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,
61 /* The following table determines how we sequence resets. Each entry
62 * represents timeout for that try. The first try can be soft or
63 * hardreset. All others are hardreset if available. In most cases
64 * the first reset w/ 10sec timeout should succeed. Following entries
65 * are mostly for error handling, hotplug and retarded devices.
67 static const unsigned long ata_eh_reset_timeouts[] = {
68 10 * HZ, /* most drives spin up by 10sec */
69 10 * HZ, /* > 99% working drives spin up before 20sec */
70 35 * HZ, /* give > 30 secs of idleness for retarded devices */
71 5 * HZ, /* and sweet one last chance */
72 /* > 1 min has elapsed, give up */
75 static void __ata_port_freeze(struct ata_port *ap);
76 static void ata_eh_finish(struct ata_port *ap);
78 static void ata_eh_handle_port_suspend(struct ata_port *ap);
79 static void ata_eh_handle_port_resume(struct ata_port *ap);
81 static void ata_eh_handle_port_suspend(struct ata_port *ap)
84 static void ata_eh_handle_port_resume(struct ata_port *ap)
86 #endif /* CONFIG_PM */
88 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
91 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
92 ATA_EH_DESC_LEN - ehi->desc_len,
97 * __ata_ehi_push_desc - push error description without adding separator
99 * @fmt: printf format string
101 * Format string according to @fmt and append it to @ehi->desc.
104 * spin_lock_irqsave(host lock)
106 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
111 __ata_ehi_pushv_desc(ehi, fmt, args);
116 * ata_ehi_push_desc - push error description with separator
118 * @fmt: printf format string
120 * Format string according to @fmt and append it to @ehi->desc.
121 * If @ehi->desc is not empty, ", " is added in-between.
124 * spin_lock_irqsave(host lock)
126 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
131 __ata_ehi_push_desc(ehi, ", ");
134 __ata_ehi_pushv_desc(ehi, fmt, args);
139 * ata_ehi_clear_desc - clean error description
145 * spin_lock_irqsave(host lock)
147 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
153 static void ata_ering_record(struct ata_ering *ering, int is_io,
154 unsigned int err_mask)
156 struct ata_ering_entry *ent;
161 ering->cursor %= ATA_ERING_SIZE;
163 ent = &ering->ring[ering->cursor];
165 ent->err_mask = err_mask;
166 ent->timestamp = get_jiffies_64();
169 static void ata_ering_clear(struct ata_ering *ering)
171 memset(ering, 0, sizeof(*ering));
174 static int ata_ering_map(struct ata_ering *ering,
175 int (*map_fn)(struct ata_ering_entry *, void *),
179 struct ata_ering_entry *ent;
183 ent = &ering->ring[idx];
186 rc = map_fn(ent, arg);
189 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
190 } while (idx != ering->cursor);
195 static unsigned int ata_eh_dev_action(struct ata_device *dev)
197 struct ata_eh_context *ehc = &dev->ap->eh_context;
199 return ehc->i.action | ehc->i.dev_action[dev->devno];
202 static void ata_eh_clear_action(struct ata_device *dev,
203 struct ata_eh_info *ehi, unsigned int action)
208 ehi->action &= ~action;
209 for (i = 0; i < ATA_MAX_DEVICES; i++)
210 ehi->dev_action[i] &= ~action;
212 /* doesn't make sense for port-wide EH actions */
213 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
215 /* break ehi->action into ehi->dev_action */
216 if (ehi->action & action) {
217 for (i = 0; i < ATA_MAX_DEVICES; i++)
218 ehi->dev_action[i] |= ehi->action & action;
219 ehi->action &= ~action;
222 /* turn off the specified per-dev action */
223 ehi->dev_action[dev->devno] &= ~action;
228 * ata_scsi_timed_out - SCSI layer time out callback
229 * @cmd: timed out SCSI command
231 * Handles SCSI layer timeout. We race with normal completion of
232 * the qc for @cmd. If the qc is already gone, we lose and let
233 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
234 * timed out and EH should be invoked. Prevent ata_qc_complete()
235 * from finishing it by setting EH_SCHEDULED and return
238 * TODO: kill this function once old EH is gone.
241 * Called from timer context
244 * EH_HANDLED or EH_NOT_HANDLED
246 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
248 struct Scsi_Host *host = cmd->device->host;
249 struct ata_port *ap = ata_shost_to_port(host);
251 struct ata_queued_cmd *qc;
252 enum scsi_eh_timer_return ret;
256 if (ap->ops->error_handler) {
257 ret = EH_NOT_HANDLED;
262 spin_lock_irqsave(ap->lock, flags);
263 qc = ata_qc_from_tag(ap, ap->active_tag);
265 WARN_ON(qc->scsicmd != cmd);
266 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
267 qc->err_mask |= AC_ERR_TIMEOUT;
268 ret = EH_NOT_HANDLED;
270 spin_unlock_irqrestore(ap->lock, flags);
273 DPRINTK("EXIT, ret=%d\n", ret);
278 * ata_scsi_error - SCSI layer error handler callback
279 * @host: SCSI host on which error occurred
281 * Handles SCSI-layer-thrown error events.
284 * Inherited from SCSI layer (none, can sleep)
289 void ata_scsi_error(struct Scsi_Host *host)
291 struct ata_port *ap = ata_shost_to_port(host);
292 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
297 /* synchronize with port task */
298 ata_port_flush_task(ap);
300 /* synchronize with host lock and sort out timeouts */
302 /* For new EH, all qcs are finished in one of three ways -
303 * normal completion, error completion, and SCSI timeout.
304 * Both cmpletions can race against SCSI timeout. When normal
305 * completion wins, the qc never reaches EH. When error
306 * completion wins, the qc has ATA_QCFLAG_FAILED set.
308 * When SCSI timeout wins, things are a bit more complex.
309 * Normal or error completion can occur after the timeout but
310 * before this point. In such cases, both types of
311 * completions are honored. A scmd is determined to have
312 * timed out iff its associated qc is active and not failed.
314 if (ap->ops->error_handler) {
315 struct scsi_cmnd *scmd, *tmp;
318 spin_lock_irqsave(ap->lock, flags);
320 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
321 struct ata_queued_cmd *qc;
323 for (i = 0; i < ATA_MAX_QUEUE; i++) {
324 qc = __ata_qc_from_tag(ap, i);
325 if (qc->flags & ATA_QCFLAG_ACTIVE &&
330 if (i < ATA_MAX_QUEUE) {
331 /* the scmd has an associated qc */
332 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
333 /* which hasn't failed yet, timeout */
334 qc->err_mask |= AC_ERR_TIMEOUT;
335 qc->flags |= ATA_QCFLAG_FAILED;
339 /* Normal completion occurred after
340 * SCSI timeout but before this point.
341 * Successfully complete it.
343 scmd->retries = scmd->allowed;
344 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
348 /* If we have timed out qcs. They belong to EH from
349 * this point but the state of the controller is
350 * unknown. Freeze the port to make sure the IRQ
351 * handler doesn't diddle with those qcs. This must
352 * be done atomically w.r.t. setting QCFLAG_FAILED.
355 __ata_port_freeze(ap);
357 spin_unlock_irqrestore(ap->lock, flags);
359 spin_unlock_wait(ap->lock);
362 /* invoke error handler */
363 if (ap->ops->error_handler) {
364 /* process port resume request */
365 ata_eh_handle_port_resume(ap);
367 /* fetch & clear EH info */
368 spin_lock_irqsave(ap->lock, flags);
370 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
371 ap->eh_context.i = ap->eh_info;
372 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
374 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
375 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
377 spin_unlock_irqrestore(ap->lock, flags);
379 /* invoke EH, skip if unloading or suspended */
380 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
381 ap->ops->error_handler(ap);
385 /* process port suspend request */
386 ata_eh_handle_port_suspend(ap);
388 /* Exception might have happend after ->error_handler
389 * recovered the port but before this point. Repeat
392 spin_lock_irqsave(ap->lock, flags);
394 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
396 ata_port_printk(ap, KERN_INFO,
397 "EH pending after completion, "
398 "repeating EH (cnt=%d)\n", repeat_cnt);
399 spin_unlock_irqrestore(ap->lock, flags);
402 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
403 "tries, giving up\n", ATA_EH_MAX_REPEAT);
404 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
407 /* this run is complete, make sure EH info is clear */
408 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
410 /* Clear host_eh_scheduled while holding ap->lock such
411 * that if exception occurs after this point but
412 * before EH completion, SCSI midlayer will
415 host->host_eh_scheduled = 0;
417 spin_unlock_irqrestore(ap->lock, flags);
419 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
420 ap->ops->eng_timeout(ap);
423 /* finish or retry handled scmd's and clean up */
424 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
426 scsi_eh_flush_done_q(&ap->eh_done_q);
429 spin_lock_irqsave(ap->lock, flags);
431 if (ap->pflags & ATA_PFLAG_LOADING)
432 ap->pflags &= ~ATA_PFLAG_LOADING;
433 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
434 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
436 if (ap->pflags & ATA_PFLAG_RECOVERED)
437 ata_port_printk(ap, KERN_INFO, "EH complete\n");
439 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
441 /* tell wait_eh that we're done */
442 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
443 wake_up_all(&ap->eh_wait_q);
445 spin_unlock_irqrestore(ap->lock, flags);
451 * ata_port_wait_eh - Wait for the currently pending EH to complete
452 * @ap: Port to wait EH for
454 * Wait until the currently pending EH is complete.
457 * Kernel thread context (may sleep).
459 void ata_port_wait_eh(struct ata_port *ap)
465 spin_lock_irqsave(ap->lock, flags);
467 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
468 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
469 spin_unlock_irqrestore(ap->lock, flags);
471 spin_lock_irqsave(ap->lock, flags);
473 finish_wait(&ap->eh_wait_q, &wait);
475 spin_unlock_irqrestore(ap->lock, flags);
477 /* make sure SCSI EH is complete */
478 if (scsi_host_in_recovery(ap->scsi_host)) {
485 * ata_qc_timeout - Handle timeout of queued command
486 * @qc: Command that timed out
488 * Some part of the kernel (currently, only the SCSI layer)
489 * has noticed that the active command on port @ap has not
490 * completed after a specified length of time. Handle this
491 * condition by disabling DMA (if necessary) and completing
492 * transactions, with error if necessary.
494 * This also handles the case of the "lost interrupt", where
495 * for some reason (possibly hardware bug, possibly driver bug)
496 * an interrupt was not delivered to the driver, even though the
497 * transaction completed successfully.
499 * TODO: kill this function once old EH is gone.
502 * Inherited from SCSI layer (none, can sleep)
504 static void ata_qc_timeout(struct ata_queued_cmd *qc)
506 struct ata_port *ap = qc->ap;
507 u8 host_stat = 0, drv_stat;
512 ap->hsm_task_state = HSM_ST_IDLE;
514 spin_lock_irqsave(ap->lock, flags);
516 switch (qc->tf.protocol) {
519 case ATA_PROT_ATAPI_DMA:
520 host_stat = ap->ops->bmdma_status(ap);
522 /* before we do anything else, clear DMA-Start bit */
523 ap->ops->bmdma_stop(qc);
529 drv_stat = ata_chk_status(ap);
531 /* ack bmdma irq events */
532 ap->ops->irq_clear(ap);
534 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
535 "stat 0x%x host_stat 0x%x\n",
536 qc->tf.command, drv_stat, host_stat);
538 /* complete taskfile transaction */
539 qc->err_mask |= AC_ERR_TIMEOUT;
543 spin_unlock_irqrestore(ap->lock, flags);
545 ata_eh_qc_complete(qc);
551 * ata_eng_timeout - Handle timeout of queued command
552 * @ap: Port on which timed-out command is active
554 * Some part of the kernel (currently, only the SCSI layer)
555 * has noticed that the active command on port @ap has not
556 * completed after a specified length of time. Handle this
557 * condition by disabling DMA (if necessary) and completing
558 * transactions, with error if necessary.
560 * This also handles the case of the "lost interrupt", where
561 * for some reason (possibly hardware bug, possibly driver bug)
562 * an interrupt was not delivered to the driver, even though the
563 * transaction completed successfully.
565 * TODO: kill this function once old EH is gone.
568 * Inherited from SCSI layer (none, can sleep)
570 void ata_eng_timeout(struct ata_port *ap)
574 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
580 * ata_qc_schedule_eh - schedule qc for error handling
581 * @qc: command to schedule error handling for
583 * Schedule error handling for @qc. EH will kick in as soon as
584 * other commands are drained.
587 * spin_lock_irqsave(host lock)
589 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
591 struct ata_port *ap = qc->ap;
593 WARN_ON(!ap->ops->error_handler);
595 qc->flags |= ATA_QCFLAG_FAILED;
596 qc->ap->pflags |= ATA_PFLAG_EH_PENDING;
598 /* The following will fail if timeout has already expired.
599 * ata_scsi_error() takes care of such scmds on EH entry.
600 * Note that ATA_QCFLAG_FAILED is unconditionally set after
601 * this function completes.
603 scsi_req_abort_cmd(qc->scsicmd);
607 * ata_port_schedule_eh - schedule error handling without a qc
608 * @ap: ATA port to schedule EH for
610 * Schedule error handling for @ap. EH will kick in as soon as
611 * all commands are drained.
614 * spin_lock_irqsave(host lock)
616 void ata_port_schedule_eh(struct ata_port *ap)
618 WARN_ON(!ap->ops->error_handler);
620 if (ap->pflags & ATA_PFLAG_INITIALIZING)
623 ap->pflags |= ATA_PFLAG_EH_PENDING;
624 scsi_schedule_eh(ap->scsi_host);
626 DPRINTK("port EH scheduled\n");
630 * ata_port_abort - abort all qc's on the port
631 * @ap: ATA port to abort qc's for
633 * Abort all active qc's of @ap and schedule EH.
636 * spin_lock_irqsave(host lock)
639 * Number of aborted qc's.
641 int ata_port_abort(struct ata_port *ap)
643 int tag, nr_aborted = 0;
645 WARN_ON(!ap->ops->error_handler);
647 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
648 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
651 qc->flags |= ATA_QCFLAG_FAILED;
658 ata_port_schedule_eh(ap);
664 * __ata_port_freeze - freeze port
665 * @ap: ATA port to freeze
667 * This function is called when HSM violation or some other
668 * condition disrupts normal operation of the port. Frozen port
669 * is not allowed to perform any operation until the port is
670 * thawed, which usually follows a successful reset.
672 * ap->ops->freeze() callback can be used for freezing the port
673 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
674 * port cannot be frozen hardware-wise, the interrupt handler
675 * must ack and clear interrupts unconditionally while the port
679 * spin_lock_irqsave(host lock)
681 static void __ata_port_freeze(struct ata_port *ap)
683 WARN_ON(!ap->ops->error_handler);
688 ap->pflags |= ATA_PFLAG_FROZEN;
690 DPRINTK("ata%u port frozen\n", ap->print_id);
694 * ata_port_freeze - abort & freeze port
695 * @ap: ATA port to freeze
697 * Abort and freeze @ap.
700 * spin_lock_irqsave(host lock)
703 * Number of aborted commands.
705 int ata_port_freeze(struct ata_port *ap)
709 WARN_ON(!ap->ops->error_handler);
711 nr_aborted = ata_port_abort(ap);
712 __ata_port_freeze(ap);
718 * ata_eh_freeze_port - EH helper to freeze port
719 * @ap: ATA port to freeze
726 void ata_eh_freeze_port(struct ata_port *ap)
730 if (!ap->ops->error_handler)
733 spin_lock_irqsave(ap->lock, flags);
734 __ata_port_freeze(ap);
735 spin_unlock_irqrestore(ap->lock, flags);
739 * ata_port_thaw_port - EH helper to thaw port
740 * @ap: ATA port to thaw
742 * Thaw frozen port @ap.
747 void ata_eh_thaw_port(struct ata_port *ap)
751 if (!ap->ops->error_handler)
754 spin_lock_irqsave(ap->lock, flags);
756 ap->pflags &= ~ATA_PFLAG_FROZEN;
761 spin_unlock_irqrestore(ap->lock, flags);
763 DPRINTK("ata%u port thawed\n", ap->print_id);
766 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
771 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
773 struct ata_port *ap = qc->ap;
774 struct scsi_cmnd *scmd = qc->scsicmd;
777 spin_lock_irqsave(ap->lock, flags);
778 qc->scsidone = ata_eh_scsidone;
779 __ata_qc_complete(qc);
780 WARN_ON(ata_tag_valid(qc->tag));
781 spin_unlock_irqrestore(ap->lock, flags);
783 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
787 * ata_eh_qc_complete - Complete an active ATA command from EH
788 * @qc: Command to complete
790 * Indicate to the mid and upper layers that an ATA command has
791 * completed. To be used from EH.
793 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
795 struct scsi_cmnd *scmd = qc->scsicmd;
796 scmd->retries = scmd->allowed;
797 __ata_eh_qc_complete(qc);
801 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
802 * @qc: Command to retry
804 * Indicate to the mid and upper layers that an ATA command
805 * should be retried. To be used from EH.
807 * SCSI midlayer limits the number of retries to scmd->allowed.
808 * scmd->retries is decremented for commands which get retried
809 * due to unrelated failures (qc->err_mask is zero).
811 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
813 struct scsi_cmnd *scmd = qc->scsicmd;
814 if (!qc->err_mask && scmd->retries)
816 __ata_eh_qc_complete(qc);
820 * ata_eh_detach_dev - detach ATA device
821 * @dev: ATA device to detach
828 static void ata_eh_detach_dev(struct ata_device *dev)
830 struct ata_port *ap = dev->ap;
833 ata_dev_disable(dev);
835 spin_lock_irqsave(ap->lock, flags);
837 dev->flags &= ~ATA_DFLAG_DETACH;
839 if (ata_scsi_offline_dev(dev)) {
840 dev->flags |= ATA_DFLAG_DETACHED;
841 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
844 /* clear per-dev EH actions */
845 ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK);
846 ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK);
848 spin_unlock_irqrestore(ap->lock, flags);
852 * ata_eh_about_to_do - about to perform eh_action
853 * @ap: target ATA port
854 * @dev: target ATA dev for per-dev action (can be NULL)
855 * @action: action about to be performed
857 * Called just before performing EH actions to clear related bits
858 * in @ap->eh_info such that eh actions are not unnecessarily
864 static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
868 struct ata_eh_info *ehi = &ap->eh_info;
869 struct ata_eh_context *ehc = &ap->eh_context;
871 spin_lock_irqsave(ap->lock, flags);
873 /* Reset is represented by combination of actions and EHI
874 * flags. Suck in all related bits before clearing eh_info to
875 * avoid losing requested action.
877 if (action & ATA_EH_RESET_MASK) {
878 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
879 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
881 /* make sure all reset actions are cleared & clear EHI flags */
882 action |= ATA_EH_RESET_MASK;
883 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
886 ata_eh_clear_action(dev, ehi, action);
888 if (!(ehc->i.flags & ATA_EHI_QUIET))
889 ap->pflags |= ATA_PFLAG_RECOVERED;
891 spin_unlock_irqrestore(ap->lock, flags);
895 * ata_eh_done - EH action complete
896 * @ap: target ATA port
897 * @dev: target ATA dev for per-dev action (can be NULL)
898 * @action: action just completed
900 * Called right after performing EH actions to clear related bits
901 * in @ap->eh_context.
906 static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
909 /* if reset is complete, clear all reset actions & reset modifier */
910 if (action & ATA_EH_RESET_MASK) {
911 action |= ATA_EH_RESET_MASK;
912 ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
915 ata_eh_clear_action(dev, &ap->eh_context.i, action);
919 * ata_err_string - convert err_mask to descriptive string
920 * @err_mask: error mask to convert to string
922 * Convert @err_mask to descriptive string. Errors are
923 * prioritized according to severity and only the most severe
930 * Descriptive string for @err_mask
932 static const char * ata_err_string(unsigned int err_mask)
934 if (err_mask & AC_ERR_HOST_BUS)
935 return "host bus error";
936 if (err_mask & AC_ERR_ATA_BUS)
937 return "ATA bus error";
938 if (err_mask & AC_ERR_TIMEOUT)
940 if (err_mask & AC_ERR_HSM)
941 return "HSM violation";
942 if (err_mask & AC_ERR_SYSTEM)
943 return "internal error";
944 if (err_mask & AC_ERR_MEDIA)
945 return "media error";
946 if (err_mask & AC_ERR_INVALID)
947 return "invalid argument";
948 if (err_mask & AC_ERR_DEV)
949 return "device error";
950 return "unknown error";
954 * ata_read_log_page - read a specific log page
955 * @dev: target device
956 * @page: page to read
957 * @buf: buffer to store read page
958 * @sectors: number of sectors to read
960 * Read log page using READ_LOG_EXT command.
963 * Kernel thread context (may sleep).
966 * 0 on success, AC_ERR_* mask otherwise.
968 static unsigned int ata_read_log_page(struct ata_device *dev,
969 u8 page, void *buf, unsigned int sectors)
971 struct ata_taskfile tf;
972 unsigned int err_mask;
974 DPRINTK("read log page - page %d\n", page);
976 ata_tf_init(dev, &tf);
977 tf.command = ATA_CMD_READ_LOG_EXT;
980 tf.hob_nsect = sectors >> 8;
981 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
982 tf.protocol = ATA_PROT_PIO;
984 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
985 buf, sectors * ATA_SECT_SIZE);
987 DPRINTK("EXIT, err_mask=%x\n", err_mask);
992 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
993 * @dev: Device to read log page 10h from
994 * @tag: Resulting tag of the failed command
995 * @tf: Resulting taskfile registers of the failed command
997 * Read log page 10h to obtain NCQ error details and clear error
1001 * Kernel thread context (may sleep).
1004 * 0 on success, -errno otherwise.
1006 static int ata_eh_read_log_10h(struct ata_device *dev,
1007 int *tag, struct ata_taskfile *tf)
1009 u8 *buf = dev->ap->sector_buf;
1010 unsigned int err_mask;
1014 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1019 for (i = 0; i < ATA_SECT_SIZE; i++)
1022 ata_dev_printk(dev, KERN_WARNING,
1023 "invalid checksum 0x%x on log page 10h\n", csum);
1028 *tag = buf[0] & 0x1f;
1030 tf->command = buf[2];
1031 tf->feature = buf[3];
1035 tf->device = buf[7];
1036 tf->hob_lbal = buf[8];
1037 tf->hob_lbam = buf[9];
1038 tf->hob_lbah = buf[10];
1039 tf->nsect = buf[12];
1040 tf->hob_nsect = buf[13];
1046 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1047 * @dev: device to perform REQUEST_SENSE to
1048 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1050 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1051 * SENSE. This function is EH helper.
1054 * Kernel thread context (may sleep).
1057 * 0 on success, AC_ERR_* mask on failure
1059 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
1061 struct ata_device *dev = qc->dev;
1062 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
1063 struct ata_port *ap = dev->ap;
1064 struct ata_taskfile tf;
1065 u8 cdb[ATAPI_CDB_LEN];
1067 DPRINTK("ATAPI request sense\n");
1069 /* FIXME: is this needed? */
1070 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1072 /* initialize sense_buf with the error register,
1073 * for the case where they are -not- overwritten
1075 sense_buf[0] = 0x70;
1076 sense_buf[2] = qc->result_tf.feature >> 4;
1078 /* some devices time out if garbage left in tf */
1079 ata_tf_init(dev, &tf);
1081 memset(cdb, 0, ATAPI_CDB_LEN);
1082 cdb[0] = REQUEST_SENSE;
1083 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1085 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1086 tf.command = ATA_CMD_PACKET;
1088 /* is it pointless to prefer PIO for "safety reasons"? */
1089 if (ap->flags & ATA_FLAG_PIO_DMA) {
1090 tf.protocol = ATA_PROT_ATAPI_DMA;
1091 tf.feature |= ATAPI_PKT_DMA;
1093 tf.protocol = ATA_PROT_ATAPI;
1094 tf.lbam = (8 * 1024) & 0xff;
1095 tf.lbah = (8 * 1024) >> 8;
1098 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1099 sense_buf, SCSI_SENSE_BUFFERSIZE);
1103 * ata_eh_analyze_serror - analyze SError for a failed port
1104 * @ap: ATA port to analyze SError for
1106 * Analyze SError if available and further determine cause of
1112 static void ata_eh_analyze_serror(struct ata_port *ap)
1114 struct ata_eh_context *ehc = &ap->eh_context;
1115 u32 serror = ehc->i.serror;
1116 unsigned int err_mask = 0, action = 0;
1118 if (serror & SERR_PERSISTENT) {
1119 err_mask |= AC_ERR_ATA_BUS;
1120 action |= ATA_EH_HARDRESET;
1123 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1124 err_mask |= AC_ERR_ATA_BUS;
1125 action |= ATA_EH_SOFTRESET;
1127 if (serror & SERR_PROTOCOL) {
1128 err_mask |= AC_ERR_HSM;
1129 action |= ATA_EH_SOFTRESET;
1131 if (serror & SERR_INTERNAL) {
1132 err_mask |= AC_ERR_SYSTEM;
1133 action |= ATA_EH_HARDRESET;
1135 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1136 ata_ehi_hotplugged(&ehc->i);
1138 ehc->i.err_mask |= err_mask;
1139 ehc->i.action |= action;
1143 * ata_eh_analyze_ncq_error - analyze NCQ error
1144 * @ap: ATA port to analyze NCQ error for
1146 * Read log page 10h, determine the offending qc and acquire
1147 * error status TF. For NCQ device errors, all LLDDs have to do
1148 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1152 * Kernel thread context (may sleep).
1154 static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1156 struct ata_eh_context *ehc = &ap->eh_context;
1157 struct ata_device *dev = ap->device;
1158 struct ata_queued_cmd *qc;
1159 struct ata_taskfile tf;
1162 /* if frozen, we can't do much */
1163 if (ap->pflags & ATA_PFLAG_FROZEN)
1166 /* is it NCQ device error? */
1167 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1170 /* has LLDD analyzed already? */
1171 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1172 qc = __ata_qc_from_tag(ap, tag);
1174 if (!(qc->flags & ATA_QCFLAG_FAILED))
1181 /* okay, this error is ours */
1182 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1184 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1185 "(errno=%d)\n", rc);
1189 if (!(ap->sactive & (1 << tag))) {
1190 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1191 "inactive tag %d\n", tag);
1195 /* we've got the perpetrator, condemn it */
1196 qc = __ata_qc_from_tag(ap, tag);
1197 memcpy(&qc->result_tf, &tf, sizeof(tf));
1198 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1199 ehc->i.err_mask &= ~AC_ERR_DEV;
1203 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1204 * @qc: qc to analyze
1205 * @tf: Taskfile registers to analyze
1207 * Analyze taskfile of @qc and further determine cause of
1208 * failure. This function also requests ATAPI sense data if
1212 * Kernel thread context (may sleep).
1215 * Determined recovery action
1217 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1218 const struct ata_taskfile *tf)
1220 unsigned int tmp, action = 0;
1221 u8 stat = tf->command, err = tf->feature;
1223 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1224 qc->err_mask |= AC_ERR_HSM;
1225 return ATA_EH_SOFTRESET;
1228 if (stat & (ATA_ERR | ATA_DF))
1229 qc->err_mask |= AC_ERR_DEV;
1233 switch (qc->dev->class) {
1236 qc->err_mask |= AC_ERR_ATA_BUS;
1238 qc->err_mask |= AC_ERR_MEDIA;
1240 qc->err_mask |= AC_ERR_INVALID;
1244 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1245 tmp = atapi_eh_request_sense(qc);
1247 /* ATA_QCFLAG_SENSE_VALID is used to
1248 * tell atapi_qc_complete() that sense
1249 * data is already valid.
1251 * TODO: interpret sense data and set
1252 * appropriate err_mask.
1254 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1256 qc->err_mask |= tmp;
1260 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1261 action |= ATA_EH_SOFTRESET;
1266 static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
1268 if (err_mask & AC_ERR_ATA_BUS)
1271 if (err_mask & AC_ERR_TIMEOUT)
1275 if (err_mask & AC_ERR_HSM)
1278 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1285 struct speed_down_verdict_arg {
1290 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1292 struct speed_down_verdict_arg *arg = void_arg;
1293 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
1295 if (ent->timestamp < arg->since)
1298 arg->nr_errors[cat]++;
1303 * ata_eh_speed_down_verdict - Determine speed down verdict
1304 * @dev: Device of interest
1306 * This function examines error ring of @dev and determines
1307 * whether NCQ needs to be turned off, transfer speed should be
1308 * stepped down, or falling back to PIO is necessary.
1310 * Cat-1 is ATA_BUS error for any command.
1312 * Cat-2 is TIMEOUT for any command or HSM violation for known
1313 * supported commands.
1315 * Cat-3 is is unclassified DEV error for known supported
1318 * NCQ needs to be turned off if there have been more than 3
1319 * Cat-2 + Cat-3 errors during last 10 minutes.
1321 * Speed down is necessary if there have been more than 3 Cat-1 +
1322 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1324 * Falling back to PIO mode is necessary if there have been more
1325 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1328 * Inherited from caller.
1331 * OR of ATA_EH_SPDN_* flags.
1333 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1335 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1336 u64 j64 = get_jiffies_64();
1337 struct speed_down_verdict_arg arg;
1338 unsigned int verdict = 0;
1340 /* scan past 10 mins of error history */
1341 memset(&arg, 0, sizeof(arg));
1342 arg.since = j64 - min(j64, j10mins);
1343 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1345 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1346 verdict |= ATA_EH_SPDN_NCQ_OFF;
1347 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1348 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1350 /* scan past 3 mins of error history */
1351 memset(&arg, 0, sizeof(arg));
1352 arg.since = j64 - min(j64, j5mins);
1353 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1355 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1356 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1362 * ata_eh_speed_down - record error and speed down if necessary
1363 * @dev: Failed device
1364 * @is_io: Did the device fail during normal IO?
1365 * @err_mask: err_mask of the error
1367 * Record error and examine error history to determine whether
1368 * adjusting transmission speed is necessary. It also sets
1369 * transmission limits appropriately if such adjustment is
1373 * Kernel thread context (may sleep).
1376 * Determined recovery action.
1378 static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1379 unsigned int err_mask)
1381 unsigned int verdict;
1382 unsigned int action = 0;
1384 /* don't bother if Cat-0 error */
1385 if (ata_eh_categorize_error(is_io, err_mask) == 0)
1388 /* record error and determine whether speed down is necessary */
1389 ata_ering_record(&dev->ering, is_io, err_mask);
1390 verdict = ata_eh_speed_down_verdict(dev);
1393 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1394 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1395 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1396 dev->flags |= ATA_DFLAG_NCQ_OFF;
1397 ata_dev_printk(dev, KERN_WARNING,
1398 "NCQ disabled due to excessive errors\n");
1403 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1404 /* speed down SATA link speed if possible */
1405 if (sata_down_spd_limit(dev->ap) == 0) {
1406 action |= ATA_EH_HARDRESET;
1410 /* lower transfer mode */
1411 if (dev->spdn_cnt < 2) {
1412 static const int dma_dnxfer_sel[] =
1413 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1414 static const int pio_dnxfer_sel[] =
1415 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1418 if (dev->xfer_shift != ATA_SHIFT_PIO)
1419 sel = dma_dnxfer_sel[dev->spdn_cnt];
1421 sel = pio_dnxfer_sel[dev->spdn_cnt];
1425 if (ata_down_xfermask_limit(dev, sel) == 0) {
1426 action |= ATA_EH_SOFTRESET;
1432 /* Fall back to PIO? Slowing down to PIO is meaningless for
1433 * SATA. Consider it only for PATA.
1435 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1436 (dev->ap->cbl != ATA_CBL_SATA) &&
1437 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1438 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1440 action |= ATA_EH_SOFTRESET;
1447 /* device has been slowed down, blow error history */
1448 ata_ering_clear(&dev->ering);
1453 * ata_eh_autopsy - analyze error and determine recovery action
1454 * @ap: ATA port to perform autopsy on
1456 * Analyze why @ap failed and determine which recovery action is
1457 * needed. This function also sets more detailed AC_ERR_* values
1458 * and fills sense data for ATAPI CHECK SENSE.
1461 * Kernel thread context (may sleep).
1463 static void ata_eh_autopsy(struct ata_port *ap)
1465 struct ata_eh_context *ehc = &ap->eh_context;
1466 unsigned int all_err_mask = 0;
1473 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1476 /* obtain and analyze SError */
1477 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1479 ehc->i.serror |= serror;
1480 ata_eh_analyze_serror(ap);
1481 } else if (rc != -EOPNOTSUPP) {
1482 /* SError read failed, force hardreset and probing */
1483 ata_ehi_schedule_probe(&ehc->i);
1484 ehc->i.action |= ATA_EH_HARDRESET;
1485 ehc->i.err_mask |= AC_ERR_OTHER;
1488 /* analyze NCQ failure */
1489 ata_eh_analyze_ncq_error(ap);
1491 /* any real error trumps AC_ERR_OTHER */
1492 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1493 ehc->i.err_mask &= ~AC_ERR_OTHER;
1495 all_err_mask |= ehc->i.err_mask;
1497 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1498 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1500 if (!(qc->flags & ATA_QCFLAG_FAILED))
1503 /* inherit upper level err_mask */
1504 qc->err_mask |= ehc->i.err_mask;
1507 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1509 /* DEV errors are probably spurious in case of ATA_BUS error */
1510 if (qc->err_mask & AC_ERR_ATA_BUS)
1511 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1514 /* any real error trumps unknown error */
1515 if (qc->err_mask & ~AC_ERR_OTHER)
1516 qc->err_mask &= ~AC_ERR_OTHER;
1518 /* SENSE_VALID trumps dev/unknown error and revalidation */
1519 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1520 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1521 ehc->i.action &= ~ATA_EH_REVALIDATE;
1524 /* accumulate error info */
1525 ehc->i.dev = qc->dev;
1526 all_err_mask |= qc->err_mask;
1527 if (qc->flags & ATA_QCFLAG_IO)
1531 /* enforce default EH actions */
1532 if (ap->pflags & ATA_PFLAG_FROZEN ||
1533 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1534 ehc->i.action |= ATA_EH_SOFTRESET;
1535 else if (all_err_mask)
1536 ehc->i.action |= ATA_EH_REVALIDATE;
1538 /* if we have offending qcs and the associated failed device */
1541 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1544 /* perform per-dev EH action only on the offending device */
1545 ehc->i.dev_action[ehc->i.dev->devno] |=
1546 ehc->i.action & ATA_EH_PERDEV_MASK;
1547 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1554 * ata_eh_report - report error handling to user
1555 * @ap: ATA port EH is going on
1557 * Report EH to user.
1562 static void ata_eh_report(struct ata_port *ap)
1564 struct ata_eh_context *ehc = &ap->eh_context;
1565 const char *frozen, *desc;
1566 int tag, nr_failed = 0;
1569 if (ehc->i.desc[0] != '\0')
1572 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1573 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1575 if (!(qc->flags & ATA_QCFLAG_FAILED))
1577 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1583 if (!nr_failed && !ehc->i.err_mask)
1587 if (ap->pflags & ATA_PFLAG_FROZEN)
1591 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1592 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1593 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1594 ehc->i.action, frozen);
1596 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
1598 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1599 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1600 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1601 ehc->i.action, frozen);
1603 ata_port_printk(ap, KERN_ERR, "%s\n", desc);
1606 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1607 static const char *dma_str[] = {
1608 [DMA_BIDIRECTIONAL] = "bidi",
1609 [DMA_TO_DEVICE] = "out",
1610 [DMA_FROM_DEVICE] = "in",
1613 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1614 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
1616 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1619 ata_dev_printk(qc->dev, KERN_ERR,
1620 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1621 "tag %d cdb 0x%x data %u %s\n "
1622 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1623 "Emask 0x%x (%s)%s\n",
1624 cmd->command, cmd->feature, cmd->nsect,
1625 cmd->lbal, cmd->lbam, cmd->lbah,
1626 cmd->hob_feature, cmd->hob_nsect,
1627 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
1628 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
1629 dma_str[qc->dma_dir],
1630 res->command, res->feature, res->nsect,
1631 res->lbal, res->lbam, res->lbah,
1632 res->hob_feature, res->hob_nsect,
1633 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1634 res->device, qc->err_mask, ata_err_string(qc->err_mask),
1635 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
1639 static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1640 unsigned int *classes, unsigned long deadline)
1644 for (i = 0; i < ATA_MAX_DEVICES; i++)
1645 classes[i] = ATA_DEV_UNKNOWN;
1647 rc = reset(ap, classes, deadline);
1651 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1652 * is complete and convert all ATA_DEV_UNKNOWN to
1655 for (i = 0; i < ATA_MAX_DEVICES; i++)
1656 if (classes[i] != ATA_DEV_UNKNOWN)
1659 if (i < ATA_MAX_DEVICES)
1660 for (i = 0; i < ATA_MAX_DEVICES; i++)
1661 if (classes[i] == ATA_DEV_UNKNOWN)
1662 classes[i] = ATA_DEV_NONE;
1667 static int ata_eh_followup_srst_needed(int rc, int classify,
1668 const unsigned int *classes)
1674 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1679 static int ata_eh_reset(struct ata_port *ap, int classify,
1680 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
1681 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1683 struct ata_eh_context *ehc = &ap->eh_context;
1684 unsigned int *classes = ehc->classes;
1685 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
1687 unsigned long deadline;
1688 unsigned int action;
1689 ata_reset_fn_t reset;
1692 /* about to reset */
1693 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1695 /* Determine which reset to use and record in ehc->i.action.
1696 * prereset() may examine and modify it.
1698 action = ehc->i.action;
1699 ehc->i.action &= ~ATA_EH_RESET_MASK;
1700 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
1701 !(action & ATA_EH_HARDRESET))))
1702 ehc->i.action |= ATA_EH_SOFTRESET;
1704 ehc->i.action |= ATA_EH_HARDRESET;
1707 rc = prereset(ap, jiffies + ATA_EH_PRERESET_TIMEOUT);
1709 if (rc == -ENOENT) {
1710 ata_port_printk(ap, KERN_DEBUG,
1711 "port disabled. ignoring.\n");
1712 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
1714 for (i = 0; i < ATA_MAX_DEVICES; i++)
1715 classes[i] = ATA_DEV_NONE;
1719 ata_port_printk(ap, KERN_ERR,
1720 "prereset failed (errno=%d)\n", rc);
1725 /* prereset() might have modified ehc->i.action */
1726 if (ehc->i.action & ATA_EH_HARDRESET)
1728 else if (ehc->i.action & ATA_EH_SOFTRESET)
1731 /* prereset told us not to reset, bang classes and return */
1732 for (i = 0; i < ATA_MAX_DEVICES; i++)
1733 classes[i] = ATA_DEV_NONE;
1738 /* did prereset() screw up? if so, fix up to avoid oopsing */
1747 deadline = jiffies + ata_eh_reset_timeouts[try++];
1749 /* shut up during boot probing */
1751 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1752 reset == softreset ? "soft" : "hard");
1754 /* mark that this EH session started with reset */
1755 if (reset == hardreset)
1756 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
1758 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
1760 rc = ata_do_reset(ap, reset, classes, deadline);
1762 if (reset == hardreset &&
1763 ata_eh_followup_srst_needed(rc, classify, classes)) {
1764 /* okay, let's do follow-up softreset */
1768 ata_port_printk(ap, KERN_ERR,
1769 "follow-up softreset required "
1770 "but no softreset avaliable\n");
1775 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
1776 rc = ata_do_reset(ap, reset, classes, deadline);
1778 if (rc == 0 && classify &&
1779 classes[0] == ATA_DEV_UNKNOWN) {
1780 ata_port_printk(ap, KERN_ERR,
1781 "classification failed\n");
1787 if (rc && try < ARRAY_SIZE(ata_eh_reset_timeouts)) {
1788 unsigned long now = jiffies;
1790 if (time_before(now, deadline)) {
1791 unsigned long delta = deadline - jiffies;
1793 ata_port_printk(ap, KERN_WARNING, "reset failed "
1794 "(errno=%d), retrying in %u secs\n",
1795 rc, (jiffies_to_msecs(delta) + 999) / 1000);
1797 schedule_timeout_uninterruptible(delta);
1801 try == ARRAY_SIZE(ata_eh_reset_timeouts) - 1)
1802 sata_down_spd_limit(ap);
1811 /* After the reset, the device state is PIO 0 and the
1812 * controller state is undefined. Record the mode.
1814 for (i = 0; i < ATA_MAX_DEVICES; i++)
1815 ap->device[i].pio_mode = XFER_PIO_0;
1817 /* record current link speed */
1818 if (sata_scr_read(ap, SCR_STATUS, &sstatus) == 0)
1819 ap->sata_spd = (sstatus >> 4) & 0xf;
1822 postreset(ap, classes);
1824 /* reset successful, schedule revalidation */
1825 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1826 ehc->i.action |= ATA_EH_REVALIDATE;
1829 /* clear hotplug flag */
1830 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
1834 static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1835 struct ata_device **r_failed_dev)
1837 struct ata_eh_context *ehc = &ap->eh_context;
1838 struct ata_device *dev;
1839 unsigned int new_mask = 0;
1840 unsigned long flags;
1845 /* For PATA drive side cable detection to work, IDENTIFY must
1846 * be done backwards such that PDIAG- is released by the slave
1847 * device before the master device is identified.
1849 for (i = ATA_MAX_DEVICES - 1; i >= 0; i--) {
1850 unsigned int action, readid_flags = 0;
1852 dev = &ap->device[i];
1853 action = ata_eh_dev_action(dev);
1855 if (ehc->i.flags & ATA_EHI_DID_RESET)
1856 readid_flags |= ATA_READID_POSTRESET;
1858 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
1859 if (ata_port_offline(ap)) {
1864 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
1865 rc = ata_dev_revalidate(dev, readid_flags);
1869 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1871 /* Configuration may have changed, reconfigure
1874 ehc->i.flags |= ATA_EHI_SETMODE;
1876 /* schedule the scsi_rescan_device() here */
1877 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
1878 } else if (dev->class == ATA_DEV_UNKNOWN &&
1879 ehc->tries[dev->devno] &&
1880 ata_class_enabled(ehc->classes[dev->devno])) {
1881 dev->class = ehc->classes[dev->devno];
1883 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1890 /* IDENTIFY was issued to non-existent
1891 * device. No need to reset. Just
1892 * thaw and kill the device.
1894 ata_eh_thaw_port(ap);
1895 dev->class = ATA_DEV_UNKNOWN;
1898 dev->class = ATA_DEV_UNKNOWN;
1904 /* PDIAG- should have been released, ask cable type if post-reset */
1905 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ap->ops->cable_detect)
1906 ap->cbl = ap->ops->cable_detect(ap);
1908 /* Configure new devices forward such that user doesn't see
1909 * device detection messages backwards.
1911 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1912 dev = &ap->device[i];
1914 if (!(new_mask & (1 << i)))
1917 ehc->i.flags |= ATA_EHI_PRINTINFO;
1918 rc = ata_dev_configure(dev);
1919 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
1923 spin_lock_irqsave(ap->lock, flags);
1924 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1925 spin_unlock_irqrestore(ap->lock, flags);
1927 /* new device discovered, configure xfermode */
1928 ehc->i.flags |= ATA_EHI_SETMODE;
1934 *r_failed_dev = dev;
1935 DPRINTK("EXIT rc=%d\n", rc);
1939 static int ata_port_nr_enabled(struct ata_port *ap)
1943 for (i = 0; i < ATA_MAX_DEVICES; i++)
1944 if (ata_dev_enabled(&ap->device[i]))
1949 static int ata_port_nr_vacant(struct ata_port *ap)
1953 for (i = 0; i < ATA_MAX_DEVICES; i++)
1954 if (ap->device[i].class == ATA_DEV_UNKNOWN)
1959 static int ata_eh_skip_recovery(struct ata_port *ap)
1961 struct ata_eh_context *ehc = &ap->eh_context;
1964 /* thaw frozen port, resume link and recover failed devices */
1965 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
1966 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
1969 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1970 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1971 struct ata_device *dev = &ap->device[i];
1973 if (dev->class == ATA_DEV_UNKNOWN &&
1974 ehc->classes[dev->devno] != ATA_DEV_NONE)
1981 static void ata_eh_handle_dev_fail(struct ata_device *dev, int err)
1983 struct ata_port *ap = dev->ap;
1984 struct ata_eh_context *ehc = &ap->eh_context;
1986 ehc->tries[dev->devno]--;
1990 /* device missing or wrong IDENTIFY data, schedule probing */
1991 ehc->i.probe_mask |= (1 << dev->devno);
1993 /* give it just one more chance */
1994 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
1996 if (ehc->tries[dev->devno] == 1) {
1997 /* This is the last chance, better to slow
1998 * down than lose it.
2000 sata_down_spd_limit(ap);
2001 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2005 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2006 /* disable device if it has used up all its chances */
2007 ata_dev_disable(dev);
2009 /* detach if offline */
2010 if (ata_port_offline(ap))
2011 ata_eh_detach_dev(dev);
2013 /* probe if requested */
2014 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2015 !(ehc->did_probe_mask & (1 << dev->devno))) {
2016 ata_eh_detach_dev(dev);
2019 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2020 ehc->did_probe_mask |= (1 << dev->devno);
2021 ehc->i.action |= ATA_EH_SOFTRESET;
2024 /* soft didn't work? be haaaaard */
2025 if (ehc->i.flags & ATA_EHI_DID_RESET)
2026 ehc->i.action |= ATA_EH_HARDRESET;
2028 ehc->i.action |= ATA_EH_SOFTRESET;
2033 * ata_eh_recover - recover host port after error
2034 * @ap: host port to recover
2035 * @prereset: prereset method (can be NULL)
2036 * @softreset: softreset method (can be NULL)
2037 * @hardreset: hardreset method (can be NULL)
2038 * @postreset: postreset method (can be NULL)
2040 * This is the alpha and omega, eum and yang, heart and soul of
2041 * libata exception handling. On entry, actions required to
2042 * recover the port and hotplug requests are recorded in
2043 * eh_context. This function executes all the operations with
2044 * appropriate retrials and fallbacks to resurrect failed
2045 * devices, detach goners and greet newcomers.
2048 * Kernel thread context (may sleep).
2051 * 0 on success, -errno on failure.
2053 static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2054 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2055 ata_postreset_fn_t postreset)
2057 struct ata_eh_context *ehc = &ap->eh_context;
2058 struct ata_device *dev;
2063 /* prep for recovery */
2064 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2065 dev = &ap->device[i];
2067 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2069 /* collect port action mask recorded in dev actions */
2070 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
2071 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
2073 /* process hotplug request */
2074 if (dev->flags & ATA_DFLAG_DETACH)
2075 ata_eh_detach_dev(dev);
2077 if (!ata_dev_enabled(dev) &&
2078 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2079 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2080 ata_eh_detach_dev(dev);
2082 ehc->did_probe_mask |= (1 << dev->devno);
2083 ehc->i.action |= ATA_EH_SOFTRESET;
2090 /* if UNLOADING, finish immediately */
2091 if (ap->pflags & ATA_PFLAG_UNLOADING)
2094 /* skip EH if possible. */
2095 if (ata_eh_skip_recovery(ap))
2098 for (i = 0; i < ATA_MAX_DEVICES; i++)
2099 ehc->classes[i] = ATA_DEV_UNKNOWN;
2102 if (ehc->i.action & ATA_EH_RESET_MASK) {
2103 ata_eh_freeze_port(ap);
2105 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
2106 softreset, hardreset, postreset);
2108 ata_port_printk(ap, KERN_ERR,
2109 "reset failed, giving up\n");
2113 ata_eh_thaw_port(ap);
2116 /* revalidate existing devices and attach new ones */
2117 rc = ata_eh_revalidate_and_attach(ap, &dev);
2121 /* configure transfer mode if necessary */
2122 if (ehc->i.flags & ATA_EHI_SETMODE) {
2123 rc = ata_set_mode(ap, &dev);
2126 ehc->i.flags &= ~ATA_EHI_SETMODE;
2132 ata_eh_handle_dev_fail(dev, rc);
2134 if (ata_port_nr_enabled(ap)) {
2135 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2136 "devices, retrying in 5 secs\n");
2139 /* no device left, repeat fast */
2147 for (i = 0; i < ATA_MAX_DEVICES; i++)
2148 ata_dev_disable(&ap->device[i]);
2151 DPRINTK("EXIT, rc=%d\n", rc);
2156 * ata_eh_finish - finish up EH
2157 * @ap: host port to finish EH for
2159 * Recovery is complete. Clean up EH states and retry or finish
2165 static void ata_eh_finish(struct ata_port *ap)
2169 /* retry or finish qcs */
2170 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2171 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2173 if (!(qc->flags & ATA_QCFLAG_FAILED))
2177 /* FIXME: Once EH migration is complete,
2178 * generate sense data in this function,
2179 * considering both err_mask and tf.
2181 if (qc->err_mask & AC_ERR_INVALID)
2182 ata_eh_qc_complete(qc);
2184 ata_eh_qc_retry(qc);
2186 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2187 ata_eh_qc_complete(qc);
2189 /* feed zero TF to sense generation */
2190 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2191 ata_eh_qc_retry(qc);
2198 * ata_do_eh - do standard error handling
2199 * @ap: host port to handle error for
2200 * @prereset: prereset method (can be NULL)
2201 * @softreset: softreset method (can be NULL)
2202 * @hardreset: hardreset method (can be NULL)
2203 * @postreset: postreset method (can be NULL)
2205 * Perform standard error handling sequence.
2208 * Kernel thread context (may sleep).
2210 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2211 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2212 ata_postreset_fn_t postreset)
2216 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
2222 * ata_eh_handle_port_suspend - perform port suspend operation
2223 * @ap: port to suspend
2228 * Kernel thread context (may sleep).
2230 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2232 unsigned long flags;
2235 /* are we suspending? */
2236 spin_lock_irqsave(ap->lock, flags);
2237 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2238 ap->pm_mesg.event == PM_EVENT_ON) {
2239 spin_unlock_irqrestore(ap->lock, flags);
2242 spin_unlock_irqrestore(ap->lock, flags);
2244 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2246 /* tell ACPI we're suspending */
2247 rc = ata_acpi_on_suspend(ap);
2252 ata_eh_freeze_port(ap);
2254 if (ap->ops->port_suspend)
2255 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2259 spin_lock_irqsave(ap->lock, flags);
2261 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2263 ap->pflags |= ATA_PFLAG_SUSPENDED;
2264 else if (ap->pflags & ATA_PFLAG_FROZEN)
2265 ata_port_schedule_eh(ap);
2267 if (ap->pm_result) {
2268 *ap->pm_result = rc;
2269 ap->pm_result = NULL;
2272 spin_unlock_irqrestore(ap->lock, flags);
2278 * ata_eh_handle_port_resume - perform port resume operation
2279 * @ap: port to resume
2284 * Kernel thread context (may sleep).
2286 static void ata_eh_handle_port_resume(struct ata_port *ap)
2288 unsigned long flags;
2291 /* are we resuming? */
2292 spin_lock_irqsave(ap->lock, flags);
2293 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2294 ap->pm_mesg.event != PM_EVENT_ON) {
2295 spin_unlock_irqrestore(ap->lock, flags);
2298 spin_unlock_irqrestore(ap->lock, flags);
2300 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
2302 if (ap->ops->port_resume)
2303 rc = ap->ops->port_resume(ap);
2305 /* tell ACPI that we're resuming */
2306 ata_acpi_on_resume(ap);
2309 spin_lock_irqsave(ap->lock, flags);
2310 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2311 if (ap->pm_result) {
2312 *ap->pm_result = rc;
2313 ap->pm_result = NULL;
2315 spin_unlock_irqrestore(ap->lock, flags);
2317 #endif /* CONFIG_PM */