2 * libata-eh.c - libata error handling
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/driver-api/libata.rst
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/export.h>
38 #include <linux/pci.h>
39 #include <scsi/scsi.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_cmnd.h>
44 #include <scsi/scsi_dbg.h>
45 #include "../scsi/scsi_transport_api.h"
47 #include <linux/libata.h>
49 #include <trace/events/libata.h>
53 /* speed down verdicts */
54 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
55 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
56 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
57 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
60 ATA_EFLAG_IS_IO = (1 << 0),
61 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
62 ATA_EFLAG_OLD_ER = (1 << 31),
64 /* error categories */
67 ATA_ECAT_TOUT_HSM = 2,
69 ATA_ECAT_DUBIOUS_NONE = 4,
70 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
71 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
72 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
75 ATA_EH_CMD_DFL_TIMEOUT = 5000,
77 /* always put at least this amount of time between resets */
78 ATA_EH_RESET_COOL_DOWN = 5000,
80 /* Waiting in ->prereset can never be reliable. It's
81 * sometimes nice to wait there but it can't be depended upon;
82 * otherwise, we wouldn't be resetting. Just give it enough
83 * time for most drives to spin up.
85 ATA_EH_PRERESET_TIMEOUT = 10000,
86 ATA_EH_FASTDRAIN_INTERVAL = 3000,
90 /* probe speed down parameters, see ata_eh_schedule_probe() */
91 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
92 ATA_EH_PROBE_TRIALS = 2,
95 /* The following table determines how we sequence resets. Each entry
96 * represents timeout for that try. The first try can be soft or
97 * hardreset. All others are hardreset if available. In most cases
98 * the first reset w/ 10sec timeout should succeed. Following entries
99 * are mostly for error handling, hotplug and those outlier devices that
100 * take an exceptionally long time to recover from reset.
102 static const unsigned long ata_eh_reset_timeouts[] = {
103 10000, /* most drives spin up by 10sec */
104 10000, /* > 99% working drives spin up before 20sec */
105 35000, /* give > 30 secs of idleness for outlier devices */
106 5000, /* and sweet one last chance */
107 ULONG_MAX, /* > 1 min has elapsed, give up */
110 static const unsigned long ata_eh_identify_timeouts[] = {
111 5000, /* covers > 99% of successes and not too boring on failures */
112 10000, /* combined time till here is enough even for media access */
113 30000, /* for true idiots */
117 static const unsigned long ata_eh_flush_timeouts[] = {
118 15000, /* be generous with flush */
120 30000, /* and even more generous */
124 static const unsigned long ata_eh_other_timeouts[] = {
125 5000, /* same rationale as identify timeout */
127 /* but no merciful 30sec for other commands, it just isn't worth it */
131 struct ata_eh_cmd_timeout_ent {
133 const unsigned long *timeouts;
136 /* The following table determines timeouts to use for EH internal
137 * commands. Each table entry is a command class and matches the
138 * commands the entry applies to and the timeout table to use.
140 * On the retry after a command timed out, the next timeout value from
141 * the table is used. If the table doesn't contain further entries,
142 * the last value is used.
144 * ehc->cmd_timeout_idx keeps track of which timeout to use per
145 * command class, so if SET_FEATURES times out on the first try, the
146 * next try will use the second timeout value only for that class.
148 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
149 static const struct ata_eh_cmd_timeout_ent
150 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
151 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
152 .timeouts = ata_eh_identify_timeouts, },
153 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_SET_FEATURES),
158 .timeouts = ata_eh_other_timeouts, },
159 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
160 .timeouts = ata_eh_other_timeouts, },
161 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
162 .timeouts = ata_eh_flush_timeouts },
166 static void __ata_port_freeze(struct ata_port *ap);
168 static void ata_eh_handle_port_suspend(struct ata_port *ap);
169 static void ata_eh_handle_port_resume(struct ata_port *ap);
170 #else /* CONFIG_PM */
171 static void ata_eh_handle_port_suspend(struct ata_port *ap)
174 static void ata_eh_handle_port_resume(struct ata_port *ap)
176 #endif /* CONFIG_PM */
178 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
179 const char *fmt, va_list args)
181 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
182 ATA_EH_DESC_LEN - ehi->desc_len,
187 * __ata_ehi_push_desc - push error description without adding separator
189 * @fmt: printf format string
191 * Format string according to @fmt and append it to @ehi->desc.
194 * spin_lock_irqsave(host lock)
196 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
201 __ata_ehi_pushv_desc(ehi, fmt, args);
206 * ata_ehi_push_desc - push error description with separator
208 * @fmt: printf format string
210 * Format string according to @fmt and append it to @ehi->desc.
211 * If @ehi->desc is not empty, ", " is added in-between.
214 * spin_lock_irqsave(host lock)
216 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
221 __ata_ehi_push_desc(ehi, ", ");
224 __ata_ehi_pushv_desc(ehi, fmt, args);
229 * ata_ehi_clear_desc - clean error description
235 * spin_lock_irqsave(host lock)
237 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
256 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
260 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
262 if (ap->link.eh_info.desc_len)
263 __ata_ehi_push_desc(&ap->link.eh_info, " ");
266 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
287 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
290 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
292 unsigned long long start, len;
294 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
296 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
299 start = (unsigned long long)pci_resource_start(pdev, bar);
300 len = (unsigned long long)pci_resource_len(pdev, bar);
303 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
305 ata_port_desc(ap, "%s 0x%llx", name,
306 start + (unsigned long long)offset);
309 #endif /* CONFIG_PCI */
311 static int ata_lookup_timeout_table(u8 cmd)
315 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
318 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
331 * Determine timeout for internal command @cmd for @dev.
337 * Determined timeout.
339 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
341 struct ata_eh_context *ehc = &dev->link->eh_context;
342 int ent = ata_lookup_timeout_table(cmd);
346 return ATA_EH_CMD_DFL_TIMEOUT;
348 idx = ehc->cmd_timeout_idx[dev->devno][ent];
349 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
364 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
366 struct ata_eh_context *ehc = &dev->link->eh_context;
367 int ent = ata_lookup_timeout_table(cmd);
373 idx = ehc->cmd_timeout_idx[dev->devno][ent];
374 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
375 ehc->cmd_timeout_idx[dev->devno][ent]++;
378 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
379 unsigned int err_mask)
381 struct ata_ering_entry *ent;
386 ering->cursor %= ATA_ERING_SIZE;
388 ent = &ering->ring[ering->cursor];
389 ent->eflags = eflags;
390 ent->err_mask = err_mask;
391 ent->timestamp = get_jiffies_64();
394 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
396 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
403 int ata_ering_map(struct ata_ering *ering,
404 int (*map_fn)(struct ata_ering_entry *, void *),
408 struct ata_ering_entry *ent;
412 ent = &ering->ring[idx];
415 rc = map_fn(ent, arg);
418 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
419 } while (idx != ering->cursor);
424 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
426 ent->eflags |= ATA_EFLAG_OLD_ER;
430 static void ata_ering_clear(struct ata_ering *ering)
432 ata_ering_map(ering, ata_ering_clear_cb, NULL);
435 static unsigned int ata_eh_dev_action(struct ata_device *dev)
437 struct ata_eh_context *ehc = &dev->link->eh_context;
439 return ehc->i.action | ehc->i.dev_action[dev->devno];
442 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
443 struct ata_eh_info *ehi, unsigned int action)
445 struct ata_device *tdev;
448 ehi->action &= ~action;
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] &= ~action;
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
455 /* break ehi->action into ehi->dev_action */
456 if (ehi->action & action) {
457 ata_for_each_dev(tdev, link, ALL)
458 ehi->dev_action[tdev->devno] |=
459 ehi->action & action;
460 ehi->action &= ~action;
463 /* turn off the specified per-dev action */
464 ehi->dev_action[dev->devno] &= ~action;
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
479 void ata_eh_acquire(struct ata_port *ap)
481 mutex_lock(&ap->host->eh_mutex);
482 WARN_ON_ONCE(ap->host->eh_owner);
483 ap->host->eh_owner = current;
487 * ata_eh_release - release EH ownership
488 * @ap: ATA port to release EH ownership for
490 * Release EH ownership for @ap if the caller. The caller must
491 * have acquired EH ownership using ata_eh_acquire() previously.
496 void ata_eh_release(struct ata_port *ap)
498 WARN_ON_ONCE(ap->host->eh_owner != current);
499 ap->host->eh_owner = NULL;
500 mutex_unlock(&ap->host->eh_mutex);
503 static void ata_eh_unload(struct ata_port *ap)
505 struct ata_link *link;
506 struct ata_device *dev;
509 /* Restore SControl IPM and SPD for the next driver and
510 * disable attached devices.
512 ata_for_each_link(link, ap, PMP_FIRST) {
513 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
514 ata_for_each_dev(dev, link, ALL)
515 ata_dev_disable(dev);
518 /* freeze and set UNLOADED */
519 spin_lock_irqsave(ap->lock, flags);
521 ata_port_freeze(ap); /* won't be thawed */
522 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
523 ap->pflags |= ATA_PFLAG_UNLOADED;
525 spin_unlock_irqrestore(ap->lock, flags);
529 * ata_scsi_error - SCSI layer error handler callback
530 * @host: SCSI host on which error occurred
532 * Handles SCSI-layer-thrown error events.
535 * Inherited from SCSI layer (none, can sleep)
540 void ata_scsi_error(struct Scsi_Host *host)
542 struct ata_port *ap = ata_shost_to_port(host);
544 LIST_HEAD(eh_work_q);
548 spin_lock_irqsave(host->host_lock, flags);
549 list_splice_init(&host->eh_cmd_q, &eh_work_q);
550 spin_unlock_irqrestore(host->host_lock, flags);
552 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
554 /* If we timed raced normal completion and there is nothing to
555 recover nr_timedout == 0 why exactly are we doing error recovery ? */
556 ata_scsi_port_error_handler(host, ap);
558 /* finish or retry handled scmd's and clean up */
559 WARN_ON(!list_empty(&eh_work_q));
565 * ata_scsi_cmd_error_handler - error callback for a list of commands
566 * @host: scsi host containing the port
567 * @ap: ATA port within the host
568 * @eh_work_q: list of commands to process
570 * process the given list of commands and return those finished to the
571 * ap->eh_done_q. This function is the first part of the libata error
572 * handler which processes a given list of failed commands.
574 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
575 struct list_head *eh_work_q)
580 /* make sure sff pio task is not running */
581 ata_sff_flush_pio_task(ap);
583 /* synchronize with host lock and sort out timeouts */
585 /* For new EH, all qcs are finished in one of three ways -
586 * normal completion, error completion, and SCSI timeout.
587 * Both completions can race against SCSI timeout. When normal
588 * completion wins, the qc never reaches EH. When error
589 * completion wins, the qc has ATA_QCFLAG_FAILED set.
591 * When SCSI timeout wins, things are a bit more complex.
592 * Normal or error completion can occur after the timeout but
593 * before this point. In such cases, both types of
594 * completions are honored. A scmd is determined to have
595 * timed out iff its associated qc is active and not failed.
597 spin_lock_irqsave(ap->lock, flags);
598 if (ap->ops->error_handler) {
599 struct scsi_cmnd *scmd, *tmp;
602 /* This must occur under the ap->lock as we don't want
603 a polled recovery to race the real interrupt handler
605 The lost_interrupt handler checks for any completed but
606 non-notified command and completes much like an IRQ handler.
608 We then fall into the error recovery code which will treat
609 this as if normal completion won the race */
611 if (ap->ops->lost_interrupt)
612 ap->ops->lost_interrupt(ap);
614 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
615 struct ata_queued_cmd *qc;
617 for (i = 0; i < ATA_MAX_QUEUE; i++) {
618 qc = __ata_qc_from_tag(ap, i);
619 if (qc->flags & ATA_QCFLAG_ACTIVE &&
624 if (i < ATA_MAX_QUEUE) {
625 /* the scmd has an associated qc */
626 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
627 /* which hasn't failed yet, timeout */
628 qc->err_mask |= AC_ERR_TIMEOUT;
629 qc->flags |= ATA_QCFLAG_FAILED;
633 /* Normal completion occurred after
634 * SCSI timeout but before this point.
635 * Successfully complete it.
637 scmd->retries = scmd->allowed;
638 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
642 /* If we have timed out qcs. They belong to EH from
643 * this point but the state of the controller is
644 * unknown. Freeze the port to make sure the IRQ
645 * handler doesn't diddle with those qcs. This must
646 * be done atomically w.r.t. setting QCFLAG_FAILED.
649 __ata_port_freeze(ap);
652 /* initialize eh_tries */
653 ap->eh_tries = ATA_EH_MAX_TRIES;
655 spin_unlock_irqrestore(ap->lock, flags);
658 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
661 * ata_scsi_port_error_handler - recover the port after the commands
662 * @host: SCSI host containing the port
665 * Handle the recovery of the port @ap after all the commands
666 * have been recovered.
668 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
672 /* invoke error handler */
673 if (ap->ops->error_handler) {
674 struct ata_link *link;
676 /* acquire EH ownership */
679 /* kill fast drain timer */
680 del_timer_sync(&ap->fastdrain_timer);
682 /* process port resume request */
683 ata_eh_handle_port_resume(ap);
685 /* fetch & clear EH info */
686 spin_lock_irqsave(ap->lock, flags);
688 ata_for_each_link(link, ap, HOST_FIRST) {
689 struct ata_eh_context *ehc = &link->eh_context;
690 struct ata_device *dev;
692 memset(&link->eh_context, 0, sizeof(link->eh_context));
693 link->eh_context.i = link->eh_info;
694 memset(&link->eh_info, 0, sizeof(link->eh_info));
696 ata_for_each_dev(dev, link, ENABLED) {
697 int devno = dev->devno;
699 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
700 if (ata_ncq_enabled(dev))
701 ehc->saved_ncq_enabled |= 1 << devno;
705 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
706 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
707 ap->excl_link = NULL; /* don't maintain exclusion over EH */
709 spin_unlock_irqrestore(ap->lock, flags);
711 /* invoke EH, skip if unloading or suspended */
712 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
713 ap->ops->error_handler(ap);
715 /* if unloading, commence suicide */
716 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
717 !(ap->pflags & ATA_PFLAG_UNLOADED))
722 /* process port suspend request */
723 ata_eh_handle_port_suspend(ap);
725 /* Exception might have happened after ->error_handler
726 * recovered the port but before this point. Repeat
729 spin_lock_irqsave(ap->lock, flags);
731 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
732 if (--ap->eh_tries) {
733 spin_unlock_irqrestore(ap->lock, flags);
737 "EH pending after %d tries, giving up\n",
739 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
742 /* this run is complete, make sure EH info is clear */
743 ata_for_each_link(link, ap, HOST_FIRST)
744 memset(&link->eh_info, 0, sizeof(link->eh_info));
746 /* end eh (clear host_eh_scheduled) while holding
747 * ap->lock such that if exception occurs after this
748 * point but before EH completion, SCSI midlayer will
753 spin_unlock_irqrestore(ap->lock, flags);
756 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
757 ap->ops->eng_timeout(ap);
760 scsi_eh_flush_done_q(&ap->eh_done_q);
763 spin_lock_irqsave(ap->lock, flags);
765 if (ap->pflags & ATA_PFLAG_LOADING)
766 ap->pflags &= ~ATA_PFLAG_LOADING;
767 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
768 !(ap->flags & ATA_FLAG_SAS_HOST))
769 schedule_delayed_work(&ap->hotplug_task, 0);
771 if (ap->pflags & ATA_PFLAG_RECOVERED)
772 ata_port_info(ap, "EH complete\n");
774 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
776 /* tell wait_eh that we're done */
777 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
778 wake_up_all(&ap->eh_wait_q);
780 spin_unlock_irqrestore(ap->lock, flags);
782 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
785 * ata_port_wait_eh - Wait for the currently pending EH to complete
786 * @ap: Port to wait EH for
788 * Wait until the currently pending EH is complete.
791 * Kernel thread context (may sleep).
793 void ata_port_wait_eh(struct ata_port *ap)
799 spin_lock_irqsave(ap->lock, flags);
801 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
802 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
803 spin_unlock_irqrestore(ap->lock, flags);
805 spin_lock_irqsave(ap->lock, flags);
807 finish_wait(&ap->eh_wait_q, &wait);
809 spin_unlock_irqrestore(ap->lock, flags);
811 /* make sure SCSI EH is complete */
812 if (scsi_host_in_recovery(ap->scsi_host)) {
817 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
819 static int ata_eh_nr_in_flight(struct ata_port *ap)
824 /* count only non-internal commands */
825 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
826 if (ata_tag_internal(tag))
828 if (ata_qc_from_tag(ap, tag))
835 void ata_eh_fastdrain_timerfn(struct timer_list *t)
837 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
841 spin_lock_irqsave(ap->lock, flags);
843 cnt = ata_eh_nr_in_flight(ap);
849 if (cnt == ap->fastdrain_cnt) {
852 /* No progress during the last interval, tag all
853 * in-flight qcs as timed out and freeze the port.
855 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
856 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
858 qc->err_mask |= AC_ERR_TIMEOUT;
863 /* some qcs have finished, give it another chance */
864 ap->fastdrain_cnt = cnt;
865 ap->fastdrain_timer.expires =
866 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
867 add_timer(&ap->fastdrain_timer);
871 spin_unlock_irqrestore(ap->lock, flags);
875 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
876 * @ap: target ATA port
877 * @fastdrain: activate fast drain
879 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
880 * is non-zero and EH wasn't pending before. Fast drain ensures
881 * that EH kicks in in timely manner.
884 * spin_lock_irqsave(host lock)
886 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
890 /* already scheduled? */
891 if (ap->pflags & ATA_PFLAG_EH_PENDING)
894 ap->pflags |= ATA_PFLAG_EH_PENDING;
899 /* do we have in-flight qcs? */
900 cnt = ata_eh_nr_in_flight(ap);
904 /* activate fast drain */
905 ap->fastdrain_cnt = cnt;
906 ap->fastdrain_timer.expires =
907 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
908 add_timer(&ap->fastdrain_timer);
912 * ata_qc_schedule_eh - schedule qc for error handling
913 * @qc: command to schedule error handling for
915 * Schedule error handling for @qc. EH will kick in as soon as
916 * other commands are drained.
919 * spin_lock_irqsave(host lock)
921 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
923 struct ata_port *ap = qc->ap;
924 struct request_queue *q = qc->scsicmd->device->request_queue;
927 WARN_ON(!ap->ops->error_handler);
929 qc->flags |= ATA_QCFLAG_FAILED;
930 ata_eh_set_pending(ap, 1);
932 /* The following will fail if timeout has already expired.
933 * ata_scsi_error() takes care of such scmds on EH entry.
934 * Note that ATA_QCFLAG_FAILED is unconditionally set after
935 * this function completes.
937 spin_lock_irqsave(q->queue_lock, flags);
938 blk_abort_request(qc->scsicmd->request);
939 spin_unlock_irqrestore(q->queue_lock, flags);
943 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
944 * @ap: ATA port to schedule EH for
946 * LOCKING: inherited from ata_port_schedule_eh
947 * spin_lock_irqsave(host lock)
949 void ata_std_sched_eh(struct ata_port *ap)
951 WARN_ON(!ap->ops->error_handler);
953 if (ap->pflags & ATA_PFLAG_INITIALIZING)
956 ata_eh_set_pending(ap, 1);
957 scsi_schedule_eh(ap->scsi_host);
959 DPRINTK("port EH scheduled\n");
961 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
964 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
965 * @ap: ATA port to end EH for
967 * In the libata object model there is a 1:1 mapping of ata_port to
968 * shost, so host fields can be directly manipulated under ap->lock, in
969 * the libsas case we need to hold a lock at the ha->level to coordinate
973 * spin_lock_irqsave(host lock)
975 void ata_std_end_eh(struct ata_port *ap)
977 struct Scsi_Host *host = ap->scsi_host;
979 host->host_eh_scheduled = 0;
981 EXPORT_SYMBOL(ata_std_end_eh);
985 * ata_port_schedule_eh - schedule error handling without a qc
986 * @ap: ATA port to schedule EH for
988 * Schedule error handling for @ap. EH will kick in as soon as
989 * all commands are drained.
992 * spin_lock_irqsave(host lock)
994 void ata_port_schedule_eh(struct ata_port *ap)
996 /* see: ata_std_sched_eh, unless you know better */
997 ap->ops->sched_eh(ap);
1000 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1002 int tag, nr_aborted = 0;
1004 WARN_ON(!ap->ops->error_handler);
1006 /* we're gonna abort all commands, no need for fast drain */
1007 ata_eh_set_pending(ap, 0);
1009 /* include internal tag in iteration */
1010 for (tag = 0; tag <= ATA_MAX_QUEUE; tag++) {
1011 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1013 if (qc && (!link || qc->dev->link == link)) {
1014 qc->flags |= ATA_QCFLAG_FAILED;
1015 ata_qc_complete(qc);
1021 ata_port_schedule_eh(ap);
1027 * ata_link_abort - abort all qc's on the link
1028 * @link: ATA link to abort qc's for
1030 * Abort all active qc's active on @link and schedule EH.
1033 * spin_lock_irqsave(host lock)
1036 * Number of aborted qc's.
1038 int ata_link_abort(struct ata_link *link)
1040 return ata_do_link_abort(link->ap, link);
1044 * ata_port_abort - abort all qc's on the port
1045 * @ap: ATA port to abort qc's for
1047 * Abort all active qc's of @ap and schedule EH.
1050 * spin_lock_irqsave(host_set lock)
1053 * Number of aborted qc's.
1055 int ata_port_abort(struct ata_port *ap)
1057 return ata_do_link_abort(ap, NULL);
1061 * __ata_port_freeze - freeze port
1062 * @ap: ATA port to freeze
1064 * This function is called when HSM violation or some other
1065 * condition disrupts normal operation of the port. Frozen port
1066 * is not allowed to perform any operation until the port is
1067 * thawed, which usually follows a successful reset.
1069 * ap->ops->freeze() callback can be used for freezing the port
1070 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1071 * port cannot be frozen hardware-wise, the interrupt handler
1072 * must ack and clear interrupts unconditionally while the port
1076 * spin_lock_irqsave(host lock)
1078 static void __ata_port_freeze(struct ata_port *ap)
1080 WARN_ON(!ap->ops->error_handler);
1082 if (ap->ops->freeze)
1083 ap->ops->freeze(ap);
1085 ap->pflags |= ATA_PFLAG_FROZEN;
1087 DPRINTK("ata%u port frozen\n", ap->print_id);
1091 * ata_port_freeze - abort & freeze port
1092 * @ap: ATA port to freeze
1094 * Abort and freeze @ap. The freeze operation must be called
1095 * first, because some hardware requires special operations
1096 * before the taskfile registers are accessible.
1099 * spin_lock_irqsave(host lock)
1102 * Number of aborted commands.
1104 int ata_port_freeze(struct ata_port *ap)
1108 WARN_ON(!ap->ops->error_handler);
1110 __ata_port_freeze(ap);
1111 nr_aborted = ata_port_abort(ap);
1117 * sata_async_notification - SATA async notification handler
1118 * @ap: ATA port where async notification is received
1120 * Handler to be called when async notification via SDB FIS is
1121 * received. This function schedules EH if necessary.
1124 * spin_lock_irqsave(host lock)
1127 * 1 if EH is scheduled, 0 otherwise.
1129 int sata_async_notification(struct ata_port *ap)
1134 if (!(ap->flags & ATA_FLAG_AN))
1137 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1139 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1141 if (!sata_pmp_attached(ap) || rc) {
1142 /* PMP is not attached or SNTF is not available */
1143 if (!sata_pmp_attached(ap)) {
1144 /* PMP is not attached. Check whether ATAPI
1145 * AN is configured. If so, notify media
1148 struct ata_device *dev = ap->link.device;
1150 if ((dev->class == ATA_DEV_ATAPI) &&
1151 (dev->flags & ATA_DFLAG_AN))
1152 ata_scsi_media_change_notify(dev);
1155 /* PMP is attached but SNTF is not available.
1156 * ATAPI async media change notification is
1157 * not used. The PMP must be reporting PHY
1158 * status change, schedule EH.
1160 ata_port_schedule_eh(ap);
1164 /* PMP is attached and SNTF is available */
1165 struct ata_link *link;
1167 /* check and notify ATAPI AN */
1168 ata_for_each_link(link, ap, EDGE) {
1169 if (!(sntf & (1 << link->pmp)))
1172 if ((link->device->class == ATA_DEV_ATAPI) &&
1173 (link->device->flags & ATA_DFLAG_AN))
1174 ata_scsi_media_change_notify(link->device);
1177 /* If PMP is reporting that PHY status of some
1178 * downstream ports has changed, schedule EH.
1180 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1181 ata_port_schedule_eh(ap);
1190 * ata_eh_freeze_port - EH helper to freeze port
1191 * @ap: ATA port to freeze
1198 void ata_eh_freeze_port(struct ata_port *ap)
1200 unsigned long flags;
1202 if (!ap->ops->error_handler)
1205 spin_lock_irqsave(ap->lock, flags);
1206 __ata_port_freeze(ap);
1207 spin_unlock_irqrestore(ap->lock, flags);
1211 * ata_port_thaw_port - EH helper to thaw port
1212 * @ap: ATA port to thaw
1214 * Thaw frozen port @ap.
1219 void ata_eh_thaw_port(struct ata_port *ap)
1221 unsigned long flags;
1223 if (!ap->ops->error_handler)
1226 spin_lock_irqsave(ap->lock, flags);
1228 ap->pflags &= ~ATA_PFLAG_FROZEN;
1233 spin_unlock_irqrestore(ap->lock, flags);
1235 DPRINTK("ata%u port thawed\n", ap->print_id);
1238 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1243 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1245 struct ata_port *ap = qc->ap;
1246 struct scsi_cmnd *scmd = qc->scsicmd;
1247 unsigned long flags;
1249 spin_lock_irqsave(ap->lock, flags);
1250 qc->scsidone = ata_eh_scsidone;
1251 __ata_qc_complete(qc);
1252 WARN_ON(ata_tag_valid(qc->tag));
1253 spin_unlock_irqrestore(ap->lock, flags);
1255 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1259 * ata_eh_qc_complete - Complete an active ATA command from EH
1260 * @qc: Command to complete
1262 * Indicate to the mid and upper layers that an ATA command has
1263 * completed. To be used from EH.
1265 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1267 struct scsi_cmnd *scmd = qc->scsicmd;
1268 scmd->retries = scmd->allowed;
1269 __ata_eh_qc_complete(qc);
1273 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1274 * @qc: Command to retry
1276 * Indicate to the mid and upper layers that an ATA command
1277 * should be retried. To be used from EH.
1279 * SCSI midlayer limits the number of retries to scmd->allowed.
1280 * scmd->allowed is incremented for commands which get retried
1281 * due to unrelated failures (qc->err_mask is zero).
1283 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1285 struct scsi_cmnd *scmd = qc->scsicmd;
1288 __ata_eh_qc_complete(qc);
1292 * ata_dev_disable - disable ATA device
1293 * @dev: ATA device to disable
1300 void ata_dev_disable(struct ata_device *dev)
1302 if (!ata_dev_enabled(dev))
1305 if (ata_msg_drv(dev->link->ap))
1306 ata_dev_warn(dev, "disabled\n");
1307 ata_acpi_on_disable(dev);
1308 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1311 /* From now till the next successful probe, ering is used to
1312 * track probe failures. Clear accumulated device error info.
1314 ata_ering_clear(&dev->ering);
1318 * ata_eh_detach_dev - detach ATA device
1319 * @dev: ATA device to detach
1326 void ata_eh_detach_dev(struct ata_device *dev)
1328 struct ata_link *link = dev->link;
1329 struct ata_port *ap = link->ap;
1330 struct ata_eh_context *ehc = &link->eh_context;
1331 unsigned long flags;
1333 ata_dev_disable(dev);
1335 spin_lock_irqsave(ap->lock, flags);
1337 dev->flags &= ~ATA_DFLAG_DETACH;
1339 if (ata_scsi_offline_dev(dev)) {
1340 dev->flags |= ATA_DFLAG_DETACHED;
1341 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1344 /* clear per-dev EH info */
1345 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1346 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1347 ehc->saved_xfer_mode[dev->devno] = 0;
1348 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1350 spin_unlock_irqrestore(ap->lock, flags);
1354 * ata_eh_about_to_do - about to perform eh_action
1355 * @link: target ATA link
1356 * @dev: target ATA dev for per-dev action (can be NULL)
1357 * @action: action about to be performed
1359 * Called just before performing EH actions to clear related bits
1360 * in @link->eh_info such that eh actions are not unnecessarily
1366 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1367 unsigned int action)
1369 struct ata_port *ap = link->ap;
1370 struct ata_eh_info *ehi = &link->eh_info;
1371 struct ata_eh_context *ehc = &link->eh_context;
1372 unsigned long flags;
1374 spin_lock_irqsave(ap->lock, flags);
1376 ata_eh_clear_action(link, dev, ehi, action);
1378 /* About to take EH action, set RECOVERED. Ignore actions on
1379 * slave links as master will do them again.
1381 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1382 ap->pflags |= ATA_PFLAG_RECOVERED;
1384 spin_unlock_irqrestore(ap->lock, flags);
1388 * ata_eh_done - EH action complete
1389 * @link: ATA link for which EH actions are complete
1390 * @dev: target ATA dev for per-dev action (can be NULL)
1391 * @action: action just completed
1393 * Called right after performing EH actions to clear related bits
1394 * in @link->eh_context.
1399 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1400 unsigned int action)
1402 struct ata_eh_context *ehc = &link->eh_context;
1404 ata_eh_clear_action(link, dev, &ehc->i, action);
1408 * ata_err_string - convert err_mask to descriptive string
1409 * @err_mask: error mask to convert to string
1411 * Convert @err_mask to descriptive string. Errors are
1412 * prioritized according to severity and only the most severe
1413 * error is reported.
1419 * Descriptive string for @err_mask
1421 static const char *ata_err_string(unsigned int err_mask)
1423 if (err_mask & AC_ERR_HOST_BUS)
1424 return "host bus error";
1425 if (err_mask & AC_ERR_ATA_BUS)
1426 return "ATA bus error";
1427 if (err_mask & AC_ERR_TIMEOUT)
1429 if (err_mask & AC_ERR_HSM)
1430 return "HSM violation";
1431 if (err_mask & AC_ERR_SYSTEM)
1432 return "internal error";
1433 if (err_mask & AC_ERR_MEDIA)
1434 return "media error";
1435 if (err_mask & AC_ERR_INVALID)
1436 return "invalid argument";
1437 if (err_mask & AC_ERR_DEV)
1438 return "device error";
1439 if (err_mask & AC_ERR_NCQ)
1441 if (err_mask & AC_ERR_NODEV_HINT)
1442 return "Polling detection error";
1443 return "unknown error";
1447 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1448 * @dev: Device to read log page 10h from
1449 * @tag: Resulting tag of the failed command
1450 * @tf: Resulting taskfile registers of the failed command
1452 * Read log page 10h to obtain NCQ error details and clear error
1456 * Kernel thread context (may sleep).
1459 * 0 on success, -errno otherwise.
1461 static int ata_eh_read_log_10h(struct ata_device *dev,
1462 int *tag, struct ata_taskfile *tf)
1464 u8 *buf = dev->link->ap->sector_buf;
1465 unsigned int err_mask;
1469 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1474 for (i = 0; i < ATA_SECT_SIZE; i++)
1477 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1483 *tag = buf[0] & 0x1f;
1485 tf->command = buf[2];
1486 tf->feature = buf[3];
1490 tf->device = buf[7];
1491 tf->hob_lbal = buf[8];
1492 tf->hob_lbam = buf[9];
1493 tf->hob_lbah = buf[10];
1494 tf->nsect = buf[12];
1495 tf->hob_nsect = buf[13];
1496 if (ata_id_has_ncq_autosense(dev->id))
1497 tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16];
1503 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1504 * @dev: target ATAPI device
1505 * @r_sense_key: out parameter for sense_key
1507 * Perform ATAPI TEST_UNIT_READY.
1510 * EH context (may sleep).
1513 * 0 on success, AC_ERR_* mask on failure.
1515 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1517 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1518 struct ata_taskfile tf;
1519 unsigned int err_mask;
1521 ata_tf_init(dev, &tf);
1523 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1524 tf.command = ATA_CMD_PACKET;
1525 tf.protocol = ATAPI_PROT_NODATA;
1527 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1528 if (err_mask == AC_ERR_DEV)
1529 *r_sense_key = tf.feature >> 4;
1534 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1535 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1536 * @cmd: scsi command for which the sense code should be set
1538 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1539 * SENSE. This function is an EH helper.
1542 * Kernel thread context (may sleep).
1544 static void ata_eh_request_sense(struct ata_queued_cmd *qc,
1545 struct scsi_cmnd *cmd)
1547 struct ata_device *dev = qc->dev;
1548 struct ata_taskfile tf;
1549 unsigned int err_mask;
1551 if (qc->ap->pflags & ATA_PFLAG_FROZEN) {
1552 ata_dev_warn(dev, "sense data available but port frozen\n");
1556 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1559 if (!ata_id_sense_reporting_enabled(dev->id)) {
1560 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1564 DPRINTK("ATA request sense\n");
1566 ata_tf_init(dev, &tf);
1567 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1568 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1569 tf.command = ATA_CMD_REQ_SENSE_DATA;
1570 tf.protocol = ATA_PROT_NODATA;
1572 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1573 /* Ignore err_mask; ATA_ERR might be set */
1574 if (tf.command & ATA_SENSE) {
1575 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1576 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1578 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1579 tf.command, err_mask);
1584 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1585 * @dev: device to perform REQUEST_SENSE to
1586 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1587 * @dfl_sense_key: default sense key to use
1589 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1590 * SENSE. This function is EH helper.
1593 * Kernel thread context (may sleep).
1596 * 0 on success, AC_ERR_* mask on failure
1598 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1599 u8 *sense_buf, u8 dfl_sense_key)
1601 u8 cdb[ATAPI_CDB_LEN] =
1602 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1603 struct ata_port *ap = dev->link->ap;
1604 struct ata_taskfile tf;
1606 DPRINTK("ATAPI request sense\n");
1608 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1610 /* initialize sense_buf with the error register,
1611 * for the case where they are -not- overwritten
1613 sense_buf[0] = 0x70;
1614 sense_buf[2] = dfl_sense_key;
1616 /* some devices time out if garbage left in tf */
1617 ata_tf_init(dev, &tf);
1619 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1620 tf.command = ATA_CMD_PACKET;
1622 /* is it pointless to prefer PIO for "safety reasons"? */
1623 if (ap->flags & ATA_FLAG_PIO_DMA) {
1624 tf.protocol = ATAPI_PROT_DMA;
1625 tf.feature |= ATAPI_PKT_DMA;
1627 tf.protocol = ATAPI_PROT_PIO;
1628 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1632 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1633 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1637 * ata_eh_analyze_serror - analyze SError for a failed port
1638 * @link: ATA link to analyze SError for
1640 * Analyze SError if available and further determine cause of
1646 static void ata_eh_analyze_serror(struct ata_link *link)
1648 struct ata_eh_context *ehc = &link->eh_context;
1649 u32 serror = ehc->i.serror;
1650 unsigned int err_mask = 0, action = 0;
1653 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1654 err_mask |= AC_ERR_ATA_BUS;
1655 action |= ATA_EH_RESET;
1657 if (serror & SERR_PROTOCOL) {
1658 err_mask |= AC_ERR_HSM;
1659 action |= ATA_EH_RESET;
1661 if (serror & SERR_INTERNAL) {
1662 err_mask |= AC_ERR_SYSTEM;
1663 action |= ATA_EH_RESET;
1666 /* Determine whether a hotplug event has occurred. Both
1667 * SError.N/X are considered hotplug events for enabled or
1668 * host links. For disabled PMP links, only N bit is
1669 * considered as X bit is left at 1 for link plugging.
1671 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1672 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1673 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1674 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1676 hotplug_mask = SERR_PHYRDY_CHG;
1678 if (serror & hotplug_mask)
1679 ata_ehi_hotplugged(&ehc->i);
1681 ehc->i.err_mask |= err_mask;
1682 ehc->i.action |= action;
1686 * ata_eh_analyze_ncq_error - analyze NCQ error
1687 * @link: ATA link to analyze NCQ error for
1689 * Read log page 10h, determine the offending qc and acquire
1690 * error status TF. For NCQ device errors, all LLDDs have to do
1691 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1695 * Kernel thread context (may sleep).
1697 void ata_eh_analyze_ncq_error(struct ata_link *link)
1699 struct ata_port *ap = link->ap;
1700 struct ata_eh_context *ehc = &link->eh_context;
1701 struct ata_device *dev = link->device;
1702 struct ata_queued_cmd *qc;
1703 struct ata_taskfile tf;
1706 /* if frozen, we can't do much */
1707 if (ap->pflags & ATA_PFLAG_FROZEN)
1710 /* is it NCQ device error? */
1711 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1714 /* has LLDD analyzed already? */
1715 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1716 qc = __ata_qc_from_tag(ap, tag);
1718 if (!(qc->flags & ATA_QCFLAG_FAILED))
1725 /* okay, this error is ours */
1726 memset(&tf, 0, sizeof(tf));
1727 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1729 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1734 if (!(link->sactive & (1 << tag))) {
1735 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1740 /* we've got the perpetrator, condemn it */
1741 qc = __ata_qc_from_tag(ap, tag);
1742 memcpy(&qc->result_tf, &tf, sizeof(tf));
1743 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1744 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1745 if ((qc->result_tf.command & ATA_SENSE) || qc->result_tf.auxiliary) {
1746 char sense_key, asc, ascq;
1748 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1749 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1750 ascq = qc->result_tf.auxiliary & 0xff;
1751 ata_scsi_set_sense(dev, qc->scsicmd, sense_key, asc, ascq);
1752 ata_scsi_set_sense_information(dev, qc->scsicmd,
1754 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1757 ehc->i.err_mask &= ~AC_ERR_DEV;
1761 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1762 * @qc: qc to analyze
1763 * @tf: Taskfile registers to analyze
1765 * Analyze taskfile of @qc and further determine cause of
1766 * failure. This function also requests ATAPI sense data if
1770 * Kernel thread context (may sleep).
1773 * Determined recovery action
1775 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1776 const struct ata_taskfile *tf)
1778 unsigned int tmp, action = 0;
1779 u8 stat = tf->command, err = tf->feature;
1781 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1782 qc->err_mask |= AC_ERR_HSM;
1783 return ATA_EH_RESET;
1786 if (stat & (ATA_ERR | ATA_DF)) {
1787 qc->err_mask |= AC_ERR_DEV;
1789 * Sense data reporting does not work if the
1790 * device fault bit is set.
1798 switch (qc->dev->class) {
1801 if (stat & ATA_SENSE)
1802 ata_eh_request_sense(qc, qc->scsicmd);
1804 qc->err_mask |= AC_ERR_ATA_BUS;
1805 if (err & (ATA_UNC | ATA_AMNF))
1806 qc->err_mask |= AC_ERR_MEDIA;
1808 qc->err_mask |= AC_ERR_INVALID;
1812 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1813 tmp = atapi_eh_request_sense(qc->dev,
1814 qc->scsicmd->sense_buffer,
1815 qc->result_tf.feature >> 4);
1817 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1819 qc->err_mask |= tmp;
1823 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1824 int ret = scsi_check_sense(qc->scsicmd);
1826 * SUCCESS here means that the sense code could be
1827 * evaluated and should be passed to the upper layers
1828 * for correct evaluation.
1829 * FAILED means the sense code could not be interpreted
1830 * and the device would need to be reset.
1831 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1832 * command would need to be retried.
1834 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1835 qc->flags |= ATA_QCFLAG_RETRY;
1836 qc->err_mask |= AC_ERR_OTHER;
1837 } else if (ret != SUCCESS) {
1838 qc->err_mask |= AC_ERR_HSM;
1841 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1842 action |= ATA_EH_RESET;
1847 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1852 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1856 base = ATA_ECAT_DUBIOUS_NONE;
1858 if (err_mask & AC_ERR_ATA_BUS)
1859 return base + ATA_ECAT_ATA_BUS;
1861 if (err_mask & AC_ERR_TIMEOUT)
1862 return base + ATA_ECAT_TOUT_HSM;
1864 if (eflags & ATA_EFLAG_IS_IO) {
1865 if (err_mask & AC_ERR_HSM)
1866 return base + ATA_ECAT_TOUT_HSM;
1868 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1869 return base + ATA_ECAT_UNK_DEV;
1875 struct speed_down_verdict_arg {
1878 int nr_errors[ATA_ECAT_NR];
1881 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1883 struct speed_down_verdict_arg *arg = void_arg;
1886 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1889 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1891 arg->nr_errors[cat]++;
1897 * ata_eh_speed_down_verdict - Determine speed down verdict
1898 * @dev: Device of interest
1900 * This function examines error ring of @dev and determines
1901 * whether NCQ needs to be turned off, transfer speed should be
1902 * stepped down, or falling back to PIO is necessary.
1904 * ECAT_ATA_BUS : ATA_BUS error for any command
1906 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1909 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1911 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1912 * data transfer hasn't been verified.
1916 * NCQ_OFF : Turn off NCQ.
1918 * SPEED_DOWN : Speed down transfer speed but don't fall back
1921 * FALLBACK_TO_PIO : Fall back to PIO.
1923 * Even if multiple verdicts are returned, only one action is
1924 * taken per error. An action triggered by non-DUBIOUS errors
1925 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1926 * This is to expedite speed down decisions right after device is
1927 * initially configured.
1929 * The following are speed down rules. #1 and #2 deal with
1932 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1933 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1935 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1936 * occurred during last 5 mins, NCQ_OFF.
1938 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1939 * occurred during last 5 mins, FALLBACK_TO_PIO
1941 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1942 * during last 10 mins, NCQ_OFF.
1944 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1945 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1948 * Inherited from caller.
1951 * OR of ATA_EH_SPDN_* flags.
1953 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1955 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1956 u64 j64 = get_jiffies_64();
1957 struct speed_down_verdict_arg arg;
1958 unsigned int verdict = 0;
1960 /* scan past 5 mins of error history */
1961 memset(&arg, 0, sizeof(arg));
1962 arg.since = j64 - min(j64, j5mins);
1963 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1965 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1966 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1967 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1968 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1970 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1971 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1972 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1974 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1975 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1976 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1977 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1979 /* scan past 10 mins of error history */
1980 memset(&arg, 0, sizeof(arg));
1981 arg.since = j64 - min(j64, j10mins);
1982 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1984 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1985 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1986 verdict |= ATA_EH_SPDN_NCQ_OFF;
1988 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1989 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1990 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1991 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1997 * ata_eh_speed_down - record error and speed down if necessary
1998 * @dev: Failed device
1999 * @eflags: mask of ATA_EFLAG_* flags
2000 * @err_mask: err_mask of the error
2002 * Record error and examine error history to determine whether
2003 * adjusting transmission speed is necessary. It also sets
2004 * transmission limits appropriately if such adjustment is
2008 * Kernel thread context (may sleep).
2011 * Determined recovery action.
2013 static unsigned int ata_eh_speed_down(struct ata_device *dev,
2014 unsigned int eflags, unsigned int err_mask)
2016 struct ata_link *link = ata_dev_phys_link(dev);
2018 unsigned int verdict;
2019 unsigned int action = 0;
2021 /* don't bother if Cat-0 error */
2022 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2025 /* record error and determine whether speed down is necessary */
2026 ata_ering_record(&dev->ering, eflags, err_mask);
2027 verdict = ata_eh_speed_down_verdict(dev);
2030 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2031 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2032 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2033 dev->flags |= ATA_DFLAG_NCQ_OFF;
2034 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2039 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2040 /* speed down SATA link speed if possible */
2041 if (sata_down_spd_limit(link, 0) == 0) {
2042 action |= ATA_EH_RESET;
2046 /* lower transfer mode */
2047 if (dev->spdn_cnt < 2) {
2048 static const int dma_dnxfer_sel[] =
2049 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2050 static const int pio_dnxfer_sel[] =
2051 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2054 if (dev->xfer_shift != ATA_SHIFT_PIO)
2055 sel = dma_dnxfer_sel[dev->spdn_cnt];
2057 sel = pio_dnxfer_sel[dev->spdn_cnt];
2061 if (ata_down_xfermask_limit(dev, sel) == 0) {
2062 action |= ATA_EH_RESET;
2068 /* Fall back to PIO? Slowing down to PIO is meaningless for
2069 * SATA ATA devices. Consider it only for PATA and SATAPI.
2071 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2072 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2073 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2074 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2076 action |= ATA_EH_RESET;
2083 /* device has been slowed down, blow error history */
2084 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2085 ata_ering_clear(&dev->ering);
2090 * ata_eh_worth_retry - analyze error and decide whether to retry
2091 * @qc: qc to possibly retry
2093 * Look at the cause of the error and decide if a retry
2094 * might be useful or not. We don't want to retry media errors
2095 * because the drive itself has probably already taken 10-30 seconds
2096 * doing its own internal retries before reporting the failure.
2098 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2100 if (qc->err_mask & AC_ERR_MEDIA)
2101 return 0; /* don't retry media errors */
2102 if (qc->flags & ATA_QCFLAG_IO)
2103 return 1; /* otherwise retry anything from fs stack */
2104 if (qc->err_mask & AC_ERR_INVALID)
2105 return 0; /* don't retry these */
2106 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2110 * ata_eh_quiet - check if we need to be quiet about a command error
2113 * Look at the qc flags anbd its scsi command request flags to determine
2114 * if we need to be quiet about the command failure.
2116 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
2119 qc->scsicmd->request->rq_flags & RQF_QUIET)
2120 qc->flags |= ATA_QCFLAG_QUIET;
2121 return qc->flags & ATA_QCFLAG_QUIET;
2125 * ata_eh_link_autopsy - analyze error and determine recovery action
2126 * @link: host link to perform autopsy on
2128 * Analyze why @link failed and determine which recovery actions
2129 * are needed. This function also sets more detailed AC_ERR_*
2130 * values and fills sense data for ATAPI CHECK SENSE.
2133 * Kernel thread context (may sleep).
2135 static void ata_eh_link_autopsy(struct ata_link *link)
2137 struct ata_port *ap = link->ap;
2138 struct ata_eh_context *ehc = &link->eh_context;
2139 struct ata_device *dev;
2140 unsigned int all_err_mask = 0, eflags = 0;
2141 int tag, nr_failed = 0, nr_quiet = 0;
2147 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2150 /* obtain and analyze SError */
2151 rc = sata_scr_read(link, SCR_ERROR, &serror);
2153 ehc->i.serror |= serror;
2154 ata_eh_analyze_serror(link);
2155 } else if (rc != -EOPNOTSUPP) {
2156 /* SError read failed, force reset and probing */
2157 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2158 ehc->i.action |= ATA_EH_RESET;
2159 ehc->i.err_mask |= AC_ERR_OTHER;
2162 /* analyze NCQ failure */
2163 ata_eh_analyze_ncq_error(link);
2165 /* any real error trumps AC_ERR_OTHER */
2166 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2167 ehc->i.err_mask &= ~AC_ERR_OTHER;
2169 all_err_mask |= ehc->i.err_mask;
2171 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2172 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2174 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2175 ata_dev_phys_link(qc->dev) != link)
2178 /* inherit upper level err_mask */
2179 qc->err_mask |= ehc->i.err_mask;
2182 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2184 /* DEV errors are probably spurious in case of ATA_BUS error */
2185 if (qc->err_mask & AC_ERR_ATA_BUS)
2186 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2189 /* any real error trumps unknown error */
2190 if (qc->err_mask & ~AC_ERR_OTHER)
2191 qc->err_mask &= ~AC_ERR_OTHER;
2194 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2195 * layers will determine whether the command is worth retrying
2196 * based on the sense data and device class/type. Otherwise,
2197 * determine directly if the command is worth retrying using its
2198 * error mask and flags.
2200 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2201 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2202 else if (ata_eh_worth_retry(qc))
2203 qc->flags |= ATA_QCFLAG_RETRY;
2205 /* accumulate error info */
2206 ehc->i.dev = qc->dev;
2207 all_err_mask |= qc->err_mask;
2208 if (qc->flags & ATA_QCFLAG_IO)
2209 eflags |= ATA_EFLAG_IS_IO;
2210 trace_ata_eh_link_autopsy_qc(qc);
2212 /* Count quiet errors */
2213 if (ata_eh_quiet(qc))
2218 /* If all failed commands requested silence, then be quiet */
2219 if (nr_quiet == nr_failed)
2220 ehc->i.flags |= ATA_EHI_QUIET;
2222 /* enforce default EH actions */
2223 if (ap->pflags & ATA_PFLAG_FROZEN ||
2224 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2225 ehc->i.action |= ATA_EH_RESET;
2226 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2227 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2228 ehc->i.action |= ATA_EH_REVALIDATE;
2230 /* If we have offending qcs and the associated failed device,
2231 * perform per-dev EH action only on the offending device.
2234 ehc->i.dev_action[ehc->i.dev->devno] |=
2235 ehc->i.action & ATA_EH_PERDEV_MASK;
2236 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2239 /* propagate timeout to host link */
2240 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2241 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2243 /* record error and consider speeding down */
2245 if (!dev && ((ata_link_max_devices(link) == 1 &&
2246 ata_dev_enabled(link->device))))
2250 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2251 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2252 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2253 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2259 * ata_eh_autopsy - analyze error and determine recovery action
2260 * @ap: host port to perform autopsy on
2262 * Analyze all links of @ap and determine why they failed and
2263 * which recovery actions are needed.
2266 * Kernel thread context (may sleep).
2268 void ata_eh_autopsy(struct ata_port *ap)
2270 struct ata_link *link;
2272 ata_for_each_link(link, ap, EDGE)
2273 ata_eh_link_autopsy(link);
2275 /* Handle the frigging slave link. Autopsy is done similarly
2276 * but actions and flags are transferred over to the master
2277 * link and handled from there.
2279 if (ap->slave_link) {
2280 struct ata_eh_context *mehc = &ap->link.eh_context;
2281 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2283 /* transfer control flags from master to slave */
2284 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2286 /* perform autopsy on the slave link */
2287 ata_eh_link_autopsy(ap->slave_link);
2289 /* transfer actions from slave to master and clear slave */
2290 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2291 mehc->i.action |= sehc->i.action;
2292 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2293 mehc->i.flags |= sehc->i.flags;
2294 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2297 /* Autopsy of fanout ports can affect host link autopsy.
2298 * Perform host link autopsy last.
2300 if (sata_pmp_attached(ap))
2301 ata_eh_link_autopsy(&ap->link);
2305 * ata_get_cmd_descript - get description for ATA command
2306 * @command: ATA command code to get description for
2308 * Return a textual description of the given command, or NULL if the
2309 * command is not known.
2314 const char *ata_get_cmd_descript(u8 command)
2316 #ifdef CONFIG_ATA_VERBOSE_ERROR
2322 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2323 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2324 { ATA_CMD_STANDBY, "STANDBY" },
2325 { ATA_CMD_IDLE, "IDLE" },
2326 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2327 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2328 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2329 { ATA_CMD_NOP, "NOP" },
2330 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2331 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2332 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2333 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2334 { ATA_CMD_SERVICE, "SERVICE" },
2335 { ATA_CMD_READ, "READ DMA" },
2336 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2337 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2338 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2339 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2340 { ATA_CMD_WRITE, "WRITE DMA" },
2341 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2342 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2343 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2344 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2345 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2346 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2347 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2348 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2349 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2350 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2351 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2352 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2353 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2354 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2355 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2356 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2357 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2358 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2359 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2360 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2361 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2362 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2363 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2364 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2365 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2366 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2367 { ATA_CMD_SLEEP, "SLEEP" },
2368 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2369 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2370 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2371 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2372 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2373 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2374 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2375 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2376 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2377 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2378 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2379 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2380 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2381 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2382 { ATA_CMD_PMP_READ, "READ BUFFER" },
2383 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2384 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2385 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2386 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2387 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2388 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2389 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2390 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2391 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2392 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2393 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2394 { ATA_CMD_SMART, "SMART" },
2395 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2396 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2397 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2398 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2399 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2400 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2401 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2402 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2403 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2404 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2405 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2406 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2407 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2408 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2409 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2410 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2411 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2412 { ATA_CMD_RESTORE, "RECALIBRATE" },
2413 { 0, NULL } /* terminate list */
2417 for (i = 0; cmd_descr[i].text; i++)
2418 if (cmd_descr[i].command == command)
2419 return cmd_descr[i].text;
2424 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2427 * ata_eh_link_report - report error handling to user
2428 * @link: ATA link EH is going on
2430 * Report EH to user.
2435 static void ata_eh_link_report(struct ata_link *link)
2437 struct ata_port *ap = link->ap;
2438 struct ata_eh_context *ehc = &link->eh_context;
2439 const char *frozen, *desc;
2440 char tries_buf[6] = "";
2441 int tag, nr_failed = 0;
2443 if (ehc->i.flags & ATA_EHI_QUIET)
2447 if (ehc->i.desc[0] != '\0')
2450 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2451 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2453 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2454 ata_dev_phys_link(qc->dev) != link ||
2455 ((qc->flags & ATA_QCFLAG_QUIET) &&
2456 qc->err_mask == AC_ERR_DEV))
2458 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2464 if (!nr_failed && !ehc->i.err_mask)
2468 if (ap->pflags & ATA_PFLAG_FROZEN)
2471 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2472 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2476 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2477 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2478 ehc->i.err_mask, link->sactive, ehc->i.serror,
2479 ehc->i.action, frozen, tries_buf);
2481 ata_dev_err(ehc->i.dev, "%s\n", desc);
2483 ata_link_err(link, "exception Emask 0x%x "
2484 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2485 ehc->i.err_mask, link->sactive, ehc->i.serror,
2486 ehc->i.action, frozen, tries_buf);
2488 ata_link_err(link, "%s\n", desc);
2491 #ifdef CONFIG_ATA_VERBOSE_ERROR
2494 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2495 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2496 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2497 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2498 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2499 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2500 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2501 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2502 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2503 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2504 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2505 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2506 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2507 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2508 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2509 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2510 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2511 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2514 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2515 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2516 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2517 char data_buf[20] = "";
2518 char cdb_buf[70] = "";
2520 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2521 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2524 if (qc->dma_dir != DMA_NONE) {
2525 static const char *dma_str[] = {
2526 [DMA_BIDIRECTIONAL] = "bidi",
2527 [DMA_TO_DEVICE] = "out",
2528 [DMA_FROM_DEVICE] = "in",
2530 const char *prot_str = NULL;
2532 switch (qc->tf.protocol) {
2533 case ATA_PROT_UNKNOWN:
2534 prot_str = "unknown";
2536 case ATA_PROT_NODATA:
2537 prot_str = "nodata";
2546 prot_str = "ncq dma";
2548 case ATA_PROT_NCQ_NODATA:
2549 prot_str = "ncq nodata";
2551 case ATAPI_PROT_NODATA:
2552 prot_str = "nodata";
2554 case ATAPI_PROT_PIO:
2557 case ATAPI_PROT_DMA:
2561 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2562 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2565 if (ata_is_atapi(qc->tf.protocol)) {
2566 const u8 *cdb = qc->cdb;
2567 size_t cdb_len = qc->dev->cdb_len;
2570 cdb = qc->scsicmd->cmnd;
2571 cdb_len = qc->scsicmd->cmd_len;
2573 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2576 const char *descr = ata_get_cmd_descript(cmd->command);
2578 ata_dev_err(qc->dev, "failed command: %s\n",
2582 ata_dev_err(qc->dev,
2583 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2585 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2586 "Emask 0x%x (%s)%s\n",
2587 cmd->command, cmd->feature, cmd->nsect,
2588 cmd->lbal, cmd->lbam, cmd->lbah,
2589 cmd->hob_feature, cmd->hob_nsect,
2590 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2591 cmd->device, qc->tag, data_buf, cdb_buf,
2592 res->command, res->feature, res->nsect,
2593 res->lbal, res->lbam, res->lbah,
2594 res->hob_feature, res->hob_nsect,
2595 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2596 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2597 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2599 #ifdef CONFIG_ATA_VERBOSE_ERROR
2600 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2601 ATA_SENSE | ATA_ERR)) {
2602 if (res->command & ATA_BUSY)
2603 ata_dev_err(qc->dev, "status: { Busy }\n");
2605 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2606 res->command & ATA_DRDY ? "DRDY " : "",
2607 res->command & ATA_DF ? "DF " : "",
2608 res->command & ATA_DRQ ? "DRQ " : "",
2609 res->command & ATA_SENSE ? "SENSE " : "",
2610 res->command & ATA_ERR ? "ERR " : "");
2613 if (cmd->command != ATA_CMD_PACKET &&
2614 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2615 ATA_IDNF | ATA_ABORTED)))
2616 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2617 res->feature & ATA_ICRC ? "ICRC " : "",
2618 res->feature & ATA_UNC ? "UNC " : "",
2619 res->feature & ATA_AMNF ? "AMNF " : "",
2620 res->feature & ATA_IDNF ? "IDNF " : "",
2621 res->feature & ATA_ABORTED ? "ABRT " : "");
2627 * ata_eh_report - report error handling to user
2628 * @ap: ATA port to report EH about
2630 * Report EH to user.
2635 void ata_eh_report(struct ata_port *ap)
2637 struct ata_link *link;
2639 ata_for_each_link(link, ap, HOST_FIRST)
2640 ata_eh_link_report(link);
2643 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2644 unsigned int *classes, unsigned long deadline,
2647 struct ata_device *dev;
2650 ata_for_each_dev(dev, link, ALL)
2651 classes[dev->devno] = ATA_DEV_UNKNOWN;
2653 return reset(link, classes, deadline);
2656 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2658 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2662 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2667 int ata_eh_reset(struct ata_link *link, int classify,
2668 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2669 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2671 struct ata_port *ap = link->ap;
2672 struct ata_link *slave = ap->slave_link;
2673 struct ata_eh_context *ehc = &link->eh_context;
2674 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2675 unsigned int *classes = ehc->classes;
2676 unsigned int lflags = link->flags;
2677 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2678 int max_tries = 0, try = 0;
2679 struct ata_link *failed_link;
2680 struct ata_device *dev;
2681 unsigned long deadline, now;
2682 ata_reset_fn_t reset;
2683 unsigned long flags;
2690 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2692 if (link->flags & ATA_LFLAG_RST_ONCE)
2694 if (link->flags & ATA_LFLAG_NO_HRST)
2696 if (link->flags & ATA_LFLAG_NO_SRST)
2699 /* make sure each reset attempt is at least COOL_DOWN apart */
2700 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2702 WARN_ON(time_after(ehc->last_reset, now));
2703 deadline = ata_deadline(ehc->last_reset,
2704 ATA_EH_RESET_COOL_DOWN);
2705 if (time_before(now, deadline))
2706 schedule_timeout_uninterruptible(deadline - now);
2709 spin_lock_irqsave(ap->lock, flags);
2710 ap->pflags |= ATA_PFLAG_RESETTING;
2711 spin_unlock_irqrestore(ap->lock, flags);
2713 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2715 ata_for_each_dev(dev, link, ALL) {
2716 /* If we issue an SRST then an ATA drive (not ATAPI)
2717 * may change configuration and be in PIO0 timing. If
2718 * we do a hard reset (or are coming from power on)
2719 * this is true for ATA or ATAPI. Until we've set a
2720 * suitable controller mode we should not touch the
2721 * bus as we may be talking too fast.
2723 dev->pio_mode = XFER_PIO_0;
2724 dev->dma_mode = 0xff;
2726 /* If the controller has a pio mode setup function
2727 * then use it to set the chipset to rights. Don't
2728 * touch the DMA setup as that will be dealt with when
2729 * configuring devices.
2731 if (ap->ops->set_piomode)
2732 ap->ops->set_piomode(ap, dev);
2735 /* prefer hardreset */
2737 ehc->i.action &= ~ATA_EH_RESET;
2740 ehc->i.action |= ATA_EH_HARDRESET;
2741 } else if (softreset) {
2743 ehc->i.action |= ATA_EH_SOFTRESET;
2747 unsigned long deadline = ata_deadline(jiffies,
2748 ATA_EH_PRERESET_TIMEOUT);
2751 sehc->i.action &= ~ATA_EH_RESET;
2752 sehc->i.action |= ehc->i.action;
2755 rc = prereset(link, deadline);
2757 /* If present, do prereset on slave link too. Reset
2758 * is skipped iff both master and slave links report
2759 * -ENOENT or clear ATA_EH_RESET.
2761 if (slave && (rc == 0 || rc == -ENOENT)) {
2764 tmp = prereset(slave, deadline);
2768 ehc->i.action |= sehc->i.action;
2772 if (rc == -ENOENT) {
2773 ata_link_dbg(link, "port disabled--ignoring\n");
2774 ehc->i.action &= ~ATA_EH_RESET;
2776 ata_for_each_dev(dev, link, ALL)
2777 classes[dev->devno] = ATA_DEV_NONE;
2782 "prereset failed (errno=%d)\n",
2787 /* prereset() might have cleared ATA_EH_RESET. If so,
2788 * bang classes, thaw and return.
2790 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2791 ata_for_each_dev(dev, link, ALL)
2792 classes[dev->devno] = ATA_DEV_NONE;
2793 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2794 ata_is_host_link(link))
2795 ata_eh_thaw_port(ap);
2805 if (ata_is_host_link(link))
2806 ata_eh_freeze_port(ap);
2808 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2812 ata_link_info(link, "%s resetting link\n",
2813 reset == softreset ? "soft" : "hard");
2815 /* mark that this EH session started with reset */
2816 ehc->last_reset = jiffies;
2817 if (reset == hardreset)
2818 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2820 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2822 rc = ata_do_reset(link, reset, classes, deadline, true);
2823 if (rc && rc != -EAGAIN) {
2828 /* hardreset slave link if existent */
2829 if (slave && reset == hardreset) {
2833 ata_link_info(slave, "hard resetting link\n");
2835 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2836 tmp = ata_do_reset(slave, reset, classes, deadline,
2844 failed_link = slave;
2850 /* perform follow-up SRST if necessary */
2851 if (reset == hardreset &&
2852 ata_eh_followup_srst_needed(link, rc)) {
2857 "follow-up softreset required but no softreset available\n");
2863 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2864 rc = ata_do_reset(link, reset, classes, deadline, true);
2873 "no reset method available, skipping reset\n");
2874 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2875 lflags |= ATA_LFLAG_ASSUME_ATA;
2879 * Post-reset processing
2881 ata_for_each_dev(dev, link, ALL) {
2882 /* After the reset, the device state is PIO 0 and the
2883 * controller state is undefined. Reset also wakes up
2884 * drives from sleeping mode.
2886 dev->pio_mode = XFER_PIO_0;
2887 dev->flags &= ~ATA_DFLAG_SLEEPING;
2889 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2892 /* apply class override */
2893 if (lflags & ATA_LFLAG_ASSUME_ATA)
2894 classes[dev->devno] = ATA_DEV_ATA;
2895 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2896 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2899 /* record current link speed */
2900 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2901 link->sata_spd = (sstatus >> 4) & 0xf;
2902 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2903 slave->sata_spd = (sstatus >> 4) & 0xf;
2906 if (ata_is_host_link(link))
2907 ata_eh_thaw_port(ap);
2909 /* postreset() should clear hardware SError. Although SError
2910 * is cleared during link resume, clearing SError here is
2911 * necessary as some PHYs raise hotplug events after SRST.
2912 * This introduces race condition where hotplug occurs between
2913 * reset and here. This race is mediated by cross checking
2914 * link onlineness and classification result later.
2917 postreset(link, classes);
2919 postreset(slave, classes);
2923 * Some controllers can't be frozen very well and may set spurious
2924 * error conditions during reset. Clear accumulated error
2925 * information and re-thaw the port if frozen. As reset is the
2926 * final recovery action and we cross check link onlineness against
2927 * device classification later, no hotplug event is lost by this.
2929 spin_lock_irqsave(link->ap->lock, flags);
2930 memset(&link->eh_info, 0, sizeof(link->eh_info));
2932 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2933 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2934 spin_unlock_irqrestore(link->ap->lock, flags);
2936 if (ap->pflags & ATA_PFLAG_FROZEN)
2937 ata_eh_thaw_port(ap);
2940 * Make sure onlineness and classification result correspond.
2941 * Hotplug could have happened during reset and some
2942 * controllers fail to wait while a drive is spinning up after
2943 * being hotplugged causing misdetection. By cross checking
2944 * link on/offlineness and classification result, those
2945 * conditions can be reliably detected and retried.
2948 ata_for_each_dev(dev, link, ALL) {
2949 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2950 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2951 ata_dev_dbg(dev, "link online but device misclassified\n");
2952 classes[dev->devno] = ATA_DEV_NONE;
2955 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2956 if (ata_class_enabled(classes[dev->devno]))
2958 "link offline, clearing class %d to NONE\n",
2959 classes[dev->devno]);
2960 classes[dev->devno] = ATA_DEV_NONE;
2961 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2963 "link status unknown, clearing UNKNOWN to NONE\n");
2964 classes[dev->devno] = ATA_DEV_NONE;
2968 if (classify && nr_unknown) {
2969 if (try < max_tries) {
2971 "link online but %d devices misclassified, retrying\n",
2978 "link online but %d devices misclassified, "
2979 "device detection might fail\n", nr_unknown);
2982 /* reset successful, schedule revalidation */
2983 ata_eh_done(link, NULL, ATA_EH_RESET);
2985 ata_eh_done(slave, NULL, ATA_EH_RESET);
2986 ehc->last_reset = jiffies; /* update to completion time */
2987 ehc->i.action |= ATA_EH_REVALIDATE;
2988 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2992 /* clear hotplug flag */
2993 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2995 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2997 spin_lock_irqsave(ap->lock, flags);
2998 ap->pflags &= ~ATA_PFLAG_RESETTING;
2999 spin_unlock_irqrestore(ap->lock, flags);
3004 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
3005 if (!ata_is_host_link(link) &&
3006 sata_scr_read(link, SCR_STATUS, &sstatus))
3009 if (try >= max_tries) {
3011 * Thaw host port even if reset failed, so that the port
3012 * can be retried on the next phy event. This risks
3013 * repeated EH runs but seems to be a better tradeoff than
3014 * shutting down a port after a botched hotplug attempt.
3016 if (ata_is_host_link(link))
3017 ata_eh_thaw_port(ap);
3022 if (time_before(now, deadline)) {
3023 unsigned long delta = deadline - now;
3025 ata_link_warn(failed_link,
3026 "reset failed (errno=%d), retrying in %u secs\n",
3027 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3031 delta = schedule_timeout_uninterruptible(delta);
3036 * While disks spinup behind PMP, some controllers fail sending SRST.
3037 * They need to be reset - as well as the PMP - before retrying.
3039 if (rc == -ERESTART) {
3040 if (ata_is_host_link(link))
3041 ata_eh_thaw_port(ap);
3045 if (try == max_tries - 1) {
3046 sata_down_spd_limit(link, 0);
3048 sata_down_spd_limit(slave, 0);
3049 } else if (rc == -EPIPE)
3050 sata_down_spd_limit(failed_link, 0);
3057 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3059 struct ata_link *link;
3060 struct ata_device *dev;
3061 unsigned long flags;
3064 * This function can be thought of as an extended version of
3065 * ata_eh_about_to_do() specially crafted to accommodate the
3066 * requirements of ATA_EH_PARK handling. Since the EH thread
3067 * does not leave the do {} while () loop in ata_eh_recover as
3068 * long as the timeout for a park request to *one* device on
3069 * the port has not expired, and since we still want to pick
3070 * up park requests to other devices on the same port or
3071 * timeout updates for the same device, we have to pull
3072 * ATA_EH_PARK actions from eh_info into eh_context.i
3073 * ourselves at the beginning of each pass over the loop.
3075 * Additionally, all write accesses to &ap->park_req_pending
3076 * through reinit_completion() (see below) or complete_all()
3077 * (see ata_scsi_park_store()) are protected by the host lock.
3078 * As a result we have that park_req_pending.done is zero on
3079 * exit from this function, i.e. when ATA_EH_PARK actions for
3080 * *all* devices on port ap have been pulled into the
3081 * respective eh_context structs. If, and only if,
3082 * park_req_pending.done is non-zero by the time we reach
3083 * wait_for_completion_timeout(), another ATA_EH_PARK action
3084 * has been scheduled for at least one of the devices on port
3085 * ap and we have to cycle over the do {} while () loop in
3086 * ata_eh_recover() again.
3089 spin_lock_irqsave(ap->lock, flags);
3090 reinit_completion(&ap->park_req_pending);
3091 ata_for_each_link(link, ap, EDGE) {
3092 ata_for_each_dev(dev, link, ALL) {
3093 struct ata_eh_info *ehi = &link->eh_info;
3095 link->eh_context.i.dev_action[dev->devno] |=
3096 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3097 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3100 spin_unlock_irqrestore(ap->lock, flags);
3103 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3105 struct ata_eh_context *ehc = &dev->link->eh_context;
3106 struct ata_taskfile tf;
3107 unsigned int err_mask;
3109 ata_tf_init(dev, &tf);
3111 ehc->unloaded_mask |= 1 << dev->devno;
3112 tf.command = ATA_CMD_IDLEIMMEDIATE;
3118 ehc->unloaded_mask &= ~(1 << dev->devno);
3119 tf.command = ATA_CMD_CHK_POWER;
3122 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3123 tf.protocol = ATA_PROT_NODATA;
3124 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3125 if (park && (err_mask || tf.lbal != 0xc4)) {
3126 ata_dev_err(dev, "head unload failed!\n");
3127 ehc->unloaded_mask &= ~(1 << dev->devno);
3131 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3132 struct ata_device **r_failed_dev)
3134 struct ata_port *ap = link->ap;
3135 struct ata_eh_context *ehc = &link->eh_context;
3136 struct ata_device *dev;
3137 unsigned int new_mask = 0;
3138 unsigned long flags;
3143 /* For PATA drive side cable detection to work, IDENTIFY must
3144 * be done backwards such that PDIAG- is released by the slave
3145 * device before the master device is identified.
3147 ata_for_each_dev(dev, link, ALL_REVERSE) {
3148 unsigned int action = ata_eh_dev_action(dev);
3149 unsigned int readid_flags = 0;
3151 if (ehc->i.flags & ATA_EHI_DID_RESET)
3152 readid_flags |= ATA_READID_POSTRESET;
3154 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3155 WARN_ON(dev->class == ATA_DEV_PMP);
3157 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3162 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3163 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3168 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3170 /* Configuration may have changed, reconfigure
3173 ehc->i.flags |= ATA_EHI_SETMODE;
3175 /* schedule the scsi_rescan_device() here */
3176 schedule_work(&(ap->scsi_rescan_task));
3177 } else if (dev->class == ATA_DEV_UNKNOWN &&
3178 ehc->tries[dev->devno] &&
3179 ata_class_enabled(ehc->classes[dev->devno])) {
3180 /* Temporarily set dev->class, it will be
3181 * permanently set once all configurations are
3182 * complete. This is necessary because new
3183 * device configuration is done in two
3186 dev->class = ehc->classes[dev->devno];
3188 if (dev->class == ATA_DEV_PMP)
3189 rc = sata_pmp_attach(dev);
3191 rc = ata_dev_read_id(dev, &dev->class,
3192 readid_flags, dev->id);
3194 /* read_id might have changed class, store and reset */
3195 ehc->classes[dev->devno] = dev->class;
3196 dev->class = ATA_DEV_UNKNOWN;
3200 /* clear error info accumulated during probe */
3201 ata_ering_clear(&dev->ering);
3202 new_mask |= 1 << dev->devno;
3205 /* IDENTIFY was issued to non-existent
3206 * device. No need to reset. Just
3207 * thaw and ignore the device.
3209 ata_eh_thaw_port(ap);
3217 /* PDIAG- should have been released, ask cable type if post-reset */
3218 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3219 if (ap->ops->cable_detect)
3220 ap->cbl = ap->ops->cable_detect(ap);
3224 /* Configure new devices forward such that user doesn't see
3225 * device detection messages backwards.
3227 ata_for_each_dev(dev, link, ALL) {
3228 if (!(new_mask & (1 << dev->devno)))
3231 dev->class = ehc->classes[dev->devno];
3233 if (dev->class == ATA_DEV_PMP)
3236 ehc->i.flags |= ATA_EHI_PRINTINFO;
3237 rc = ata_dev_configure(dev);
3238 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3240 dev->class = ATA_DEV_UNKNOWN;
3244 spin_lock_irqsave(ap->lock, flags);
3245 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3246 spin_unlock_irqrestore(ap->lock, flags);
3248 /* new device discovered, configure xfermode */
3249 ehc->i.flags |= ATA_EHI_SETMODE;
3255 *r_failed_dev = dev;
3256 DPRINTK("EXIT rc=%d\n", rc);
3261 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3262 * @link: link on which timings will be programmed
3263 * @r_failed_dev: out parameter for failed device
3265 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3266 * ata_set_mode() fails, pointer to the failing device is
3267 * returned in @r_failed_dev.
3270 * PCI/etc. bus probe sem.
3273 * 0 on success, negative errno otherwise
3275 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3277 struct ata_port *ap = link->ap;
3278 struct ata_device *dev;
3281 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3282 ata_for_each_dev(dev, link, ENABLED) {
3283 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3284 struct ata_ering_entry *ent;
3286 ent = ata_ering_top(&dev->ering);
3288 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3292 /* has private set_mode? */
3293 if (ap->ops->set_mode)
3294 rc = ap->ops->set_mode(link, r_failed_dev);
3296 rc = ata_do_set_mode(link, r_failed_dev);
3298 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3299 ata_for_each_dev(dev, link, ENABLED) {
3300 struct ata_eh_context *ehc = &link->eh_context;
3301 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3302 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3304 if (dev->xfer_mode != saved_xfer_mode ||
3305 ata_ncq_enabled(dev) != saved_ncq)
3306 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3313 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3314 * @dev: ATAPI device to clear UA for
3316 * Resets and other operations can make an ATAPI device raise
3317 * UNIT ATTENTION which causes the next operation to fail. This
3318 * function clears UA.
3321 * EH context (may sleep).
3324 * 0 on success, -errno on failure.
3326 static int atapi_eh_clear_ua(struct ata_device *dev)
3330 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3331 u8 *sense_buffer = dev->link->ap->sector_buf;
3333 unsigned int err_mask;
3335 err_mask = atapi_eh_tur(dev, &sense_key);
3336 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3338 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3343 if (!err_mask || sense_key != UNIT_ATTENTION)
3346 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3348 ata_dev_warn(dev, "failed to clear "
3349 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3354 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3361 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3362 * @dev: ATA device which may need FLUSH retry
3364 * If @dev failed FLUSH, it needs to be reported upper layer
3365 * immediately as it means that @dev failed to remap and already
3366 * lost at least a sector and further FLUSH retrials won't make
3367 * any difference to the lost sector. However, if FLUSH failed
3368 * for other reasons, for example transmission error, FLUSH needs
3371 * This function determines whether FLUSH failure retry is
3372 * necessary and performs it if so.
3375 * 0 if EH can continue, -errno if EH needs to be repeated.
3377 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3379 struct ata_link *link = dev->link;
3380 struct ata_port *ap = link->ap;
3381 struct ata_queued_cmd *qc;
3382 struct ata_taskfile tf;
3383 unsigned int err_mask;
3386 /* did flush fail for this device? */
3387 if (!ata_tag_valid(link->active_tag))
3390 qc = __ata_qc_from_tag(ap, link->active_tag);
3391 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3392 qc->tf.command != ATA_CMD_FLUSH))
3395 /* if the device failed it, it should be reported to upper layers */
3396 if (qc->err_mask & AC_ERR_DEV)
3399 /* flush failed for some other reason, give it another shot */
3400 ata_tf_init(dev, &tf);
3402 tf.command = qc->tf.command;
3403 tf.flags |= ATA_TFLAG_DEVICE;
3404 tf.protocol = ATA_PROT_NODATA;
3406 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3407 tf.command, qc->err_mask);
3409 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3412 * FLUSH is complete but there's no way to
3413 * successfully complete a failed command from EH.
3414 * Making sure retry is allowed at least once and
3415 * retrying it should do the trick - whatever was in
3416 * the cache is already on the platter and this won't
3417 * cause infinite loop.
3419 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3421 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3425 /* if device failed it, report it to upper layers */
3426 if (err_mask & AC_ERR_DEV) {
3427 qc->err_mask |= AC_ERR_DEV;
3429 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3437 * ata_eh_set_lpm - configure SATA interface power management
3438 * @link: link to configure power management
3439 * @policy: the link power management policy
3440 * @r_failed_dev: out parameter for failed device
3442 * Enable SATA Interface power management. This will enable
3443 * Device Interface Power Management (DIPM) for min_power and
3444 * medium_power_with_dipm policies, and then call driver specific
3445 * callbacks for enabling Host Initiated Power management.
3451 * 0 on success, -errno on failure.
3453 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3454 struct ata_device **r_failed_dev)
3456 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3457 struct ata_eh_context *ehc = &link->eh_context;
3458 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3459 enum ata_lpm_policy old_policy = link->lpm_policy;
3460 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3461 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3462 unsigned int err_mask;
3465 /* if the link or host doesn't do LPM, noop */
3466 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3470 * DIPM is enabled only for MIN_POWER as some devices
3471 * misbehave when the host NACKs transition to SLUMBER. Order
3472 * device and link configurations such that the host always
3473 * allows DIPM requests.
3475 ata_for_each_dev(dev, link, ENABLED) {
3476 bool hipm = ata_id_has_hipm(dev->id);
3477 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3479 /* find the first enabled and LPM enabled devices */
3483 if (!lpm_dev && (hipm || dipm))
3486 hints &= ~ATA_LPM_EMPTY;
3488 hints &= ~ATA_LPM_HIPM;
3490 /* disable DIPM before changing link config */
3491 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3492 err_mask = ata_dev_set_feature(dev,
3493 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3494 if (err_mask && err_mask != AC_ERR_DEV) {
3496 "failed to disable DIPM, Emask 0x%x\n",
3505 rc = ap->ops->set_lpm(link, policy, hints);
3506 if (!rc && ap->slave_link)
3507 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3509 rc = sata_pmp_set_lpm(link, policy, hints);
3512 * Attribute link config failure to the first (LPM) enabled
3513 * device on the link.
3516 if (rc == -EOPNOTSUPP) {
3517 link->flags |= ATA_LFLAG_NO_LPM;
3520 dev = lpm_dev ? lpm_dev : link_dev;
3525 * Low level driver acked the transition. Issue DIPM command
3526 * with the new policy set.
3528 link->lpm_policy = policy;
3529 if (ap && ap->slave_link)
3530 ap->slave_link->lpm_policy = policy;
3532 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3533 ata_for_each_dev(dev, link, ENABLED) {
3534 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3535 ata_id_has_dipm(dev->id)) {
3536 err_mask = ata_dev_set_feature(dev,
3537 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3538 if (err_mask && err_mask != AC_ERR_DEV) {
3540 "failed to enable DIPM, Emask 0x%x\n",
3548 link->last_lpm_change = jiffies;
3549 link->flags |= ATA_LFLAG_CHANGED;
3554 /* restore the old policy */
3555 link->lpm_policy = old_policy;
3556 if (ap && ap->slave_link)
3557 ap->slave_link->lpm_policy = old_policy;
3559 /* if no device or only one more chance is left, disable LPM */
3560 if (!dev || ehc->tries[dev->devno] <= 2) {
3561 ata_link_warn(link, "disabling LPM on the link\n");
3562 link->flags |= ATA_LFLAG_NO_LPM;
3565 *r_failed_dev = dev;
3569 int ata_link_nr_enabled(struct ata_link *link)
3571 struct ata_device *dev;
3574 ata_for_each_dev(dev, link, ENABLED)
3579 static int ata_link_nr_vacant(struct ata_link *link)
3581 struct ata_device *dev;
3584 ata_for_each_dev(dev, link, ALL)
3585 if (dev->class == ATA_DEV_UNKNOWN)
3590 static int ata_eh_skip_recovery(struct ata_link *link)
3592 struct ata_port *ap = link->ap;
3593 struct ata_eh_context *ehc = &link->eh_context;
3594 struct ata_device *dev;
3596 /* skip disabled links */
3597 if (link->flags & ATA_LFLAG_DISABLED)
3600 /* skip if explicitly requested */
3601 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3604 /* thaw frozen port and recover failed devices */
3605 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3608 /* reset at least once if reset is requested */
3609 if ((ehc->i.action & ATA_EH_RESET) &&
3610 !(ehc->i.flags & ATA_EHI_DID_RESET))
3613 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3614 ata_for_each_dev(dev, link, ALL) {
3615 if (dev->class == ATA_DEV_UNKNOWN &&
3616 ehc->classes[dev->devno] != ATA_DEV_NONE)
3623 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3625 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3626 u64 now = get_jiffies_64();
3627 int *trials = void_arg;
3629 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3630 (ent->timestamp < now - min(now, interval)))
3637 static int ata_eh_schedule_probe(struct ata_device *dev)
3639 struct ata_eh_context *ehc = &dev->link->eh_context;
3640 struct ata_link *link = ata_dev_phys_link(dev);
3643 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3644 (ehc->did_probe_mask & (1 << dev->devno)))
3647 ata_eh_detach_dev(dev);
3649 ehc->did_probe_mask |= (1 << dev->devno);
3650 ehc->i.action |= ATA_EH_RESET;
3651 ehc->saved_xfer_mode[dev->devno] = 0;
3652 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3654 /* the link maybe in a deep sleep, wake it up */
3655 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3656 if (ata_is_host_link(link))
3657 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3660 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3664 /* Record and count probe trials on the ering. The specific
3665 * error mask used is irrelevant. Because a successful device
3666 * detection clears the ering, this count accumulates only if
3667 * there are consecutive failed probes.
3669 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3670 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3671 * forced to 1.5Gbps.
3673 * This is to work around cases where failed link speed
3674 * negotiation results in device misdetection leading to
3675 * infinite DEVXCHG or PHRDY CHG events.
3677 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3678 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3680 if (trials > ATA_EH_PROBE_TRIALS)
3681 sata_down_spd_limit(link, 1);
3686 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3688 struct ata_eh_context *ehc = &dev->link->eh_context;
3690 /* -EAGAIN from EH routine indicates retry without prejudice.
3691 * The requester is responsible for ensuring forward progress.
3694 ehc->tries[dev->devno]--;
3698 /* device missing or wrong IDENTIFY data, schedule probing */
3699 ehc->i.probe_mask |= (1 << dev->devno);
3702 /* give it just one more chance */
3703 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3706 if (ehc->tries[dev->devno] == 1) {
3707 /* This is the last chance, better to slow
3708 * down than lose it.
3710 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3711 if (dev->pio_mode > XFER_PIO_0)
3712 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3716 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3717 /* disable device if it has used up all its chances */
3718 ata_dev_disable(dev);
3720 /* detach if offline */
3721 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3722 ata_eh_detach_dev(dev);
3724 /* schedule probe if necessary */
3725 if (ata_eh_schedule_probe(dev)) {
3726 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3727 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3728 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3733 ehc->i.action |= ATA_EH_RESET;
3739 * ata_eh_recover - recover host port after error
3740 * @ap: host port to recover
3741 * @prereset: prereset method (can be NULL)
3742 * @softreset: softreset method (can be NULL)
3743 * @hardreset: hardreset method (can be NULL)
3744 * @postreset: postreset method (can be NULL)
3745 * @r_failed_link: out parameter for failed link
3747 * This is the alpha and omega, eum and yang, heart and soul of
3748 * libata exception handling. On entry, actions required to
3749 * recover each link and hotplug requests are recorded in the
3750 * link's eh_context. This function executes all the operations
3751 * with appropriate retrials and fallbacks to resurrect failed
3752 * devices, detach goners and greet newcomers.
3755 * Kernel thread context (may sleep).
3758 * 0 on success, -errno on failure.
3760 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3761 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3762 ata_postreset_fn_t postreset,
3763 struct ata_link **r_failed_link)
3765 struct ata_link *link;
3766 struct ata_device *dev;
3768 unsigned long flags, deadline;
3772 /* prep for recovery */
3773 ata_for_each_link(link, ap, EDGE) {
3774 struct ata_eh_context *ehc = &link->eh_context;
3776 /* re-enable link? */
3777 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3778 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3779 spin_lock_irqsave(ap->lock, flags);
3780 link->flags &= ~ATA_LFLAG_DISABLED;
3781 spin_unlock_irqrestore(ap->lock, flags);
3782 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3785 ata_for_each_dev(dev, link, ALL) {
3786 if (link->flags & ATA_LFLAG_NO_RETRY)
3787 ehc->tries[dev->devno] = 1;
3789 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3791 /* collect port action mask recorded in dev actions */
3792 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3793 ~ATA_EH_PERDEV_MASK;
3794 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3796 /* process hotplug request */
3797 if (dev->flags & ATA_DFLAG_DETACH)
3798 ata_eh_detach_dev(dev);
3800 /* schedule probe if necessary */
3801 if (!ata_dev_enabled(dev))
3802 ata_eh_schedule_probe(dev);
3809 /* if UNLOADING, finish immediately */
3810 if (ap->pflags & ATA_PFLAG_UNLOADING)
3814 ata_for_each_link(link, ap, EDGE) {
3815 struct ata_eh_context *ehc = &link->eh_context;
3817 /* skip EH if possible. */
3818 if (ata_eh_skip_recovery(link))
3821 ata_for_each_dev(dev, link, ALL)
3822 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3826 ata_for_each_link(link, ap, EDGE) {
3827 struct ata_eh_context *ehc = &link->eh_context;
3829 if (!(ehc->i.action & ATA_EH_RESET))
3832 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3833 prereset, softreset, hardreset, postreset);
3835 ata_link_err(link, "reset failed, giving up\n");
3844 * clears ATA_EH_PARK in eh_info and resets
3845 * ap->park_req_pending
3847 ata_eh_pull_park_action(ap);
3850 ata_for_each_link(link, ap, EDGE) {
3851 ata_for_each_dev(dev, link, ALL) {
3852 struct ata_eh_context *ehc = &link->eh_context;
3855 if (dev->class != ATA_DEV_ATA &&
3856 dev->class != ATA_DEV_ZAC)
3858 if (!(ehc->i.dev_action[dev->devno] &
3861 tmp = dev->unpark_deadline;
3862 if (time_before(deadline, tmp))
3864 else if (time_before_eq(tmp, jiffies))
3866 if (ehc->unloaded_mask & (1 << dev->devno))
3869 ata_eh_park_issue_cmd(dev, 1);
3874 if (time_before_eq(deadline, now))
3878 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3882 ata_for_each_link(link, ap, EDGE) {
3883 ata_for_each_dev(dev, link, ALL) {
3884 if (!(link->eh_context.unloaded_mask &
3888 ata_eh_park_issue_cmd(dev, 0);
3889 ata_eh_done(link, dev, ATA_EH_PARK);
3895 ata_for_each_link(link, ap, PMP_FIRST) {
3896 struct ata_eh_context *ehc = &link->eh_context;
3898 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3901 /* revalidate existing devices and attach new ones */
3902 rc = ata_eh_revalidate_and_attach(link, &dev);
3906 /* if PMP got attached, return, pmp EH will take care of it */
3907 if (link->device->class == ATA_DEV_PMP) {
3912 /* configure transfer mode if necessary */
3913 if (ehc->i.flags & ATA_EHI_SETMODE) {
3914 rc = ata_set_mode(link, &dev);
3917 ehc->i.flags &= ~ATA_EHI_SETMODE;
3920 /* If reset has been issued, clear UA to avoid
3921 * disrupting the current users of the device.
3923 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3924 ata_for_each_dev(dev, link, ALL) {
3925 if (dev->class != ATA_DEV_ATAPI)
3927 rc = atapi_eh_clear_ua(dev);
3930 if (zpodd_dev_enabled(dev))
3931 zpodd_post_poweron(dev);
3935 /* retry flush if necessary */
3936 ata_for_each_dev(dev, link, ALL) {
3937 if (dev->class != ATA_DEV_ATA &&
3938 dev->class != ATA_DEV_ZAC)
3940 rc = ata_eh_maybe_retry_flush(dev);
3946 /* configure link power saving */
3947 if (link->lpm_policy != ap->target_lpm_policy) {
3948 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3953 /* this link is okay now */
3960 ata_eh_handle_dev_fail(dev, rc);
3962 if (ap->pflags & ATA_PFLAG_FROZEN) {
3963 /* PMP reset requires working host port.
3964 * Can't retry if it's frozen.
3966 if (sata_pmp_attached(ap))
3976 if (rc && r_failed_link)
3977 *r_failed_link = link;
3979 DPRINTK("EXIT, rc=%d\n", rc);
3984 * ata_eh_finish - finish up EH
3985 * @ap: host port to finish EH for
3987 * Recovery is complete. Clean up EH states and retry or finish
3993 void ata_eh_finish(struct ata_port *ap)
3997 /* retry or finish qcs */
3998 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3999 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
4001 if (!(qc->flags & ATA_QCFLAG_FAILED))
4005 /* FIXME: Once EH migration is complete,
4006 * generate sense data in this function,
4007 * considering both err_mask and tf.
4009 if (qc->flags & ATA_QCFLAG_RETRY)
4010 ata_eh_qc_retry(qc);
4012 ata_eh_qc_complete(qc);
4014 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
4015 ata_eh_qc_complete(qc);
4017 /* feed zero TF to sense generation */
4018 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
4019 ata_eh_qc_retry(qc);
4024 /* make sure nr_active_links is zero after EH */
4025 WARN_ON(ap->nr_active_links);
4026 ap->nr_active_links = 0;
4030 * ata_do_eh - do standard error handling
4031 * @ap: host port to handle error for
4033 * @prereset: prereset method (can be NULL)
4034 * @softreset: softreset method (can be NULL)
4035 * @hardreset: hardreset method (can be NULL)
4036 * @postreset: postreset method (can be NULL)
4038 * Perform standard error handling sequence.
4041 * Kernel thread context (may sleep).
4043 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4044 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4045 ata_postreset_fn_t postreset)
4047 struct ata_device *dev;
4053 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4056 ata_for_each_dev(dev, &ap->link, ALL)
4057 ata_dev_disable(dev);
4064 * ata_std_error_handler - standard error handler
4065 * @ap: host port to handle error for
4067 * Standard error handler
4070 * Kernel thread context (may sleep).
4072 void ata_std_error_handler(struct ata_port *ap)
4074 struct ata_port_operations *ops = ap->ops;
4075 ata_reset_fn_t hardreset = ops->hardreset;
4077 /* ignore built-in hardreset if SCR access is not available */
4078 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4081 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4086 * ata_eh_handle_port_suspend - perform port suspend operation
4087 * @ap: port to suspend
4092 * Kernel thread context (may sleep).
4094 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4096 unsigned long flags;
4098 struct ata_device *dev;
4100 /* are we suspending? */
4101 spin_lock_irqsave(ap->lock, flags);
4102 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4103 ap->pm_mesg.event & PM_EVENT_RESUME) {
4104 spin_unlock_irqrestore(ap->lock, flags);
4107 spin_unlock_irqrestore(ap->lock, flags);
4109 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4112 * If we have a ZPODD attached, check its zero
4113 * power ready status before the port is frozen.
4114 * Only needed for runtime suspend.
4116 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4117 ata_for_each_dev(dev, &ap->link, ENABLED) {
4118 if (zpodd_dev_enabled(dev))
4119 zpodd_on_suspend(dev);
4123 /* tell ACPI we're suspending */
4124 rc = ata_acpi_on_suspend(ap);
4129 ata_eh_freeze_port(ap);
4131 if (ap->ops->port_suspend)
4132 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4134 ata_acpi_set_state(ap, ap->pm_mesg);
4136 /* update the flags */
4137 spin_lock_irqsave(ap->lock, flags);
4139 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4141 ap->pflags |= ATA_PFLAG_SUSPENDED;
4142 else if (ap->pflags & ATA_PFLAG_FROZEN)
4143 ata_port_schedule_eh(ap);
4145 spin_unlock_irqrestore(ap->lock, flags);
4151 * ata_eh_handle_port_resume - perform port resume operation
4152 * @ap: port to resume
4157 * Kernel thread context (may sleep).
4159 static void ata_eh_handle_port_resume(struct ata_port *ap)
4161 struct ata_link *link;
4162 struct ata_device *dev;
4163 unsigned long flags;
4165 /* are we resuming? */
4166 spin_lock_irqsave(ap->lock, flags);
4167 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4168 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4169 spin_unlock_irqrestore(ap->lock, flags);
4172 spin_unlock_irqrestore(ap->lock, flags);
4174 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4177 * Error timestamps are in jiffies which doesn't run while
4178 * suspended and PHY events during resume isn't too uncommon.
4179 * When the two are combined, it can lead to unnecessary speed
4180 * downs if the machine is suspended and resumed repeatedly.
4181 * Clear error history.
4183 ata_for_each_link(link, ap, HOST_FIRST)
4184 ata_for_each_dev(dev, link, ALL)
4185 ata_ering_clear(&dev->ering);
4187 ata_acpi_set_state(ap, ap->pm_mesg);
4189 if (ap->ops->port_resume)
4190 ap->ops->port_resume(ap);
4192 /* tell ACPI that we're resuming */
4193 ata_acpi_on_resume(ap);
4195 /* update the flags */
4196 spin_lock_irqsave(ap->lock, flags);
4197 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4198 spin_unlock_irqrestore(ap->lock, flags);
4200 #endif /* CONFIG_PM */