1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * libata-eh.c - libata error handling
5 * Copyright 2006 Tejun Heo <htejun@gmail.com>
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);
155 static void ata_eh_handle_port_suspend(struct ata_port *ap);
156 static void ata_eh_handle_port_resume(struct ata_port *ap);
157 #else /* CONFIG_PM */
158 static void ata_eh_handle_port_suspend(struct ata_port *ap)
161 static void ata_eh_handle_port_resume(struct ata_port *ap)
163 #endif /* CONFIG_PM */
165 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
166 const char *fmt, va_list args)
168 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
169 ATA_EH_DESC_LEN - ehi->desc_len,
174 * __ata_ehi_push_desc - push error description without adding separator
176 * @fmt: printf format string
178 * Format string according to @fmt and append it to @ehi->desc.
181 * spin_lock_irqsave(host lock)
183 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
188 __ata_ehi_pushv_desc(ehi, fmt, args);
191 EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
194 * ata_ehi_push_desc - push error description with separator
196 * @fmt: printf format string
198 * Format string according to @fmt and append it to @ehi->desc.
199 * If @ehi->desc is not empty, ", " is added in-between.
202 * spin_lock_irqsave(host lock)
204 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
209 __ata_ehi_push_desc(ehi, ", ");
212 __ata_ehi_pushv_desc(ehi, fmt, args);
215 EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
218 * ata_ehi_clear_desc - clean error description
224 * spin_lock_irqsave(host lock)
226 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
231 EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
234 * ata_port_desc - append port description
235 * @ap: target ATA port
236 * @fmt: printf format string
238 * Format string according to @fmt and append it to port
239 * description. If port description is not empty, " " is added
240 * in-between. This function is to be used while initializing
241 * ata_host. The description is printed on host registration.
246 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
250 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
252 if (ap->link.eh_info.desc_len)
253 __ata_ehi_push_desc(&ap->link.eh_info, " ");
256 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
259 EXPORT_SYMBOL_GPL(ata_port_desc);
263 * ata_port_pbar_desc - append PCI BAR description
264 * @ap: target ATA port
265 * @bar: target PCI BAR
266 * @offset: offset into PCI BAR
267 * @name: name of the area
269 * If @offset is negative, this function formats a string which
270 * contains the name, address, size and type of the BAR and
271 * appends it to the port description. If @offset is zero or
272 * positive, only name and offsetted address is appended.
277 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
280 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
282 unsigned long long start, len;
284 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
286 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
289 start = (unsigned long long)pci_resource_start(pdev, bar);
290 len = (unsigned long long)pci_resource_len(pdev, bar);
293 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
295 ata_port_desc(ap, "%s 0x%llx", name,
296 start + (unsigned long long)offset);
298 EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
299 #endif /* CONFIG_PCI */
301 static int ata_lookup_timeout_table(u8 cmd)
305 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
308 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
317 * ata_internal_cmd_timeout - determine timeout for an internal command
318 * @dev: target device
319 * @cmd: internal command to be issued
321 * Determine timeout for internal command @cmd for @dev.
327 * Determined timeout.
329 unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
331 struct ata_eh_context *ehc = &dev->link->eh_context;
332 int ent = ata_lookup_timeout_table(cmd);
336 return ATA_EH_CMD_DFL_TIMEOUT;
338 idx = ehc->cmd_timeout_idx[dev->devno][ent];
339 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
343 * ata_internal_cmd_timed_out - notification for internal command timeout
344 * @dev: target device
345 * @cmd: internal command which timed out
347 * Notify EH that internal command @cmd for @dev timed out. This
348 * function should be called only for commands whose timeouts are
349 * determined using ata_internal_cmd_timeout().
354 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
356 struct ata_eh_context *ehc = &dev->link->eh_context;
357 int ent = ata_lookup_timeout_table(cmd);
363 idx = ehc->cmd_timeout_idx[dev->devno][ent];
364 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
365 ehc->cmd_timeout_idx[dev->devno][ent]++;
368 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
369 unsigned int err_mask)
371 struct ata_ering_entry *ent;
376 ering->cursor %= ATA_ERING_SIZE;
378 ent = &ering->ring[ering->cursor];
379 ent->eflags = eflags;
380 ent->err_mask = err_mask;
381 ent->timestamp = get_jiffies_64();
384 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
386 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
393 int ata_ering_map(struct ata_ering *ering,
394 int (*map_fn)(struct ata_ering_entry *, void *),
398 struct ata_ering_entry *ent;
402 ent = &ering->ring[idx];
405 rc = map_fn(ent, arg);
408 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
409 } while (idx != ering->cursor);
414 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
416 ent->eflags |= ATA_EFLAG_OLD_ER;
420 static void ata_ering_clear(struct ata_ering *ering)
422 ata_ering_map(ering, ata_ering_clear_cb, NULL);
425 static unsigned int ata_eh_dev_action(struct ata_device *dev)
427 struct ata_eh_context *ehc = &dev->link->eh_context;
429 return ehc->i.action | ehc->i.dev_action[dev->devno];
432 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
433 struct ata_eh_info *ehi, unsigned int action)
435 struct ata_device *tdev;
438 ehi->action &= ~action;
439 ata_for_each_dev(tdev, link, ALL)
440 ehi->dev_action[tdev->devno] &= ~action;
442 /* doesn't make sense for port-wide EH actions */
443 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
445 /* break ehi->action into ehi->dev_action */
446 if (ehi->action & action) {
447 ata_for_each_dev(tdev, link, ALL)
448 ehi->dev_action[tdev->devno] |=
449 ehi->action & action;
450 ehi->action &= ~action;
453 /* turn off the specified per-dev action */
454 ehi->dev_action[dev->devno] &= ~action;
459 * ata_eh_acquire - acquire EH ownership
460 * @ap: ATA port to acquire EH ownership for
462 * Acquire EH ownership for @ap. This is the basic exclusion
463 * mechanism for ports sharing a host. Only one port hanging off
464 * the same host can claim the ownership of EH.
469 void ata_eh_acquire(struct ata_port *ap)
471 mutex_lock(&ap->host->eh_mutex);
472 WARN_ON_ONCE(ap->host->eh_owner);
473 ap->host->eh_owner = current;
477 * ata_eh_release - release EH ownership
478 * @ap: ATA port to release EH ownership for
480 * Release EH ownership for @ap if the caller. The caller must
481 * have acquired EH ownership using ata_eh_acquire() previously.
486 void ata_eh_release(struct ata_port *ap)
488 WARN_ON_ONCE(ap->host->eh_owner != current);
489 ap->host->eh_owner = NULL;
490 mutex_unlock(&ap->host->eh_mutex);
493 static void ata_eh_unload(struct ata_port *ap)
495 struct ata_link *link;
496 struct ata_device *dev;
499 /* Restore SControl IPM and SPD for the next driver and
500 * disable attached devices.
502 ata_for_each_link(link, ap, PMP_FIRST) {
503 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
504 ata_for_each_dev(dev, link, ALL)
505 ata_dev_disable(dev);
508 /* freeze and set UNLOADED */
509 spin_lock_irqsave(ap->lock, flags);
511 ata_port_freeze(ap); /* won't be thawed */
512 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
513 ap->pflags |= ATA_PFLAG_UNLOADED;
515 spin_unlock_irqrestore(ap->lock, flags);
519 * ata_scsi_error - SCSI layer error handler callback
520 * @host: SCSI host on which error occurred
522 * Handles SCSI-layer-thrown error events.
525 * Inherited from SCSI layer (none, can sleep)
530 void ata_scsi_error(struct Scsi_Host *host)
532 struct ata_port *ap = ata_shost_to_port(host);
534 LIST_HEAD(eh_work_q);
536 spin_lock_irqsave(host->host_lock, flags);
537 list_splice_init(&host->eh_cmd_q, &eh_work_q);
538 spin_unlock_irqrestore(host->host_lock, flags);
540 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
542 /* If we timed raced normal completion and there is nothing to
543 recover nr_timedout == 0 why exactly are we doing error recovery ? */
544 ata_scsi_port_error_handler(host, ap);
546 /* finish or retry handled scmd's and clean up */
547 WARN_ON(!list_empty(&eh_work_q));
552 * ata_scsi_cmd_error_handler - error callback for a list of commands
553 * @host: scsi host containing the port
554 * @ap: ATA port within the host
555 * @eh_work_q: list of commands to process
557 * process the given list of commands and return those finished to the
558 * ap->eh_done_q. This function is the first part of the libata error
559 * handler which processes a given list of failed commands.
561 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
562 struct list_head *eh_work_q)
567 /* make sure sff pio task is not running */
568 ata_sff_flush_pio_task(ap);
570 /* synchronize with host lock and sort out timeouts */
572 /* For new EH, all qcs are finished in one of three ways -
573 * normal completion, error completion, and SCSI timeout.
574 * Both completions can race against SCSI timeout. When normal
575 * completion wins, the qc never reaches EH. When error
576 * completion wins, the qc has ATA_QCFLAG_FAILED set.
578 * When SCSI timeout wins, things are a bit more complex.
579 * Normal or error completion can occur after the timeout but
580 * before this point. In such cases, both types of
581 * completions are honored. A scmd is determined to have
582 * timed out iff its associated qc is active and not failed.
584 spin_lock_irqsave(ap->lock, flags);
585 if (ap->ops->error_handler) {
586 struct scsi_cmnd *scmd, *tmp;
589 /* This must occur under the ap->lock as we don't want
590 a polled recovery to race the real interrupt handler
592 The lost_interrupt handler checks for any completed but
593 non-notified command and completes much like an IRQ handler.
595 We then fall into the error recovery code which will treat
596 this as if normal completion won the race */
598 if (ap->ops->lost_interrupt)
599 ap->ops->lost_interrupt(ap);
601 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
602 struct ata_queued_cmd *qc;
604 ata_qc_for_each_raw(ap, qc, i) {
605 if (qc->flags & ATA_QCFLAG_ACTIVE &&
610 if (i < ATA_MAX_QUEUE) {
611 /* the scmd has an associated qc */
612 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
613 /* which hasn't failed yet, timeout */
614 qc->err_mask |= AC_ERR_TIMEOUT;
615 qc->flags |= ATA_QCFLAG_FAILED;
619 /* Normal completion occurred after
620 * SCSI timeout but before this point.
621 * Successfully complete it.
623 scmd->retries = scmd->allowed;
624 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
628 /* If we have timed out qcs. They belong to EH from
629 * this point but the state of the controller is
630 * unknown. Freeze the port to make sure the IRQ
631 * handler doesn't diddle with those qcs. This must
632 * be done atomically w.r.t. setting QCFLAG_FAILED.
635 __ata_port_freeze(ap);
638 /* initialize eh_tries */
639 ap->eh_tries = ATA_EH_MAX_TRIES;
641 spin_unlock_irqrestore(ap->lock, flags);
644 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
647 * ata_scsi_port_error_handler - recover the port after the commands
648 * @host: SCSI host containing the port
651 * Handle the recovery of the port @ap after all the commands
652 * have been recovered.
654 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
658 /* invoke error handler */
659 if (ap->ops->error_handler) {
660 struct ata_link *link;
662 /* acquire EH ownership */
665 /* kill fast drain timer */
666 del_timer_sync(&ap->fastdrain_timer);
668 /* process port resume request */
669 ata_eh_handle_port_resume(ap);
671 /* fetch & clear EH info */
672 spin_lock_irqsave(ap->lock, flags);
674 ata_for_each_link(link, ap, HOST_FIRST) {
675 struct ata_eh_context *ehc = &link->eh_context;
676 struct ata_device *dev;
678 memset(&link->eh_context, 0, sizeof(link->eh_context));
679 link->eh_context.i = link->eh_info;
680 memset(&link->eh_info, 0, sizeof(link->eh_info));
682 ata_for_each_dev(dev, link, ENABLED) {
683 int devno = dev->devno;
685 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
686 if (ata_ncq_enabled(dev))
687 ehc->saved_ncq_enabled |= 1 << devno;
691 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
692 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
693 ap->excl_link = NULL; /* don't maintain exclusion over EH */
695 spin_unlock_irqrestore(ap->lock, flags);
697 /* invoke EH, skip if unloading or suspended */
698 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
699 ap->ops->error_handler(ap);
701 /* if unloading, commence suicide */
702 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
703 !(ap->pflags & ATA_PFLAG_UNLOADED))
708 /* process port suspend request */
709 ata_eh_handle_port_suspend(ap);
711 /* Exception might have happened after ->error_handler
712 * recovered the port but before this point. Repeat
715 spin_lock_irqsave(ap->lock, flags);
717 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
718 if (--ap->eh_tries) {
719 spin_unlock_irqrestore(ap->lock, flags);
723 "EH pending after %d tries, giving up\n",
725 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
728 /* this run is complete, make sure EH info is clear */
729 ata_for_each_link(link, ap, HOST_FIRST)
730 memset(&link->eh_info, 0, sizeof(link->eh_info));
732 /* end eh (clear host_eh_scheduled) while holding
733 * ap->lock such that if exception occurs after this
734 * point but before EH completion, SCSI midlayer will
739 spin_unlock_irqrestore(ap->lock, flags);
742 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
743 ap->ops->eng_timeout(ap);
746 scsi_eh_flush_done_q(&ap->eh_done_q);
749 spin_lock_irqsave(ap->lock, flags);
751 if (ap->pflags & ATA_PFLAG_LOADING)
752 ap->pflags &= ~ATA_PFLAG_LOADING;
753 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
754 !(ap->flags & ATA_FLAG_SAS_HOST))
755 schedule_delayed_work(&ap->hotplug_task, 0);
757 if (ap->pflags & ATA_PFLAG_RECOVERED)
758 ata_port_info(ap, "EH complete\n");
760 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
762 /* tell wait_eh that we're done */
763 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
764 wake_up_all(&ap->eh_wait_q);
766 spin_unlock_irqrestore(ap->lock, flags);
768 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
771 * ata_port_wait_eh - Wait for the currently pending EH to complete
772 * @ap: Port to wait EH for
774 * Wait until the currently pending EH is complete.
777 * Kernel thread context (may sleep).
779 void ata_port_wait_eh(struct ata_port *ap)
785 spin_lock_irqsave(ap->lock, flags);
787 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
788 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
789 spin_unlock_irqrestore(ap->lock, flags);
791 spin_lock_irqsave(ap->lock, flags);
793 finish_wait(&ap->eh_wait_q, &wait);
795 spin_unlock_irqrestore(ap->lock, flags);
797 /* make sure SCSI EH is complete */
798 if (scsi_host_in_recovery(ap->scsi_host)) {
803 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
805 static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
807 struct ata_queued_cmd *qc;
811 /* count only non-internal commands */
812 ata_qc_for_each(ap, qc, tag) {
820 void ata_eh_fastdrain_timerfn(struct timer_list *t)
822 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
826 spin_lock_irqsave(ap->lock, flags);
828 cnt = ata_eh_nr_in_flight(ap);
834 if (cnt == ap->fastdrain_cnt) {
835 struct ata_queued_cmd *qc;
838 /* No progress during the last interval, tag all
839 * in-flight qcs as timed out and freeze the port.
841 ata_qc_for_each(ap, qc, tag) {
843 qc->err_mask |= AC_ERR_TIMEOUT;
848 /* some qcs have finished, give it another chance */
849 ap->fastdrain_cnt = cnt;
850 ap->fastdrain_timer.expires =
851 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
852 add_timer(&ap->fastdrain_timer);
856 spin_unlock_irqrestore(ap->lock, flags);
860 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
861 * @ap: target ATA port
862 * @fastdrain: activate fast drain
864 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
865 * is non-zero and EH wasn't pending before. Fast drain ensures
866 * that EH kicks in in timely manner.
869 * spin_lock_irqsave(host lock)
871 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
875 /* already scheduled? */
876 if (ap->pflags & ATA_PFLAG_EH_PENDING)
879 ap->pflags |= ATA_PFLAG_EH_PENDING;
884 /* do we have in-flight qcs? */
885 cnt = ata_eh_nr_in_flight(ap);
889 /* activate fast drain */
890 ap->fastdrain_cnt = cnt;
891 ap->fastdrain_timer.expires =
892 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
893 add_timer(&ap->fastdrain_timer);
897 * ata_qc_schedule_eh - schedule qc for error handling
898 * @qc: command to schedule error handling for
900 * Schedule error handling for @qc. EH will kick in as soon as
901 * other commands are drained.
904 * spin_lock_irqsave(host lock)
906 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
908 struct ata_port *ap = qc->ap;
910 WARN_ON(!ap->ops->error_handler);
912 qc->flags |= ATA_QCFLAG_FAILED;
913 ata_eh_set_pending(ap, 1);
915 /* The following will fail if timeout has already expired.
916 * ata_scsi_error() takes care of such scmds on EH entry.
917 * Note that ATA_QCFLAG_FAILED is unconditionally set after
918 * this function completes.
920 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
924 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
925 * @ap: ATA port to schedule EH for
927 * LOCKING: inherited from ata_port_schedule_eh
928 * spin_lock_irqsave(host lock)
930 void ata_std_sched_eh(struct ata_port *ap)
932 WARN_ON(!ap->ops->error_handler);
934 if (ap->pflags & ATA_PFLAG_INITIALIZING)
937 ata_eh_set_pending(ap, 1);
938 scsi_schedule_eh(ap->scsi_host);
940 trace_ata_std_sched_eh(ap);
942 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
945 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
946 * @ap: ATA port to end EH for
948 * In the libata object model there is a 1:1 mapping of ata_port to
949 * shost, so host fields can be directly manipulated under ap->lock, in
950 * the libsas case we need to hold a lock at the ha->level to coordinate
954 * spin_lock_irqsave(host lock)
956 void ata_std_end_eh(struct ata_port *ap)
958 struct Scsi_Host *host = ap->scsi_host;
960 host->host_eh_scheduled = 0;
962 EXPORT_SYMBOL(ata_std_end_eh);
966 * ata_port_schedule_eh - schedule error handling without a qc
967 * @ap: ATA port to schedule EH for
969 * Schedule error handling for @ap. EH will kick in as soon as
970 * all commands are drained.
973 * spin_lock_irqsave(host lock)
975 void ata_port_schedule_eh(struct ata_port *ap)
977 /* see: ata_std_sched_eh, unless you know better */
978 ap->ops->sched_eh(ap);
980 EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
982 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
984 struct ata_queued_cmd *qc;
985 int tag, nr_aborted = 0;
987 WARN_ON(!ap->ops->error_handler);
989 /* we're gonna abort all commands, no need for fast drain */
990 ata_eh_set_pending(ap, 0);
992 /* include internal tag in iteration */
993 ata_qc_for_each_with_internal(ap, qc, tag) {
994 if (qc && (!link || qc->dev->link == link)) {
995 qc->flags |= ATA_QCFLAG_FAILED;
1002 ata_port_schedule_eh(ap);
1008 * ata_link_abort - abort all qc's on the link
1009 * @link: ATA link to abort qc's for
1011 * Abort all active qc's active on @link and schedule EH.
1014 * spin_lock_irqsave(host lock)
1017 * Number of aborted qc's.
1019 int ata_link_abort(struct ata_link *link)
1021 return ata_do_link_abort(link->ap, link);
1023 EXPORT_SYMBOL_GPL(ata_link_abort);
1026 * ata_port_abort - abort all qc's on the port
1027 * @ap: ATA port to abort qc's for
1029 * Abort all active qc's of @ap and schedule EH.
1032 * spin_lock_irqsave(host_set lock)
1035 * Number of aborted qc's.
1037 int ata_port_abort(struct ata_port *ap)
1039 return ata_do_link_abort(ap, NULL);
1041 EXPORT_SYMBOL_GPL(ata_port_abort);
1044 * __ata_port_freeze - freeze port
1045 * @ap: ATA port to freeze
1047 * This function is called when HSM violation or some other
1048 * condition disrupts normal operation of the port. Frozen port
1049 * is not allowed to perform any operation until the port is
1050 * thawed, which usually follows a successful reset.
1052 * ap->ops->freeze() callback can be used for freezing the port
1053 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1054 * port cannot be frozen hardware-wise, the interrupt handler
1055 * must ack and clear interrupts unconditionally while the port
1059 * spin_lock_irqsave(host lock)
1061 static void __ata_port_freeze(struct ata_port *ap)
1063 WARN_ON(!ap->ops->error_handler);
1065 if (ap->ops->freeze)
1066 ap->ops->freeze(ap);
1068 ap->pflags |= ATA_PFLAG_FROZEN;
1070 trace_ata_port_freeze(ap);
1074 * ata_port_freeze - abort & freeze port
1075 * @ap: ATA port to freeze
1077 * Abort and freeze @ap. The freeze operation must be called
1078 * first, because some hardware requires special operations
1079 * before the taskfile registers are accessible.
1082 * spin_lock_irqsave(host lock)
1085 * Number of aborted commands.
1087 int ata_port_freeze(struct ata_port *ap)
1089 WARN_ON(!ap->ops->error_handler);
1091 __ata_port_freeze(ap);
1093 return ata_port_abort(ap);
1095 EXPORT_SYMBOL_GPL(ata_port_freeze);
1098 * ata_eh_freeze_port - EH helper to freeze port
1099 * @ap: ATA port to freeze
1106 void ata_eh_freeze_port(struct ata_port *ap)
1108 unsigned long flags;
1110 if (!ap->ops->error_handler)
1113 spin_lock_irqsave(ap->lock, flags);
1114 __ata_port_freeze(ap);
1115 spin_unlock_irqrestore(ap->lock, flags);
1117 EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1120 * ata_eh_thaw_port - EH helper to thaw port
1121 * @ap: ATA port to thaw
1123 * Thaw frozen port @ap.
1128 void ata_eh_thaw_port(struct ata_port *ap)
1130 unsigned long flags;
1132 if (!ap->ops->error_handler)
1135 spin_lock_irqsave(ap->lock, flags);
1137 ap->pflags &= ~ATA_PFLAG_FROZEN;
1142 spin_unlock_irqrestore(ap->lock, flags);
1144 trace_ata_port_thaw(ap);
1147 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1152 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1154 struct ata_port *ap = qc->ap;
1155 struct scsi_cmnd *scmd = qc->scsicmd;
1156 unsigned long flags;
1158 spin_lock_irqsave(ap->lock, flags);
1159 qc->scsidone = ata_eh_scsidone;
1160 __ata_qc_complete(qc);
1161 WARN_ON(ata_tag_valid(qc->tag));
1162 spin_unlock_irqrestore(ap->lock, flags);
1164 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1168 * ata_eh_qc_complete - Complete an active ATA command from EH
1169 * @qc: Command to complete
1171 * Indicate to the mid and upper layers that an ATA command has
1172 * completed. To be used from EH.
1174 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1176 struct scsi_cmnd *scmd = qc->scsicmd;
1177 scmd->retries = scmd->allowed;
1178 __ata_eh_qc_complete(qc);
1182 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1183 * @qc: Command to retry
1185 * Indicate to the mid and upper layers that an ATA command
1186 * should be retried. To be used from EH.
1188 * SCSI midlayer limits the number of retries to scmd->allowed.
1189 * scmd->allowed is incremented for commands which get retried
1190 * due to unrelated failures (qc->err_mask is zero).
1192 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1194 struct scsi_cmnd *scmd = qc->scsicmd;
1197 __ata_eh_qc_complete(qc);
1201 * ata_dev_disable - disable ATA device
1202 * @dev: ATA device to disable
1209 void ata_dev_disable(struct ata_device *dev)
1211 if (!ata_dev_enabled(dev))
1214 ata_dev_warn(dev, "disable device\n");
1215 ata_acpi_on_disable(dev);
1216 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1219 /* From now till the next successful probe, ering is used to
1220 * track probe failures. Clear accumulated device error info.
1222 ata_ering_clear(&dev->ering);
1224 EXPORT_SYMBOL_GPL(ata_dev_disable);
1227 * ata_eh_detach_dev - detach ATA device
1228 * @dev: ATA device to detach
1235 void ata_eh_detach_dev(struct ata_device *dev)
1237 struct ata_link *link = dev->link;
1238 struct ata_port *ap = link->ap;
1239 struct ata_eh_context *ehc = &link->eh_context;
1240 unsigned long flags;
1242 ata_dev_disable(dev);
1244 spin_lock_irqsave(ap->lock, flags);
1246 dev->flags &= ~ATA_DFLAG_DETACH;
1248 if (ata_scsi_offline_dev(dev)) {
1249 dev->flags |= ATA_DFLAG_DETACHED;
1250 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1253 /* clear per-dev EH info */
1254 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1255 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1256 ehc->saved_xfer_mode[dev->devno] = 0;
1257 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1259 spin_unlock_irqrestore(ap->lock, flags);
1263 * ata_eh_about_to_do - about to perform eh_action
1264 * @link: target ATA link
1265 * @dev: target ATA dev for per-dev action (can be NULL)
1266 * @action: action about to be performed
1268 * Called just before performing EH actions to clear related bits
1269 * in @link->eh_info such that eh actions are not unnecessarily
1275 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1276 unsigned int action)
1278 struct ata_port *ap = link->ap;
1279 struct ata_eh_info *ehi = &link->eh_info;
1280 struct ata_eh_context *ehc = &link->eh_context;
1281 unsigned long flags;
1283 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1285 spin_lock_irqsave(ap->lock, flags);
1287 ata_eh_clear_action(link, dev, ehi, action);
1289 /* About to take EH action, set RECOVERED. Ignore actions on
1290 * slave links as master will do them again.
1292 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1293 ap->pflags |= ATA_PFLAG_RECOVERED;
1295 spin_unlock_irqrestore(ap->lock, flags);
1299 * ata_eh_done - EH action complete
1300 * @link: ATA link for which EH actions are complete
1301 * @dev: target ATA dev for per-dev action (can be NULL)
1302 * @action: action just completed
1304 * Called right after performing EH actions to clear related bits
1305 * in @link->eh_context.
1310 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1311 unsigned int action)
1313 struct ata_eh_context *ehc = &link->eh_context;
1315 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1317 ata_eh_clear_action(link, dev, &ehc->i, action);
1321 * ata_err_string - convert err_mask to descriptive string
1322 * @err_mask: error mask to convert to string
1324 * Convert @err_mask to descriptive string. Errors are
1325 * prioritized according to severity and only the most severe
1326 * error is reported.
1332 * Descriptive string for @err_mask
1334 static const char *ata_err_string(unsigned int err_mask)
1336 if (err_mask & AC_ERR_HOST_BUS)
1337 return "host bus error";
1338 if (err_mask & AC_ERR_ATA_BUS)
1339 return "ATA bus error";
1340 if (err_mask & AC_ERR_TIMEOUT)
1342 if (err_mask & AC_ERR_HSM)
1343 return "HSM violation";
1344 if (err_mask & AC_ERR_SYSTEM)
1345 return "internal error";
1346 if (err_mask & AC_ERR_MEDIA)
1347 return "media error";
1348 if (err_mask & AC_ERR_INVALID)
1349 return "invalid argument";
1350 if (err_mask & AC_ERR_DEV)
1351 return "device error";
1352 if (err_mask & AC_ERR_NCQ)
1354 if (err_mask & AC_ERR_NODEV_HINT)
1355 return "Polling detection error";
1356 return "unknown error";
1360 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1361 * @dev: target ATAPI device
1362 * @r_sense_key: out parameter for sense_key
1364 * Perform ATAPI TEST_UNIT_READY.
1367 * EH context (may sleep).
1370 * 0 on success, AC_ERR_* mask on failure.
1372 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1374 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1375 struct ata_taskfile tf;
1376 unsigned int err_mask;
1378 ata_tf_init(dev, &tf);
1380 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1381 tf.command = ATA_CMD_PACKET;
1382 tf.protocol = ATAPI_PROT_NODATA;
1384 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1385 if (err_mask == AC_ERR_DEV)
1386 *r_sense_key = tf.error >> 4;
1391 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1392 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1394 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1395 * SENSE. This function is an EH helper.
1398 * Kernel thread context (may sleep).
1400 static void ata_eh_request_sense(struct ata_queued_cmd *qc)
1402 struct scsi_cmnd *cmd = qc->scsicmd;
1403 struct ata_device *dev = qc->dev;
1404 struct ata_taskfile tf;
1405 unsigned int err_mask;
1407 if (qc->ap->pflags & ATA_PFLAG_FROZEN) {
1408 ata_dev_warn(dev, "sense data available but port frozen\n");
1412 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1415 if (!ata_id_sense_reporting_enabled(dev->id)) {
1416 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1420 ata_tf_init(dev, &tf);
1421 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1422 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1423 tf.command = ATA_CMD_REQ_SENSE_DATA;
1424 tf.protocol = ATA_PROT_NODATA;
1426 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1427 /* Ignore err_mask; ATA_ERR might be set */
1428 if (tf.status & ATA_SENSE) {
1429 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1430 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1432 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1433 tf.status, err_mask);
1438 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1439 * @dev: device to perform REQUEST_SENSE to
1440 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1441 * @dfl_sense_key: default sense key to use
1443 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1444 * SENSE. This function is EH helper.
1447 * Kernel thread context (may sleep).
1450 * 0 on success, AC_ERR_* mask on failure
1452 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1453 u8 *sense_buf, u8 dfl_sense_key)
1455 u8 cdb[ATAPI_CDB_LEN] =
1456 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1457 struct ata_port *ap = dev->link->ap;
1458 struct ata_taskfile tf;
1460 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1462 /* initialize sense_buf with the error register,
1463 * for the case where they are -not- overwritten
1465 sense_buf[0] = 0x70;
1466 sense_buf[2] = dfl_sense_key;
1468 /* some devices time out if garbage left in tf */
1469 ata_tf_init(dev, &tf);
1471 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1472 tf.command = ATA_CMD_PACKET;
1474 /* is it pointless to prefer PIO for "safety reasons"? */
1475 if (ap->flags & ATA_FLAG_PIO_DMA) {
1476 tf.protocol = ATAPI_PROT_DMA;
1477 tf.feature |= ATAPI_PKT_DMA;
1479 tf.protocol = ATAPI_PROT_PIO;
1480 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1484 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1485 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1489 * ata_eh_analyze_serror - analyze SError for a failed port
1490 * @link: ATA link to analyze SError for
1492 * Analyze SError if available and further determine cause of
1498 static void ata_eh_analyze_serror(struct ata_link *link)
1500 struct ata_eh_context *ehc = &link->eh_context;
1501 u32 serror = ehc->i.serror;
1502 unsigned int err_mask = 0, action = 0;
1505 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1506 err_mask |= AC_ERR_ATA_BUS;
1507 action |= ATA_EH_RESET;
1509 if (serror & SERR_PROTOCOL) {
1510 err_mask |= AC_ERR_HSM;
1511 action |= ATA_EH_RESET;
1513 if (serror & SERR_INTERNAL) {
1514 err_mask |= AC_ERR_SYSTEM;
1515 action |= ATA_EH_RESET;
1518 /* Determine whether a hotplug event has occurred. Both
1519 * SError.N/X are considered hotplug events for enabled or
1520 * host links. For disabled PMP links, only N bit is
1521 * considered as X bit is left at 1 for link plugging.
1523 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1524 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1525 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1526 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1528 hotplug_mask = SERR_PHYRDY_CHG;
1530 if (serror & hotplug_mask)
1531 ata_ehi_hotplugged(&ehc->i);
1533 ehc->i.err_mask |= err_mask;
1534 ehc->i.action |= action;
1538 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1539 * @qc: qc to analyze
1541 * Analyze taskfile of @qc and further determine cause of
1542 * failure. This function also requests ATAPI sense data if
1546 * Kernel thread context (may sleep).
1549 * Determined recovery action
1551 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1553 const struct ata_taskfile *tf = &qc->result_tf;
1554 unsigned int tmp, action = 0;
1555 u8 stat = tf->status, err = tf->error;
1557 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1558 qc->err_mask |= AC_ERR_HSM;
1559 return ATA_EH_RESET;
1562 if (stat & (ATA_ERR | ATA_DF)) {
1563 qc->err_mask |= AC_ERR_DEV;
1565 * Sense data reporting does not work if the
1566 * device fault bit is set.
1574 switch (qc->dev->class) {
1576 if (stat & ATA_SENSE)
1577 ata_eh_request_sense(qc);
1581 qc->err_mask |= AC_ERR_ATA_BUS;
1582 if (err & (ATA_UNC | ATA_AMNF))
1583 qc->err_mask |= AC_ERR_MEDIA;
1585 qc->err_mask |= AC_ERR_INVALID;
1589 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1590 tmp = atapi_eh_request_sense(qc->dev,
1591 qc->scsicmd->sense_buffer,
1592 qc->result_tf.error >> 4);
1594 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1596 qc->err_mask |= tmp;
1600 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1601 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
1603 * SUCCESS here means that the sense code could be
1604 * evaluated and should be passed to the upper layers
1605 * for correct evaluation.
1606 * FAILED means the sense code could not be interpreted
1607 * and the device would need to be reset.
1608 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1609 * command would need to be retried.
1611 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1612 qc->flags |= ATA_QCFLAG_RETRY;
1613 qc->err_mask |= AC_ERR_OTHER;
1614 } else if (ret != SUCCESS) {
1615 qc->err_mask |= AC_ERR_HSM;
1618 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1619 action |= ATA_EH_RESET;
1624 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1629 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1633 base = ATA_ECAT_DUBIOUS_NONE;
1635 if (err_mask & AC_ERR_ATA_BUS)
1636 return base + ATA_ECAT_ATA_BUS;
1638 if (err_mask & AC_ERR_TIMEOUT)
1639 return base + ATA_ECAT_TOUT_HSM;
1641 if (eflags & ATA_EFLAG_IS_IO) {
1642 if (err_mask & AC_ERR_HSM)
1643 return base + ATA_ECAT_TOUT_HSM;
1645 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1646 return base + ATA_ECAT_UNK_DEV;
1652 struct speed_down_verdict_arg {
1655 int nr_errors[ATA_ECAT_NR];
1658 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1660 struct speed_down_verdict_arg *arg = void_arg;
1663 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1666 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1668 arg->nr_errors[cat]++;
1674 * ata_eh_speed_down_verdict - Determine speed down verdict
1675 * @dev: Device of interest
1677 * This function examines error ring of @dev and determines
1678 * whether NCQ needs to be turned off, transfer speed should be
1679 * stepped down, or falling back to PIO is necessary.
1681 * ECAT_ATA_BUS : ATA_BUS error for any command
1683 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1686 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1688 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1689 * data transfer hasn't been verified.
1693 * NCQ_OFF : Turn off NCQ.
1695 * SPEED_DOWN : Speed down transfer speed but don't fall back
1698 * FALLBACK_TO_PIO : Fall back to PIO.
1700 * Even if multiple verdicts are returned, only one action is
1701 * taken per error. An action triggered by non-DUBIOUS errors
1702 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1703 * This is to expedite speed down decisions right after device is
1704 * initially configured.
1706 * The following are speed down rules. #1 and #2 deal with
1709 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1710 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1712 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1713 * occurred during last 5 mins, NCQ_OFF.
1715 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1716 * occurred during last 5 mins, FALLBACK_TO_PIO
1718 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1719 * during last 10 mins, NCQ_OFF.
1721 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1722 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1725 * Inherited from caller.
1728 * OR of ATA_EH_SPDN_* flags.
1730 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1732 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1733 u64 j64 = get_jiffies_64();
1734 struct speed_down_verdict_arg arg;
1735 unsigned int verdict = 0;
1737 /* scan past 5 mins of error history */
1738 memset(&arg, 0, sizeof(arg));
1739 arg.since = j64 - min(j64, j5mins);
1740 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1742 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1743 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1744 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1745 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1747 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1748 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1749 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1751 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1752 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1753 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1754 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1756 /* scan past 10 mins of error history */
1757 memset(&arg, 0, sizeof(arg));
1758 arg.since = j64 - min(j64, j10mins);
1759 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1761 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1762 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1763 verdict |= ATA_EH_SPDN_NCQ_OFF;
1765 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1766 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1767 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1768 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1774 * ata_eh_speed_down - record error and speed down if necessary
1775 * @dev: Failed device
1776 * @eflags: mask of ATA_EFLAG_* flags
1777 * @err_mask: err_mask of the error
1779 * Record error and examine error history to determine whether
1780 * adjusting transmission speed is necessary. It also sets
1781 * transmission limits appropriately if such adjustment is
1785 * Kernel thread context (may sleep).
1788 * Determined recovery action.
1790 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1791 unsigned int eflags, unsigned int err_mask)
1793 struct ata_link *link = ata_dev_phys_link(dev);
1795 unsigned int verdict;
1796 unsigned int action = 0;
1798 /* don't bother if Cat-0 error */
1799 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1802 /* record error and determine whether speed down is necessary */
1803 ata_ering_record(&dev->ering, eflags, err_mask);
1804 verdict = ata_eh_speed_down_verdict(dev);
1807 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1808 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1809 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1810 dev->flags |= ATA_DFLAG_NCQ_OFF;
1811 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1816 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1817 /* speed down SATA link speed if possible */
1818 if (sata_down_spd_limit(link, 0) == 0) {
1819 action |= ATA_EH_RESET;
1823 /* lower transfer mode */
1824 if (dev->spdn_cnt < 2) {
1825 static const int dma_dnxfer_sel[] =
1826 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1827 static const int pio_dnxfer_sel[] =
1828 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1831 if (dev->xfer_shift != ATA_SHIFT_PIO)
1832 sel = dma_dnxfer_sel[dev->spdn_cnt];
1834 sel = pio_dnxfer_sel[dev->spdn_cnt];
1838 if (ata_down_xfermask_limit(dev, sel) == 0) {
1839 action |= ATA_EH_RESET;
1845 /* Fall back to PIO? Slowing down to PIO is meaningless for
1846 * SATA ATA devices. Consider it only for PATA and SATAPI.
1848 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1849 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1850 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1851 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1853 action |= ATA_EH_RESET;
1860 /* device has been slowed down, blow error history */
1861 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1862 ata_ering_clear(&dev->ering);
1867 * ata_eh_worth_retry - analyze error and decide whether to retry
1868 * @qc: qc to possibly retry
1870 * Look at the cause of the error and decide if a retry
1871 * might be useful or not. We don't want to retry media errors
1872 * because the drive itself has probably already taken 10-30 seconds
1873 * doing its own internal retries before reporting the failure.
1875 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1877 if (qc->err_mask & AC_ERR_MEDIA)
1878 return 0; /* don't retry media errors */
1879 if (qc->flags & ATA_QCFLAG_IO)
1880 return 1; /* otherwise retry anything from fs stack */
1881 if (qc->err_mask & AC_ERR_INVALID)
1882 return 0; /* don't retry these */
1883 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1887 * ata_eh_quiet - check if we need to be quiet about a command error
1890 * Look at the qc flags anbd its scsi command request flags to determine
1891 * if we need to be quiet about the command failure.
1893 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1895 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1896 qc->flags |= ATA_QCFLAG_QUIET;
1897 return qc->flags & ATA_QCFLAG_QUIET;
1901 * ata_eh_link_autopsy - analyze error and determine recovery action
1902 * @link: host link to perform autopsy on
1904 * Analyze why @link failed and determine which recovery actions
1905 * are needed. This function also sets more detailed AC_ERR_*
1906 * values and fills sense data for ATAPI CHECK SENSE.
1909 * Kernel thread context (may sleep).
1911 static void ata_eh_link_autopsy(struct ata_link *link)
1913 struct ata_port *ap = link->ap;
1914 struct ata_eh_context *ehc = &link->eh_context;
1915 struct ata_queued_cmd *qc;
1916 struct ata_device *dev;
1917 unsigned int all_err_mask = 0, eflags = 0;
1918 int tag, nr_failed = 0, nr_quiet = 0;
1922 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1925 /* obtain and analyze SError */
1926 rc = sata_scr_read(link, SCR_ERROR, &serror);
1928 ehc->i.serror |= serror;
1929 ata_eh_analyze_serror(link);
1930 } else if (rc != -EOPNOTSUPP) {
1931 /* SError read failed, force reset and probing */
1932 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1933 ehc->i.action |= ATA_EH_RESET;
1934 ehc->i.err_mask |= AC_ERR_OTHER;
1937 /* analyze NCQ failure */
1938 ata_eh_analyze_ncq_error(link);
1940 /* any real error trumps AC_ERR_OTHER */
1941 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1942 ehc->i.err_mask &= ~AC_ERR_OTHER;
1944 all_err_mask |= ehc->i.err_mask;
1946 ata_qc_for_each_raw(ap, qc, tag) {
1947 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1948 ata_dev_phys_link(qc->dev) != link)
1951 /* inherit upper level err_mask */
1952 qc->err_mask |= ehc->i.err_mask;
1955 ehc->i.action |= ata_eh_analyze_tf(qc);
1957 /* DEV errors are probably spurious in case of ATA_BUS error */
1958 if (qc->err_mask & AC_ERR_ATA_BUS)
1959 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1962 /* any real error trumps unknown error */
1963 if (qc->err_mask & ~AC_ERR_OTHER)
1964 qc->err_mask &= ~AC_ERR_OTHER;
1967 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
1968 * layers will determine whether the command is worth retrying
1969 * based on the sense data and device class/type. Otherwise,
1970 * determine directly if the command is worth retrying using its
1971 * error mask and flags.
1973 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1974 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1975 else if (ata_eh_worth_retry(qc))
1976 qc->flags |= ATA_QCFLAG_RETRY;
1978 /* accumulate error info */
1979 ehc->i.dev = qc->dev;
1980 all_err_mask |= qc->err_mask;
1981 if (qc->flags & ATA_QCFLAG_IO)
1982 eflags |= ATA_EFLAG_IS_IO;
1983 trace_ata_eh_link_autopsy_qc(qc);
1985 /* Count quiet errors */
1986 if (ata_eh_quiet(qc))
1991 /* If all failed commands requested silence, then be quiet */
1992 if (nr_quiet == nr_failed)
1993 ehc->i.flags |= ATA_EHI_QUIET;
1995 /* enforce default EH actions */
1996 if (ap->pflags & ATA_PFLAG_FROZEN ||
1997 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1998 ehc->i.action |= ATA_EH_RESET;
1999 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2000 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2001 ehc->i.action |= ATA_EH_REVALIDATE;
2003 /* If we have offending qcs and the associated failed device,
2004 * perform per-dev EH action only on the offending device.
2007 ehc->i.dev_action[ehc->i.dev->devno] |=
2008 ehc->i.action & ATA_EH_PERDEV_MASK;
2009 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2012 /* propagate timeout to host link */
2013 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2014 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2016 /* record error and consider speeding down */
2018 if (!dev && ((ata_link_max_devices(link) == 1 &&
2019 ata_dev_enabled(link->device))))
2023 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2024 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2025 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2026 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2031 * ata_eh_autopsy - analyze error and determine recovery action
2032 * @ap: host port to perform autopsy on
2034 * Analyze all links of @ap and determine why they failed and
2035 * which recovery actions are needed.
2038 * Kernel thread context (may sleep).
2040 void ata_eh_autopsy(struct ata_port *ap)
2042 struct ata_link *link;
2044 ata_for_each_link(link, ap, EDGE)
2045 ata_eh_link_autopsy(link);
2047 /* Handle the frigging slave link. Autopsy is done similarly
2048 * but actions and flags are transferred over to the master
2049 * link and handled from there.
2051 if (ap->slave_link) {
2052 struct ata_eh_context *mehc = &ap->link.eh_context;
2053 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2055 /* transfer control flags from master to slave */
2056 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2058 /* perform autopsy on the slave link */
2059 ata_eh_link_autopsy(ap->slave_link);
2061 /* transfer actions from slave to master and clear slave */
2062 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2063 mehc->i.action |= sehc->i.action;
2064 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2065 mehc->i.flags |= sehc->i.flags;
2066 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2069 /* Autopsy of fanout ports can affect host link autopsy.
2070 * Perform host link autopsy last.
2072 if (sata_pmp_attached(ap))
2073 ata_eh_link_autopsy(&ap->link);
2077 * ata_get_cmd_name - get name for ATA command
2078 * @command: ATA command code to get name for
2080 * Return a textual name of the given command or "unknown"
2085 const char *ata_get_cmd_name(u8 command)
2087 #ifdef CONFIG_ATA_VERBOSE_ERROR
2093 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2094 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2095 { ATA_CMD_STANDBY, "STANDBY" },
2096 { ATA_CMD_IDLE, "IDLE" },
2097 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2098 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2099 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2100 { ATA_CMD_NOP, "NOP" },
2101 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2102 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2103 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2104 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2105 { ATA_CMD_SERVICE, "SERVICE" },
2106 { ATA_CMD_READ, "READ DMA" },
2107 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2108 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2109 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2110 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2111 { ATA_CMD_WRITE, "WRITE DMA" },
2112 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2113 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2114 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2115 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2116 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2117 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2118 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2119 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2120 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2121 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2122 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2123 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2124 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2125 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2126 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2127 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2128 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2129 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2130 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2131 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2132 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2133 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2134 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2135 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2136 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2137 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2138 { ATA_CMD_SLEEP, "SLEEP" },
2139 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2140 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2141 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2142 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2143 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2144 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2145 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2146 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2147 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2148 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2149 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2150 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2151 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2152 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2153 { ATA_CMD_PMP_READ, "READ BUFFER" },
2154 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2155 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2156 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2157 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2158 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2159 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2160 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2161 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2162 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2163 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2164 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2165 { ATA_CMD_SMART, "SMART" },
2166 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2167 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2168 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2169 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2170 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2171 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2172 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2173 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2174 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2175 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2176 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2177 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2178 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2179 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2180 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2181 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2182 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2183 { ATA_CMD_RESTORE, "RECALIBRATE" },
2184 { 0, NULL } /* terminate list */
2188 for (i = 0; cmd_descr[i].text; i++)
2189 if (cmd_descr[i].command == command)
2190 return cmd_descr[i].text;
2195 EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2198 * ata_eh_link_report - report error handling to user
2199 * @link: ATA link EH is going on
2201 * Report EH to user.
2206 static void ata_eh_link_report(struct ata_link *link)
2208 struct ata_port *ap = link->ap;
2209 struct ata_eh_context *ehc = &link->eh_context;
2210 struct ata_queued_cmd *qc;
2211 const char *frozen, *desc;
2212 char tries_buf[6] = "";
2213 int tag, nr_failed = 0;
2215 if (ehc->i.flags & ATA_EHI_QUIET)
2219 if (ehc->i.desc[0] != '\0')
2222 ata_qc_for_each_raw(ap, qc, tag) {
2223 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2224 ata_dev_phys_link(qc->dev) != link ||
2225 ((qc->flags & ATA_QCFLAG_QUIET) &&
2226 qc->err_mask == AC_ERR_DEV))
2228 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2234 if (!nr_failed && !ehc->i.err_mask)
2238 if (ap->pflags & ATA_PFLAG_FROZEN)
2241 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2242 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2246 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2247 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2248 ehc->i.err_mask, link->sactive, ehc->i.serror,
2249 ehc->i.action, frozen, tries_buf);
2251 ata_dev_err(ehc->i.dev, "%s\n", desc);
2253 ata_link_err(link, "exception Emask 0x%x "
2254 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2255 ehc->i.err_mask, link->sactive, ehc->i.serror,
2256 ehc->i.action, frozen, tries_buf);
2258 ata_link_err(link, "%s\n", desc);
2261 #ifdef CONFIG_ATA_VERBOSE_ERROR
2264 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2265 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2266 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2267 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2268 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2269 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2270 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2271 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2272 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2273 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2274 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2275 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2276 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2277 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2278 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2279 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2280 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2281 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2284 ata_qc_for_each_raw(ap, qc, tag) {
2285 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2286 char data_buf[20] = "";
2287 char cdb_buf[70] = "";
2289 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2290 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2293 if (qc->dma_dir != DMA_NONE) {
2294 static const char *dma_str[] = {
2295 [DMA_BIDIRECTIONAL] = "bidi",
2296 [DMA_TO_DEVICE] = "out",
2297 [DMA_FROM_DEVICE] = "in",
2299 const char *prot_str = NULL;
2301 switch (qc->tf.protocol) {
2302 case ATA_PROT_UNKNOWN:
2303 prot_str = "unknown";
2305 case ATA_PROT_NODATA:
2306 prot_str = "nodata";
2315 prot_str = "ncq dma";
2317 case ATA_PROT_NCQ_NODATA:
2318 prot_str = "ncq nodata";
2320 case ATAPI_PROT_NODATA:
2321 prot_str = "nodata";
2323 case ATAPI_PROT_PIO:
2326 case ATAPI_PROT_DMA:
2330 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2331 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2334 if (ata_is_atapi(qc->tf.protocol)) {
2335 const u8 *cdb = qc->cdb;
2336 size_t cdb_len = qc->dev->cdb_len;
2339 cdb = qc->scsicmd->cmnd;
2340 cdb_len = qc->scsicmd->cmd_len;
2342 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2345 ata_dev_err(qc->dev, "failed command: %s\n",
2346 ata_get_cmd_name(cmd->command));
2348 ata_dev_err(qc->dev,
2349 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2351 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2352 "Emask 0x%x (%s)%s\n",
2353 cmd->command, cmd->feature, cmd->nsect,
2354 cmd->lbal, cmd->lbam, cmd->lbah,
2355 cmd->hob_feature, cmd->hob_nsect,
2356 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2357 cmd->device, qc->tag, data_buf, cdb_buf,
2358 res->status, res->error, res->nsect,
2359 res->lbal, res->lbam, res->lbah,
2360 res->hob_feature, res->hob_nsect,
2361 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2362 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2363 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2365 #ifdef CONFIG_ATA_VERBOSE_ERROR
2366 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2367 ATA_SENSE | ATA_ERR)) {
2368 if (res->status & ATA_BUSY)
2369 ata_dev_err(qc->dev, "status: { Busy }\n");
2371 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2372 res->status & ATA_DRDY ? "DRDY " : "",
2373 res->status & ATA_DF ? "DF " : "",
2374 res->status & ATA_DRQ ? "DRQ " : "",
2375 res->status & ATA_SENSE ? "SENSE " : "",
2376 res->status & ATA_ERR ? "ERR " : "");
2379 if (cmd->command != ATA_CMD_PACKET &&
2380 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2382 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2383 res->error & ATA_ICRC ? "ICRC " : "",
2384 res->error & ATA_UNC ? "UNC " : "",
2385 res->error & ATA_AMNF ? "AMNF " : "",
2386 res->error & ATA_IDNF ? "IDNF " : "",
2387 res->error & ATA_ABORTED ? "ABRT " : "");
2393 * ata_eh_report - report error handling to user
2394 * @ap: ATA port to report EH about
2396 * Report EH to user.
2401 void ata_eh_report(struct ata_port *ap)
2403 struct ata_link *link;
2405 ata_for_each_link(link, ap, HOST_FIRST)
2406 ata_eh_link_report(link);
2409 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2410 unsigned int *classes, unsigned long deadline,
2413 struct ata_device *dev;
2416 ata_for_each_dev(dev, link, ALL)
2417 classes[dev->devno] = ATA_DEV_UNKNOWN;
2419 return reset(link, classes, deadline);
2422 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2424 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2428 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2433 int ata_eh_reset(struct ata_link *link, int classify,
2434 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2435 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2437 struct ata_port *ap = link->ap;
2438 struct ata_link *slave = ap->slave_link;
2439 struct ata_eh_context *ehc = &link->eh_context;
2440 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2441 unsigned int *classes = ehc->classes;
2442 unsigned int lflags = link->flags;
2443 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2444 int max_tries = 0, try = 0;
2445 struct ata_link *failed_link;
2446 struct ata_device *dev;
2447 unsigned long deadline, now;
2448 ata_reset_fn_t reset;
2449 unsigned long flags;
2456 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2458 if (link->flags & ATA_LFLAG_RST_ONCE)
2460 if (link->flags & ATA_LFLAG_NO_HRST)
2462 if (link->flags & ATA_LFLAG_NO_SRST)
2465 /* make sure each reset attempt is at least COOL_DOWN apart */
2466 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2468 WARN_ON(time_after(ehc->last_reset, now));
2469 deadline = ata_deadline(ehc->last_reset,
2470 ATA_EH_RESET_COOL_DOWN);
2471 if (time_before(now, deadline))
2472 schedule_timeout_uninterruptible(deadline - now);
2475 spin_lock_irqsave(ap->lock, flags);
2476 ap->pflags |= ATA_PFLAG_RESETTING;
2477 spin_unlock_irqrestore(ap->lock, flags);
2479 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2481 ata_for_each_dev(dev, link, ALL) {
2482 /* If we issue an SRST then an ATA drive (not ATAPI)
2483 * may change configuration and be in PIO0 timing. If
2484 * we do a hard reset (or are coming from power on)
2485 * this is true for ATA or ATAPI. Until we've set a
2486 * suitable controller mode we should not touch the
2487 * bus as we may be talking too fast.
2489 dev->pio_mode = XFER_PIO_0;
2490 dev->dma_mode = 0xff;
2492 /* If the controller has a pio mode setup function
2493 * then use it to set the chipset to rights. Don't
2494 * touch the DMA setup as that will be dealt with when
2495 * configuring devices.
2497 if (ap->ops->set_piomode)
2498 ap->ops->set_piomode(ap, dev);
2501 /* prefer hardreset */
2503 ehc->i.action &= ~ATA_EH_RESET;
2506 ehc->i.action |= ATA_EH_HARDRESET;
2507 } else if (softreset) {
2509 ehc->i.action |= ATA_EH_SOFTRESET;
2513 unsigned long deadline = ata_deadline(jiffies,
2514 ATA_EH_PRERESET_TIMEOUT);
2517 sehc->i.action &= ~ATA_EH_RESET;
2518 sehc->i.action |= ehc->i.action;
2521 rc = prereset(link, deadline);
2523 /* If present, do prereset on slave link too. Reset
2524 * is skipped iff both master and slave links report
2525 * -ENOENT or clear ATA_EH_RESET.
2527 if (slave && (rc == 0 || rc == -ENOENT)) {
2530 tmp = prereset(slave, deadline);
2534 ehc->i.action |= sehc->i.action;
2538 if (rc == -ENOENT) {
2539 ata_link_dbg(link, "port disabled--ignoring\n");
2540 ehc->i.action &= ~ATA_EH_RESET;
2542 ata_for_each_dev(dev, link, ALL)
2543 classes[dev->devno] = ATA_DEV_NONE;
2548 "prereset failed (errno=%d)\n",
2553 /* prereset() might have cleared ATA_EH_RESET. If so,
2554 * bang classes, thaw and return.
2556 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2557 ata_for_each_dev(dev, link, ALL)
2558 classes[dev->devno] = ATA_DEV_NONE;
2559 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2560 ata_is_host_link(link))
2561 ata_eh_thaw_port(ap);
2571 if (ata_is_host_link(link))
2572 ata_eh_freeze_port(ap);
2574 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2578 ata_link_info(link, "%s resetting link\n",
2579 reset == softreset ? "soft" : "hard");
2581 /* mark that this EH session started with reset */
2582 ehc->last_reset = jiffies;
2583 if (reset == hardreset) {
2584 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2585 trace_ata_link_hardreset_begin(link, classes, deadline);
2587 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2588 trace_ata_link_softreset_begin(link, classes, deadline);
2591 rc = ata_do_reset(link, reset, classes, deadline, true);
2592 if (reset == hardreset)
2593 trace_ata_link_hardreset_end(link, classes, rc);
2595 trace_ata_link_softreset_end(link, classes, rc);
2596 if (rc && rc != -EAGAIN) {
2601 /* hardreset slave link if existent */
2602 if (slave && reset == hardreset) {
2606 ata_link_info(slave, "hard resetting link\n");
2608 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2609 trace_ata_slave_hardreset_begin(slave, classes,
2611 tmp = ata_do_reset(slave, reset, classes, deadline,
2613 trace_ata_slave_hardreset_end(slave, classes, tmp);
2621 failed_link = slave;
2627 /* perform follow-up SRST if necessary */
2628 if (reset == hardreset &&
2629 ata_eh_followup_srst_needed(link, rc)) {
2634 "follow-up softreset required but no softreset available\n");
2640 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2641 trace_ata_link_softreset_begin(link, classes, deadline);
2642 rc = ata_do_reset(link, reset, classes, deadline, true);
2643 trace_ata_link_softreset_end(link, classes, rc);
2652 "no reset method available, skipping reset\n");
2653 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2654 lflags |= ATA_LFLAG_ASSUME_ATA;
2658 * Post-reset processing
2660 ata_for_each_dev(dev, link, ALL) {
2661 /* After the reset, the device state is PIO 0 and the
2662 * controller state is undefined. Reset also wakes up
2663 * drives from sleeping mode.
2665 dev->pio_mode = XFER_PIO_0;
2666 dev->flags &= ~ATA_DFLAG_SLEEPING;
2668 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2671 /* apply class override */
2672 if (lflags & ATA_LFLAG_ASSUME_ATA)
2673 classes[dev->devno] = ATA_DEV_ATA;
2674 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2675 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2678 /* record current link speed */
2679 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2680 link->sata_spd = (sstatus >> 4) & 0xf;
2681 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2682 slave->sata_spd = (sstatus >> 4) & 0xf;
2685 if (ata_is_host_link(link))
2686 ata_eh_thaw_port(ap);
2688 /* postreset() should clear hardware SError. Although SError
2689 * is cleared during link resume, clearing SError here is
2690 * necessary as some PHYs raise hotplug events after SRST.
2691 * This introduces race condition where hotplug occurs between
2692 * reset and here. This race is mediated by cross checking
2693 * link onlineness and classification result later.
2696 postreset(link, classes);
2697 trace_ata_link_postreset(link, classes, rc);
2699 postreset(slave, classes);
2700 trace_ata_slave_postreset(slave, classes, rc);
2705 * Some controllers can't be frozen very well and may set spurious
2706 * error conditions during reset. Clear accumulated error
2707 * information and re-thaw the port if frozen. As reset is the
2708 * final recovery action and we cross check link onlineness against
2709 * device classification later, no hotplug event is lost by this.
2711 spin_lock_irqsave(link->ap->lock, flags);
2712 memset(&link->eh_info, 0, sizeof(link->eh_info));
2714 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2715 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2716 spin_unlock_irqrestore(link->ap->lock, flags);
2718 if (ap->pflags & ATA_PFLAG_FROZEN)
2719 ata_eh_thaw_port(ap);
2722 * Make sure onlineness and classification result correspond.
2723 * Hotplug could have happened during reset and some
2724 * controllers fail to wait while a drive is spinning up after
2725 * being hotplugged causing misdetection. By cross checking
2726 * link on/offlineness and classification result, those
2727 * conditions can be reliably detected and retried.
2730 ata_for_each_dev(dev, link, ALL) {
2731 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2732 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2733 ata_dev_dbg(dev, "link online but device misclassified\n");
2734 classes[dev->devno] = ATA_DEV_NONE;
2737 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2738 if (ata_class_enabled(classes[dev->devno]))
2740 "link offline, clearing class %d to NONE\n",
2741 classes[dev->devno]);
2742 classes[dev->devno] = ATA_DEV_NONE;
2743 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2745 "link status unknown, clearing UNKNOWN to NONE\n");
2746 classes[dev->devno] = ATA_DEV_NONE;
2750 if (classify && nr_unknown) {
2751 if (try < max_tries) {
2753 "link online but %d devices misclassified, retrying\n",
2760 "link online but %d devices misclassified, "
2761 "device detection might fail\n", nr_unknown);
2764 /* reset successful, schedule revalidation */
2765 ata_eh_done(link, NULL, ATA_EH_RESET);
2767 ata_eh_done(slave, NULL, ATA_EH_RESET);
2768 ehc->last_reset = jiffies; /* update to completion time */
2769 ehc->i.action |= ATA_EH_REVALIDATE;
2770 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2774 /* clear hotplug flag */
2775 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2777 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2779 spin_lock_irqsave(ap->lock, flags);
2780 ap->pflags &= ~ATA_PFLAG_RESETTING;
2781 spin_unlock_irqrestore(ap->lock, flags);
2786 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2787 if (!ata_is_host_link(link) &&
2788 sata_scr_read(link, SCR_STATUS, &sstatus))
2791 if (try >= max_tries) {
2793 * Thaw host port even if reset failed, so that the port
2794 * can be retried on the next phy event. This risks
2795 * repeated EH runs but seems to be a better tradeoff than
2796 * shutting down a port after a botched hotplug attempt.
2798 if (ata_is_host_link(link))
2799 ata_eh_thaw_port(ap);
2804 if (time_before(now, deadline)) {
2805 unsigned long delta = deadline - now;
2807 ata_link_warn(failed_link,
2808 "reset failed (errno=%d), retrying in %u secs\n",
2809 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2813 delta = schedule_timeout_uninterruptible(delta);
2818 * While disks spinup behind PMP, some controllers fail sending SRST.
2819 * They need to be reset - as well as the PMP - before retrying.
2821 if (rc == -ERESTART) {
2822 if (ata_is_host_link(link))
2823 ata_eh_thaw_port(ap);
2827 if (try == max_tries - 1) {
2828 sata_down_spd_limit(link, 0);
2830 sata_down_spd_limit(slave, 0);
2831 } else if (rc == -EPIPE)
2832 sata_down_spd_limit(failed_link, 0);
2839 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2841 struct ata_link *link;
2842 struct ata_device *dev;
2843 unsigned long flags;
2846 * This function can be thought of as an extended version of
2847 * ata_eh_about_to_do() specially crafted to accommodate the
2848 * requirements of ATA_EH_PARK handling. Since the EH thread
2849 * does not leave the do {} while () loop in ata_eh_recover as
2850 * long as the timeout for a park request to *one* device on
2851 * the port has not expired, and since we still want to pick
2852 * up park requests to other devices on the same port or
2853 * timeout updates for the same device, we have to pull
2854 * ATA_EH_PARK actions from eh_info into eh_context.i
2855 * ourselves at the beginning of each pass over the loop.
2857 * Additionally, all write accesses to &ap->park_req_pending
2858 * through reinit_completion() (see below) or complete_all()
2859 * (see ata_scsi_park_store()) are protected by the host lock.
2860 * As a result we have that park_req_pending.done is zero on
2861 * exit from this function, i.e. when ATA_EH_PARK actions for
2862 * *all* devices on port ap have been pulled into the
2863 * respective eh_context structs. If, and only if,
2864 * park_req_pending.done is non-zero by the time we reach
2865 * wait_for_completion_timeout(), another ATA_EH_PARK action
2866 * has been scheduled for at least one of the devices on port
2867 * ap and we have to cycle over the do {} while () loop in
2868 * ata_eh_recover() again.
2871 spin_lock_irqsave(ap->lock, flags);
2872 reinit_completion(&ap->park_req_pending);
2873 ata_for_each_link(link, ap, EDGE) {
2874 ata_for_each_dev(dev, link, ALL) {
2875 struct ata_eh_info *ehi = &link->eh_info;
2877 link->eh_context.i.dev_action[dev->devno] |=
2878 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2879 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2882 spin_unlock_irqrestore(ap->lock, flags);
2885 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2887 struct ata_eh_context *ehc = &dev->link->eh_context;
2888 struct ata_taskfile tf;
2889 unsigned int err_mask;
2891 ata_tf_init(dev, &tf);
2893 ehc->unloaded_mask |= 1 << dev->devno;
2894 tf.command = ATA_CMD_IDLEIMMEDIATE;
2900 ehc->unloaded_mask &= ~(1 << dev->devno);
2901 tf.command = ATA_CMD_CHK_POWER;
2904 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2905 tf.protocol = ATA_PROT_NODATA;
2906 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2907 if (park && (err_mask || tf.lbal != 0xc4)) {
2908 ata_dev_err(dev, "head unload failed!\n");
2909 ehc->unloaded_mask &= ~(1 << dev->devno);
2913 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2914 struct ata_device **r_failed_dev)
2916 struct ata_port *ap = link->ap;
2917 struct ata_eh_context *ehc = &link->eh_context;
2918 struct ata_device *dev;
2919 unsigned int new_mask = 0;
2920 unsigned long flags;
2923 /* For PATA drive side cable detection to work, IDENTIFY must
2924 * be done backwards such that PDIAG- is released by the slave
2925 * device before the master device is identified.
2927 ata_for_each_dev(dev, link, ALL_REVERSE) {
2928 unsigned int action = ata_eh_dev_action(dev);
2929 unsigned int readid_flags = 0;
2931 if (ehc->i.flags & ATA_EHI_DID_RESET)
2932 readid_flags |= ATA_READID_POSTRESET;
2934 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2935 WARN_ON(dev->class == ATA_DEV_PMP);
2937 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2942 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2943 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2948 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2950 /* Configuration may have changed, reconfigure
2953 ehc->i.flags |= ATA_EHI_SETMODE;
2955 /* schedule the scsi_rescan_device() here */
2956 schedule_work(&(ap->scsi_rescan_task));
2957 } else if (dev->class == ATA_DEV_UNKNOWN &&
2958 ehc->tries[dev->devno] &&
2959 ata_class_enabled(ehc->classes[dev->devno])) {
2960 /* Temporarily set dev->class, it will be
2961 * permanently set once all configurations are
2962 * complete. This is necessary because new
2963 * device configuration is done in two
2966 dev->class = ehc->classes[dev->devno];
2968 if (dev->class == ATA_DEV_PMP)
2969 rc = sata_pmp_attach(dev);
2971 rc = ata_dev_read_id(dev, &dev->class,
2972 readid_flags, dev->id);
2974 /* read_id might have changed class, store and reset */
2975 ehc->classes[dev->devno] = dev->class;
2976 dev->class = ATA_DEV_UNKNOWN;
2980 /* clear error info accumulated during probe */
2981 ata_ering_clear(&dev->ering);
2982 new_mask |= 1 << dev->devno;
2985 /* IDENTIFY was issued to non-existent
2986 * device. No need to reset. Just
2987 * thaw and ignore the device.
2989 ata_eh_thaw_port(ap);
2997 /* PDIAG- should have been released, ask cable type if post-reset */
2998 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2999 if (ap->ops->cable_detect)
3000 ap->cbl = ap->ops->cable_detect(ap);
3004 /* Configure new devices forward such that user doesn't see
3005 * device detection messages backwards.
3007 ata_for_each_dev(dev, link, ALL) {
3008 if (!(new_mask & (1 << dev->devno)))
3011 dev->class = ehc->classes[dev->devno];
3013 if (dev->class == ATA_DEV_PMP)
3016 ehc->i.flags |= ATA_EHI_PRINTINFO;
3017 rc = ata_dev_configure(dev);
3018 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3020 dev->class = ATA_DEV_UNKNOWN;
3024 spin_lock_irqsave(ap->lock, flags);
3025 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3026 spin_unlock_irqrestore(ap->lock, flags);
3028 /* new device discovered, configure xfermode */
3029 ehc->i.flags |= ATA_EHI_SETMODE;
3035 *r_failed_dev = dev;
3040 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3041 * @link: link on which timings will be programmed
3042 * @r_failed_dev: out parameter for failed device
3044 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3045 * ata_set_mode() fails, pointer to the failing device is
3046 * returned in @r_failed_dev.
3049 * PCI/etc. bus probe sem.
3052 * 0 on success, negative errno otherwise
3054 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3056 struct ata_port *ap = link->ap;
3057 struct ata_device *dev;
3060 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3061 ata_for_each_dev(dev, link, ENABLED) {
3062 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3063 struct ata_ering_entry *ent;
3065 ent = ata_ering_top(&dev->ering);
3067 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3071 /* has private set_mode? */
3072 if (ap->ops->set_mode)
3073 rc = ap->ops->set_mode(link, r_failed_dev);
3075 rc = ata_do_set_mode(link, r_failed_dev);
3077 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3078 ata_for_each_dev(dev, link, ENABLED) {
3079 struct ata_eh_context *ehc = &link->eh_context;
3080 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3081 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3083 if (dev->xfer_mode != saved_xfer_mode ||
3084 ata_ncq_enabled(dev) != saved_ncq)
3085 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3092 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3093 * @dev: ATAPI device to clear UA for
3095 * Resets and other operations can make an ATAPI device raise
3096 * UNIT ATTENTION which causes the next operation to fail. This
3097 * function clears UA.
3100 * EH context (may sleep).
3103 * 0 on success, -errno on failure.
3105 static int atapi_eh_clear_ua(struct ata_device *dev)
3109 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3110 u8 *sense_buffer = dev->link->ap->sector_buf;
3112 unsigned int err_mask;
3114 err_mask = atapi_eh_tur(dev, &sense_key);
3115 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3117 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3122 if (!err_mask || sense_key != UNIT_ATTENTION)
3125 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3127 ata_dev_warn(dev, "failed to clear "
3128 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3133 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3140 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3141 * @dev: ATA device which may need FLUSH retry
3143 * If @dev failed FLUSH, it needs to be reported upper layer
3144 * immediately as it means that @dev failed to remap and already
3145 * lost at least a sector and further FLUSH retrials won't make
3146 * any difference to the lost sector. However, if FLUSH failed
3147 * for other reasons, for example transmission error, FLUSH needs
3150 * This function determines whether FLUSH failure retry is
3151 * necessary and performs it if so.
3154 * 0 if EH can continue, -errno if EH needs to be repeated.
3156 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3158 struct ata_link *link = dev->link;
3159 struct ata_port *ap = link->ap;
3160 struct ata_queued_cmd *qc;
3161 struct ata_taskfile tf;
3162 unsigned int err_mask;
3165 /* did flush fail for this device? */
3166 if (!ata_tag_valid(link->active_tag))
3169 qc = __ata_qc_from_tag(ap, link->active_tag);
3170 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3171 qc->tf.command != ATA_CMD_FLUSH))
3174 /* if the device failed it, it should be reported to upper layers */
3175 if (qc->err_mask & AC_ERR_DEV)
3178 /* flush failed for some other reason, give it another shot */
3179 ata_tf_init(dev, &tf);
3181 tf.command = qc->tf.command;
3182 tf.flags |= ATA_TFLAG_DEVICE;
3183 tf.protocol = ATA_PROT_NODATA;
3185 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3186 tf.command, qc->err_mask);
3188 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3191 * FLUSH is complete but there's no way to
3192 * successfully complete a failed command from EH.
3193 * Making sure retry is allowed at least once and
3194 * retrying it should do the trick - whatever was in
3195 * the cache is already on the platter and this won't
3196 * cause infinite loop.
3198 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3200 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3204 /* if device failed it, report it to upper layers */
3205 if (err_mask & AC_ERR_DEV) {
3206 qc->err_mask |= AC_ERR_DEV;
3208 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3216 * ata_eh_set_lpm - configure SATA interface power management
3217 * @link: link to configure power management
3218 * @policy: the link power management policy
3219 * @r_failed_dev: out parameter for failed device
3221 * Enable SATA Interface power management. This will enable
3222 * Device Interface Power Management (DIPM) for min_power and
3223 * medium_power_with_dipm policies, and then call driver specific
3224 * callbacks for enabling Host Initiated Power management.
3230 * 0 on success, -errno on failure.
3232 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3233 struct ata_device **r_failed_dev)
3235 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3236 struct ata_eh_context *ehc = &link->eh_context;
3237 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3238 enum ata_lpm_policy old_policy = link->lpm_policy;
3239 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3240 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3241 unsigned int err_mask;
3244 /* if the link or host doesn't do LPM, noop */
3245 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3246 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3250 * DIPM is enabled only for MIN_POWER as some devices
3251 * misbehave when the host NACKs transition to SLUMBER. Order
3252 * device and link configurations such that the host always
3253 * allows DIPM requests.
3255 ata_for_each_dev(dev, link, ENABLED) {
3256 bool hipm = ata_id_has_hipm(dev->id);
3257 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3259 /* find the first enabled and LPM enabled devices */
3263 if (!lpm_dev && (hipm || dipm))
3266 hints &= ~ATA_LPM_EMPTY;
3268 hints &= ~ATA_LPM_HIPM;
3270 /* disable DIPM before changing link config */
3271 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3272 err_mask = ata_dev_set_feature(dev,
3273 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3274 if (err_mask && err_mask != AC_ERR_DEV) {
3276 "failed to disable DIPM, Emask 0x%x\n",
3285 rc = ap->ops->set_lpm(link, policy, hints);
3286 if (!rc && ap->slave_link)
3287 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3289 rc = sata_pmp_set_lpm(link, policy, hints);
3292 * Attribute link config failure to the first (LPM) enabled
3293 * device on the link.
3296 if (rc == -EOPNOTSUPP) {
3297 link->flags |= ATA_LFLAG_NO_LPM;
3300 dev = lpm_dev ? lpm_dev : link_dev;
3305 * Low level driver acked the transition. Issue DIPM command
3306 * with the new policy set.
3308 link->lpm_policy = policy;
3309 if (ap && ap->slave_link)
3310 ap->slave_link->lpm_policy = policy;
3312 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3313 ata_for_each_dev(dev, link, ENABLED) {
3314 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3315 ata_id_has_dipm(dev->id)) {
3316 err_mask = ata_dev_set_feature(dev,
3317 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3318 if (err_mask && err_mask != AC_ERR_DEV) {
3320 "failed to enable DIPM, Emask 0x%x\n",
3328 link->last_lpm_change = jiffies;
3329 link->flags |= ATA_LFLAG_CHANGED;
3334 /* restore the old policy */
3335 link->lpm_policy = old_policy;
3336 if (ap && ap->slave_link)
3337 ap->slave_link->lpm_policy = old_policy;
3339 /* if no device or only one more chance is left, disable LPM */
3340 if (!dev || ehc->tries[dev->devno] <= 2) {
3341 ata_link_warn(link, "disabling LPM on the link\n");
3342 link->flags |= ATA_LFLAG_NO_LPM;
3345 *r_failed_dev = dev;
3349 int ata_link_nr_enabled(struct ata_link *link)
3351 struct ata_device *dev;
3354 ata_for_each_dev(dev, link, ENABLED)
3359 static int ata_link_nr_vacant(struct ata_link *link)
3361 struct ata_device *dev;
3364 ata_for_each_dev(dev, link, ALL)
3365 if (dev->class == ATA_DEV_UNKNOWN)
3370 static int ata_eh_skip_recovery(struct ata_link *link)
3372 struct ata_port *ap = link->ap;
3373 struct ata_eh_context *ehc = &link->eh_context;
3374 struct ata_device *dev;
3376 /* skip disabled links */
3377 if (link->flags & ATA_LFLAG_DISABLED)
3380 /* skip if explicitly requested */
3381 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3384 /* thaw frozen port and recover failed devices */
3385 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3388 /* reset at least once if reset is requested */
3389 if ((ehc->i.action & ATA_EH_RESET) &&
3390 !(ehc->i.flags & ATA_EHI_DID_RESET))
3393 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3394 ata_for_each_dev(dev, link, ALL) {
3395 if (dev->class == ATA_DEV_UNKNOWN &&
3396 ehc->classes[dev->devno] != ATA_DEV_NONE)
3403 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3405 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3406 u64 now = get_jiffies_64();
3407 int *trials = void_arg;
3409 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3410 (ent->timestamp < now - min(now, interval)))
3417 static int ata_eh_schedule_probe(struct ata_device *dev)
3419 struct ata_eh_context *ehc = &dev->link->eh_context;
3420 struct ata_link *link = ata_dev_phys_link(dev);
3423 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3424 (ehc->did_probe_mask & (1 << dev->devno)))
3427 ata_eh_detach_dev(dev);
3429 ehc->did_probe_mask |= (1 << dev->devno);
3430 ehc->i.action |= ATA_EH_RESET;
3431 ehc->saved_xfer_mode[dev->devno] = 0;
3432 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3434 /* the link maybe in a deep sleep, wake it up */
3435 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3436 if (ata_is_host_link(link))
3437 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3440 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3444 /* Record and count probe trials on the ering. The specific
3445 * error mask used is irrelevant. Because a successful device
3446 * detection clears the ering, this count accumulates only if
3447 * there are consecutive failed probes.
3449 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3450 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3451 * forced to 1.5Gbps.
3453 * This is to work around cases where failed link speed
3454 * negotiation results in device misdetection leading to
3455 * infinite DEVXCHG or PHRDY CHG events.
3457 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3458 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3460 if (trials > ATA_EH_PROBE_TRIALS)
3461 sata_down_spd_limit(link, 1);
3466 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3468 struct ata_eh_context *ehc = &dev->link->eh_context;
3470 /* -EAGAIN from EH routine indicates retry without prejudice.
3471 * The requester is responsible for ensuring forward progress.
3474 ehc->tries[dev->devno]--;
3478 /* device missing or wrong IDENTIFY data, schedule probing */
3479 ehc->i.probe_mask |= (1 << dev->devno);
3482 /* give it just one more chance */
3483 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3486 if (ehc->tries[dev->devno] == 1) {
3487 /* This is the last chance, better to slow
3488 * down than lose it.
3490 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3491 if (dev->pio_mode > XFER_PIO_0)
3492 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3496 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3497 /* disable device if it has used up all its chances */
3498 ata_dev_disable(dev);
3500 /* detach if offline */
3501 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3502 ata_eh_detach_dev(dev);
3504 /* schedule probe if necessary */
3505 if (ata_eh_schedule_probe(dev)) {
3506 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3507 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3508 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3513 ehc->i.action |= ATA_EH_RESET;
3519 * ata_eh_recover - recover host port after error
3520 * @ap: host port to recover
3521 * @prereset: prereset method (can be NULL)
3522 * @softreset: softreset method (can be NULL)
3523 * @hardreset: hardreset method (can be NULL)
3524 * @postreset: postreset method (can be NULL)
3525 * @r_failed_link: out parameter for failed link
3527 * This is the alpha and omega, eum and yang, heart and soul of
3528 * libata exception handling. On entry, actions required to
3529 * recover each link and hotplug requests are recorded in the
3530 * link's eh_context. This function executes all the operations
3531 * with appropriate retrials and fallbacks to resurrect failed
3532 * devices, detach goners and greet newcomers.
3535 * Kernel thread context (may sleep).
3538 * 0 on success, -errno on failure.
3540 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3541 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3542 ata_postreset_fn_t postreset,
3543 struct ata_link **r_failed_link)
3545 struct ata_link *link;
3546 struct ata_device *dev;
3548 unsigned long flags, deadline;
3550 /* prep for recovery */
3551 ata_for_each_link(link, ap, EDGE) {
3552 struct ata_eh_context *ehc = &link->eh_context;
3554 /* re-enable link? */
3555 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3556 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3557 spin_lock_irqsave(ap->lock, flags);
3558 link->flags &= ~ATA_LFLAG_DISABLED;
3559 spin_unlock_irqrestore(ap->lock, flags);
3560 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3563 ata_for_each_dev(dev, link, ALL) {
3564 if (link->flags & ATA_LFLAG_NO_RETRY)
3565 ehc->tries[dev->devno] = 1;
3567 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3569 /* collect port action mask recorded in dev actions */
3570 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3571 ~ATA_EH_PERDEV_MASK;
3572 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3574 /* process hotplug request */
3575 if (dev->flags & ATA_DFLAG_DETACH)
3576 ata_eh_detach_dev(dev);
3578 /* schedule probe if necessary */
3579 if (!ata_dev_enabled(dev))
3580 ata_eh_schedule_probe(dev);
3587 /* if UNLOADING, finish immediately */
3588 if (ap->pflags & ATA_PFLAG_UNLOADING)
3592 ata_for_each_link(link, ap, EDGE) {
3593 struct ata_eh_context *ehc = &link->eh_context;
3595 /* skip EH if possible. */
3596 if (ata_eh_skip_recovery(link))
3599 ata_for_each_dev(dev, link, ALL)
3600 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3604 ata_for_each_link(link, ap, EDGE) {
3605 struct ata_eh_context *ehc = &link->eh_context;
3607 if (!(ehc->i.action & ATA_EH_RESET))
3610 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3611 prereset, softreset, hardreset, postreset);
3613 ata_link_err(link, "reset failed, giving up\n");
3622 * clears ATA_EH_PARK in eh_info and resets
3623 * ap->park_req_pending
3625 ata_eh_pull_park_action(ap);
3628 ata_for_each_link(link, ap, EDGE) {
3629 ata_for_each_dev(dev, link, ALL) {
3630 struct ata_eh_context *ehc = &link->eh_context;
3633 if (dev->class != ATA_DEV_ATA &&
3634 dev->class != ATA_DEV_ZAC)
3636 if (!(ehc->i.dev_action[dev->devno] &
3639 tmp = dev->unpark_deadline;
3640 if (time_before(deadline, tmp))
3642 else if (time_before_eq(tmp, jiffies))
3644 if (ehc->unloaded_mask & (1 << dev->devno))
3647 ata_eh_park_issue_cmd(dev, 1);
3652 if (time_before_eq(deadline, now))
3656 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3660 ata_for_each_link(link, ap, EDGE) {
3661 ata_for_each_dev(dev, link, ALL) {
3662 if (!(link->eh_context.unloaded_mask &
3666 ata_eh_park_issue_cmd(dev, 0);
3667 ata_eh_done(link, dev, ATA_EH_PARK);
3673 ata_for_each_link(link, ap, PMP_FIRST) {
3674 struct ata_eh_context *ehc = &link->eh_context;
3676 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3679 /* revalidate existing devices and attach new ones */
3680 rc = ata_eh_revalidate_and_attach(link, &dev);
3684 /* if PMP got attached, return, pmp EH will take care of it */
3685 if (link->device->class == ATA_DEV_PMP) {
3690 /* configure transfer mode if necessary */
3691 if (ehc->i.flags & ATA_EHI_SETMODE) {
3692 rc = ata_set_mode(link, &dev);
3695 ehc->i.flags &= ~ATA_EHI_SETMODE;
3698 /* If reset has been issued, clear UA to avoid
3699 * disrupting the current users of the device.
3701 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3702 ata_for_each_dev(dev, link, ALL) {
3703 if (dev->class != ATA_DEV_ATAPI)
3705 rc = atapi_eh_clear_ua(dev);
3708 if (zpodd_dev_enabled(dev))
3709 zpodd_post_poweron(dev);
3713 /* retry flush if necessary */
3714 ata_for_each_dev(dev, link, ALL) {
3715 if (dev->class != ATA_DEV_ATA &&
3716 dev->class != ATA_DEV_ZAC)
3718 rc = ata_eh_maybe_retry_flush(dev);
3724 /* configure link power saving */
3725 if (link->lpm_policy != ap->target_lpm_policy) {
3726 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3731 /* this link is okay now */
3738 ata_eh_handle_dev_fail(dev, rc);
3740 if (ap->pflags & ATA_PFLAG_FROZEN) {
3741 /* PMP reset requires working host port.
3742 * Can't retry if it's frozen.
3744 if (sata_pmp_attached(ap))
3754 if (rc && r_failed_link)
3755 *r_failed_link = link;
3761 * ata_eh_finish - finish up EH
3762 * @ap: host port to finish EH for
3764 * Recovery is complete. Clean up EH states and retry or finish
3770 void ata_eh_finish(struct ata_port *ap)
3772 struct ata_queued_cmd *qc;
3775 /* retry or finish qcs */
3776 ata_qc_for_each_raw(ap, qc, tag) {
3777 if (!(qc->flags & ATA_QCFLAG_FAILED))
3781 /* FIXME: Once EH migration is complete,
3782 * generate sense data in this function,
3783 * considering both err_mask and tf.
3785 if (qc->flags & ATA_QCFLAG_RETRY)
3786 ata_eh_qc_retry(qc);
3788 ata_eh_qc_complete(qc);
3790 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3791 ata_eh_qc_complete(qc);
3793 /* feed zero TF to sense generation */
3794 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3795 ata_eh_qc_retry(qc);
3800 /* make sure nr_active_links is zero after EH */
3801 WARN_ON(ap->nr_active_links);
3802 ap->nr_active_links = 0;
3806 * ata_do_eh - do standard error handling
3807 * @ap: host port to handle error for
3809 * @prereset: prereset method (can be NULL)
3810 * @softreset: softreset method (can be NULL)
3811 * @hardreset: hardreset method (can be NULL)
3812 * @postreset: postreset method (can be NULL)
3814 * Perform standard error handling sequence.
3817 * Kernel thread context (may sleep).
3819 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3820 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3821 ata_postreset_fn_t postreset)
3823 struct ata_device *dev;
3829 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3832 ata_for_each_dev(dev, &ap->link, ALL)
3833 ata_dev_disable(dev);
3840 * ata_std_error_handler - standard error handler
3841 * @ap: host port to handle error for
3843 * Standard error handler
3846 * Kernel thread context (may sleep).
3848 void ata_std_error_handler(struct ata_port *ap)
3850 struct ata_port_operations *ops = ap->ops;
3851 ata_reset_fn_t hardreset = ops->hardreset;
3853 /* ignore built-in hardreset if SCR access is not available */
3854 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3857 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3859 EXPORT_SYMBOL_GPL(ata_std_error_handler);
3863 * ata_eh_handle_port_suspend - perform port suspend operation
3864 * @ap: port to suspend
3869 * Kernel thread context (may sleep).
3871 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3873 unsigned long flags;
3875 struct ata_device *dev;
3877 /* are we suspending? */
3878 spin_lock_irqsave(ap->lock, flags);
3879 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3880 ap->pm_mesg.event & PM_EVENT_RESUME) {
3881 spin_unlock_irqrestore(ap->lock, flags);
3884 spin_unlock_irqrestore(ap->lock, flags);
3886 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3889 * If we have a ZPODD attached, check its zero
3890 * power ready status before the port is frozen.
3891 * Only needed for runtime suspend.
3893 if (PMSG_IS_AUTO(ap->pm_mesg)) {
3894 ata_for_each_dev(dev, &ap->link, ENABLED) {
3895 if (zpodd_dev_enabled(dev))
3896 zpodd_on_suspend(dev);
3901 ata_eh_freeze_port(ap);
3903 if (ap->ops->port_suspend)
3904 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3906 ata_acpi_set_state(ap, ap->pm_mesg);
3908 /* update the flags */
3909 spin_lock_irqsave(ap->lock, flags);
3911 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3913 ap->pflags |= ATA_PFLAG_SUSPENDED;
3914 else if (ap->pflags & ATA_PFLAG_FROZEN)
3915 ata_port_schedule_eh(ap);
3917 spin_unlock_irqrestore(ap->lock, flags);
3923 * ata_eh_handle_port_resume - perform port resume operation
3924 * @ap: port to resume
3929 * Kernel thread context (may sleep).
3931 static void ata_eh_handle_port_resume(struct ata_port *ap)
3933 struct ata_link *link;
3934 struct ata_device *dev;
3935 unsigned long flags;
3937 /* are we resuming? */
3938 spin_lock_irqsave(ap->lock, flags);
3939 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3940 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
3941 spin_unlock_irqrestore(ap->lock, flags);
3944 spin_unlock_irqrestore(ap->lock, flags);
3946 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3949 * Error timestamps are in jiffies which doesn't run while
3950 * suspended and PHY events during resume isn't too uncommon.
3951 * When the two are combined, it can lead to unnecessary speed
3952 * downs if the machine is suspended and resumed repeatedly.
3953 * Clear error history.
3955 ata_for_each_link(link, ap, HOST_FIRST)
3956 ata_for_each_dev(dev, link, ALL)
3957 ata_ering_clear(&dev->ering);
3959 ata_acpi_set_state(ap, ap->pm_mesg);
3961 if (ap->ops->port_resume)
3962 ap->ops->port_resume(ap);
3964 /* tell ACPI that we're resuming */
3965 ata_acpi_on_resume(ap);
3967 /* update the flags */
3968 spin_lock_irqsave(ap->lock, flags);
3969 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3970 spin_unlock_irqrestore(ap->lock, flags);
3972 #endif /* CONFIG_PM */