: response_code, sense_key,
1838 * asc, ascq and additional_length (only for descriptor format).
1840 * Typically this function can be called after a device has
1841 * responded to a SCSI command with the CHECK_CONDITION status.
1844 * 1 if valid sense data information found, else 0;
1846 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1847 struct scsi_sense_hdr
*sshdr
)
1849 if (!sense_buffer
|| !sb_len
|| (sense_buffer
[0] & 0x70) != 0x70)
1852 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1854 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1855 if (sshdr
->response_code
>= 0x72) {
1860 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1862 sshdr
->asc
= sense_buffer
[2];
1864 sshdr
->ascq
= sense_buffer
[3];
1866 sshdr
->additional_length
= sense_buffer
[7];
1872 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1874 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1875 sb_len
: (sense_buffer
[7] + 8);
1877 sshdr
->asc
= sense_buffer
[12];
1879 sshdr
->ascq
= sense_buffer
[13];
1885 EXPORT_SYMBOL(scsi_normalize_sense
);
1887 int scsi_request_normalize_sense(struct scsi_request
*sreq
,
1888 struct scsi_sense_hdr
*sshdr
)
1890 return scsi_normalize_sense(sreq
->sr_sense_buffer
,
1891 sizeof(sreq
->sr_sense_buffer
), sshdr
);
1893 EXPORT_SYMBOL(scsi_request_normalize_sense
);
1895 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1896 struct scsi_sense_hdr
*sshdr
)
1898 return scsi_normalize_sense(cmd
->sense_buffer
,
1899 sizeof(cmd
->sense_buffer
), sshdr
);
1901 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1904 * scsi_sense_desc_find - search for a given descriptor type in
1905 * descriptor sense data format.
1907 * @sense_buffer: byte array of descriptor format sense data
1908 * @sb_len: number of valid bytes in sense_buffer
1909 * @desc_type: value of descriptor type to find
1910 * (e.g. 0 -> information)
1913 * only valid when sense data is in descriptor format
1916 * pointer to start of (first) descriptor if found else NULL
1918 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
1921 int add_sen_len
, add_len
, desc_len
, k
;
1924 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
1926 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
1928 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
1929 add_sen_len
: (sb_len
- 8);
1930 descp
= &sense_buffer
[8];
1931 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
1933 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
1934 desc_len
= add_len
+ 2;
1935 if (descp
[0] == desc_type
)
1937 if (add_len
< 0) // short descriptor ??
1942 EXPORT_SYMBOL(scsi_sense_desc_find
);
1945 * scsi_get_sense_info_fld - attempts to get information field from
1946 * sense data (either fixed or descriptor format)
1948 * @sense_buffer: byte array of sense data
1949 * @sb_len: number of valid bytes in sense_buffer
1950 * @info_out: pointer to 64 integer where 8 or 4 byte information
1951 * field will be placed if found.
1954 * 1 if information field found, 0 if not found.
1956 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
1965 switch (sense_buffer
[0] & 0x7f) {
1968 if (sense_buffer
[0] & 0x80) {
1969 *info_out
= (sense_buffer
[3] << 24) +
1970 (sense_buffer
[4] << 16) +
1971 (sense_buffer
[5] << 8) + sense_buffer
[6];
1977 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
1979 if (ucp
&& (0xa == ucp
[1])) {
1981 for (j
= 0; j
< 8; ++j
) {
1994 EXPORT_SYMBOL(scsi_get_sense_info_fld
);