1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * libata-eh.c - libata error handling
7 * libata documentation is available via 'make {ps|pdf}docs',
8 * as Documentation/driver-api/libata.rst
10 * Hardware documentation available from http://www.t13.org/ and
11 * http://www.sata-io.org/
14 #include <linux/kernel.h>
15 #include <linux/blkdev.h>
16 #include <linux/export.h>
17 #include <linux/pci.h>
18 #include <scsi/scsi.h>
19 #include <scsi/scsi_host.h>
20 #include <scsi/scsi_eh.h>
21 #include <scsi/scsi_device.h>
22 #include <scsi/scsi_cmnd.h>
23 #include <scsi/scsi_dbg.h>
24 #include "../scsi/scsi_transport_api.h"
26 #include <linux/libata.h>
28 #include <trace/events/libata.h>
32 /* speed down verdicts */
33 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
34 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
35 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
36 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
39 ATA_EFLAG_IS_IO = (1 << 0),
40 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
41 ATA_EFLAG_OLD_ER = (1 << 31),
43 /* error categories */
46 ATA_ECAT_TOUT_HSM = 2,
48 ATA_ECAT_DUBIOUS_NONE = 4,
49 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
50 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
51 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
54 ATA_EH_CMD_DFL_TIMEOUT = 5000,
56 /* always put at least this amount of time between resets */
57 ATA_EH_RESET_COOL_DOWN = 5000,
59 /* Waiting in ->prereset can never be reliable. It's
60 * sometimes nice to wait there but it can't be depended upon;
61 * otherwise, we wouldn't be resetting. Just give it enough
62 * time for most drives to spin up.
64 ATA_EH_PRERESET_TIMEOUT = 10000,
65 ATA_EH_FASTDRAIN_INTERVAL = 3000,
69 /* probe speed down parameters, see ata_eh_schedule_probe() */
70 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
71 ATA_EH_PROBE_TRIALS = 2,
74 /* The following table determines how we sequence resets. Each entry
75 * represents timeout for that try. The first try can be soft or
76 * hardreset. All others are hardreset if available. In most cases
77 * the first reset w/ 10sec timeout should succeed. Following entries
78 * are mostly for error handling, hotplug and those outlier devices that
79 * take an exceptionally long time to recover from reset.
81 static const unsigned long ata_eh_reset_timeouts[] = {
82 10000, /* most drives spin up by 10sec */
83 10000, /* > 99% working drives spin up before 20sec */
84 35000, /* give > 30 secs of idleness for outlier devices */
85 5000, /* and sweet one last chance */
86 ULONG_MAX, /* > 1 min has elapsed, give up */
89 static const unsigned int ata_eh_identify_timeouts[] = {
90 5000, /* covers > 99% of successes and not too boring on failures */
91 10000, /* combined time till here is enough even for media access */
92 30000, /* for true idiots */
96 static const unsigned int ata_eh_revalidate_timeouts[] = {
97 15000, /* Some drives are slow to read log pages when waking-up */
98 15000, /* combined time till here is enough even for media access */
102 static const unsigned int ata_eh_flush_timeouts[] = {
103 15000, /* be generous with flush */
105 30000, /* and even more generous */
109 static const unsigned int ata_eh_other_timeouts[] = {
110 5000, /* same rationale as identify timeout */
112 /* but no merciful 30sec for other commands, it just isn't worth it */
116 struct ata_eh_cmd_timeout_ent {
118 const unsigned int *timeouts;
121 /* The following table determines timeouts to use for EH internal
122 * commands. Each table entry is a command class and matches the
123 * commands the entry applies to and the timeout table to use.
125 * On the retry after a command timed out, the next timeout value from
126 * the table is used. If the table doesn't contain further entries,
127 * the last value is used.
129 * ehc->cmd_timeout_idx keeps track of which timeout to use per
130 * command class, so if SET_FEATURES times out on the first try, the
131 * next try will use the second timeout value only for that class.
133 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
134 static const struct ata_eh_cmd_timeout_ent
135 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
136 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
137 .timeouts = ata_eh_identify_timeouts, },
138 { .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT),
139 .timeouts = ata_eh_revalidate_timeouts, },
140 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
141 .timeouts = ata_eh_other_timeouts, },
142 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_SET_FEATURES),
145 .timeouts = ata_eh_other_timeouts, },
146 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
149 .timeouts = ata_eh_flush_timeouts },
153 static void __ata_port_freeze(struct ata_port *ap);
154 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
155 struct ata_device **r_failed_dev);
157 static void ata_eh_handle_port_suspend(struct ata_port *ap);
158 static void ata_eh_handle_port_resume(struct ata_port *ap);
159 #else /* CONFIG_PM */
160 static void ata_eh_handle_port_suspend(struct ata_port *ap)
163 static void ata_eh_handle_port_resume(struct ata_port *ap)
165 #endif /* CONFIG_PM */
167 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
168 const char *fmt, va_list args)
170 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
171 ATA_EH_DESC_LEN - ehi->desc_len,
176 * __ata_ehi_push_desc - push error description without adding separator
178 * @fmt: printf format string
180 * Format string according to @fmt and append it to @ehi->desc.
183 * spin_lock_irqsave(host lock)
185 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
190 __ata_ehi_pushv_desc(ehi, fmt, args);
193 EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
196 * ata_ehi_push_desc - push error description with separator
198 * @fmt: printf format string
200 * Format string according to @fmt and append it to @ehi->desc.
201 * If @ehi->desc is not empty, ", " is added in-between.
204 * spin_lock_irqsave(host lock)
206 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
211 __ata_ehi_push_desc(ehi, ", ");
214 __ata_ehi_pushv_desc(ehi, fmt, args);
217 EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
220 * ata_ehi_clear_desc - clean error description
226 * spin_lock_irqsave(host lock)
228 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
233 EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
236 * ata_port_desc - append port description
237 * @ap: target ATA port
238 * @fmt: printf format string
240 * Format string according to @fmt and append it to port
241 * description. If port description is not empty, " " is added
242 * in-between. This function is to be used while initializing
243 * ata_host. The description is printed on host registration.
248 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
252 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
254 if (ap->link.eh_info.desc_len)
255 __ata_ehi_push_desc(&ap->link.eh_info, " ");
258 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
261 EXPORT_SYMBOL_GPL(ata_port_desc);
265 * ata_port_pbar_desc - append PCI BAR description
266 * @ap: target ATA port
267 * @bar: target PCI BAR
268 * @offset: offset into PCI BAR
269 * @name: name of the area
271 * If @offset is negative, this function formats a string which
272 * contains the name, address, size and type of the BAR and
273 * appends it to the port description. If @offset is zero or
274 * positive, only name and offsetted address is appended.
279 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
282 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
284 unsigned long long start, len;
286 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
288 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
291 start = (unsigned long long)pci_resource_start(pdev, bar);
292 len = (unsigned long long)pci_resource_len(pdev, bar);
295 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
297 ata_port_desc(ap, "%s 0x%llx", name,
298 start + (unsigned long long)offset);
300 EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
301 #endif /* CONFIG_PCI */
303 static int ata_lookup_timeout_table(u8 cmd)
307 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
310 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
319 * ata_internal_cmd_timeout - determine timeout for an internal command
320 * @dev: target device
321 * @cmd: internal command to be issued
323 * Determine timeout for internal command @cmd for @dev.
329 * Determined timeout.
331 unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
333 struct ata_eh_context *ehc = &dev->link->eh_context;
334 int ent = ata_lookup_timeout_table(cmd);
338 return ATA_EH_CMD_DFL_TIMEOUT;
340 idx = ehc->cmd_timeout_idx[dev->devno][ent];
341 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
345 * ata_internal_cmd_timed_out - notification for internal command timeout
346 * @dev: target device
347 * @cmd: internal command which timed out
349 * Notify EH that internal command @cmd for @dev timed out. This
350 * function should be called only for commands whose timeouts are
351 * determined using ata_internal_cmd_timeout().
356 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
358 struct ata_eh_context *ehc = &dev->link->eh_context;
359 int ent = ata_lookup_timeout_table(cmd);
365 idx = ehc->cmd_timeout_idx[dev->devno][ent];
366 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
367 ehc->cmd_timeout_idx[dev->devno][ent]++;
370 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
371 unsigned int err_mask)
373 struct ata_ering_entry *ent;
378 ering->cursor %= ATA_ERING_SIZE;
380 ent = &ering->ring[ering->cursor];
381 ent->eflags = eflags;
382 ent->err_mask = err_mask;
383 ent->timestamp = get_jiffies_64();
386 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
388 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
395 int ata_ering_map(struct ata_ering *ering,
396 int (*map_fn)(struct ata_ering_entry *, void *),
400 struct ata_ering_entry *ent;
404 ent = &ering->ring[idx];
407 rc = map_fn(ent, arg);
410 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
411 } while (idx != ering->cursor);
416 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
418 ent->eflags |= ATA_EFLAG_OLD_ER;
422 static void ata_ering_clear(struct ata_ering *ering)
424 ata_ering_map(ering, ata_ering_clear_cb, NULL);
427 static unsigned int ata_eh_dev_action(struct ata_device *dev)
429 struct ata_eh_context *ehc = &dev->link->eh_context;
431 return ehc->i.action | ehc->i.dev_action[dev->devno];
434 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
435 struct ata_eh_info *ehi, unsigned int action)
437 struct ata_device *tdev;
440 ehi->action &= ~action;
441 ata_for_each_dev(tdev, link, ALL)
442 ehi->dev_action[tdev->devno] &= ~action;
444 /* doesn't make sense for port-wide EH actions */
445 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
447 /* break ehi->action into ehi->dev_action */
448 if (ehi->action & action) {
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] |=
451 ehi->action & action;
452 ehi->action &= ~action;
455 /* turn off the specified per-dev action */
456 ehi->dev_action[dev->devno] &= ~action;
461 * ata_eh_acquire - acquire EH ownership
462 * @ap: ATA port to acquire EH ownership for
464 * Acquire EH ownership for @ap. This is the basic exclusion
465 * mechanism for ports sharing a host. Only one port hanging off
466 * the same host can claim the ownership of EH.
471 void ata_eh_acquire(struct ata_port *ap)
473 mutex_lock(&ap->host->eh_mutex);
474 WARN_ON_ONCE(ap->host->eh_owner);
475 ap->host->eh_owner = current;
479 * ata_eh_release - release EH ownership
480 * @ap: ATA port to release EH ownership for
482 * Release EH ownership for @ap if the caller. The caller must
483 * have acquired EH ownership using ata_eh_acquire() previously.
488 void ata_eh_release(struct ata_port *ap)
490 WARN_ON_ONCE(ap->host->eh_owner != current);
491 ap->host->eh_owner = NULL;
492 mutex_unlock(&ap->host->eh_mutex);
495 static void ata_eh_unload(struct ata_port *ap)
497 struct ata_link *link;
498 struct ata_device *dev;
501 /* Restore SControl IPM and SPD for the next driver and
502 * disable attached devices.
504 ata_for_each_link(link, ap, PMP_FIRST) {
505 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
506 ata_for_each_dev(dev, link, ALL)
507 ata_dev_disable(dev);
510 /* freeze and set UNLOADED */
511 spin_lock_irqsave(ap->lock, flags);
513 ata_port_freeze(ap); /* won't be thawed */
514 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
515 ap->pflags |= ATA_PFLAG_UNLOADED;
517 spin_unlock_irqrestore(ap->lock, flags);
521 * ata_scsi_error - SCSI layer error handler callback
522 * @host: SCSI host on which error occurred
524 * Handles SCSI-layer-thrown error events.
527 * Inherited from SCSI layer (none, can sleep)
532 void ata_scsi_error(struct Scsi_Host *host)
534 struct ata_port *ap = ata_shost_to_port(host);
536 LIST_HEAD(eh_work_q);
538 spin_lock_irqsave(host->host_lock, flags);
539 list_splice_init(&host->eh_cmd_q, &eh_work_q);
540 spin_unlock_irqrestore(host->host_lock, flags);
542 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
544 /* If we timed raced normal completion and there is nothing to
545 recover nr_timedout == 0 why exactly are we doing error recovery ? */
546 ata_scsi_port_error_handler(host, ap);
548 /* finish or retry handled scmd's and clean up */
549 WARN_ON(!list_empty(&eh_work_q));
554 * ata_scsi_cmd_error_handler - error callback for a list of commands
555 * @host: scsi host containing the port
556 * @ap: ATA port within the host
557 * @eh_work_q: list of commands to process
559 * process the given list of commands and return those finished to the
560 * ap->eh_done_q. This function is the first part of the libata error
561 * handler which processes a given list of failed commands.
563 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
564 struct list_head *eh_work_q)
568 struct scsi_cmnd *scmd, *tmp;
571 /* make sure sff pio task is not running */
572 ata_sff_flush_pio_task(ap);
574 if (!ap->ops->error_handler)
577 /* synchronize with host lock and sort out timeouts */
580 * For new EH, all qcs are finished in one of three ways -
581 * normal completion, error completion, and SCSI timeout.
582 * Both completions can race against SCSI timeout. When normal
583 * completion wins, the qc never reaches EH. When error
584 * completion wins, the qc has ATA_QCFLAG_EH set.
586 * When SCSI timeout wins, things are a bit more complex.
587 * Normal or error completion can occur after the timeout but
588 * before this point. In such cases, both types of
589 * completions are honored. A scmd is determined to have
590 * timed out iff its associated qc is active and not failed.
592 spin_lock_irqsave(ap->lock, flags);
595 * This must occur under the ap->lock as we don't want
596 * a polled recovery to race the real interrupt handler
598 * The lost_interrupt handler checks for any completed but
599 * non-notified command and completes much like an IRQ handler.
601 * We then fall into the error recovery code which will treat
602 * this as if normal completion won the race
604 if (ap->ops->lost_interrupt)
605 ap->ops->lost_interrupt(ap);
607 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
608 struct ata_queued_cmd *qc;
610 ata_qc_for_each_raw(ap, qc, i) {
611 if (qc->flags & ATA_QCFLAG_ACTIVE &&
616 if (i < ATA_MAX_QUEUE) {
617 /* the scmd has an associated qc */
618 if (!(qc->flags & ATA_QCFLAG_EH)) {
619 /* which hasn't failed yet, timeout */
620 qc->err_mask |= AC_ERR_TIMEOUT;
621 qc->flags |= ATA_QCFLAG_EH;
625 /* Normal completion occurred after
626 * SCSI timeout but before this point.
627 * Successfully complete it.
629 scmd->retries = scmd->allowed;
630 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
635 * If we have timed out qcs. They belong to EH from
636 * this point but the state of the controller is
637 * unknown. Freeze the port to make sure the IRQ
638 * handler doesn't diddle with those qcs. This must
639 * be done atomically w.r.t. setting ATA_QCFLAG_EH.
642 __ata_port_freeze(ap);
644 /* initialize eh_tries */
645 ap->eh_tries = ATA_EH_MAX_TRIES;
647 spin_unlock_irqrestore(ap->lock, flags);
649 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
652 * ata_scsi_port_error_handler - recover the port after the commands
653 * @host: SCSI host containing the port
656 * Handle the recovery of the port @ap after all the commands
657 * have been recovered.
659 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
663 /* invoke error handler */
664 if (ap->ops->error_handler) {
665 struct ata_link *link;
667 /* acquire EH ownership */
670 /* kill fast drain timer */
671 del_timer_sync(&ap->fastdrain_timer);
673 /* process port resume request */
674 ata_eh_handle_port_resume(ap);
676 /* fetch & clear EH info */
677 spin_lock_irqsave(ap->lock, flags);
679 ata_for_each_link(link, ap, HOST_FIRST) {
680 struct ata_eh_context *ehc = &link->eh_context;
681 struct ata_device *dev;
683 memset(&link->eh_context, 0, sizeof(link->eh_context));
684 link->eh_context.i = link->eh_info;
685 memset(&link->eh_info, 0, sizeof(link->eh_info));
687 ata_for_each_dev(dev, link, ENABLED) {
688 int devno = dev->devno;
690 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
691 if (ata_ncq_enabled(dev))
692 ehc->saved_ncq_enabled |= 1 << devno;
696 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
697 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
698 ap->excl_link = NULL; /* don't maintain exclusion over EH */
700 spin_unlock_irqrestore(ap->lock, flags);
702 /* invoke EH, skip if unloading or suspended */
703 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
704 ap->ops->error_handler(ap);
706 /* if unloading, commence suicide */
707 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
708 !(ap->pflags & ATA_PFLAG_UNLOADED))
713 /* process port suspend request */
714 ata_eh_handle_port_suspend(ap);
716 /* Exception might have happened after ->error_handler
717 * recovered the port but before this point. Repeat
720 spin_lock_irqsave(ap->lock, flags);
722 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
723 if (--ap->eh_tries) {
724 spin_unlock_irqrestore(ap->lock, flags);
728 "EH pending after %d tries, giving up\n",
730 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
733 /* this run is complete, make sure EH info is clear */
734 ata_for_each_link(link, ap, HOST_FIRST)
735 memset(&link->eh_info, 0, sizeof(link->eh_info));
737 /* end eh (clear host_eh_scheduled) while holding
738 * ap->lock such that if exception occurs after this
739 * point but before EH completion, SCSI midlayer will
744 spin_unlock_irqrestore(ap->lock, flags);
747 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
748 ap->ops->eng_timeout(ap);
751 scsi_eh_flush_done_q(&ap->eh_done_q);
754 spin_lock_irqsave(ap->lock, flags);
756 if (ap->pflags & ATA_PFLAG_LOADING)
757 ap->pflags &= ~ATA_PFLAG_LOADING;
758 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
759 !(ap->flags & ATA_FLAG_SAS_HOST))
760 schedule_delayed_work(&ap->hotplug_task, 0);
762 if (ap->pflags & ATA_PFLAG_RECOVERED)
763 ata_port_info(ap, "EH complete\n");
765 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
767 /* tell wait_eh that we're done */
768 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
769 wake_up_all(&ap->eh_wait_q);
771 spin_unlock_irqrestore(ap->lock, flags);
773 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
776 * ata_port_wait_eh - Wait for the currently pending EH to complete
777 * @ap: Port to wait EH for
779 * Wait until the currently pending EH is complete.
782 * Kernel thread context (may sleep).
784 void ata_port_wait_eh(struct ata_port *ap)
790 spin_lock_irqsave(ap->lock, flags);
792 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
793 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
794 spin_unlock_irqrestore(ap->lock, flags);
796 spin_lock_irqsave(ap->lock, flags);
798 finish_wait(&ap->eh_wait_q, &wait);
800 spin_unlock_irqrestore(ap->lock, flags);
802 /* make sure SCSI EH is complete */
803 if (scsi_host_in_recovery(ap->scsi_host)) {
808 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
810 static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
812 struct ata_queued_cmd *qc;
816 /* count only non-internal commands */
817 ata_qc_for_each(ap, qc, tag) {
825 void ata_eh_fastdrain_timerfn(struct timer_list *t)
827 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
831 spin_lock_irqsave(ap->lock, flags);
833 cnt = ata_eh_nr_in_flight(ap);
839 if (cnt == ap->fastdrain_cnt) {
840 struct ata_queued_cmd *qc;
843 /* No progress during the last interval, tag all
844 * in-flight qcs as timed out and freeze the port.
846 ata_qc_for_each(ap, qc, tag) {
848 qc->err_mask |= AC_ERR_TIMEOUT;
853 /* some qcs have finished, give it another chance */
854 ap->fastdrain_cnt = cnt;
855 ap->fastdrain_timer.expires =
856 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
857 add_timer(&ap->fastdrain_timer);
861 spin_unlock_irqrestore(ap->lock, flags);
865 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
866 * @ap: target ATA port
867 * @fastdrain: activate fast drain
869 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
870 * is non-zero and EH wasn't pending before. Fast drain ensures
871 * that EH kicks in in timely manner.
874 * spin_lock_irqsave(host lock)
876 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
880 /* already scheduled? */
881 if (ap->pflags & ATA_PFLAG_EH_PENDING)
884 ap->pflags |= ATA_PFLAG_EH_PENDING;
889 /* do we have in-flight qcs? */
890 cnt = ata_eh_nr_in_flight(ap);
894 /* activate fast drain */
895 ap->fastdrain_cnt = cnt;
896 ap->fastdrain_timer.expires =
897 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
898 add_timer(&ap->fastdrain_timer);
902 * ata_qc_schedule_eh - schedule qc for error handling
903 * @qc: command to schedule error handling for
905 * Schedule error handling for @qc. EH will kick in as soon as
906 * other commands are drained.
909 * spin_lock_irqsave(host lock)
911 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
913 struct ata_port *ap = qc->ap;
915 WARN_ON(!ap->ops->error_handler);
917 qc->flags |= ATA_QCFLAG_EH;
918 ata_eh_set_pending(ap, 1);
920 /* The following will fail if timeout has already expired.
921 * ata_scsi_error() takes care of such scmds on EH entry.
922 * Note that ATA_QCFLAG_EH is unconditionally set after
923 * this function completes.
925 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
929 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
930 * @ap: ATA port to schedule EH for
932 * LOCKING: inherited from ata_port_schedule_eh
933 * spin_lock_irqsave(host lock)
935 void ata_std_sched_eh(struct ata_port *ap)
937 WARN_ON(!ap->ops->error_handler);
939 if (ap->pflags & ATA_PFLAG_INITIALIZING)
942 ata_eh_set_pending(ap, 1);
943 scsi_schedule_eh(ap->scsi_host);
945 trace_ata_std_sched_eh(ap);
947 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
950 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
951 * @ap: ATA port to end EH for
953 * In the libata object model there is a 1:1 mapping of ata_port to
954 * shost, so host fields can be directly manipulated under ap->lock, in
955 * the libsas case we need to hold a lock at the ha->level to coordinate
959 * spin_lock_irqsave(host lock)
961 void ata_std_end_eh(struct ata_port *ap)
963 struct Scsi_Host *host = ap->scsi_host;
965 host->host_eh_scheduled = 0;
967 EXPORT_SYMBOL(ata_std_end_eh);
971 * ata_port_schedule_eh - schedule error handling without a qc
972 * @ap: ATA port to schedule EH for
974 * Schedule error handling for @ap. EH will kick in as soon as
975 * all commands are drained.
978 * spin_lock_irqsave(host lock)
980 void ata_port_schedule_eh(struct ata_port *ap)
982 /* see: ata_std_sched_eh, unless you know better */
983 ap->ops->sched_eh(ap);
985 EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
987 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
989 struct ata_queued_cmd *qc;
990 int tag, nr_aborted = 0;
992 WARN_ON(!ap->ops->error_handler);
994 /* we're gonna abort all commands, no need for fast drain */
995 ata_eh_set_pending(ap, 0);
997 /* include internal tag in iteration */
998 ata_qc_for_each_with_internal(ap, qc, tag) {
999 if (qc && (!link || qc->dev->link == link)) {
1000 qc->flags |= ATA_QCFLAG_EH;
1001 ata_qc_complete(qc);
1007 ata_port_schedule_eh(ap);
1013 * ata_link_abort - abort all qc's on the link
1014 * @link: ATA link to abort qc's for
1016 * Abort all active qc's active on @link and schedule EH.
1019 * spin_lock_irqsave(host lock)
1022 * Number of aborted qc's.
1024 int ata_link_abort(struct ata_link *link)
1026 return ata_do_link_abort(link->ap, link);
1028 EXPORT_SYMBOL_GPL(ata_link_abort);
1031 * ata_port_abort - abort all qc's on the port
1032 * @ap: ATA port to abort qc's for
1034 * Abort all active qc's of @ap and schedule EH.
1037 * spin_lock_irqsave(host_set lock)
1040 * Number of aborted qc's.
1042 int ata_port_abort(struct ata_port *ap)
1044 return ata_do_link_abort(ap, NULL);
1046 EXPORT_SYMBOL_GPL(ata_port_abort);
1049 * __ata_port_freeze - freeze port
1050 * @ap: ATA port to freeze
1052 * This function is called when HSM violation or some other
1053 * condition disrupts normal operation of the port. Frozen port
1054 * is not allowed to perform any operation until the port is
1055 * thawed, which usually follows a successful reset.
1057 * ap->ops->freeze() callback can be used for freezing the port
1058 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1059 * port cannot be frozen hardware-wise, the interrupt handler
1060 * must ack and clear interrupts unconditionally while the port
1064 * spin_lock_irqsave(host lock)
1066 static void __ata_port_freeze(struct ata_port *ap)
1068 WARN_ON(!ap->ops->error_handler);
1070 if (ap->ops->freeze)
1071 ap->ops->freeze(ap);
1073 ap->pflags |= ATA_PFLAG_FROZEN;
1075 trace_ata_port_freeze(ap);
1079 * ata_port_freeze - abort & freeze port
1080 * @ap: ATA port to freeze
1082 * Abort and freeze @ap. The freeze operation must be called
1083 * first, because some hardware requires special operations
1084 * before the taskfile registers are accessible.
1087 * spin_lock_irqsave(host lock)
1090 * Number of aborted commands.
1092 int ata_port_freeze(struct ata_port *ap)
1094 WARN_ON(!ap->ops->error_handler);
1096 __ata_port_freeze(ap);
1098 return ata_port_abort(ap);
1100 EXPORT_SYMBOL_GPL(ata_port_freeze);
1103 * ata_eh_freeze_port - EH helper to freeze port
1104 * @ap: ATA port to freeze
1111 void ata_eh_freeze_port(struct ata_port *ap)
1113 unsigned long flags;
1115 if (!ap->ops->error_handler)
1118 spin_lock_irqsave(ap->lock, flags);
1119 __ata_port_freeze(ap);
1120 spin_unlock_irqrestore(ap->lock, flags);
1122 EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1125 * ata_eh_thaw_port - EH helper to thaw port
1126 * @ap: ATA port to thaw
1128 * Thaw frozen port @ap.
1133 void ata_eh_thaw_port(struct ata_port *ap)
1135 unsigned long flags;
1137 if (!ap->ops->error_handler)
1140 spin_lock_irqsave(ap->lock, flags);
1142 ap->pflags &= ~ATA_PFLAG_FROZEN;
1147 spin_unlock_irqrestore(ap->lock, flags);
1149 trace_ata_port_thaw(ap);
1152 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1157 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1159 struct ata_port *ap = qc->ap;
1160 struct scsi_cmnd *scmd = qc->scsicmd;
1161 unsigned long flags;
1163 spin_lock_irqsave(ap->lock, flags);
1164 qc->scsidone = ata_eh_scsidone;
1165 __ata_qc_complete(qc);
1166 WARN_ON(ata_tag_valid(qc->tag));
1167 spin_unlock_irqrestore(ap->lock, flags);
1169 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1173 * ata_eh_qc_complete - Complete an active ATA command from EH
1174 * @qc: Command to complete
1176 * Indicate to the mid and upper layers that an ATA command has
1177 * completed. To be used from EH.
1179 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1181 struct scsi_cmnd *scmd = qc->scsicmd;
1182 scmd->retries = scmd->allowed;
1183 __ata_eh_qc_complete(qc);
1187 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1188 * @qc: Command to retry
1190 * Indicate to the mid and upper layers that an ATA command
1191 * should be retried. To be used from EH.
1193 * SCSI midlayer limits the number of retries to scmd->allowed.
1194 * scmd->allowed is incremented for commands which get retried
1195 * due to unrelated failures (qc->err_mask is zero).
1197 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1199 struct scsi_cmnd *scmd = qc->scsicmd;
1202 __ata_eh_qc_complete(qc);
1206 * ata_dev_disable - disable ATA device
1207 * @dev: ATA device to disable
1214 void ata_dev_disable(struct ata_device *dev)
1216 if (!ata_dev_enabled(dev))
1219 ata_dev_warn(dev, "disable device\n");
1220 ata_acpi_on_disable(dev);
1221 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1224 /* From now till the next successful probe, ering is used to
1225 * track probe failures. Clear accumulated device error info.
1227 ata_ering_clear(&dev->ering);
1229 EXPORT_SYMBOL_GPL(ata_dev_disable);
1232 * ata_eh_detach_dev - detach ATA device
1233 * @dev: ATA device to detach
1240 void ata_eh_detach_dev(struct ata_device *dev)
1242 struct ata_link *link = dev->link;
1243 struct ata_port *ap = link->ap;
1244 struct ata_eh_context *ehc = &link->eh_context;
1245 unsigned long flags;
1247 ata_dev_disable(dev);
1249 spin_lock_irqsave(ap->lock, flags);
1251 dev->flags &= ~ATA_DFLAG_DETACH;
1253 if (ata_scsi_offline_dev(dev)) {
1254 dev->flags |= ATA_DFLAG_DETACHED;
1255 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1258 /* clear per-dev EH info */
1259 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1260 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1261 ehc->saved_xfer_mode[dev->devno] = 0;
1262 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1264 spin_unlock_irqrestore(ap->lock, flags);
1268 * ata_eh_about_to_do - about to perform eh_action
1269 * @link: target ATA link
1270 * @dev: target ATA dev for per-dev action (can be NULL)
1271 * @action: action about to be performed
1273 * Called just before performing EH actions to clear related bits
1274 * in @link->eh_info such that eh actions are not unnecessarily
1280 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1281 unsigned int action)
1283 struct ata_port *ap = link->ap;
1284 struct ata_eh_info *ehi = &link->eh_info;
1285 struct ata_eh_context *ehc = &link->eh_context;
1286 unsigned long flags;
1288 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1290 spin_lock_irqsave(ap->lock, flags);
1292 ata_eh_clear_action(link, dev, ehi, action);
1294 /* About to take EH action, set RECOVERED. Ignore actions on
1295 * slave links as master will do them again.
1297 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1298 ap->pflags |= ATA_PFLAG_RECOVERED;
1300 spin_unlock_irqrestore(ap->lock, flags);
1304 * ata_eh_done - EH action complete
1305 * @link: ATA link for which EH actions are complete
1306 * @dev: target ATA dev for per-dev action (can be NULL)
1307 * @action: action just completed
1309 * Called right after performing EH actions to clear related bits
1310 * in @link->eh_context.
1315 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1316 unsigned int action)
1318 struct ata_eh_context *ehc = &link->eh_context;
1320 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1322 ata_eh_clear_action(link, dev, &ehc->i, action);
1326 * ata_err_string - convert err_mask to descriptive string
1327 * @err_mask: error mask to convert to string
1329 * Convert @err_mask to descriptive string. Errors are
1330 * prioritized according to severity and only the most severe
1331 * error is reported.
1337 * Descriptive string for @err_mask
1339 static const char *ata_err_string(unsigned int err_mask)
1341 if (err_mask & AC_ERR_HOST_BUS)
1342 return "host bus error";
1343 if (err_mask & AC_ERR_ATA_BUS)
1344 return "ATA bus error";
1345 if (err_mask & AC_ERR_TIMEOUT)
1347 if (err_mask & AC_ERR_HSM)
1348 return "HSM violation";
1349 if (err_mask & AC_ERR_SYSTEM)
1350 return "internal error";
1351 if (err_mask & AC_ERR_MEDIA)
1352 return "media error";
1353 if (err_mask & AC_ERR_INVALID)
1354 return "invalid argument";
1355 if (err_mask & AC_ERR_DEV)
1356 return "device error";
1357 if (err_mask & AC_ERR_NCQ)
1359 if (err_mask & AC_ERR_NODEV_HINT)
1360 return "Polling detection error";
1361 return "unknown error";
1365 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1366 * @dev: target ATAPI device
1367 * @r_sense_key: out parameter for sense_key
1369 * Perform ATAPI TEST_UNIT_READY.
1372 * EH context (may sleep).
1375 * 0 on success, AC_ERR_* mask on failure.
1377 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1379 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1380 struct ata_taskfile tf;
1381 unsigned int err_mask;
1383 ata_tf_init(dev, &tf);
1385 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1386 tf.command = ATA_CMD_PACKET;
1387 tf.protocol = ATAPI_PROT_NODATA;
1389 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1390 if (err_mask == AC_ERR_DEV)
1391 *r_sense_key = tf.error >> 4;
1396 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1397 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1399 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1400 * SENSE. This function is an EH helper.
1403 * Kernel thread context (may sleep).
1405 static void ata_eh_request_sense(struct ata_queued_cmd *qc)
1407 struct scsi_cmnd *cmd = qc->scsicmd;
1408 struct ata_device *dev = qc->dev;
1409 struct ata_taskfile tf;
1410 unsigned int err_mask;
1412 if (ata_port_is_frozen(qc->ap)) {
1413 ata_dev_warn(dev, "sense data available but port frozen\n");
1417 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1420 if (!ata_id_sense_reporting_enabled(dev->id)) {
1421 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1425 ata_tf_init(dev, &tf);
1426 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1427 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1428 tf.command = ATA_CMD_REQ_SENSE_DATA;
1429 tf.protocol = ATA_PROT_NODATA;
1431 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1432 /* Ignore err_mask; ATA_ERR might be set */
1433 if (tf.status & ATA_SENSE) {
1434 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1435 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1436 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1439 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1440 tf.status, err_mask);
1445 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1446 * @dev: device to perform REQUEST_SENSE to
1447 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1448 * @dfl_sense_key: default sense key to use
1450 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1451 * SENSE. This function is EH helper.
1454 * Kernel thread context (may sleep).
1457 * 0 on success, AC_ERR_* mask on failure
1459 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1460 u8 *sense_buf, u8 dfl_sense_key)
1462 u8 cdb[ATAPI_CDB_LEN] =
1463 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1464 struct ata_port *ap = dev->link->ap;
1465 struct ata_taskfile tf;
1467 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1469 /* initialize sense_buf with the error register,
1470 * for the case where they are -not- overwritten
1472 sense_buf[0] = 0x70;
1473 sense_buf[2] = dfl_sense_key;
1475 /* some devices time out if garbage left in tf */
1476 ata_tf_init(dev, &tf);
1478 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1479 tf.command = ATA_CMD_PACKET;
1481 /* is it pointless to prefer PIO for "safety reasons"? */
1482 if (ap->flags & ATA_FLAG_PIO_DMA) {
1483 tf.protocol = ATAPI_PROT_DMA;
1484 tf.feature |= ATAPI_PKT_DMA;
1486 tf.protocol = ATAPI_PROT_PIO;
1487 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1491 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1492 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1496 * ata_eh_analyze_serror - analyze SError for a failed port
1497 * @link: ATA link to analyze SError for
1499 * Analyze SError if available and further determine cause of
1505 static void ata_eh_analyze_serror(struct ata_link *link)
1507 struct ata_eh_context *ehc = &link->eh_context;
1508 u32 serror = ehc->i.serror;
1509 unsigned int err_mask = 0, action = 0;
1512 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1513 err_mask |= AC_ERR_ATA_BUS;
1514 action |= ATA_EH_RESET;
1516 if (serror & SERR_PROTOCOL) {
1517 err_mask |= AC_ERR_HSM;
1518 action |= ATA_EH_RESET;
1520 if (serror & SERR_INTERNAL) {
1521 err_mask |= AC_ERR_SYSTEM;
1522 action |= ATA_EH_RESET;
1525 /* Determine whether a hotplug event has occurred. Both
1526 * SError.N/X are considered hotplug events for enabled or
1527 * host links. For disabled PMP links, only N bit is
1528 * considered as X bit is left at 1 for link plugging.
1530 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1531 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1532 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1533 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1535 hotplug_mask = SERR_PHYRDY_CHG;
1537 if (serror & hotplug_mask)
1538 ata_ehi_hotplugged(&ehc->i);
1540 ehc->i.err_mask |= err_mask;
1541 ehc->i.action |= action;
1545 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1546 * @qc: qc to analyze
1548 * Analyze taskfile of @qc and further determine cause of
1549 * failure. This function also requests ATAPI sense data if
1553 * Kernel thread context (may sleep).
1556 * Determined recovery action
1558 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1560 const struct ata_taskfile *tf = &qc->result_tf;
1561 unsigned int tmp, action = 0;
1562 u8 stat = tf->status, err = tf->error;
1564 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1565 qc->err_mask |= AC_ERR_HSM;
1566 return ATA_EH_RESET;
1569 if (stat & (ATA_ERR | ATA_DF)) {
1570 qc->err_mask |= AC_ERR_DEV;
1572 * Sense data reporting does not work if the
1573 * device fault bit is set.
1581 switch (qc->dev->class) {
1585 * Fetch the sense data explicitly if:
1586 * -It was a non-NCQ command that failed, or
1587 * -It was a NCQ command that failed, but the sense data
1588 * was not included in the NCQ command error log
1589 * (i.e. NCQ autosense is not supported by the device).
1591 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) && (stat & ATA_SENSE))
1592 ata_eh_request_sense(qc);
1594 qc->err_mask |= AC_ERR_ATA_BUS;
1595 if (err & (ATA_UNC | ATA_AMNF))
1596 qc->err_mask |= AC_ERR_MEDIA;
1598 qc->err_mask |= AC_ERR_INVALID;
1602 if (!ata_port_is_frozen(qc->ap)) {
1603 tmp = atapi_eh_request_sense(qc->dev,
1604 qc->scsicmd->sense_buffer,
1605 qc->result_tf.error >> 4);
1607 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1609 qc->err_mask |= tmp;
1613 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1614 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
1616 * SUCCESS here means that the sense code could be
1617 * evaluated and should be passed to the upper layers
1618 * for correct evaluation.
1619 * FAILED means the sense code could not be interpreted
1620 * and the device would need to be reset.
1621 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1622 * command would need to be retried.
1624 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1625 qc->flags |= ATA_QCFLAG_RETRY;
1626 qc->err_mask |= AC_ERR_OTHER;
1627 } else if (ret != SUCCESS) {
1628 qc->err_mask |= AC_ERR_HSM;
1631 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1632 action |= ATA_EH_RESET;
1637 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1642 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1646 base = ATA_ECAT_DUBIOUS_NONE;
1648 if (err_mask & AC_ERR_ATA_BUS)
1649 return base + ATA_ECAT_ATA_BUS;
1651 if (err_mask & AC_ERR_TIMEOUT)
1652 return base + ATA_ECAT_TOUT_HSM;
1654 if (eflags & ATA_EFLAG_IS_IO) {
1655 if (err_mask & AC_ERR_HSM)
1656 return base + ATA_ECAT_TOUT_HSM;
1658 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1659 return base + ATA_ECAT_UNK_DEV;
1665 struct speed_down_verdict_arg {
1668 int nr_errors[ATA_ECAT_NR];
1671 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1673 struct speed_down_verdict_arg *arg = void_arg;
1676 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1679 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1681 arg->nr_errors[cat]++;
1687 * ata_eh_speed_down_verdict - Determine speed down verdict
1688 * @dev: Device of interest
1690 * This function examines error ring of @dev and determines
1691 * whether NCQ needs to be turned off, transfer speed should be
1692 * stepped down, or falling back to PIO is necessary.
1694 * ECAT_ATA_BUS : ATA_BUS error for any command
1696 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1699 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1701 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1702 * data transfer hasn't been verified.
1706 * NCQ_OFF : Turn off NCQ.
1708 * SPEED_DOWN : Speed down transfer speed but don't fall back
1711 * FALLBACK_TO_PIO : Fall back to PIO.
1713 * Even if multiple verdicts are returned, only one action is
1714 * taken per error. An action triggered by non-DUBIOUS errors
1715 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1716 * This is to expedite speed down decisions right after device is
1717 * initially configured.
1719 * The following are speed down rules. #1 and #2 deal with
1722 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1723 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1725 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1726 * occurred during last 5 mins, NCQ_OFF.
1728 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1729 * occurred during last 5 mins, FALLBACK_TO_PIO
1731 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1732 * during last 10 mins, NCQ_OFF.
1734 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1735 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1738 * Inherited from caller.
1741 * OR of ATA_EH_SPDN_* flags.
1743 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1745 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1746 u64 j64 = get_jiffies_64();
1747 struct speed_down_verdict_arg arg;
1748 unsigned int verdict = 0;
1750 /* scan past 5 mins of error history */
1751 memset(&arg, 0, sizeof(arg));
1752 arg.since = j64 - min(j64, j5mins);
1753 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1755 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1756 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1757 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1758 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1760 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1761 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1762 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1764 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1765 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1766 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1767 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1769 /* scan past 10 mins of error history */
1770 memset(&arg, 0, sizeof(arg));
1771 arg.since = j64 - min(j64, j10mins);
1772 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1774 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1775 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1776 verdict |= ATA_EH_SPDN_NCQ_OFF;
1778 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1779 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1780 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1781 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1787 * ata_eh_speed_down - record error and speed down if necessary
1788 * @dev: Failed device
1789 * @eflags: mask of ATA_EFLAG_* flags
1790 * @err_mask: err_mask of the error
1792 * Record error and examine error history to determine whether
1793 * adjusting transmission speed is necessary. It also sets
1794 * transmission limits appropriately if such adjustment is
1798 * Kernel thread context (may sleep).
1801 * Determined recovery action.
1803 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1804 unsigned int eflags, unsigned int err_mask)
1806 struct ata_link *link = ata_dev_phys_link(dev);
1808 unsigned int verdict;
1809 unsigned int action = 0;
1811 /* don't bother if Cat-0 error */
1812 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1815 /* record error and determine whether speed down is necessary */
1816 ata_ering_record(&dev->ering, eflags, err_mask);
1817 verdict = ata_eh_speed_down_verdict(dev);
1820 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1821 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1822 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1823 dev->flags |= ATA_DFLAG_NCQ_OFF;
1824 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1829 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1830 /* speed down SATA link speed if possible */
1831 if (sata_down_spd_limit(link, 0) == 0) {
1832 action |= ATA_EH_RESET;
1836 /* lower transfer mode */
1837 if (dev->spdn_cnt < 2) {
1838 static const int dma_dnxfer_sel[] =
1839 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1840 static const int pio_dnxfer_sel[] =
1841 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1844 if (dev->xfer_shift != ATA_SHIFT_PIO)
1845 sel = dma_dnxfer_sel[dev->spdn_cnt];
1847 sel = pio_dnxfer_sel[dev->spdn_cnt];
1851 if (ata_down_xfermask_limit(dev, sel) == 0) {
1852 action |= ATA_EH_RESET;
1858 /* Fall back to PIO? Slowing down to PIO is meaningless for
1859 * SATA ATA devices. Consider it only for PATA and SATAPI.
1861 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1862 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1863 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1864 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1866 action |= ATA_EH_RESET;
1873 /* device has been slowed down, blow error history */
1874 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1875 ata_ering_clear(&dev->ering);
1880 * ata_eh_worth_retry - analyze error and decide whether to retry
1881 * @qc: qc to possibly retry
1883 * Look at the cause of the error and decide if a retry
1884 * might be useful or not. We don't want to retry media errors
1885 * because the drive itself has probably already taken 10-30 seconds
1886 * doing its own internal retries before reporting the failure.
1888 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1890 if (qc->err_mask & AC_ERR_MEDIA)
1891 return 0; /* don't retry media errors */
1892 if (qc->flags & ATA_QCFLAG_IO)
1893 return 1; /* otherwise retry anything from fs stack */
1894 if (qc->err_mask & AC_ERR_INVALID)
1895 return 0; /* don't retry these */
1896 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1900 * ata_eh_quiet - check if we need to be quiet about a command error
1903 * Look at the qc flags anbd its scsi command request flags to determine
1904 * if we need to be quiet about the command failure.
1906 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1908 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1909 qc->flags |= ATA_QCFLAG_QUIET;
1910 return qc->flags & ATA_QCFLAG_QUIET;
1914 * ata_eh_link_autopsy - analyze error and determine recovery action
1915 * @link: host link to perform autopsy on
1917 * Analyze why @link failed and determine which recovery actions
1918 * are needed. This function also sets more detailed AC_ERR_*
1919 * values and fills sense data for ATAPI CHECK SENSE.
1922 * Kernel thread context (may sleep).
1924 static void ata_eh_link_autopsy(struct ata_link *link)
1926 struct ata_port *ap = link->ap;
1927 struct ata_eh_context *ehc = &link->eh_context;
1928 struct ata_queued_cmd *qc;
1929 struct ata_device *dev;
1930 unsigned int all_err_mask = 0, eflags = 0;
1931 int tag, nr_failed = 0, nr_quiet = 0;
1935 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1938 /* obtain and analyze SError */
1939 rc = sata_scr_read(link, SCR_ERROR, &serror);
1941 ehc->i.serror |= serror;
1942 ata_eh_analyze_serror(link);
1943 } else if (rc != -EOPNOTSUPP) {
1944 /* SError read failed, force reset and probing */
1945 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1946 ehc->i.action |= ATA_EH_RESET;
1947 ehc->i.err_mask |= AC_ERR_OTHER;
1950 /* analyze NCQ failure */
1951 ata_eh_analyze_ncq_error(link);
1953 /* any real error trumps AC_ERR_OTHER */
1954 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1955 ehc->i.err_mask &= ~AC_ERR_OTHER;
1957 all_err_mask |= ehc->i.err_mask;
1959 ata_qc_for_each_raw(ap, qc, tag) {
1960 if (!(qc->flags & ATA_QCFLAG_EH) ||
1961 qc->flags & ATA_QCFLAG_RETRY ||
1962 ata_dev_phys_link(qc->dev) != link)
1965 /* inherit upper level err_mask */
1966 qc->err_mask |= ehc->i.err_mask;
1969 ehc->i.action |= ata_eh_analyze_tf(qc);
1971 /* DEV errors are probably spurious in case of ATA_BUS error */
1972 if (qc->err_mask & AC_ERR_ATA_BUS)
1973 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1976 /* any real error trumps unknown error */
1977 if (qc->err_mask & ~AC_ERR_OTHER)
1978 qc->err_mask &= ~AC_ERR_OTHER;
1981 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
1982 * layers will determine whether the command is worth retrying
1983 * based on the sense data and device class/type. Otherwise,
1984 * determine directly if the command is worth retrying using its
1985 * error mask and flags.
1987 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1988 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1989 else if (ata_eh_worth_retry(qc))
1990 qc->flags |= ATA_QCFLAG_RETRY;
1992 /* accumulate error info */
1993 ehc->i.dev = qc->dev;
1994 all_err_mask |= qc->err_mask;
1995 if (qc->flags & ATA_QCFLAG_IO)
1996 eflags |= ATA_EFLAG_IS_IO;
1997 trace_ata_eh_link_autopsy_qc(qc);
1999 /* Count quiet errors */
2000 if (ata_eh_quiet(qc))
2005 /* If all failed commands requested silence, then be quiet */
2006 if (nr_quiet == nr_failed)
2007 ehc->i.flags |= ATA_EHI_QUIET;
2009 /* enforce default EH actions */
2010 if (ata_port_is_frozen(ap) ||
2011 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2012 ehc->i.action |= ATA_EH_RESET;
2013 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2014 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2015 ehc->i.action |= ATA_EH_REVALIDATE;
2017 /* If we have offending qcs and the associated failed device,
2018 * perform per-dev EH action only on the offending device.
2021 ehc->i.dev_action[ehc->i.dev->devno] |=
2022 ehc->i.action & ATA_EH_PERDEV_MASK;
2023 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2026 /* propagate timeout to host link */
2027 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2028 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2030 /* record error and consider speeding down */
2032 if (!dev && ((ata_link_max_devices(link) == 1 &&
2033 ata_dev_enabled(link->device))))
2037 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2038 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2039 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2040 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2045 * ata_eh_autopsy - analyze error and determine recovery action
2046 * @ap: host port to perform autopsy on
2048 * Analyze all links of @ap and determine why they failed and
2049 * which recovery actions are needed.
2052 * Kernel thread context (may sleep).
2054 void ata_eh_autopsy(struct ata_port *ap)
2056 struct ata_link *link;
2058 ata_for_each_link(link, ap, EDGE)
2059 ata_eh_link_autopsy(link);
2061 /* Handle the frigging slave link. Autopsy is done similarly
2062 * but actions and flags are transferred over to the master
2063 * link and handled from there.
2065 if (ap->slave_link) {
2066 struct ata_eh_context *mehc = &ap->link.eh_context;
2067 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2069 /* transfer control flags from master to slave */
2070 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2072 /* perform autopsy on the slave link */
2073 ata_eh_link_autopsy(ap->slave_link);
2075 /* transfer actions from slave to master and clear slave */
2076 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2077 mehc->i.action |= sehc->i.action;
2078 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2079 mehc->i.flags |= sehc->i.flags;
2080 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2083 /* Autopsy of fanout ports can affect host link autopsy.
2084 * Perform host link autopsy last.
2086 if (sata_pmp_attached(ap))
2087 ata_eh_link_autopsy(&ap->link);
2091 * ata_get_cmd_name - get name for ATA command
2092 * @command: ATA command code to get name for
2094 * Return a textual name of the given command or "unknown"
2099 const char *ata_get_cmd_name(u8 command)
2101 #ifdef CONFIG_ATA_VERBOSE_ERROR
2107 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2108 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2109 { ATA_CMD_STANDBY, "STANDBY" },
2110 { ATA_CMD_IDLE, "IDLE" },
2111 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2112 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2113 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2114 { ATA_CMD_NOP, "NOP" },
2115 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2116 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2117 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2118 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2119 { ATA_CMD_SERVICE, "SERVICE" },
2120 { ATA_CMD_READ, "READ DMA" },
2121 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2122 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2123 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2124 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2125 { ATA_CMD_WRITE, "WRITE DMA" },
2126 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2127 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2128 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2129 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2130 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2131 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2132 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2133 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2134 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" },
2135 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2136 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2137 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2138 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2139 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2140 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2141 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2142 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2143 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2144 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2145 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2146 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2147 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2148 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2149 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2150 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2151 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2152 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2153 { ATA_CMD_SLEEP, "SLEEP" },
2154 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2155 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2156 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2157 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2158 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2159 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2160 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2161 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2162 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2163 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2164 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2165 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2166 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2167 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2168 { ATA_CMD_PMP_READ, "READ BUFFER" },
2169 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2170 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2171 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2172 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2173 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2174 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2175 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2176 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2177 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2178 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2179 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2180 { ATA_CMD_SMART, "SMART" },
2181 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2182 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2183 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2184 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2185 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2186 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2187 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2188 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2189 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2190 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2191 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2192 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2193 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2194 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2195 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2196 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2197 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2198 { ATA_CMD_RESTORE, "RECALIBRATE" },
2199 { 0, NULL } /* terminate list */
2203 for (i = 0; cmd_descr[i].text; i++)
2204 if (cmd_descr[i].command == command)
2205 return cmd_descr[i].text;
2210 EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2213 * ata_eh_link_report - report error handling to user
2214 * @link: ATA link EH is going on
2216 * Report EH to user.
2221 static void ata_eh_link_report(struct ata_link *link)
2223 struct ata_port *ap = link->ap;
2224 struct ata_eh_context *ehc = &link->eh_context;
2225 struct ata_queued_cmd *qc;
2226 const char *frozen, *desc;
2227 char tries_buf[6] = "";
2228 int tag, nr_failed = 0;
2230 if (ehc->i.flags & ATA_EHI_QUIET)
2234 if (ehc->i.desc[0] != '\0')
2237 ata_qc_for_each_raw(ap, qc, tag) {
2238 if (!(qc->flags & ATA_QCFLAG_EH) ||
2239 ata_dev_phys_link(qc->dev) != link ||
2240 ((qc->flags & ATA_QCFLAG_QUIET) &&
2241 qc->err_mask == AC_ERR_DEV))
2243 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2249 if (!nr_failed && !ehc->i.err_mask)
2253 if (ata_port_is_frozen(ap))
2256 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2257 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2261 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2262 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2263 ehc->i.err_mask, link->sactive, ehc->i.serror,
2264 ehc->i.action, frozen, tries_buf);
2266 ata_dev_err(ehc->i.dev, "%s\n", desc);
2268 ata_link_err(link, "exception Emask 0x%x "
2269 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2270 ehc->i.err_mask, link->sactive, ehc->i.serror,
2271 ehc->i.action, frozen, tries_buf);
2273 ata_link_err(link, "%s\n", desc);
2276 #ifdef CONFIG_ATA_VERBOSE_ERROR
2279 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2280 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2281 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2282 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2283 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2284 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2285 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2286 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2287 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2288 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2289 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2290 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2291 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2292 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2293 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2294 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2295 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2296 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2299 ata_qc_for_each_raw(ap, qc, tag) {
2300 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2301 char data_buf[20] = "";
2302 char cdb_buf[70] = "";
2304 if (!(qc->flags & ATA_QCFLAG_EH) ||
2305 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2308 if (qc->dma_dir != DMA_NONE) {
2309 static const char *dma_str[] = {
2310 [DMA_BIDIRECTIONAL] = "bidi",
2311 [DMA_TO_DEVICE] = "out",
2312 [DMA_FROM_DEVICE] = "in",
2314 const char *prot_str = NULL;
2316 switch (qc->tf.protocol) {
2317 case ATA_PROT_UNKNOWN:
2318 prot_str = "unknown";
2320 case ATA_PROT_NODATA:
2321 prot_str = "nodata";
2330 prot_str = "ncq dma";
2332 case ATA_PROT_NCQ_NODATA:
2333 prot_str = "ncq nodata";
2335 case ATAPI_PROT_NODATA:
2336 prot_str = "nodata";
2338 case ATAPI_PROT_PIO:
2341 case ATAPI_PROT_DMA:
2345 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2346 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2349 if (ata_is_atapi(qc->tf.protocol)) {
2350 const u8 *cdb = qc->cdb;
2351 size_t cdb_len = qc->dev->cdb_len;
2354 cdb = qc->scsicmd->cmnd;
2355 cdb_len = qc->scsicmd->cmd_len;
2357 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2360 ata_dev_err(qc->dev, "failed command: %s\n",
2361 ata_get_cmd_name(cmd->command));
2363 ata_dev_err(qc->dev,
2364 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2366 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2367 "Emask 0x%x (%s)%s\n",
2368 cmd->command, cmd->feature, cmd->nsect,
2369 cmd->lbal, cmd->lbam, cmd->lbah,
2370 cmd->hob_feature, cmd->hob_nsect,
2371 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2372 cmd->device, qc->tag, data_buf, cdb_buf,
2373 res->status, res->error, res->nsect,
2374 res->lbal, res->lbam, res->lbah,
2375 res->hob_feature, res->hob_nsect,
2376 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2377 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2378 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2380 #ifdef CONFIG_ATA_VERBOSE_ERROR
2381 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2382 ATA_SENSE | ATA_ERR)) {
2383 if (res->status & ATA_BUSY)
2384 ata_dev_err(qc->dev, "status: { Busy }\n");
2386 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2387 res->status & ATA_DRDY ? "DRDY " : "",
2388 res->status & ATA_DF ? "DF " : "",
2389 res->status & ATA_DRQ ? "DRQ " : "",
2390 res->status & ATA_SENSE ? "SENSE " : "",
2391 res->status & ATA_ERR ? "ERR " : "");
2394 if (cmd->command != ATA_CMD_PACKET &&
2395 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2397 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2398 res->error & ATA_ICRC ? "ICRC " : "",
2399 res->error & ATA_UNC ? "UNC " : "",
2400 res->error & ATA_AMNF ? "AMNF " : "",
2401 res->error & ATA_IDNF ? "IDNF " : "",
2402 res->error & ATA_ABORTED ? "ABRT " : "");
2408 * ata_eh_report - report error handling to user
2409 * @ap: ATA port to report EH about
2411 * Report EH to user.
2416 void ata_eh_report(struct ata_port *ap)
2418 struct ata_link *link;
2420 ata_for_each_link(link, ap, HOST_FIRST)
2421 ata_eh_link_report(link);
2424 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2425 unsigned int *classes, unsigned long deadline,
2428 struct ata_device *dev;
2431 ata_for_each_dev(dev, link, ALL)
2432 classes[dev->devno] = ATA_DEV_UNKNOWN;
2434 return reset(link, classes, deadline);
2437 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2439 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2443 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2448 int ata_eh_reset(struct ata_link *link, int classify,
2449 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2450 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2452 struct ata_port *ap = link->ap;
2453 struct ata_link *slave = ap->slave_link;
2454 struct ata_eh_context *ehc = &link->eh_context;
2455 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2456 unsigned int *classes = ehc->classes;
2457 unsigned int lflags = link->flags;
2458 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2459 int max_tries = 0, try = 0;
2460 struct ata_link *failed_link;
2461 struct ata_device *dev;
2462 unsigned long deadline, now;
2463 ata_reset_fn_t reset;
2464 unsigned long flags;
2471 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2473 if (link->flags & ATA_LFLAG_RST_ONCE)
2475 if (link->flags & ATA_LFLAG_NO_HRST)
2477 if (link->flags & ATA_LFLAG_NO_SRST)
2480 /* make sure each reset attempt is at least COOL_DOWN apart */
2481 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2483 WARN_ON(time_after(ehc->last_reset, now));
2484 deadline = ata_deadline(ehc->last_reset,
2485 ATA_EH_RESET_COOL_DOWN);
2486 if (time_before(now, deadline))
2487 schedule_timeout_uninterruptible(deadline - now);
2490 spin_lock_irqsave(ap->lock, flags);
2491 ap->pflags |= ATA_PFLAG_RESETTING;
2492 spin_unlock_irqrestore(ap->lock, flags);
2494 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2496 ata_for_each_dev(dev, link, ALL) {
2497 /* If we issue an SRST then an ATA drive (not ATAPI)
2498 * may change configuration and be in PIO0 timing. If
2499 * we do a hard reset (or are coming from power on)
2500 * this is true for ATA or ATAPI. Until we've set a
2501 * suitable controller mode we should not touch the
2502 * bus as we may be talking too fast.
2504 dev->pio_mode = XFER_PIO_0;
2505 dev->dma_mode = 0xff;
2507 /* If the controller has a pio mode setup function
2508 * then use it to set the chipset to rights. Don't
2509 * touch the DMA setup as that will be dealt with when
2510 * configuring devices.
2512 if (ap->ops->set_piomode)
2513 ap->ops->set_piomode(ap, dev);
2516 /* prefer hardreset */
2518 ehc->i.action &= ~ATA_EH_RESET;
2521 ehc->i.action |= ATA_EH_HARDRESET;
2522 } else if (softreset) {
2524 ehc->i.action |= ATA_EH_SOFTRESET;
2528 unsigned long deadline = ata_deadline(jiffies,
2529 ATA_EH_PRERESET_TIMEOUT);
2532 sehc->i.action &= ~ATA_EH_RESET;
2533 sehc->i.action |= ehc->i.action;
2536 rc = prereset(link, deadline);
2538 /* If present, do prereset on slave link too. Reset
2539 * is skipped iff both master and slave links report
2540 * -ENOENT or clear ATA_EH_RESET.
2542 if (slave && (rc == 0 || rc == -ENOENT)) {
2545 tmp = prereset(slave, deadline);
2549 ehc->i.action |= sehc->i.action;
2553 if (rc == -ENOENT) {
2554 ata_link_dbg(link, "port disabled--ignoring\n");
2555 ehc->i.action &= ~ATA_EH_RESET;
2557 ata_for_each_dev(dev, link, ALL)
2558 classes[dev->devno] = ATA_DEV_NONE;
2563 "prereset failed (errno=%d)\n",
2568 /* prereset() might have cleared ATA_EH_RESET. If so,
2569 * bang classes, thaw and return.
2571 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2572 ata_for_each_dev(dev, link, ALL)
2573 classes[dev->devno] = ATA_DEV_NONE;
2574 if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2575 ata_eh_thaw_port(ap);
2585 if (ata_is_host_link(link))
2586 ata_eh_freeze_port(ap);
2588 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2592 ata_link_info(link, "%s resetting link\n",
2593 reset == softreset ? "soft" : "hard");
2595 /* mark that this EH session started with reset */
2596 ehc->last_reset = jiffies;
2597 if (reset == hardreset) {
2598 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2599 trace_ata_link_hardreset_begin(link, classes, deadline);
2601 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2602 trace_ata_link_softreset_begin(link, classes, deadline);
2605 rc = ata_do_reset(link, reset, classes, deadline, true);
2606 if (reset == hardreset)
2607 trace_ata_link_hardreset_end(link, classes, rc);
2609 trace_ata_link_softreset_end(link, classes, rc);
2610 if (rc && rc != -EAGAIN) {
2615 /* hardreset slave link if existent */
2616 if (slave && reset == hardreset) {
2620 ata_link_info(slave, "hard resetting link\n");
2622 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2623 trace_ata_slave_hardreset_begin(slave, classes,
2625 tmp = ata_do_reset(slave, reset, classes, deadline,
2627 trace_ata_slave_hardreset_end(slave, classes, tmp);
2635 failed_link = slave;
2641 /* perform follow-up SRST if necessary */
2642 if (reset == hardreset &&
2643 ata_eh_followup_srst_needed(link, rc)) {
2648 "follow-up softreset required but no softreset available\n");
2654 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2655 trace_ata_link_softreset_begin(link, classes, deadline);
2656 rc = ata_do_reset(link, reset, classes, deadline, true);
2657 trace_ata_link_softreset_end(link, classes, rc);
2666 "no reset method available, skipping reset\n");
2667 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2668 lflags |= ATA_LFLAG_ASSUME_ATA;
2672 * Post-reset processing
2674 ata_for_each_dev(dev, link, ALL) {
2675 /* After the reset, the device state is PIO 0 and the
2676 * controller state is undefined. Reset also wakes up
2677 * drives from sleeping mode.
2679 dev->pio_mode = XFER_PIO_0;
2680 dev->flags &= ~ATA_DFLAG_SLEEPING;
2682 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2685 /* apply class override */
2686 if (lflags & ATA_LFLAG_ASSUME_ATA)
2687 classes[dev->devno] = ATA_DEV_ATA;
2688 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2689 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2692 /* record current link speed */
2693 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2694 link->sata_spd = (sstatus >> 4) & 0xf;
2695 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2696 slave->sata_spd = (sstatus >> 4) & 0xf;
2699 if (ata_is_host_link(link))
2700 ata_eh_thaw_port(ap);
2702 /* postreset() should clear hardware SError. Although SError
2703 * is cleared during link resume, clearing SError here is
2704 * necessary as some PHYs raise hotplug events after SRST.
2705 * This introduces race condition where hotplug occurs between
2706 * reset and here. This race is mediated by cross checking
2707 * link onlineness and classification result later.
2710 postreset(link, classes);
2711 trace_ata_link_postreset(link, classes, rc);
2713 postreset(slave, classes);
2714 trace_ata_slave_postreset(slave, classes, rc);
2719 * Some controllers can't be frozen very well and may set spurious
2720 * error conditions during reset. Clear accumulated error
2721 * information and re-thaw the port if frozen. As reset is the
2722 * final recovery action and we cross check link onlineness against
2723 * device classification later, no hotplug event is lost by this.
2725 spin_lock_irqsave(link->ap->lock, flags);
2726 memset(&link->eh_info, 0, sizeof(link->eh_info));
2728 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2729 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2730 spin_unlock_irqrestore(link->ap->lock, flags);
2732 if (ata_port_is_frozen(ap))
2733 ata_eh_thaw_port(ap);
2736 * Make sure onlineness and classification result correspond.
2737 * Hotplug could have happened during reset and some
2738 * controllers fail to wait while a drive is spinning up after
2739 * being hotplugged causing misdetection. By cross checking
2740 * link on/offlineness and classification result, those
2741 * conditions can be reliably detected and retried.
2744 ata_for_each_dev(dev, link, ALL) {
2745 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2746 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2747 ata_dev_dbg(dev, "link online but device misclassified\n");
2748 classes[dev->devno] = ATA_DEV_NONE;
2751 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2752 if (ata_class_enabled(classes[dev->devno]))
2754 "link offline, clearing class %d to NONE\n",
2755 classes[dev->devno]);
2756 classes[dev->devno] = ATA_DEV_NONE;
2757 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2759 "link status unknown, clearing UNKNOWN to NONE\n");
2760 classes[dev->devno] = ATA_DEV_NONE;
2764 if (classify && nr_unknown) {
2765 if (try < max_tries) {
2767 "link online but %d devices misclassified, retrying\n",
2774 "link online but %d devices misclassified, "
2775 "device detection might fail\n", nr_unknown);
2778 /* reset successful, schedule revalidation */
2779 ata_eh_done(link, NULL, ATA_EH_RESET);
2781 ata_eh_done(slave, NULL, ATA_EH_RESET);
2782 ehc->last_reset = jiffies; /* update to completion time */
2783 ehc->i.action |= ATA_EH_REVALIDATE;
2784 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2788 /* clear hotplug flag */
2789 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2791 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2793 spin_lock_irqsave(ap->lock, flags);
2794 ap->pflags &= ~ATA_PFLAG_RESETTING;
2795 spin_unlock_irqrestore(ap->lock, flags);
2800 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2801 if (!ata_is_host_link(link) &&
2802 sata_scr_read(link, SCR_STATUS, &sstatus))
2805 if (try >= max_tries) {
2807 * Thaw host port even if reset failed, so that the port
2808 * can be retried on the next phy event. This risks
2809 * repeated EH runs but seems to be a better tradeoff than
2810 * shutting down a port after a botched hotplug attempt.
2812 if (ata_is_host_link(link))
2813 ata_eh_thaw_port(ap);
2818 if (time_before(now, deadline)) {
2819 unsigned long delta = deadline - now;
2821 ata_link_warn(failed_link,
2822 "reset failed (errno=%d), retrying in %u secs\n",
2823 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2827 delta = schedule_timeout_uninterruptible(delta);
2832 * While disks spinup behind PMP, some controllers fail sending SRST.
2833 * They need to be reset - as well as the PMP - before retrying.
2835 if (rc == -ERESTART) {
2836 if (ata_is_host_link(link))
2837 ata_eh_thaw_port(ap);
2841 if (try == max_tries - 1) {
2842 sata_down_spd_limit(link, 0);
2844 sata_down_spd_limit(slave, 0);
2845 } else if (rc == -EPIPE)
2846 sata_down_spd_limit(failed_link, 0);
2853 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2855 struct ata_link *link;
2856 struct ata_device *dev;
2857 unsigned long flags;
2860 * This function can be thought of as an extended version of
2861 * ata_eh_about_to_do() specially crafted to accommodate the
2862 * requirements of ATA_EH_PARK handling. Since the EH thread
2863 * does not leave the do {} while () loop in ata_eh_recover as
2864 * long as the timeout for a park request to *one* device on
2865 * the port has not expired, and since we still want to pick
2866 * up park requests to other devices on the same port or
2867 * timeout updates for the same device, we have to pull
2868 * ATA_EH_PARK actions from eh_info into eh_context.i
2869 * ourselves at the beginning of each pass over the loop.
2871 * Additionally, all write accesses to &ap->park_req_pending
2872 * through reinit_completion() (see below) or complete_all()
2873 * (see ata_scsi_park_store()) are protected by the host lock.
2874 * As a result we have that park_req_pending.done is zero on
2875 * exit from this function, i.e. when ATA_EH_PARK actions for
2876 * *all* devices on port ap have been pulled into the
2877 * respective eh_context structs. If, and only if,
2878 * park_req_pending.done is non-zero by the time we reach
2879 * wait_for_completion_timeout(), another ATA_EH_PARK action
2880 * has been scheduled for at least one of the devices on port
2881 * ap and we have to cycle over the do {} while () loop in
2882 * ata_eh_recover() again.
2885 spin_lock_irqsave(ap->lock, flags);
2886 reinit_completion(&ap->park_req_pending);
2887 ata_for_each_link(link, ap, EDGE) {
2888 ata_for_each_dev(dev, link, ALL) {
2889 struct ata_eh_info *ehi = &link->eh_info;
2891 link->eh_context.i.dev_action[dev->devno] |=
2892 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2893 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2896 spin_unlock_irqrestore(ap->lock, flags);
2899 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2901 struct ata_eh_context *ehc = &dev->link->eh_context;
2902 struct ata_taskfile tf;
2903 unsigned int err_mask;
2905 ata_tf_init(dev, &tf);
2907 ehc->unloaded_mask |= 1 << dev->devno;
2908 tf.command = ATA_CMD_IDLEIMMEDIATE;
2914 ehc->unloaded_mask &= ~(1 << dev->devno);
2915 tf.command = ATA_CMD_CHK_POWER;
2918 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2919 tf.protocol = ATA_PROT_NODATA;
2920 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2921 if (park && (err_mask || tf.lbal != 0xc4)) {
2922 ata_dev_err(dev, "head unload failed!\n");
2923 ehc->unloaded_mask &= ~(1 << dev->devno);
2927 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2928 struct ata_device **r_failed_dev)
2930 struct ata_port *ap = link->ap;
2931 struct ata_eh_context *ehc = &link->eh_context;
2932 struct ata_device *dev;
2933 unsigned int new_mask = 0;
2934 unsigned long flags;
2937 /* For PATA drive side cable detection to work, IDENTIFY must
2938 * be done backwards such that PDIAG- is released by the slave
2939 * device before the master device is identified.
2941 ata_for_each_dev(dev, link, ALL_REVERSE) {
2942 unsigned int action = ata_eh_dev_action(dev);
2943 unsigned int readid_flags = 0;
2945 if (ehc->i.flags & ATA_EHI_DID_RESET)
2946 readid_flags |= ATA_READID_POSTRESET;
2948 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2949 WARN_ON(dev->class == ATA_DEV_PMP);
2952 * The link may be in a deep sleep, wake it up.
2954 * If the link is in deep sleep, ata_phys_link_offline()
2955 * will return true, causing the revalidation to fail,
2956 * which leads to a (potentially) needless hard reset.
2958 * ata_eh_recover() will later restore the link policy
2959 * to ap->target_lpm_policy after revalidation is done.
2961 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
2962 rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER,
2968 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2973 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2974 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2979 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2981 /* Configuration may have changed, reconfigure
2984 ehc->i.flags |= ATA_EHI_SETMODE;
2986 /* schedule the scsi_rescan_device() here */
2987 schedule_work(&(ap->scsi_rescan_task));
2988 } else if (dev->class == ATA_DEV_UNKNOWN &&
2989 ehc->tries[dev->devno] &&
2990 ata_class_enabled(ehc->classes[dev->devno])) {
2991 /* Temporarily set dev->class, it will be
2992 * permanently set once all configurations are
2993 * complete. This is necessary because new
2994 * device configuration is done in two
2997 dev->class = ehc->classes[dev->devno];
2999 if (dev->class == ATA_DEV_PMP)
3000 rc = sata_pmp_attach(dev);
3002 rc = ata_dev_read_id(dev, &dev->class,
3003 readid_flags, dev->id);
3005 /* read_id might have changed class, store and reset */
3006 ehc->classes[dev->devno] = dev->class;
3007 dev->class = ATA_DEV_UNKNOWN;
3011 /* clear error info accumulated during probe */
3012 ata_ering_clear(&dev->ering);
3013 new_mask |= 1 << dev->devno;
3016 /* IDENTIFY was issued to non-existent
3017 * device. No need to reset. Just
3018 * thaw and ignore the device.
3020 ata_eh_thaw_port(ap);
3028 /* PDIAG- should have been released, ask cable type if post-reset */
3029 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3030 if (ap->ops->cable_detect)
3031 ap->cbl = ap->ops->cable_detect(ap);
3035 /* Configure new devices forward such that user doesn't see
3036 * device detection messages backwards.
3038 ata_for_each_dev(dev, link, ALL) {
3039 if (!(new_mask & (1 << dev->devno)))
3042 dev->class = ehc->classes[dev->devno];
3044 if (dev->class == ATA_DEV_PMP)
3047 ehc->i.flags |= ATA_EHI_PRINTINFO;
3048 rc = ata_dev_configure(dev);
3049 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3051 dev->class = ATA_DEV_UNKNOWN;
3055 spin_lock_irqsave(ap->lock, flags);
3056 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3057 spin_unlock_irqrestore(ap->lock, flags);
3059 /* new device discovered, configure xfermode */
3060 ehc->i.flags |= ATA_EHI_SETMODE;
3066 *r_failed_dev = dev;
3071 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3072 * @link: link on which timings will be programmed
3073 * @r_failed_dev: out parameter for failed device
3075 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3076 * ata_set_mode() fails, pointer to the failing device is
3077 * returned in @r_failed_dev.
3080 * PCI/etc. bus probe sem.
3083 * 0 on success, negative errno otherwise
3085 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3087 struct ata_port *ap = link->ap;
3088 struct ata_device *dev;
3091 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3092 ata_for_each_dev(dev, link, ENABLED) {
3093 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3094 struct ata_ering_entry *ent;
3096 ent = ata_ering_top(&dev->ering);
3098 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3102 /* has private set_mode? */
3103 if (ap->ops->set_mode)
3104 rc = ap->ops->set_mode(link, r_failed_dev);
3106 rc = ata_do_set_mode(link, r_failed_dev);
3108 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3109 ata_for_each_dev(dev, link, ENABLED) {
3110 struct ata_eh_context *ehc = &link->eh_context;
3111 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3112 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3114 if (dev->xfer_mode != saved_xfer_mode ||
3115 ata_ncq_enabled(dev) != saved_ncq)
3116 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3123 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3124 * @dev: ATAPI device to clear UA for
3126 * Resets and other operations can make an ATAPI device raise
3127 * UNIT ATTENTION which causes the next operation to fail. This
3128 * function clears UA.
3131 * EH context (may sleep).
3134 * 0 on success, -errno on failure.
3136 static int atapi_eh_clear_ua(struct ata_device *dev)
3140 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3141 u8 *sense_buffer = dev->link->ap->sector_buf;
3143 unsigned int err_mask;
3145 err_mask = atapi_eh_tur(dev, &sense_key);
3146 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3148 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3153 if (!err_mask || sense_key != UNIT_ATTENTION)
3156 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3158 ata_dev_warn(dev, "failed to clear "
3159 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3164 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3171 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3172 * @dev: ATA device which may need FLUSH retry
3174 * If @dev failed FLUSH, it needs to be reported upper layer
3175 * immediately as it means that @dev failed to remap and already
3176 * lost at least a sector and further FLUSH retrials won't make
3177 * any difference to the lost sector. However, if FLUSH failed
3178 * for other reasons, for example transmission error, FLUSH needs
3181 * This function determines whether FLUSH failure retry is
3182 * necessary and performs it if so.
3185 * 0 if EH can continue, -errno if EH needs to be repeated.
3187 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3189 struct ata_link *link = dev->link;
3190 struct ata_port *ap = link->ap;
3191 struct ata_queued_cmd *qc;
3192 struct ata_taskfile tf;
3193 unsigned int err_mask;
3196 /* did flush fail for this device? */
3197 if (!ata_tag_valid(link->active_tag))
3200 qc = __ata_qc_from_tag(ap, link->active_tag);
3201 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3202 qc->tf.command != ATA_CMD_FLUSH))
3205 /* if the device failed it, it should be reported to upper layers */
3206 if (qc->err_mask & AC_ERR_DEV)
3209 /* flush failed for some other reason, give it another shot */
3210 ata_tf_init(dev, &tf);
3212 tf.command = qc->tf.command;
3213 tf.flags |= ATA_TFLAG_DEVICE;
3214 tf.protocol = ATA_PROT_NODATA;
3216 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3217 tf.command, qc->err_mask);
3219 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3222 * FLUSH is complete but there's no way to
3223 * successfully complete a failed command from EH.
3224 * Making sure retry is allowed at least once and
3225 * retrying it should do the trick - whatever was in
3226 * the cache is already on the platter and this won't
3227 * cause infinite loop.
3229 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3231 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3235 /* if device failed it, report it to upper layers */
3236 if (err_mask & AC_ERR_DEV) {
3237 qc->err_mask |= AC_ERR_DEV;
3239 if (!ata_port_is_frozen(ap))
3247 * ata_eh_set_lpm - configure SATA interface power management
3248 * @link: link to configure power management
3249 * @policy: the link power management policy
3250 * @r_failed_dev: out parameter for failed device
3252 * Enable SATA Interface power management. This will enable
3253 * Device Interface Power Management (DIPM) for min_power and
3254 * medium_power_with_dipm policies, and then call driver specific
3255 * callbacks for enabling Host Initiated Power management.
3261 * 0 on success, -errno on failure.
3263 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3264 struct ata_device **r_failed_dev)
3266 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3267 struct ata_eh_context *ehc = &link->eh_context;
3268 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3269 enum ata_lpm_policy old_policy = link->lpm_policy;
3270 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3271 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3272 unsigned int err_mask;
3275 /* if the link or host doesn't do LPM, noop */
3276 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3277 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3281 * DIPM is enabled only for MIN_POWER as some devices
3282 * misbehave when the host NACKs transition to SLUMBER. Order
3283 * device and link configurations such that the host always
3284 * allows DIPM requests.
3286 ata_for_each_dev(dev, link, ENABLED) {
3287 bool hipm = ata_id_has_hipm(dev->id);
3288 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3290 /* find the first enabled and LPM enabled devices */
3294 if (!lpm_dev && (hipm || dipm))
3297 hints &= ~ATA_LPM_EMPTY;
3299 hints &= ~ATA_LPM_HIPM;
3301 /* disable DIPM before changing link config */
3302 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3303 err_mask = ata_dev_set_feature(dev,
3304 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3305 if (err_mask && err_mask != AC_ERR_DEV) {
3307 "failed to disable DIPM, Emask 0x%x\n",
3316 rc = ap->ops->set_lpm(link, policy, hints);
3317 if (!rc && ap->slave_link)
3318 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3320 rc = sata_pmp_set_lpm(link, policy, hints);
3323 * Attribute link config failure to the first (LPM) enabled
3324 * device on the link.
3327 if (rc == -EOPNOTSUPP) {
3328 link->flags |= ATA_LFLAG_NO_LPM;
3331 dev = lpm_dev ? lpm_dev : link_dev;
3336 * Low level driver acked the transition. Issue DIPM command
3337 * with the new policy set.
3339 link->lpm_policy = policy;
3340 if (ap && ap->slave_link)
3341 ap->slave_link->lpm_policy = policy;
3343 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3344 ata_for_each_dev(dev, link, ENABLED) {
3345 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3346 ata_id_has_dipm(dev->id)) {
3347 err_mask = ata_dev_set_feature(dev,
3348 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3349 if (err_mask && err_mask != AC_ERR_DEV) {
3351 "failed to enable DIPM, Emask 0x%x\n",
3359 link->last_lpm_change = jiffies;
3360 link->flags |= ATA_LFLAG_CHANGED;
3365 /* restore the old policy */
3366 link->lpm_policy = old_policy;
3367 if (ap && ap->slave_link)
3368 ap->slave_link->lpm_policy = old_policy;
3370 /* if no device or only one more chance is left, disable LPM */
3371 if (!dev || ehc->tries[dev->devno] <= 2) {
3372 ata_link_warn(link, "disabling LPM on the link\n");
3373 link->flags |= ATA_LFLAG_NO_LPM;
3376 *r_failed_dev = dev;
3380 int ata_link_nr_enabled(struct ata_link *link)
3382 struct ata_device *dev;
3385 ata_for_each_dev(dev, link, ENABLED)
3390 static int ata_link_nr_vacant(struct ata_link *link)
3392 struct ata_device *dev;
3395 ata_for_each_dev(dev, link, ALL)
3396 if (dev->class == ATA_DEV_UNKNOWN)
3401 static int ata_eh_skip_recovery(struct ata_link *link)
3403 struct ata_port *ap = link->ap;
3404 struct ata_eh_context *ehc = &link->eh_context;
3405 struct ata_device *dev;
3407 /* skip disabled links */
3408 if (link->flags & ATA_LFLAG_DISABLED)
3411 /* skip if explicitly requested */
3412 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3415 /* thaw frozen port and recover failed devices */
3416 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3419 /* reset at least once if reset is requested */
3420 if ((ehc->i.action & ATA_EH_RESET) &&
3421 !(ehc->i.flags & ATA_EHI_DID_RESET))
3424 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3425 ata_for_each_dev(dev, link, ALL) {
3426 if (dev->class == ATA_DEV_UNKNOWN &&
3427 ehc->classes[dev->devno] != ATA_DEV_NONE)
3434 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3436 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3437 u64 now = get_jiffies_64();
3438 int *trials = void_arg;
3440 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3441 (ent->timestamp < now - min(now, interval)))
3448 static int ata_eh_schedule_probe(struct ata_device *dev)
3450 struct ata_eh_context *ehc = &dev->link->eh_context;
3451 struct ata_link *link = ata_dev_phys_link(dev);
3454 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3455 (ehc->did_probe_mask & (1 << dev->devno)))
3458 ata_eh_detach_dev(dev);
3460 ehc->did_probe_mask |= (1 << dev->devno);
3461 ehc->i.action |= ATA_EH_RESET;
3462 ehc->saved_xfer_mode[dev->devno] = 0;
3463 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3465 /* the link maybe in a deep sleep, wake it up */
3466 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3467 if (ata_is_host_link(link))
3468 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3471 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3475 /* Record and count probe trials on the ering. The specific
3476 * error mask used is irrelevant. Because a successful device
3477 * detection clears the ering, this count accumulates only if
3478 * there are consecutive failed probes.
3480 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3481 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3482 * forced to 1.5Gbps.
3484 * This is to work around cases where failed link speed
3485 * negotiation results in device misdetection leading to
3486 * infinite DEVXCHG or PHRDY CHG events.
3488 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3489 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3491 if (trials > ATA_EH_PROBE_TRIALS)
3492 sata_down_spd_limit(link, 1);
3497 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3499 struct ata_eh_context *ehc = &dev->link->eh_context;
3501 /* -EAGAIN from EH routine indicates retry without prejudice.
3502 * The requester is responsible for ensuring forward progress.
3505 ehc->tries[dev->devno]--;
3509 /* device missing or wrong IDENTIFY data, schedule probing */
3510 ehc->i.probe_mask |= (1 << dev->devno);
3513 /* give it just one more chance */
3514 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3517 if (ehc->tries[dev->devno] == 1) {
3518 /* This is the last chance, better to slow
3519 * down than lose it.
3521 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3522 if (dev->pio_mode > XFER_PIO_0)
3523 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3527 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3528 /* disable device if it has used up all its chances */
3529 ata_dev_disable(dev);
3531 /* detach if offline */
3532 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3533 ata_eh_detach_dev(dev);
3535 /* schedule probe if necessary */
3536 if (ata_eh_schedule_probe(dev)) {
3537 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3538 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3539 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3544 ehc->i.action |= ATA_EH_RESET;
3550 * ata_eh_recover - recover host port after error
3551 * @ap: host port to recover
3552 * @prereset: prereset method (can be NULL)
3553 * @softreset: softreset method (can be NULL)
3554 * @hardreset: hardreset method (can be NULL)
3555 * @postreset: postreset method (can be NULL)
3556 * @r_failed_link: out parameter for failed link
3558 * This is the alpha and omega, eum and yang, heart and soul of
3559 * libata exception handling. On entry, actions required to
3560 * recover each link and hotplug requests are recorded in the
3561 * link's eh_context. This function executes all the operations
3562 * with appropriate retrials and fallbacks to resurrect failed
3563 * devices, detach goners and greet newcomers.
3566 * Kernel thread context (may sleep).
3569 * 0 on success, -errno on failure.
3571 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3572 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3573 ata_postreset_fn_t postreset,
3574 struct ata_link **r_failed_link)
3576 struct ata_link *link;
3577 struct ata_device *dev;
3579 unsigned long flags, deadline;
3581 /* prep for recovery */
3582 ata_for_each_link(link, ap, EDGE) {
3583 struct ata_eh_context *ehc = &link->eh_context;
3585 /* re-enable link? */
3586 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3587 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3588 spin_lock_irqsave(ap->lock, flags);
3589 link->flags &= ~ATA_LFLAG_DISABLED;
3590 spin_unlock_irqrestore(ap->lock, flags);
3591 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3594 ata_for_each_dev(dev, link, ALL) {
3595 if (link->flags & ATA_LFLAG_NO_RETRY)
3596 ehc->tries[dev->devno] = 1;
3598 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3600 /* collect port action mask recorded in dev actions */
3601 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3602 ~ATA_EH_PERDEV_MASK;
3603 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3605 /* process hotplug request */
3606 if (dev->flags & ATA_DFLAG_DETACH)
3607 ata_eh_detach_dev(dev);
3609 /* schedule probe if necessary */
3610 if (!ata_dev_enabled(dev))
3611 ata_eh_schedule_probe(dev);
3618 /* if UNLOADING, finish immediately */
3619 if (ap->pflags & ATA_PFLAG_UNLOADING)
3623 ata_for_each_link(link, ap, EDGE) {
3624 struct ata_eh_context *ehc = &link->eh_context;
3626 /* skip EH if possible. */
3627 if (ata_eh_skip_recovery(link))
3630 ata_for_each_dev(dev, link, ALL)
3631 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3635 ata_for_each_link(link, ap, EDGE) {
3636 struct ata_eh_context *ehc = &link->eh_context;
3638 if (!(ehc->i.action & ATA_EH_RESET))
3641 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3642 prereset, softreset, hardreset, postreset);
3644 ata_link_err(link, "reset failed, giving up\n");
3653 * clears ATA_EH_PARK in eh_info and resets
3654 * ap->park_req_pending
3656 ata_eh_pull_park_action(ap);
3659 ata_for_each_link(link, ap, EDGE) {
3660 ata_for_each_dev(dev, link, ALL) {
3661 struct ata_eh_context *ehc = &link->eh_context;
3664 if (dev->class != ATA_DEV_ATA &&
3665 dev->class != ATA_DEV_ZAC)
3667 if (!(ehc->i.dev_action[dev->devno] &
3670 tmp = dev->unpark_deadline;
3671 if (time_before(deadline, tmp))
3673 else if (time_before_eq(tmp, jiffies))
3675 if (ehc->unloaded_mask & (1 << dev->devno))
3678 ata_eh_park_issue_cmd(dev, 1);
3683 if (time_before_eq(deadline, now))
3687 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3691 ata_for_each_link(link, ap, EDGE) {
3692 ata_for_each_dev(dev, link, ALL) {
3693 if (!(link->eh_context.unloaded_mask &
3697 ata_eh_park_issue_cmd(dev, 0);
3698 ata_eh_done(link, dev, ATA_EH_PARK);
3704 ata_for_each_link(link, ap, PMP_FIRST) {
3705 struct ata_eh_context *ehc = &link->eh_context;
3707 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3710 /* revalidate existing devices and attach new ones */
3711 rc = ata_eh_revalidate_and_attach(link, &dev);
3715 /* if PMP got attached, return, pmp EH will take care of it */
3716 if (link->device->class == ATA_DEV_PMP) {
3721 /* configure transfer mode if necessary */
3722 if (ehc->i.flags & ATA_EHI_SETMODE) {
3723 rc = ata_set_mode(link, &dev);
3726 ehc->i.flags &= ~ATA_EHI_SETMODE;
3729 /* If reset has been issued, clear UA to avoid
3730 * disrupting the current users of the device.
3732 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3733 ata_for_each_dev(dev, link, ALL) {
3734 if (dev->class != ATA_DEV_ATAPI)
3736 rc = atapi_eh_clear_ua(dev);
3739 if (zpodd_dev_enabled(dev))
3740 zpodd_post_poweron(dev);
3744 /* retry flush if necessary */
3745 ata_for_each_dev(dev, link, ALL) {
3746 if (dev->class != ATA_DEV_ATA &&
3747 dev->class != ATA_DEV_ZAC)
3749 rc = ata_eh_maybe_retry_flush(dev);
3755 /* configure link power saving */
3756 if (link->lpm_policy != ap->target_lpm_policy) {
3757 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3762 /* this link is okay now */
3769 ata_eh_handle_dev_fail(dev, rc);
3771 if (ata_port_is_frozen(ap)) {
3772 /* PMP reset requires working host port.
3773 * Can't retry if it's frozen.
3775 if (sata_pmp_attached(ap))
3785 if (rc && r_failed_link)
3786 *r_failed_link = link;
3792 * ata_eh_finish - finish up EH
3793 * @ap: host port to finish EH for
3795 * Recovery is complete. Clean up EH states and retry or finish
3801 void ata_eh_finish(struct ata_port *ap)
3803 struct ata_queued_cmd *qc;
3806 /* retry or finish qcs */
3807 ata_qc_for_each_raw(ap, qc, tag) {
3808 if (!(qc->flags & ATA_QCFLAG_EH))
3812 /* FIXME: Once EH migration is complete,
3813 * generate sense data in this function,
3814 * considering both err_mask and tf.
3816 if (qc->flags & ATA_QCFLAG_RETRY)
3817 ata_eh_qc_retry(qc);
3819 ata_eh_qc_complete(qc);
3821 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3822 ata_eh_qc_complete(qc);
3824 /* feed zero TF to sense generation */
3825 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3826 ata_eh_qc_retry(qc);
3831 /* make sure nr_active_links is zero after EH */
3832 WARN_ON(ap->nr_active_links);
3833 ap->nr_active_links = 0;
3837 * ata_do_eh - do standard error handling
3838 * @ap: host port to handle error for
3840 * @prereset: prereset method (can be NULL)
3841 * @softreset: softreset method (can be NULL)
3842 * @hardreset: hardreset method (can be NULL)
3843 * @postreset: postreset method (can be NULL)
3845 * Perform standard error handling sequence.
3848 * Kernel thread context (may sleep).
3850 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3851 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3852 ata_postreset_fn_t postreset)
3854 struct ata_device *dev;
3860 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3863 ata_for_each_dev(dev, &ap->link, ALL)
3864 ata_dev_disable(dev);
3871 * ata_std_error_handler - standard error handler
3872 * @ap: host port to handle error for
3874 * Standard error handler
3877 * Kernel thread context (may sleep).
3879 void ata_std_error_handler(struct ata_port *ap)
3881 struct ata_port_operations *ops = ap->ops;
3882 ata_reset_fn_t hardreset = ops->hardreset;
3884 /* ignore built-in hardreset if SCR access is not available */
3885 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3888 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3890 EXPORT_SYMBOL_GPL(ata_std_error_handler);
3894 * ata_eh_handle_port_suspend - perform port suspend operation
3895 * @ap: port to suspend
3900 * Kernel thread context (may sleep).
3902 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3904 unsigned long flags;
3906 struct ata_device *dev;
3908 /* are we suspending? */
3909 spin_lock_irqsave(ap->lock, flags);
3910 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3911 ap->pm_mesg.event & PM_EVENT_RESUME) {
3912 spin_unlock_irqrestore(ap->lock, flags);
3915 spin_unlock_irqrestore(ap->lock, flags);
3917 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3920 * If we have a ZPODD attached, check its zero
3921 * power ready status before the port is frozen.
3922 * Only needed for runtime suspend.
3924 if (PMSG_IS_AUTO(ap->pm_mesg)) {
3925 ata_for_each_dev(dev, &ap->link, ENABLED) {
3926 if (zpodd_dev_enabled(dev))
3927 zpodd_on_suspend(dev);
3932 ata_eh_freeze_port(ap);
3934 if (ap->ops->port_suspend)
3935 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3937 ata_acpi_set_state(ap, ap->pm_mesg);
3939 /* update the flags */
3940 spin_lock_irqsave(ap->lock, flags);
3942 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3944 ap->pflags |= ATA_PFLAG_SUSPENDED;
3945 else if (ata_port_is_frozen(ap))
3946 ata_port_schedule_eh(ap);
3948 spin_unlock_irqrestore(ap->lock, flags);
3954 * ata_eh_handle_port_resume - perform port resume operation
3955 * @ap: port to resume
3960 * Kernel thread context (may sleep).
3962 static void ata_eh_handle_port_resume(struct ata_port *ap)
3964 struct ata_link *link;
3965 struct ata_device *dev;
3966 unsigned long flags;
3968 /* are we resuming? */
3969 spin_lock_irqsave(ap->lock, flags);
3970 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3971 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
3972 spin_unlock_irqrestore(ap->lock, flags);
3975 spin_unlock_irqrestore(ap->lock, flags);
3977 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3980 * Error timestamps are in jiffies which doesn't run while
3981 * suspended and PHY events during resume isn't too uncommon.
3982 * When the two are combined, it can lead to unnecessary speed
3983 * downs if the machine is suspended and resumed repeatedly.
3984 * Clear error history.
3986 ata_for_each_link(link, ap, HOST_FIRST)
3987 ata_for_each_dev(dev, link, ALL)
3988 ata_ering_clear(&dev->ering);
3990 ata_acpi_set_state(ap, ap->pm_mesg);
3992 if (ap->ops->port_resume)
3993 ap->ops->port_resume(ap);
3995 /* tell ACPI that we're resuming */
3996 ata_acpi_on_resume(ap);
3998 /* update the flags */
3999 spin_lock_irqsave(ap->lock, flags);
4000 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4001 spin_unlock_irqrestore(ap->lock, flags);
4003 #endif /* CONFIG_PM */