Staging: rtl8192e: Add #include <linux/vmalloc.h>
[linux-2.6/mini2440.git] / drivers / scsi / scsi_error.c
blob877204daf549644338fb2b55494944af7127a8ea
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/kernel.h>
22 #include <linux/freezer.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
39 #include "scsi_transport_api.h"
41 #define SENSE_TIMEOUT (10*HZ)
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
47 #define BUS_RESET_SETTLE_TIME (10)
48 #define HOST_RESET_SETTLE_TIME (10)
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host *shost)
53 if (shost->host_busy == shost->host_failed) {
54 wake_up_process(shost->ehandler);
55 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
60 /**
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
64 * Schedule SCSI EH without scmd.
66 void scsi_schedule_eh(struct Scsi_Host *shost)
68 unsigned long flags;
70 spin_lock_irqsave(shost->host_lock, flags);
72 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
73 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
74 shost->host_eh_scheduled++;
75 scsi_eh_wakeup(shost);
78 spin_unlock_irqrestore(shost->host_lock, flags);
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
82 /**
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
87 * Return value:
88 * 0 on failure.
90 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
92 struct Scsi_Host *shost = scmd->device->host;
93 unsigned long flags;
94 int ret = 0;
96 if (!shost->ehandler)
97 return 0;
99 spin_lock_irqsave(shost->host_lock, flags);
100 if (scsi_host_set_state(shost, SHOST_RECOVERY))
101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
102 goto out_unlock;
104 ret = 1;
105 scmd->eh_eflags |= eh_flag;
106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
107 shost->host_failed++;
108 scsi_eh_wakeup(shost);
109 out_unlock:
110 spin_unlock_irqrestore(shost->host_lock, flags);
111 return ret;
115 * scsi_times_out - Timeout function for normal scsi commands.
116 * @req: request that is timing out.
118 * Notes:
119 * We do not need to lock this. There is the potential for a race
120 * only in that the normal completion handling might run, but if the
121 * normal completion function determines that the timer has already
122 * fired, then it mustn't do anything.
124 enum blk_eh_timer_return scsi_times_out(struct request *req)
126 struct scsi_cmnd *scmd = req->special;
127 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
129 scsi_log_completion(scmd, TIMEOUT_ERROR);
131 if (scmd->device->host->transportt->eh_timed_out)
132 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
133 else if (scmd->device->host->hostt->eh_timed_out)
134 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
136 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
137 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
138 scmd->result |= DID_TIME_OUT << 16;
139 rtn = BLK_EH_HANDLED;
142 return rtn;
146 * scsi_block_when_processing_errors - Prevent cmds from being queued.
147 * @sdev: Device on which we are performing recovery.
149 * Description:
150 * We block until the host is out of error recovery, and then check to
151 * see whether the host or the device is offline.
153 * Return value:
154 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
156 int scsi_block_when_processing_errors(struct scsi_device *sdev)
158 int online;
160 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
162 online = scsi_device_online(sdev);
164 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
165 online));
167 return online;
169 EXPORT_SYMBOL(scsi_block_when_processing_errors);
171 #ifdef CONFIG_SCSI_LOGGING
173 * scsi_eh_prt_fail_stats - Log info on failures.
174 * @shost: scsi host being recovered.
175 * @work_q: Queue of scsi cmds to process.
177 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
178 struct list_head *work_q)
180 struct scsi_cmnd *scmd;
181 struct scsi_device *sdev;
182 int total_failures = 0;
183 int cmd_failed = 0;
184 int cmd_cancel = 0;
185 int devices_failed = 0;
187 shost_for_each_device(sdev, shost) {
188 list_for_each_entry(scmd, work_q, eh_entry) {
189 if (scmd->device == sdev) {
190 ++total_failures;
191 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
192 ++cmd_cancel;
193 else
194 ++cmd_failed;
198 if (cmd_cancel || cmd_failed) {
199 SCSI_LOG_ERROR_RECOVERY(3,
200 sdev_printk(KERN_INFO, sdev,
201 "%s: cmds failed: %d, cancel: %d\n",
202 __func__, cmd_failed,
203 cmd_cancel));
204 cmd_cancel = 0;
205 cmd_failed = 0;
206 ++devices_failed;
210 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
211 " devices require eh work\n",
212 total_failures, devices_failed));
214 #endif
217 * scsi_check_sense - Examine scsi cmd sense
218 * @scmd: Cmd to have sense checked.
220 * Return value:
221 * SUCCESS or FAILED or NEEDS_RETRY
223 * Notes:
224 * When a deferred error is detected the current command has
225 * not been executed and needs retrying.
227 static int scsi_check_sense(struct scsi_cmnd *scmd)
229 struct scsi_device *sdev = scmd->device;
230 struct scsi_sense_hdr sshdr;
232 if (! scsi_command_normalize_sense(scmd, &sshdr))
233 return FAILED; /* no valid sense data */
235 if (scsi_sense_is_deferred(&sshdr))
236 return NEEDS_RETRY;
238 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
239 sdev->scsi_dh_data->scsi_dh->check_sense) {
240 int rc;
242 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
243 if (rc != SCSI_RETURN_NOT_HANDLED)
244 return rc;
245 /* handler does not care. Drop down to default handling */
249 * Previous logic looked for FILEMARK, EOM or ILI which are
250 * mainly associated with tapes and returned SUCCESS.
252 if (sshdr.response_code == 0x70) {
253 /* fixed format */
254 if (scmd->sense_buffer[2] & 0xe0)
255 return SUCCESS;
256 } else {
258 * descriptor format: look for "stream commands sense data
259 * descriptor" (see SSC-3). Assume single sense data
260 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
262 if ((sshdr.additional_length > 3) &&
263 (scmd->sense_buffer[8] == 0x4) &&
264 (scmd->sense_buffer[11] & 0xe0))
265 return SUCCESS;
268 switch (sshdr.sense_key) {
269 case NO_SENSE:
270 return SUCCESS;
271 case RECOVERED_ERROR:
272 return /* soft_error */ SUCCESS;
274 case ABORTED_COMMAND:
275 if (sshdr.asc == 0x10) /* DIF */
276 return SUCCESS;
278 return NEEDS_RETRY;
279 case NOT_READY:
280 case UNIT_ATTENTION:
282 * if we are expecting a cc/ua because of a bus reset that we
283 * performed, treat this just as a retry. otherwise this is
284 * information that we should pass up to the upper-level driver
285 * so that we can deal with it there.
287 if (scmd->device->expecting_cc_ua) {
288 scmd->device->expecting_cc_ua = 0;
289 return NEEDS_RETRY;
292 * if the device is in the process of becoming ready, we
293 * should retry.
295 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
296 return NEEDS_RETRY;
298 * if the device is not started, we need to wake
299 * the error handler to start the motor
301 if (scmd->device->allow_restart &&
302 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
303 return FAILED;
304 return SUCCESS;
306 /* these three are not supported */
307 case COPY_ABORTED:
308 case VOLUME_OVERFLOW:
309 case MISCOMPARE:
310 return SUCCESS;
312 case MEDIUM_ERROR:
313 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
314 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
315 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
316 return SUCCESS;
318 return NEEDS_RETRY;
320 case HARDWARE_ERROR:
321 if (scmd->device->retry_hwerror)
322 return ADD_TO_MLQUEUE;
323 else
324 return SUCCESS;
326 case ILLEGAL_REQUEST:
327 case BLANK_CHECK:
328 case DATA_PROTECT:
329 default:
330 return SUCCESS;
335 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
336 * @scmd: SCSI cmd to examine.
338 * Notes:
339 * This is *only* called when we are examining the status of commands
340 * queued during error recovery. the main difference here is that we
341 * don't allow for the possibility of retries here, and we are a lot
342 * more restrictive about what we consider acceptable.
344 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
347 * first check the host byte, to see if there is anything in there
348 * that would indicate what we need to do.
350 if (host_byte(scmd->result) == DID_RESET) {
352 * rats. we are already in the error handler, so we now
353 * get to try and figure out what to do next. if the sense
354 * is valid, we have a pretty good idea of what to do.
355 * if not, we mark it as FAILED.
357 return scsi_check_sense(scmd);
359 if (host_byte(scmd->result) != DID_OK)
360 return FAILED;
363 * next, check the message byte.
365 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
366 return FAILED;
369 * now, check the status byte to see if this indicates
370 * anything special.
372 switch (status_byte(scmd->result)) {
373 case GOOD:
374 case COMMAND_TERMINATED:
375 return SUCCESS;
376 case CHECK_CONDITION:
377 return scsi_check_sense(scmd);
378 case CONDITION_GOOD:
379 case INTERMEDIATE_GOOD:
380 case INTERMEDIATE_C_GOOD:
382 * who knows? FIXME(eric)
384 return SUCCESS;
385 case RESERVATION_CONFLICT:
387 * let issuer deal with this, it could be just fine
389 return SUCCESS;
390 case BUSY:
391 case QUEUE_FULL:
392 default:
393 return FAILED;
395 return FAILED;
399 * scsi_eh_done - Completion function for error handling.
400 * @scmd: Cmd that is done.
402 static void scsi_eh_done(struct scsi_cmnd *scmd)
404 struct completion *eh_action;
406 SCSI_LOG_ERROR_RECOVERY(3,
407 printk("%s scmd: %p result: %x\n",
408 __func__, scmd, scmd->result));
410 eh_action = scmd->device->host->eh_action;
411 if (eh_action)
412 complete(eh_action);
416 * scsi_try_host_reset - ask host adapter to reset itself
417 * @scmd: SCSI cmd to send hsot reset.
419 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
421 unsigned long flags;
422 int rtn;
424 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
425 __func__));
427 if (!scmd->device->host->hostt->eh_host_reset_handler)
428 return FAILED;
430 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
432 if (rtn == SUCCESS) {
433 if (!scmd->device->host->hostt->skip_settle_delay)
434 ssleep(HOST_RESET_SETTLE_TIME);
435 spin_lock_irqsave(scmd->device->host->host_lock, flags);
436 scsi_report_bus_reset(scmd->device->host,
437 scmd_channel(scmd));
438 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
441 return rtn;
445 * scsi_try_bus_reset - ask host to perform a bus reset
446 * @scmd: SCSI cmd to send bus reset.
448 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
450 unsigned long flags;
451 int rtn;
453 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
454 __func__));
456 if (!scmd->device->host->hostt->eh_bus_reset_handler)
457 return FAILED;
459 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
461 if (rtn == SUCCESS) {
462 if (!scmd->device->host->hostt->skip_settle_delay)
463 ssleep(BUS_RESET_SETTLE_TIME);
464 spin_lock_irqsave(scmd->device->host->host_lock, flags);
465 scsi_report_bus_reset(scmd->device->host,
466 scmd_channel(scmd));
467 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
470 return rtn;
473 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
475 sdev->was_reset = 1;
476 sdev->expecting_cc_ua = 1;
480 * scsi_try_target_reset - Ask host to perform a target reset
481 * @scmd: SCSI cmd used to send a target reset
483 * Notes:
484 * There is no timeout for this operation. if this operation is
485 * unreliable for a given host, then the host itself needs to put a
486 * timer on it, and set the host back to a consistent state prior to
487 * returning.
489 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
491 unsigned long flags;
492 int rtn;
494 if (!scmd->device->host->hostt->eh_target_reset_handler)
495 return FAILED;
497 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
498 if (rtn == SUCCESS) {
499 spin_lock_irqsave(scmd->device->host->host_lock, flags);
500 __starget_for_each_device(scsi_target(scmd->device), NULL,
501 __scsi_report_device_reset);
502 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
505 return rtn;
509 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
510 * @scmd: SCSI cmd used to send BDR
512 * Notes:
513 * There is no timeout for this operation. if this operation is
514 * unreliable for a given host, then the host itself needs to put a
515 * timer on it, and set the host back to a consistent state prior to
516 * returning.
518 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
520 int rtn;
522 if (!scmd->device->host->hostt->eh_device_reset_handler)
523 return FAILED;
525 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
526 if (rtn == SUCCESS)
527 __scsi_report_device_reset(scmd->device, NULL);
528 return rtn;
531 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
533 if (!scmd->device->host->hostt->eh_abort_handler)
534 return FAILED;
536 return scmd->device->host->hostt->eh_abort_handler(scmd);
540 * scsi_try_to_abort_cmd - Ask host to abort a running command.
541 * @scmd: SCSI cmd to abort from Lower Level.
543 * Notes:
544 * This function will not return until the user's completion function
545 * has been called. there is no timeout on this operation. if the
546 * author of the low-level driver wishes this operation to be timed,
547 * they can provide this facility themselves. helper functions in
548 * scsi_error.c can be supplied to make this easier to do.
550 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
553 * scsi_done was called just after the command timed out and before
554 * we had a chance to process it. (db)
556 if (scmd->serial_number == 0)
557 return SUCCESS;
558 return __scsi_try_to_abort_cmd(scmd);
561 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
563 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
564 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
565 if (scsi_try_target_reset(scmd) != SUCCESS)
566 if (scsi_try_bus_reset(scmd) != SUCCESS)
567 scsi_try_host_reset(scmd);
571 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
572 * @scmd: SCSI command structure to hijack
573 * @ses: structure to save restore information
574 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
575 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
576 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
578 * This function is used to save a scsi command information before re-execution
579 * as part of the error recovery process. If @sense_bytes is 0 the command
580 * sent must be one that does not transfer any data. If @sense_bytes != 0
581 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
582 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
584 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
585 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
587 struct scsi_device *sdev = scmd->device;
590 * We need saved copies of a number of fields - this is because
591 * error handling may need to overwrite these with different values
592 * to run different commands, and once error handling is complete,
593 * we will need to restore these values prior to running the actual
594 * command.
596 ses->cmd_len = scmd->cmd_len;
597 ses->cmnd = scmd->cmnd;
598 ses->data_direction = scmd->sc_data_direction;
599 ses->sdb = scmd->sdb;
600 ses->next_rq = scmd->request->next_rq;
601 ses->result = scmd->result;
602 ses->underflow = scmd->underflow;
603 ses->prot_op = scmd->prot_op;
605 scmd->prot_op = SCSI_PROT_NORMAL;
606 scmd->cmnd = ses->eh_cmnd;
607 memset(scmd->cmnd, 0, BLK_MAX_CDB);
608 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
609 scmd->request->next_rq = NULL;
611 if (sense_bytes) {
612 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
613 sense_bytes);
614 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
615 scmd->sdb.length);
616 scmd->sdb.table.sgl = &ses->sense_sgl;
617 scmd->sc_data_direction = DMA_FROM_DEVICE;
618 scmd->sdb.table.nents = 1;
619 scmd->cmnd[0] = REQUEST_SENSE;
620 scmd->cmnd[4] = scmd->sdb.length;
621 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
622 } else {
623 scmd->sc_data_direction = DMA_NONE;
624 if (cmnd) {
625 BUG_ON(cmnd_size > BLK_MAX_CDB);
626 memcpy(scmd->cmnd, cmnd, cmnd_size);
627 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
631 scmd->underflow = 0;
633 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
634 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
635 (sdev->lun << 5 & 0xe0);
638 * Zero the sense buffer. The scsi spec mandates that any
639 * untransferred sense data should be interpreted as being zero.
641 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
643 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
646 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
647 * @scmd: SCSI command structure to restore
648 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
650 * Undo any damage done by above scsi_eh_prep_cmnd().
652 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
655 * Restore original data
657 scmd->cmd_len = ses->cmd_len;
658 scmd->cmnd = ses->cmnd;
659 scmd->sc_data_direction = ses->data_direction;
660 scmd->sdb = ses->sdb;
661 scmd->request->next_rq = ses->next_rq;
662 scmd->result = ses->result;
663 scmd->underflow = ses->underflow;
664 scmd->prot_op = ses->prot_op;
666 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
669 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
670 * @scmd: SCSI command structure to hijack
671 * @cmnd: CDB to send
672 * @cmnd_size: size in bytes of @cmnd
673 * @timeout: timeout for this request
674 * @sense_bytes: size of sense data to copy or 0
676 * This function is used to send a scsi command down to a target device
677 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
679 * Return value:
680 * SUCCESS or FAILED or NEEDS_RETRY
682 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
683 int cmnd_size, int timeout, unsigned sense_bytes)
685 struct scsi_device *sdev = scmd->device;
686 struct Scsi_Host *shost = sdev->host;
687 DECLARE_COMPLETION_ONSTACK(done);
688 unsigned long timeleft;
689 unsigned long flags;
690 struct scsi_eh_save ses;
691 int rtn;
693 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
694 shost->eh_action = &done;
696 spin_lock_irqsave(shost->host_lock, flags);
697 scsi_log_send(scmd);
698 shost->hostt->queuecommand(scmd, scsi_eh_done);
699 spin_unlock_irqrestore(shost->host_lock, flags);
701 timeleft = wait_for_completion_timeout(&done, timeout);
703 shost->eh_action = NULL;
705 scsi_log_completion(scmd, SUCCESS);
707 SCSI_LOG_ERROR_RECOVERY(3,
708 printk("%s: scmd: %p, timeleft: %ld\n",
709 __func__, scmd, timeleft));
712 * If there is time left scsi_eh_done got called, and we will
713 * examine the actual status codes to see whether the command
714 * actually did complete normally, else tell the host to forget
715 * about this command.
717 if (timeleft) {
718 rtn = scsi_eh_completed_normally(scmd);
719 SCSI_LOG_ERROR_RECOVERY(3,
720 printk("%s: scsi_eh_completed_normally %x\n",
721 __func__, rtn));
723 switch (rtn) {
724 case SUCCESS:
725 case NEEDS_RETRY:
726 case FAILED:
727 break;
728 default:
729 rtn = FAILED;
730 break;
732 } else {
733 scsi_abort_eh_cmnd(scmd);
734 rtn = FAILED;
737 scsi_eh_restore_cmnd(scmd, &ses);
738 return rtn;
742 * scsi_request_sense - Request sense data from a particular target.
743 * @scmd: SCSI cmd for request sense.
745 * Notes:
746 * Some hosts automatically obtain this information, others require
747 * that we obtain it on our own. This function will *not* return until
748 * the command either times out, or it completes.
750 static int scsi_request_sense(struct scsi_cmnd *scmd)
752 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
756 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
757 * @scmd: Original SCSI cmd that eh has finished.
758 * @done_q: Queue for processed commands.
760 * Notes:
761 * We don't want to use the normal command completion while we are are
762 * still handling errors - it may cause other commands to be queued,
763 * and that would disturb what we are doing. Thus we really want to
764 * keep a list of pending commands for final completion, and once we
765 * are ready to leave error handling we handle completion for real.
767 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
769 scmd->device->host->host_failed--;
770 scmd->eh_eflags = 0;
771 list_move_tail(&scmd->eh_entry, done_q);
773 EXPORT_SYMBOL(scsi_eh_finish_cmd);
776 * scsi_eh_get_sense - Get device sense data.
777 * @work_q: Queue of commands to process.
778 * @done_q: Queue of processed commands.
780 * Description:
781 * See if we need to request sense information. if so, then get it
782 * now, so we have a better idea of what to do.
784 * Notes:
785 * This has the unfortunate side effect that if a shost adapter does
786 * not automatically request sense information, we end up shutting
787 * it down before we request it.
789 * All drivers should request sense information internally these days,
790 * so for now all I have to say is tough noogies if you end up in here.
792 * XXX: Long term this code should go away, but that needs an audit of
793 * all LLDDs first.
795 int scsi_eh_get_sense(struct list_head *work_q,
796 struct list_head *done_q)
798 struct scsi_cmnd *scmd, *next;
799 int rtn;
801 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
802 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
803 SCSI_SENSE_VALID(scmd))
804 continue;
806 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
807 "%s: requesting sense\n",
808 current->comm));
809 rtn = scsi_request_sense(scmd);
810 if (rtn != SUCCESS)
811 continue;
813 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
814 " result %x\n", scmd,
815 scmd->result));
816 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
818 rtn = scsi_decide_disposition(scmd);
821 * if the result was normal, then just pass it along to the
822 * upper level.
824 if (rtn == SUCCESS)
825 /* we don't want this command reissued, just
826 * finished with the sense data, so set
827 * retries to the max allowed to ensure it
828 * won't get reissued */
829 scmd->retries = scmd->allowed;
830 else if (rtn != NEEDS_RETRY)
831 continue;
833 scsi_eh_finish_cmd(scmd, done_q);
836 return list_empty(work_q);
838 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
841 * scsi_eh_tur - Send TUR to device.
842 * @scmd: &scsi_cmnd to send TUR
844 * Return value:
845 * 0 - Device is ready. 1 - Device NOT ready.
847 static int scsi_eh_tur(struct scsi_cmnd *scmd)
849 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
850 int retry_cnt = 1, rtn;
852 retry_tur:
853 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
855 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
856 __func__, scmd, rtn));
858 switch (rtn) {
859 case NEEDS_RETRY:
860 if (retry_cnt--)
861 goto retry_tur;
862 /*FALLTHRU*/
863 case SUCCESS:
864 return 0;
865 default:
866 return 1;
871 * scsi_eh_abort_cmds - abort pending commands.
872 * @work_q: &list_head for pending commands.
873 * @done_q: &list_head for processed commands.
875 * Decription:
876 * Try and see whether or not it makes sense to try and abort the
877 * running command. This only works out to be the case if we have one
878 * command that has timed out. If the command simply failed, it makes
879 * no sense to try and abort the command, since as far as the shost
880 * adapter is concerned, it isn't running.
882 static int scsi_eh_abort_cmds(struct list_head *work_q,
883 struct list_head *done_q)
885 struct scsi_cmnd *scmd, *next;
886 int rtn;
888 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
889 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
890 continue;
891 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
892 "0x%p\n", current->comm,
893 scmd));
894 rtn = scsi_try_to_abort_cmd(scmd);
895 if (rtn == SUCCESS) {
896 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
897 if (!scsi_device_online(scmd->device) ||
898 !scsi_eh_tur(scmd)) {
899 scsi_eh_finish_cmd(scmd, done_q);
902 } else
903 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
904 " cmd failed:"
905 "0x%p\n",
906 current->comm,
907 scmd));
910 return list_empty(work_q);
914 * scsi_eh_try_stu - Send START_UNIT to device.
915 * @scmd: &scsi_cmnd to send START_UNIT
917 * Return value:
918 * 0 - Device is ready. 1 - Device NOT ready.
920 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
922 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
924 if (scmd->device->allow_restart) {
925 int i, rtn = NEEDS_RETRY;
927 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
928 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
930 if (rtn == SUCCESS)
931 return 0;
934 return 1;
938 * scsi_eh_stu - send START_UNIT if needed
939 * @shost: &scsi host being recovered.
940 * @work_q: &list_head for pending commands.
941 * @done_q: &list_head for processed commands.
943 * Notes:
944 * If commands are failing due to not ready, initializing command required,
945 * try revalidating the device, which will end up sending a start unit.
947 static int scsi_eh_stu(struct Scsi_Host *shost,
948 struct list_head *work_q,
949 struct list_head *done_q)
951 struct scsi_cmnd *scmd, *stu_scmd, *next;
952 struct scsi_device *sdev;
954 shost_for_each_device(sdev, shost) {
955 stu_scmd = NULL;
956 list_for_each_entry(scmd, work_q, eh_entry)
957 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
958 scsi_check_sense(scmd) == FAILED ) {
959 stu_scmd = scmd;
960 break;
963 if (!stu_scmd)
964 continue;
966 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
967 " 0x%p\n", current->comm, sdev));
969 if (!scsi_eh_try_stu(stu_scmd)) {
970 if (!scsi_device_online(sdev) ||
971 !scsi_eh_tur(stu_scmd)) {
972 list_for_each_entry_safe(scmd, next,
973 work_q, eh_entry) {
974 if (scmd->device == sdev)
975 scsi_eh_finish_cmd(scmd, done_q);
978 } else {
979 SCSI_LOG_ERROR_RECOVERY(3,
980 printk("%s: START_UNIT failed to sdev:"
981 " 0x%p\n", current->comm, sdev));
985 return list_empty(work_q);
990 * scsi_eh_bus_device_reset - send bdr if needed
991 * @shost: scsi host being recovered.
992 * @work_q: &list_head for pending commands.
993 * @done_q: &list_head for processed commands.
995 * Notes:
996 * Try a bus device reset. Still, look to see whether we have multiple
997 * devices that are jammed or not - if we have multiple devices, it
998 * makes no sense to try bus_device_reset - we really would need to try
999 * a bus_reset instead.
1001 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1002 struct list_head *work_q,
1003 struct list_head *done_q)
1005 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1006 struct scsi_device *sdev;
1007 int rtn;
1009 shost_for_each_device(sdev, shost) {
1010 bdr_scmd = NULL;
1011 list_for_each_entry(scmd, work_q, eh_entry)
1012 if (scmd->device == sdev) {
1013 bdr_scmd = scmd;
1014 break;
1017 if (!bdr_scmd)
1018 continue;
1020 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1021 " 0x%p\n", current->comm,
1022 sdev));
1023 rtn = scsi_try_bus_device_reset(bdr_scmd);
1024 if (rtn == SUCCESS) {
1025 if (!scsi_device_online(sdev) ||
1026 !scsi_eh_tur(bdr_scmd)) {
1027 list_for_each_entry_safe(scmd, next,
1028 work_q, eh_entry) {
1029 if (scmd->device == sdev)
1030 scsi_eh_finish_cmd(scmd,
1031 done_q);
1034 } else {
1035 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1036 " failed sdev:"
1037 "0x%p\n",
1038 current->comm,
1039 sdev));
1043 return list_empty(work_q);
1047 * scsi_eh_target_reset - send target reset if needed
1048 * @shost: scsi host being recovered.
1049 * @work_q: &list_head for pending commands.
1050 * @done_q: &list_head for processed commands.
1052 * Notes:
1053 * Try a target reset.
1055 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1056 struct list_head *work_q,
1057 struct list_head *done_q)
1059 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1060 unsigned int id = 0;
1061 int rtn;
1063 do {
1064 tgtr_scmd = NULL;
1065 list_for_each_entry(scmd, work_q, eh_entry) {
1066 if (id == scmd_id(scmd)) {
1067 tgtr_scmd = scmd;
1068 break;
1071 if (!tgtr_scmd) {
1072 /* not one exactly equal; find the next highest */
1073 list_for_each_entry(scmd, work_q, eh_entry) {
1074 if (scmd_id(scmd) > id &&
1075 (!tgtr_scmd ||
1076 scmd_id(tgtr_scmd) > scmd_id(scmd)))
1077 tgtr_scmd = scmd;
1080 if (!tgtr_scmd)
1081 /* no more commands, that's it */
1082 break;
1084 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1085 "to target %d\n",
1086 current->comm, id));
1087 rtn = scsi_try_target_reset(tgtr_scmd);
1088 if (rtn == SUCCESS) {
1089 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1090 if (id == scmd_id(scmd))
1091 if (!scsi_device_online(scmd->device) ||
1092 !scsi_eh_tur(tgtr_scmd))
1093 scsi_eh_finish_cmd(scmd,
1094 done_q);
1096 } else
1097 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1098 " failed target: "
1099 "%d\n",
1100 current->comm, id));
1101 id++;
1102 } while(id != 0);
1104 return list_empty(work_q);
1108 * scsi_eh_bus_reset - send a bus reset
1109 * @shost: &scsi host being recovered.
1110 * @work_q: &list_head for pending commands.
1111 * @done_q: &list_head for processed commands.
1113 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1114 struct list_head *work_q,
1115 struct list_head *done_q)
1117 struct scsi_cmnd *scmd, *chan_scmd, *next;
1118 unsigned int channel;
1119 int rtn;
1122 * we really want to loop over the various channels, and do this on
1123 * a channel by channel basis. we should also check to see if any
1124 * of the failed commands are on soft_reset devices, and if so, skip
1125 * the reset.
1128 for (channel = 0; channel <= shost->max_channel; channel++) {
1129 chan_scmd = NULL;
1130 list_for_each_entry(scmd, work_q, eh_entry) {
1131 if (channel == scmd_channel(scmd)) {
1132 chan_scmd = scmd;
1133 break;
1135 * FIXME add back in some support for
1136 * soft_reset devices.
1141 if (!chan_scmd)
1142 continue;
1143 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1144 " %d\n", current->comm,
1145 channel));
1146 rtn = scsi_try_bus_reset(chan_scmd);
1147 if (rtn == SUCCESS) {
1148 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1149 if (channel == scmd_channel(scmd))
1150 if (!scsi_device_online(scmd->device) ||
1151 !scsi_eh_tur(scmd))
1152 scsi_eh_finish_cmd(scmd,
1153 done_q);
1155 } else {
1156 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1157 " failed chan: %d\n",
1158 current->comm,
1159 channel));
1162 return list_empty(work_q);
1166 * scsi_eh_host_reset - send a host reset
1167 * @work_q: list_head for processed commands.
1168 * @done_q: list_head for processed commands.
1170 static int scsi_eh_host_reset(struct list_head *work_q,
1171 struct list_head *done_q)
1173 struct scsi_cmnd *scmd, *next;
1174 int rtn;
1176 if (!list_empty(work_q)) {
1177 scmd = list_entry(work_q->next,
1178 struct scsi_cmnd, eh_entry);
1180 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1181 , current->comm));
1183 rtn = scsi_try_host_reset(scmd);
1184 if (rtn == SUCCESS) {
1185 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1186 if (!scsi_device_online(scmd->device) ||
1187 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1188 !scsi_eh_tur(scmd))
1189 scsi_eh_finish_cmd(scmd, done_q);
1191 } else {
1192 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1193 " failed\n",
1194 current->comm));
1197 return list_empty(work_q);
1201 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1202 * @work_q: list_head for processed commands.
1203 * @done_q: list_head for processed commands.
1205 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1206 struct list_head *done_q)
1208 struct scsi_cmnd *scmd, *next;
1210 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1211 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1212 "not ready after error recovery\n");
1213 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1214 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1216 * FIXME: Handle lost cmds.
1219 scsi_eh_finish_cmd(scmd, done_q);
1221 return;
1225 * scsi_noretry_cmd - determinte if command should be failed fast
1226 * @scmd: SCSI cmd to examine.
1228 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1230 switch (host_byte(scmd->result)) {
1231 case DID_OK:
1232 break;
1233 case DID_BUS_BUSY:
1234 return blk_failfast_transport(scmd->request);
1235 case DID_PARITY:
1236 return blk_failfast_dev(scmd->request);
1237 case DID_ERROR:
1238 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1239 status_byte(scmd->result) == RESERVATION_CONFLICT)
1240 return 0;
1241 /* fall through */
1242 case DID_SOFT_ERROR:
1243 return blk_failfast_driver(scmd->request);
1246 switch (status_byte(scmd->result)) {
1247 case CHECK_CONDITION:
1249 * assume caller has checked sense and determinted
1250 * the check condition was retryable.
1252 return blk_failfast_dev(scmd->request);
1255 return 0;
1259 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1260 * @scmd: SCSI cmd to examine.
1262 * Notes:
1263 * This is *only* called when we are examining the status after sending
1264 * out the actual data command. any commands that are queued for error
1265 * recovery (e.g. test_unit_ready) do *not* come through here.
1267 * When this routine returns failed, it means the error handler thread
1268 * is woken. In cases where the error code indicates an error that
1269 * doesn't require the error handler read (i.e. we don't need to
1270 * abort/reset), this function should return SUCCESS.
1272 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1274 int rtn;
1277 * if the device is offline, then we clearly just pass the result back
1278 * up to the top level.
1280 if (!scsi_device_online(scmd->device)) {
1281 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1282 " as SUCCESS\n",
1283 __func__));
1284 return SUCCESS;
1288 * first check the host byte, to see if there is anything in there
1289 * that would indicate what we need to do.
1291 switch (host_byte(scmd->result)) {
1292 case DID_PASSTHROUGH:
1294 * no matter what, pass this through to the upper layer.
1295 * nuke this special code so that it looks like we are saying
1296 * did_ok.
1298 scmd->result &= 0xff00ffff;
1299 return SUCCESS;
1300 case DID_OK:
1302 * looks good. drop through, and check the next byte.
1304 break;
1305 case DID_NO_CONNECT:
1306 case DID_BAD_TARGET:
1307 case DID_ABORT:
1309 * note - this means that we just report the status back
1310 * to the top level driver, not that we actually think
1311 * that it indicates SUCCESS.
1313 return SUCCESS;
1315 * when the low level driver returns did_soft_error,
1316 * it is responsible for keeping an internal retry counter
1317 * in order to avoid endless loops (db)
1319 * actually this is a bug in this function here. we should
1320 * be mindful of the maximum number of retries specified
1321 * and not get stuck in a loop.
1323 case DID_SOFT_ERROR:
1324 goto maybe_retry;
1325 case DID_IMM_RETRY:
1326 return NEEDS_RETRY;
1328 case DID_REQUEUE:
1329 return ADD_TO_MLQUEUE;
1330 case DID_TRANSPORT_DISRUPTED:
1332 * LLD/transport was disrupted during processing of the IO.
1333 * The transport class is now blocked/blocking,
1334 * and the transport will decide what to do with the IO
1335 * based on its timers and recovery capablilities if
1336 * there are enough retries.
1338 goto maybe_retry;
1339 case DID_TRANSPORT_FAILFAST:
1341 * The transport decided to failfast the IO (most likely
1342 * the fast io fail tmo fired), so send IO directly upwards.
1344 return SUCCESS;
1345 case DID_ERROR:
1346 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1347 status_byte(scmd->result) == RESERVATION_CONFLICT)
1349 * execute reservation conflict processing code
1350 * lower down
1352 break;
1353 /* fallthrough */
1355 case DID_BUS_BUSY:
1356 case DID_PARITY:
1357 goto maybe_retry;
1358 case DID_TIME_OUT:
1360 * when we scan the bus, we get timeout messages for
1361 * these commands if there is no device available.
1362 * other hosts report did_no_connect for the same thing.
1364 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1365 scmd->cmnd[0] == INQUIRY)) {
1366 return SUCCESS;
1367 } else {
1368 return FAILED;
1370 case DID_RESET:
1371 return SUCCESS;
1372 default:
1373 return FAILED;
1377 * next, check the message byte.
1379 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1380 return FAILED;
1383 * check the status byte to see if this indicates anything special.
1385 switch (status_byte(scmd->result)) {
1386 case QUEUE_FULL:
1388 * the case of trying to send too many commands to a
1389 * tagged queueing device.
1391 case BUSY:
1393 * device can't talk to us at the moment. Should only
1394 * occur (SAM-3) when the task queue is empty, so will cause
1395 * the empty queue handling to trigger a stall in the
1396 * device.
1398 return ADD_TO_MLQUEUE;
1399 case GOOD:
1400 case COMMAND_TERMINATED:
1401 return SUCCESS;
1402 case TASK_ABORTED:
1403 goto maybe_retry;
1404 case CHECK_CONDITION:
1405 rtn = scsi_check_sense(scmd);
1406 if (rtn == NEEDS_RETRY)
1407 goto maybe_retry;
1408 /* if rtn == FAILED, we have no sense information;
1409 * returning FAILED will wake the error handler thread
1410 * to collect the sense and redo the decide
1411 * disposition */
1412 return rtn;
1413 case CONDITION_GOOD:
1414 case INTERMEDIATE_GOOD:
1415 case INTERMEDIATE_C_GOOD:
1416 case ACA_ACTIVE:
1418 * who knows? FIXME(eric)
1420 return SUCCESS;
1422 case RESERVATION_CONFLICT:
1423 sdev_printk(KERN_INFO, scmd->device,
1424 "reservation conflict\n");
1425 return SUCCESS; /* causes immediate i/o error */
1426 default:
1427 return FAILED;
1429 return FAILED;
1431 maybe_retry:
1433 /* we requeue for retry because the error was retryable, and
1434 * the request was not marked fast fail. Note that above,
1435 * even if the request is marked fast fail, we still requeue
1436 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1437 if ((++scmd->retries) <= scmd->allowed
1438 && !scsi_noretry_cmd(scmd)) {
1439 return NEEDS_RETRY;
1440 } else {
1442 * no more retries - report this one back to upper level.
1444 return SUCCESS;
1448 static void eh_lock_door_done(struct request *req, int uptodate)
1450 __blk_put_request(req->q, req);
1454 * scsi_eh_lock_door - Prevent medium removal for the specified device
1455 * @sdev: SCSI device to prevent medium removal
1457 * Locking:
1458 * We must be called from process context.
1460 * Notes:
1461 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1462 * head of the devices request queue, and continue.
1464 static void scsi_eh_lock_door(struct scsi_device *sdev)
1466 struct request *req;
1469 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1470 * request becomes available
1472 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1474 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1475 req->cmd[1] = 0;
1476 req->cmd[2] = 0;
1477 req->cmd[3] = 0;
1478 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1479 req->cmd[5] = 0;
1481 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1483 req->cmd_type = REQ_TYPE_BLOCK_PC;
1484 req->cmd_flags |= REQ_QUIET;
1485 req->timeout = 10 * HZ;
1486 req->retries = 5;
1488 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1492 * scsi_restart_operations - restart io operations to the specified host.
1493 * @shost: Host we are restarting.
1495 * Notes:
1496 * When we entered the error handler, we blocked all further i/o to
1497 * this device. we need to 'reverse' this process.
1499 static void scsi_restart_operations(struct Scsi_Host *shost)
1501 struct scsi_device *sdev;
1502 unsigned long flags;
1505 * If the door was locked, we need to insert a door lock request
1506 * onto the head of the SCSI request queue for the device. There
1507 * is no point trying to lock the door of an off-line device.
1509 shost_for_each_device(sdev, shost) {
1510 if (scsi_device_online(sdev) && sdev->locked)
1511 scsi_eh_lock_door(sdev);
1515 * next free up anything directly waiting upon the host. this
1516 * will be requests for character device operations, and also for
1517 * ioctls to queued block devices.
1519 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1520 __func__));
1522 spin_lock_irqsave(shost->host_lock, flags);
1523 if (scsi_host_set_state(shost, SHOST_RUNNING))
1524 if (scsi_host_set_state(shost, SHOST_CANCEL))
1525 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1526 spin_unlock_irqrestore(shost->host_lock, flags);
1528 wake_up(&shost->host_wait);
1531 * finally we need to re-initiate requests that may be pending. we will
1532 * have had everything blocked while error handling is taking place, and
1533 * now that error recovery is done, we will need to ensure that these
1534 * requests are started.
1536 scsi_run_host_queues(shost);
1540 * scsi_eh_ready_devs - check device ready state and recover if not.
1541 * @shost: host to be recovered.
1542 * @work_q: &list_head for pending commands.
1543 * @done_q: &list_head for processed commands.
1545 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1546 struct list_head *work_q,
1547 struct list_head *done_q)
1549 if (!scsi_eh_stu(shost, work_q, done_q))
1550 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1551 if (!scsi_eh_target_reset(shost, work_q, done_q))
1552 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1553 if (!scsi_eh_host_reset(work_q, done_q))
1554 scsi_eh_offline_sdevs(work_q,
1555 done_q);
1557 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1560 * scsi_eh_flush_done_q - finish processed commands or retry them.
1561 * @done_q: list_head of processed commands.
1563 void scsi_eh_flush_done_q(struct list_head *done_q)
1565 struct scsi_cmnd *scmd, *next;
1567 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1568 list_del_init(&scmd->eh_entry);
1569 if (scsi_device_online(scmd->device) &&
1570 !scsi_noretry_cmd(scmd) &&
1571 (++scmd->retries <= scmd->allowed)) {
1572 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1573 " retry cmd: %p\n",
1574 current->comm,
1575 scmd));
1576 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1577 } else {
1579 * If just we got sense for the device (called
1580 * scsi_eh_get_sense), scmd->result is already
1581 * set, do not set DRIVER_TIMEOUT.
1583 if (!scmd->result)
1584 scmd->result |= (DRIVER_TIMEOUT << 24);
1585 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1586 " cmd: %p\n",
1587 current->comm, scmd));
1588 scsi_finish_command(scmd);
1592 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1595 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1596 * @shost: Host to unjam.
1598 * Notes:
1599 * When we come in here, we *know* that all commands on the bus have
1600 * either completed, failed or timed out. we also know that no further
1601 * commands are being sent to the host, so things are relatively quiet
1602 * and we have freedom to fiddle with things as we wish.
1604 * This is only the *default* implementation. it is possible for
1605 * individual drivers to supply their own version of this function, and
1606 * if the maintainer wishes to do this, it is strongly suggested that
1607 * this function be taken as a template and modified. this function
1608 * was designed to correctly handle problems for about 95% of the
1609 * different cases out there, and it should always provide at least a
1610 * reasonable amount of error recovery.
1612 * Any command marked 'failed' or 'timeout' must eventually have
1613 * scsi_finish_cmd() called for it. we do all of the retry stuff
1614 * here, so when we restart the host after we return it should have an
1615 * empty queue.
1617 static void scsi_unjam_host(struct Scsi_Host *shost)
1619 unsigned long flags;
1620 LIST_HEAD(eh_work_q);
1621 LIST_HEAD(eh_done_q);
1623 spin_lock_irqsave(shost->host_lock, flags);
1624 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1625 spin_unlock_irqrestore(shost->host_lock, flags);
1627 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1629 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1630 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1631 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1633 scsi_eh_flush_done_q(&eh_done_q);
1637 * scsi_error_handler - SCSI error handler thread
1638 * @data: Host for which we are running.
1640 * Notes:
1641 * This is the main error handling loop. This is run as a kernel thread
1642 * for every SCSI host and handles all error handling activity.
1644 int scsi_error_handler(void *data)
1646 struct Scsi_Host *shost = data;
1649 * We use TASK_INTERRUPTIBLE so that the thread is not
1650 * counted against the load average as a running process.
1651 * We never actually get interrupted because kthread_run
1652 * disables signal delivery for the created thread.
1654 set_current_state(TASK_INTERRUPTIBLE);
1655 while (!kthread_should_stop()) {
1656 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1657 shost->host_failed != shost->host_busy) {
1658 SCSI_LOG_ERROR_RECOVERY(1,
1659 printk("Error handler scsi_eh_%d sleeping\n",
1660 shost->host_no));
1661 schedule();
1662 set_current_state(TASK_INTERRUPTIBLE);
1663 continue;
1666 __set_current_state(TASK_RUNNING);
1667 SCSI_LOG_ERROR_RECOVERY(1,
1668 printk("Error handler scsi_eh_%d waking up\n",
1669 shost->host_no));
1672 * We have a host that is failing for some reason. Figure out
1673 * what we need to do to get it up and online again (if we can).
1674 * If we fail, we end up taking the thing offline.
1676 if (shost->transportt->eh_strategy_handler)
1677 shost->transportt->eh_strategy_handler(shost);
1678 else
1679 scsi_unjam_host(shost);
1682 * Note - if the above fails completely, the action is to take
1683 * individual devices offline and flush the queue of any
1684 * outstanding requests that may have been pending. When we
1685 * restart, we restart any I/O to any other devices on the bus
1686 * which are still online.
1688 scsi_restart_operations(shost);
1689 set_current_state(TASK_INTERRUPTIBLE);
1691 __set_current_state(TASK_RUNNING);
1693 SCSI_LOG_ERROR_RECOVERY(1,
1694 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1695 shost->ehandler = NULL;
1696 return 0;
1700 * Function: scsi_report_bus_reset()
1702 * Purpose: Utility function used by low-level drivers to report that
1703 * they have observed a bus reset on the bus being handled.
1705 * Arguments: shost - Host in question
1706 * channel - channel on which reset was observed.
1708 * Returns: Nothing
1710 * Lock status: Host lock must be held.
1712 * Notes: This only needs to be called if the reset is one which
1713 * originates from an unknown location. Resets originated
1714 * by the mid-level itself don't need to call this, but there
1715 * should be no harm.
1717 * The main purpose of this is to make sure that a CHECK_CONDITION
1718 * is properly treated.
1720 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1722 struct scsi_device *sdev;
1724 __shost_for_each_device(sdev, shost) {
1725 if (channel == sdev_channel(sdev))
1726 __scsi_report_device_reset(sdev, NULL);
1729 EXPORT_SYMBOL(scsi_report_bus_reset);
1732 * Function: scsi_report_device_reset()
1734 * Purpose: Utility function used by low-level drivers to report that
1735 * they have observed a device reset on the device being handled.
1737 * Arguments: shost - Host in question
1738 * channel - channel on which reset was observed
1739 * target - target on which reset was observed
1741 * Returns: Nothing
1743 * Lock status: Host lock must be held
1745 * Notes: This only needs to be called if the reset is one which
1746 * originates from an unknown location. Resets originated
1747 * by the mid-level itself don't need to call this, but there
1748 * should be no harm.
1750 * The main purpose of this is to make sure that a CHECK_CONDITION
1751 * is properly treated.
1753 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1755 struct scsi_device *sdev;
1757 __shost_for_each_device(sdev, shost) {
1758 if (channel == sdev_channel(sdev) &&
1759 target == sdev_id(sdev))
1760 __scsi_report_device_reset(sdev, NULL);
1763 EXPORT_SYMBOL(scsi_report_device_reset);
1765 static void
1766 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1771 * Function: scsi_reset_provider
1773 * Purpose: Send requested reset to a bus or device at any phase.
1775 * Arguments: device - device to send reset to
1776 * flag - reset type (see scsi.h)
1778 * Returns: SUCCESS/FAILURE.
1780 * Notes: This is used by the SCSI Generic driver to provide
1781 * Bus/Device reset capability.
1784 scsi_reset_provider(struct scsi_device *dev, int flag)
1786 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1787 struct Scsi_Host *shost = dev->host;
1788 struct request req;
1789 unsigned long flags;
1790 int rtn;
1792 blk_rq_init(NULL, &req);
1793 scmd->request = &req;
1795 scmd->cmnd = req.cmd;
1797 scmd->scsi_done = scsi_reset_provider_done_command;
1798 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1800 scmd->cmd_len = 0;
1802 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1804 spin_lock_irqsave(shost->host_lock, flags);
1805 shost->tmf_in_progress = 1;
1806 spin_unlock_irqrestore(shost->host_lock, flags);
1808 switch (flag) {
1809 case SCSI_TRY_RESET_DEVICE:
1810 rtn = scsi_try_bus_device_reset(scmd);
1811 if (rtn == SUCCESS)
1812 break;
1813 /* FALLTHROUGH */
1814 case SCSI_TRY_RESET_TARGET:
1815 rtn = scsi_try_target_reset(scmd);
1816 if (rtn == SUCCESS)
1817 break;
1818 /* FALLTHROUGH */
1819 case SCSI_TRY_RESET_BUS:
1820 rtn = scsi_try_bus_reset(scmd);
1821 if (rtn == SUCCESS)
1822 break;
1823 /* FALLTHROUGH */
1824 case SCSI_TRY_RESET_HOST:
1825 rtn = scsi_try_host_reset(scmd);
1826 break;
1827 default:
1828 rtn = FAILED;
1831 spin_lock_irqsave(shost->host_lock, flags);
1832 shost->tmf_in_progress = 0;
1833 spin_unlock_irqrestore(shost->host_lock, flags);
1836 * be sure to wake up anyone who was sleeping or had their queue
1837 * suspended while we performed the TMF.
1839 SCSI_LOG_ERROR_RECOVERY(3,
1840 printk("%s: waking up host to restart after TMF\n",
1841 __func__));
1843 wake_up(&shost->host_wait);
1845 scsi_run_host_queues(shost);
1847 scsi_next_command(scmd);
1848 return rtn;
1850 EXPORT_SYMBOL(scsi_reset_provider);
1853 * scsi_normalize_sense - normalize main elements from either fixed or
1854 * descriptor sense data format into a common format.
1856 * @sense_buffer: byte array containing sense data returned by device
1857 * @sb_len: number of valid bytes in sense_buffer
1858 * @sshdr: pointer to instance of structure that common
1859 * elements are written to.
1861 * Notes:
1862 * The "main elements" from sense data are: response_code, sense_key,
1863 * asc, ascq and additional_length (only for descriptor format).
1865 * Typically this function can be called after a device has
1866 * responded to a SCSI command with the CHECK_CONDITION status.
1868 * Return value:
1869 * 1 if valid sense data information found, else 0;
1871 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1872 struct scsi_sense_hdr *sshdr)
1874 if (!sense_buffer || !sb_len)
1875 return 0;
1877 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1879 sshdr->response_code = (sense_buffer[0] & 0x7f);
1881 if (!scsi_sense_valid(sshdr))
1882 return 0;
1884 if (sshdr->response_code >= 0x72) {
1886 * descriptor format
1888 if (sb_len > 1)
1889 sshdr->sense_key = (sense_buffer[1] & 0xf);
1890 if (sb_len > 2)
1891 sshdr->asc = sense_buffer[2];
1892 if (sb_len > 3)
1893 sshdr->ascq = sense_buffer[3];
1894 if (sb_len > 7)
1895 sshdr->additional_length = sense_buffer[7];
1896 } else {
1898 * fixed format
1900 if (sb_len > 2)
1901 sshdr->sense_key = (sense_buffer[2] & 0xf);
1902 if (sb_len > 7) {
1903 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1904 sb_len : (sense_buffer[7] + 8);
1905 if (sb_len > 12)
1906 sshdr->asc = sense_buffer[12];
1907 if (sb_len > 13)
1908 sshdr->ascq = sense_buffer[13];
1912 return 1;
1914 EXPORT_SYMBOL(scsi_normalize_sense);
1916 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1917 struct scsi_sense_hdr *sshdr)
1919 return scsi_normalize_sense(cmd->sense_buffer,
1920 SCSI_SENSE_BUFFERSIZE, sshdr);
1922 EXPORT_SYMBOL(scsi_command_normalize_sense);
1925 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1926 * @sense_buffer: byte array of descriptor format sense data
1927 * @sb_len: number of valid bytes in sense_buffer
1928 * @desc_type: value of descriptor type to find
1929 * (e.g. 0 -> information)
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 - get information field from sense data (either fixed or descriptor format)
1965 * @sense_buffer: byte array of sense data
1966 * @sb_len: number of valid bytes in sense_buffer
1967 * @info_out: pointer to 64 integer where 8 or 4 byte information
1968 * field will be placed if found.
1970 * Return value:
1971 * 1 if information field found, 0 if not found.
1973 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1974 u64 * info_out)
1976 int j;
1977 const u8 * ucp;
1978 u64 ull;
1980 if (sb_len < 7)
1981 return 0;
1982 switch (sense_buffer[0] & 0x7f) {
1983 case 0x70:
1984 case 0x71:
1985 if (sense_buffer[0] & 0x80) {
1986 *info_out = (sense_buffer[3] << 24) +
1987 (sense_buffer[4] << 16) +
1988 (sense_buffer[5] << 8) + sense_buffer[6];
1989 return 1;
1990 } else
1991 return 0;
1992 case 0x72:
1993 case 0x73:
1994 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1995 0 /* info desc */);
1996 if (ucp && (0xa == ucp[1])) {
1997 ull = 0;
1998 for (j = 0; j < 8; ++j) {
1999 if (j > 0)
2000 ull <<= 8;
2001 ull |= ucp[4 + j];
2003 *info_out = ull;
2004 return 1;
2005 } else
2006 return 0;
2007 default:
2008 return 0;
2011 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2014 * scsi_build_sense_buffer - build sense data in a buffer
2015 * @desc: Sense format (non zero == descriptor format,
2016 * 0 == fixed format)
2017 * @buf: Where to build sense data
2018 * @key: Sense key
2019 * @asc: Additional sense code
2020 * @ascq: Additional sense code qualifier
2023 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2025 if (desc) {
2026 buf[0] = 0x72; /* descriptor, current */
2027 buf[1] = key;
2028 buf[2] = asc;
2029 buf[3] = ascq;
2030 buf[7] = 0;
2031 } else {
2032 buf[0] = 0x70; /* fixed, current */
2033 buf[2] = key;
2034 buf[7] = 0xa;
2035 buf[12] = asc;
2036 buf[13] = ascq;
2039 EXPORT_SYMBOL(scsi_build_sense_buffer);