[SCSI] remove scsi_eh_eflags_ macros
[linux-2.6/suspend2-2.6.18.git] / drivers / scsi / scsi_error.c
blobe289b323853953fffc2ee038c1e04eab0f936db4
1 /*
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
13 * minor cleanups.
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) {
52 up(shost->eh_wait);
53 SCSI_LOG_ERROR_RECOVERY(5,
54 printk("Waking error handler thread\n"));
58 /**
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.
63 * Return value:
64 * 0 on failure.
65 **/
66 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
68 struct Scsi_Host *shost = scmd->device->host;
69 unsigned long flags;
71 if (shost->eh_wait == NULL)
72 return 0;
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);
79 shost->host_failed++;
80 scsi_eh_wakeup(shost);
81 spin_unlock_irqrestore(shost->host_lock, flags);
82 return 1;
85 /**
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.
91 * Notes:
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.
95 **/
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
124 * Notes:
125 * This should be turned into an inline function.
127 * Return value:
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)
133 int rtn;
135 rtn = del_timer(&scmd->eh_timeout);
137 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
138 " rtn: %d\n", __FUNCTION__,
139 scmd, rtn));
141 scmd->eh_timeout.data = (unsigned long)NULL;
142 scmd->eh_timeout.function = NULL;
144 return rtn;
146 EXPORT_SYMBOL(scsi_delete_timer);
149 * scsi_times_out - Timeout function for normal scsi commands.
150 * @scmd: Cmd that is timing out.
152 * Notes:
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)) {
164 case EH_HANDLED:
165 __scsi_done(scmd);
166 return;
167 case EH_RESET_TIMER:
168 /* This allows a single retry even of a command
169 * with allowed == 0 */
170 if (scmd->retries++ > scmd->allowed)
171 break;
172 scsi_add_timer(scmd, scmd->timeout_per_command,
173 scsi_times_out);
174 return;
175 case EH_NOT_HANDLED:
176 break;
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.
189 * Description:
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.
193 * Return value:
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)
198 int online;
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__,
205 online));
207 return online;
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;
223 int cmd_failed = 0;
224 int cmd_cancel = 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) {
230 ++total_failures;
231 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
232 ++cmd_cancel;
233 else
234 ++cmd_failed;
238 if (cmd_cancel || cmd_failed) {
239 SCSI_LOG_ERROR_RECOVERY(3,
240 printk("%s: %d:%d:%d:%d cmds failed: %d,"
241 " cancel: %d\n",
242 __FUNCTION__, shost->host_no,
243 sdev->channel, sdev->id, sdev->lun,
244 cmd_failed, cmd_cancel));
245 cmd_cancel = 0;
246 cmd_failed = 0;
247 ++devices_failed;
251 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
252 " devices require eh work\n",
253 total_failures, devices_failed));
255 #endif
258 * scsi_check_sense - Examine scsi cmd sense
259 * @scmd: Cmd to have sense checked.
261 * Return value:
262 * SUCCESS or FAILED or NEEDS_RETRY
264 * Notes:
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))
276 return NEEDS_RETRY;
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) {
283 /* fixed format */
284 if (scmd->sense_buffer[2] & 0xe0)
285 return SUCCESS;
286 } else {
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))
295 return SUCCESS;
298 switch (sshdr.sense_key) {
299 case NO_SENSE:
300 return SUCCESS;
301 case RECOVERED_ERROR:
302 return /* soft_error */ SUCCESS;
304 case ABORTED_COMMAND:
305 return NEEDS_RETRY;
306 case NOT_READY:
307 case UNIT_ATTENTION:
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;
316 return NEEDS_RETRY;
319 * if the device is in the process of becoming ready, we
320 * should retry.
322 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
323 return NEEDS_RETRY;
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))
330 return FAILED;
331 return SUCCESS;
333 /* these three are not supported */
334 case COPY_ABORTED:
335 case VOLUME_OVERFLOW:
336 case MISCOMPARE:
337 return SUCCESS;
339 case MEDIUM_ERROR:
340 return NEEDS_RETRY;
342 case HARDWARE_ERROR:
343 if (scmd->device->retry_hwerror)
344 return NEEDS_RETRY;
345 else
346 return SUCCESS;
348 case ILLEGAL_REQUEST:
349 case BLANK_CHECK:
350 case DATA_PROTECT:
351 default:
352 return SUCCESS;
357 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
358 * @scmd: SCSI cmd to examine.
360 * Notes:
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)
382 return FAILED;
385 * next, check the message byte.
387 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
388 return FAILED;
391 * now, check the status byte to see if this indicates
392 * anything special.
394 switch (status_byte(scmd->result)) {
395 case GOOD:
396 case COMMAND_TERMINATED:
397 return SUCCESS;
398 case CHECK_CONDITION:
399 return scsi_check_sense(scmd);
400 case CONDITION_GOOD:
401 case INTERMEDIATE_GOOD:
402 case INTERMEDIATE_C_GOOD:
404 * who knows? FIXME(eric)
406 return SUCCESS;
407 case BUSY:
408 case QUEUE_FULL:
409 case RESERVATION_CONFLICT:
410 default:
411 return FAILED;
413 return FAILED;
417 * scsi_eh_times_out - timeout function for error handling.
418 * @scmd: Cmd that is timing out.
420 * Notes:
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__,
429 scmd));
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.
461 * Notes:
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.
465 * Return value:
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);
473 unsigned long flags;
474 int rtn = SUCCESS;
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);
493 scsi_log_send(scmd);
494 shost->hostt->queuecommand(scmd, scsi_eh_done);
495 spin_unlock_irqrestore(shost->host_lock, flags);
497 down(&sem);
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
513 * to abort it. (db)
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;
523 rtn = FAILED;
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",
537 __FUNCTION__, rtn));
538 switch (rtn) {
539 case SUCCESS:
540 case NEEDS_RETRY:
541 case FAILED:
542 break;
543 default:
544 rtn = FAILED;
545 break;
549 return rtn;
553 * scsi_request_sense - Request sense data from a particular target.
554 * @scmd: SCSI cmd for request sense.
556 * Notes:
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;
566 int saved_result;
567 int rtn;
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",
576 __FUNCTION__);
577 return FAILED;
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;
592 scmd->use_sg = 0;
593 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
594 scmd->sc_data_direction = DMA_FROM_DEVICE;
595 scmd->underflow = 0;
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));
605 kfree(scsi_result);
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;
613 return rtn;
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.
621 * Notes:
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--;
632 scmd->eh_eflags = 0;
635 * set this back so that the upper level can correctly free up
636 * things.
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..
647 * Description:
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.
651 * Notes:
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
660 * all LLDDs first.
662 static int scsi_eh_get_sense(struct list_head *work_q,
663 struct list_head *done_q)
665 struct list_head *lh, *lh_sf;
666 struct scsi_cmnd *scmd;
667 int rtn;
669 list_for_each_safe(lh, lh_sf, work_q) {
670 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
671 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
672 SCSI_SENSE_VALID(scmd))
673 continue;
675 SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
676 " for id: %d\n",
677 current->comm,
678 scmd->device->id));
679 rtn = scsi_request_sense(scmd);
680 if (rtn != SUCCESS)
681 continue;
683 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
684 " result %x\n", scmd,
685 scmd->result));
686 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
688 rtn = scsi_decide_disposition(scmd);
691 * if the result was normal, then just pass it along to the
692 * upper level.
694 if (rtn == SUCCESS)
695 /* we don't want this command reissued, just
696 * finished with the sense data, so set
697 * retries to the max allowed to ensure it
698 * won't get reissued */
699 scmd->retries = scmd->allowed;
700 else if (rtn != NEEDS_RETRY)
701 continue;
703 scsi_eh_finish_cmd(scmd, done_q);
706 return list_empty(work_q);
710 * scsi_try_to_abort_cmd - Ask host to abort a running command.
711 * @scmd: SCSI cmd to abort from Lower Level.
713 * Notes:
714 * This function will not return until the user's completion function
715 * has been called. there is no timeout on this operation. if the
716 * author of the low-level driver wishes this operation to be timed,
717 * they can provide this facility themselves. helper functions in
718 * scsi_error.c can be supplied to make this easier to do.
720 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
722 if (!scmd->device->host->hostt->eh_abort_handler)
723 return FAILED;
726 * scsi_done was called just after the command timed out and before
727 * we had a chance to process it. (db)
729 if (scmd->serial_number == 0)
730 return SUCCESS;
731 return scmd->device->host->hostt->eh_abort_handler(scmd);
735 * scsi_eh_tur - Send TUR to device.
736 * @scmd: Scsi cmd to send TUR
738 * Return value:
739 * 0 - Device is ready. 1 - Device NOT ready.
741 static int scsi_eh_tur(struct scsi_cmnd *scmd)
743 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
744 int retry_cnt = 1, rtn;
745 int saved_result;
747 retry_tur:
748 memcpy(scmd->cmnd, tur_command, sizeof(tur_command));
751 * zero the sense buffer. the scsi spec mandates that any
752 * untransferred sense data should be interpreted as being zero.
754 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
756 saved_result = scmd->result;
757 scmd->request_buffer = NULL;
758 scmd->request_bufflen = 0;
759 scmd->use_sg = 0;
760 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
761 scmd->underflow = 0;
762 scmd->sc_data_direction = DMA_NONE;
764 rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT);
767 * when we eventually call scsi_finish, we really wish to complete
768 * the original request, so let's restore the original data. (db)
770 scsi_setup_cmd_retry(scmd);
771 scmd->result = saved_result;
774 * hey, we are done. let's look to see what happened.
776 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
777 __FUNCTION__, scmd, rtn));
778 if (rtn == SUCCESS)
779 return 0;
780 else if (rtn == NEEDS_RETRY)
781 if (retry_cnt--)
782 goto retry_tur;
783 return 1;
787 * scsi_eh_abort_cmds - abort canceled commands.
788 * @shost: scsi host being recovered.
789 * @eh_done_q: list_head for processed commands.
791 * Decription:
792 * Try and see whether or not it makes sense to try and abort the
793 * running command. this only works out to be the case if we have one
794 * command that has timed out. if the command simply failed, it makes
795 * no sense to try and abort the command, since as far as the shost
796 * adapter is concerned, it isn't running.
798 static int scsi_eh_abort_cmds(struct list_head *work_q,
799 struct list_head *done_q)
801 struct list_head *lh, *lh_sf;
802 struct scsi_cmnd *scmd;
803 int rtn;
805 list_for_each_safe(lh, lh_sf, work_q) {
806 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
807 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
808 continue;
809 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
810 "0x%p\n", current->comm,
811 scmd));
812 rtn = scsi_try_to_abort_cmd(scmd);
813 if (rtn == SUCCESS) {
814 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
815 if (!scsi_device_online(scmd->device) ||
816 !scsi_eh_tur(scmd)) {
817 scsi_eh_finish_cmd(scmd, done_q);
820 } else
821 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
822 " cmd failed:"
823 "0x%p\n",
824 current->comm,
825 scmd));
828 return list_empty(work_q);
832 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
833 * @scmd: SCSI cmd used to send BDR
835 * Notes:
836 * There is no timeout for this operation. if this operation is
837 * unreliable for a given host, then the host itself needs to put a
838 * timer on it, and set the host back to a consistent state prior to
839 * returning.
841 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
843 int rtn;
845 if (!scmd->device->host->hostt->eh_device_reset_handler)
846 return FAILED;
848 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
849 if (rtn == SUCCESS) {
850 scmd->device->was_reset = 1;
851 scmd->device->expecting_cc_ua = 1;
854 return rtn;
858 * scsi_eh_try_stu - Send START_UNIT to device.
859 * @scmd: Scsi cmd to send START_UNIT
861 * Return value:
862 * 0 - Device is ready. 1 - Device NOT ready.
864 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
866 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
867 int rtn;
868 int saved_result;
870 if (!scmd->device->allow_restart)
871 return 1;
873 memcpy(scmd->cmnd, stu_command, sizeof(stu_command));
876 * zero the sense buffer. the scsi spec mandates that any
877 * untransferred sense data should be interpreted as being zero.
879 memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
881 saved_result = scmd->result;
882 scmd->request_buffer = NULL;
883 scmd->request_bufflen = 0;
884 scmd->use_sg = 0;
885 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
886 scmd->underflow = 0;
887 scmd->sc_data_direction = DMA_NONE;
889 rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT);
892 * when we eventually call scsi_finish, we really wish to complete
893 * the original request, so let's restore the original data. (db)
895 scsi_setup_cmd_retry(scmd);
896 scmd->result = saved_result;
899 * hey, we are done. let's look to see what happened.
901 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
902 __FUNCTION__, scmd, rtn));
903 if (rtn == SUCCESS)
904 return 0;
905 return 1;
909 * scsi_eh_stu - send START_UNIT if needed
910 * @shost: scsi host being recovered.
911 * @eh_done_q: list_head for processed commands.
913 * Notes:
914 * If commands are failing due to not ready, initializing command required,
915 * try revalidating the device, which will end up sending a start unit.
917 static int scsi_eh_stu(struct Scsi_Host *shost,
918 struct list_head *work_q,
919 struct list_head *done_q)
921 struct list_head *lh, *lh_sf;
922 struct scsi_cmnd *scmd, *stu_scmd;
923 struct scsi_device *sdev;
925 shost_for_each_device(sdev, shost) {
926 stu_scmd = NULL;
927 list_for_each_entry(scmd, work_q, eh_entry)
928 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
929 scsi_check_sense(scmd) == FAILED ) {
930 stu_scmd = scmd;
931 break;
934 if (!stu_scmd)
935 continue;
937 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
938 " 0x%p\n", current->comm, sdev));
940 if (!scsi_eh_try_stu(stu_scmd)) {
941 if (!scsi_device_online(sdev) ||
942 !scsi_eh_tur(stu_scmd)) {
943 list_for_each_safe(lh, lh_sf, work_q) {
944 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
945 if (scmd->device == sdev)
946 scsi_eh_finish_cmd(scmd, done_q);
949 } else {
950 SCSI_LOG_ERROR_RECOVERY(3,
951 printk("%s: START_UNIT failed to sdev:"
952 " 0x%p\n", current->comm, sdev));
956 return list_empty(work_q);
961 * scsi_eh_bus_device_reset - send bdr if needed
962 * @shost: scsi host being recovered.
963 * @eh_done_q: list_head for processed commands.
965 * Notes:
966 * Try a bus device reset. still, look to see whether we have multiple
967 * devices that are jammed or not - if we have multiple devices, it
968 * makes no sense to try bus_device_reset - we really would need to try
969 * a bus_reset instead.
971 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
972 struct list_head *work_q,
973 struct list_head *done_q)
975 struct list_head *lh, *lh_sf;
976 struct scsi_cmnd *scmd, *bdr_scmd;
977 struct scsi_device *sdev;
978 int rtn;
980 shost_for_each_device(sdev, shost) {
981 bdr_scmd = NULL;
982 list_for_each_entry(scmd, work_q, eh_entry)
983 if (scmd->device == sdev) {
984 bdr_scmd = scmd;
985 break;
988 if (!bdr_scmd)
989 continue;
991 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
992 " 0x%p\n", current->comm,
993 sdev));
994 rtn = scsi_try_bus_device_reset(bdr_scmd);
995 if (rtn == SUCCESS) {
996 if (!scsi_device_online(sdev) ||
997 !scsi_eh_tur(bdr_scmd)) {
998 list_for_each_safe(lh, lh_sf,
999 work_q) {
1000 scmd = list_entry(lh, struct
1001 scsi_cmnd,
1002 eh_entry);
1003 if (scmd->device == sdev)
1004 scsi_eh_finish_cmd(scmd,
1005 done_q);
1008 } else {
1009 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1010 " failed sdev:"
1011 "0x%p\n",
1012 current->comm,
1013 sdev));
1017 return list_empty(work_q);
1021 * scsi_try_bus_reset - ask host to perform a bus reset
1022 * @scmd: SCSI cmd to send bus reset.
1024 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
1026 unsigned long flags;
1027 int rtn;
1029 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1030 __FUNCTION__));
1032 if (!scmd->device->host->hostt->eh_bus_reset_handler)
1033 return FAILED;
1035 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
1037 if (rtn == SUCCESS) {
1038 if (!scmd->device->host->hostt->skip_settle_delay)
1039 ssleep(BUS_RESET_SETTLE_TIME);
1040 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1041 scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
1042 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1045 return rtn;
1049 * scsi_try_host_reset - ask host adapter to reset itself
1050 * @scmd: SCSI cmd to send hsot reset.
1052 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
1054 unsigned long flags;
1055 int rtn;
1057 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1058 __FUNCTION__));
1060 if (!scmd->device->host->hostt->eh_host_reset_handler)
1061 return FAILED;
1063 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
1065 if (rtn == SUCCESS) {
1066 if (!scmd->device->host->hostt->skip_settle_delay)
1067 ssleep(HOST_RESET_SETTLE_TIME);
1068 spin_lock_irqsave(scmd->device->host->host_lock, flags);
1069 scsi_report_bus_reset(scmd->device->host, scmd->device->channel);
1070 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
1073 return rtn;
1077 * scsi_eh_bus_reset - send a bus reset
1078 * @shost: scsi host being recovered.
1079 * @eh_done_q: list_head for processed commands.
1081 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1082 struct list_head *work_q,
1083 struct list_head *done_q)
1085 struct list_head *lh, *lh_sf;
1086 struct scsi_cmnd *scmd;
1087 struct scsi_cmnd *chan_scmd;
1088 unsigned int channel;
1089 int rtn;
1092 * we really want to loop over the various channels, and do this on
1093 * a channel by channel basis. we should also check to see if any
1094 * of the failed commands are on soft_reset devices, and if so, skip
1095 * the reset.
1098 for (channel = 0; channel <= shost->max_channel; channel++) {
1099 chan_scmd = NULL;
1100 list_for_each_entry(scmd, work_q, eh_entry) {
1101 if (channel == scmd->device->channel) {
1102 chan_scmd = scmd;
1103 break;
1105 * FIXME add back in some support for
1106 * soft_reset devices.
1111 if (!chan_scmd)
1112 continue;
1113 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1114 " %d\n", current->comm,
1115 channel));
1116 rtn = scsi_try_bus_reset(chan_scmd);
1117 if (rtn == SUCCESS) {
1118 list_for_each_safe(lh, lh_sf, work_q) {
1119 scmd = list_entry(lh, struct scsi_cmnd,
1120 eh_entry);
1121 if (channel == scmd->device->channel)
1122 if (!scsi_device_online(scmd->device) ||
1123 !scsi_eh_tur(scmd))
1124 scsi_eh_finish_cmd(scmd,
1125 done_q);
1127 } else {
1128 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1129 " failed chan: %d\n",
1130 current->comm,
1131 channel));
1134 return list_empty(work_q);
1138 * scsi_eh_host_reset - send a host reset
1139 * @work_q: list_head for processed commands.
1140 * @done_q: list_head for processed commands.
1142 static int scsi_eh_host_reset(struct list_head *work_q,
1143 struct list_head *done_q)
1145 int rtn;
1146 struct list_head *lh, *lh_sf;
1147 struct scsi_cmnd *scmd;
1149 if (!list_empty(work_q)) {
1150 scmd = list_entry(work_q->next,
1151 struct scsi_cmnd, eh_entry);
1153 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1154 , current->comm));
1156 rtn = scsi_try_host_reset(scmd);
1157 if (rtn == SUCCESS) {
1158 list_for_each_safe(lh, lh_sf, work_q) {
1159 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1160 if (!scsi_device_online(scmd->device) ||
1161 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1162 !scsi_eh_tur(scmd))
1163 scsi_eh_finish_cmd(scmd, done_q);
1165 } else {
1166 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1167 " failed\n",
1168 current->comm));
1171 return list_empty(work_q);
1175 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1176 * @work_q: list_head for processed commands.
1177 * @done_q: list_head for processed commands.
1180 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1181 struct list_head *done_q)
1183 struct list_head *lh, *lh_sf;
1184 struct scsi_cmnd *scmd;
1186 list_for_each_safe(lh, lh_sf, work_q) {
1187 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1188 printk(KERN_INFO "scsi: Device offlined - not"
1189 " ready after error recovery: host"
1190 " %d channel %d id %d lun %d\n",
1191 scmd->device->host->host_no,
1192 scmd->device->channel,
1193 scmd->device->id,
1194 scmd->device->lun);
1195 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1196 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1198 * FIXME: Handle lost cmds.
1201 scsi_eh_finish_cmd(scmd, done_q);
1203 return;
1207 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1208 * @scmd: SCSI cmd to examine.
1210 * Notes:
1211 * This is *only* called when we are examining the status after sending
1212 * out the actual data command. any commands that are queued for error
1213 * recovery (e.g. test_unit_ready) do *not* come through here.
1215 * When this routine returns failed, it means the error handler thread
1216 * is woken. In cases where the error code indicates an error that
1217 * doesn't require the error handler read (i.e. we don't need to
1218 * abort/reset), this function should return SUCCESS.
1220 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1222 int rtn;
1225 * if the device is offline, then we clearly just pass the result back
1226 * up to the top level.
1228 if (!scsi_device_online(scmd->device)) {
1229 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1230 " as SUCCESS\n",
1231 __FUNCTION__));
1232 return SUCCESS;
1236 * first check the host byte, to see if there is anything in there
1237 * that would indicate what we need to do.
1239 switch (host_byte(scmd->result)) {
1240 case DID_PASSTHROUGH:
1242 * no matter what, pass this through to the upper layer.
1243 * nuke this special code so that it looks like we are saying
1244 * did_ok.
1246 scmd->result &= 0xff00ffff;
1247 return SUCCESS;
1248 case DID_OK:
1250 * looks good. drop through, and check the next byte.
1252 break;
1253 case DID_NO_CONNECT:
1254 case DID_BAD_TARGET:
1255 case DID_ABORT:
1257 * note - this means that we just report the status back
1258 * to the top level driver, not that we actually think
1259 * that it indicates SUCCESS.
1261 return SUCCESS;
1263 * when the low level driver returns did_soft_error,
1264 * it is responsible for keeping an internal retry counter
1265 * in order to avoid endless loops (db)
1267 * actually this is a bug in this function here. we should
1268 * be mindful of the maximum number of retries specified
1269 * and not get stuck in a loop.
1271 case DID_SOFT_ERROR:
1272 goto maybe_retry;
1273 case DID_IMM_RETRY:
1274 return NEEDS_RETRY;
1276 case DID_REQUEUE:
1277 return ADD_TO_MLQUEUE;
1279 case DID_ERROR:
1280 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1281 status_byte(scmd->result) == RESERVATION_CONFLICT)
1283 * execute reservation conflict processing code
1284 * lower down
1286 break;
1287 /* fallthrough */
1289 case DID_BUS_BUSY:
1290 case DID_PARITY:
1291 goto maybe_retry;
1292 case DID_TIME_OUT:
1294 * when we scan the bus, we get timeout messages for
1295 * these commands if there is no device available.
1296 * other hosts report did_no_connect for the same thing.
1298 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1299 scmd->cmnd[0] == INQUIRY)) {
1300 return SUCCESS;
1301 } else {
1302 return FAILED;
1304 case DID_RESET:
1305 return SUCCESS;
1306 default:
1307 return FAILED;
1311 * next, check the message byte.
1313 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1314 return FAILED;
1317 * check the status byte to see if this indicates anything special.
1319 switch (status_byte(scmd->result)) {
1320 case QUEUE_FULL:
1322 * the case of trying to send too many commands to a
1323 * tagged queueing device.
1325 case BUSY:
1327 * device can't talk to us at the moment. Should only
1328 * occur (SAM-3) when the task queue is empty, so will cause
1329 * the empty queue handling to trigger a stall in the
1330 * device.
1332 return ADD_TO_MLQUEUE;
1333 case GOOD:
1334 case COMMAND_TERMINATED:
1335 case TASK_ABORTED:
1336 return SUCCESS;
1337 case CHECK_CONDITION:
1338 rtn = scsi_check_sense(scmd);
1339 if (rtn == NEEDS_RETRY)
1340 goto maybe_retry;
1341 /* if rtn == FAILED, we have no sense information;
1342 * returning FAILED will wake the error handler thread
1343 * to collect the sense and redo the decide
1344 * disposition */
1345 return rtn;
1346 case CONDITION_GOOD:
1347 case INTERMEDIATE_GOOD:
1348 case INTERMEDIATE_C_GOOD:
1349 case ACA_ACTIVE:
1351 * who knows? FIXME(eric)
1353 return SUCCESS;
1355 case RESERVATION_CONFLICT:
1356 printk(KERN_INFO "scsi: reservation conflict: host"
1357 " %d channel %d id %d lun %d\n",
1358 scmd->device->host->host_no, scmd->device->channel,
1359 scmd->device->id, scmd->device->lun);
1360 return SUCCESS; /* causes immediate i/o error */
1361 default:
1362 return FAILED;
1364 return FAILED;
1366 maybe_retry:
1368 /* we requeue for retry because the error was retryable, and
1369 * the request was not marked fast fail. Note that above,
1370 * even if the request is marked fast fail, we still requeue
1371 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1372 if ((++scmd->retries) < scmd->allowed
1373 && !blk_noretry_request(scmd->request)) {
1374 return NEEDS_RETRY;
1375 } else {
1377 * no more retries - report this one back to upper level.
1379 return SUCCESS;
1384 * scsi_eh_lock_done - done function for eh door lock request
1385 * @scmd: SCSI command block for the door lock request
1387 * Notes:
1388 * We completed the asynchronous door lock request, and it has either
1389 * locked the door or failed. We must free the command structures
1390 * associated with this request.
1392 static void scsi_eh_lock_done(struct scsi_cmnd *scmd)
1394 struct scsi_request *sreq = scmd->sc_request;
1396 scsi_release_request(sreq);
1401 * scsi_eh_lock_door - Prevent medium removal for the specified device
1402 * @sdev: SCSI device to prevent medium removal
1404 * Locking:
1405 * We must be called from process context; scsi_allocate_request()
1406 * may sleep.
1408 * Notes:
1409 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1410 * head of the devices request queue, and continue.
1412 * Bugs:
1413 * scsi_allocate_request() may sleep waiting for existing requests to
1414 * be processed. However, since we haven't kicked off any request
1415 * processing for this host, this may deadlock.
1417 * If scsi_allocate_request() fails for what ever reason, we
1418 * completely forget to lock the door.
1420 static void scsi_eh_lock_door(struct scsi_device *sdev)
1422 struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL);
1424 if (unlikely(!sreq)) {
1425 printk(KERN_ERR "%s: request allocate failed,"
1426 "prevent media removal cmd not sent\n", __FUNCTION__);
1427 return;
1430 sreq->sr_cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1431 sreq->sr_cmnd[1] = 0;
1432 sreq->sr_cmnd[2] = 0;
1433 sreq->sr_cmnd[3] = 0;
1434 sreq->sr_cmnd[4] = SCSI_REMOVAL_PREVENT;
1435 sreq->sr_cmnd[5] = 0;
1436 sreq->sr_data_direction = DMA_NONE;
1437 sreq->sr_bufflen = 0;
1438 sreq->sr_buffer = NULL;
1439 sreq->sr_allowed = 5;
1440 sreq->sr_done = scsi_eh_lock_done;
1441 sreq->sr_timeout_per_command = 10 * HZ;
1442 sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]);
1444 scsi_insert_special_req(sreq, 1);
1449 * scsi_restart_operations - restart io operations to the specified host.
1450 * @shost: Host we are restarting.
1452 * Notes:
1453 * When we entered the error handler, we blocked all further i/o to
1454 * this device. we need to 'reverse' this process.
1456 static void scsi_restart_operations(struct Scsi_Host *shost)
1458 struct scsi_device *sdev;
1461 * If the door was locked, we need to insert a door lock request
1462 * onto the head of the SCSI request queue for the device. There
1463 * is no point trying to lock the door of an off-line device.
1465 shost_for_each_device(sdev, shost) {
1466 if (scsi_device_online(sdev) && sdev->locked)
1467 scsi_eh_lock_door(sdev);
1471 * next free up anything directly waiting upon the host. this
1472 * will be requests for character device operations, and also for
1473 * ioctls to queued block devices.
1475 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1476 __FUNCTION__));
1478 clear_bit(SHOST_RECOVERY, &shost->shost_state);
1480 wake_up(&shost->host_wait);
1483 * finally we need to re-initiate requests that may be pending. we will
1484 * have had everything blocked while error handling is taking place, and
1485 * now that error recovery is done, we will need to ensure that these
1486 * requests are started.
1488 scsi_run_host_queues(shost);
1492 * scsi_eh_ready_devs - check device ready state and recover if not.
1493 * @shost: host to be recovered.
1494 * @eh_done_q: list_head for processed commands.
1497 static void scsi_eh_ready_devs(struct Scsi_Host *shost,
1498 struct list_head *work_q,
1499 struct list_head *done_q)
1501 if (!scsi_eh_stu(shost, work_q, done_q))
1502 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1503 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1504 if (!scsi_eh_host_reset(work_q, done_q))
1505 scsi_eh_offline_sdevs(work_q, done_q);
1509 * scsi_eh_flush_done_q - finish processed commands or retry them.
1510 * @done_q: list_head of processed commands.
1513 static void scsi_eh_flush_done_q(struct list_head *done_q)
1515 struct list_head *lh, *lh_sf;
1516 struct scsi_cmnd *scmd;
1518 list_for_each_safe(lh, lh_sf, done_q) {
1519 scmd = list_entry(lh, struct scsi_cmnd, eh_entry);
1520 list_del_init(lh);
1521 if (scsi_device_online(scmd->device) &&
1522 !blk_noretry_request(scmd->request) &&
1523 (++scmd->retries < scmd->allowed)) {
1524 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1525 " retry cmd: %p\n",
1526 current->comm,
1527 scmd));
1528 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1529 } else {
1531 * If just we got sense for the device (called
1532 * scsi_eh_get_sense), scmd->result is already
1533 * set, do not set DRIVER_TIMEOUT.
1535 if (!scmd->result)
1536 scmd->result |= (DRIVER_TIMEOUT << 24);
1537 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1538 " cmd: %p\n",
1539 current->comm, scmd));
1540 scsi_finish_command(scmd);
1546 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1547 * @shost: Host to unjam.
1549 * Notes:
1550 * When we come in here, we *know* that all commands on the bus have
1551 * either completed, failed or timed out. we also know that no further
1552 * commands are being sent to the host, so things are relatively quiet
1553 * and we have freedom to fiddle with things as we wish.
1555 * This is only the *default* implementation. it is possible for
1556 * individual drivers to supply their own version of this function, and
1557 * if the maintainer wishes to do this, it is strongly suggested that
1558 * this function be taken as a template and modified. this function
1559 * was designed to correctly handle problems for about 95% of the
1560 * different cases out there, and it should always provide at least a
1561 * reasonable amount of error recovery.
1563 * Any command marked 'failed' or 'timeout' must eventually have
1564 * scsi_finish_cmd() called for it. we do all of the retry stuff
1565 * here, so when we restart the host after we return it should have an
1566 * empty queue.
1568 static void scsi_unjam_host(struct Scsi_Host *shost)
1570 unsigned long flags;
1571 LIST_HEAD(eh_work_q);
1572 LIST_HEAD(eh_done_q);
1574 spin_lock_irqsave(shost->host_lock, flags);
1575 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1576 spin_unlock_irqrestore(shost->host_lock, flags);
1578 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1580 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1581 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1582 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1584 scsi_eh_flush_done_q(&eh_done_q);
1588 * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
1589 * @data: Host for which we are running.
1591 * Notes:
1592 * This is always run in the context of a kernel thread. The idea is
1593 * that we start this thing up when the kernel starts up (one per host
1594 * that we detect), and it immediately goes to sleep and waits for some
1595 * event (i.e. failure). When this takes place, we have the job of
1596 * trying to unjam the bus and restarting things.
1598 int scsi_error_handler(void *data)
1600 struct Scsi_Host *shost = (struct Scsi_Host *) data;
1601 int rtn;
1602 DECLARE_MUTEX_LOCKED(sem);
1605 * Flush resources
1608 daemonize("scsi_eh_%d", shost->host_no);
1610 current->flags |= PF_NOFREEZE;
1612 shost->eh_wait = &sem;
1613 shost->ehandler = current;
1616 * Wake up the thread that created us.
1618 SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
1619 " scsi_eh_%d\n",shost->host_no));
1621 complete(shost->eh_notify);
1623 while (1) {
1625 * If we get a signal, it means we are supposed to go
1626 * away and die. This typically happens if the user is
1627 * trying to unload a module.
1629 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1630 " scsi_eh_%d"
1631 " sleeping\n",shost->host_no));
1634 * Note - we always use down_interruptible with the semaphore
1635 * even if the module was loaded as part of the kernel. The
1636 * reason is that down() will cause this thread to be counted
1637 * in the load average as a running process, and down
1638 * interruptible doesn't. Given that we need to allow this
1639 * thread to die if the driver was loaded as a module, using
1640 * semaphores isn't unreasonable.
1642 down_interruptible(&sem);
1643 if (shost->eh_kill)
1644 break;
1646 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1647 " scsi_eh_%d waking"
1648 " up\n",shost->host_no));
1650 shost->eh_active = 1;
1653 * We have a host that is failing for some reason. Figure out
1654 * what we need to do to get it up and online again (if we can).
1655 * If we fail, we end up taking the thing offline.
1657 if (shost->hostt->eh_strategy_handler)
1658 rtn = shost->hostt->eh_strategy_handler(shost);
1659 else
1660 scsi_unjam_host(shost);
1662 shost->eh_active = 0;
1665 * Note - if the above fails completely, the action is to take
1666 * individual devices offline and flush the queue of any
1667 * outstanding requests that may have been pending. When we
1668 * restart, we restart any I/O to any other devices on the bus
1669 * which are still online.
1671 scsi_restart_operations(shost);
1675 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
1676 " exiting\n",shost->host_no));
1679 * Make sure that nobody tries to wake us up again.
1681 shost->eh_wait = NULL;
1684 * Knock this down too. From this point on, the host is flying
1685 * without a pilot. If this is because the module is being unloaded,
1686 * that's fine. If the user sent a signal to this thing, we are
1687 * potentially in real danger.
1689 shost->eh_active = 0;
1690 shost->ehandler = NULL;
1693 * If anyone is waiting for us to exit (i.e. someone trying to unload
1694 * a driver), then wake up that process to let them know we are on
1695 * the way out the door.
1697 complete_and_exit(shost->eh_notify, 0);
1698 return 0;
1702 * Function: scsi_report_bus_reset()
1704 * Purpose: Utility function used by low-level drivers to report that
1705 * they have observed a bus reset on the bus being handled.
1707 * Arguments: shost - Host in question
1708 * channel - channel on which reset was observed.
1710 * Returns: Nothing
1712 * Lock status: Host lock must be held.
1714 * Notes: This only needs to be called if the reset is one which
1715 * originates from an unknown location. Resets originated
1716 * by the mid-level itself don't need to call this, but there
1717 * should be no harm.
1719 * The main purpose of this is to make sure that a CHECK_CONDITION
1720 * is properly treated.
1722 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1724 struct scsi_device *sdev;
1726 __shost_for_each_device(sdev, shost) {
1727 if (channel == sdev->channel) {
1728 sdev->was_reset = 1;
1729 sdev->expecting_cc_ua = 1;
1733 EXPORT_SYMBOL(scsi_report_bus_reset);
1736 * Function: scsi_report_device_reset()
1738 * Purpose: Utility function used by low-level drivers to report that
1739 * they have observed a device reset on the device being handled.
1741 * Arguments: shost - Host in question
1742 * channel - channel on which reset was observed
1743 * target - target on which reset was observed
1745 * Returns: Nothing
1747 * Lock status: Host lock must be held
1749 * Notes: This only needs to be called if the reset is one which
1750 * originates from an unknown location. Resets originated
1751 * by the mid-level itself don't need to call this, but there
1752 * should be no harm.
1754 * The main purpose of this is to make sure that a CHECK_CONDITION
1755 * is properly treated.
1757 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1759 struct scsi_device *sdev;
1761 __shost_for_each_device(sdev, shost) {
1762 if (channel == sdev->channel &&
1763 target == sdev->id) {
1764 sdev->was_reset = 1;
1765 sdev->expecting_cc_ua = 1;
1769 EXPORT_SYMBOL(scsi_report_device_reset);
1771 static void
1772 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1777 * Function: scsi_reset_provider
1779 * Purpose: Send requested reset to a bus or device at any phase.
1781 * Arguments: device - device to send reset to
1782 * flag - reset type (see scsi.h)
1784 * Returns: SUCCESS/FAILURE.
1786 * Notes: This is used by the SCSI Generic driver to provide
1787 * Bus/Device reset capability.
1790 scsi_reset_provider(struct scsi_device *dev, int flag)
1792 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1793 struct request req;
1794 int rtn;
1796 scmd->request = &req;
1797 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1798 scmd->request->rq_status = RQ_SCSI_BUSY;
1800 memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1802 scmd->scsi_done = scsi_reset_provider_done_command;
1803 scmd->done = NULL;
1804 scmd->buffer = NULL;
1805 scmd->bufflen = 0;
1806 scmd->request_buffer = NULL;
1807 scmd->request_bufflen = 0;
1809 scmd->cmd_len = 0;
1811 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1812 scmd->sc_request = NULL;
1813 scmd->sc_magic = SCSI_CMND_MAGIC;
1815 init_timer(&scmd->eh_timeout);
1818 * Sometimes the command can get back into the timer chain,
1819 * so use the pid as an identifier.
1821 scmd->pid = 0;
1823 switch (flag) {
1824 case SCSI_TRY_RESET_DEVICE:
1825 rtn = scsi_try_bus_device_reset(scmd);
1826 if (rtn == SUCCESS)
1827 break;
1828 /* FALLTHROUGH */
1829 case SCSI_TRY_RESET_BUS:
1830 rtn = scsi_try_bus_reset(scmd);
1831 if (rtn == SUCCESS)
1832 break;
1833 /* FALLTHROUGH */
1834 case SCSI_TRY_RESET_HOST:
1835 rtn = scsi_try_host_reset(scmd);
1836 break;
1837 default:
1838 rtn = FAILED;
1841 scsi_next_command(scmd);
1842 return rtn;
1844 EXPORT_SYMBOL(scsi_reset_provider);
1847 * scsi_normalize_sense - normalize main elements from either fixed or
1848 * descriptor sense data format into a common format.
1850 * @sense_buffer: byte array containing sense data returned by device
1851 * @sb_len: number of valid bytes in sense_buffer
1852 * @sshdr: pointer to instance of structure that common
1853 * elements are written to.
1855 * Notes:
1856 * The "main elements" from sense data are: response_code, sense_key,
1857 * asc, ascq and additional_length (only for descriptor format).
1859 * Typically this function can be called after a device has
1860 * responded to a SCSI command with the CHECK_CONDITION status.
1862 * Return value:
1863 * 1 if valid sense data information found, else 0;
1865 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1866 struct scsi_sense_hdr *sshdr)
1868 if (!sense_buffer || !sb_len || (sense_buffer[0] & 0x70) != 0x70)
1869 return 0;
1871 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1873 sshdr->response_code = (sense_buffer[0] & 0x7f);
1874 if (sshdr->response_code >= 0x72) {
1876 * descriptor format
1878 if (sb_len > 1)
1879 sshdr->sense_key = (sense_buffer[1] & 0xf);
1880 if (sb_len > 2)
1881 sshdr->asc = sense_buffer[2];
1882 if (sb_len > 3)
1883 sshdr->ascq = sense_buffer[3];
1884 if (sb_len > 7)
1885 sshdr->additional_length = sense_buffer[7];
1886 } else {
1888 * fixed format
1890 if (sb_len > 2)
1891 sshdr->sense_key = (sense_buffer[2] & 0xf);
1892 if (sb_len > 7) {
1893 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1894 sb_len : (sense_buffer[7] + 8);
1895 if (sb_len > 12)
1896 sshdr->asc = sense_buffer[12];
1897 if (sb_len > 13)
1898 sshdr->ascq = sense_buffer[13];
1902 return 1;
1904 EXPORT_SYMBOL(scsi_normalize_sense);
1906 int scsi_request_normalize_sense(struct scsi_request *sreq,
1907 struct scsi_sense_hdr *sshdr)
1909 return scsi_normalize_sense(sreq->sr_sense_buffer,
1910 sizeof(sreq->sr_sense_buffer), sshdr);
1912 EXPORT_SYMBOL(scsi_request_normalize_sense);
1914 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1915 struct scsi_sense_hdr *sshdr)
1917 return scsi_normalize_sense(cmd->sense_buffer,
1918 sizeof(cmd->sense_buffer), sshdr);
1920 EXPORT_SYMBOL(scsi_command_normalize_sense);
1923 * scsi_sense_desc_find - search for a given descriptor type in
1924 * 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)
1931 * Notes:
1932 * only valid when sense data is in descriptor format
1934 * Return value:
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,
1938 int desc_type)
1940 int add_sen_len, add_len, desc_len, k;
1941 const u8 * descp;
1943 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1944 return NULL;
1945 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1946 return NULL;
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) {
1951 descp += 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)
1955 return descp;
1956 if (add_len < 0) // short descriptor ??
1957 break;
1959 return NULL;
1961 EXPORT_SYMBOL(scsi_sense_desc_find);
1964 * scsi_get_sense_info_fld - attempts to get information field from
1965 * sense data (either fixed or descriptor format)
1967 * @sense_buffer: byte array of sense data
1968 * @sb_len: number of valid bytes in sense_buffer
1969 * @info_out: pointer to 64 integer where 8 or 4 byte information
1970 * field will be placed if found.
1972 * Return value:
1973 * 1 if information field found, 0 if not found.
1975 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1976 u64 * info_out)
1978 int j;
1979 const u8 * ucp;
1980 u64 ull;
1982 if (sb_len < 7)
1983 return 0;
1984 switch (sense_buffer[0] & 0x7f) {
1985 case 0x70:
1986 case 0x71:
1987 if (sense_buffer[0] & 0x80) {
1988 *info_out = (sense_buffer[3] << 24) +
1989 (sense_buffer[4] << 16) +
1990 (sense_buffer[5] << 8) + sense_buffer[6];
1991 return 1;
1992 } else
1993 return 0;
1994 case 0x72:
1995 case 0x73:
1996 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1997 0 /* info desc */);
1998 if (ucp && (0xa == ucp[1])) {
1999 ull = 0;
2000 for (j = 0; j < 8; ++j) {
2001 if (j > 0)
2002 ull <<= 8;
2003 ull |= ucp[4 + j];
2005 *info_out = ull;
2006 return 1;
2007 } else
2008 return 0;
2009 default:
2010 return 0;
2013 EXPORT_SYMBOL(scsi_get_sense_info_fld);