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 <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include "../scsi/scsi_transport_api.h"
45 #include <linux/libata.h>
50 /* speed down verdicts */
51 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
54 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
57 ATA_EFLAG_IS_IO = (1 << 0),
58 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
60 /* error categories */
63 ATA_ECAT_TOUT_HSM = 2,
65 ATA_ECAT_DUBIOUS_NONE = 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_EH_CMD_DFL_TIMEOUT = 5000,
73 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN = 5000,
76 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
81 ATA_EH_PRERESET_TIMEOUT = 10000,
82 ATA_EH_FASTDRAIN_INTERVAL = 3000,
86 /* probe speed down parameters, see ata_eh_schedule_probe() */
87 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
88 ATA_EH_PROBE_TRIALS = 2,
91 /* The following table determines how we sequence resets. Each entry
92 * represents timeout for that try. The first try can be soft or
93 * hardreset. All others are hardreset if available. In most cases
94 * the first reset w/ 10sec timeout should succeed. Following entries
95 * are mostly for error handling, hotplug and retarded devices.
97 static const unsigned long ata_eh_reset_timeouts[] = {
98 10000, /* most drives spin up by 10sec */
99 10000, /* > 99% working drives spin up before 20sec */
100 35000, /* give > 30 secs of idleness for retarded devices */
101 5000, /* and sweet one last chance */
102 ULONG_MAX, /* > 1 min has elapsed, give up */
105 static const unsigned long ata_eh_identify_timeouts[] = {
106 5000, /* covers > 99% of successes and not too boring on failures */
107 10000, /* combined time till here is enough even for media access */
108 30000, /* for true idiots */
112 static const unsigned long ata_eh_other_timeouts[] = {
113 5000, /* same rationale as identify timeout */
115 /* but no merciful 30sec for other commands, it just isn't worth it */
119 struct ata_eh_cmd_timeout_ent {
121 const unsigned long *timeouts;
124 /* The following table determines timeouts to use for EH internal
125 * commands. Each table entry is a command class and matches the
126 * commands the entry applies to and the timeout table to use.
128 * On the retry after a command timed out, the next timeout value from
129 * the table is used. If the table doesn't contain further entries,
130 * the last value is used.
132 * ehc->cmd_timeout_idx keeps track of which timeout to use per
133 * command class, so if SET_FEATURES times out on the first try, the
134 * next try will use the second timeout value only for that class.
136 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
137 static const struct ata_eh_cmd_timeout_ent
138 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
139 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
140 .timeouts = ata_eh_identify_timeouts, },
141 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
142 .timeouts = ata_eh_other_timeouts, },
143 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
144 .timeouts = ata_eh_other_timeouts, },
145 { .commands = CMDS(ATA_CMD_SET_FEATURES),
146 .timeouts = ata_eh_other_timeouts, },
147 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
148 .timeouts = ata_eh_other_timeouts, },
152 static void __ata_port_freeze(struct ata_port *ap);
154 static void ata_eh_handle_port_suspend(struct ata_port *ap);
155 static void ata_eh_handle_port_resume(struct ata_port *ap);
156 #else /* CONFIG_PM */
157 static void ata_eh_handle_port_suspend(struct ata_port *ap)
160 static void ata_eh_handle_port_resume(struct ata_port *ap)
162 #endif /* CONFIG_PM */
164 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
167 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
168 ATA_EH_DESC_LEN - ehi->desc_len,
173 * __ata_ehi_push_desc - push error description without adding separator
175 * @fmt: printf format string
177 * Format string according to @fmt and append it to @ehi->desc.
180 * spin_lock_irqsave(host lock)
182 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
187 __ata_ehi_pushv_desc(ehi, fmt, args);
192 * ata_ehi_push_desc - push error description with separator
194 * @fmt: printf format string
196 * Format string according to @fmt and append it to @ehi->desc.
197 * If @ehi->desc is not empty, ", " is added in-between.
200 * spin_lock_irqsave(host lock)
202 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
207 __ata_ehi_push_desc(ehi, ", ");
210 __ata_ehi_pushv_desc(ehi, fmt, args);
215 * ata_ehi_clear_desc - clean error description
221 * spin_lock_irqsave(host lock)
223 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
230 * ata_port_desc - append port description
231 * @ap: target ATA port
232 * @fmt: printf format string
234 * Format string according to @fmt and append it to port
235 * description. If port description is not empty, " " is added
236 * in-between. This function is to be used while initializing
237 * ata_host. The description is printed on host registration.
242 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
246 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
248 if (ap->link.eh_info.desc_len)
249 __ata_ehi_push_desc(&ap->link.eh_info, " ");
252 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
259 * ata_port_pbar_desc - append PCI BAR description
260 * @ap: target ATA port
261 * @bar: target PCI BAR
262 * @offset: offset into PCI BAR
263 * @name: name of the area
265 * If @offset is negative, this function formats a string which
266 * contains the name, address, size and type of the BAR and
267 * appends it to the port description. If @offset is zero or
268 * positive, only name and offsetted address is appended.
273 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
276 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
278 unsigned long long start, len;
280 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
282 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
285 start = (unsigned long long)pci_resource_start(pdev, bar);
286 len = (unsigned long long)pci_resource_len(pdev, bar);
289 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
291 ata_port_desc(ap, "%s 0x%llx", name,
292 start + (unsigned long long)offset);
295 #endif /* CONFIG_PCI */
297 static int ata_lookup_timeout_table(u8 cmd)
301 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
304 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
313 * ata_internal_cmd_timeout - determine timeout for an internal command
314 * @dev: target device
315 * @cmd: internal command to be issued
317 * Determine timeout for internal command @cmd for @dev.
323 * Determined timeout.
325 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
327 struct ata_eh_context *ehc = &dev->link->eh_context;
328 int ent = ata_lookup_timeout_table(cmd);
332 return ATA_EH_CMD_DFL_TIMEOUT;
334 idx = ehc->cmd_timeout_idx[dev->devno][ent];
335 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
339 * ata_internal_cmd_timed_out - notification for internal command timeout
340 * @dev: target device
341 * @cmd: internal command which timed out
343 * Notify EH that internal command @cmd for @dev timed out. This
344 * function should be called only for commands whose timeouts are
345 * determined using ata_internal_cmd_timeout().
350 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
352 struct ata_eh_context *ehc = &dev->link->eh_context;
353 int ent = ata_lookup_timeout_table(cmd);
359 idx = ehc->cmd_timeout_idx[dev->devno][ent];
360 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
361 ehc->cmd_timeout_idx[dev->devno][ent]++;
364 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
365 unsigned int err_mask)
367 struct ata_ering_entry *ent;
372 ering->cursor %= ATA_ERING_SIZE;
374 ent = &ering->ring[ering->cursor];
375 ent->eflags = eflags;
376 ent->err_mask = err_mask;
377 ent->timestamp = get_jiffies_64();
380 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
382 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
389 static void ata_ering_clear(struct ata_ering *ering)
391 memset(ering, 0, sizeof(*ering));
394 static int ata_ering_map(struct ata_ering *ering,
395 int (*map_fn)(struct ata_ering_entry *, void *),
399 struct ata_ering_entry *ent;
403 ent = &ering->ring[idx];
406 rc = map_fn(ent, arg);
409 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
410 } while (idx != ering->cursor);
415 static unsigned int ata_eh_dev_action(struct ata_device *dev)
417 struct ata_eh_context *ehc = &dev->link->eh_context;
419 return ehc->i.action | ehc->i.dev_action[dev->devno];
422 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
423 struct ata_eh_info *ehi, unsigned int action)
425 struct ata_device *tdev;
428 ehi->action &= ~action;
429 ata_for_each_dev(tdev, link, ALL)
430 ehi->dev_action[tdev->devno] &= ~action;
432 /* doesn't make sense for port-wide EH actions */
433 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
435 /* break ehi->action into ehi->dev_action */
436 if (ehi->action & action) {
437 ata_for_each_dev(tdev, link, ALL)
438 ehi->dev_action[tdev->devno] |=
439 ehi->action & action;
440 ehi->action &= ~action;
443 /* turn off the specified per-dev action */
444 ehi->dev_action[dev->devno] &= ~action;
449 * ata_scsi_timed_out - SCSI layer time out callback
450 * @cmd: timed out SCSI command
452 * Handles SCSI layer timeout. We race with normal completion of
453 * the qc for @cmd. If the qc is already gone, we lose and let
454 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
455 * timed out and EH should be invoked. Prevent ata_qc_complete()
456 * from finishing it by setting EH_SCHEDULED and return
459 * TODO: kill this function once old EH is gone.
462 * Called from timer context
465 * EH_HANDLED or EH_NOT_HANDLED
467 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
469 struct Scsi_Host *host = cmd->device->host;
470 struct ata_port *ap = ata_shost_to_port(host);
472 struct ata_queued_cmd *qc;
473 enum blk_eh_timer_return ret;
477 if (ap->ops->error_handler) {
478 ret = BLK_EH_NOT_HANDLED;
482 ret = BLK_EH_HANDLED;
483 spin_lock_irqsave(ap->lock, flags);
484 qc = ata_qc_from_tag(ap, ap->link.active_tag);
486 WARN_ON(qc->scsicmd != cmd);
487 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
488 qc->err_mask |= AC_ERR_TIMEOUT;
489 ret = BLK_EH_NOT_HANDLED;
491 spin_unlock_irqrestore(ap->lock, flags);
494 DPRINTK("EXIT, ret=%d\n", ret);
498 static void ata_eh_unload(struct ata_port *ap)
500 struct ata_link *link;
501 struct ata_device *dev;
504 /* Restore SControl IPM and SPD for the next driver and
505 * disable attached devices.
507 ata_for_each_link(link, ap, PMP_FIRST) {
508 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
509 ata_for_each_dev(dev, link, ALL)
510 ata_dev_disable(dev);
513 /* freeze and set UNLOADED */
514 spin_lock_irqsave(ap->lock, flags);
516 ata_port_freeze(ap); /* won't be thawed */
517 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
518 ap->pflags |= ATA_PFLAG_UNLOADED;
520 spin_unlock_irqrestore(ap->lock, flags);
524 * ata_scsi_error - SCSI layer error handler callback
525 * @host: SCSI host on which error occurred
527 * Handles SCSI-layer-thrown error events.
530 * Inherited from SCSI layer (none, can sleep)
535 void ata_scsi_error(struct Scsi_Host *host)
537 struct ata_port *ap = ata_shost_to_port(host);
543 /* synchronize with port task */
544 ata_port_flush_task(ap);
546 /* synchronize with host lock and sort out timeouts */
548 /* For new EH, all qcs are finished in one of three ways -
549 * normal completion, error completion, and SCSI timeout.
550 * Both completions can race against SCSI timeout. When normal
551 * completion wins, the qc never reaches EH. When error
552 * completion wins, the qc has ATA_QCFLAG_FAILED set.
554 * When SCSI timeout wins, things are a bit more complex.
555 * Normal or error completion can occur after the timeout but
556 * before this point. In such cases, both types of
557 * completions are honored. A scmd is determined to have
558 * timed out iff its associated qc is active and not failed.
560 if (ap->ops->error_handler) {
561 struct scsi_cmnd *scmd, *tmp;
564 spin_lock_irqsave(ap->lock, flags);
566 /* This must occur under the ap->lock as we don't want
567 a polled recovery to race the real interrupt handler
569 The lost_interrupt handler checks for any completed but
570 non-notified command and completes much like an IRQ handler.
572 We then fall into the error recovery code which will treat
573 this as if normal completion won the race */
575 if (ap->ops->lost_interrupt)
576 ap->ops->lost_interrupt(ap);
578 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
579 struct ata_queued_cmd *qc;
581 for (i = 0; i < ATA_MAX_QUEUE; i++) {
582 qc = __ata_qc_from_tag(ap, i);
583 if (qc->flags & ATA_QCFLAG_ACTIVE &&
588 if (i < ATA_MAX_QUEUE) {
589 /* the scmd has an associated qc */
590 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
591 /* which hasn't failed yet, timeout */
592 qc->err_mask |= AC_ERR_TIMEOUT;
593 qc->flags |= ATA_QCFLAG_FAILED;
597 /* Normal completion occurred after
598 * SCSI timeout but before this point.
599 * Successfully complete it.
601 scmd->retries = scmd->allowed;
602 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
606 /* If we have timed out qcs. They belong to EH from
607 * this point but the state of the controller is
608 * unknown. Freeze the port to make sure the IRQ
609 * handler doesn't diddle with those qcs. This must
610 * be done atomically w.r.t. setting QCFLAG_FAILED.
613 __ata_port_freeze(ap);
615 spin_unlock_irqrestore(ap->lock, flags);
617 /* initialize eh_tries */
618 ap->eh_tries = ATA_EH_MAX_TRIES;
620 spin_unlock_wait(ap->lock);
622 /* If we timed raced normal completion and there is nothing to
623 recover nr_timedout == 0 why exactly are we doing error recovery ? */
626 /* invoke error handler */
627 if (ap->ops->error_handler) {
628 struct ata_link *link;
630 /* kill fast drain timer */
631 del_timer_sync(&ap->fastdrain_timer);
633 /* process port resume request */
634 ata_eh_handle_port_resume(ap);
636 /* fetch & clear EH info */
637 spin_lock_irqsave(ap->lock, flags);
639 ata_for_each_link(link, ap, HOST_FIRST) {
640 struct ata_eh_context *ehc = &link->eh_context;
641 struct ata_device *dev;
643 memset(&link->eh_context, 0, sizeof(link->eh_context));
644 link->eh_context.i = link->eh_info;
645 memset(&link->eh_info, 0, sizeof(link->eh_info));
647 ata_for_each_dev(dev, link, ENABLED) {
648 int devno = dev->devno;
650 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
651 if (ata_ncq_enabled(dev))
652 ehc->saved_ncq_enabled |= 1 << devno;
656 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
657 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
658 ap->excl_link = NULL; /* don't maintain exclusion over EH */
660 spin_unlock_irqrestore(ap->lock, flags);
662 /* invoke EH, skip if unloading or suspended */
663 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
664 ap->ops->error_handler(ap);
666 /* if unloading, commence suicide */
667 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
668 !(ap->pflags & ATA_PFLAG_UNLOADED))
673 /* process port suspend request */
674 ata_eh_handle_port_suspend(ap);
676 /* Exception might have happend after ->error_handler
677 * recovered the port but before this point. Repeat
680 spin_lock_irqsave(ap->lock, flags);
682 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
683 if (--ap->eh_tries) {
684 spin_unlock_irqrestore(ap->lock, flags);
687 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
688 "tries, giving up\n", ATA_EH_MAX_TRIES);
689 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
692 /* this run is complete, make sure EH info is clear */
693 ata_for_each_link(link, ap, HOST_FIRST)
694 memset(&link->eh_info, 0, sizeof(link->eh_info));
696 /* Clear host_eh_scheduled while holding ap->lock such
697 * that if exception occurs after this point but
698 * before EH completion, SCSI midlayer will
701 host->host_eh_scheduled = 0;
703 spin_unlock_irqrestore(ap->lock, flags);
705 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
706 ap->ops->eng_timeout(ap);
709 /* finish or retry handled scmd's and clean up */
710 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
712 scsi_eh_flush_done_q(&ap->eh_done_q);
715 spin_lock_irqsave(ap->lock, flags);
717 if (ap->pflags & ATA_PFLAG_LOADING)
718 ap->pflags &= ~ATA_PFLAG_LOADING;
719 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
720 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
722 if (ap->pflags & ATA_PFLAG_RECOVERED)
723 ata_port_printk(ap, KERN_INFO, "EH complete\n");
725 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
727 /* tell wait_eh that we're done */
728 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
729 wake_up_all(&ap->eh_wait_q);
731 spin_unlock_irqrestore(ap->lock, flags);
737 * ata_port_wait_eh - Wait for the currently pending EH to complete
738 * @ap: Port to wait EH for
740 * Wait until the currently pending EH is complete.
743 * Kernel thread context (may sleep).
745 void ata_port_wait_eh(struct ata_port *ap)
751 spin_lock_irqsave(ap->lock, flags);
753 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
754 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
755 spin_unlock_irqrestore(ap->lock, flags);
757 spin_lock_irqsave(ap->lock, flags);
759 finish_wait(&ap->eh_wait_q, &wait);
761 spin_unlock_irqrestore(ap->lock, flags);
763 /* make sure SCSI EH is complete */
764 if (scsi_host_in_recovery(ap->scsi_host)) {
770 static int ata_eh_nr_in_flight(struct ata_port *ap)
775 /* count only non-internal commands */
776 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
777 if (ata_qc_from_tag(ap, tag))
783 void ata_eh_fastdrain_timerfn(unsigned long arg)
785 struct ata_port *ap = (void *)arg;
789 spin_lock_irqsave(ap->lock, flags);
791 cnt = ata_eh_nr_in_flight(ap);
797 if (cnt == ap->fastdrain_cnt) {
800 /* No progress during the last interval, tag all
801 * in-flight qcs as timed out and freeze the port.
803 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
804 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
806 qc->err_mask |= AC_ERR_TIMEOUT;
811 /* some qcs have finished, give it another chance */
812 ap->fastdrain_cnt = cnt;
813 ap->fastdrain_timer.expires =
814 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
815 add_timer(&ap->fastdrain_timer);
819 spin_unlock_irqrestore(ap->lock, flags);
823 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
824 * @ap: target ATA port
825 * @fastdrain: activate fast drain
827 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
828 * is non-zero and EH wasn't pending before. Fast drain ensures
829 * that EH kicks in in timely manner.
832 * spin_lock_irqsave(host lock)
834 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
838 /* already scheduled? */
839 if (ap->pflags & ATA_PFLAG_EH_PENDING)
842 ap->pflags |= ATA_PFLAG_EH_PENDING;
847 /* do we have in-flight qcs? */
848 cnt = ata_eh_nr_in_flight(ap);
852 /* activate fast drain */
853 ap->fastdrain_cnt = cnt;
854 ap->fastdrain_timer.expires =
855 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
856 add_timer(&ap->fastdrain_timer);
860 * ata_qc_schedule_eh - schedule qc for error handling
861 * @qc: command to schedule error handling for
863 * Schedule error handling for @qc. EH will kick in as soon as
864 * other commands are drained.
867 * spin_lock_irqsave(host lock)
869 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
871 struct ata_port *ap = qc->ap;
873 WARN_ON(!ap->ops->error_handler);
875 qc->flags |= ATA_QCFLAG_FAILED;
876 ata_eh_set_pending(ap, 1);
878 /* The following will fail if timeout has already expired.
879 * ata_scsi_error() takes care of such scmds on EH entry.
880 * Note that ATA_QCFLAG_FAILED is unconditionally set after
881 * this function completes.
883 blk_abort_request(qc->scsicmd->request);
887 * ata_port_schedule_eh - schedule error handling without a qc
888 * @ap: ATA port to schedule EH for
890 * Schedule error handling for @ap. EH will kick in as soon as
891 * all commands are drained.
894 * spin_lock_irqsave(host lock)
896 void ata_port_schedule_eh(struct ata_port *ap)
898 WARN_ON(!ap->ops->error_handler);
900 if (ap->pflags & ATA_PFLAG_INITIALIZING)
903 ata_eh_set_pending(ap, 1);
904 scsi_schedule_eh(ap->scsi_host);
906 DPRINTK("port EH scheduled\n");
909 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
911 int tag, nr_aborted = 0;
913 WARN_ON(!ap->ops->error_handler);
915 /* we're gonna abort all commands, no need for fast drain */
916 ata_eh_set_pending(ap, 0);
918 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
919 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
921 if (qc && (!link || qc->dev->link == link)) {
922 qc->flags |= ATA_QCFLAG_FAILED;
929 ata_port_schedule_eh(ap);
935 * ata_link_abort - abort all qc's on the link
936 * @link: ATA link to abort qc's for
938 * Abort all active qc's active on @link and schedule EH.
941 * spin_lock_irqsave(host lock)
944 * Number of aborted qc's.
946 int ata_link_abort(struct ata_link *link)
948 return ata_do_link_abort(link->ap, link);
952 * ata_port_abort - abort all qc's on the port
953 * @ap: ATA port to abort qc's for
955 * Abort all active qc's of @ap and schedule EH.
958 * spin_lock_irqsave(host_set lock)
961 * Number of aborted qc's.
963 int ata_port_abort(struct ata_port *ap)
965 return ata_do_link_abort(ap, NULL);
969 * __ata_port_freeze - freeze port
970 * @ap: ATA port to freeze
972 * This function is called when HSM violation or some other
973 * condition disrupts normal operation of the port. Frozen port
974 * is not allowed to perform any operation until the port is
975 * thawed, which usually follows a successful reset.
977 * ap->ops->freeze() callback can be used for freezing the port
978 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
979 * port cannot be frozen hardware-wise, the interrupt handler
980 * must ack and clear interrupts unconditionally while the port
984 * spin_lock_irqsave(host lock)
986 static void __ata_port_freeze(struct ata_port *ap)
988 WARN_ON(!ap->ops->error_handler);
993 ap->pflags |= ATA_PFLAG_FROZEN;
995 DPRINTK("ata%u port frozen\n", ap->print_id);
999 * ata_port_freeze - abort & freeze port
1000 * @ap: ATA port to freeze
1002 * Abort and freeze @ap.
1005 * spin_lock_irqsave(host lock)
1008 * Number of aborted commands.
1010 int ata_port_freeze(struct ata_port *ap)
1014 WARN_ON(!ap->ops->error_handler);
1016 nr_aborted = ata_port_abort(ap);
1017 __ata_port_freeze(ap);
1023 * sata_async_notification - SATA async notification handler
1024 * @ap: ATA port where async notification is received
1026 * Handler to be called when async notification via SDB FIS is
1027 * received. This function schedules EH if necessary.
1030 * spin_lock_irqsave(host lock)
1033 * 1 if EH is scheduled, 0 otherwise.
1035 int sata_async_notification(struct ata_port *ap)
1040 if (!(ap->flags & ATA_FLAG_AN))
1043 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1045 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1047 if (!sata_pmp_attached(ap) || rc) {
1048 /* PMP is not attached or SNTF is not available */
1049 if (!sata_pmp_attached(ap)) {
1050 /* PMP is not attached. Check whether ATAPI
1051 * AN is configured. If so, notify media
1054 struct ata_device *dev = ap->link.device;
1056 if ((dev->class == ATA_DEV_ATAPI) &&
1057 (dev->flags & ATA_DFLAG_AN))
1058 ata_scsi_media_change_notify(dev);
1061 /* PMP is attached but SNTF is not available.
1062 * ATAPI async media change notification is
1063 * not used. The PMP must be reporting PHY
1064 * status change, schedule EH.
1066 ata_port_schedule_eh(ap);
1070 /* PMP is attached and SNTF is available */
1071 struct ata_link *link;
1073 /* check and notify ATAPI AN */
1074 ata_for_each_link(link, ap, EDGE) {
1075 if (!(sntf & (1 << link->pmp)))
1078 if ((link->device->class == ATA_DEV_ATAPI) &&
1079 (link->device->flags & ATA_DFLAG_AN))
1080 ata_scsi_media_change_notify(link->device);
1083 /* If PMP is reporting that PHY status of some
1084 * downstream ports has changed, schedule EH.
1086 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1087 ata_port_schedule_eh(ap);
1096 * ata_eh_freeze_port - EH helper to freeze port
1097 * @ap: ATA port to freeze
1104 void ata_eh_freeze_port(struct ata_port *ap)
1106 unsigned long flags;
1108 if (!ap->ops->error_handler)
1111 spin_lock_irqsave(ap->lock, flags);
1112 __ata_port_freeze(ap);
1113 spin_unlock_irqrestore(ap->lock, flags);
1117 * ata_port_thaw_port - EH helper to thaw port
1118 * @ap: ATA port to thaw
1120 * Thaw frozen port @ap.
1125 void ata_eh_thaw_port(struct ata_port *ap)
1127 unsigned long flags;
1129 if (!ap->ops->error_handler)
1132 spin_lock_irqsave(ap->lock, flags);
1134 ap->pflags &= ~ATA_PFLAG_FROZEN;
1139 spin_unlock_irqrestore(ap->lock, flags);
1141 DPRINTK("ata%u port thawed\n", ap->print_id);
1144 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1149 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1151 struct ata_port *ap = qc->ap;
1152 struct scsi_cmnd *scmd = qc->scsicmd;
1153 unsigned long flags;
1155 spin_lock_irqsave(ap->lock, flags);
1156 qc->scsidone = ata_eh_scsidone;
1157 __ata_qc_complete(qc);
1158 WARN_ON(ata_tag_valid(qc->tag));
1159 spin_unlock_irqrestore(ap->lock, flags);
1161 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1165 * ata_eh_qc_complete - Complete an active ATA command from EH
1166 * @qc: Command to complete
1168 * Indicate to the mid and upper layers that an ATA command has
1169 * completed. To be used from EH.
1171 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1173 struct scsi_cmnd *scmd = qc->scsicmd;
1174 scmd->retries = scmd->allowed;
1175 __ata_eh_qc_complete(qc);
1179 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1180 * @qc: Command to retry
1182 * Indicate to the mid and upper layers that an ATA command
1183 * should be retried. To be used from EH.
1185 * SCSI midlayer limits the number of retries to scmd->allowed.
1186 * scmd->retries is decremented for commands which get retried
1187 * due to unrelated failures (qc->err_mask is zero).
1189 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1191 struct scsi_cmnd *scmd = qc->scsicmd;
1192 if (!qc->err_mask && scmd->retries)
1194 __ata_eh_qc_complete(qc);
1198 * ata_dev_disable - disable ATA device
1199 * @dev: ATA device to disable
1206 void ata_dev_disable(struct ata_device *dev)
1208 if (!ata_dev_enabled(dev))
1211 if (ata_msg_drv(dev->link->ap))
1212 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1213 ata_acpi_on_disable(dev);
1214 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1217 /* From now till the next successful probe, ering is used to
1218 * track probe failures. Clear accumulated device error info.
1220 ata_ering_clear(&dev->ering);
1224 * ata_eh_detach_dev - detach ATA device
1225 * @dev: ATA device to detach
1232 void ata_eh_detach_dev(struct ata_device *dev)
1234 struct ata_link *link = dev->link;
1235 struct ata_port *ap = link->ap;
1236 struct ata_eh_context *ehc = &link->eh_context;
1237 unsigned long flags;
1239 ata_dev_disable(dev);
1241 spin_lock_irqsave(ap->lock, flags);
1243 dev->flags &= ~ATA_DFLAG_DETACH;
1245 if (ata_scsi_offline_dev(dev)) {
1246 dev->flags |= ATA_DFLAG_DETACHED;
1247 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1250 /* clear per-dev EH info */
1251 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1252 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1253 ehc->saved_xfer_mode[dev->devno] = 0;
1254 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1256 spin_unlock_irqrestore(ap->lock, flags);
1260 * ata_eh_about_to_do - about to perform eh_action
1261 * @link: target ATA link
1262 * @dev: target ATA dev for per-dev action (can be NULL)
1263 * @action: action about to be performed
1265 * Called just before performing EH actions to clear related bits
1266 * in @link->eh_info such that eh actions are not unnecessarily
1272 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1273 unsigned int action)
1275 struct ata_port *ap = link->ap;
1276 struct ata_eh_info *ehi = &link->eh_info;
1277 struct ata_eh_context *ehc = &link->eh_context;
1278 unsigned long flags;
1280 spin_lock_irqsave(ap->lock, flags);
1282 ata_eh_clear_action(link, dev, ehi, action);
1284 /* About to take EH action, set RECOVERED. Ignore actions on
1285 * slave links as master will do them again.
1287 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1288 ap->pflags |= ATA_PFLAG_RECOVERED;
1290 spin_unlock_irqrestore(ap->lock, flags);
1294 * ata_eh_done - EH action complete
1295 * @ap: target ATA port
1296 * @dev: target ATA dev for per-dev action (can be NULL)
1297 * @action: action just completed
1299 * Called right after performing EH actions to clear related bits
1300 * in @link->eh_context.
1305 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1306 unsigned int action)
1308 struct ata_eh_context *ehc = &link->eh_context;
1310 ata_eh_clear_action(link, dev, &ehc->i, action);
1314 * ata_err_string - convert err_mask to descriptive string
1315 * @err_mask: error mask to convert to string
1317 * Convert @err_mask to descriptive string. Errors are
1318 * prioritized according to severity and only the most severe
1319 * error is reported.
1325 * Descriptive string for @err_mask
1327 static const char *ata_err_string(unsigned int err_mask)
1329 if (err_mask & AC_ERR_HOST_BUS)
1330 return "host bus error";
1331 if (err_mask & AC_ERR_ATA_BUS)
1332 return "ATA bus error";
1333 if (err_mask & AC_ERR_TIMEOUT)
1335 if (err_mask & AC_ERR_HSM)
1336 return "HSM violation";
1337 if (err_mask & AC_ERR_SYSTEM)
1338 return "internal error";
1339 if (err_mask & AC_ERR_MEDIA)
1340 return "media error";
1341 if (err_mask & AC_ERR_INVALID)
1342 return "invalid argument";
1343 if (err_mask & AC_ERR_DEV)
1344 return "device error";
1345 return "unknown error";
1349 * ata_read_log_page - read a specific log page
1350 * @dev: target device
1351 * @page: page to read
1352 * @buf: buffer to store read page
1353 * @sectors: number of sectors to read
1355 * Read log page using READ_LOG_EXT command.
1358 * Kernel thread context (may sleep).
1361 * 0 on success, AC_ERR_* mask otherwise.
1363 static unsigned int ata_read_log_page(struct ata_device *dev,
1364 u8 page, void *buf, unsigned int sectors)
1366 struct ata_taskfile tf;
1367 unsigned int err_mask;
1369 DPRINTK("read log page - page %d\n", page);
1371 ata_tf_init(dev, &tf);
1372 tf.command = ATA_CMD_READ_LOG_EXT;
1375 tf.hob_nsect = sectors >> 8;
1376 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1377 tf.protocol = ATA_PROT_PIO;
1379 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1380 buf, sectors * ATA_SECT_SIZE, 0);
1382 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1387 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1388 * @dev: Device to read log page 10h from
1389 * @tag: Resulting tag of the failed command
1390 * @tf: Resulting taskfile registers of the failed command
1392 * Read log page 10h to obtain NCQ error details and clear error
1396 * Kernel thread context (may sleep).
1399 * 0 on success, -errno otherwise.
1401 static int ata_eh_read_log_10h(struct ata_device *dev,
1402 int *tag, struct ata_taskfile *tf)
1404 u8 *buf = dev->link->ap->sector_buf;
1405 unsigned int err_mask;
1409 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1414 for (i = 0; i < ATA_SECT_SIZE; i++)
1417 ata_dev_printk(dev, KERN_WARNING,
1418 "invalid checksum 0x%x on log page 10h\n", csum);
1423 *tag = buf[0] & 0x1f;
1425 tf->command = buf[2];
1426 tf->feature = buf[3];
1430 tf->device = buf[7];
1431 tf->hob_lbal = buf[8];
1432 tf->hob_lbam = buf[9];
1433 tf->hob_lbah = buf[10];
1434 tf->nsect = buf[12];
1435 tf->hob_nsect = buf[13];
1441 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1442 * @dev: target ATAPI device
1443 * @r_sense_key: out parameter for sense_key
1445 * Perform ATAPI TEST_UNIT_READY.
1448 * EH context (may sleep).
1451 * 0 on success, AC_ERR_* mask on failure.
1453 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1455 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1456 struct ata_taskfile tf;
1457 unsigned int err_mask;
1459 ata_tf_init(dev, &tf);
1461 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1462 tf.command = ATA_CMD_PACKET;
1463 tf.protocol = ATAPI_PROT_NODATA;
1465 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1466 if (err_mask == AC_ERR_DEV)
1467 *r_sense_key = tf.feature >> 4;
1472 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1473 * @dev: device to perform REQUEST_SENSE to
1474 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1475 * @dfl_sense_key: default sense key to use
1477 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1478 * SENSE. This function is EH helper.
1481 * Kernel thread context (may sleep).
1484 * 0 on success, AC_ERR_* mask on failure
1486 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1487 u8 *sense_buf, u8 dfl_sense_key)
1489 u8 cdb[ATAPI_CDB_LEN] =
1490 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1491 struct ata_port *ap = dev->link->ap;
1492 struct ata_taskfile tf;
1494 DPRINTK("ATAPI request sense\n");
1496 /* FIXME: is this needed? */
1497 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1499 /* initialize sense_buf with the error register,
1500 * for the case where they are -not- overwritten
1502 sense_buf[0] = 0x70;
1503 sense_buf[2] = dfl_sense_key;
1505 /* some devices time out if garbage left in tf */
1506 ata_tf_init(dev, &tf);
1508 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1509 tf.command = ATA_CMD_PACKET;
1511 /* is it pointless to prefer PIO for "safety reasons"? */
1512 if (ap->flags & ATA_FLAG_PIO_DMA) {
1513 tf.protocol = ATAPI_PROT_DMA;
1514 tf.feature |= ATAPI_PKT_DMA;
1516 tf.protocol = ATAPI_PROT_PIO;
1517 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1521 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1522 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1526 * ata_eh_analyze_serror - analyze SError for a failed port
1527 * @link: ATA link to analyze SError for
1529 * Analyze SError if available and further determine cause of
1535 static void ata_eh_analyze_serror(struct ata_link *link)
1537 struct ata_eh_context *ehc = &link->eh_context;
1538 u32 serror = ehc->i.serror;
1539 unsigned int err_mask = 0, action = 0;
1542 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1543 err_mask |= AC_ERR_ATA_BUS;
1544 action |= ATA_EH_RESET;
1546 if (serror & SERR_PROTOCOL) {
1547 err_mask |= AC_ERR_HSM;
1548 action |= ATA_EH_RESET;
1550 if (serror & SERR_INTERNAL) {
1551 err_mask |= AC_ERR_SYSTEM;
1552 action |= ATA_EH_RESET;
1555 /* Determine whether a hotplug event has occurred. Both
1556 * SError.N/X are considered hotplug events for enabled or
1557 * host links. For disabled PMP links, only N bit is
1558 * considered as X bit is left at 1 for link plugging.
1562 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1563 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1565 hotplug_mask = SERR_PHYRDY_CHG;
1567 if (serror & hotplug_mask)
1568 ata_ehi_hotplugged(&ehc->i);
1570 ehc->i.err_mask |= err_mask;
1571 ehc->i.action |= action;
1575 * ata_eh_analyze_ncq_error - analyze NCQ error
1576 * @link: ATA link to analyze NCQ error for
1578 * Read log page 10h, determine the offending qc and acquire
1579 * error status TF. For NCQ device errors, all LLDDs have to do
1580 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1584 * Kernel thread context (may sleep).
1586 void ata_eh_analyze_ncq_error(struct ata_link *link)
1588 struct ata_port *ap = link->ap;
1589 struct ata_eh_context *ehc = &link->eh_context;
1590 struct ata_device *dev = link->device;
1591 struct ata_queued_cmd *qc;
1592 struct ata_taskfile tf;
1595 /* if frozen, we can't do much */
1596 if (ap->pflags & ATA_PFLAG_FROZEN)
1599 /* is it NCQ device error? */
1600 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1603 /* has LLDD analyzed already? */
1604 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1605 qc = __ata_qc_from_tag(ap, tag);
1607 if (!(qc->flags & ATA_QCFLAG_FAILED))
1614 /* okay, this error is ours */
1615 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1617 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1618 "(errno=%d)\n", rc);
1622 if (!(link->sactive & (1 << tag))) {
1623 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1624 "inactive tag %d\n", tag);
1628 /* we've got the perpetrator, condemn it */
1629 qc = __ata_qc_from_tag(ap, tag);
1630 memcpy(&qc->result_tf, &tf, sizeof(tf));
1631 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1632 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1633 ehc->i.err_mask &= ~AC_ERR_DEV;
1637 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1638 * @qc: qc to analyze
1639 * @tf: Taskfile registers to analyze
1641 * Analyze taskfile of @qc and further determine cause of
1642 * failure. This function also requests ATAPI sense data if
1646 * Kernel thread context (may sleep).
1649 * Determined recovery action
1651 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1652 const struct ata_taskfile *tf)
1654 unsigned int tmp, action = 0;
1655 u8 stat = tf->command, err = tf->feature;
1657 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1658 qc->err_mask |= AC_ERR_HSM;
1659 return ATA_EH_RESET;
1662 if (stat & (ATA_ERR | ATA_DF))
1663 qc->err_mask |= AC_ERR_DEV;
1667 switch (qc->dev->class) {
1670 qc->err_mask |= AC_ERR_ATA_BUS;
1672 qc->err_mask |= AC_ERR_MEDIA;
1674 qc->err_mask |= AC_ERR_INVALID;
1678 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1679 tmp = atapi_eh_request_sense(qc->dev,
1680 qc->scsicmd->sense_buffer,
1681 qc->result_tf.feature >> 4);
1683 /* ATA_QCFLAG_SENSE_VALID is used to
1684 * tell atapi_qc_complete() that sense
1685 * data is already valid.
1687 * TODO: interpret sense data and set
1688 * appropriate err_mask.
1690 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1692 qc->err_mask |= tmp;
1696 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1697 action |= ATA_EH_RESET;
1702 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1707 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1711 base = ATA_ECAT_DUBIOUS_NONE;
1713 if (err_mask & AC_ERR_ATA_BUS)
1714 return base + ATA_ECAT_ATA_BUS;
1716 if (err_mask & AC_ERR_TIMEOUT)
1717 return base + ATA_ECAT_TOUT_HSM;
1719 if (eflags & ATA_EFLAG_IS_IO) {
1720 if (err_mask & AC_ERR_HSM)
1721 return base + ATA_ECAT_TOUT_HSM;
1723 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1724 return base + ATA_ECAT_UNK_DEV;
1730 struct speed_down_verdict_arg {
1733 int nr_errors[ATA_ECAT_NR];
1736 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1738 struct speed_down_verdict_arg *arg = void_arg;
1741 if (ent->timestamp < arg->since)
1744 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1746 arg->nr_errors[cat]++;
1752 * ata_eh_speed_down_verdict - Determine speed down verdict
1753 * @dev: Device of interest
1755 * This function examines error ring of @dev and determines
1756 * whether NCQ needs to be turned off, transfer speed should be
1757 * stepped down, or falling back to PIO is necessary.
1759 * ECAT_ATA_BUS : ATA_BUS error for any command
1761 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1764 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1766 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1767 * data transfer hasn't been verified.
1771 * NCQ_OFF : Turn off NCQ.
1773 * SPEED_DOWN : Speed down transfer speed but don't fall back
1776 * FALLBACK_TO_PIO : Fall back to PIO.
1778 * Even if multiple verdicts are returned, only one action is
1779 * taken per error. An action triggered by non-DUBIOUS errors
1780 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1781 * This is to expedite speed down decisions right after device is
1782 * initially configured.
1784 * The followings are speed down rules. #1 and #2 deal with
1787 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1788 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1790 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1791 * occurred during last 5 mins, NCQ_OFF.
1793 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1794 * ocurred during last 5 mins, FALLBACK_TO_PIO
1796 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1797 * during last 10 mins, NCQ_OFF.
1799 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1800 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1803 * Inherited from caller.
1806 * OR of ATA_EH_SPDN_* flags.
1808 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1810 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1811 u64 j64 = get_jiffies_64();
1812 struct speed_down_verdict_arg arg;
1813 unsigned int verdict = 0;
1815 /* scan past 5 mins of error history */
1816 memset(&arg, 0, sizeof(arg));
1817 arg.since = j64 - min(j64, j5mins);
1818 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1820 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1821 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1822 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1823 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1825 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1826 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1827 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1829 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1830 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1831 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1832 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1834 /* scan past 10 mins of error history */
1835 memset(&arg, 0, sizeof(arg));
1836 arg.since = j64 - min(j64, j10mins);
1837 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1839 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1840 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1841 verdict |= ATA_EH_SPDN_NCQ_OFF;
1843 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1844 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1845 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1846 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1852 * ata_eh_speed_down - record error and speed down if necessary
1853 * @dev: Failed device
1854 * @eflags: mask of ATA_EFLAG_* flags
1855 * @err_mask: err_mask of the error
1857 * Record error and examine error history to determine whether
1858 * adjusting transmission speed is necessary. It also sets
1859 * transmission limits appropriately if such adjustment is
1863 * Kernel thread context (may sleep).
1866 * Determined recovery action.
1868 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1869 unsigned int eflags, unsigned int err_mask)
1871 struct ata_link *link = ata_dev_phys_link(dev);
1873 unsigned int verdict;
1874 unsigned int action = 0;
1876 /* don't bother if Cat-0 error */
1877 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1880 /* record error and determine whether speed down is necessary */
1881 ata_ering_record(&dev->ering, eflags, err_mask);
1882 verdict = ata_eh_speed_down_verdict(dev);
1885 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1886 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1887 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1888 dev->flags |= ATA_DFLAG_NCQ_OFF;
1889 ata_dev_printk(dev, KERN_WARNING,
1890 "NCQ disabled due to excessive errors\n");
1895 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1896 /* speed down SATA link speed if possible */
1897 if (sata_down_spd_limit(link, 0) == 0) {
1898 action |= ATA_EH_RESET;
1902 /* lower transfer mode */
1903 if (dev->spdn_cnt < 2) {
1904 static const int dma_dnxfer_sel[] =
1905 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1906 static const int pio_dnxfer_sel[] =
1907 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1910 if (dev->xfer_shift != ATA_SHIFT_PIO)
1911 sel = dma_dnxfer_sel[dev->spdn_cnt];
1913 sel = pio_dnxfer_sel[dev->spdn_cnt];
1917 if (ata_down_xfermask_limit(dev, sel) == 0) {
1918 action |= ATA_EH_RESET;
1924 /* Fall back to PIO? Slowing down to PIO is meaningless for
1925 * SATA ATA devices. Consider it only for PATA and SATAPI.
1927 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1928 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1929 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1930 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1932 action |= ATA_EH_RESET;
1939 /* device has been slowed down, blow error history */
1940 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1941 ata_ering_clear(&dev->ering);
1946 * ata_eh_link_autopsy - analyze error and determine recovery action
1947 * @link: host link to perform autopsy on
1949 * Analyze why @link failed and determine which recovery actions
1950 * are needed. This function also sets more detailed AC_ERR_*
1951 * values and fills sense data for ATAPI CHECK SENSE.
1954 * Kernel thread context (may sleep).
1956 static void ata_eh_link_autopsy(struct ata_link *link)
1958 struct ata_port *ap = link->ap;
1959 struct ata_eh_context *ehc = &link->eh_context;
1960 struct ata_device *dev;
1961 unsigned int all_err_mask = 0, eflags = 0;
1968 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1971 /* obtain and analyze SError */
1972 rc = sata_scr_read(link, SCR_ERROR, &serror);
1974 ehc->i.serror |= serror;
1975 ata_eh_analyze_serror(link);
1976 } else if (rc != -EOPNOTSUPP) {
1977 /* SError read failed, force reset and probing */
1978 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1979 ehc->i.action |= ATA_EH_RESET;
1980 ehc->i.err_mask |= AC_ERR_OTHER;
1983 /* analyze NCQ failure */
1984 ata_eh_analyze_ncq_error(link);
1986 /* any real error trumps AC_ERR_OTHER */
1987 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1988 ehc->i.err_mask &= ~AC_ERR_OTHER;
1990 all_err_mask |= ehc->i.err_mask;
1992 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1993 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1995 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1996 ata_dev_phys_link(qc->dev) != link)
1999 /* inherit upper level err_mask */
2000 qc->err_mask |= ehc->i.err_mask;
2003 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2005 /* DEV errors are probably spurious in case of ATA_BUS error */
2006 if (qc->err_mask & AC_ERR_ATA_BUS)
2007 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2010 /* any real error trumps unknown error */
2011 if (qc->err_mask & ~AC_ERR_OTHER)
2012 qc->err_mask &= ~AC_ERR_OTHER;
2014 /* SENSE_VALID trumps dev/unknown error and revalidation */
2015 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2016 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2018 /* determine whether the command is worth retrying */
2019 if (!(qc->err_mask & AC_ERR_INVALID) &&
2020 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
2021 qc->flags |= ATA_QCFLAG_RETRY;
2023 /* accumulate error info */
2024 ehc->i.dev = qc->dev;
2025 all_err_mask |= qc->err_mask;
2026 if (qc->flags & ATA_QCFLAG_IO)
2027 eflags |= ATA_EFLAG_IS_IO;
2030 /* enforce default EH actions */
2031 if (ap->pflags & ATA_PFLAG_FROZEN ||
2032 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2033 ehc->i.action |= ATA_EH_RESET;
2034 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2035 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2036 ehc->i.action |= ATA_EH_REVALIDATE;
2038 /* If we have offending qcs and the associated failed device,
2039 * perform per-dev EH action only on the offending device.
2042 ehc->i.dev_action[ehc->i.dev->devno] |=
2043 ehc->i.action & ATA_EH_PERDEV_MASK;
2044 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2047 /* propagate timeout to host link */
2048 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2049 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2051 /* record error and consider speeding down */
2053 if (!dev && ((ata_link_max_devices(link) == 1 &&
2054 ata_dev_enabled(link->device))))
2058 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2059 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2060 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2067 * ata_eh_autopsy - analyze error and determine recovery action
2068 * @ap: host port to perform autopsy on
2070 * Analyze all links of @ap and determine why they failed and
2071 * which recovery actions are needed.
2074 * Kernel thread context (may sleep).
2076 void ata_eh_autopsy(struct ata_port *ap)
2078 struct ata_link *link;
2080 ata_for_each_link(link, ap, EDGE)
2081 ata_eh_link_autopsy(link);
2083 /* Handle the frigging slave link. Autopsy is done similarly
2084 * but actions and flags are transferred over to the master
2085 * link and handled from there.
2087 if (ap->slave_link) {
2088 struct ata_eh_context *mehc = &ap->link.eh_context;
2089 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2091 /* transfer control flags from master to slave */
2092 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2094 /* perform autopsy on the slave link */
2095 ata_eh_link_autopsy(ap->slave_link);
2097 /* transfer actions from slave to master and clear slave */
2098 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2099 mehc->i.action |= sehc->i.action;
2100 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2101 mehc->i.flags |= sehc->i.flags;
2102 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2105 /* Autopsy of fanout ports can affect host link autopsy.
2106 * Perform host link autopsy last.
2108 if (sata_pmp_attached(ap))
2109 ata_eh_link_autopsy(&ap->link);
2113 * ata_eh_link_report - report error handling to user
2114 * @link: ATA link EH is going on
2116 * Report EH to user.
2121 static void ata_eh_link_report(struct ata_link *link)
2123 struct ata_port *ap = link->ap;
2124 struct ata_eh_context *ehc = &link->eh_context;
2125 const char *frozen, *desc;
2127 int tag, nr_failed = 0;
2129 if (ehc->i.flags & ATA_EHI_QUIET)
2133 if (ehc->i.desc[0] != '\0')
2136 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2137 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2139 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2140 ata_dev_phys_link(qc->dev) != link ||
2141 ((qc->flags & ATA_QCFLAG_QUIET) &&
2142 qc->err_mask == AC_ERR_DEV))
2144 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2150 if (!nr_failed && !ehc->i.err_mask)
2154 if (ap->pflags & ATA_PFLAG_FROZEN)
2157 memset(tries_buf, 0, sizeof(tries_buf));
2158 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2159 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2163 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2164 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2165 ehc->i.err_mask, link->sactive, ehc->i.serror,
2166 ehc->i.action, frozen, tries_buf);
2168 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2170 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2171 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2172 ehc->i.err_mask, link->sactive, ehc->i.serror,
2173 ehc->i.action, frozen, tries_buf);
2175 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2179 ata_link_printk(link, KERN_ERR,
2180 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2181 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2182 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2183 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2184 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2185 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2186 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2187 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2188 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2189 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2190 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2191 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2192 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2193 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2194 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2195 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2196 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2197 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2199 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2200 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2201 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2202 const u8 *cdb = qc->cdb;
2203 char data_buf[20] = "";
2204 char cdb_buf[70] = "";
2206 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2207 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2210 if (qc->dma_dir != DMA_NONE) {
2211 static const char *dma_str[] = {
2212 [DMA_BIDIRECTIONAL] = "bidi",
2213 [DMA_TO_DEVICE] = "out",
2214 [DMA_FROM_DEVICE] = "in",
2216 static const char *prot_str[] = {
2217 [ATA_PROT_PIO] = "pio",
2218 [ATA_PROT_DMA] = "dma",
2219 [ATA_PROT_NCQ] = "ncq",
2220 [ATAPI_PROT_PIO] = "pio",
2221 [ATAPI_PROT_DMA] = "dma",
2224 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2225 prot_str[qc->tf.protocol], qc->nbytes,
2226 dma_str[qc->dma_dir]);
2229 if (ata_is_atapi(qc->tf.protocol))
2230 snprintf(cdb_buf, sizeof(cdb_buf),
2231 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2232 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2233 cdb[0], cdb[1], cdb[2], cdb[3],
2234 cdb[4], cdb[5], cdb[6], cdb[7],
2235 cdb[8], cdb[9], cdb[10], cdb[11],
2236 cdb[12], cdb[13], cdb[14], cdb[15]);
2238 ata_dev_printk(qc->dev, KERN_ERR,
2239 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2241 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2242 "Emask 0x%x (%s)%s\n",
2243 cmd->command, cmd->feature, cmd->nsect,
2244 cmd->lbal, cmd->lbam, cmd->lbah,
2245 cmd->hob_feature, cmd->hob_nsect,
2246 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2247 cmd->device, qc->tag, data_buf, cdb_buf,
2248 res->command, res->feature, res->nsect,
2249 res->lbal, res->lbam, res->lbah,
2250 res->hob_feature, res->hob_nsect,
2251 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2252 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2253 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2255 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2257 if (res->command & ATA_BUSY)
2258 ata_dev_printk(qc->dev, KERN_ERR,
2259 "status: { Busy }\n");
2261 ata_dev_printk(qc->dev, KERN_ERR,
2262 "status: { %s%s%s%s}\n",
2263 res->command & ATA_DRDY ? "DRDY " : "",
2264 res->command & ATA_DF ? "DF " : "",
2265 res->command & ATA_DRQ ? "DRQ " : "",
2266 res->command & ATA_ERR ? "ERR " : "");
2269 if (cmd->command != ATA_CMD_PACKET &&
2270 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2272 ata_dev_printk(qc->dev, KERN_ERR,
2273 "error: { %s%s%s%s}\n",
2274 res->feature & ATA_ICRC ? "ICRC " : "",
2275 res->feature & ATA_UNC ? "UNC " : "",
2276 res->feature & ATA_IDNF ? "IDNF " : "",
2277 res->feature & ATA_ABORTED ? "ABRT " : "");
2282 * ata_eh_report - report error handling to user
2283 * @ap: ATA port to report EH about
2285 * Report EH to user.
2290 void ata_eh_report(struct ata_port *ap)
2292 struct ata_link *link;
2294 ata_for_each_link(link, ap, HOST_FIRST)
2295 ata_eh_link_report(link);
2298 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2299 unsigned int *classes, unsigned long deadline,
2302 struct ata_device *dev;
2305 ata_for_each_dev(dev, link, ALL)
2306 classes[dev->devno] = ATA_DEV_UNKNOWN;
2308 return reset(link, classes, deadline);
2311 static int ata_eh_followup_srst_needed(struct ata_link *link,
2312 int rc, const unsigned int *classes)
2314 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2318 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2323 int ata_eh_reset(struct ata_link *link, int classify,
2324 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2325 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2327 struct ata_port *ap = link->ap;
2328 struct ata_link *slave = ap->slave_link;
2329 struct ata_eh_context *ehc = &link->eh_context;
2330 struct ata_eh_context *sehc = &slave->eh_context;
2331 unsigned int *classes = ehc->classes;
2332 unsigned int lflags = link->flags;
2333 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2334 int max_tries = 0, try = 0;
2335 struct ata_link *failed_link;
2336 struct ata_device *dev;
2337 unsigned long deadline, now;
2338 ata_reset_fn_t reset;
2339 unsigned long flags;
2346 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2348 if (link->flags & ATA_LFLAG_NO_HRST)
2350 if (link->flags & ATA_LFLAG_NO_SRST)
2353 /* make sure each reset attemp is at least COOL_DOWN apart */
2354 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2356 WARN_ON(time_after(ehc->last_reset, now));
2357 deadline = ata_deadline(ehc->last_reset,
2358 ATA_EH_RESET_COOL_DOWN);
2359 if (time_before(now, deadline))
2360 schedule_timeout_uninterruptible(deadline - now);
2363 spin_lock_irqsave(ap->lock, flags);
2364 ap->pflags |= ATA_PFLAG_RESETTING;
2365 spin_unlock_irqrestore(ap->lock, flags);
2367 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2369 ata_for_each_dev(dev, link, ALL) {
2370 /* If we issue an SRST then an ATA drive (not ATAPI)
2371 * may change configuration and be in PIO0 timing. If
2372 * we do a hard reset (or are coming from power on)
2373 * this is true for ATA or ATAPI. Until we've set a
2374 * suitable controller mode we should not touch the
2375 * bus as we may be talking too fast.
2377 dev->pio_mode = XFER_PIO_0;
2379 /* If the controller has a pio mode setup function
2380 * then use it to set the chipset to rights. Don't
2381 * touch the DMA setup as that will be dealt with when
2382 * configuring devices.
2384 if (ap->ops->set_piomode)
2385 ap->ops->set_piomode(ap, dev);
2388 /* prefer hardreset */
2390 ehc->i.action &= ~ATA_EH_RESET;
2393 ehc->i.action |= ATA_EH_HARDRESET;
2394 } else if (softreset) {
2396 ehc->i.action |= ATA_EH_SOFTRESET;
2400 unsigned long deadline = ata_deadline(jiffies,
2401 ATA_EH_PRERESET_TIMEOUT);
2404 sehc->i.action &= ~ATA_EH_RESET;
2405 sehc->i.action |= ehc->i.action;
2408 rc = prereset(link, deadline);
2410 /* If present, do prereset on slave link too. Reset
2411 * is skipped iff both master and slave links report
2412 * -ENOENT or clear ATA_EH_RESET.
2414 if (slave && (rc == 0 || rc == -ENOENT)) {
2417 tmp = prereset(slave, deadline);
2421 ehc->i.action |= sehc->i.action;
2425 if (rc == -ENOENT) {
2426 ata_link_printk(link, KERN_DEBUG,
2427 "port disabled. ignoring.\n");
2428 ehc->i.action &= ~ATA_EH_RESET;
2430 ata_for_each_dev(dev, link, ALL)
2431 classes[dev->devno] = ATA_DEV_NONE;
2435 ata_link_printk(link, KERN_ERR,
2436 "prereset failed (errno=%d)\n", rc);
2440 /* prereset() might have cleared ATA_EH_RESET. If so,
2441 * bang classes, thaw and return.
2443 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2444 ata_for_each_dev(dev, link, ALL)
2445 classes[dev->devno] = ATA_DEV_NONE;
2446 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2447 ata_is_host_link(link))
2448 ata_eh_thaw_port(ap);
2458 if (ata_is_host_link(link))
2459 ata_eh_freeze_port(ap);
2461 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2465 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2466 reset == softreset ? "soft" : "hard");
2468 /* mark that this EH session started with reset */
2469 ehc->last_reset = jiffies;
2470 if (reset == hardreset)
2471 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2473 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2475 rc = ata_do_reset(link, reset, classes, deadline, true);
2476 if (rc && rc != -EAGAIN) {
2481 /* hardreset slave link if existent */
2482 if (slave && reset == hardreset) {
2486 ata_link_printk(slave, KERN_INFO,
2487 "hard resetting link\n");
2489 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2490 tmp = ata_do_reset(slave, reset, classes, deadline,
2498 failed_link = slave;
2504 /* perform follow-up SRST if necessary */
2505 if (reset == hardreset &&
2506 ata_eh_followup_srst_needed(link, rc, classes)) {
2510 ata_link_printk(link, KERN_ERR,
2511 "follow-up softreset required "
2512 "but no softreset avaliable\n");
2518 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2519 rc = ata_do_reset(link, reset, classes, deadline, true);
2523 ata_link_printk(link, KERN_INFO, "no reset method "
2524 "available, skipping reset\n");
2525 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2526 lflags |= ATA_LFLAG_ASSUME_ATA;
2530 * Post-reset processing
2532 ata_for_each_dev(dev, link, ALL) {
2533 /* After the reset, the device state is PIO 0 and the
2534 * controller state is undefined. Reset also wakes up
2535 * drives from sleeping mode.
2537 dev->pio_mode = XFER_PIO_0;
2538 dev->flags &= ~ATA_DFLAG_SLEEPING;
2540 if (!ata_phys_link_offline(ata_dev_phys_link(dev))) {
2541 /* apply class override */
2542 if (lflags & ATA_LFLAG_ASSUME_ATA)
2543 classes[dev->devno] = ATA_DEV_ATA;
2544 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2545 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2547 classes[dev->devno] = ATA_DEV_NONE;
2550 /* record current link speed */
2551 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2552 link->sata_spd = (sstatus >> 4) & 0xf;
2553 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2554 slave->sata_spd = (sstatus >> 4) & 0xf;
2557 if (ata_is_host_link(link))
2558 ata_eh_thaw_port(ap);
2560 /* postreset() should clear hardware SError. Although SError
2561 * is cleared during link resume, clearing SError here is
2562 * necessary as some PHYs raise hotplug events after SRST.
2563 * This introduces race condition where hotplug occurs between
2564 * reset and here. This race is mediated by cross checking
2565 * link onlineness and classification result later.
2568 postreset(link, classes);
2570 postreset(slave, classes);
2573 /* clear cached SError */
2574 spin_lock_irqsave(link->ap->lock, flags);
2575 link->eh_info.serror = 0;
2577 slave->eh_info.serror = 0;
2578 spin_unlock_irqrestore(link->ap->lock, flags);
2580 /* Make sure onlineness and classification result correspond.
2581 * Hotplug could have happened during reset and some
2582 * controllers fail to wait while a drive is spinning up after
2583 * being hotplugged causing misdetection. By cross checking
2584 * link onlineness and classification result, those conditions
2585 * can be reliably detected and retried.
2588 ata_for_each_dev(dev, link, ALL) {
2589 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2590 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2591 classes[dev->devno] = ATA_DEV_NONE;
2592 if (ata_phys_link_online(ata_dev_phys_link(dev)))
2597 if (classify && nr_unknown) {
2598 if (try < max_tries) {
2599 ata_link_printk(link, KERN_WARNING, "link online but "
2600 "device misclassified, retrying\n");
2605 ata_link_printk(link, KERN_WARNING,
2606 "link online but device misclassified, "
2607 "device detection might fail\n");
2610 /* reset successful, schedule revalidation */
2611 ata_eh_done(link, NULL, ATA_EH_RESET);
2613 ata_eh_done(slave, NULL, ATA_EH_RESET);
2614 ehc->last_reset = jiffies; /* update to completion time */
2615 ehc->i.action |= ATA_EH_REVALIDATE;
2619 /* clear hotplug flag */
2620 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2622 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2624 spin_lock_irqsave(ap->lock, flags);
2625 ap->pflags &= ~ATA_PFLAG_RESETTING;
2626 spin_unlock_irqrestore(ap->lock, flags);
2631 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2632 if (!ata_is_host_link(link) &&
2633 sata_scr_read(link, SCR_STATUS, &sstatus))
2636 if (rc == -ERESTART || try >= max_tries)
2640 if (time_before(now, deadline)) {
2641 unsigned long delta = deadline - now;
2643 ata_link_printk(failed_link, KERN_WARNING,
2644 "reset failed (errno=%d), retrying in %u secs\n",
2645 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2648 delta = schedule_timeout_uninterruptible(delta);
2651 if (try == max_tries - 1) {
2652 sata_down_spd_limit(link, 0);
2654 sata_down_spd_limit(slave, 0);
2655 } else if (rc == -EPIPE)
2656 sata_down_spd_limit(failed_link, 0);
2663 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2665 struct ata_link *link;
2666 struct ata_device *dev;
2667 unsigned long flags;
2670 * This function can be thought of as an extended version of
2671 * ata_eh_about_to_do() specially crafted to accommodate the
2672 * requirements of ATA_EH_PARK handling. Since the EH thread
2673 * does not leave the do {} while () loop in ata_eh_recover as
2674 * long as the timeout for a park request to *one* device on
2675 * the port has not expired, and since we still want to pick
2676 * up park requests to other devices on the same port or
2677 * timeout updates for the same device, we have to pull
2678 * ATA_EH_PARK actions from eh_info into eh_context.i
2679 * ourselves at the beginning of each pass over the loop.
2681 * Additionally, all write accesses to &ap->park_req_pending
2682 * through INIT_COMPLETION() (see below) or complete_all()
2683 * (see ata_scsi_park_store()) are protected by the host lock.
2684 * As a result we have that park_req_pending.done is zero on
2685 * exit from this function, i.e. when ATA_EH_PARK actions for
2686 * *all* devices on port ap have been pulled into the
2687 * respective eh_context structs. If, and only if,
2688 * park_req_pending.done is non-zero by the time we reach
2689 * wait_for_completion_timeout(), another ATA_EH_PARK action
2690 * has been scheduled for at least one of the devices on port
2691 * ap and we have to cycle over the do {} while () loop in
2692 * ata_eh_recover() again.
2695 spin_lock_irqsave(ap->lock, flags);
2696 INIT_COMPLETION(ap->park_req_pending);
2697 ata_for_each_link(link, ap, EDGE) {
2698 ata_for_each_dev(dev, link, ALL) {
2699 struct ata_eh_info *ehi = &link->eh_info;
2701 link->eh_context.i.dev_action[dev->devno] |=
2702 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2703 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2706 spin_unlock_irqrestore(ap->lock, flags);
2709 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2711 struct ata_eh_context *ehc = &dev->link->eh_context;
2712 struct ata_taskfile tf;
2713 unsigned int err_mask;
2715 ata_tf_init(dev, &tf);
2717 ehc->unloaded_mask |= 1 << dev->devno;
2718 tf.command = ATA_CMD_IDLEIMMEDIATE;
2724 ehc->unloaded_mask &= ~(1 << dev->devno);
2725 tf.command = ATA_CMD_CHK_POWER;
2728 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2729 tf.protocol |= ATA_PROT_NODATA;
2730 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2731 if (park && (err_mask || tf.lbal != 0xc4)) {
2732 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2733 ehc->unloaded_mask &= ~(1 << dev->devno);
2737 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2738 struct ata_device **r_failed_dev)
2740 struct ata_port *ap = link->ap;
2741 struct ata_eh_context *ehc = &link->eh_context;
2742 struct ata_device *dev;
2743 unsigned int new_mask = 0;
2744 unsigned long flags;
2749 /* For PATA drive side cable detection to work, IDENTIFY must
2750 * be done backwards such that PDIAG- is released by the slave
2751 * device before the master device is identified.
2753 ata_for_each_dev(dev, link, ALL_REVERSE) {
2754 unsigned int action = ata_eh_dev_action(dev);
2755 unsigned int readid_flags = 0;
2757 if (ehc->i.flags & ATA_EHI_DID_RESET)
2758 readid_flags |= ATA_READID_POSTRESET;
2760 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2761 WARN_ON(dev->class == ATA_DEV_PMP);
2763 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2768 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2769 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2774 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2776 /* Configuration may have changed, reconfigure
2779 ehc->i.flags |= ATA_EHI_SETMODE;
2781 /* schedule the scsi_rescan_device() here */
2782 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2783 } else if (dev->class == ATA_DEV_UNKNOWN &&
2784 ehc->tries[dev->devno] &&
2785 ata_class_enabled(ehc->classes[dev->devno])) {
2786 dev->class = ehc->classes[dev->devno];
2788 if (dev->class == ATA_DEV_PMP)
2789 rc = sata_pmp_attach(dev);
2791 rc = ata_dev_read_id(dev, &dev->class,
2792 readid_flags, dev->id);
2795 /* clear error info accumulated during probe */
2796 ata_ering_clear(&dev->ering);
2797 new_mask |= 1 << dev->devno;
2800 /* IDENTIFY was issued to non-existent
2801 * device. No need to reset. Just
2802 * thaw and kill the device.
2804 ata_eh_thaw_port(ap);
2805 dev->class = ATA_DEV_UNKNOWN;
2808 dev->class = ATA_DEV_UNKNOWN;
2814 /* PDIAG- should have been released, ask cable type if post-reset */
2815 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2816 if (ap->ops->cable_detect)
2817 ap->cbl = ap->ops->cable_detect(ap);
2821 /* Configure new devices forward such that user doesn't see
2822 * device detection messages backwards.
2824 ata_for_each_dev(dev, link, ALL) {
2825 if (!(new_mask & (1 << dev->devno)) ||
2826 dev->class == ATA_DEV_PMP)
2829 ehc->i.flags |= ATA_EHI_PRINTINFO;
2830 rc = ata_dev_configure(dev);
2831 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2835 spin_lock_irqsave(ap->lock, flags);
2836 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2837 spin_unlock_irqrestore(ap->lock, flags);
2839 /* new device discovered, configure xfermode */
2840 ehc->i.flags |= ATA_EHI_SETMODE;
2846 *r_failed_dev = dev;
2847 DPRINTK("EXIT rc=%d\n", rc);
2852 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2853 * @link: link on which timings will be programmed
2854 * @r_failed_dev: out paramter for failed device
2856 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2857 * ata_set_mode() fails, pointer to the failing device is
2858 * returned in @r_failed_dev.
2861 * PCI/etc. bus probe sem.
2864 * 0 on success, negative errno otherwise
2866 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2868 struct ata_port *ap = link->ap;
2869 struct ata_device *dev;
2872 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2873 ata_for_each_dev(dev, link, ENABLED) {
2874 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2875 struct ata_ering_entry *ent;
2877 ent = ata_ering_top(&dev->ering);
2879 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2883 /* has private set_mode? */
2884 if (ap->ops->set_mode)
2885 rc = ap->ops->set_mode(link, r_failed_dev);
2887 rc = ata_do_set_mode(link, r_failed_dev);
2889 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2890 ata_for_each_dev(dev, link, ENABLED) {
2891 struct ata_eh_context *ehc = &link->eh_context;
2892 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2893 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2895 if (dev->xfer_mode != saved_xfer_mode ||
2896 ata_ncq_enabled(dev) != saved_ncq)
2897 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2904 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2905 * @dev: ATAPI device to clear UA for
2907 * Resets and other operations can make an ATAPI device raise
2908 * UNIT ATTENTION which causes the next operation to fail. This
2909 * function clears UA.
2912 * EH context (may sleep).
2915 * 0 on success, -errno on failure.
2917 static int atapi_eh_clear_ua(struct ata_device *dev)
2921 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
2922 u8 *sense_buffer = dev->link->ap->sector_buf;
2924 unsigned int err_mask;
2926 err_mask = atapi_eh_tur(dev, &sense_key);
2927 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
2928 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
2929 "failed (err_mask=0x%x)\n", err_mask);
2933 if (!err_mask || sense_key != UNIT_ATTENTION)
2936 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
2938 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
2939 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
2944 ata_dev_printk(dev, KERN_WARNING,
2945 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
2950 static int ata_link_nr_enabled(struct ata_link *link)
2952 struct ata_device *dev;
2955 ata_for_each_dev(dev, link, ENABLED)
2960 static int ata_link_nr_vacant(struct ata_link *link)
2962 struct ata_device *dev;
2965 ata_for_each_dev(dev, link, ALL)
2966 if (dev->class == ATA_DEV_UNKNOWN)
2971 static int ata_eh_skip_recovery(struct ata_link *link)
2973 struct ata_port *ap = link->ap;
2974 struct ata_eh_context *ehc = &link->eh_context;
2975 struct ata_device *dev;
2977 /* skip disabled links */
2978 if (link->flags & ATA_LFLAG_DISABLED)
2981 /* thaw frozen port and recover failed devices */
2982 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2985 /* reset at least once if reset is requested */
2986 if ((ehc->i.action & ATA_EH_RESET) &&
2987 !(ehc->i.flags & ATA_EHI_DID_RESET))
2990 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2991 ata_for_each_dev(dev, link, ALL) {
2992 if (dev->class == ATA_DEV_UNKNOWN &&
2993 ehc->classes[dev->devno] != ATA_DEV_NONE)
3000 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3002 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3003 u64 now = get_jiffies_64();
3004 int *trials = void_arg;
3006 if (ent->timestamp < now - min(now, interval))
3013 static int ata_eh_schedule_probe(struct ata_device *dev)
3015 struct ata_eh_context *ehc = &dev->link->eh_context;
3016 struct ata_link *link = ata_dev_phys_link(dev);
3019 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3020 (ehc->did_probe_mask & (1 << dev->devno)))
3023 ata_eh_detach_dev(dev);
3025 ehc->did_probe_mask |= (1 << dev->devno);
3026 ehc->i.action |= ATA_EH_RESET;
3027 ehc->saved_xfer_mode[dev->devno] = 0;
3028 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3030 /* Record and count probe trials on the ering. The specific
3031 * error mask used is irrelevant. Because a successful device
3032 * detection clears the ering, this count accumulates only if
3033 * there are consecutive failed probes.
3035 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3036 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3037 * forced to 1.5Gbps.
3039 * This is to work around cases where failed link speed
3040 * negotiation results in device misdetection leading to
3041 * infinite DEVXCHG or PHRDY CHG events.
3043 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3044 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3046 if (trials > ATA_EH_PROBE_TRIALS)
3047 sata_down_spd_limit(link, 1);
3052 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3054 struct ata_eh_context *ehc = &dev->link->eh_context;
3056 /* -EAGAIN from EH routine indicates retry without prejudice.
3057 * The requester is responsible for ensuring forward progress.
3060 ehc->tries[dev->devno]--;
3064 /* device missing or wrong IDENTIFY data, schedule probing */
3065 ehc->i.probe_mask |= (1 << dev->devno);
3067 /* give it just one more chance */
3068 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3070 if (ehc->tries[dev->devno] == 1) {
3071 /* This is the last chance, better to slow
3072 * down than lose it.
3074 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3075 if (dev->pio_mode > XFER_PIO_0)
3076 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3080 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3081 /* disable device if it has used up all its chances */
3082 ata_dev_disable(dev);
3084 /* detach if offline */
3085 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3086 ata_eh_detach_dev(dev);
3088 /* schedule probe if necessary */
3089 if (ata_eh_schedule_probe(dev)) {
3090 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3091 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3092 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3097 ehc->i.action |= ATA_EH_RESET;
3103 * ata_eh_recover - recover host port after error
3104 * @ap: host port to recover
3105 * @prereset: prereset method (can be NULL)
3106 * @softreset: softreset method (can be NULL)
3107 * @hardreset: hardreset method (can be NULL)
3108 * @postreset: postreset method (can be NULL)
3109 * @r_failed_link: out parameter for failed link
3111 * This is the alpha and omega, eum and yang, heart and soul of
3112 * libata exception handling. On entry, actions required to
3113 * recover each link and hotplug requests are recorded in the
3114 * link's eh_context. This function executes all the operations
3115 * with appropriate retrials and fallbacks to resurrect failed
3116 * devices, detach goners and greet newcomers.
3119 * Kernel thread context (may sleep).
3122 * 0 on success, -errno on failure.
3124 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3125 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3126 ata_postreset_fn_t postreset,
3127 struct ata_link **r_failed_link)
3129 struct ata_link *link;
3130 struct ata_device *dev;
3133 unsigned long flags, deadline;
3137 /* prep for recovery */
3138 ata_for_each_link(link, ap, EDGE) {
3139 struct ata_eh_context *ehc = &link->eh_context;
3141 /* re-enable link? */
3142 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3143 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3144 spin_lock_irqsave(ap->lock, flags);
3145 link->flags &= ~ATA_LFLAG_DISABLED;
3146 spin_unlock_irqrestore(ap->lock, flags);
3147 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3150 ata_for_each_dev(dev, link, ALL) {
3151 if (link->flags & ATA_LFLAG_NO_RETRY)
3152 ehc->tries[dev->devno] = 1;
3154 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3156 /* collect port action mask recorded in dev actions */
3157 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3158 ~ATA_EH_PERDEV_MASK;
3159 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3161 /* process hotplug request */
3162 if (dev->flags & ATA_DFLAG_DETACH)
3163 ata_eh_detach_dev(dev);
3165 /* schedule probe if necessary */
3166 if (!ata_dev_enabled(dev))
3167 ata_eh_schedule_probe(dev);
3175 /* if UNLOADING, finish immediately */
3176 if (ap->pflags & ATA_PFLAG_UNLOADING)
3180 ata_for_each_link(link, ap, EDGE) {
3181 struct ata_eh_context *ehc = &link->eh_context;
3183 /* skip EH if possible. */
3184 if (ata_eh_skip_recovery(link))
3187 ata_for_each_dev(dev, link, ALL)
3188 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3192 ata_for_each_link(link, ap, EDGE) {
3193 struct ata_eh_context *ehc = &link->eh_context;
3195 if (!(ehc->i.action & ATA_EH_RESET))
3198 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3199 prereset, softreset, hardreset, postreset);
3201 ata_link_printk(link, KERN_ERR,
3202 "reset failed, giving up\n");
3211 * clears ATA_EH_PARK in eh_info and resets
3212 * ap->park_req_pending
3214 ata_eh_pull_park_action(ap);
3217 ata_for_each_link(link, ap, EDGE) {
3218 ata_for_each_dev(dev, link, ALL) {
3219 struct ata_eh_context *ehc = &link->eh_context;
3222 if (dev->class != ATA_DEV_ATA)
3224 if (!(ehc->i.dev_action[dev->devno] &
3227 tmp = dev->unpark_deadline;
3228 if (time_before(deadline, tmp))
3230 else if (time_before_eq(tmp, jiffies))
3232 if (ehc->unloaded_mask & (1 << dev->devno))
3235 ata_eh_park_issue_cmd(dev, 1);
3240 if (time_before_eq(deadline, now))
3243 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3246 ata_for_each_link(link, ap, EDGE) {
3247 ata_for_each_dev(dev, link, ALL) {
3248 if (!(link->eh_context.unloaded_mask &
3252 ata_eh_park_issue_cmd(dev, 0);
3253 ata_eh_done(link, dev, ATA_EH_PARK);
3258 ata_for_each_link(link, ap, EDGE) {
3259 struct ata_eh_context *ehc = &link->eh_context;
3261 /* revalidate existing devices and attach new ones */
3262 rc = ata_eh_revalidate_and_attach(link, &dev);
3266 /* if PMP got attached, return, pmp EH will take care of it */
3267 if (link->device->class == ATA_DEV_PMP) {
3272 /* configure transfer mode if necessary */
3273 if (ehc->i.flags & ATA_EHI_SETMODE) {
3274 rc = ata_set_mode(link, &dev);
3277 ehc->i.flags &= ~ATA_EHI_SETMODE;
3280 /* If reset has been issued, clear UA to avoid
3281 * disrupting the current users of the device.
3283 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3284 ata_for_each_dev(dev, link, ALL) {
3285 if (dev->class != ATA_DEV_ATAPI)
3287 rc = atapi_eh_clear_ua(dev);
3293 /* configure link power saving */
3294 if (ehc->i.action & ATA_EH_LPM)
3295 ata_for_each_dev(dev, link, ALL)
3296 ata_dev_enable_pm(dev, ap->pm_policy);
3298 /* this link is okay now */
3304 ata_eh_handle_dev_fail(dev, rc);
3306 if (ap->pflags & ATA_PFLAG_FROZEN) {
3307 /* PMP reset requires working host port.
3308 * Can't retry if it's frozen.
3310 if (sata_pmp_attached(ap))
3320 if (rc && r_failed_link)
3321 *r_failed_link = link;
3323 DPRINTK("EXIT, rc=%d\n", rc);
3328 * ata_eh_finish - finish up EH
3329 * @ap: host port to finish EH for
3331 * Recovery is complete. Clean up EH states and retry or finish
3337 void ata_eh_finish(struct ata_port *ap)
3341 /* retry or finish qcs */
3342 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3343 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3345 if (!(qc->flags & ATA_QCFLAG_FAILED))
3349 /* FIXME: Once EH migration is complete,
3350 * generate sense data in this function,
3351 * considering both err_mask and tf.
3353 if (qc->flags & ATA_QCFLAG_RETRY)
3354 ata_eh_qc_retry(qc);
3356 ata_eh_qc_complete(qc);
3358 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3359 ata_eh_qc_complete(qc);
3361 /* feed zero TF to sense generation */
3362 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3363 ata_eh_qc_retry(qc);
3368 /* make sure nr_active_links is zero after EH */
3369 WARN_ON(ap->nr_active_links);
3370 ap->nr_active_links = 0;
3374 * ata_do_eh - do standard error handling
3375 * @ap: host port to handle error for
3377 * @prereset: prereset method (can be NULL)
3378 * @softreset: softreset method (can be NULL)
3379 * @hardreset: hardreset method (can be NULL)
3380 * @postreset: postreset method (can be NULL)
3382 * Perform standard error handling sequence.
3385 * Kernel thread context (may sleep).
3387 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3388 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3389 ata_postreset_fn_t postreset)
3391 struct ata_device *dev;
3397 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3400 ata_for_each_dev(dev, &ap->link, ALL)
3401 ata_dev_disable(dev);
3408 * ata_std_error_handler - standard error handler
3409 * @ap: host port to handle error for
3411 * Standard error handler
3414 * Kernel thread context (may sleep).
3416 void ata_std_error_handler(struct ata_port *ap)
3418 struct ata_port_operations *ops = ap->ops;
3419 ata_reset_fn_t hardreset = ops->hardreset;
3421 /* ignore built-in hardreset if SCR access is not available */
3422 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3425 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3430 * ata_eh_handle_port_suspend - perform port suspend operation
3431 * @ap: port to suspend
3436 * Kernel thread context (may sleep).
3438 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3440 unsigned long flags;
3443 /* are we suspending? */
3444 spin_lock_irqsave(ap->lock, flags);
3445 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3446 ap->pm_mesg.event == PM_EVENT_ON) {
3447 spin_unlock_irqrestore(ap->lock, flags);
3450 spin_unlock_irqrestore(ap->lock, flags);
3452 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3454 /* tell ACPI we're suspending */
3455 rc = ata_acpi_on_suspend(ap);
3460 ata_eh_freeze_port(ap);
3462 if (ap->ops->port_suspend)
3463 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3465 ata_acpi_set_state(ap, PMSG_SUSPEND);
3468 spin_lock_irqsave(ap->lock, flags);
3470 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3472 ap->pflags |= ATA_PFLAG_SUSPENDED;
3473 else if (ap->pflags & ATA_PFLAG_FROZEN)
3474 ata_port_schedule_eh(ap);
3476 if (ap->pm_result) {
3477 *ap->pm_result = rc;
3478 ap->pm_result = NULL;
3481 spin_unlock_irqrestore(ap->lock, flags);
3487 * ata_eh_handle_port_resume - perform port resume operation
3488 * @ap: port to resume
3493 * Kernel thread context (may sleep).
3495 static void ata_eh_handle_port_resume(struct ata_port *ap)
3497 unsigned long flags;
3500 /* are we resuming? */
3501 spin_lock_irqsave(ap->lock, flags);
3502 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3503 ap->pm_mesg.event != PM_EVENT_ON) {
3504 spin_unlock_irqrestore(ap->lock, flags);
3507 spin_unlock_irqrestore(ap->lock, flags);
3509 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3511 ata_acpi_set_state(ap, PMSG_ON);
3513 if (ap->ops->port_resume)
3514 rc = ap->ops->port_resume(ap);
3516 /* tell ACPI that we're resuming */
3517 ata_acpi_on_resume(ap);
3520 spin_lock_irqsave(ap->lock, flags);
3521 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3522 if (ap->pm_result) {
3523 *ap->pm_result = rc;
3524 ap->pm_result = NULL;
3526 spin_unlock_irqrestore(ap->lock, flags);
3528 #endif /* CONFIG_PM */