2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/timer.h>
20 #include <linux/string.h>
21 #include <linux/kernel.h>
22 #include <linux/freezer.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
39 #include "scsi_transport_api.h"
41 #define SENSE_TIMEOUT (10*HZ)
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
47 #define BUS_RESET_SETTLE_TIME (10)
48 #define HOST_RESET_SETTLE_TIME (10)
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host *shost)
53 if (shost->host_busy == shost->host_failed) {
54 wake_up_process(shost->ehandler);
55 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
64 * Schedule SCSI EH without scmd.
66 void scsi_schedule_eh(struct Scsi_Host *shost)
70 spin_lock_irqsave(shost->host_lock, flags);
72 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
73 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
74 shost->host_eh_scheduled++;
75 scsi_eh_wakeup(shost);
78 spin_unlock_irqrestore(shost->host_lock, flags);
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
90 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
92 struct Scsi_Host *shost = scmd->device->host;
99 spin_lock_irqsave(shost->host_lock, flags);
100 if (scsi_host_set_state(shost, SHOST_RECOVERY))
101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
105 scmd->eh_eflags |= eh_flag;
106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
107 shost->host_failed++;
108 scsi_eh_wakeup(shost);
110 spin_unlock_irqrestore(shost->host_lock, flags);
115 * scsi_add_timer - Start timeout timer for a single scsi command.
116 * @scmd: scsi command that is about to start running.
117 * @timeout: amount of time to allow this command to run.
118 * @complete: timeout function to call if timer isn't canceled.
121 * This should be turned into an inline function. Each scsi command
122 * has its own timer, and as it is added to the queue, we set up the
123 * timer. When the command completes, we cancel the timer.
125 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
126 void (*complete)(struct scsi_cmnd *))
130 * If the clock was already running for this command, then
131 * first delete the timer. The timer handling code gets rather
132 * confused if we don't do this.
134 if (scmd->eh_timeout.function)
135 del_timer(&scmd->eh_timeout);
137 scmd->eh_timeout.data = (unsigned long)scmd;
138 scmd->eh_timeout.expires = jiffies + timeout;
139 scmd->eh_timeout.function = (void (*)(unsigned long)) complete;
141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
142 " %d, (%p)\n", __func__,
143 scmd, timeout, complete));
145 add_timer(&scmd->eh_timeout);
149 * scsi_delete_timer - Delete/cancel timer for a given function.
150 * @scmd: Cmd that we are canceling timer for
153 * This should be turned into an inline function.
156 * 1 if we were able to detach the timer. 0 if we blew it, and the
157 * timer function has already started to run.
159 int scsi_delete_timer(struct scsi_cmnd *scmd)
163 rtn = del_timer(&scmd->eh_timeout);
165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
166 " rtn: %d\n", __func__,
169 scmd->eh_timeout.data = (unsigned long)NULL;
170 scmd->eh_timeout.function = NULL;
176 * scsi_times_out - Timeout function for normal scsi commands.
177 * @scmd: Cmd that is timing out.
180 * We do not need to lock this. There is the potential for a race
181 * only in that the normal completion handling might run, but if the
182 * normal completion function determines that the timer has already
183 * fired, then it mustn't do anything.
185 void scsi_times_out(struct scsi_cmnd *scmd)
187 enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
189 scsi_log_completion(scmd, TIMEOUT_ERROR);
191 if (scmd->device->host->transportt->eh_timed_out)
192 eh_timed_out = scmd->device->host->transportt->eh_timed_out;
193 else if (scmd->device->host->hostt->eh_timed_out)
194 eh_timed_out = scmd->device->host->hostt->eh_timed_out;
199 switch (eh_timed_out(scmd)) {
204 scsi_add_timer(scmd, scmd->timeout_per_command,
211 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
212 scmd->result |= DID_TIME_OUT << 16;
218 * scsi_block_when_processing_errors - Prevent cmds from being queued.
219 * @sdev: Device on which we are performing recovery.
222 * We block until the host is out of error recovery, and then check to
223 * see whether the host or the device is offline.
226 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
228 int scsi_block_when_processing_errors(struct scsi_device *sdev)
232 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
234 online = scsi_device_online(sdev);
236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
241 EXPORT_SYMBOL(scsi_block_when_processing_errors);
243 #ifdef CONFIG_SCSI_LOGGING
245 * scsi_eh_prt_fail_stats - Log info on failures.
246 * @shost: scsi host being recovered.
247 * @work_q: Queue of scsi cmds to process.
249 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
250 struct list_head *work_q)
252 struct scsi_cmnd *scmd;
253 struct scsi_device *sdev;
254 int total_failures = 0;
257 int devices_failed = 0;
259 shost_for_each_device(sdev, shost) {
260 list_for_each_entry(scmd, work_q, eh_entry) {
261 if (scmd->device == sdev) {
263 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
270 if (cmd_cancel || cmd_failed) {
271 SCSI_LOG_ERROR_RECOVERY(3,
272 sdev_printk(KERN_INFO, sdev,
273 "%s: cmds failed: %d, cancel: %d\n",
274 __func__, cmd_failed,
282 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
283 " devices require eh work\n",
284 total_failures, devices_failed));
289 * scsi_check_sense - Examine scsi cmd sense
290 * @scmd: Cmd to have sense checked.
293 * SUCCESS or FAILED or NEEDS_RETRY
296 * When a deferred error is detected the current command has
297 * not been executed and needs retrying.
299 static int scsi_check_sense(struct scsi_cmnd *scmd)
301 struct scsi_device *sdev = scmd->device;
302 struct scsi_sense_hdr sshdr;
304 if (! scsi_command_normalize_sense(scmd, &sshdr))
305 return FAILED; /* no valid sense data */
307 if (scsi_sense_is_deferred(&sshdr))
310 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
311 sdev->scsi_dh_data->scsi_dh->check_sense) {
314 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
315 if (rc != SCSI_RETURN_NOT_HANDLED)
317 /* handler does not care. Drop down to default handling */
321 * Previous logic looked for FILEMARK, EOM or ILI which are
322 * mainly associated with tapes and returned SUCCESS.
324 if (sshdr.response_code == 0x70) {
326 if (scmd->sense_buffer[2] & 0xe0)
330 * descriptor format: look for "stream commands sense data
331 * descriptor" (see SSC-3). Assume single sense data
332 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
334 if ((sshdr.additional_length > 3) &&
335 (scmd->sense_buffer[8] == 0x4) &&
336 (scmd->sense_buffer[11] & 0xe0))
340 switch (sshdr.sense_key) {
343 case RECOVERED_ERROR:
344 return /* soft_error */ SUCCESS;
346 case ABORTED_COMMAND:
347 if (sshdr.asc == 0x10) /* DIF */
354 * if we are expecting a cc/ua because of a bus reset that we
355 * performed, treat this just as a retry. otherwise this is
356 * information that we should pass up to the upper-level driver
357 * so that we can deal with it there.
359 if (scmd->device->expecting_cc_ua) {
360 scmd->device->expecting_cc_ua = 0;
364 * if the device is in the process of becoming ready, we
367 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
370 * if the device is not started, we need to wake
371 * the error handler to start the motor
373 if (scmd->device->allow_restart &&
374 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
378 /* these three are not supported */
380 case VOLUME_OVERFLOW:
385 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
386 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
387 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
393 if (scmd->device->retry_hwerror)
398 case ILLEGAL_REQUEST:
407 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
408 * @scmd: SCSI cmd to examine.
411 * This is *only* called when we are examining the status of commands
412 * queued during error recovery. the main difference here is that we
413 * don't allow for the possibility of retries here, and we are a lot
414 * more restrictive about what we consider acceptable.
416 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
419 * first check the host byte, to see if there is anything in there
420 * that would indicate what we need to do.
422 if (host_byte(scmd->result) == DID_RESET) {
424 * rats. we are already in the error handler, so we now
425 * get to try and figure out what to do next. if the sense
426 * is valid, we have a pretty good idea of what to do.
427 * if not, we mark it as FAILED.
429 return scsi_check_sense(scmd);
431 if (host_byte(scmd->result) != DID_OK)
435 * next, check the message byte.
437 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
441 * now, check the status byte to see if this indicates
444 switch (status_byte(scmd->result)) {
446 case COMMAND_TERMINATED:
448 case CHECK_CONDITION:
449 return scsi_check_sense(scmd);
451 case INTERMEDIATE_GOOD:
452 case INTERMEDIATE_C_GOOD:
454 * who knows? FIXME(eric)
459 case RESERVATION_CONFLICT:
467 * scsi_eh_done - Completion function for error handling.
468 * @scmd: Cmd that is done.
470 static void scsi_eh_done(struct scsi_cmnd *scmd)
472 struct completion *eh_action;
474 SCSI_LOG_ERROR_RECOVERY(3,
475 printk("%s scmd: %p result: %x\n",
476 __func__, scmd, scmd->result));
478 eh_action = scmd->device->host->eh_action;
484 * scsi_try_host_reset - ask host adapter to reset itself
485 * @scmd: SCSI cmd to send hsot reset.
487 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
492 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
495 if (!scmd->device->host->hostt->eh_host_reset_handler)
498 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
500 if (rtn == SUCCESS) {
501 if (!scmd->device->host->hostt->skip_settle_delay)
502 ssleep(HOST_RESET_SETTLE_TIME);
503 spin_lock_irqsave(scmd->device->host->host_lock, flags);
504 scsi_report_bus_reset(scmd->device->host,
506 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
513 * scsi_try_bus_reset - ask host to perform a bus reset
514 * @scmd: SCSI cmd to send bus reset.
516 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
521 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
524 if (!scmd->device->host->hostt->eh_bus_reset_handler)
527 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
529 if (rtn == SUCCESS) {
530 if (!scmd->device->host->hostt->skip_settle_delay)
531 ssleep(BUS_RESET_SETTLE_TIME);
532 spin_lock_irqsave(scmd->device->host->host_lock, flags);
533 scsi_report_bus_reset(scmd->device->host,
535 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
541 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
544 sdev->expecting_cc_ua = 1;
548 * scsi_try_target_reset - Ask host to perform a target reset
549 * @scmd: SCSI cmd used to send a target reset
552 * There is no timeout for this operation. if this operation is
553 * unreliable for a given host, then the host itself needs to put a
554 * timer on it, and set the host back to a consistent state prior to
557 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
562 if (!scmd->device->host->hostt->eh_target_reset_handler)
565 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
566 if (rtn == SUCCESS) {
567 spin_lock_irqsave(scmd->device->host->host_lock, flags);
568 __starget_for_each_device(scsi_target(scmd->device), NULL,
569 __scsi_report_device_reset);
570 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
577 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
578 * @scmd: SCSI cmd used to send BDR
581 * There is no timeout for this operation. if this operation is
582 * unreliable for a given host, then the host itself needs to put a
583 * timer on it, and set the host back to a consistent state prior to
586 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
590 if (!scmd->device->host->hostt->eh_device_reset_handler)
593 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
595 __scsi_report_device_reset(scmd->device, NULL);
599 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
601 if (!scmd->device->host->hostt->eh_abort_handler)
604 return scmd->device->host->hostt->eh_abort_handler(scmd);
608 * scsi_try_to_abort_cmd - Ask host to abort a running command.
609 * @scmd: SCSI cmd to abort from Lower Level.
612 * This function will not return until the user's completion function
613 * has been called. there is no timeout on this operation. if the
614 * author of the low-level driver wishes this operation to be timed,
615 * they can provide this facility themselves. helper functions in
616 * scsi_error.c can be supplied to make this easier to do.
618 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
621 * scsi_done was called just after the command timed out and before
622 * we had a chance to process it. (db)
624 if (scmd->serial_number == 0)
626 return __scsi_try_to_abort_cmd(scmd);
629 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
631 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
632 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
633 if (scsi_try_target_reset(scmd) != SUCCESS)
634 if (scsi_try_bus_reset(scmd) != SUCCESS)
635 scsi_try_host_reset(scmd);
639 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
640 * @scmd: SCSI command structure to hijack
641 * @ses: structure to save restore information
642 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
643 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
644 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
646 * This function is used to save a scsi command information before re-execution
647 * as part of the error recovery process. If @sense_bytes is 0 the command
648 * sent must be one that does not transfer any data. If @sense_bytes != 0
649 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
650 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
652 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
653 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
655 struct scsi_device *sdev = scmd->device;
658 * We need saved copies of a number of fields - this is because
659 * error handling may need to overwrite these with different values
660 * to run different commands, and once error handling is complete,
661 * we will need to restore these values prior to running the actual
664 ses->cmd_len = scmd->cmd_len;
665 ses->cmnd = scmd->cmnd;
666 ses->data_direction = scmd->sc_data_direction;
667 ses->sdb = scmd->sdb;
668 ses->next_rq = scmd->request->next_rq;
669 ses->result = scmd->result;
670 ses->underflow = scmd->underflow;
671 ses->prot_op = scmd->prot_op;
673 scmd->prot_op = SCSI_PROT_NORMAL;
674 scmd->cmnd = ses->eh_cmnd;
675 memset(scmd->cmnd, 0, BLK_MAX_CDB);
676 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
677 scmd->request->next_rq = NULL;
680 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
682 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
684 scmd->sdb.table.sgl = &ses->sense_sgl;
685 scmd->sc_data_direction = DMA_FROM_DEVICE;
686 scmd->sdb.table.nents = 1;
687 scmd->cmnd[0] = REQUEST_SENSE;
688 scmd->cmnd[4] = scmd->sdb.length;
689 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
691 scmd->sc_data_direction = DMA_NONE;
693 BUG_ON(cmnd_size > BLK_MAX_CDB);
694 memcpy(scmd->cmnd, cmnd, cmnd_size);
695 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
701 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
702 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
703 (sdev->lun << 5 & 0xe0);
706 * Zero the sense buffer. The scsi spec mandates that any
707 * untransferred sense data should be interpreted as being zero.
709 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
711 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
714 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
715 * @scmd: SCSI command structure to restore
716 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
718 * Undo any damage done by above scsi_prep_eh_cmnd().
720 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
723 * Restore original data
725 scmd->cmd_len = ses->cmd_len;
726 scmd->cmnd = ses->cmnd;
727 scmd->sc_data_direction = ses->data_direction;
728 scmd->sdb = ses->sdb;
729 scmd->request->next_rq = ses->next_rq;
730 scmd->result = ses->result;
731 scmd->underflow = ses->underflow;
732 scmd->prot_op = ses->prot_op;
734 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
737 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
738 * @scmd: SCSI command structure to hijack
740 * @cmnd_size: size in bytes of @cmnd
741 * @timeout: timeout for this request
742 * @sense_bytes: size of sense data to copy or 0
744 * This function is used to send a scsi command down to a target device
745 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
748 * SUCCESS or FAILED or NEEDS_RETRY
750 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
751 int cmnd_size, int timeout, unsigned sense_bytes)
753 struct scsi_device *sdev = scmd->device;
754 struct Scsi_Host *shost = sdev->host;
755 DECLARE_COMPLETION_ONSTACK(done);
756 unsigned long timeleft;
758 struct scsi_eh_save ses;
761 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
762 shost->eh_action = &done;
764 spin_lock_irqsave(shost->host_lock, flags);
766 shost->hostt->queuecommand(scmd, scsi_eh_done);
767 spin_unlock_irqrestore(shost->host_lock, flags);
769 timeleft = wait_for_completion_timeout(&done, timeout);
771 shost->eh_action = NULL;
773 scsi_log_completion(scmd, SUCCESS);
775 SCSI_LOG_ERROR_RECOVERY(3,
776 printk("%s: scmd: %p, timeleft: %ld\n",
777 __func__, scmd, timeleft));
780 * If there is time left scsi_eh_done got called, and we will
781 * examine the actual status codes to see whether the command
782 * actually did complete normally, else tell the host to forget
783 * about this command.
786 rtn = scsi_eh_completed_normally(scmd);
787 SCSI_LOG_ERROR_RECOVERY(3,
788 printk("%s: scsi_eh_completed_normally %x\n",
801 scsi_abort_eh_cmnd(scmd);
805 scsi_eh_restore_cmnd(scmd, &ses);
810 * scsi_request_sense - Request sense data from a particular target.
811 * @scmd: SCSI cmd for request sense.
814 * Some hosts automatically obtain this information, others require
815 * that we obtain it on our own. This function will *not* return until
816 * the command either times out, or it completes.
818 static int scsi_request_sense(struct scsi_cmnd *scmd)
820 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
824 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
825 * @scmd: Original SCSI cmd that eh has finished.
826 * @done_q: Queue for processed commands.
829 * We don't want to use the normal command completion while we are are
830 * still handling errors - it may cause other commands to be queued,
831 * and that would disturb what we are doing. Thus we really want to
832 * keep a list of pending commands for final completion, and once we
833 * are ready to leave error handling we handle completion for real.
835 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
837 scmd->device->host->host_failed--;
839 list_move_tail(&scmd->eh_entry, done_q);
841 EXPORT_SYMBOL(scsi_eh_finish_cmd);
844 * scsi_eh_get_sense - Get device sense data.
845 * @work_q: Queue of commands to process.
846 * @done_q: Queue of processed commands.
849 * See if we need to request sense information. if so, then get it
850 * now, so we have a better idea of what to do.
853 * This has the unfortunate side effect that if a shost adapter does
854 * not automatically request sense information, we end up shutting
855 * it down before we request it.
857 * All drivers should request sense information internally these days,
858 * so for now all I have to say is tough noogies if you end up in here.
860 * XXX: Long term this code should go away, but that needs an audit of
863 int scsi_eh_get_sense(struct list_head *work_q,
864 struct list_head *done_q)
866 struct scsi_cmnd *scmd, *next;
869 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
870 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
871 SCSI_SENSE_VALID(scmd))
874 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
875 "%s: requesting sense\n",
877 rtn = scsi_request_sense(scmd);
881 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
882 " result %x\n", scmd,
884 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
886 rtn = scsi_decide_disposition(scmd);
889 * if the result was normal, then just pass it along to the
893 /* we don't want this command reissued, just
894 * finished with the sense data, so set
895 * retries to the max allowed to ensure it
896 * won't get reissued */
897 scmd->retries = scmd->allowed;
898 else if (rtn != NEEDS_RETRY)
901 scsi_eh_finish_cmd(scmd, done_q);
904 return list_empty(work_q);
906 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
909 * scsi_eh_tur - Send TUR to device.
910 * @scmd: &scsi_cmnd to send TUR
913 * 0 - Device is ready. 1 - Device NOT ready.
915 static int scsi_eh_tur(struct scsi_cmnd *scmd)
917 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
918 int retry_cnt = 1, rtn;
921 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
923 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
924 __func__, scmd, rtn));
939 * scsi_eh_abort_cmds - abort pending commands.
940 * @work_q: &list_head for pending commands.
941 * @done_q: &list_head for processed commands.
944 * Try and see whether or not it makes sense to try and abort the
945 * running command. This only works out to be the case if we have one
946 * command that has timed out. If the command simply failed, it makes
947 * no sense to try and abort the command, since as far as the shost
948 * adapter is concerned, it isn't running.
950 static int scsi_eh_abort_cmds(struct list_head *work_q,
951 struct list_head *done_q)
953 struct scsi_cmnd *scmd, *next;
956 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
957 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
959 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
960 "0x%p\n", current->comm,
962 rtn = scsi_try_to_abort_cmd(scmd);
963 if (rtn == SUCCESS) {
964 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
965 if (!scsi_device_online(scmd->device) ||
966 !scsi_eh_tur(scmd)) {
967 scsi_eh_finish_cmd(scmd, done_q);
971 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
978 return list_empty(work_q);
982 * scsi_eh_try_stu - Send START_UNIT to device.
983 * @scmd: &scsi_cmnd to send START_UNIT
986 * 0 - Device is ready. 1 - Device NOT ready.
988 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
990 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
992 if (scmd->device->allow_restart) {
993 int i, rtn = NEEDS_RETRY;
995 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
996 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
997 scmd->device->timeout, 0);
1007 * scsi_eh_stu - send START_UNIT if needed
1008 * @shost: &scsi host being recovered.
1009 * @work_q: &list_head for pending commands.
1010 * @done_q: &list_head for processed commands.
1013 * If commands are failing due to not ready, initializing command required,
1014 * try revalidating the device, which will end up sending a start unit.
1016 static int scsi_eh_stu(struct Scsi_Host *shost,
1017 struct list_head *work_q,
1018 struct list_head *done_q)
1020 struct scsi_cmnd *scmd, *stu_scmd, *next;
1021 struct scsi_device *sdev;
1023 shost_for_each_device(sdev, shost) {
1025 list_for_each_entry(scmd, work_q, eh_entry)
1026 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1027 scsi_check_sense(scmd) == FAILED ) {
1035 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1036 " 0x%p\n", current->comm, sdev));
1038 if (!scsi_eh_try_stu(stu_scmd)) {
1039 if (!scsi_device_online(sdev) ||
1040 !scsi_eh_tur(stu_scmd)) {
1041 list_for_each_entry_safe(scmd, next,
1043 if (scmd->device == sdev)
1044 scsi_eh_finish_cmd(scmd, done_q);
1048 SCSI_LOG_ERROR_RECOVERY(3,
1049 printk("%s: START_UNIT failed to sdev:"
1050 " 0x%p\n", current->comm, sdev));
1054 return list_empty(work_q);
1059 * scsi_eh_bus_device_reset - send bdr if needed
1060 * @shost: scsi host being recovered.
1061 * @work_q: &list_head for pending commands.
1062 * @done_q: &list_head for processed commands.
1065 * Try a bus device reset. Still, look to see whether we have multiple
1066 * devices that are jammed or not - if we have multiple devices, it
1067 * makes no sense to try bus_device_reset - we really would need to try
1068 * a bus_reset instead.
1070 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1071 struct list_head *work_q,
1072 struct list_head *done_q)
1074 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1075 struct scsi_device *sdev;
1078 shost_for_each_device(sdev, shost) {
1080 list_for_each_entry(scmd, work_q, eh_entry)
1081 if (scmd->device == sdev) {
1089 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1090 " 0x%p\n", current->comm,
1092 rtn = scsi_try_bus_device_reset(bdr_scmd);
1093 if (rtn == SUCCESS) {
1094 if (!scsi_device_online(sdev) ||
1095 !scsi_eh_tur(bdr_scmd)) {
1096 list_for_each_entry_safe(scmd, next,
1098 if (scmd->device == sdev)
1099 scsi_eh_finish_cmd(scmd,
1104 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1112 return list_empty(work_q);
1116 * scsi_eh_target_reset - send target reset if needed
1117 * @shost: scsi host being recovered.
1118 * @work_q: &list_head for pending commands.
1119 * @done_q: &list_head for processed commands.
1122 * Try a target reset.
1124 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1125 struct list_head *work_q,
1126 struct list_head *done_q)
1128 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1132 for (id = 0; id <= shost->max_id; id++) {
1134 list_for_each_entry(scmd, work_q, eh_entry) {
1135 if (id == scmd_id(scmd)) {
1143 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1145 current->comm, id));
1146 rtn = scsi_try_target_reset(tgtr_scmd);
1147 if (rtn == SUCCESS) {
1148 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1149 if (id == scmd_id(scmd))
1150 if (!scsi_device_online(scmd->device) ||
1151 !scsi_eh_tur(tgtr_scmd))
1152 scsi_eh_finish_cmd(scmd,
1156 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1159 current->comm, id));
1162 return list_empty(work_q);
1166 * scsi_eh_bus_reset - send a bus reset
1167 * @shost: &scsi host being recovered.
1168 * @work_q: &list_head for pending commands.
1169 * @done_q: &list_head for processed commands.
1171 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1172 struct list_head *work_q,
1173 struct list_head *done_q)
1175 struct scsi_cmnd *scmd, *chan_scmd, *next;
1176 unsigned int channel;
1180 * we really want to loop over the various channels, and do this on
1181 * a channel by channel basis. we should also check to see if any
1182 * of the failed commands are on soft_reset devices, and if so, skip
1186 for (channel = 0; channel <= shost->max_channel; channel++) {
1188 list_for_each_entry(scmd, work_q, eh_entry) {
1189 if (channel == scmd_channel(scmd)) {
1193 * FIXME add back in some support for
1194 * soft_reset devices.
1201 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1202 " %d\n", current->comm,
1204 rtn = scsi_try_bus_reset(chan_scmd);
1205 if (rtn == SUCCESS) {
1206 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1207 if (channel == scmd_channel(scmd))
1208 if (!scsi_device_online(scmd->device) ||
1210 scsi_eh_finish_cmd(scmd,
1214 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1215 " failed chan: %d\n",
1220 return list_empty(work_q);
1224 * scsi_eh_host_reset - send a host reset
1225 * @work_q: list_head for processed commands.
1226 * @done_q: list_head for processed commands.
1228 static int scsi_eh_host_reset(struct list_head *work_q,
1229 struct list_head *done_q)
1231 struct scsi_cmnd *scmd, *next;
1234 if (!list_empty(work_q)) {
1235 scmd = list_entry(work_q->next,
1236 struct scsi_cmnd, eh_entry);
1238 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1241 rtn = scsi_try_host_reset(scmd);
1242 if (rtn == SUCCESS) {
1243 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1244 if (!scsi_device_online(scmd->device) ||
1245 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1247 scsi_eh_finish_cmd(scmd, done_q);
1250 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1255 return list_empty(work_q);
1259 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1260 * @work_q: list_head for processed commands.
1261 * @done_q: list_head for processed commands.
1263 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1264 struct list_head *done_q)
1266 struct scsi_cmnd *scmd, *next;
1268 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1269 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1270 "not ready after error recovery\n");
1271 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1272 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1274 * FIXME: Handle lost cmds.
1277 scsi_eh_finish_cmd(scmd, done_q);
1283 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1284 * @scmd: SCSI cmd to examine.
1287 * This is *only* called when we are examining the status after sending
1288 * out the actual data command. any commands that are queued for error
1289 * recovery (e.g. test_unit_ready) do *not* come through here.
1291 * When this routine returns failed, it means the error handler thread
1292 * is woken. In cases where the error code indicates an error that
1293 * doesn't require the error handler read (i.e. we don't need to
1294 * abort/reset), this function should return SUCCESS.
1296 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1301 * if the device is offline, then we clearly just pass the result back
1302 * up to the top level.
1304 if (!scsi_device_online(scmd->device)) {
1305 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1312 * first check the host byte, to see if there is anything in there
1313 * that would indicate what we need to do.
1315 switch (host_byte(scmd->result)) {
1316 case DID_PASSTHROUGH:
1318 * no matter what, pass this through to the upper layer.
1319 * nuke this special code so that it looks like we are saying
1322 scmd->result &= 0xff00ffff;
1326 * looks good. drop through, and check the next byte.
1329 case DID_NO_CONNECT:
1330 case DID_BAD_TARGET:
1333 * note - this means that we just report the status back
1334 * to the top level driver, not that we actually think
1335 * that it indicates SUCCESS.
1339 * when the low level driver returns did_soft_error,
1340 * it is responsible for keeping an internal retry counter
1341 * in order to avoid endless loops (db)
1343 * actually this is a bug in this function here. we should
1344 * be mindful of the maximum number of retries specified
1345 * and not get stuck in a loop.
1347 case DID_SOFT_ERROR:
1353 return ADD_TO_MLQUEUE;
1356 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1357 status_byte(scmd->result) == RESERVATION_CONFLICT)
1359 * execute reservation conflict processing code
1370 * when we scan the bus, we get timeout messages for
1371 * these commands if there is no device available.
1372 * other hosts report did_no_connect for the same thing.
1374 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1375 scmd->cmnd[0] == INQUIRY)) {
1387 * next, check the message byte.
1389 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1393 * check the status byte to see if this indicates anything special.
1395 switch (status_byte(scmd->result)) {
1398 * the case of trying to send too many commands to a
1399 * tagged queueing device.
1403 * device can't talk to us at the moment. Should only
1404 * occur (SAM-3) when the task queue is empty, so will cause
1405 * the empty queue handling to trigger a stall in the
1408 return ADD_TO_MLQUEUE;
1410 case COMMAND_TERMINATED:
1413 case CHECK_CONDITION:
1414 rtn = scsi_check_sense(scmd);
1415 if (rtn == NEEDS_RETRY)
1417 /* if rtn == FAILED, we have no sense information;
1418 * returning FAILED will wake the error handler thread
1419 * to collect the sense and redo the decide
1422 case CONDITION_GOOD:
1423 case INTERMEDIATE_GOOD:
1424 case INTERMEDIATE_C_GOOD:
1427 * who knows? FIXME(eric)
1431 case RESERVATION_CONFLICT:
1432 sdev_printk(KERN_INFO, scmd->device,
1433 "reservation conflict\n");
1434 return SUCCESS; /* causes immediate i/o error */
1442 /* we requeue for retry because the error was retryable, and
1443 * the request was not marked fast fail. Note that above,
1444 * even if the request is marked fast fail, we still requeue
1445 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1446 if ((++scmd->retries) <= scmd->allowed
1447 && !blk_noretry_request(scmd->request)) {
1451 * no more retries - report this one back to upper level.
1458 * scsi_eh_lock_door - Prevent medium removal for the specified device
1459 * @sdev: SCSI device to prevent medium removal
1462 * We must be called from process context; scsi_allocate_request()
1466 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1467 * head of the devices request queue, and continue.
1470 * scsi_allocate_request() may sleep waiting for existing requests to
1471 * be processed. However, since we haven't kicked off any request
1472 * processing for this host, this may deadlock.
1474 * If scsi_allocate_request() fails for what ever reason, we
1475 * completely forget to lock the door.
1477 static void scsi_eh_lock_door(struct scsi_device *sdev)
1479 unsigned char cmnd[MAX_COMMAND_SIZE];
1481 cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1485 cmnd[4] = SCSI_REMOVAL_PREVENT;
1488 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
1489 5, NULL, NULL, GFP_KERNEL);
1494 * scsi_restart_operations - restart io operations to the specified host.
1495 * @shost: Host we are restarting.
1498 * When we entered the error handler, we blocked all further i/o to
1499 * this device. we need to 'reverse' this process.
1501 static void scsi_restart_operations(struct Scsi_Host *shost)
1503 struct scsi_device *sdev;
1504 unsigned long flags;
1507 * If the door was locked, we need to insert a door lock request
1508 * onto the head of the SCSI request queue for the device. There
1509 * is no point trying to lock the door of an off-line device.
1511 shost_for_each_device(sdev, shost) {
1512 if (scsi_device_online(sdev) && sdev->locked)
1513 scsi_eh_lock_door(sdev);
1517 * next free up anything directly waiting upon the host. this
1518 * will be requests for character device operations, and also for
1519 * ioctls to queued block devices.
1521 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1524 spin_lock_irqsave(shost->host_lock, flags);
1525 if (scsi_host_set_state(shost, SHOST_RUNNING))
1526 if (scsi_host_set_state(shost, SHOST_CANCEL))
1527 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1528 spin_unlock_irqrestore(shost->host_lock, flags);
1530 wake_up(&shost->host_wait);
1533 * finally we need to re-initiate requests that may be pending. we will
1534 * have had everything blocked while error handling is taking place, and
1535 * now that error recovery is done, we will need to ensure that these
1536 * requests are started.
1538 scsi_run_host_queues(shost);
1542 * scsi_eh_ready_devs - check device ready state and recover if not.
1543 * @shost: host to be recovered.
1544 * @work_q: &list_head for pending commands.
1545 * @done_q: &list_head for processed commands.
1547 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1548 struct list_head *work_q,
1549 struct list_head *done_q)
1551 if (!scsi_eh_stu(shost, work_q, done_q))
1552 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1553 if (!scsi_eh_target_reset(shost, work_q, done_q))
1554 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1555 if (!scsi_eh_host_reset(work_q, done_q))
1556 scsi_eh_offline_sdevs(work_q,
1559 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1562 * scsi_eh_flush_done_q - finish processed commands or retry them.
1563 * @done_q: list_head of processed commands.
1565 void scsi_eh_flush_done_q(struct list_head *done_q)
1567 struct scsi_cmnd *scmd, *next;
1569 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1570 list_del_init(&scmd->eh_entry);
1571 if (scsi_device_online(scmd->device) &&
1572 !blk_noretry_request(scmd->request) &&
1573 (++scmd->retries <= scmd->allowed)) {
1574 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1578 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1581 * If just we got sense for the device (called
1582 * scsi_eh_get_sense), scmd->result is already
1583 * set, do not set DRIVER_TIMEOUT.
1586 scmd->result |= (DRIVER_TIMEOUT << 24);
1587 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1589 current->comm, scmd));
1590 scsi_finish_command(scmd);
1594 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1597 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1598 * @shost: Host to unjam.
1601 * When we come in here, we *know* that all commands on the bus have
1602 * either completed, failed or timed out. we also know that no further
1603 * commands are being sent to the host, so things are relatively quiet
1604 * and we have freedom to fiddle with things as we wish.
1606 * This is only the *default* implementation. it is possible for
1607 * individual drivers to supply their own version of this function, and
1608 * if the maintainer wishes to do this, it is strongly suggested that
1609 * this function be taken as a template and modified. this function
1610 * was designed to correctly handle problems for about 95% of the
1611 * different cases out there, and it should always provide at least a
1612 * reasonable amount of error recovery.
1614 * Any command marked 'failed' or 'timeout' must eventually have
1615 * scsi_finish_cmd() called for it. we do all of the retry stuff
1616 * here, so when we restart the host after we return it should have an
1619 static void scsi_unjam_host(struct Scsi_Host *shost)
1621 unsigned long flags;
1622 LIST_HEAD(eh_work_q);
1623 LIST_HEAD(eh_done_q);
1625 spin_lock_irqsave(shost->host_lock, flags);
1626 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1627 spin_unlock_irqrestore(shost->host_lock, flags);
1629 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1631 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1632 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1633 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1635 scsi_eh_flush_done_q(&eh_done_q);
1639 * scsi_error_handler - SCSI error handler thread
1640 * @data: Host for which we are running.
1643 * This is the main error handling loop. This is run as a kernel thread
1644 * for every SCSI host and handles all error handling activity.
1646 int scsi_error_handler(void *data)
1648 struct Scsi_Host *shost = data;
1651 * We use TASK_INTERRUPTIBLE so that the thread is not
1652 * counted against the load average as a running process.
1653 * We never actually get interrupted because kthread_run
1654 * disables singal delivery for the created thread.
1656 set_current_state(TASK_INTERRUPTIBLE);
1657 while (!kthread_should_stop()) {
1658 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1659 shost->host_failed != shost->host_busy) {
1660 SCSI_LOG_ERROR_RECOVERY(1,
1661 printk("Error handler scsi_eh_%d sleeping\n",
1664 set_current_state(TASK_INTERRUPTIBLE);
1668 __set_current_state(TASK_RUNNING);
1669 SCSI_LOG_ERROR_RECOVERY(1,
1670 printk("Error handler scsi_eh_%d waking up\n",
1674 * We have a host that is failing for some reason. Figure out
1675 * what we need to do to get it up and online again (if we can).
1676 * If we fail, we end up taking the thing offline.
1678 if (shost->transportt->eh_strategy_handler)
1679 shost->transportt->eh_strategy_handler(shost);
1681 scsi_unjam_host(shost);
1684 * Note - if the above fails completely, the action is to take
1685 * individual devices offline and flush the queue of any
1686 * outstanding requests that may have been pending. When we
1687 * restart, we restart any I/O to any other devices on the bus
1688 * which are still online.
1690 scsi_restart_operations(shost);
1691 set_current_state(TASK_INTERRUPTIBLE);
1693 __set_current_state(TASK_RUNNING);
1695 SCSI_LOG_ERROR_RECOVERY(1,
1696 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1697 shost->ehandler = NULL;
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.
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(sdev))
1728 __scsi_report_device_reset(sdev, NULL);
1731 EXPORT_SYMBOL(scsi_report_bus_reset);
1734 * Function: scsi_report_device_reset()
1736 * Purpose: Utility function used by low-level drivers to report that
1737 * they have observed a device reset on the device being handled.
1739 * Arguments: shost - Host in question
1740 * channel - channel on which reset was observed
1741 * target - target on which reset was observed
1745 * Lock status: Host lock must be held
1747 * Notes: This only needs to be called if the reset is one which
1748 * originates from an unknown location. Resets originated
1749 * by the mid-level itself don't need to call this, but there
1750 * should be no harm.
1752 * The main purpose of this is to make sure that a CHECK_CONDITION
1753 * is properly treated.
1755 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1757 struct scsi_device *sdev;
1759 __shost_for_each_device(sdev, shost) {
1760 if (channel == sdev_channel(sdev) &&
1761 target == sdev_id(sdev))
1762 __scsi_report_device_reset(sdev, NULL);
1765 EXPORT_SYMBOL(scsi_report_device_reset);
1768 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1773 * Function: scsi_reset_provider
1775 * Purpose: Send requested reset to a bus or device at any phase.
1777 * Arguments: device - device to send reset to
1778 * flag - reset type (see scsi.h)
1780 * Returns: SUCCESS/FAILURE.
1782 * Notes: This is used by the SCSI Generic driver to provide
1783 * Bus/Device reset capability.
1786 scsi_reset_provider(struct scsi_device *dev, int flag)
1788 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1789 struct Scsi_Host *shost = dev->host;
1791 unsigned long flags;
1794 blk_rq_init(NULL, &req);
1795 scmd->request = &req;
1796 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1798 scmd->cmnd = req.cmd;
1800 scmd->scsi_done = scsi_reset_provider_done_command;
1801 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1805 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1807 init_timer(&scmd->eh_timeout);
1809 spin_lock_irqsave(shost->host_lock, flags);
1810 shost->tmf_in_progress = 1;
1811 spin_unlock_irqrestore(shost->host_lock, flags);
1814 case SCSI_TRY_RESET_DEVICE:
1815 rtn = scsi_try_bus_device_reset(scmd);
1819 case SCSI_TRY_RESET_TARGET:
1820 rtn = scsi_try_target_reset(scmd);
1824 case SCSI_TRY_RESET_BUS:
1825 rtn = scsi_try_bus_reset(scmd);
1829 case SCSI_TRY_RESET_HOST:
1830 rtn = scsi_try_host_reset(scmd);
1836 spin_lock_irqsave(shost->host_lock, flags);
1837 shost->tmf_in_progress = 0;
1838 spin_unlock_irqrestore(shost->host_lock, flags);
1841 * be sure to wake up anyone who was sleeping or had their queue
1842 * suspended while we performed the TMF.
1844 SCSI_LOG_ERROR_RECOVERY(3,
1845 printk("%s: waking up host to restart after TMF\n",
1848 wake_up(&shost->host_wait);
1850 scsi_run_host_queues(shost);
1852 scsi_next_command(scmd);
1855 EXPORT_SYMBOL(scsi_reset_provider);
1858 * scsi_normalize_sense - normalize main elements from either fixed or
1859 * descriptor sense data format into a common format.
1861 * @sense_buffer: byte array containing sense data returned by device
1862 * @sb_len: number of valid bytes in sense_buffer
1863 * @sshdr: pointer to instance of structure that common
1864 * elements are written to.
1867 * The "main elements" from sense data are: response_code, sense_key,
1868 * asc, ascq and additional_length (only for descriptor format).
1870 * Typically this function can be called after a device has
1871 * responded to a SCSI command with the CHECK_CONDITION status.
1874 * 1 if valid sense data information found, else 0;
1876 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1877 struct scsi_sense_hdr *sshdr)
1879 if (!sense_buffer || !sb_len)
1882 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1884 sshdr->response_code = (sense_buffer[0] & 0x7f);
1886 if (!scsi_sense_valid(sshdr))
1889 if (sshdr->response_code >= 0x72) {
1894 sshdr->sense_key = (sense_buffer[1] & 0xf);
1896 sshdr->asc = sense_buffer[2];
1898 sshdr->ascq = sense_buffer[3];
1900 sshdr->additional_length = sense_buffer[7];
1906 sshdr->sense_key = (sense_buffer[2] & 0xf);
1908 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1909 sb_len : (sense_buffer[7] + 8);
1911 sshdr->asc = sense_buffer[12];
1913 sshdr->ascq = sense_buffer[13];
1919 EXPORT_SYMBOL(scsi_normalize_sense);
1921 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1922 struct scsi_sense_hdr *sshdr)
1924 return scsi_normalize_sense(cmd->sense_buffer,
1925 SCSI_SENSE_BUFFERSIZE, sshdr);
1927 EXPORT_SYMBOL(scsi_command_normalize_sense);
1930 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1931 * @sense_buffer: byte array of descriptor format sense data
1932 * @sb_len: number of valid bytes in sense_buffer
1933 * @desc_type: value of descriptor type to find
1934 * (e.g. 0 -> information)
1937 * only valid when sense data is in descriptor format
1940 * pointer to start of (first) descriptor if found else NULL
1942 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1945 int add_sen_len, add_len, desc_len, k;
1948 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1950 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1952 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1953 add_sen_len : (sb_len - 8);
1954 descp = &sense_buffer[8];
1955 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1957 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1958 desc_len = add_len + 2;
1959 if (descp[0] == desc_type)
1961 if (add_len < 0) // short descriptor ??
1966 EXPORT_SYMBOL(scsi_sense_desc_find);
1969 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1970 * @sense_buffer: byte array of sense data
1971 * @sb_len: number of valid bytes in sense_buffer
1972 * @info_out: pointer to 64 integer where 8 or 4 byte information
1973 * field will be placed if found.
1976 * 1 if information field found, 0 if not found.
1978 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1987 switch (sense_buffer[0] & 0x7f) {
1990 if (sense_buffer[0] & 0x80) {
1991 *info_out = (sense_buffer[3] << 24) +
1992 (sense_buffer[4] << 16) +
1993 (sense_buffer[5] << 8) + sense_buffer[6];
1999 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2001 if (ucp && (0xa == ucp[1])) {
2003 for (j = 0; j < 8; ++j) {
2016 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2019 * scsi_build_sense_buffer - build sense data in a buffer
2020 * @desc: Sense format (non zero == descriptor format,
2021 * 0 == fixed format)
2022 * @buf: Where to build sense data
2024 * @asc: Additional sense code
2025 * @ascq: Additional sense code qualifier
2028 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2031 buf[0] = 0x72; /* descriptor, current */
2037 buf[0] = 0x70; /* fixed, current */
2044 EXPORT_SYMBOL(scsi_build_sense_buffer);