1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * libata-eh.c - libata error handling
5 * Maintained by: Tejun Heo <tj@kernel.org>
6 * Please ALWAYS copy linux-ide@vger.kernel.org
9 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * libata documentation is available via 'make {ps|pdf}docs',
12 * as Documentation/driver-api/libata.rst
14 * Hardware documentation available from http://www.t13.org/ and
15 * http://www.sata-io.org/
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/export.h>
21 #include <linux/pci.h>
22 #include <scsi/scsi.h>
23 #include <scsi/scsi_host.h>
24 #include <scsi/scsi_eh.h>
25 #include <scsi/scsi_device.h>
26 #include <scsi/scsi_cmnd.h>
27 #include <scsi/scsi_dbg.h>
28 #include "../scsi/scsi_transport_api.h"
30 #include <linux/libata.h>
32 #include <trace/events/libata.h>
36 /* speed down verdicts */
37 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
38 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
39 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
40 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
43 ATA_EFLAG_IS_IO = (1 << 0),
44 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
45 ATA_EFLAG_OLD_ER = (1 << 31),
47 /* error categories */
50 ATA_ECAT_TOUT_HSM = 2,
52 ATA_ECAT_DUBIOUS_NONE = 4,
53 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
54 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
55 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
58 ATA_EH_CMD_DFL_TIMEOUT = 5000,
60 /* always put at least this amount of time between resets */
61 ATA_EH_RESET_COOL_DOWN = 5000,
63 /* Waiting in ->prereset can never be reliable. It's
64 * sometimes nice to wait there but it can't be depended upon;
65 * otherwise, we wouldn't be resetting. Just give it enough
66 * time for most drives to spin up.
68 ATA_EH_PRERESET_TIMEOUT = 10000,
69 ATA_EH_FASTDRAIN_INTERVAL = 3000,
73 /* probe speed down parameters, see ata_eh_schedule_probe() */
74 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
75 ATA_EH_PROBE_TRIALS = 2,
78 /* The following table determines how we sequence resets. Each entry
79 * represents timeout for that try. The first try can be soft or
80 * hardreset. All others are hardreset if available. In most cases
81 * the first reset w/ 10sec timeout should succeed. Following entries
82 * are mostly for error handling, hotplug and those outlier devices that
83 * take an exceptionally long time to recover from reset.
85 static const unsigned long ata_eh_reset_timeouts[] = {
86 10000, /* most drives spin up by 10sec */
87 10000, /* > 99% working drives spin up before 20sec */
88 35000, /* give > 30 secs of idleness for outlier devices */
89 5000, /* and sweet one last chance */
90 ULONG_MAX, /* > 1 min has elapsed, give up */
93 static const unsigned long ata_eh_identify_timeouts[] = {
94 5000, /* covers > 99% of successes and not too boring on failures */
95 10000, /* combined time till here is enough even for media access */
96 30000, /* for true idiots */
100 static const unsigned long ata_eh_flush_timeouts[] = {
101 15000, /* be generous with flush */
103 30000, /* and even more generous */
107 static const unsigned long ata_eh_other_timeouts[] = {
108 5000, /* same rationale as identify timeout */
110 /* but no merciful 30sec for other commands, it just isn't worth it */
114 struct ata_eh_cmd_timeout_ent {
116 const unsigned long *timeouts;
119 /* The following table determines timeouts to use for EH internal
120 * commands. Each table entry is a command class and matches the
121 * commands the entry applies to and the timeout table to use.
123 * On the retry after a command timed out, the next timeout value from
124 * the table is used. If the table doesn't contain further entries,
125 * the last value is used.
127 * ehc->cmd_timeout_idx keeps track of which timeout to use per
128 * command class, so if SET_FEATURES times out on the first try, the
129 * next try will use the second timeout value only for that class.
131 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
132 static const struct ata_eh_cmd_timeout_ent
133 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
134 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
135 .timeouts = ata_eh_identify_timeouts, },
136 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
137 .timeouts = ata_eh_other_timeouts, },
138 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
139 .timeouts = ata_eh_other_timeouts, },
140 { .commands = CMDS(ATA_CMD_SET_FEATURES),
141 .timeouts = ata_eh_other_timeouts, },
142 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
145 .timeouts = ata_eh_flush_timeouts },
149 static void __ata_port_freeze(struct ata_port *ap);
151 static void ata_eh_handle_port_suspend(struct ata_port *ap);
152 static void ata_eh_handle_port_resume(struct ata_port *ap);
153 #else /* CONFIG_PM */
154 static void ata_eh_handle_port_suspend(struct ata_port *ap)
157 static void ata_eh_handle_port_resume(struct ata_port *ap)
159 #endif /* CONFIG_PM */
161 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
162 const char *fmt, va_list args)
164 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
165 ATA_EH_DESC_LEN - ehi->desc_len,
170 * __ata_ehi_push_desc - push error description without adding separator
172 * @fmt: printf format string
174 * Format string according to @fmt and append it to @ehi->desc.
177 * spin_lock_irqsave(host lock)
179 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
184 __ata_ehi_pushv_desc(ehi, fmt, args);
189 * ata_ehi_push_desc - push error description with separator
191 * @fmt: printf format string
193 * Format string according to @fmt and append it to @ehi->desc.
194 * If @ehi->desc is not empty, ", " is added in-between.
197 * spin_lock_irqsave(host lock)
199 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
204 __ata_ehi_push_desc(ehi, ", ");
207 __ata_ehi_pushv_desc(ehi, fmt, args);
212 * ata_ehi_clear_desc - clean error description
218 * spin_lock_irqsave(host lock)
220 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
227 * ata_port_desc - append port description
228 * @ap: target ATA port
229 * @fmt: printf format string
231 * Format string according to @fmt and append it to port
232 * description. If port description is not empty, " " is added
233 * in-between. This function is to be used while initializing
234 * ata_host. The description is printed on host registration.
239 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
243 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
245 if (ap->link.eh_info.desc_len)
246 __ata_ehi_push_desc(&ap->link.eh_info, " ");
249 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
256 * ata_port_pbar_desc - append PCI BAR description
257 * @ap: target ATA port
258 * @bar: target PCI BAR
259 * @offset: offset into PCI BAR
260 * @name: name of the area
262 * If @offset is negative, this function formats a string which
263 * contains the name, address, size and type of the BAR and
264 * appends it to the port description. If @offset is zero or
265 * positive, only name and offsetted address is appended.
270 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
273 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
275 unsigned long long start, len;
277 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
279 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
282 start = (unsigned long long)pci_resource_start(pdev, bar);
283 len = (unsigned long long)pci_resource_len(pdev, bar);
286 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
288 ata_port_desc(ap, "%s 0x%llx", name,
289 start + (unsigned long long)offset);
292 #endif /* CONFIG_PCI */
294 static int ata_lookup_timeout_table(u8 cmd)
298 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
301 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
310 * ata_internal_cmd_timeout - determine timeout for an internal command
311 * @dev: target device
312 * @cmd: internal command to be issued
314 * Determine timeout for internal command @cmd for @dev.
320 * Determined timeout.
322 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
324 struct ata_eh_context *ehc = &dev->link->eh_context;
325 int ent = ata_lookup_timeout_table(cmd);
329 return ATA_EH_CMD_DFL_TIMEOUT;
331 idx = ehc->cmd_timeout_idx[dev->devno][ent];
332 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
336 * ata_internal_cmd_timed_out - notification for internal command timeout
337 * @dev: target device
338 * @cmd: internal command which timed out
340 * Notify EH that internal command @cmd for @dev timed out. This
341 * function should be called only for commands whose timeouts are
342 * determined using ata_internal_cmd_timeout().
347 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
349 struct ata_eh_context *ehc = &dev->link->eh_context;
350 int ent = ata_lookup_timeout_table(cmd);
356 idx = ehc->cmd_timeout_idx[dev->devno][ent];
357 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
358 ehc->cmd_timeout_idx[dev->devno][ent]++;
361 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
362 unsigned int err_mask)
364 struct ata_ering_entry *ent;
369 ering->cursor %= ATA_ERING_SIZE;
371 ent = &ering->ring[ering->cursor];
372 ent->eflags = eflags;
373 ent->err_mask = err_mask;
374 ent->timestamp = get_jiffies_64();
377 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
379 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
386 int ata_ering_map(struct ata_ering *ering,
387 int (*map_fn)(struct ata_ering_entry *, void *),
391 struct ata_ering_entry *ent;
395 ent = &ering->ring[idx];
398 rc = map_fn(ent, arg);
401 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
402 } while (idx != ering->cursor);
407 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
409 ent->eflags |= ATA_EFLAG_OLD_ER;
413 static void ata_ering_clear(struct ata_ering *ering)
415 ata_ering_map(ering, ata_ering_clear_cb, NULL);
418 static unsigned int ata_eh_dev_action(struct ata_device *dev)
420 struct ata_eh_context *ehc = &dev->link->eh_context;
422 return ehc->i.action | ehc->i.dev_action[dev->devno];
425 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
426 struct ata_eh_info *ehi, unsigned int action)
428 struct ata_device *tdev;
431 ehi->action &= ~action;
432 ata_for_each_dev(tdev, link, ALL)
433 ehi->dev_action[tdev->devno] &= ~action;
435 /* doesn't make sense for port-wide EH actions */
436 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
438 /* break ehi->action into ehi->dev_action */
439 if (ehi->action & action) {
440 ata_for_each_dev(tdev, link, ALL)
441 ehi->dev_action[tdev->devno] |=
442 ehi->action & action;
443 ehi->action &= ~action;
446 /* turn off the specified per-dev action */
447 ehi->dev_action[dev->devno] &= ~action;
452 * ata_eh_acquire - acquire EH ownership
453 * @ap: ATA port to acquire EH ownership for
455 * Acquire EH ownership for @ap. This is the basic exclusion
456 * mechanism for ports sharing a host. Only one port hanging off
457 * the same host can claim the ownership of EH.
462 void ata_eh_acquire(struct ata_port *ap)
464 mutex_lock(&ap->host->eh_mutex);
465 WARN_ON_ONCE(ap->host->eh_owner);
466 ap->host->eh_owner = current;
470 * ata_eh_release - release EH ownership
471 * @ap: ATA port to release EH ownership for
473 * Release EH ownership for @ap if the caller. The caller must
474 * have acquired EH ownership using ata_eh_acquire() previously.
479 void ata_eh_release(struct ata_port *ap)
481 WARN_ON_ONCE(ap->host->eh_owner != current);
482 ap->host->eh_owner = NULL;
483 mutex_unlock(&ap->host->eh_mutex);
486 static void ata_eh_unload(struct ata_port *ap)
488 struct ata_link *link;
489 struct ata_device *dev;
492 /* Restore SControl IPM and SPD for the next driver and
493 * disable attached devices.
495 ata_for_each_link(link, ap, PMP_FIRST) {
496 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
497 ata_for_each_dev(dev, link, ALL)
498 ata_dev_disable(dev);
501 /* freeze and set UNLOADED */
502 spin_lock_irqsave(ap->lock, flags);
504 ata_port_freeze(ap); /* won't be thawed */
505 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
506 ap->pflags |= ATA_PFLAG_UNLOADED;
508 spin_unlock_irqrestore(ap->lock, flags);
512 * ata_scsi_error - SCSI layer error handler callback
513 * @host: SCSI host on which error occurred
515 * Handles SCSI-layer-thrown error events.
518 * Inherited from SCSI layer (none, can sleep)
523 void ata_scsi_error(struct Scsi_Host *host)
525 struct ata_port *ap = ata_shost_to_port(host);
527 LIST_HEAD(eh_work_q);
531 spin_lock_irqsave(host->host_lock, flags);
532 list_splice_init(&host->eh_cmd_q, &eh_work_q);
533 spin_unlock_irqrestore(host->host_lock, flags);
535 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
537 /* If we timed raced normal completion and there is nothing to
538 recover nr_timedout == 0 why exactly are we doing error recovery ? */
539 ata_scsi_port_error_handler(host, ap);
541 /* finish or retry handled scmd's and clean up */
542 WARN_ON(!list_empty(&eh_work_q));
548 * ata_scsi_cmd_error_handler - error callback for a list of commands
549 * @host: scsi host containing the port
550 * @ap: ATA port within the host
551 * @eh_work_q: list of commands to process
553 * process the given list of commands and return those finished to the
554 * ap->eh_done_q. This function is the first part of the libata error
555 * handler which processes a given list of failed commands.
557 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
558 struct list_head *eh_work_q)
563 /* make sure sff pio task is not running */
564 ata_sff_flush_pio_task(ap);
566 /* synchronize with host lock and sort out timeouts */
568 /* For new EH, all qcs are finished in one of three ways -
569 * normal completion, error completion, and SCSI timeout.
570 * Both completions can race against SCSI timeout. When normal
571 * completion wins, the qc never reaches EH. When error
572 * completion wins, the qc has ATA_QCFLAG_FAILED set.
574 * When SCSI timeout wins, things are a bit more complex.
575 * Normal or error completion can occur after the timeout but
576 * before this point. In such cases, both types of
577 * completions are honored. A scmd is determined to have
578 * timed out iff its associated qc is active and not failed.
580 spin_lock_irqsave(ap->lock, flags);
581 if (ap->ops->error_handler) {
582 struct scsi_cmnd *scmd, *tmp;
585 /* This must occur under the ap->lock as we don't want
586 a polled recovery to race the real interrupt handler
588 The lost_interrupt handler checks for any completed but
589 non-notified command and completes much like an IRQ handler.
591 We then fall into the error recovery code which will treat
592 this as if normal completion won the race */
594 if (ap->ops->lost_interrupt)
595 ap->ops->lost_interrupt(ap);
597 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
598 struct ata_queued_cmd *qc;
600 ata_qc_for_each_raw(ap, qc, i) {
601 if (qc->flags & ATA_QCFLAG_ACTIVE &&
606 if (i < ATA_MAX_QUEUE) {
607 /* the scmd has an associated qc */
608 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
609 /* which hasn't failed yet, timeout */
610 qc->err_mask |= AC_ERR_TIMEOUT;
611 qc->flags |= ATA_QCFLAG_FAILED;
615 /* Normal completion occurred after
616 * SCSI timeout but before this point.
617 * Successfully complete it.
619 scmd->retries = scmd->allowed;
620 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
624 /* If we have timed out qcs. They belong to EH from
625 * this point but the state of the controller is
626 * unknown. Freeze the port to make sure the IRQ
627 * handler doesn't diddle with those qcs. This must
628 * be done atomically w.r.t. setting QCFLAG_FAILED.
631 __ata_port_freeze(ap);
634 /* initialize eh_tries */
635 ap->eh_tries = ATA_EH_MAX_TRIES;
637 spin_unlock_irqrestore(ap->lock, flags);
640 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
643 * ata_scsi_port_error_handler - recover the port after the commands
644 * @host: SCSI host containing the port
647 * Handle the recovery of the port @ap after all the commands
648 * have been recovered.
650 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
654 /* invoke error handler */
655 if (ap->ops->error_handler) {
656 struct ata_link *link;
658 /* acquire EH ownership */
661 /* kill fast drain timer */
662 del_timer_sync(&ap->fastdrain_timer);
664 /* process port resume request */
665 ata_eh_handle_port_resume(ap);
667 /* fetch & clear EH info */
668 spin_lock_irqsave(ap->lock, flags);
670 ata_for_each_link(link, ap, HOST_FIRST) {
671 struct ata_eh_context *ehc = &link->eh_context;
672 struct ata_device *dev;
674 memset(&link->eh_context, 0, sizeof(link->eh_context));
675 link->eh_context.i = link->eh_info;
676 memset(&link->eh_info, 0, sizeof(link->eh_info));
678 ata_for_each_dev(dev, link, ENABLED) {
679 int devno = dev->devno;
681 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
682 if (ata_ncq_enabled(dev))
683 ehc->saved_ncq_enabled |= 1 << devno;
687 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
688 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
689 ap->excl_link = NULL; /* don't maintain exclusion over EH */
691 spin_unlock_irqrestore(ap->lock, flags);
693 /* invoke EH, skip if unloading or suspended */
694 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
695 ap->ops->error_handler(ap);
697 /* if unloading, commence suicide */
698 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
699 !(ap->pflags & ATA_PFLAG_UNLOADED))
704 /* process port suspend request */
705 ata_eh_handle_port_suspend(ap);
707 /* Exception might have happened after ->error_handler
708 * recovered the port but before this point. Repeat
711 spin_lock_irqsave(ap->lock, flags);
713 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
714 if (--ap->eh_tries) {
715 spin_unlock_irqrestore(ap->lock, flags);
719 "EH pending after %d tries, giving up\n",
721 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
724 /* this run is complete, make sure EH info is clear */
725 ata_for_each_link(link, ap, HOST_FIRST)
726 memset(&link->eh_info, 0, sizeof(link->eh_info));
728 /* end eh (clear host_eh_scheduled) while holding
729 * ap->lock such that if exception occurs after this
730 * point but before EH completion, SCSI midlayer will
735 spin_unlock_irqrestore(ap->lock, flags);
738 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
739 ap->ops->eng_timeout(ap);
742 scsi_eh_flush_done_q(&ap->eh_done_q);
745 spin_lock_irqsave(ap->lock, flags);
747 if (ap->pflags & ATA_PFLAG_LOADING)
748 ap->pflags &= ~ATA_PFLAG_LOADING;
749 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
750 !(ap->flags & ATA_FLAG_SAS_HOST))
751 schedule_delayed_work(&ap->hotplug_task, 0);
753 if (ap->pflags & ATA_PFLAG_RECOVERED)
754 ata_port_info(ap, "EH complete\n");
756 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
758 /* tell wait_eh that we're done */
759 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
760 wake_up_all(&ap->eh_wait_q);
762 spin_unlock_irqrestore(ap->lock, flags);
764 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
767 * ata_port_wait_eh - Wait for the currently pending EH to complete
768 * @ap: Port to wait EH for
770 * Wait until the currently pending EH is complete.
773 * Kernel thread context (may sleep).
775 void ata_port_wait_eh(struct ata_port *ap)
781 spin_lock_irqsave(ap->lock, flags);
783 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
784 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
785 spin_unlock_irqrestore(ap->lock, flags);
787 spin_lock_irqsave(ap->lock, flags);
789 finish_wait(&ap->eh_wait_q, &wait);
791 spin_unlock_irqrestore(ap->lock, flags);
793 /* make sure SCSI EH is complete */
794 if (scsi_host_in_recovery(ap->scsi_host)) {
799 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
801 static int ata_eh_nr_in_flight(struct ata_port *ap)
803 struct ata_queued_cmd *qc;
807 /* count only non-internal commands */
808 ata_qc_for_each(ap, qc, tag) {
816 void ata_eh_fastdrain_timerfn(struct timer_list *t)
818 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
822 spin_lock_irqsave(ap->lock, flags);
824 cnt = ata_eh_nr_in_flight(ap);
830 if (cnt == ap->fastdrain_cnt) {
831 struct ata_queued_cmd *qc;
834 /* No progress during the last interval, tag all
835 * in-flight qcs as timed out and freeze the port.
837 ata_qc_for_each(ap, qc, tag) {
839 qc->err_mask |= AC_ERR_TIMEOUT;
844 /* some qcs have finished, give it another chance */
845 ap->fastdrain_cnt = cnt;
846 ap->fastdrain_timer.expires =
847 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
848 add_timer(&ap->fastdrain_timer);
852 spin_unlock_irqrestore(ap->lock, flags);
856 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
857 * @ap: target ATA port
858 * @fastdrain: activate fast drain
860 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
861 * is non-zero and EH wasn't pending before. Fast drain ensures
862 * that EH kicks in in timely manner.
865 * spin_lock_irqsave(host lock)
867 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
871 /* already scheduled? */
872 if (ap->pflags & ATA_PFLAG_EH_PENDING)
875 ap->pflags |= ATA_PFLAG_EH_PENDING;
880 /* do we have in-flight qcs? */
881 cnt = ata_eh_nr_in_flight(ap);
885 /* activate fast drain */
886 ap->fastdrain_cnt = cnt;
887 ap->fastdrain_timer.expires =
888 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
889 add_timer(&ap->fastdrain_timer);
893 * ata_qc_schedule_eh - schedule qc for error handling
894 * @qc: command to schedule error handling for
896 * Schedule error handling for @qc. EH will kick in as soon as
897 * other commands are drained.
900 * spin_lock_irqsave(host lock)
902 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
904 struct ata_port *ap = qc->ap;
906 WARN_ON(!ap->ops->error_handler);
908 qc->flags |= ATA_QCFLAG_FAILED;
909 ata_eh_set_pending(ap, 1);
911 /* The following will fail if timeout has already expired.
912 * ata_scsi_error() takes care of such scmds on EH entry.
913 * Note that ATA_QCFLAG_FAILED is unconditionally set after
914 * this function completes.
916 blk_abort_request(qc->scsicmd->request);
920 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
921 * @ap: ATA port to schedule EH for
923 * LOCKING: inherited from ata_port_schedule_eh
924 * spin_lock_irqsave(host lock)
926 void ata_std_sched_eh(struct ata_port *ap)
928 WARN_ON(!ap->ops->error_handler);
930 if (ap->pflags & ATA_PFLAG_INITIALIZING)
933 ata_eh_set_pending(ap, 1);
934 scsi_schedule_eh(ap->scsi_host);
936 DPRINTK("port EH scheduled\n");
938 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
941 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
942 * @ap: ATA port to end EH for
944 * In the libata object model there is a 1:1 mapping of ata_port to
945 * shost, so host fields can be directly manipulated under ap->lock, in
946 * the libsas case we need to hold a lock at the ha->level to coordinate
950 * spin_lock_irqsave(host lock)
952 void ata_std_end_eh(struct ata_port *ap)
954 struct Scsi_Host *host = ap->scsi_host;
956 host->host_eh_scheduled = 0;
958 EXPORT_SYMBOL(ata_std_end_eh);
962 * ata_port_schedule_eh - schedule error handling without a qc
963 * @ap: ATA port to schedule EH for
965 * Schedule error handling for @ap. EH will kick in as soon as
966 * all commands are drained.
969 * spin_lock_irqsave(host lock)
971 void ata_port_schedule_eh(struct ata_port *ap)
973 /* see: ata_std_sched_eh, unless you know better */
974 ap->ops->sched_eh(ap);
977 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
979 struct ata_queued_cmd *qc;
980 int tag, nr_aborted = 0;
982 WARN_ON(!ap->ops->error_handler);
984 /* we're gonna abort all commands, no need for fast drain */
985 ata_eh_set_pending(ap, 0);
987 /* include internal tag in iteration */
988 ata_qc_for_each_with_internal(ap, qc, tag) {
989 if (qc && (!link || qc->dev->link == link)) {
990 qc->flags |= ATA_QCFLAG_FAILED;
997 ata_port_schedule_eh(ap);
1003 * ata_link_abort - abort all qc's on the link
1004 * @link: ATA link to abort qc's for
1006 * Abort all active qc's active on @link and schedule EH.
1009 * spin_lock_irqsave(host lock)
1012 * Number of aborted qc's.
1014 int ata_link_abort(struct ata_link *link)
1016 return ata_do_link_abort(link->ap, link);
1020 * ata_port_abort - abort all qc's on the port
1021 * @ap: ATA port to abort qc's for
1023 * Abort all active qc's of @ap and schedule EH.
1026 * spin_lock_irqsave(host_set lock)
1029 * Number of aborted qc's.
1031 int ata_port_abort(struct ata_port *ap)
1033 return ata_do_link_abort(ap, NULL);
1037 * __ata_port_freeze - freeze port
1038 * @ap: ATA port to freeze
1040 * This function is called when HSM violation or some other
1041 * condition disrupts normal operation of the port. Frozen port
1042 * is not allowed to perform any operation until the port is
1043 * thawed, which usually follows a successful reset.
1045 * ap->ops->freeze() callback can be used for freezing the port
1046 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1047 * port cannot be frozen hardware-wise, the interrupt handler
1048 * must ack and clear interrupts unconditionally while the port
1052 * spin_lock_irqsave(host lock)
1054 static void __ata_port_freeze(struct ata_port *ap)
1056 WARN_ON(!ap->ops->error_handler);
1058 if (ap->ops->freeze)
1059 ap->ops->freeze(ap);
1061 ap->pflags |= ATA_PFLAG_FROZEN;
1063 DPRINTK("ata%u port frozen\n", ap->print_id);
1067 * ata_port_freeze - abort & freeze port
1068 * @ap: ATA port to freeze
1070 * Abort and freeze @ap. The freeze operation must be called
1071 * first, because some hardware requires special operations
1072 * before the taskfile registers are accessible.
1075 * spin_lock_irqsave(host lock)
1078 * Number of aborted commands.
1080 int ata_port_freeze(struct ata_port *ap)
1084 WARN_ON(!ap->ops->error_handler);
1086 __ata_port_freeze(ap);
1087 nr_aborted = ata_port_abort(ap);
1093 * sata_async_notification - SATA async notification handler
1094 * @ap: ATA port where async notification is received
1096 * Handler to be called when async notification via SDB FIS is
1097 * received. This function schedules EH if necessary.
1100 * spin_lock_irqsave(host lock)
1103 * 1 if EH is scheduled, 0 otherwise.
1105 int sata_async_notification(struct ata_port *ap)
1110 if (!(ap->flags & ATA_FLAG_AN))
1113 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1115 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1117 if (!sata_pmp_attached(ap) || rc) {
1118 /* PMP is not attached or SNTF is not available */
1119 if (!sata_pmp_attached(ap)) {
1120 /* PMP is not attached. Check whether ATAPI
1121 * AN is configured. If so, notify media
1124 struct ata_device *dev = ap->link.device;
1126 if ((dev->class == ATA_DEV_ATAPI) &&
1127 (dev->flags & ATA_DFLAG_AN))
1128 ata_scsi_media_change_notify(dev);
1131 /* PMP is attached but SNTF is not available.
1132 * ATAPI async media change notification is
1133 * not used. The PMP must be reporting PHY
1134 * status change, schedule EH.
1136 ata_port_schedule_eh(ap);
1140 /* PMP is attached and SNTF is available */
1141 struct ata_link *link;
1143 /* check and notify ATAPI AN */
1144 ata_for_each_link(link, ap, EDGE) {
1145 if (!(sntf & (1 << link->pmp)))
1148 if ((link->device->class == ATA_DEV_ATAPI) &&
1149 (link->device->flags & ATA_DFLAG_AN))
1150 ata_scsi_media_change_notify(link->device);
1153 /* If PMP is reporting that PHY status of some
1154 * downstream ports has changed, schedule EH.
1156 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1157 ata_port_schedule_eh(ap);
1166 * ata_eh_freeze_port - EH helper to freeze port
1167 * @ap: ATA port to freeze
1174 void ata_eh_freeze_port(struct ata_port *ap)
1176 unsigned long flags;
1178 if (!ap->ops->error_handler)
1181 spin_lock_irqsave(ap->lock, flags);
1182 __ata_port_freeze(ap);
1183 spin_unlock_irqrestore(ap->lock, flags);
1187 * ata_port_thaw_port - EH helper to thaw port
1188 * @ap: ATA port to thaw
1190 * Thaw frozen port @ap.
1195 void ata_eh_thaw_port(struct ata_port *ap)
1197 unsigned long flags;
1199 if (!ap->ops->error_handler)
1202 spin_lock_irqsave(ap->lock, flags);
1204 ap->pflags &= ~ATA_PFLAG_FROZEN;
1209 spin_unlock_irqrestore(ap->lock, flags);
1211 DPRINTK("ata%u port thawed\n", ap->print_id);
1214 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1219 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1221 struct ata_port *ap = qc->ap;
1222 struct scsi_cmnd *scmd = qc->scsicmd;
1223 unsigned long flags;
1225 spin_lock_irqsave(ap->lock, flags);
1226 qc->scsidone = ata_eh_scsidone;
1227 __ata_qc_complete(qc);
1228 WARN_ON(ata_tag_valid(qc->tag));
1229 spin_unlock_irqrestore(ap->lock, flags);
1231 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1235 * ata_eh_qc_complete - Complete an active ATA command from EH
1236 * @qc: Command to complete
1238 * Indicate to the mid and upper layers that an ATA command has
1239 * completed. To be used from EH.
1241 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1243 struct scsi_cmnd *scmd = qc->scsicmd;
1244 scmd->retries = scmd->allowed;
1245 __ata_eh_qc_complete(qc);
1249 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1250 * @qc: Command to retry
1252 * Indicate to the mid and upper layers that an ATA command
1253 * should be retried. To be used from EH.
1255 * SCSI midlayer limits the number of retries to scmd->allowed.
1256 * scmd->allowed is incremented for commands which get retried
1257 * due to unrelated failures (qc->err_mask is zero).
1259 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1261 struct scsi_cmnd *scmd = qc->scsicmd;
1264 __ata_eh_qc_complete(qc);
1268 * ata_dev_disable - disable ATA device
1269 * @dev: ATA device to disable
1276 void ata_dev_disable(struct ata_device *dev)
1278 if (!ata_dev_enabled(dev))
1281 if (ata_msg_drv(dev->link->ap))
1282 ata_dev_warn(dev, "disabled\n");
1283 ata_acpi_on_disable(dev);
1284 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1287 /* From now till the next successful probe, ering is used to
1288 * track probe failures. Clear accumulated device error info.
1290 ata_ering_clear(&dev->ering);
1294 * ata_eh_detach_dev - detach ATA device
1295 * @dev: ATA device to detach
1302 void ata_eh_detach_dev(struct ata_device *dev)
1304 struct ata_link *link = dev->link;
1305 struct ata_port *ap = link->ap;
1306 struct ata_eh_context *ehc = &link->eh_context;
1307 unsigned long flags;
1309 ata_dev_disable(dev);
1311 spin_lock_irqsave(ap->lock, flags);
1313 dev->flags &= ~ATA_DFLAG_DETACH;
1315 if (ata_scsi_offline_dev(dev)) {
1316 dev->flags |= ATA_DFLAG_DETACHED;
1317 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1320 /* clear per-dev EH info */
1321 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1322 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1323 ehc->saved_xfer_mode[dev->devno] = 0;
1324 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1326 spin_unlock_irqrestore(ap->lock, flags);
1330 * ata_eh_about_to_do - about to perform eh_action
1331 * @link: target ATA link
1332 * @dev: target ATA dev for per-dev action (can be NULL)
1333 * @action: action about to be performed
1335 * Called just before performing EH actions to clear related bits
1336 * in @link->eh_info such that eh actions are not unnecessarily
1342 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1343 unsigned int action)
1345 struct ata_port *ap = link->ap;
1346 struct ata_eh_info *ehi = &link->eh_info;
1347 struct ata_eh_context *ehc = &link->eh_context;
1348 unsigned long flags;
1350 spin_lock_irqsave(ap->lock, flags);
1352 ata_eh_clear_action(link, dev, ehi, action);
1354 /* About to take EH action, set RECOVERED. Ignore actions on
1355 * slave links as master will do them again.
1357 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1358 ap->pflags |= ATA_PFLAG_RECOVERED;
1360 spin_unlock_irqrestore(ap->lock, flags);
1364 * ata_eh_done - EH action complete
1365 * @link: ATA link for which EH actions are complete
1366 * @dev: target ATA dev for per-dev action (can be NULL)
1367 * @action: action just completed
1369 * Called right after performing EH actions to clear related bits
1370 * in @link->eh_context.
1375 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1376 unsigned int action)
1378 struct ata_eh_context *ehc = &link->eh_context;
1380 ata_eh_clear_action(link, dev, &ehc->i, action);
1384 * ata_err_string - convert err_mask to descriptive string
1385 * @err_mask: error mask to convert to string
1387 * Convert @err_mask to descriptive string. Errors are
1388 * prioritized according to severity and only the most severe
1389 * error is reported.
1395 * Descriptive string for @err_mask
1397 static const char *ata_err_string(unsigned int err_mask)
1399 if (err_mask & AC_ERR_HOST_BUS)
1400 return "host bus error";
1401 if (err_mask & AC_ERR_ATA_BUS)
1402 return "ATA bus error";
1403 if (err_mask & AC_ERR_TIMEOUT)
1405 if (err_mask & AC_ERR_HSM)
1406 return "HSM violation";
1407 if (err_mask & AC_ERR_SYSTEM)
1408 return "internal error";
1409 if (err_mask & AC_ERR_MEDIA)
1410 return "media error";
1411 if (err_mask & AC_ERR_INVALID)
1412 return "invalid argument";
1413 if (err_mask & AC_ERR_DEV)
1414 return "device error";
1415 if (err_mask & AC_ERR_NCQ)
1417 if (err_mask & AC_ERR_NODEV_HINT)
1418 return "Polling detection error";
1419 return "unknown error";
1423 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1424 * @dev: Device to read log page 10h from
1425 * @tag: Resulting tag of the failed command
1426 * @tf: Resulting taskfile registers of the failed command
1428 * Read log page 10h to obtain NCQ error details and clear error
1432 * Kernel thread context (may sleep).
1435 * 0 on success, -errno otherwise.
1437 static int ata_eh_read_log_10h(struct ata_device *dev,
1438 int *tag, struct ata_taskfile *tf)
1440 u8 *buf = dev->link->ap->sector_buf;
1441 unsigned int err_mask;
1445 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1450 for (i = 0; i < ATA_SECT_SIZE; i++)
1453 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1459 *tag = buf[0] & 0x1f;
1461 tf->command = buf[2];
1462 tf->feature = buf[3];
1466 tf->device = buf[7];
1467 tf->hob_lbal = buf[8];
1468 tf->hob_lbam = buf[9];
1469 tf->hob_lbah = buf[10];
1470 tf->nsect = buf[12];
1471 tf->hob_nsect = buf[13];
1472 if (ata_id_has_ncq_autosense(dev->id))
1473 tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16];
1479 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1480 * @dev: target ATAPI device
1481 * @r_sense_key: out parameter for sense_key
1483 * Perform ATAPI TEST_UNIT_READY.
1486 * EH context (may sleep).
1489 * 0 on success, AC_ERR_* mask on failure.
1491 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1493 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1494 struct ata_taskfile tf;
1495 unsigned int err_mask;
1497 ata_tf_init(dev, &tf);
1499 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1500 tf.command = ATA_CMD_PACKET;
1501 tf.protocol = ATAPI_PROT_NODATA;
1503 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1504 if (err_mask == AC_ERR_DEV)
1505 *r_sense_key = tf.feature >> 4;
1510 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1511 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1512 * @cmd: scsi command for which the sense code should be set
1514 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1515 * SENSE. This function is an EH helper.
1518 * Kernel thread context (may sleep).
1520 static void ata_eh_request_sense(struct ata_queued_cmd *qc,
1521 struct scsi_cmnd *cmd)
1523 struct ata_device *dev = qc->dev;
1524 struct ata_taskfile tf;
1525 unsigned int err_mask;
1527 if (qc->ap->pflags & ATA_PFLAG_FROZEN) {
1528 ata_dev_warn(dev, "sense data available but port frozen\n");
1532 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1535 if (!ata_id_sense_reporting_enabled(dev->id)) {
1536 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1540 DPRINTK("ATA request sense\n");
1542 ata_tf_init(dev, &tf);
1543 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1544 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1545 tf.command = ATA_CMD_REQ_SENSE_DATA;
1546 tf.protocol = ATA_PROT_NODATA;
1548 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1549 /* Ignore err_mask; ATA_ERR might be set */
1550 if (tf.command & ATA_SENSE) {
1551 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1552 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1554 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1555 tf.command, err_mask);
1560 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1561 * @dev: device to perform REQUEST_SENSE to
1562 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1563 * @dfl_sense_key: default sense key to use
1565 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1566 * SENSE. This function is EH helper.
1569 * Kernel thread context (may sleep).
1572 * 0 on success, AC_ERR_* mask on failure
1574 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1575 u8 *sense_buf, u8 dfl_sense_key)
1577 u8 cdb[ATAPI_CDB_LEN] =
1578 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1579 struct ata_port *ap = dev->link->ap;
1580 struct ata_taskfile tf;
1582 DPRINTK("ATAPI request sense\n");
1584 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1586 /* initialize sense_buf with the error register,
1587 * for the case where they are -not- overwritten
1589 sense_buf[0] = 0x70;
1590 sense_buf[2] = dfl_sense_key;
1592 /* some devices time out if garbage left in tf */
1593 ata_tf_init(dev, &tf);
1595 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1596 tf.command = ATA_CMD_PACKET;
1598 /* is it pointless to prefer PIO for "safety reasons"? */
1599 if (ap->flags & ATA_FLAG_PIO_DMA) {
1600 tf.protocol = ATAPI_PROT_DMA;
1601 tf.feature |= ATAPI_PKT_DMA;
1603 tf.protocol = ATAPI_PROT_PIO;
1604 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1608 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1609 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1613 * ata_eh_analyze_serror - analyze SError for a failed port
1614 * @link: ATA link to analyze SError for
1616 * Analyze SError if available and further determine cause of
1622 static void ata_eh_analyze_serror(struct ata_link *link)
1624 struct ata_eh_context *ehc = &link->eh_context;
1625 u32 serror = ehc->i.serror;
1626 unsigned int err_mask = 0, action = 0;
1629 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1630 err_mask |= AC_ERR_ATA_BUS;
1631 action |= ATA_EH_RESET;
1633 if (serror & SERR_PROTOCOL) {
1634 err_mask |= AC_ERR_HSM;
1635 action |= ATA_EH_RESET;
1637 if (serror & SERR_INTERNAL) {
1638 err_mask |= AC_ERR_SYSTEM;
1639 action |= ATA_EH_RESET;
1642 /* Determine whether a hotplug event has occurred. Both
1643 * SError.N/X are considered hotplug events for enabled or
1644 * host links. For disabled PMP links, only N bit is
1645 * considered as X bit is left at 1 for link plugging.
1647 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1648 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1649 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1650 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1652 hotplug_mask = SERR_PHYRDY_CHG;
1654 if (serror & hotplug_mask)
1655 ata_ehi_hotplugged(&ehc->i);
1657 ehc->i.err_mask |= err_mask;
1658 ehc->i.action |= action;
1662 * ata_eh_analyze_ncq_error - analyze NCQ error
1663 * @link: ATA link to analyze NCQ error for
1665 * Read log page 10h, determine the offending qc and acquire
1666 * error status TF. For NCQ device errors, all LLDDs have to do
1667 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1671 * Kernel thread context (may sleep).
1673 void ata_eh_analyze_ncq_error(struct ata_link *link)
1675 struct ata_port *ap = link->ap;
1676 struct ata_eh_context *ehc = &link->eh_context;
1677 struct ata_device *dev = link->device;
1678 struct ata_queued_cmd *qc;
1679 struct ata_taskfile tf;
1682 /* if frozen, we can't do much */
1683 if (ap->pflags & ATA_PFLAG_FROZEN)
1686 /* is it NCQ device error? */
1687 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1690 /* has LLDD analyzed already? */
1691 ata_qc_for_each_raw(ap, qc, tag) {
1692 if (!(qc->flags & ATA_QCFLAG_FAILED))
1699 /* okay, this error is ours */
1700 memset(&tf, 0, sizeof(tf));
1701 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1703 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1708 if (!(link->sactive & (1 << tag))) {
1709 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1714 /* we've got the perpetrator, condemn it */
1715 qc = __ata_qc_from_tag(ap, tag);
1716 memcpy(&qc->result_tf, &tf, sizeof(tf));
1717 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1718 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1719 if ((qc->result_tf.command & ATA_SENSE) || qc->result_tf.auxiliary) {
1720 char sense_key, asc, ascq;
1722 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1723 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1724 ascq = qc->result_tf.auxiliary & 0xff;
1725 ata_scsi_set_sense(dev, qc->scsicmd, sense_key, asc, ascq);
1726 ata_scsi_set_sense_information(dev, qc->scsicmd,
1728 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1731 ehc->i.err_mask &= ~AC_ERR_DEV;
1735 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1736 * @qc: qc to analyze
1737 * @tf: Taskfile registers to analyze
1739 * Analyze taskfile of @qc and further determine cause of
1740 * failure. This function also requests ATAPI sense data if
1744 * Kernel thread context (may sleep).
1747 * Determined recovery action
1749 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1750 const struct ata_taskfile *tf)
1752 unsigned int tmp, action = 0;
1753 u8 stat = tf->command, err = tf->feature;
1755 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1756 qc->err_mask |= AC_ERR_HSM;
1757 return ATA_EH_RESET;
1760 if (stat & (ATA_ERR | ATA_DF)) {
1761 qc->err_mask |= AC_ERR_DEV;
1763 * Sense data reporting does not work if the
1764 * device fault bit is set.
1772 switch (qc->dev->class) {
1775 if (stat & ATA_SENSE)
1776 ata_eh_request_sense(qc, qc->scsicmd);
1778 qc->err_mask |= AC_ERR_ATA_BUS;
1779 if (err & (ATA_UNC | ATA_AMNF))
1780 qc->err_mask |= AC_ERR_MEDIA;
1782 qc->err_mask |= AC_ERR_INVALID;
1786 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1787 tmp = atapi_eh_request_sense(qc->dev,
1788 qc->scsicmd->sense_buffer,
1789 qc->result_tf.feature >> 4);
1791 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1793 qc->err_mask |= tmp;
1797 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1798 int ret = scsi_check_sense(qc->scsicmd);
1800 * SUCCESS here means that the sense code could be
1801 * evaluated and should be passed to the upper layers
1802 * for correct evaluation.
1803 * FAILED means the sense code could not be interpreted
1804 * and the device would need to be reset.
1805 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1806 * command would need to be retried.
1808 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1809 qc->flags |= ATA_QCFLAG_RETRY;
1810 qc->err_mask |= AC_ERR_OTHER;
1811 } else if (ret != SUCCESS) {
1812 qc->err_mask |= AC_ERR_HSM;
1815 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1816 action |= ATA_EH_RESET;
1821 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1826 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1830 base = ATA_ECAT_DUBIOUS_NONE;
1832 if (err_mask & AC_ERR_ATA_BUS)
1833 return base + ATA_ECAT_ATA_BUS;
1835 if (err_mask & AC_ERR_TIMEOUT)
1836 return base + ATA_ECAT_TOUT_HSM;
1838 if (eflags & ATA_EFLAG_IS_IO) {
1839 if (err_mask & AC_ERR_HSM)
1840 return base + ATA_ECAT_TOUT_HSM;
1842 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1843 return base + ATA_ECAT_UNK_DEV;
1849 struct speed_down_verdict_arg {
1852 int nr_errors[ATA_ECAT_NR];
1855 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1857 struct speed_down_verdict_arg *arg = void_arg;
1860 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1863 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1865 arg->nr_errors[cat]++;
1871 * ata_eh_speed_down_verdict - Determine speed down verdict
1872 * @dev: Device of interest
1874 * This function examines error ring of @dev and determines
1875 * whether NCQ needs to be turned off, transfer speed should be
1876 * stepped down, or falling back to PIO is necessary.
1878 * ECAT_ATA_BUS : ATA_BUS error for any command
1880 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1883 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1885 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1886 * data transfer hasn't been verified.
1890 * NCQ_OFF : Turn off NCQ.
1892 * SPEED_DOWN : Speed down transfer speed but don't fall back
1895 * FALLBACK_TO_PIO : Fall back to PIO.
1897 * Even if multiple verdicts are returned, only one action is
1898 * taken per error. An action triggered by non-DUBIOUS errors
1899 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1900 * This is to expedite speed down decisions right after device is
1901 * initially configured.
1903 * The following are speed down rules. #1 and #2 deal with
1906 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1907 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1909 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1910 * occurred during last 5 mins, NCQ_OFF.
1912 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1913 * occurred during last 5 mins, FALLBACK_TO_PIO
1915 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1916 * during last 10 mins, NCQ_OFF.
1918 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1919 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1922 * Inherited from caller.
1925 * OR of ATA_EH_SPDN_* flags.
1927 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1929 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1930 u64 j64 = get_jiffies_64();
1931 struct speed_down_verdict_arg arg;
1932 unsigned int verdict = 0;
1934 /* scan past 5 mins of error history */
1935 memset(&arg, 0, sizeof(arg));
1936 arg.since = j64 - min(j64, j5mins);
1937 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1939 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1940 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1941 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1942 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1944 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1945 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1946 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1948 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1949 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1950 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1951 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1953 /* scan past 10 mins of error history */
1954 memset(&arg, 0, sizeof(arg));
1955 arg.since = j64 - min(j64, j10mins);
1956 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1958 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1959 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1960 verdict |= ATA_EH_SPDN_NCQ_OFF;
1962 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1963 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1964 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1965 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1971 * ata_eh_speed_down - record error and speed down if necessary
1972 * @dev: Failed device
1973 * @eflags: mask of ATA_EFLAG_* flags
1974 * @err_mask: err_mask of the error
1976 * Record error and examine error history to determine whether
1977 * adjusting transmission speed is necessary. It also sets
1978 * transmission limits appropriately if such adjustment is
1982 * Kernel thread context (may sleep).
1985 * Determined recovery action.
1987 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1988 unsigned int eflags, unsigned int err_mask)
1990 struct ata_link *link = ata_dev_phys_link(dev);
1992 unsigned int verdict;
1993 unsigned int action = 0;
1995 /* don't bother if Cat-0 error */
1996 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1999 /* record error and determine whether speed down is necessary */
2000 ata_ering_record(&dev->ering, eflags, err_mask);
2001 verdict = ata_eh_speed_down_verdict(dev);
2004 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2005 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2006 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2007 dev->flags |= ATA_DFLAG_NCQ_OFF;
2008 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2013 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2014 /* speed down SATA link speed if possible */
2015 if (sata_down_spd_limit(link, 0) == 0) {
2016 action |= ATA_EH_RESET;
2020 /* lower transfer mode */
2021 if (dev->spdn_cnt < 2) {
2022 static const int dma_dnxfer_sel[] =
2023 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2024 static const int pio_dnxfer_sel[] =
2025 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2028 if (dev->xfer_shift != ATA_SHIFT_PIO)
2029 sel = dma_dnxfer_sel[dev->spdn_cnt];
2031 sel = pio_dnxfer_sel[dev->spdn_cnt];
2035 if (ata_down_xfermask_limit(dev, sel) == 0) {
2036 action |= ATA_EH_RESET;
2042 /* Fall back to PIO? Slowing down to PIO is meaningless for
2043 * SATA ATA devices. Consider it only for PATA and SATAPI.
2045 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2046 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2047 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2048 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2050 action |= ATA_EH_RESET;
2057 /* device has been slowed down, blow error history */
2058 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2059 ata_ering_clear(&dev->ering);
2064 * ata_eh_worth_retry - analyze error and decide whether to retry
2065 * @qc: qc to possibly retry
2067 * Look at the cause of the error and decide if a retry
2068 * might be useful or not. We don't want to retry media errors
2069 * because the drive itself has probably already taken 10-30 seconds
2070 * doing its own internal retries before reporting the failure.
2072 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2074 if (qc->err_mask & AC_ERR_MEDIA)
2075 return 0; /* don't retry media errors */
2076 if (qc->flags & ATA_QCFLAG_IO)
2077 return 1; /* otherwise retry anything from fs stack */
2078 if (qc->err_mask & AC_ERR_INVALID)
2079 return 0; /* don't retry these */
2080 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2084 * ata_eh_quiet - check if we need to be quiet about a command error
2087 * Look at the qc flags anbd its scsi command request flags to determine
2088 * if we need to be quiet about the command failure.
2090 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
2093 qc->scsicmd->request->rq_flags & RQF_QUIET)
2094 qc->flags |= ATA_QCFLAG_QUIET;
2095 return qc->flags & ATA_QCFLAG_QUIET;
2099 * ata_eh_link_autopsy - analyze error and determine recovery action
2100 * @link: host link to perform autopsy on
2102 * Analyze why @link failed and determine which recovery actions
2103 * are needed. This function also sets more detailed AC_ERR_*
2104 * values and fills sense data for ATAPI CHECK SENSE.
2107 * Kernel thread context (may sleep).
2109 static void ata_eh_link_autopsy(struct ata_link *link)
2111 struct ata_port *ap = link->ap;
2112 struct ata_eh_context *ehc = &link->eh_context;
2113 struct ata_queued_cmd *qc;
2114 struct ata_device *dev;
2115 unsigned int all_err_mask = 0, eflags = 0;
2116 int tag, nr_failed = 0, nr_quiet = 0;
2122 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2125 /* obtain and analyze SError */
2126 rc = sata_scr_read(link, SCR_ERROR, &serror);
2128 ehc->i.serror |= serror;
2129 ata_eh_analyze_serror(link);
2130 } else if (rc != -EOPNOTSUPP) {
2131 /* SError read failed, force reset and probing */
2132 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2133 ehc->i.action |= ATA_EH_RESET;
2134 ehc->i.err_mask |= AC_ERR_OTHER;
2137 /* analyze NCQ failure */
2138 ata_eh_analyze_ncq_error(link);
2140 /* any real error trumps AC_ERR_OTHER */
2141 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2142 ehc->i.err_mask &= ~AC_ERR_OTHER;
2144 all_err_mask |= ehc->i.err_mask;
2146 ata_qc_for_each_raw(ap, qc, tag) {
2147 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2148 ata_dev_phys_link(qc->dev) != link)
2151 /* inherit upper level err_mask */
2152 qc->err_mask |= ehc->i.err_mask;
2155 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2157 /* DEV errors are probably spurious in case of ATA_BUS error */
2158 if (qc->err_mask & AC_ERR_ATA_BUS)
2159 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2162 /* any real error trumps unknown error */
2163 if (qc->err_mask & ~AC_ERR_OTHER)
2164 qc->err_mask &= ~AC_ERR_OTHER;
2167 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2168 * layers will determine whether the command is worth retrying
2169 * based on the sense data and device class/type. Otherwise,
2170 * determine directly if the command is worth retrying using its
2171 * error mask and flags.
2173 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2174 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2175 else if (ata_eh_worth_retry(qc))
2176 qc->flags |= ATA_QCFLAG_RETRY;
2178 /* accumulate error info */
2179 ehc->i.dev = qc->dev;
2180 all_err_mask |= qc->err_mask;
2181 if (qc->flags & ATA_QCFLAG_IO)
2182 eflags |= ATA_EFLAG_IS_IO;
2183 trace_ata_eh_link_autopsy_qc(qc);
2185 /* Count quiet errors */
2186 if (ata_eh_quiet(qc))
2191 /* If all failed commands requested silence, then be quiet */
2192 if (nr_quiet == nr_failed)
2193 ehc->i.flags |= ATA_EHI_QUIET;
2195 /* enforce default EH actions */
2196 if (ap->pflags & ATA_PFLAG_FROZEN ||
2197 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2198 ehc->i.action |= ATA_EH_RESET;
2199 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2200 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2201 ehc->i.action |= ATA_EH_REVALIDATE;
2203 /* If we have offending qcs and the associated failed device,
2204 * perform per-dev EH action only on the offending device.
2207 ehc->i.dev_action[ehc->i.dev->devno] |=
2208 ehc->i.action & ATA_EH_PERDEV_MASK;
2209 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2212 /* propagate timeout to host link */
2213 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2214 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2216 /* record error and consider speeding down */
2218 if (!dev && ((ata_link_max_devices(link) == 1 &&
2219 ata_dev_enabled(link->device))))
2223 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2224 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2225 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2226 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2232 * ata_eh_autopsy - analyze error and determine recovery action
2233 * @ap: host port to perform autopsy on
2235 * Analyze all links of @ap and determine why they failed and
2236 * which recovery actions are needed.
2239 * Kernel thread context (may sleep).
2241 void ata_eh_autopsy(struct ata_port *ap)
2243 struct ata_link *link;
2245 ata_for_each_link(link, ap, EDGE)
2246 ata_eh_link_autopsy(link);
2248 /* Handle the frigging slave link. Autopsy is done similarly
2249 * but actions and flags are transferred over to the master
2250 * link and handled from there.
2252 if (ap->slave_link) {
2253 struct ata_eh_context *mehc = &ap->link.eh_context;
2254 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2256 /* transfer control flags from master to slave */
2257 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2259 /* perform autopsy on the slave link */
2260 ata_eh_link_autopsy(ap->slave_link);
2262 /* transfer actions from slave to master and clear slave */
2263 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2264 mehc->i.action |= sehc->i.action;
2265 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2266 mehc->i.flags |= sehc->i.flags;
2267 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2270 /* Autopsy of fanout ports can affect host link autopsy.
2271 * Perform host link autopsy last.
2273 if (sata_pmp_attached(ap))
2274 ata_eh_link_autopsy(&ap->link);
2278 * ata_get_cmd_descript - get description for ATA command
2279 * @command: ATA command code to get description for
2281 * Return a textual description of the given command, or NULL if the
2282 * command is not known.
2287 const char *ata_get_cmd_descript(u8 command)
2289 #ifdef CONFIG_ATA_VERBOSE_ERROR
2295 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2296 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2297 { ATA_CMD_STANDBY, "STANDBY" },
2298 { ATA_CMD_IDLE, "IDLE" },
2299 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2300 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2301 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2302 { ATA_CMD_NOP, "NOP" },
2303 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2304 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2305 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2306 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2307 { ATA_CMD_SERVICE, "SERVICE" },
2308 { ATA_CMD_READ, "READ DMA" },
2309 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2310 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2311 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2312 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2313 { ATA_CMD_WRITE, "WRITE DMA" },
2314 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2315 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2316 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2317 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2318 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2319 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2320 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2321 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2322 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2323 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2324 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2325 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2326 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2327 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2328 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2329 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2330 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2331 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2332 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2333 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2334 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2335 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2336 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2337 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2338 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2339 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2340 { ATA_CMD_SLEEP, "SLEEP" },
2341 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2342 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2343 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2344 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2345 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2346 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2347 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2348 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2349 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2350 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2351 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2352 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2353 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2354 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2355 { ATA_CMD_PMP_READ, "READ BUFFER" },
2356 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2357 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2358 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2359 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2360 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2361 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2362 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2363 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2364 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2365 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2366 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2367 { ATA_CMD_SMART, "SMART" },
2368 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2369 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2370 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2371 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2372 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2373 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2374 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2375 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2376 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2377 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2378 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2379 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2380 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2381 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2382 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2383 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2384 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2385 { ATA_CMD_RESTORE, "RECALIBRATE" },
2386 { 0, NULL } /* terminate list */
2390 for (i = 0; cmd_descr[i].text; i++)
2391 if (cmd_descr[i].command == command)
2392 return cmd_descr[i].text;
2397 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2400 * ata_eh_link_report - report error handling to user
2401 * @link: ATA link EH is going on
2403 * Report EH to user.
2408 static void ata_eh_link_report(struct ata_link *link)
2410 struct ata_port *ap = link->ap;
2411 struct ata_eh_context *ehc = &link->eh_context;
2412 struct ata_queued_cmd *qc;
2413 const char *frozen, *desc;
2414 char tries_buf[6] = "";
2415 int tag, nr_failed = 0;
2417 if (ehc->i.flags & ATA_EHI_QUIET)
2421 if (ehc->i.desc[0] != '\0')
2424 ata_qc_for_each_raw(ap, qc, tag) {
2425 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2426 ata_dev_phys_link(qc->dev) != link ||
2427 ((qc->flags & ATA_QCFLAG_QUIET) &&
2428 qc->err_mask == AC_ERR_DEV))
2430 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2436 if (!nr_failed && !ehc->i.err_mask)
2440 if (ap->pflags & ATA_PFLAG_FROZEN)
2443 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2444 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2448 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2449 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2450 ehc->i.err_mask, link->sactive, ehc->i.serror,
2451 ehc->i.action, frozen, tries_buf);
2453 ata_dev_err(ehc->i.dev, "%s\n", desc);
2455 ata_link_err(link, "exception Emask 0x%x "
2456 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2457 ehc->i.err_mask, link->sactive, ehc->i.serror,
2458 ehc->i.action, frozen, tries_buf);
2460 ata_link_err(link, "%s\n", desc);
2463 #ifdef CONFIG_ATA_VERBOSE_ERROR
2466 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2467 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2468 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2469 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2470 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2471 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2472 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2473 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2474 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2475 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2476 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2477 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2478 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2479 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2480 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2481 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2482 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2483 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2486 ata_qc_for_each_raw(ap, qc, tag) {
2487 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2488 char data_buf[20] = "";
2489 char cdb_buf[70] = "";
2491 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2492 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2495 if (qc->dma_dir != DMA_NONE) {
2496 static const char *dma_str[] = {
2497 [DMA_BIDIRECTIONAL] = "bidi",
2498 [DMA_TO_DEVICE] = "out",
2499 [DMA_FROM_DEVICE] = "in",
2501 const char *prot_str = NULL;
2503 switch (qc->tf.protocol) {
2504 case ATA_PROT_UNKNOWN:
2505 prot_str = "unknown";
2507 case ATA_PROT_NODATA:
2508 prot_str = "nodata";
2517 prot_str = "ncq dma";
2519 case ATA_PROT_NCQ_NODATA:
2520 prot_str = "ncq nodata";
2522 case ATAPI_PROT_NODATA:
2523 prot_str = "nodata";
2525 case ATAPI_PROT_PIO:
2528 case ATAPI_PROT_DMA:
2532 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2533 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2536 if (ata_is_atapi(qc->tf.protocol)) {
2537 const u8 *cdb = qc->cdb;
2538 size_t cdb_len = qc->dev->cdb_len;
2541 cdb = qc->scsicmd->cmnd;
2542 cdb_len = qc->scsicmd->cmd_len;
2544 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2547 const char *descr = ata_get_cmd_descript(cmd->command);
2549 ata_dev_err(qc->dev, "failed command: %s\n",
2553 ata_dev_err(qc->dev,
2554 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2556 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2557 "Emask 0x%x (%s)%s\n",
2558 cmd->command, cmd->feature, cmd->nsect,
2559 cmd->lbal, cmd->lbam, cmd->lbah,
2560 cmd->hob_feature, cmd->hob_nsect,
2561 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2562 cmd->device, qc->tag, data_buf, cdb_buf,
2563 res->command, res->feature, res->nsect,
2564 res->lbal, res->lbam, res->lbah,
2565 res->hob_feature, res->hob_nsect,
2566 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2567 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2568 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2570 #ifdef CONFIG_ATA_VERBOSE_ERROR
2571 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2572 ATA_SENSE | ATA_ERR)) {
2573 if (res->command & ATA_BUSY)
2574 ata_dev_err(qc->dev, "status: { Busy }\n");
2576 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2577 res->command & ATA_DRDY ? "DRDY " : "",
2578 res->command & ATA_DF ? "DF " : "",
2579 res->command & ATA_DRQ ? "DRQ " : "",
2580 res->command & ATA_SENSE ? "SENSE " : "",
2581 res->command & ATA_ERR ? "ERR " : "");
2584 if (cmd->command != ATA_CMD_PACKET &&
2585 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2586 ATA_IDNF | ATA_ABORTED)))
2587 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2588 res->feature & ATA_ICRC ? "ICRC " : "",
2589 res->feature & ATA_UNC ? "UNC " : "",
2590 res->feature & ATA_AMNF ? "AMNF " : "",
2591 res->feature & ATA_IDNF ? "IDNF " : "",
2592 res->feature & ATA_ABORTED ? "ABRT " : "");
2598 * ata_eh_report - report error handling to user
2599 * @ap: ATA port to report EH about
2601 * Report EH to user.
2606 void ata_eh_report(struct ata_port *ap)
2608 struct ata_link *link;
2610 ata_for_each_link(link, ap, HOST_FIRST)
2611 ata_eh_link_report(link);
2614 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2615 unsigned int *classes, unsigned long deadline,
2618 struct ata_device *dev;
2621 ata_for_each_dev(dev, link, ALL)
2622 classes[dev->devno] = ATA_DEV_UNKNOWN;
2624 return reset(link, classes, deadline);
2627 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2629 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2633 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2638 int ata_eh_reset(struct ata_link *link, int classify,
2639 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2640 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2642 struct ata_port *ap = link->ap;
2643 struct ata_link *slave = ap->slave_link;
2644 struct ata_eh_context *ehc = &link->eh_context;
2645 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2646 unsigned int *classes = ehc->classes;
2647 unsigned int lflags = link->flags;
2648 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2649 int max_tries = 0, try = 0;
2650 struct ata_link *failed_link;
2651 struct ata_device *dev;
2652 unsigned long deadline, now;
2653 ata_reset_fn_t reset;
2654 unsigned long flags;
2661 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2663 if (link->flags & ATA_LFLAG_RST_ONCE)
2665 if (link->flags & ATA_LFLAG_NO_HRST)
2667 if (link->flags & ATA_LFLAG_NO_SRST)
2670 /* make sure each reset attempt is at least COOL_DOWN apart */
2671 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2673 WARN_ON(time_after(ehc->last_reset, now));
2674 deadline = ata_deadline(ehc->last_reset,
2675 ATA_EH_RESET_COOL_DOWN);
2676 if (time_before(now, deadline))
2677 schedule_timeout_uninterruptible(deadline - now);
2680 spin_lock_irqsave(ap->lock, flags);
2681 ap->pflags |= ATA_PFLAG_RESETTING;
2682 spin_unlock_irqrestore(ap->lock, flags);
2684 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2686 ata_for_each_dev(dev, link, ALL) {
2687 /* If we issue an SRST then an ATA drive (not ATAPI)
2688 * may change configuration and be in PIO0 timing. If
2689 * we do a hard reset (or are coming from power on)
2690 * this is true for ATA or ATAPI. Until we've set a
2691 * suitable controller mode we should not touch the
2692 * bus as we may be talking too fast.
2694 dev->pio_mode = XFER_PIO_0;
2695 dev->dma_mode = 0xff;
2697 /* If the controller has a pio mode setup function
2698 * then use it to set the chipset to rights. Don't
2699 * touch the DMA setup as that will be dealt with when
2700 * configuring devices.
2702 if (ap->ops->set_piomode)
2703 ap->ops->set_piomode(ap, dev);
2706 /* prefer hardreset */
2708 ehc->i.action &= ~ATA_EH_RESET;
2711 ehc->i.action |= ATA_EH_HARDRESET;
2712 } else if (softreset) {
2714 ehc->i.action |= ATA_EH_SOFTRESET;
2718 unsigned long deadline = ata_deadline(jiffies,
2719 ATA_EH_PRERESET_TIMEOUT);
2722 sehc->i.action &= ~ATA_EH_RESET;
2723 sehc->i.action |= ehc->i.action;
2726 rc = prereset(link, deadline);
2728 /* If present, do prereset on slave link too. Reset
2729 * is skipped iff both master and slave links report
2730 * -ENOENT or clear ATA_EH_RESET.
2732 if (slave && (rc == 0 || rc == -ENOENT)) {
2735 tmp = prereset(slave, deadline);
2739 ehc->i.action |= sehc->i.action;
2743 if (rc == -ENOENT) {
2744 ata_link_dbg(link, "port disabled--ignoring\n");
2745 ehc->i.action &= ~ATA_EH_RESET;
2747 ata_for_each_dev(dev, link, ALL)
2748 classes[dev->devno] = ATA_DEV_NONE;
2753 "prereset failed (errno=%d)\n",
2758 /* prereset() might have cleared ATA_EH_RESET. If so,
2759 * bang classes, thaw and return.
2761 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2762 ata_for_each_dev(dev, link, ALL)
2763 classes[dev->devno] = ATA_DEV_NONE;
2764 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2765 ata_is_host_link(link))
2766 ata_eh_thaw_port(ap);
2776 if (ata_is_host_link(link))
2777 ata_eh_freeze_port(ap);
2779 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2783 ata_link_info(link, "%s resetting link\n",
2784 reset == softreset ? "soft" : "hard");
2786 /* mark that this EH session started with reset */
2787 ehc->last_reset = jiffies;
2788 if (reset == hardreset)
2789 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2791 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2793 rc = ata_do_reset(link, reset, classes, deadline, true);
2794 if (rc && rc != -EAGAIN) {
2799 /* hardreset slave link if existent */
2800 if (slave && reset == hardreset) {
2804 ata_link_info(slave, "hard resetting link\n");
2806 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2807 tmp = ata_do_reset(slave, reset, classes, deadline,
2815 failed_link = slave;
2821 /* perform follow-up SRST if necessary */
2822 if (reset == hardreset &&
2823 ata_eh_followup_srst_needed(link, rc)) {
2828 "follow-up softreset required but no softreset available\n");
2834 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2835 rc = ata_do_reset(link, reset, classes, deadline, true);
2844 "no reset method available, skipping reset\n");
2845 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2846 lflags |= ATA_LFLAG_ASSUME_ATA;
2850 * Post-reset processing
2852 ata_for_each_dev(dev, link, ALL) {
2853 /* After the reset, the device state is PIO 0 and the
2854 * controller state is undefined. Reset also wakes up
2855 * drives from sleeping mode.
2857 dev->pio_mode = XFER_PIO_0;
2858 dev->flags &= ~ATA_DFLAG_SLEEPING;
2860 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2863 /* apply class override */
2864 if (lflags & ATA_LFLAG_ASSUME_ATA)
2865 classes[dev->devno] = ATA_DEV_ATA;
2866 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2867 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2870 /* record current link speed */
2871 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2872 link->sata_spd = (sstatus >> 4) & 0xf;
2873 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2874 slave->sata_spd = (sstatus >> 4) & 0xf;
2877 if (ata_is_host_link(link))
2878 ata_eh_thaw_port(ap);
2880 /* postreset() should clear hardware SError. Although SError
2881 * is cleared during link resume, clearing SError here is
2882 * necessary as some PHYs raise hotplug events after SRST.
2883 * This introduces race condition where hotplug occurs between
2884 * reset and here. This race is mediated by cross checking
2885 * link onlineness and classification result later.
2888 postreset(link, classes);
2890 postreset(slave, classes);
2894 * Some controllers can't be frozen very well and may set spurious
2895 * error conditions during reset. Clear accumulated error
2896 * information and re-thaw the port if frozen. As reset is the
2897 * final recovery action and we cross check link onlineness against
2898 * device classification later, no hotplug event is lost by this.
2900 spin_lock_irqsave(link->ap->lock, flags);
2901 memset(&link->eh_info, 0, sizeof(link->eh_info));
2903 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2904 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2905 spin_unlock_irqrestore(link->ap->lock, flags);
2907 if (ap->pflags & ATA_PFLAG_FROZEN)
2908 ata_eh_thaw_port(ap);
2911 * Make sure onlineness and classification result correspond.
2912 * Hotplug could have happened during reset and some
2913 * controllers fail to wait while a drive is spinning up after
2914 * being hotplugged causing misdetection. By cross checking
2915 * link on/offlineness and classification result, those
2916 * conditions can be reliably detected and retried.
2919 ata_for_each_dev(dev, link, ALL) {
2920 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2921 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2922 ata_dev_dbg(dev, "link online but device misclassified\n");
2923 classes[dev->devno] = ATA_DEV_NONE;
2926 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2927 if (ata_class_enabled(classes[dev->devno]))
2929 "link offline, clearing class %d to NONE\n",
2930 classes[dev->devno]);
2931 classes[dev->devno] = ATA_DEV_NONE;
2932 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2934 "link status unknown, clearing UNKNOWN to NONE\n");
2935 classes[dev->devno] = ATA_DEV_NONE;
2939 if (classify && nr_unknown) {
2940 if (try < max_tries) {
2942 "link online but %d devices misclassified, retrying\n",
2949 "link online but %d devices misclassified, "
2950 "device detection might fail\n", nr_unknown);
2953 /* reset successful, schedule revalidation */
2954 ata_eh_done(link, NULL, ATA_EH_RESET);
2956 ata_eh_done(slave, NULL, ATA_EH_RESET);
2957 ehc->last_reset = jiffies; /* update to completion time */
2958 ehc->i.action |= ATA_EH_REVALIDATE;
2959 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2963 /* clear hotplug flag */
2964 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2966 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2968 spin_lock_irqsave(ap->lock, flags);
2969 ap->pflags &= ~ATA_PFLAG_RESETTING;
2970 spin_unlock_irqrestore(ap->lock, flags);
2975 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2976 if (!ata_is_host_link(link) &&
2977 sata_scr_read(link, SCR_STATUS, &sstatus))
2980 if (try >= max_tries) {
2982 * Thaw host port even if reset failed, so that the port
2983 * can be retried on the next phy event. This risks
2984 * repeated EH runs but seems to be a better tradeoff than
2985 * shutting down a port after a botched hotplug attempt.
2987 if (ata_is_host_link(link))
2988 ata_eh_thaw_port(ap);
2993 if (time_before(now, deadline)) {
2994 unsigned long delta = deadline - now;
2996 ata_link_warn(failed_link,
2997 "reset failed (errno=%d), retrying in %u secs\n",
2998 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3002 delta = schedule_timeout_uninterruptible(delta);
3007 * While disks spinup behind PMP, some controllers fail sending SRST.
3008 * They need to be reset - as well as the PMP - before retrying.
3010 if (rc == -ERESTART) {
3011 if (ata_is_host_link(link))
3012 ata_eh_thaw_port(ap);
3016 if (try == max_tries - 1) {
3017 sata_down_spd_limit(link, 0);
3019 sata_down_spd_limit(slave, 0);
3020 } else if (rc == -EPIPE)
3021 sata_down_spd_limit(failed_link, 0);
3028 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3030 struct ata_link *link;
3031 struct ata_device *dev;
3032 unsigned long flags;
3035 * This function can be thought of as an extended version of
3036 * ata_eh_about_to_do() specially crafted to accommodate the
3037 * requirements of ATA_EH_PARK handling. Since the EH thread
3038 * does not leave the do {} while () loop in ata_eh_recover as
3039 * long as the timeout for a park request to *one* device on
3040 * the port has not expired, and since we still want to pick
3041 * up park requests to other devices on the same port or
3042 * timeout updates for the same device, we have to pull
3043 * ATA_EH_PARK actions from eh_info into eh_context.i
3044 * ourselves at the beginning of each pass over the loop.
3046 * Additionally, all write accesses to &ap->park_req_pending
3047 * through reinit_completion() (see below) or complete_all()
3048 * (see ata_scsi_park_store()) are protected by the host lock.
3049 * As a result we have that park_req_pending.done is zero on
3050 * exit from this function, i.e. when ATA_EH_PARK actions for
3051 * *all* devices on port ap have been pulled into the
3052 * respective eh_context structs. If, and only if,
3053 * park_req_pending.done is non-zero by the time we reach
3054 * wait_for_completion_timeout(), another ATA_EH_PARK action
3055 * has been scheduled for at least one of the devices on port
3056 * ap and we have to cycle over the do {} while () loop in
3057 * ata_eh_recover() again.
3060 spin_lock_irqsave(ap->lock, flags);
3061 reinit_completion(&ap->park_req_pending);
3062 ata_for_each_link(link, ap, EDGE) {
3063 ata_for_each_dev(dev, link, ALL) {
3064 struct ata_eh_info *ehi = &link->eh_info;
3066 link->eh_context.i.dev_action[dev->devno] |=
3067 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3068 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3071 spin_unlock_irqrestore(ap->lock, flags);
3074 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3076 struct ata_eh_context *ehc = &dev->link->eh_context;
3077 struct ata_taskfile tf;
3078 unsigned int err_mask;
3080 ata_tf_init(dev, &tf);
3082 ehc->unloaded_mask |= 1 << dev->devno;
3083 tf.command = ATA_CMD_IDLEIMMEDIATE;
3089 ehc->unloaded_mask &= ~(1 << dev->devno);
3090 tf.command = ATA_CMD_CHK_POWER;
3093 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3094 tf.protocol = ATA_PROT_NODATA;
3095 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3096 if (park && (err_mask || tf.lbal != 0xc4)) {
3097 ata_dev_err(dev, "head unload failed!\n");
3098 ehc->unloaded_mask &= ~(1 << dev->devno);
3102 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3103 struct ata_device **r_failed_dev)
3105 struct ata_port *ap = link->ap;
3106 struct ata_eh_context *ehc = &link->eh_context;
3107 struct ata_device *dev;
3108 unsigned int new_mask = 0;
3109 unsigned long flags;
3114 /* For PATA drive side cable detection to work, IDENTIFY must
3115 * be done backwards such that PDIAG- is released by the slave
3116 * device before the master device is identified.
3118 ata_for_each_dev(dev, link, ALL_REVERSE) {
3119 unsigned int action = ata_eh_dev_action(dev);
3120 unsigned int readid_flags = 0;
3122 if (ehc->i.flags & ATA_EHI_DID_RESET)
3123 readid_flags |= ATA_READID_POSTRESET;
3125 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3126 WARN_ON(dev->class == ATA_DEV_PMP);
3128 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3133 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3134 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3139 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3141 /* Configuration may have changed, reconfigure
3144 ehc->i.flags |= ATA_EHI_SETMODE;
3146 /* schedule the scsi_rescan_device() here */
3147 schedule_work(&(ap->scsi_rescan_task));
3148 } else if (dev->class == ATA_DEV_UNKNOWN &&
3149 ehc->tries[dev->devno] &&
3150 ata_class_enabled(ehc->classes[dev->devno])) {
3151 /* Temporarily set dev->class, it will be
3152 * permanently set once all configurations are
3153 * complete. This is necessary because new
3154 * device configuration is done in two
3157 dev->class = ehc->classes[dev->devno];
3159 if (dev->class == ATA_DEV_PMP)
3160 rc = sata_pmp_attach(dev);
3162 rc = ata_dev_read_id(dev, &dev->class,
3163 readid_flags, dev->id);
3165 /* read_id might have changed class, store and reset */
3166 ehc->classes[dev->devno] = dev->class;
3167 dev->class = ATA_DEV_UNKNOWN;
3171 /* clear error info accumulated during probe */
3172 ata_ering_clear(&dev->ering);
3173 new_mask |= 1 << dev->devno;
3176 /* IDENTIFY was issued to non-existent
3177 * device. No need to reset. Just
3178 * thaw and ignore the device.
3180 ata_eh_thaw_port(ap);
3188 /* PDIAG- should have been released, ask cable type if post-reset */
3189 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3190 if (ap->ops->cable_detect)
3191 ap->cbl = ap->ops->cable_detect(ap);
3195 /* Configure new devices forward such that user doesn't see
3196 * device detection messages backwards.
3198 ata_for_each_dev(dev, link, ALL) {
3199 if (!(new_mask & (1 << dev->devno)))
3202 dev->class = ehc->classes[dev->devno];
3204 if (dev->class == ATA_DEV_PMP)
3207 ehc->i.flags |= ATA_EHI_PRINTINFO;
3208 rc = ata_dev_configure(dev);
3209 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3211 dev->class = ATA_DEV_UNKNOWN;
3215 spin_lock_irqsave(ap->lock, flags);
3216 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3217 spin_unlock_irqrestore(ap->lock, flags);
3219 /* new device discovered, configure xfermode */
3220 ehc->i.flags |= ATA_EHI_SETMODE;
3226 *r_failed_dev = dev;
3227 DPRINTK("EXIT rc=%d\n", rc);
3232 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3233 * @link: link on which timings will be programmed
3234 * @r_failed_dev: out parameter for failed device
3236 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3237 * ata_set_mode() fails, pointer to the failing device is
3238 * returned in @r_failed_dev.
3241 * PCI/etc. bus probe sem.
3244 * 0 on success, negative errno otherwise
3246 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3248 struct ata_port *ap = link->ap;
3249 struct ata_device *dev;
3252 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3253 ata_for_each_dev(dev, link, ENABLED) {
3254 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3255 struct ata_ering_entry *ent;
3257 ent = ata_ering_top(&dev->ering);
3259 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3263 /* has private set_mode? */
3264 if (ap->ops->set_mode)
3265 rc = ap->ops->set_mode(link, r_failed_dev);
3267 rc = ata_do_set_mode(link, r_failed_dev);
3269 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3270 ata_for_each_dev(dev, link, ENABLED) {
3271 struct ata_eh_context *ehc = &link->eh_context;
3272 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3273 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3275 if (dev->xfer_mode != saved_xfer_mode ||
3276 ata_ncq_enabled(dev) != saved_ncq)
3277 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3284 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3285 * @dev: ATAPI device to clear UA for
3287 * Resets and other operations can make an ATAPI device raise
3288 * UNIT ATTENTION which causes the next operation to fail. This
3289 * function clears UA.
3292 * EH context (may sleep).
3295 * 0 on success, -errno on failure.
3297 static int atapi_eh_clear_ua(struct ata_device *dev)
3301 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3302 u8 *sense_buffer = dev->link->ap->sector_buf;
3304 unsigned int err_mask;
3306 err_mask = atapi_eh_tur(dev, &sense_key);
3307 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3309 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3314 if (!err_mask || sense_key != UNIT_ATTENTION)
3317 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3319 ata_dev_warn(dev, "failed to clear "
3320 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3325 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3332 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3333 * @dev: ATA device which may need FLUSH retry
3335 * If @dev failed FLUSH, it needs to be reported upper layer
3336 * immediately as it means that @dev failed to remap and already
3337 * lost at least a sector and further FLUSH retrials won't make
3338 * any difference to the lost sector. However, if FLUSH failed
3339 * for other reasons, for example transmission error, FLUSH needs
3342 * This function determines whether FLUSH failure retry is
3343 * necessary and performs it if so.
3346 * 0 if EH can continue, -errno if EH needs to be repeated.
3348 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3350 struct ata_link *link = dev->link;
3351 struct ata_port *ap = link->ap;
3352 struct ata_queued_cmd *qc;
3353 struct ata_taskfile tf;
3354 unsigned int err_mask;
3357 /* did flush fail for this device? */
3358 if (!ata_tag_valid(link->active_tag))
3361 qc = __ata_qc_from_tag(ap, link->active_tag);
3362 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3363 qc->tf.command != ATA_CMD_FLUSH))
3366 /* if the device failed it, it should be reported to upper layers */
3367 if (qc->err_mask & AC_ERR_DEV)
3370 /* flush failed for some other reason, give it another shot */
3371 ata_tf_init(dev, &tf);
3373 tf.command = qc->tf.command;
3374 tf.flags |= ATA_TFLAG_DEVICE;
3375 tf.protocol = ATA_PROT_NODATA;
3377 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3378 tf.command, qc->err_mask);
3380 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3383 * FLUSH is complete but there's no way to
3384 * successfully complete a failed command from EH.
3385 * Making sure retry is allowed at least once and
3386 * retrying it should do the trick - whatever was in
3387 * the cache is already on the platter and this won't
3388 * cause infinite loop.
3390 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3392 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3396 /* if device failed it, report it to upper layers */
3397 if (err_mask & AC_ERR_DEV) {
3398 qc->err_mask |= AC_ERR_DEV;
3400 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3408 * ata_eh_set_lpm - configure SATA interface power management
3409 * @link: link to configure power management
3410 * @policy: the link power management policy
3411 * @r_failed_dev: out parameter for failed device
3413 * Enable SATA Interface power management. This will enable
3414 * Device Interface Power Management (DIPM) for min_power and
3415 * medium_power_with_dipm policies, and then call driver specific
3416 * callbacks for enabling Host Initiated Power management.
3422 * 0 on success, -errno on failure.
3424 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3425 struct ata_device **r_failed_dev)
3427 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3428 struct ata_eh_context *ehc = &link->eh_context;
3429 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3430 enum ata_lpm_policy old_policy = link->lpm_policy;
3431 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3432 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3433 unsigned int err_mask;
3436 /* if the link or host doesn't do LPM, noop */
3437 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3441 * DIPM is enabled only for MIN_POWER as some devices
3442 * misbehave when the host NACKs transition to SLUMBER. Order
3443 * device and link configurations such that the host always
3444 * allows DIPM requests.
3446 ata_for_each_dev(dev, link, ENABLED) {
3447 bool hipm = ata_id_has_hipm(dev->id);
3448 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3450 /* find the first enabled and LPM enabled devices */
3454 if (!lpm_dev && (hipm || dipm))
3457 hints &= ~ATA_LPM_EMPTY;
3459 hints &= ~ATA_LPM_HIPM;
3461 /* disable DIPM before changing link config */
3462 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3463 err_mask = ata_dev_set_feature(dev,
3464 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3465 if (err_mask && err_mask != AC_ERR_DEV) {
3467 "failed to disable DIPM, Emask 0x%x\n",
3476 rc = ap->ops->set_lpm(link, policy, hints);
3477 if (!rc && ap->slave_link)
3478 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3480 rc = sata_pmp_set_lpm(link, policy, hints);
3483 * Attribute link config failure to the first (LPM) enabled
3484 * device on the link.
3487 if (rc == -EOPNOTSUPP) {
3488 link->flags |= ATA_LFLAG_NO_LPM;
3491 dev = lpm_dev ? lpm_dev : link_dev;
3496 * Low level driver acked the transition. Issue DIPM command
3497 * with the new policy set.
3499 link->lpm_policy = policy;
3500 if (ap && ap->slave_link)
3501 ap->slave_link->lpm_policy = policy;
3503 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3504 ata_for_each_dev(dev, link, ENABLED) {
3505 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3506 ata_id_has_dipm(dev->id)) {
3507 err_mask = ata_dev_set_feature(dev,
3508 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3509 if (err_mask && err_mask != AC_ERR_DEV) {
3511 "failed to enable DIPM, Emask 0x%x\n",
3519 link->last_lpm_change = jiffies;
3520 link->flags |= ATA_LFLAG_CHANGED;
3525 /* restore the old policy */
3526 link->lpm_policy = old_policy;
3527 if (ap && ap->slave_link)
3528 ap->slave_link->lpm_policy = old_policy;
3530 /* if no device or only one more chance is left, disable LPM */
3531 if (!dev || ehc->tries[dev->devno] <= 2) {
3532 ata_link_warn(link, "disabling LPM on the link\n");
3533 link->flags |= ATA_LFLAG_NO_LPM;
3536 *r_failed_dev = dev;
3540 int ata_link_nr_enabled(struct ata_link *link)
3542 struct ata_device *dev;
3545 ata_for_each_dev(dev, link, ENABLED)
3550 static int ata_link_nr_vacant(struct ata_link *link)
3552 struct ata_device *dev;
3555 ata_for_each_dev(dev, link, ALL)
3556 if (dev->class == ATA_DEV_UNKNOWN)
3561 static int ata_eh_skip_recovery(struct ata_link *link)
3563 struct ata_port *ap = link->ap;
3564 struct ata_eh_context *ehc = &link->eh_context;
3565 struct ata_device *dev;
3567 /* skip disabled links */
3568 if (link->flags & ATA_LFLAG_DISABLED)
3571 /* skip if explicitly requested */
3572 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3575 /* thaw frozen port and recover failed devices */
3576 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3579 /* reset at least once if reset is requested */
3580 if ((ehc->i.action & ATA_EH_RESET) &&
3581 !(ehc->i.flags & ATA_EHI_DID_RESET))
3584 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3585 ata_for_each_dev(dev, link, ALL) {
3586 if (dev->class == ATA_DEV_UNKNOWN &&
3587 ehc->classes[dev->devno] != ATA_DEV_NONE)
3594 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3596 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3597 u64 now = get_jiffies_64();
3598 int *trials = void_arg;
3600 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3601 (ent->timestamp < now - min(now, interval)))
3608 static int ata_eh_schedule_probe(struct ata_device *dev)
3610 struct ata_eh_context *ehc = &dev->link->eh_context;
3611 struct ata_link *link = ata_dev_phys_link(dev);
3614 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3615 (ehc->did_probe_mask & (1 << dev->devno)))
3618 ata_eh_detach_dev(dev);
3620 ehc->did_probe_mask |= (1 << dev->devno);
3621 ehc->i.action |= ATA_EH_RESET;
3622 ehc->saved_xfer_mode[dev->devno] = 0;
3623 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3625 /* the link maybe in a deep sleep, wake it up */
3626 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3627 if (ata_is_host_link(link))
3628 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3631 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3635 /* Record and count probe trials on the ering. The specific
3636 * error mask used is irrelevant. Because a successful device
3637 * detection clears the ering, this count accumulates only if
3638 * there are consecutive failed probes.
3640 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3641 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3642 * forced to 1.5Gbps.
3644 * This is to work around cases where failed link speed
3645 * negotiation results in device misdetection leading to
3646 * infinite DEVXCHG or PHRDY CHG events.
3648 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3649 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3651 if (trials > ATA_EH_PROBE_TRIALS)
3652 sata_down_spd_limit(link, 1);
3657 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3659 struct ata_eh_context *ehc = &dev->link->eh_context;
3661 /* -EAGAIN from EH routine indicates retry without prejudice.
3662 * The requester is responsible for ensuring forward progress.
3665 ehc->tries[dev->devno]--;
3669 /* device missing or wrong IDENTIFY data, schedule probing */
3670 ehc->i.probe_mask |= (1 << dev->devno);
3673 /* give it just one more chance */
3674 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3677 if (ehc->tries[dev->devno] == 1) {
3678 /* This is the last chance, better to slow
3679 * down than lose it.
3681 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3682 if (dev->pio_mode > XFER_PIO_0)
3683 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3687 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3688 /* disable device if it has used up all its chances */
3689 ata_dev_disable(dev);
3691 /* detach if offline */
3692 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3693 ata_eh_detach_dev(dev);
3695 /* schedule probe if necessary */
3696 if (ata_eh_schedule_probe(dev)) {
3697 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3698 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3699 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3704 ehc->i.action |= ATA_EH_RESET;
3710 * ata_eh_recover - recover host port after error
3711 * @ap: host port to recover
3712 * @prereset: prereset method (can be NULL)
3713 * @softreset: softreset method (can be NULL)
3714 * @hardreset: hardreset method (can be NULL)
3715 * @postreset: postreset method (can be NULL)
3716 * @r_failed_link: out parameter for failed link
3718 * This is the alpha and omega, eum and yang, heart and soul of
3719 * libata exception handling. On entry, actions required to
3720 * recover each link and hotplug requests are recorded in the
3721 * link's eh_context. This function executes all the operations
3722 * with appropriate retrials and fallbacks to resurrect failed
3723 * devices, detach goners and greet newcomers.
3726 * Kernel thread context (may sleep).
3729 * 0 on success, -errno on failure.
3731 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3732 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3733 ata_postreset_fn_t postreset,
3734 struct ata_link **r_failed_link)
3736 struct ata_link *link;
3737 struct ata_device *dev;
3739 unsigned long flags, deadline;
3743 /* prep for recovery */
3744 ata_for_each_link(link, ap, EDGE) {
3745 struct ata_eh_context *ehc = &link->eh_context;
3747 /* re-enable link? */
3748 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3749 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3750 spin_lock_irqsave(ap->lock, flags);
3751 link->flags &= ~ATA_LFLAG_DISABLED;
3752 spin_unlock_irqrestore(ap->lock, flags);
3753 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3756 ata_for_each_dev(dev, link, ALL) {
3757 if (link->flags & ATA_LFLAG_NO_RETRY)
3758 ehc->tries[dev->devno] = 1;
3760 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3762 /* collect port action mask recorded in dev actions */
3763 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3764 ~ATA_EH_PERDEV_MASK;
3765 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3767 /* process hotplug request */
3768 if (dev->flags & ATA_DFLAG_DETACH)
3769 ata_eh_detach_dev(dev);
3771 /* schedule probe if necessary */
3772 if (!ata_dev_enabled(dev))
3773 ata_eh_schedule_probe(dev);
3780 /* if UNLOADING, finish immediately */
3781 if (ap->pflags & ATA_PFLAG_UNLOADING)
3785 ata_for_each_link(link, ap, EDGE) {
3786 struct ata_eh_context *ehc = &link->eh_context;
3788 /* skip EH if possible. */
3789 if (ata_eh_skip_recovery(link))
3792 ata_for_each_dev(dev, link, ALL)
3793 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3797 ata_for_each_link(link, ap, EDGE) {
3798 struct ata_eh_context *ehc = &link->eh_context;
3800 if (!(ehc->i.action & ATA_EH_RESET))
3803 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3804 prereset, softreset, hardreset, postreset);
3806 ata_link_err(link, "reset failed, giving up\n");
3815 * clears ATA_EH_PARK in eh_info and resets
3816 * ap->park_req_pending
3818 ata_eh_pull_park_action(ap);
3821 ata_for_each_link(link, ap, EDGE) {
3822 ata_for_each_dev(dev, link, ALL) {
3823 struct ata_eh_context *ehc = &link->eh_context;
3826 if (dev->class != ATA_DEV_ATA &&
3827 dev->class != ATA_DEV_ZAC)
3829 if (!(ehc->i.dev_action[dev->devno] &
3832 tmp = dev->unpark_deadline;
3833 if (time_before(deadline, tmp))
3835 else if (time_before_eq(tmp, jiffies))
3837 if (ehc->unloaded_mask & (1 << dev->devno))
3840 ata_eh_park_issue_cmd(dev, 1);
3845 if (time_before_eq(deadline, now))
3849 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3853 ata_for_each_link(link, ap, EDGE) {
3854 ata_for_each_dev(dev, link, ALL) {
3855 if (!(link->eh_context.unloaded_mask &
3859 ata_eh_park_issue_cmd(dev, 0);
3860 ata_eh_done(link, dev, ATA_EH_PARK);
3866 ata_for_each_link(link, ap, PMP_FIRST) {
3867 struct ata_eh_context *ehc = &link->eh_context;
3869 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3872 /* revalidate existing devices and attach new ones */
3873 rc = ata_eh_revalidate_and_attach(link, &dev);
3877 /* if PMP got attached, return, pmp EH will take care of it */
3878 if (link->device->class == ATA_DEV_PMP) {
3883 /* configure transfer mode if necessary */
3884 if (ehc->i.flags & ATA_EHI_SETMODE) {
3885 rc = ata_set_mode(link, &dev);
3888 ehc->i.flags &= ~ATA_EHI_SETMODE;
3891 /* If reset has been issued, clear UA to avoid
3892 * disrupting the current users of the device.
3894 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3895 ata_for_each_dev(dev, link, ALL) {
3896 if (dev->class != ATA_DEV_ATAPI)
3898 rc = atapi_eh_clear_ua(dev);
3901 if (zpodd_dev_enabled(dev))
3902 zpodd_post_poweron(dev);
3906 /* retry flush if necessary */
3907 ata_for_each_dev(dev, link, ALL) {
3908 if (dev->class != ATA_DEV_ATA &&
3909 dev->class != ATA_DEV_ZAC)
3911 rc = ata_eh_maybe_retry_flush(dev);
3917 /* configure link power saving */
3918 if (link->lpm_policy != ap->target_lpm_policy) {
3919 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3924 /* this link is okay now */
3931 ata_eh_handle_dev_fail(dev, rc);
3933 if (ap->pflags & ATA_PFLAG_FROZEN) {
3934 /* PMP reset requires working host port.
3935 * Can't retry if it's frozen.
3937 if (sata_pmp_attached(ap))
3947 if (rc && r_failed_link)
3948 *r_failed_link = link;
3950 DPRINTK("EXIT, rc=%d\n", rc);
3955 * ata_eh_finish - finish up EH
3956 * @ap: host port to finish EH for
3958 * Recovery is complete. Clean up EH states and retry or finish
3964 void ata_eh_finish(struct ata_port *ap)
3966 struct ata_queued_cmd *qc;
3969 /* retry or finish qcs */
3970 ata_qc_for_each_raw(ap, qc, tag) {
3971 if (!(qc->flags & ATA_QCFLAG_FAILED))
3975 /* FIXME: Once EH migration is complete,
3976 * generate sense data in this function,
3977 * considering both err_mask and tf.
3979 if (qc->flags & ATA_QCFLAG_RETRY)
3980 ata_eh_qc_retry(qc);
3982 ata_eh_qc_complete(qc);
3984 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3985 ata_eh_qc_complete(qc);
3987 /* feed zero TF to sense generation */
3988 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3989 ata_eh_qc_retry(qc);
3994 /* make sure nr_active_links is zero after EH */
3995 WARN_ON(ap->nr_active_links);
3996 ap->nr_active_links = 0;
4000 * ata_do_eh - do standard error handling
4001 * @ap: host port to handle error for
4003 * @prereset: prereset method (can be NULL)
4004 * @softreset: softreset method (can be NULL)
4005 * @hardreset: hardreset method (can be NULL)
4006 * @postreset: postreset method (can be NULL)
4008 * Perform standard error handling sequence.
4011 * Kernel thread context (may sleep).
4013 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4014 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4015 ata_postreset_fn_t postreset)
4017 struct ata_device *dev;
4023 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4026 ata_for_each_dev(dev, &ap->link, ALL)
4027 ata_dev_disable(dev);
4034 * ata_std_error_handler - standard error handler
4035 * @ap: host port to handle error for
4037 * Standard error handler
4040 * Kernel thread context (may sleep).
4042 void ata_std_error_handler(struct ata_port *ap)
4044 struct ata_port_operations *ops = ap->ops;
4045 ata_reset_fn_t hardreset = ops->hardreset;
4047 /* ignore built-in hardreset if SCR access is not available */
4048 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4051 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4056 * ata_eh_handle_port_suspend - perform port suspend operation
4057 * @ap: port to suspend
4062 * Kernel thread context (may sleep).
4064 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4066 unsigned long flags;
4068 struct ata_device *dev;
4070 /* are we suspending? */
4071 spin_lock_irqsave(ap->lock, flags);
4072 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4073 ap->pm_mesg.event & PM_EVENT_RESUME) {
4074 spin_unlock_irqrestore(ap->lock, flags);
4077 spin_unlock_irqrestore(ap->lock, flags);
4079 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4082 * If we have a ZPODD attached, check its zero
4083 * power ready status before the port is frozen.
4084 * Only needed for runtime suspend.
4086 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4087 ata_for_each_dev(dev, &ap->link, ENABLED) {
4088 if (zpodd_dev_enabled(dev))
4089 zpodd_on_suspend(dev);
4093 /* tell ACPI we're suspending */
4094 rc = ata_acpi_on_suspend(ap);
4099 ata_eh_freeze_port(ap);
4101 if (ap->ops->port_suspend)
4102 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4104 ata_acpi_set_state(ap, ap->pm_mesg);
4106 /* update the flags */
4107 spin_lock_irqsave(ap->lock, flags);
4109 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4111 ap->pflags |= ATA_PFLAG_SUSPENDED;
4112 else if (ap->pflags & ATA_PFLAG_FROZEN)
4113 ata_port_schedule_eh(ap);
4115 spin_unlock_irqrestore(ap->lock, flags);
4121 * ata_eh_handle_port_resume - perform port resume operation
4122 * @ap: port to resume
4127 * Kernel thread context (may sleep).
4129 static void ata_eh_handle_port_resume(struct ata_port *ap)
4131 struct ata_link *link;
4132 struct ata_device *dev;
4133 unsigned long flags;
4135 /* are we resuming? */
4136 spin_lock_irqsave(ap->lock, flags);
4137 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4138 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4139 spin_unlock_irqrestore(ap->lock, flags);
4142 spin_unlock_irqrestore(ap->lock, flags);
4144 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4147 * Error timestamps are in jiffies which doesn't run while
4148 * suspended and PHY events during resume isn't too uncommon.
4149 * When the two are combined, it can lead to unnecessary speed
4150 * downs if the machine is suspended and resumed repeatedly.
4151 * Clear error history.
4153 ata_for_each_link(link, ap, HOST_FIRST)
4154 ata_for_each_dev(dev, link, ALL)
4155 ata_ering_clear(&dev->ering);
4157 ata_acpi_set_state(ap, ap->pm_mesg);
4159 if (ap->ops->port_resume)
4160 ap->ops->port_resume(ap);
4162 /* tell ACPI that we're resuming */
4163 ata_acpi_on_resume(ap);
4165 /* update the flags */
4166 spin_lock_irqsave(ap->lock, flags);
4167 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4168 spin_unlock_irqrestore(ap->lock, flags);
4170 #endif /* CONFIG_PM */