2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_eh.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_cmnd.h>
41 #include "../scsi/scsi_transport_api.h"
43 #include <linux/libata.h>
48 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
49 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
50 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
53 /* Waiting in ->prereset can never be reliable. It's sometimes nice
54 * to wait there but it can't be depended upon; otherwise, we wouldn't
55 * be resetting. Just give it enough time for most drives to spin up.
58 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
59 ATA_EH_FASTDRAIN_INTERVAL = 3 * HZ,
62 /* The following table determines how we sequence resets. Each entry
63 * represents timeout for that try. The first try can be soft or
64 * hardreset. All others are hardreset if available. In most cases
65 * the first reset w/ 10sec timeout should succeed. Following entries
66 * are mostly for error handling, hotplug and retarded devices.
68 static const unsigned long ata_eh_reset_timeouts[] = {
69 10 * HZ, /* most drives spin up by 10sec */
70 10 * HZ, /* > 99% working drives spin up before 20sec */
71 35 * HZ, /* give > 30 secs of idleness for retarded devices */
72 5 * HZ, /* and sweet one last chance */
73 /* > 1 min has elapsed, give up */
76 static void __ata_port_freeze(struct ata_port *ap);
77 static void ata_eh_finish(struct ata_port *ap);
79 static void ata_eh_handle_port_suspend(struct ata_port *ap);
80 static void ata_eh_handle_port_resume(struct ata_port *ap);
82 static void ata_eh_handle_port_suspend(struct ata_port *ap)
85 static void ata_eh_handle_port_resume(struct ata_port *ap)
87 #endif /* CONFIG_PM */
89 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
92 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
93 ATA_EH_DESC_LEN - ehi->desc_len,
98 * __ata_ehi_push_desc - push error description without adding separator
100 * @fmt: printf format string
102 * Format string according to @fmt and append it to @ehi->desc.
105 * spin_lock_irqsave(host lock)
107 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
112 __ata_ehi_pushv_desc(ehi, fmt, args);
117 * ata_ehi_push_desc - push error description with separator
119 * @fmt: printf format string
121 * Format string according to @fmt and append it to @ehi->desc.
122 * If @ehi->desc is not empty, ", " is added in-between.
125 * spin_lock_irqsave(host lock)
127 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
132 __ata_ehi_push_desc(ehi, ", ");
135 __ata_ehi_pushv_desc(ehi, fmt, args);
140 * ata_ehi_clear_desc - clean error description
146 * spin_lock_irqsave(host lock)
148 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
154 static void ata_ering_record(struct ata_ering *ering, int is_io,
155 unsigned int err_mask)
157 struct ata_ering_entry *ent;
162 ering->cursor %= ATA_ERING_SIZE;
164 ent = &ering->ring[ering->cursor];
166 ent->err_mask = err_mask;
167 ent->timestamp = get_jiffies_64();
170 static void ata_ering_clear(struct ata_ering *ering)
172 memset(ering, 0, sizeof(*ering));
175 static int ata_ering_map(struct ata_ering *ering,
176 int (*map_fn)(struct ata_ering_entry *, void *),
180 struct ata_ering_entry *ent;
184 ent = &ering->ring[idx];
187 rc = map_fn(ent, arg);
190 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
191 } while (idx != ering->cursor);
196 static unsigned int ata_eh_dev_action(struct ata_device *dev)
198 struct ata_eh_context *ehc = &dev->link->eh_context;
200 return ehc->i.action | ehc->i.dev_action[dev->devno];
203 static void ata_eh_clear_action(struct ata_device *dev,
204 struct ata_eh_info *ehi, unsigned int action)
209 ehi->action &= ~action;
210 for (i = 0; i < ATA_MAX_DEVICES; i++)
211 ehi->dev_action[i] &= ~action;
213 /* doesn't make sense for port-wide EH actions */
214 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
216 /* break ehi->action into ehi->dev_action */
217 if (ehi->action & action) {
218 for (i = 0; i < ATA_MAX_DEVICES; i++)
219 ehi->dev_action[i] |= ehi->action & action;
220 ehi->action &= ~action;
223 /* turn off the specified per-dev action */
224 ehi->dev_action[dev->devno] &= ~action;
229 * ata_scsi_timed_out - SCSI layer time out callback
230 * @cmd: timed out SCSI command
232 * Handles SCSI layer timeout. We race with normal completion of
233 * the qc for @cmd. If the qc is already gone, we lose and let
234 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
235 * timed out and EH should be invoked. Prevent ata_qc_complete()
236 * from finishing it by setting EH_SCHEDULED and return
239 * TODO: kill this function once old EH is gone.
242 * Called from timer context
245 * EH_HANDLED or EH_NOT_HANDLED
247 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
249 struct Scsi_Host *host = cmd->device->host;
250 struct ata_port *ap = ata_shost_to_port(host);
252 struct ata_queued_cmd *qc;
253 enum scsi_eh_timer_return ret;
257 if (ap->ops->error_handler) {
258 ret = EH_NOT_HANDLED;
263 spin_lock_irqsave(ap->lock, flags);
264 qc = ata_qc_from_tag(ap, ap->link.active_tag);
266 WARN_ON(qc->scsicmd != cmd);
267 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
268 qc->err_mask |= AC_ERR_TIMEOUT;
269 ret = EH_NOT_HANDLED;
271 spin_unlock_irqrestore(ap->lock, flags);
274 DPRINTK("EXIT, ret=%d\n", ret);
279 * ata_scsi_error - SCSI layer error handler callback
280 * @host: SCSI host on which error occurred
282 * Handles SCSI-layer-thrown error events.
285 * Inherited from SCSI layer (none, can sleep)
290 void ata_scsi_error(struct Scsi_Host *host)
292 struct ata_port *ap = ata_shost_to_port(host);
293 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
298 /* synchronize with port task */
299 ata_port_flush_task(ap);
301 /* synchronize with host lock and sort out timeouts */
303 /* For new EH, all qcs are finished in one of three ways -
304 * normal completion, error completion, and SCSI timeout.
305 * Both cmpletions can race against SCSI timeout. When normal
306 * completion wins, the qc never reaches EH. When error
307 * completion wins, the qc has ATA_QCFLAG_FAILED set.
309 * When SCSI timeout wins, things are a bit more complex.
310 * Normal or error completion can occur after the timeout but
311 * before this point. In such cases, both types of
312 * completions are honored. A scmd is determined to have
313 * timed out iff its associated qc is active and not failed.
315 if (ap->ops->error_handler) {
316 struct scsi_cmnd *scmd, *tmp;
319 spin_lock_irqsave(ap->lock, flags);
321 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
322 struct ata_queued_cmd *qc;
324 for (i = 0; i < ATA_MAX_QUEUE; i++) {
325 qc = __ata_qc_from_tag(ap, i);
326 if (qc->flags & ATA_QCFLAG_ACTIVE &&
331 if (i < ATA_MAX_QUEUE) {
332 /* the scmd has an associated qc */
333 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
334 /* which hasn't failed yet, timeout */
335 qc->err_mask |= AC_ERR_TIMEOUT;
336 qc->flags |= ATA_QCFLAG_FAILED;
340 /* Normal completion occurred after
341 * SCSI timeout but before this point.
342 * Successfully complete it.
344 scmd->retries = scmd->allowed;
345 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
349 /* If we have timed out qcs. They belong to EH from
350 * this point but the state of the controller is
351 * unknown. Freeze the port to make sure the IRQ
352 * handler doesn't diddle with those qcs. This must
353 * be done atomically w.r.t. setting QCFLAG_FAILED.
356 __ata_port_freeze(ap);
358 spin_unlock_irqrestore(ap->lock, flags);
360 spin_unlock_wait(ap->lock);
363 /* invoke error handler */
364 if (ap->ops->error_handler) {
365 /* kill fast drain timer */
366 del_timer_sync(&ap->fastdrain_timer);
368 /* process port resume request */
369 ata_eh_handle_port_resume(ap);
371 /* fetch & clear EH info */
372 spin_lock_irqsave(ap->lock, flags);
374 memset(&ap->link.eh_context, 0, sizeof(ap->link.eh_context));
375 ap->link.eh_context.i = ap->link.eh_info;
376 memset(&ap->link.eh_info, 0, sizeof(ap->link.eh_info));
378 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
379 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
381 spin_unlock_irqrestore(ap->lock, flags);
383 /* invoke EH, skip if unloading or suspended */
384 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
385 ap->ops->error_handler(ap);
389 /* process port suspend request */
390 ata_eh_handle_port_suspend(ap);
392 /* Exception might have happend after ->error_handler
393 * recovered the port but before this point. Repeat
396 spin_lock_irqsave(ap->lock, flags);
398 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
400 ata_port_printk(ap, KERN_INFO,
401 "EH pending after completion, "
402 "repeating EH (cnt=%d)\n", repeat_cnt);
403 spin_unlock_irqrestore(ap->lock, flags);
406 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
407 "tries, giving up\n", ATA_EH_MAX_REPEAT);
408 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
411 /* this run is complete, make sure EH info is clear */
412 memset(&ap->link.eh_info, 0, sizeof(ap->link.eh_info));
414 /* Clear host_eh_scheduled while holding ap->lock such
415 * that if exception occurs after this point but
416 * before EH completion, SCSI midlayer will
419 host->host_eh_scheduled = 0;
421 spin_unlock_irqrestore(ap->lock, flags);
423 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
424 ap->ops->eng_timeout(ap);
427 /* finish or retry handled scmd's and clean up */
428 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
430 scsi_eh_flush_done_q(&ap->eh_done_q);
433 spin_lock_irqsave(ap->lock, flags);
435 if (ap->pflags & ATA_PFLAG_LOADING)
436 ap->pflags &= ~ATA_PFLAG_LOADING;
437 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
438 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
440 if (ap->pflags & ATA_PFLAG_RECOVERED)
441 ata_port_printk(ap, KERN_INFO, "EH complete\n");
443 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
445 /* tell wait_eh that we're done */
446 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
447 wake_up_all(&ap->eh_wait_q);
449 spin_unlock_irqrestore(ap->lock, flags);
455 * ata_port_wait_eh - Wait for the currently pending EH to complete
456 * @ap: Port to wait EH for
458 * Wait until the currently pending EH is complete.
461 * Kernel thread context (may sleep).
463 void ata_port_wait_eh(struct ata_port *ap)
469 spin_lock_irqsave(ap->lock, flags);
471 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
472 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
473 spin_unlock_irqrestore(ap->lock, flags);
475 spin_lock_irqsave(ap->lock, flags);
477 finish_wait(&ap->eh_wait_q, &wait);
479 spin_unlock_irqrestore(ap->lock, flags);
481 /* make sure SCSI EH is complete */
482 if (scsi_host_in_recovery(ap->scsi_host)) {
489 * ata_qc_timeout - Handle timeout of queued command
490 * @qc: Command that timed out
492 * Some part of the kernel (currently, only the SCSI layer)
493 * has noticed that the active command on port @ap has not
494 * completed after a specified length of time. Handle this
495 * condition by disabling DMA (if necessary) and completing
496 * transactions, with error if necessary.
498 * This also handles the case of the "lost interrupt", where
499 * for some reason (possibly hardware bug, possibly driver bug)
500 * an interrupt was not delivered to the driver, even though the
501 * transaction completed successfully.
503 * TODO: kill this function once old EH is gone.
506 * Inherited from SCSI layer (none, can sleep)
508 static void ata_qc_timeout(struct ata_queued_cmd *qc)
510 struct ata_port *ap = qc->ap;
511 u8 host_stat = 0, drv_stat;
516 ap->hsm_task_state = HSM_ST_IDLE;
518 spin_lock_irqsave(ap->lock, flags);
520 switch (qc->tf.protocol) {
523 case ATA_PROT_ATAPI_DMA:
524 host_stat = ap->ops->bmdma_status(ap);
526 /* before we do anything else, clear DMA-Start bit */
527 ap->ops->bmdma_stop(qc);
533 drv_stat = ata_chk_status(ap);
535 /* ack bmdma irq events */
536 ap->ops->irq_clear(ap);
538 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
539 "stat 0x%x host_stat 0x%x\n",
540 qc->tf.command, drv_stat, host_stat);
542 /* complete taskfile transaction */
543 qc->err_mask |= AC_ERR_TIMEOUT;
547 spin_unlock_irqrestore(ap->lock, flags);
549 ata_eh_qc_complete(qc);
555 * ata_eng_timeout - Handle timeout of queued command
556 * @ap: Port on which timed-out command is active
558 * Some part of the kernel (currently, only the SCSI layer)
559 * has noticed that the active command on port @ap has not
560 * completed after a specified length of time. Handle this
561 * condition by disabling DMA (if necessary) and completing
562 * transactions, with error if necessary.
564 * This also handles the case of the "lost interrupt", where
565 * for some reason (possibly hardware bug, possibly driver bug)
566 * an interrupt was not delivered to the driver, even though the
567 * transaction completed successfully.
569 * TODO: kill this function once old EH is gone.
572 * Inherited from SCSI layer (none, can sleep)
574 void ata_eng_timeout(struct ata_port *ap)
578 ata_qc_timeout(ata_qc_from_tag(ap, ap->link.active_tag));
583 static int ata_eh_nr_in_flight(struct ata_port *ap)
588 /* count only non-internal commands */
589 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
590 if (ata_qc_from_tag(ap, tag))
596 void ata_eh_fastdrain_timerfn(unsigned long arg)
598 struct ata_port *ap = (void *)arg;
602 spin_lock_irqsave(ap->lock, flags);
604 cnt = ata_eh_nr_in_flight(ap);
610 if (cnt == ap->fastdrain_cnt) {
613 /* No progress during the last interval, tag all
614 * in-flight qcs as timed out and freeze the port.
616 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
617 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
619 qc->err_mask |= AC_ERR_TIMEOUT;
624 /* some qcs have finished, give it another chance */
625 ap->fastdrain_cnt = cnt;
626 ap->fastdrain_timer.expires =
627 jiffies + ATA_EH_FASTDRAIN_INTERVAL;
628 add_timer(&ap->fastdrain_timer);
632 spin_unlock_irqrestore(ap->lock, flags);
636 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
637 * @ap: target ATA port
638 * @fastdrain: activate fast drain
640 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
641 * is non-zero and EH wasn't pending before. Fast drain ensures
642 * that EH kicks in in timely manner.
645 * spin_lock_irqsave(host lock)
647 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
651 /* already scheduled? */
652 if (ap->pflags & ATA_PFLAG_EH_PENDING)
655 ap->pflags |= ATA_PFLAG_EH_PENDING;
660 /* do we have in-flight qcs? */
661 cnt = ata_eh_nr_in_flight(ap);
665 /* activate fast drain */
666 ap->fastdrain_cnt = cnt;
667 ap->fastdrain_timer.expires = jiffies + ATA_EH_FASTDRAIN_INTERVAL;
668 add_timer(&ap->fastdrain_timer);
672 * ata_qc_schedule_eh - schedule qc for error handling
673 * @qc: command to schedule error handling for
675 * Schedule error handling for @qc. EH will kick in as soon as
676 * other commands are drained.
679 * spin_lock_irqsave(host lock)
681 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
683 struct ata_port *ap = qc->ap;
685 WARN_ON(!ap->ops->error_handler);
687 qc->flags |= ATA_QCFLAG_FAILED;
688 ata_eh_set_pending(ap, 1);
690 /* The following will fail if timeout has already expired.
691 * ata_scsi_error() takes care of such scmds on EH entry.
692 * Note that ATA_QCFLAG_FAILED is unconditionally set after
693 * this function completes.
695 scsi_req_abort_cmd(qc->scsicmd);
699 * ata_port_schedule_eh - schedule error handling without a qc
700 * @ap: ATA port to schedule EH for
702 * Schedule error handling for @ap. EH will kick in as soon as
703 * all commands are drained.
706 * spin_lock_irqsave(host lock)
708 void ata_port_schedule_eh(struct ata_port *ap)
710 WARN_ON(!ap->ops->error_handler);
712 if (ap->pflags & ATA_PFLAG_INITIALIZING)
715 ata_eh_set_pending(ap, 1);
716 scsi_schedule_eh(ap->scsi_host);
718 DPRINTK("port EH scheduled\n");
722 * ata_port_abort - abort all qc's on the port
723 * @ap: ATA port to abort qc's for
725 * Abort all active qc's of @ap and schedule EH.
728 * spin_lock_irqsave(host lock)
731 * Number of aborted qc's.
733 int ata_port_abort(struct ata_port *ap)
735 int tag, nr_aborted = 0;
737 WARN_ON(!ap->ops->error_handler);
739 /* we're gonna abort all commands, no need for fast drain */
740 ata_eh_set_pending(ap, 0);
742 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
743 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
746 qc->flags |= ATA_QCFLAG_FAILED;
753 ata_port_schedule_eh(ap);
759 * __ata_port_freeze - freeze port
760 * @ap: ATA port to freeze
762 * This function is called when HSM violation or some other
763 * condition disrupts normal operation of the port. Frozen port
764 * is not allowed to perform any operation until the port is
765 * thawed, which usually follows a successful reset.
767 * ap->ops->freeze() callback can be used for freezing the port
768 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
769 * port cannot be frozen hardware-wise, the interrupt handler
770 * must ack and clear interrupts unconditionally while the port
774 * spin_lock_irqsave(host lock)
776 static void __ata_port_freeze(struct ata_port *ap)
778 WARN_ON(!ap->ops->error_handler);
783 ap->pflags |= ATA_PFLAG_FROZEN;
785 DPRINTK("ata%u port frozen\n", ap->print_id);
789 * ata_port_freeze - abort & freeze port
790 * @ap: ATA port to freeze
792 * Abort and freeze @ap.
795 * spin_lock_irqsave(host lock)
798 * Number of aborted commands.
800 int ata_port_freeze(struct ata_port *ap)
804 WARN_ON(!ap->ops->error_handler);
806 nr_aborted = ata_port_abort(ap);
807 __ata_port_freeze(ap);
813 * ata_eh_freeze_port - EH helper to freeze port
814 * @ap: ATA port to freeze
821 void ata_eh_freeze_port(struct ata_port *ap)
825 if (!ap->ops->error_handler)
828 spin_lock_irqsave(ap->lock, flags);
829 __ata_port_freeze(ap);
830 spin_unlock_irqrestore(ap->lock, flags);
834 * ata_port_thaw_port - EH helper to thaw port
835 * @ap: ATA port to thaw
837 * Thaw frozen port @ap.
842 void ata_eh_thaw_port(struct ata_port *ap)
846 if (!ap->ops->error_handler)
849 spin_lock_irqsave(ap->lock, flags);
851 ap->pflags &= ~ATA_PFLAG_FROZEN;
856 spin_unlock_irqrestore(ap->lock, flags);
858 DPRINTK("ata%u port thawed\n", ap->print_id);
861 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
866 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
868 struct ata_port *ap = qc->ap;
869 struct scsi_cmnd *scmd = qc->scsicmd;
872 spin_lock_irqsave(ap->lock, flags);
873 qc->scsidone = ata_eh_scsidone;
874 __ata_qc_complete(qc);
875 WARN_ON(ata_tag_valid(qc->tag));
876 spin_unlock_irqrestore(ap->lock, flags);
878 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
882 * ata_eh_qc_complete - Complete an active ATA command from EH
883 * @qc: Command to complete
885 * Indicate to the mid and upper layers that an ATA command has
886 * completed. To be used from EH.
888 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
890 struct scsi_cmnd *scmd = qc->scsicmd;
891 scmd->retries = scmd->allowed;
892 __ata_eh_qc_complete(qc);
896 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
897 * @qc: Command to retry
899 * Indicate to the mid and upper layers that an ATA command
900 * should be retried. To be used from EH.
902 * SCSI midlayer limits the number of retries to scmd->allowed.
903 * scmd->retries is decremented for commands which get retried
904 * due to unrelated failures (qc->err_mask is zero).
906 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
908 struct scsi_cmnd *scmd = qc->scsicmd;
909 if (!qc->err_mask && scmd->retries)
911 __ata_eh_qc_complete(qc);
915 * ata_eh_detach_dev - detach ATA device
916 * @dev: ATA device to detach
923 static void ata_eh_detach_dev(struct ata_device *dev)
925 struct ata_port *ap = dev->link->ap;
928 ata_dev_disable(dev);
930 spin_lock_irqsave(ap->lock, flags);
932 dev->flags &= ~ATA_DFLAG_DETACH;
934 if (ata_scsi_offline_dev(dev)) {
935 dev->flags |= ATA_DFLAG_DETACHED;
936 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
939 /* clear per-dev EH actions */
940 ata_eh_clear_action(dev, &dev->link->eh_info, ATA_EH_PERDEV_MASK);
941 ata_eh_clear_action(dev, &dev->link->eh_context.i, ATA_EH_PERDEV_MASK);
943 spin_unlock_irqrestore(ap->lock, flags);
947 * ata_eh_about_to_do - about to perform eh_action
948 * @ap: target ATA port
949 * @dev: target ATA dev for per-dev action (can be NULL)
950 * @action: action about to be performed
952 * Called just before performing EH actions to clear related bits
953 * in @ap->link.eh_info such that eh actions are not
954 * unnecessarily repeated.
959 static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
963 struct ata_eh_info *ehi = &ap->link.eh_info;
964 struct ata_eh_context *ehc = &ap->link.eh_context;
966 spin_lock_irqsave(ap->lock, flags);
968 /* Reset is represented by combination of actions and EHI
969 * flags. Suck in all related bits before clearing eh_info to
970 * avoid losing requested action.
972 if (action & ATA_EH_RESET_MASK) {
973 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
974 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
976 /* make sure all reset actions are cleared & clear EHI flags */
977 action |= ATA_EH_RESET_MASK;
978 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
981 ata_eh_clear_action(dev, ehi, action);
983 if (!(ehc->i.flags & ATA_EHI_QUIET))
984 ap->pflags |= ATA_PFLAG_RECOVERED;
986 spin_unlock_irqrestore(ap->lock, flags);
990 * ata_eh_done - EH action complete
991 * @ap: target ATA port
992 * @dev: target ATA dev for per-dev action (can be NULL)
993 * @action: action just completed
995 * Called right after performing EH actions to clear related bits
996 * in @ap->link.eh_context.
1001 static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
1002 unsigned int action)
1004 struct ata_eh_context *ehc = &ap->link.eh_context;
1006 /* if reset is complete, clear all reset actions & reset modifier */
1007 if (action & ATA_EH_RESET_MASK) {
1008 action |= ATA_EH_RESET_MASK;
1009 ehc->i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
1012 ata_eh_clear_action(dev, &ehc->i, action);
1016 * ata_err_string - convert err_mask to descriptive string
1017 * @err_mask: error mask to convert to string
1019 * Convert @err_mask to descriptive string. Errors are
1020 * prioritized according to severity and only the most severe
1021 * error is reported.
1027 * Descriptive string for @err_mask
1029 static const char * ata_err_string(unsigned int err_mask)
1031 if (err_mask & AC_ERR_HOST_BUS)
1032 return "host bus error";
1033 if (err_mask & AC_ERR_ATA_BUS)
1034 return "ATA bus error";
1035 if (err_mask & AC_ERR_TIMEOUT)
1037 if (err_mask & AC_ERR_HSM)
1038 return "HSM violation";
1039 if (err_mask & AC_ERR_SYSTEM)
1040 return "internal error";
1041 if (err_mask & AC_ERR_MEDIA)
1042 return "media error";
1043 if (err_mask & AC_ERR_INVALID)
1044 return "invalid argument";
1045 if (err_mask & AC_ERR_DEV)
1046 return "device error";
1047 return "unknown error";
1051 * ata_read_log_page - read a specific log page
1052 * @dev: target device
1053 * @page: page to read
1054 * @buf: buffer to store read page
1055 * @sectors: number of sectors to read
1057 * Read log page using READ_LOG_EXT command.
1060 * Kernel thread context (may sleep).
1063 * 0 on success, AC_ERR_* mask otherwise.
1065 static unsigned int ata_read_log_page(struct ata_device *dev,
1066 u8 page, void *buf, unsigned int sectors)
1068 struct ata_taskfile tf;
1069 unsigned int err_mask;
1071 DPRINTK("read log page - page %d\n", page);
1073 ata_tf_init(dev, &tf);
1074 tf.command = ATA_CMD_READ_LOG_EXT;
1077 tf.hob_nsect = sectors >> 8;
1078 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1079 tf.protocol = ATA_PROT_PIO;
1081 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1082 buf, sectors * ATA_SECT_SIZE);
1084 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1089 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1090 * @dev: Device to read log page 10h from
1091 * @tag: Resulting tag of the failed command
1092 * @tf: Resulting taskfile registers of the failed command
1094 * Read log page 10h to obtain NCQ error details and clear error
1098 * Kernel thread context (may sleep).
1101 * 0 on success, -errno otherwise.
1103 static int ata_eh_read_log_10h(struct ata_device *dev,
1104 int *tag, struct ata_taskfile *tf)
1106 u8 *buf = dev->link->ap->sector_buf;
1107 unsigned int err_mask;
1111 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1116 for (i = 0; i < ATA_SECT_SIZE; i++)
1119 ata_dev_printk(dev, KERN_WARNING,
1120 "invalid checksum 0x%x on log page 10h\n", csum);
1125 *tag = buf[0] & 0x1f;
1127 tf->command = buf[2];
1128 tf->feature = buf[3];
1132 tf->device = buf[7];
1133 tf->hob_lbal = buf[8];
1134 tf->hob_lbam = buf[9];
1135 tf->hob_lbah = buf[10];
1136 tf->nsect = buf[12];
1137 tf->hob_nsect = buf[13];
1143 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1144 * @dev: device to perform REQUEST_SENSE to
1145 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1147 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1148 * SENSE. This function is EH helper.
1151 * Kernel thread context (may sleep).
1154 * 0 on success, AC_ERR_* mask on failure
1156 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
1158 struct ata_device *dev = qc->dev;
1159 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
1160 struct ata_port *ap = dev->link->ap;
1161 struct ata_taskfile tf;
1162 u8 cdb[ATAPI_CDB_LEN];
1164 DPRINTK("ATAPI request sense\n");
1166 /* FIXME: is this needed? */
1167 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1169 /* initialize sense_buf with the error register,
1170 * for the case where they are -not- overwritten
1172 sense_buf[0] = 0x70;
1173 sense_buf[2] = qc->result_tf.feature >> 4;
1175 /* some devices time out if garbage left in tf */
1176 ata_tf_init(dev, &tf);
1178 memset(cdb, 0, ATAPI_CDB_LEN);
1179 cdb[0] = REQUEST_SENSE;
1180 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1182 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1183 tf.command = ATA_CMD_PACKET;
1185 /* is it pointless to prefer PIO for "safety reasons"? */
1186 if (ap->flags & ATA_FLAG_PIO_DMA) {
1187 tf.protocol = ATA_PROT_ATAPI_DMA;
1188 tf.feature |= ATAPI_PKT_DMA;
1190 tf.protocol = ATA_PROT_ATAPI;
1191 tf.lbam = (8 * 1024) & 0xff;
1192 tf.lbah = (8 * 1024) >> 8;
1195 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1196 sense_buf, SCSI_SENSE_BUFFERSIZE);
1200 * ata_eh_analyze_serror - analyze SError for a failed port
1201 * @ap: ATA port to analyze SError for
1203 * Analyze SError if available and further determine cause of
1209 static void ata_eh_analyze_serror(struct ata_port *ap)
1211 struct ata_eh_context *ehc = &ap->link.eh_context;
1212 u32 serror = ehc->i.serror;
1213 unsigned int err_mask = 0, action = 0;
1215 if (serror & SERR_PERSISTENT) {
1216 err_mask |= AC_ERR_ATA_BUS;
1217 action |= ATA_EH_HARDRESET;
1220 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1221 err_mask |= AC_ERR_ATA_BUS;
1222 action |= ATA_EH_SOFTRESET;
1224 if (serror & SERR_PROTOCOL) {
1225 err_mask |= AC_ERR_HSM;
1226 action |= ATA_EH_SOFTRESET;
1228 if (serror & SERR_INTERNAL) {
1229 err_mask |= AC_ERR_SYSTEM;
1230 action |= ATA_EH_HARDRESET;
1232 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1233 ata_ehi_hotplugged(&ehc->i);
1235 ehc->i.err_mask |= err_mask;
1236 ehc->i.action |= action;
1240 * ata_eh_analyze_ncq_error - analyze NCQ error
1241 * @ap: ATA port to analyze NCQ error for
1243 * Read log page 10h, determine the offending qc and acquire
1244 * error status TF. For NCQ device errors, all LLDDs have to do
1245 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1249 * Kernel thread context (may sleep).
1251 static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1253 struct ata_eh_context *ehc = &ap->link.eh_context;
1254 struct ata_device *dev = ap->link.device;
1255 struct ata_queued_cmd *qc;
1256 struct ata_taskfile tf;
1259 /* if frozen, we can't do much */
1260 if (ap->pflags & ATA_PFLAG_FROZEN)
1263 /* is it NCQ device error? */
1264 if (!ap->link.sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1267 /* has LLDD analyzed already? */
1268 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1269 qc = __ata_qc_from_tag(ap, tag);
1271 if (!(qc->flags & ATA_QCFLAG_FAILED))
1278 /* okay, this error is ours */
1279 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1281 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1282 "(errno=%d)\n", rc);
1286 if (!(ap->link.sactive & (1 << tag))) {
1287 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1288 "inactive tag %d\n", tag);
1292 /* we've got the perpetrator, condemn it */
1293 qc = __ata_qc_from_tag(ap, tag);
1294 memcpy(&qc->result_tf, &tf, sizeof(tf));
1295 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1296 ehc->i.err_mask &= ~AC_ERR_DEV;
1300 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1301 * @qc: qc to analyze
1302 * @tf: Taskfile registers to analyze
1304 * Analyze taskfile of @qc and further determine cause of
1305 * failure. This function also requests ATAPI sense data if
1309 * Kernel thread context (may sleep).
1312 * Determined recovery action
1314 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1315 const struct ata_taskfile *tf)
1317 unsigned int tmp, action = 0;
1318 u8 stat = tf->command, err = tf->feature;
1320 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1321 qc->err_mask |= AC_ERR_HSM;
1322 return ATA_EH_SOFTRESET;
1325 if (stat & (ATA_ERR | ATA_DF))
1326 qc->err_mask |= AC_ERR_DEV;
1330 switch (qc->dev->class) {
1333 qc->err_mask |= AC_ERR_ATA_BUS;
1335 qc->err_mask |= AC_ERR_MEDIA;
1337 qc->err_mask |= AC_ERR_INVALID;
1341 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1342 tmp = atapi_eh_request_sense(qc);
1344 /* ATA_QCFLAG_SENSE_VALID is used to
1345 * tell atapi_qc_complete() that sense
1346 * data is already valid.
1348 * TODO: interpret sense data and set
1349 * appropriate err_mask.
1351 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1353 qc->err_mask |= tmp;
1357 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1358 action |= ATA_EH_SOFTRESET;
1363 static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
1365 if (err_mask & AC_ERR_ATA_BUS)
1368 if (err_mask & AC_ERR_TIMEOUT)
1372 if (err_mask & AC_ERR_HSM)
1375 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1382 struct speed_down_verdict_arg {
1387 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1389 struct speed_down_verdict_arg *arg = void_arg;
1390 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
1392 if (ent->timestamp < arg->since)
1395 arg->nr_errors[cat]++;
1400 * ata_eh_speed_down_verdict - Determine speed down verdict
1401 * @dev: Device of interest
1403 * This function examines error ring of @dev and determines
1404 * whether NCQ needs to be turned off, transfer speed should be
1405 * stepped down, or falling back to PIO is necessary.
1407 * Cat-1 is ATA_BUS error for any command.
1409 * Cat-2 is TIMEOUT for any command or HSM violation for known
1410 * supported commands.
1412 * Cat-3 is is unclassified DEV error for known supported
1415 * NCQ needs to be turned off if there have been more than 3
1416 * Cat-2 + Cat-3 errors during last 10 minutes.
1418 * Speed down is necessary if there have been more than 3 Cat-1 +
1419 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1421 * Falling back to PIO mode is necessary if there have been more
1422 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1425 * Inherited from caller.
1428 * OR of ATA_EH_SPDN_* flags.
1430 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1432 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1433 u64 j64 = get_jiffies_64();
1434 struct speed_down_verdict_arg arg;
1435 unsigned int verdict = 0;
1437 /* scan past 10 mins of error history */
1438 memset(&arg, 0, sizeof(arg));
1439 arg.since = j64 - min(j64, j10mins);
1440 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1442 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1443 verdict |= ATA_EH_SPDN_NCQ_OFF;
1444 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1445 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1447 /* scan past 3 mins of error history */
1448 memset(&arg, 0, sizeof(arg));
1449 arg.since = j64 - min(j64, j5mins);
1450 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1452 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1453 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1459 * ata_eh_speed_down - record error and speed down if necessary
1460 * @dev: Failed device
1461 * @is_io: Did the device fail during normal IO?
1462 * @err_mask: err_mask of the error
1464 * Record error and examine error history to determine whether
1465 * adjusting transmission speed is necessary. It also sets
1466 * transmission limits appropriately if such adjustment is
1470 * Kernel thread context (may sleep).
1473 * Determined recovery action.
1475 static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1476 unsigned int err_mask)
1478 unsigned int verdict;
1479 unsigned int action = 0;
1481 /* don't bother if Cat-0 error */
1482 if (ata_eh_categorize_error(is_io, err_mask) == 0)
1485 /* record error and determine whether speed down is necessary */
1486 ata_ering_record(&dev->ering, is_io, err_mask);
1487 verdict = ata_eh_speed_down_verdict(dev);
1490 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1491 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1492 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1493 dev->flags |= ATA_DFLAG_NCQ_OFF;
1494 ata_dev_printk(dev, KERN_WARNING,
1495 "NCQ disabled due to excessive errors\n");
1500 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1501 /* speed down SATA link speed if possible */
1502 if (sata_down_spd_limit(dev->link->ap) == 0) {
1503 action |= ATA_EH_HARDRESET;
1507 /* lower transfer mode */
1508 if (dev->spdn_cnt < 2) {
1509 static const int dma_dnxfer_sel[] =
1510 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1511 static const int pio_dnxfer_sel[] =
1512 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1515 if (dev->xfer_shift != ATA_SHIFT_PIO)
1516 sel = dma_dnxfer_sel[dev->spdn_cnt];
1518 sel = pio_dnxfer_sel[dev->spdn_cnt];
1522 if (ata_down_xfermask_limit(dev, sel) == 0) {
1523 action |= ATA_EH_SOFTRESET;
1529 /* Fall back to PIO? Slowing down to PIO is meaningless for
1530 * SATA. Consider it only for PATA.
1532 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1533 (dev->link->ap->cbl != ATA_CBL_SATA) &&
1534 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1535 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1537 action |= ATA_EH_SOFTRESET;
1544 /* device has been slowed down, blow error history */
1545 ata_ering_clear(&dev->ering);
1550 * ata_eh_autopsy - analyze error and determine recovery action
1551 * @ap: ATA port to perform autopsy on
1553 * Analyze why @ap failed and determine which recovery action is
1554 * needed. This function also sets more detailed AC_ERR_* values
1555 * and fills sense data for ATAPI CHECK SENSE.
1558 * Kernel thread context (may sleep).
1560 static void ata_eh_autopsy(struct ata_port *ap)
1562 struct ata_eh_context *ehc = &ap->link.eh_context;
1563 unsigned int all_err_mask = 0;
1570 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1573 /* obtain and analyze SError */
1574 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1576 ehc->i.serror |= serror;
1577 ata_eh_analyze_serror(ap);
1578 } else if (rc != -EOPNOTSUPP) {
1579 /* SError read failed, force hardreset and probing */
1580 ata_ehi_schedule_probe(&ehc->i);
1581 ehc->i.action |= ATA_EH_HARDRESET;
1582 ehc->i.err_mask |= AC_ERR_OTHER;
1585 /* analyze NCQ failure */
1586 ata_eh_analyze_ncq_error(ap);
1588 /* any real error trumps AC_ERR_OTHER */
1589 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1590 ehc->i.err_mask &= ~AC_ERR_OTHER;
1592 all_err_mask |= ehc->i.err_mask;
1594 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1595 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1597 if (!(qc->flags & ATA_QCFLAG_FAILED))
1600 /* inherit upper level err_mask */
1601 qc->err_mask |= ehc->i.err_mask;
1604 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1606 /* DEV errors are probably spurious in case of ATA_BUS error */
1607 if (qc->err_mask & AC_ERR_ATA_BUS)
1608 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1611 /* any real error trumps unknown error */
1612 if (qc->err_mask & ~AC_ERR_OTHER)
1613 qc->err_mask &= ~AC_ERR_OTHER;
1615 /* SENSE_VALID trumps dev/unknown error and revalidation */
1616 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1617 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1618 ehc->i.action &= ~ATA_EH_REVALIDATE;
1621 /* accumulate error info */
1622 ehc->i.dev = qc->dev;
1623 all_err_mask |= qc->err_mask;
1624 if (qc->flags & ATA_QCFLAG_IO)
1628 /* enforce default EH actions */
1629 if (ap->pflags & ATA_PFLAG_FROZEN ||
1630 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1631 ehc->i.action |= ATA_EH_SOFTRESET;
1632 else if (all_err_mask)
1633 ehc->i.action |= ATA_EH_REVALIDATE;
1635 /* if we have offending qcs and the associated failed device */
1638 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1641 /* perform per-dev EH action only on the offending device */
1642 ehc->i.dev_action[ehc->i.dev->devno] |=
1643 ehc->i.action & ATA_EH_PERDEV_MASK;
1644 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1651 * ata_eh_report - report error handling to user
1652 * @ap: ATA port EH is going on
1654 * Report EH to user.
1659 static void ata_eh_report(struct ata_port *ap)
1661 struct ata_eh_context *ehc = &ap->link.eh_context;
1662 const char *frozen, *desc;
1663 int tag, nr_failed = 0;
1666 if (ehc->i.desc[0] != '\0')
1669 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1670 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1672 if (!(qc->flags & ATA_QCFLAG_FAILED))
1674 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1680 if (!nr_failed && !ehc->i.err_mask)
1684 if (ap->pflags & ATA_PFLAG_FROZEN)
1688 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1689 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1690 ehc->i.err_mask, ap->link.sactive,
1691 ehc->i.serror, ehc->i.action, frozen);
1693 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
1695 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1696 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1697 ehc->i.err_mask, ap->link.sactive,
1698 ehc->i.serror, ehc->i.action, frozen);
1700 ata_port_printk(ap, KERN_ERR, "%s\n", desc);
1703 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1704 static const char *dma_str[] = {
1705 [DMA_BIDIRECTIONAL] = "bidi",
1706 [DMA_TO_DEVICE] = "out",
1707 [DMA_FROM_DEVICE] = "in",
1710 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1711 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
1713 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1716 ata_dev_printk(qc->dev, KERN_ERR,
1717 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1718 "tag %d cdb 0x%x data %u %s\n "
1719 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1720 "Emask 0x%x (%s)%s\n",
1721 cmd->command, cmd->feature, cmd->nsect,
1722 cmd->lbal, cmd->lbam, cmd->lbah,
1723 cmd->hob_feature, cmd->hob_nsect,
1724 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
1725 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
1726 dma_str[qc->dma_dir],
1727 res->command, res->feature, res->nsect,
1728 res->lbal, res->lbam, res->lbah,
1729 res->hob_feature, res->hob_nsect,
1730 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1731 res->device, qc->err_mask, ata_err_string(qc->err_mask),
1732 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
1736 static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1737 unsigned int *classes, unsigned long deadline)
1741 for (i = 0; i < ATA_MAX_DEVICES; i++)
1742 classes[i] = ATA_DEV_UNKNOWN;
1744 rc = reset(ap, classes, deadline);
1748 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1749 * is complete and convert all ATA_DEV_UNKNOWN to
1752 for (i = 0; i < ATA_MAX_DEVICES; i++)
1753 if (classes[i] != ATA_DEV_UNKNOWN)
1756 if (i < ATA_MAX_DEVICES)
1757 for (i = 0; i < ATA_MAX_DEVICES; i++)
1758 if (classes[i] == ATA_DEV_UNKNOWN)
1759 classes[i] = ATA_DEV_NONE;
1764 static int ata_eh_followup_srst_needed(int rc, int classify,
1765 const unsigned int *classes)
1771 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1776 static int ata_eh_reset(struct ata_port *ap, int classify,
1777 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
1778 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1780 struct ata_eh_context *ehc = &ap->link.eh_context;
1781 unsigned int *classes = ehc->classes;
1782 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
1784 unsigned long deadline;
1785 unsigned int action;
1786 ata_reset_fn_t reset;
1789 /* about to reset */
1790 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1792 /* Determine which reset to use and record in ehc->i.action.
1793 * prereset() may examine and modify it.
1795 action = ehc->i.action;
1796 ehc->i.action &= ~ATA_EH_RESET_MASK;
1797 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
1798 !(action & ATA_EH_HARDRESET))))
1799 ehc->i.action |= ATA_EH_SOFTRESET;
1801 ehc->i.action |= ATA_EH_HARDRESET;
1804 rc = prereset(ap, jiffies + ATA_EH_PRERESET_TIMEOUT);
1806 if (rc == -ENOENT) {
1807 ata_port_printk(ap, KERN_DEBUG,
1808 "port disabled. ignoring.\n");
1809 ehc->i.action &= ~ATA_EH_RESET_MASK;
1811 for (i = 0; i < ATA_MAX_DEVICES; i++)
1812 classes[i] = ATA_DEV_NONE;
1816 ata_port_printk(ap, KERN_ERR,
1817 "prereset failed (errno=%d)\n", rc);
1822 /* prereset() might have modified ehc->i.action */
1823 if (ehc->i.action & ATA_EH_HARDRESET)
1825 else if (ehc->i.action & ATA_EH_SOFTRESET)
1828 /* prereset told us not to reset, bang classes and return */
1829 for (i = 0; i < ATA_MAX_DEVICES; i++)
1830 classes[i] = ATA_DEV_NONE;
1835 /* did prereset() screw up? if so, fix up to avoid oopsing */
1844 deadline = jiffies + ata_eh_reset_timeouts[try++];
1846 /* shut up during boot probing */
1848 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1849 reset == softreset ? "soft" : "hard");
1851 /* mark that this EH session started with reset */
1852 if (reset == hardreset)
1853 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
1855 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
1857 rc = ata_do_reset(ap, reset, classes, deadline);
1859 if (reset == hardreset &&
1860 ata_eh_followup_srst_needed(rc, classify, classes)) {
1861 /* okay, let's do follow-up softreset */
1865 ata_port_printk(ap, KERN_ERR,
1866 "follow-up softreset required "
1867 "but no softreset avaliable\n");
1872 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
1873 rc = ata_do_reset(ap, reset, classes, deadline);
1875 if (rc == 0 && classify &&
1876 classes[0] == ATA_DEV_UNKNOWN) {
1877 ata_port_printk(ap, KERN_ERR,
1878 "classification failed\n");
1884 if (rc && try < ARRAY_SIZE(ata_eh_reset_timeouts)) {
1885 unsigned long now = jiffies;
1887 if (time_before(now, deadline)) {
1888 unsigned long delta = deadline - jiffies;
1890 ata_port_printk(ap, KERN_WARNING, "reset failed "
1891 "(errno=%d), retrying in %u secs\n",
1892 rc, (jiffies_to_msecs(delta) + 999) / 1000);
1894 schedule_timeout_uninterruptible(delta);
1898 try == ARRAY_SIZE(ata_eh_reset_timeouts) - 1)
1899 sata_down_spd_limit(ap);
1908 /* After the reset, the device state is PIO 0 and the
1909 * controller state is undefined. Record the mode.
1911 for (i = 0; i < ATA_MAX_DEVICES; i++)
1912 ap->link.device[i].pio_mode = XFER_PIO_0;
1914 /* record current link speed */
1915 if (sata_scr_read(ap, SCR_STATUS, &sstatus) == 0)
1916 ap->link.sata_spd = (sstatus >> 4) & 0xf;
1919 postreset(ap, classes);
1921 /* reset successful, schedule revalidation */
1922 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1923 ehc->i.action |= ATA_EH_REVALIDATE;
1926 /* clear hotplug flag */
1927 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
1931 static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1932 struct ata_device **r_failed_dev)
1934 struct ata_eh_context *ehc = &ap->link.eh_context;
1935 struct ata_device *dev;
1936 unsigned int new_mask = 0;
1937 unsigned long flags;
1942 /* For PATA drive side cable detection to work, IDENTIFY must
1943 * be done backwards such that PDIAG- is released by the slave
1944 * device before the master device is identified.
1946 for (i = ATA_MAX_DEVICES - 1; i >= 0; i--) {
1947 unsigned int action, readid_flags = 0;
1949 dev = &ap->link.device[i];
1950 action = ata_eh_dev_action(dev);
1952 if (ehc->i.flags & ATA_EHI_DID_RESET)
1953 readid_flags |= ATA_READID_POSTRESET;
1955 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
1956 if (ata_port_offline(ap)) {
1961 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
1962 rc = ata_dev_revalidate(dev, readid_flags);
1966 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1968 /* Configuration may have changed, reconfigure
1971 ehc->i.flags |= ATA_EHI_SETMODE;
1973 /* schedule the scsi_rescan_device() here */
1974 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
1975 } else if (dev->class == ATA_DEV_UNKNOWN &&
1976 ehc->tries[dev->devno] &&
1977 ata_class_enabled(ehc->classes[dev->devno])) {
1978 dev->class = ehc->classes[dev->devno];
1980 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1987 /* IDENTIFY was issued to non-existent
1988 * device. No need to reset. Just
1989 * thaw and kill the device.
1991 ata_eh_thaw_port(ap);
1992 dev->class = ATA_DEV_UNKNOWN;
1995 dev->class = ATA_DEV_UNKNOWN;
2001 /* PDIAG- should have been released, ask cable type if post-reset */
2002 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ap->ops->cable_detect)
2003 ap->cbl = ap->ops->cable_detect(ap);
2005 /* Configure new devices forward such that user doesn't see
2006 * device detection messages backwards.
2008 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2009 dev = &ap->link.device[i];
2011 if (!(new_mask & (1 << i)))
2014 ehc->i.flags |= ATA_EHI_PRINTINFO;
2015 rc = ata_dev_configure(dev);
2016 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2020 spin_lock_irqsave(ap->lock, flags);
2021 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2022 spin_unlock_irqrestore(ap->lock, flags);
2024 /* new device discovered, configure xfermode */
2025 ehc->i.flags |= ATA_EHI_SETMODE;
2031 *r_failed_dev = dev;
2032 DPRINTK("EXIT rc=%d\n", rc);
2036 static int ata_port_nr_enabled(struct ata_port *ap)
2040 for (i = 0; i < ATA_MAX_DEVICES; i++)
2041 if (ata_dev_enabled(&ap->link.device[i]))
2046 static int ata_port_nr_vacant(struct ata_port *ap)
2050 for (i = 0; i < ATA_MAX_DEVICES; i++)
2051 if (ap->link.device[i].class == ATA_DEV_UNKNOWN)
2056 static int ata_eh_skip_recovery(struct ata_port *ap)
2058 struct ata_eh_context *ehc = &ap->link.eh_context;
2061 /* thaw frozen port, resume link and recover failed devices */
2062 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
2063 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
2066 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2067 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2068 struct ata_device *dev = &ap->link.device[i];
2070 if (dev->class == ATA_DEV_UNKNOWN &&
2071 ehc->classes[dev->devno] != ATA_DEV_NONE)
2078 static void ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2080 struct ata_port *ap = dev->link->ap;
2081 struct ata_eh_context *ehc = &dev->link->eh_context;
2083 ehc->tries[dev->devno]--;
2087 /* device missing or wrong IDENTIFY data, schedule probing */
2088 ehc->i.probe_mask |= (1 << dev->devno);
2090 /* give it just one more chance */
2091 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2093 if (ehc->tries[dev->devno] == 1) {
2094 /* This is the last chance, better to slow
2095 * down than lose it.
2097 sata_down_spd_limit(ap);
2098 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2102 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2103 /* disable device if it has used up all its chances */
2104 ata_dev_disable(dev);
2106 /* detach if offline */
2107 if (ata_port_offline(ap))
2108 ata_eh_detach_dev(dev);
2110 /* probe if requested */
2111 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2112 !(ehc->did_probe_mask & (1 << dev->devno))) {
2113 ata_eh_detach_dev(dev);
2116 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2117 ehc->did_probe_mask |= (1 << dev->devno);
2118 ehc->i.action |= ATA_EH_SOFTRESET;
2121 /* soft didn't work? be haaaaard */
2122 if (ehc->i.flags & ATA_EHI_DID_RESET)
2123 ehc->i.action |= ATA_EH_HARDRESET;
2125 ehc->i.action |= ATA_EH_SOFTRESET;
2130 * ata_eh_recover - recover host port after error
2131 * @ap: host port to recover
2132 * @prereset: prereset method (can be NULL)
2133 * @softreset: softreset method (can be NULL)
2134 * @hardreset: hardreset method (can be NULL)
2135 * @postreset: postreset method (can be NULL)
2137 * This is the alpha and omega, eum and yang, heart and soul of
2138 * libata exception handling. On entry, actions required to
2139 * recover the port and hotplug requests are recorded in
2140 * eh_context. This function executes all the operations with
2141 * appropriate retrials and fallbacks to resurrect failed
2142 * devices, detach goners and greet newcomers.
2145 * Kernel thread context (may sleep).
2148 * 0 on success, -errno on failure.
2150 static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2151 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2152 ata_postreset_fn_t postreset)
2154 struct ata_eh_context *ehc = &ap->link.eh_context;
2155 struct ata_device *dev;
2160 /* prep for recovery */
2161 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2162 dev = &ap->link.device[i];
2164 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2166 /* collect port action mask recorded in dev actions */
2167 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
2168 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
2170 /* process hotplug request */
2171 if (dev->flags & ATA_DFLAG_DETACH)
2172 ata_eh_detach_dev(dev);
2174 if (!ata_dev_enabled(dev) &&
2175 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2176 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2177 ata_eh_detach_dev(dev);
2179 ehc->did_probe_mask |= (1 << dev->devno);
2180 ehc->i.action |= ATA_EH_SOFTRESET;
2187 /* if UNLOADING, finish immediately */
2188 if (ap->pflags & ATA_PFLAG_UNLOADING)
2191 /* skip EH if possible. */
2192 if (ata_eh_skip_recovery(ap))
2195 for (i = 0; i < ATA_MAX_DEVICES; i++)
2196 ehc->classes[i] = ATA_DEV_UNKNOWN;
2199 if (ehc->i.action & ATA_EH_RESET_MASK) {
2200 ata_eh_freeze_port(ap);
2202 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
2203 softreset, hardreset, postreset);
2205 ata_port_printk(ap, KERN_ERR,
2206 "reset failed, giving up\n");
2210 ata_eh_thaw_port(ap);
2213 /* revalidate existing devices and attach new ones */
2214 rc = ata_eh_revalidate_and_attach(ap, &dev);
2218 /* configure transfer mode if necessary */
2219 if (ehc->i.flags & ATA_EHI_SETMODE) {
2220 rc = ata_set_mode(ap, &dev);
2223 ehc->i.flags &= ~ATA_EHI_SETMODE;
2229 ata_eh_handle_dev_fail(dev, rc);
2231 if (ata_port_nr_enabled(ap)) {
2232 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2233 "devices, retrying in 5 secs\n");
2236 /* no device left, repeat fast */
2244 for (i = 0; i < ATA_MAX_DEVICES; i++)
2245 ata_dev_disable(&ap->link.device[i]);
2248 DPRINTK("EXIT, rc=%d\n", rc);
2253 * ata_eh_finish - finish up EH
2254 * @ap: host port to finish EH for
2256 * Recovery is complete. Clean up EH states and retry or finish
2262 static void ata_eh_finish(struct ata_port *ap)
2266 /* retry or finish qcs */
2267 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2268 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2270 if (!(qc->flags & ATA_QCFLAG_FAILED))
2274 /* FIXME: Once EH migration is complete,
2275 * generate sense data in this function,
2276 * considering both err_mask and tf.
2278 if (qc->err_mask & AC_ERR_INVALID)
2279 ata_eh_qc_complete(qc);
2281 ata_eh_qc_retry(qc);
2283 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2284 ata_eh_qc_complete(qc);
2286 /* feed zero TF to sense generation */
2287 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2288 ata_eh_qc_retry(qc);
2295 * ata_do_eh - do standard error handling
2296 * @ap: host port to handle error for
2297 * @prereset: prereset method (can be NULL)
2298 * @softreset: softreset method (can be NULL)
2299 * @hardreset: hardreset method (can be NULL)
2300 * @postreset: postreset method (can be NULL)
2302 * Perform standard error handling sequence.
2305 * Kernel thread context (may sleep).
2307 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2308 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2309 ata_postreset_fn_t postreset)
2313 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
2319 * ata_eh_handle_port_suspend - perform port suspend operation
2320 * @ap: port to suspend
2325 * Kernel thread context (may sleep).
2327 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2329 unsigned long flags;
2332 /* are we suspending? */
2333 spin_lock_irqsave(ap->lock, flags);
2334 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2335 ap->pm_mesg.event == PM_EVENT_ON) {
2336 spin_unlock_irqrestore(ap->lock, flags);
2339 spin_unlock_irqrestore(ap->lock, flags);
2341 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2343 /* tell ACPI we're suspending */
2344 rc = ata_acpi_on_suspend(ap);
2349 ata_eh_freeze_port(ap);
2351 if (ap->ops->port_suspend)
2352 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2356 spin_lock_irqsave(ap->lock, flags);
2358 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2360 ap->pflags |= ATA_PFLAG_SUSPENDED;
2361 else if (ap->pflags & ATA_PFLAG_FROZEN)
2362 ata_port_schedule_eh(ap);
2364 if (ap->pm_result) {
2365 *ap->pm_result = rc;
2366 ap->pm_result = NULL;
2369 spin_unlock_irqrestore(ap->lock, flags);
2375 * ata_eh_handle_port_resume - perform port resume operation
2376 * @ap: port to resume
2381 * Kernel thread context (may sleep).
2383 static void ata_eh_handle_port_resume(struct ata_port *ap)
2385 unsigned long flags;
2388 /* are we resuming? */
2389 spin_lock_irqsave(ap->lock, flags);
2390 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2391 ap->pm_mesg.event != PM_EVENT_ON) {
2392 spin_unlock_irqrestore(ap->lock, flags);
2395 spin_unlock_irqrestore(ap->lock, flags);
2397 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
2399 if (ap->ops->port_resume)
2400 rc = ap->ops->port_resume(ap);
2402 /* tell ACPI that we're resuming */
2403 ata_acpi_on_resume(ap);
2406 spin_lock_irqsave(ap->lock, flags);
2407 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2408 if (ap->pm_result) {
2409 *ap->pm_result = rc;
2410 ap->pm_result = NULL;
2412 spin_unlock_irqrestore(ap->lock, flags);
2414 #endif /* CONFIG_PM */