2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include "../scsi/scsi_transport_api.h"
45 #include <linux/libata.h>
50 /* speed down verdicts */
51 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
54 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
57 ATA_EFLAG_IS_IO = (1 << 0),
58 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
60 /* error categories */
63 ATA_ECAT_TOUT_HSM = 2,
65 ATA_ECAT_DUBIOUS_NONE = 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_EH_CMD_DFL_TIMEOUT = 5000,
73 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN = 5000,
76 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
81 ATA_EH_PRERESET_TIMEOUT = 10000,
82 ATA_EH_FASTDRAIN_INTERVAL = 3000,
87 /* The following table determines how we sequence resets. Each entry
88 * represents timeout for that try. The first try can be soft or
89 * hardreset. All others are hardreset if available. In most cases
90 * the first reset w/ 10sec timeout should succeed. Following entries
91 * are mostly for error handling, hotplug and retarded devices.
93 static const unsigned long ata_eh_reset_timeouts[] = {
94 10000, /* most drives spin up by 10sec */
95 10000, /* > 99% working drives spin up before 20sec */
96 35000, /* give > 30 secs of idleness for retarded devices */
97 5000, /* and sweet one last chance */
98 ULONG_MAX, /* > 1 min has elapsed, give up */
101 static const unsigned long ata_eh_identify_timeouts[] = {
102 5000, /* covers > 99% of successes and not too boring on failures */
103 10000, /* combined time till here is enough even for media access */
104 30000, /* for true idiots */
108 static const unsigned long ata_eh_other_timeouts[] = {
109 5000, /* same rationale as identify timeout */
111 /* but no merciful 30sec for other commands, it just isn't worth it */
115 struct ata_eh_cmd_timeout_ent {
117 const unsigned long *timeouts;
120 /* The following table determines timeouts to use for EH internal
121 * commands. Each table entry is a command class and matches the
122 * commands the entry applies to and the timeout table to use.
124 * On the retry after a command timed out, the next timeout value from
125 * the table is used. If the table doesn't contain further entries,
126 * the last value is used.
128 * ehc->cmd_timeout_idx keeps track of which timeout to use per
129 * command class, so if SET_FEATURES times out on the first try, the
130 * next try will use the second timeout value only for that class.
132 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
133 static const struct ata_eh_cmd_timeout_ent
134 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
135 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
136 .timeouts = ata_eh_identify_timeouts, },
137 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
138 .timeouts = ata_eh_other_timeouts, },
139 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
140 .timeouts = ata_eh_other_timeouts, },
141 { .commands = CMDS(ATA_CMD_SET_FEATURES),
142 .timeouts = ata_eh_other_timeouts, },
143 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
144 .timeouts = ata_eh_other_timeouts, },
148 static void __ata_port_freeze(struct ata_port *ap);
150 static void ata_eh_handle_port_suspend(struct ata_port *ap);
151 static void ata_eh_handle_port_resume(struct ata_port *ap);
152 #else /* CONFIG_PM */
153 static void ata_eh_handle_port_suspend(struct ata_port *ap)
156 static void ata_eh_handle_port_resume(struct ata_port *ap)
158 #endif /* CONFIG_PM */
160 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
163 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
164 ATA_EH_DESC_LEN - ehi->desc_len,
169 * __ata_ehi_push_desc - push error description without adding separator
171 * @fmt: printf format string
173 * Format string according to @fmt and append it to @ehi->desc.
176 * spin_lock_irqsave(host lock)
178 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
183 __ata_ehi_pushv_desc(ehi, fmt, args);
188 * ata_ehi_push_desc - push error description with separator
190 * @fmt: printf format string
192 * Format string according to @fmt and append it to @ehi->desc.
193 * If @ehi->desc is not empty, ", " is added in-between.
196 * spin_lock_irqsave(host lock)
198 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
203 __ata_ehi_push_desc(ehi, ", ");
206 __ata_ehi_pushv_desc(ehi, fmt, args);
211 * ata_ehi_clear_desc - clean error description
217 * spin_lock_irqsave(host lock)
219 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
226 * ata_port_desc - append port description
227 * @ap: target ATA port
228 * @fmt: printf format string
230 * Format string according to @fmt and append it to port
231 * description. If port description is not empty, " " is added
232 * in-between. This function is to be used while initializing
233 * ata_host. The description is printed on host registration.
238 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
242 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
244 if (ap->link.eh_info.desc_len)
245 __ata_ehi_push_desc(&ap->link.eh_info, " ");
248 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
255 * ata_port_pbar_desc - append PCI BAR description
256 * @ap: target ATA port
257 * @bar: target PCI BAR
258 * @offset: offset into PCI BAR
259 * @name: name of the area
261 * If @offset is negative, this function formats a string which
262 * contains the name, address, size and type of the BAR and
263 * appends it to the port description. If @offset is zero or
264 * positive, only name and offsetted address is appended.
269 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
272 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
274 unsigned long long start, len;
276 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
278 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
281 start = (unsigned long long)pci_resource_start(pdev, bar);
282 len = (unsigned long long)pci_resource_len(pdev, bar);
285 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
287 ata_port_desc(ap, "%s 0x%llx", name,
288 start + (unsigned long long)offset);
291 #endif /* CONFIG_PCI */
293 static int ata_lookup_timeout_table(u8 cmd)
297 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
300 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
309 * ata_internal_cmd_timeout - determine timeout for an internal command
310 * @dev: target device
311 * @cmd: internal command to be issued
313 * Determine timeout for internal command @cmd for @dev.
319 * Determined timeout.
321 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
323 struct ata_eh_context *ehc = &dev->link->eh_context;
324 int ent = ata_lookup_timeout_table(cmd);
328 return ATA_EH_CMD_DFL_TIMEOUT;
330 idx = ehc->cmd_timeout_idx[dev->devno][ent];
331 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
335 * ata_internal_cmd_timed_out - notification for internal command timeout
336 * @dev: target device
337 * @cmd: internal command which timed out
339 * Notify EH that internal command @cmd for @dev timed out. This
340 * function should be called only for commands whose timeouts are
341 * determined using ata_internal_cmd_timeout().
346 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
348 struct ata_eh_context *ehc = &dev->link->eh_context;
349 int ent = ata_lookup_timeout_table(cmd);
355 idx = ehc->cmd_timeout_idx[dev->devno][ent];
356 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
357 ehc->cmd_timeout_idx[dev->devno][ent]++;
360 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
361 unsigned int err_mask)
363 struct ata_ering_entry *ent;
368 ering->cursor %= ATA_ERING_SIZE;
370 ent = &ering->ring[ering->cursor];
371 ent->eflags = eflags;
372 ent->err_mask = err_mask;
373 ent->timestamp = get_jiffies_64();
376 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
378 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
385 static void ata_ering_clear(struct ata_ering *ering)
387 memset(ering, 0, sizeof(*ering));
390 static int ata_ering_map(struct ata_ering *ering,
391 int (*map_fn)(struct ata_ering_entry *, void *),
395 struct ata_ering_entry *ent;
399 ent = &ering->ring[idx];
402 rc = map_fn(ent, arg);
405 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
406 } while (idx != ering->cursor);
411 static unsigned int ata_eh_dev_action(struct ata_device *dev)
413 struct ata_eh_context *ehc = &dev->link->eh_context;
415 return ehc->i.action | ehc->i.dev_action[dev->devno];
418 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
419 struct ata_eh_info *ehi, unsigned int action)
421 struct ata_device *tdev;
424 ehi->action &= ~action;
425 ata_for_each_dev(tdev, link, ALL)
426 ehi->dev_action[tdev->devno] &= ~action;
428 /* doesn't make sense for port-wide EH actions */
429 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
431 /* break ehi->action into ehi->dev_action */
432 if (ehi->action & action) {
433 ata_for_each_dev(tdev, link, ALL)
434 ehi->dev_action[tdev->devno] |=
435 ehi->action & action;
436 ehi->action &= ~action;
439 /* turn off the specified per-dev action */
440 ehi->dev_action[dev->devno] &= ~action;
445 * ata_scsi_timed_out - SCSI layer time out callback
446 * @cmd: timed out SCSI command
448 * Handles SCSI layer timeout. We race with normal completion of
449 * the qc for @cmd. If the qc is already gone, we lose and let
450 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
451 * timed out and EH should be invoked. Prevent ata_qc_complete()
452 * from finishing it by setting EH_SCHEDULED and return
455 * TODO: kill this function once old EH is gone.
458 * Called from timer context
461 * EH_HANDLED or EH_NOT_HANDLED
463 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
465 struct Scsi_Host *host = cmd->device->host;
466 struct ata_port *ap = ata_shost_to_port(host);
468 struct ata_queued_cmd *qc;
469 enum blk_eh_timer_return ret;
473 if (ap->ops->error_handler) {
474 ret = BLK_EH_NOT_HANDLED;
478 ret = BLK_EH_HANDLED;
479 spin_lock_irqsave(ap->lock, flags);
480 qc = ata_qc_from_tag(ap, ap->link.active_tag);
482 WARN_ON(qc->scsicmd != cmd);
483 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
484 qc->err_mask |= AC_ERR_TIMEOUT;
485 ret = BLK_EH_NOT_HANDLED;
487 spin_unlock_irqrestore(ap->lock, flags);
490 DPRINTK("EXIT, ret=%d\n", ret);
494 static void ata_eh_unload(struct ata_port *ap)
496 struct ata_link *link;
497 struct ata_device *dev;
500 /* Restore SControl IPM and SPD for the next driver and
501 * disable attached devices.
503 ata_for_each_link(link, ap, PMP_FIRST) {
504 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
505 ata_for_each_dev(dev, link, ALL)
506 ata_dev_disable(dev);
509 /* freeze and set UNLOADED */
510 spin_lock_irqsave(ap->lock, flags);
512 ata_port_freeze(ap); /* won't be thawed */
513 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
514 ap->pflags |= ATA_PFLAG_UNLOADED;
516 spin_unlock_irqrestore(ap->lock, flags);
520 * ata_scsi_error - SCSI layer error handler callback
521 * @host: SCSI host on which error occurred
523 * Handles SCSI-layer-thrown error events.
526 * Inherited from SCSI layer (none, can sleep)
531 void ata_scsi_error(struct Scsi_Host *host)
533 struct ata_port *ap = ata_shost_to_port(host);
539 /* synchronize with port task */
540 ata_port_flush_task(ap);
542 /* synchronize with host lock and sort out timeouts */
544 /* For new EH, all qcs are finished in one of three ways -
545 * normal completion, error completion, and SCSI timeout.
546 * Both cmpletions can race against SCSI timeout. When normal
547 * completion wins, the qc never reaches EH. When error
548 * completion wins, the qc has ATA_QCFLAG_FAILED set.
550 * When SCSI timeout wins, things are a bit more complex.
551 * Normal or error completion can occur after the timeout but
552 * before this point. In such cases, both types of
553 * completions are honored. A scmd is determined to have
554 * timed out iff its associated qc is active and not failed.
556 if (ap->ops->error_handler) {
557 struct scsi_cmnd *scmd, *tmp;
560 spin_lock_irqsave(ap->lock, flags);
562 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
563 struct ata_queued_cmd *qc;
565 for (i = 0; i < ATA_MAX_QUEUE; i++) {
566 qc = __ata_qc_from_tag(ap, i);
567 if (qc->flags & ATA_QCFLAG_ACTIVE &&
572 if (i < ATA_MAX_QUEUE) {
573 /* the scmd has an associated qc */
574 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
575 /* which hasn't failed yet, timeout */
576 qc->err_mask |= AC_ERR_TIMEOUT;
577 qc->flags |= ATA_QCFLAG_FAILED;
581 /* Normal completion occurred after
582 * SCSI timeout but before this point.
583 * Successfully complete it.
585 scmd->retries = scmd->allowed;
586 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
590 /* If we have timed out qcs. They belong to EH from
591 * this point but the state of the controller is
592 * unknown. Freeze the port to make sure the IRQ
593 * handler doesn't diddle with those qcs. This must
594 * be done atomically w.r.t. setting QCFLAG_FAILED.
597 __ata_port_freeze(ap);
599 spin_unlock_irqrestore(ap->lock, flags);
601 /* initialize eh_tries */
602 ap->eh_tries = ATA_EH_MAX_TRIES;
604 spin_unlock_wait(ap->lock);
607 /* invoke error handler */
608 if (ap->ops->error_handler) {
609 struct ata_link *link;
611 /* kill fast drain timer */
612 del_timer_sync(&ap->fastdrain_timer);
614 /* process port resume request */
615 ata_eh_handle_port_resume(ap);
617 /* fetch & clear EH info */
618 spin_lock_irqsave(ap->lock, flags);
620 ata_for_each_link(link, ap, HOST_FIRST) {
621 struct ata_eh_context *ehc = &link->eh_context;
622 struct ata_device *dev;
624 memset(&link->eh_context, 0, sizeof(link->eh_context));
625 link->eh_context.i = link->eh_info;
626 memset(&link->eh_info, 0, sizeof(link->eh_info));
628 ata_for_each_dev(dev, link, ENABLED) {
629 int devno = dev->devno;
631 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
632 if (ata_ncq_enabled(dev))
633 ehc->saved_ncq_enabled |= 1 << devno;
637 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
638 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
639 ap->excl_link = NULL; /* don't maintain exclusion over EH */
641 spin_unlock_irqrestore(ap->lock, flags);
643 /* invoke EH, skip if unloading or suspended */
644 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
645 ap->ops->error_handler(ap);
647 /* if unloading, commence suicide */
648 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
649 !(ap->pflags & ATA_PFLAG_UNLOADED))
654 /* process port suspend request */
655 ata_eh_handle_port_suspend(ap);
657 /* Exception might have happend after ->error_handler
658 * recovered the port but before this point. Repeat
661 spin_lock_irqsave(ap->lock, flags);
663 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
664 if (--ap->eh_tries) {
665 spin_unlock_irqrestore(ap->lock, flags);
668 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
669 "tries, giving up\n", ATA_EH_MAX_TRIES);
670 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
673 /* this run is complete, make sure EH info is clear */
674 ata_for_each_link(link, ap, HOST_FIRST)
675 memset(&link->eh_info, 0, sizeof(link->eh_info));
677 /* Clear host_eh_scheduled while holding ap->lock such
678 * that if exception occurs after this point but
679 * before EH completion, SCSI midlayer will
682 host->host_eh_scheduled = 0;
684 spin_unlock_irqrestore(ap->lock, flags);
686 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
687 ap->ops->eng_timeout(ap);
690 /* finish or retry handled scmd's and clean up */
691 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
693 scsi_eh_flush_done_q(&ap->eh_done_q);
696 spin_lock_irqsave(ap->lock, flags);
698 if (ap->pflags & ATA_PFLAG_LOADING)
699 ap->pflags &= ~ATA_PFLAG_LOADING;
700 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
701 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
703 if (ap->pflags & ATA_PFLAG_RECOVERED)
704 ata_port_printk(ap, KERN_INFO, "EH complete\n");
706 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
708 /* tell wait_eh that we're done */
709 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
710 wake_up_all(&ap->eh_wait_q);
712 spin_unlock_irqrestore(ap->lock, flags);
718 * ata_port_wait_eh - Wait for the currently pending EH to complete
719 * @ap: Port to wait EH for
721 * Wait until the currently pending EH is complete.
724 * Kernel thread context (may sleep).
726 void ata_port_wait_eh(struct ata_port *ap)
732 spin_lock_irqsave(ap->lock, flags);
734 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
735 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
736 spin_unlock_irqrestore(ap->lock, flags);
738 spin_lock_irqsave(ap->lock, flags);
740 finish_wait(&ap->eh_wait_q, &wait);
742 spin_unlock_irqrestore(ap->lock, flags);
744 /* make sure SCSI EH is complete */
745 if (scsi_host_in_recovery(ap->scsi_host)) {
751 static int ata_eh_nr_in_flight(struct ata_port *ap)
756 /* count only non-internal commands */
757 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
758 if (ata_qc_from_tag(ap, tag))
764 void ata_eh_fastdrain_timerfn(unsigned long arg)
766 struct ata_port *ap = (void *)arg;
770 spin_lock_irqsave(ap->lock, flags);
772 cnt = ata_eh_nr_in_flight(ap);
778 if (cnt == ap->fastdrain_cnt) {
781 /* No progress during the last interval, tag all
782 * in-flight qcs as timed out and freeze the port.
784 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
785 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
787 qc->err_mask |= AC_ERR_TIMEOUT;
792 /* some qcs have finished, give it another chance */
793 ap->fastdrain_cnt = cnt;
794 ap->fastdrain_timer.expires =
795 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
796 add_timer(&ap->fastdrain_timer);
800 spin_unlock_irqrestore(ap->lock, flags);
804 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
805 * @ap: target ATA port
806 * @fastdrain: activate fast drain
808 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
809 * is non-zero and EH wasn't pending before. Fast drain ensures
810 * that EH kicks in in timely manner.
813 * spin_lock_irqsave(host lock)
815 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
819 /* already scheduled? */
820 if (ap->pflags & ATA_PFLAG_EH_PENDING)
823 ap->pflags |= ATA_PFLAG_EH_PENDING;
828 /* do we have in-flight qcs? */
829 cnt = ata_eh_nr_in_flight(ap);
833 /* activate fast drain */
834 ap->fastdrain_cnt = cnt;
835 ap->fastdrain_timer.expires =
836 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
837 add_timer(&ap->fastdrain_timer);
841 * ata_qc_schedule_eh - schedule qc for error handling
842 * @qc: command to schedule error handling for
844 * Schedule error handling for @qc. EH will kick in as soon as
845 * other commands are drained.
848 * spin_lock_irqsave(host lock)
850 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
852 struct ata_port *ap = qc->ap;
854 WARN_ON(!ap->ops->error_handler);
856 qc->flags |= ATA_QCFLAG_FAILED;
857 ata_eh_set_pending(ap, 1);
859 /* The following will fail if timeout has already expired.
860 * ata_scsi_error() takes care of such scmds on EH entry.
861 * Note that ATA_QCFLAG_FAILED is unconditionally set after
862 * this function completes.
864 blk_abort_request(qc->scsicmd->request);
868 * ata_port_schedule_eh - schedule error handling without a qc
869 * @ap: ATA port to schedule EH for
871 * Schedule error handling for @ap. EH will kick in as soon as
872 * all commands are drained.
875 * spin_lock_irqsave(host lock)
877 void ata_port_schedule_eh(struct ata_port *ap)
879 WARN_ON(!ap->ops->error_handler);
881 if (ap->pflags & ATA_PFLAG_INITIALIZING)
884 ata_eh_set_pending(ap, 1);
885 scsi_schedule_eh(ap->scsi_host);
887 DPRINTK("port EH scheduled\n");
890 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
892 int tag, nr_aborted = 0;
894 WARN_ON(!ap->ops->error_handler);
896 /* we're gonna abort all commands, no need for fast drain */
897 ata_eh_set_pending(ap, 0);
899 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
900 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
902 if (qc && (!link || qc->dev->link == link)) {
903 qc->flags |= ATA_QCFLAG_FAILED;
910 ata_port_schedule_eh(ap);
916 * ata_link_abort - abort all qc's on the link
917 * @link: ATA link to abort qc's for
919 * Abort all active qc's active on @link and schedule EH.
922 * spin_lock_irqsave(host lock)
925 * Number of aborted qc's.
927 int ata_link_abort(struct ata_link *link)
929 return ata_do_link_abort(link->ap, link);
933 * ata_port_abort - abort all qc's on the port
934 * @ap: ATA port to abort qc's for
936 * Abort all active qc's of @ap and schedule EH.
939 * spin_lock_irqsave(host_set lock)
942 * Number of aborted qc's.
944 int ata_port_abort(struct ata_port *ap)
946 return ata_do_link_abort(ap, NULL);
950 * __ata_port_freeze - freeze port
951 * @ap: ATA port to freeze
953 * This function is called when HSM violation or some other
954 * condition disrupts normal operation of the port. Frozen port
955 * is not allowed to perform any operation until the port is
956 * thawed, which usually follows a successful reset.
958 * ap->ops->freeze() callback can be used for freezing the port
959 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
960 * port cannot be frozen hardware-wise, the interrupt handler
961 * must ack and clear interrupts unconditionally while the port
965 * spin_lock_irqsave(host lock)
967 static void __ata_port_freeze(struct ata_port *ap)
969 WARN_ON(!ap->ops->error_handler);
974 ap->pflags |= ATA_PFLAG_FROZEN;
976 DPRINTK("ata%u port frozen\n", ap->print_id);
980 * ata_port_freeze - abort & freeze port
981 * @ap: ATA port to freeze
983 * Abort and freeze @ap.
986 * spin_lock_irqsave(host lock)
989 * Number of aborted commands.
991 int ata_port_freeze(struct ata_port *ap)
995 WARN_ON(!ap->ops->error_handler);
997 nr_aborted = ata_port_abort(ap);
998 __ata_port_freeze(ap);
1004 * sata_async_notification - SATA async notification handler
1005 * @ap: ATA port where async notification is received
1007 * Handler to be called when async notification via SDB FIS is
1008 * received. This function schedules EH if necessary.
1011 * spin_lock_irqsave(host lock)
1014 * 1 if EH is scheduled, 0 otherwise.
1016 int sata_async_notification(struct ata_port *ap)
1021 if (!(ap->flags & ATA_FLAG_AN))
1024 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1026 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1028 if (!sata_pmp_attached(ap) || rc) {
1029 /* PMP is not attached or SNTF is not available */
1030 if (!sata_pmp_attached(ap)) {
1031 /* PMP is not attached. Check whether ATAPI
1032 * AN is configured. If so, notify media
1035 struct ata_device *dev = ap->link.device;
1037 if ((dev->class == ATA_DEV_ATAPI) &&
1038 (dev->flags & ATA_DFLAG_AN))
1039 ata_scsi_media_change_notify(dev);
1042 /* PMP is attached but SNTF is not available.
1043 * ATAPI async media change notification is
1044 * not used. The PMP must be reporting PHY
1045 * status change, schedule EH.
1047 ata_port_schedule_eh(ap);
1051 /* PMP is attached and SNTF is available */
1052 struct ata_link *link;
1054 /* check and notify ATAPI AN */
1055 ata_for_each_link(link, ap, EDGE) {
1056 if (!(sntf & (1 << link->pmp)))
1059 if ((link->device->class == ATA_DEV_ATAPI) &&
1060 (link->device->flags & ATA_DFLAG_AN))
1061 ata_scsi_media_change_notify(link->device);
1064 /* If PMP is reporting that PHY status of some
1065 * downstream ports has changed, schedule EH.
1067 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1068 ata_port_schedule_eh(ap);
1077 * ata_eh_freeze_port - EH helper to freeze port
1078 * @ap: ATA port to freeze
1085 void ata_eh_freeze_port(struct ata_port *ap)
1087 unsigned long flags;
1089 if (!ap->ops->error_handler)
1092 spin_lock_irqsave(ap->lock, flags);
1093 __ata_port_freeze(ap);
1094 spin_unlock_irqrestore(ap->lock, flags);
1098 * ata_port_thaw_port - EH helper to thaw port
1099 * @ap: ATA port to thaw
1101 * Thaw frozen port @ap.
1106 void ata_eh_thaw_port(struct ata_port *ap)
1108 unsigned long flags;
1110 if (!ap->ops->error_handler)
1113 spin_lock_irqsave(ap->lock, flags);
1115 ap->pflags &= ~ATA_PFLAG_FROZEN;
1120 spin_unlock_irqrestore(ap->lock, flags);
1122 DPRINTK("ata%u port thawed\n", ap->print_id);
1125 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1130 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1132 struct ata_port *ap = qc->ap;
1133 struct scsi_cmnd *scmd = qc->scsicmd;
1134 unsigned long flags;
1136 spin_lock_irqsave(ap->lock, flags);
1137 qc->scsidone = ata_eh_scsidone;
1138 __ata_qc_complete(qc);
1139 WARN_ON(ata_tag_valid(qc->tag));
1140 spin_unlock_irqrestore(ap->lock, flags);
1142 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1146 * ata_eh_qc_complete - Complete an active ATA command from EH
1147 * @qc: Command to complete
1149 * Indicate to the mid and upper layers that an ATA command has
1150 * completed. To be used from EH.
1152 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1154 struct scsi_cmnd *scmd = qc->scsicmd;
1155 scmd->retries = scmd->allowed;
1156 __ata_eh_qc_complete(qc);
1160 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1161 * @qc: Command to retry
1163 * Indicate to the mid and upper layers that an ATA command
1164 * should be retried. To be used from EH.
1166 * SCSI midlayer limits the number of retries to scmd->allowed.
1167 * scmd->retries is decremented for commands which get retried
1168 * due to unrelated failures (qc->err_mask is zero).
1170 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1172 struct scsi_cmnd *scmd = qc->scsicmd;
1173 if (!qc->err_mask && scmd->retries)
1175 __ata_eh_qc_complete(qc);
1179 * ata_dev_disable - disable ATA device
1180 * @dev: ATA device to disable
1187 void ata_dev_disable(struct ata_device *dev)
1189 if (!ata_dev_enabled(dev))
1192 if (ata_msg_drv(dev->link->ap))
1193 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1194 ata_acpi_on_disable(dev);
1195 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1200 * ata_eh_detach_dev - detach ATA device
1201 * @dev: ATA device to detach
1208 void ata_eh_detach_dev(struct ata_device *dev)
1210 struct ata_link *link = dev->link;
1211 struct ata_port *ap = link->ap;
1212 struct ata_eh_context *ehc = &link->eh_context;
1213 unsigned long flags;
1215 ata_dev_disable(dev);
1217 spin_lock_irqsave(ap->lock, flags);
1219 dev->flags &= ~ATA_DFLAG_DETACH;
1221 if (ata_scsi_offline_dev(dev)) {
1222 dev->flags |= ATA_DFLAG_DETACHED;
1223 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1226 /* clear per-dev EH info */
1227 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1228 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1229 ehc->saved_xfer_mode[dev->devno] = 0;
1230 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1232 spin_unlock_irqrestore(ap->lock, flags);
1236 * ata_eh_about_to_do - about to perform eh_action
1237 * @link: target ATA link
1238 * @dev: target ATA dev for per-dev action (can be NULL)
1239 * @action: action about to be performed
1241 * Called just before performing EH actions to clear related bits
1242 * in @link->eh_info such that eh actions are not unnecessarily
1248 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1249 unsigned int action)
1251 struct ata_port *ap = link->ap;
1252 struct ata_eh_info *ehi = &link->eh_info;
1253 struct ata_eh_context *ehc = &link->eh_context;
1254 unsigned long flags;
1256 spin_lock_irqsave(ap->lock, flags);
1258 ata_eh_clear_action(link, dev, ehi, action);
1260 /* About to take EH action, set RECOVERED. Ignore actions on
1261 * slave links as master will do them again.
1263 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1264 ap->pflags |= ATA_PFLAG_RECOVERED;
1266 spin_unlock_irqrestore(ap->lock, flags);
1270 * ata_eh_done - EH action complete
1271 * @ap: target ATA port
1272 * @dev: target ATA dev for per-dev action (can be NULL)
1273 * @action: action just completed
1275 * Called right after performing EH actions to clear related bits
1276 * in @link->eh_context.
1281 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1282 unsigned int action)
1284 struct ata_eh_context *ehc = &link->eh_context;
1286 ata_eh_clear_action(link, dev, &ehc->i, action);
1290 * ata_err_string - convert err_mask to descriptive string
1291 * @err_mask: error mask to convert to string
1293 * Convert @err_mask to descriptive string. Errors are
1294 * prioritized according to severity and only the most severe
1295 * error is reported.
1301 * Descriptive string for @err_mask
1303 static const char *ata_err_string(unsigned int err_mask)
1305 if (err_mask & AC_ERR_HOST_BUS)
1306 return "host bus error";
1307 if (err_mask & AC_ERR_ATA_BUS)
1308 return "ATA bus error";
1309 if (err_mask & AC_ERR_TIMEOUT)
1311 if (err_mask & AC_ERR_HSM)
1312 return "HSM violation";
1313 if (err_mask & AC_ERR_SYSTEM)
1314 return "internal error";
1315 if (err_mask & AC_ERR_MEDIA)
1316 return "media error";
1317 if (err_mask & AC_ERR_INVALID)
1318 return "invalid argument";
1319 if (err_mask & AC_ERR_DEV)
1320 return "device error";
1321 return "unknown error";
1325 * ata_read_log_page - read a specific log page
1326 * @dev: target device
1327 * @page: page to read
1328 * @buf: buffer to store read page
1329 * @sectors: number of sectors to read
1331 * Read log page using READ_LOG_EXT command.
1334 * Kernel thread context (may sleep).
1337 * 0 on success, AC_ERR_* mask otherwise.
1339 static unsigned int ata_read_log_page(struct ata_device *dev,
1340 u8 page, void *buf, unsigned int sectors)
1342 struct ata_taskfile tf;
1343 unsigned int err_mask;
1345 DPRINTK("read log page - page %d\n", page);
1347 ata_tf_init(dev, &tf);
1348 tf.command = ATA_CMD_READ_LOG_EXT;
1351 tf.hob_nsect = sectors >> 8;
1352 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1353 tf.protocol = ATA_PROT_PIO;
1355 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1356 buf, sectors * ATA_SECT_SIZE, 0);
1358 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1363 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1364 * @dev: Device to read log page 10h from
1365 * @tag: Resulting tag of the failed command
1366 * @tf: Resulting taskfile registers of the failed command
1368 * Read log page 10h to obtain NCQ error details and clear error
1372 * Kernel thread context (may sleep).
1375 * 0 on success, -errno otherwise.
1377 static int ata_eh_read_log_10h(struct ata_device *dev,
1378 int *tag, struct ata_taskfile *tf)
1380 u8 *buf = dev->link->ap->sector_buf;
1381 unsigned int err_mask;
1385 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1390 for (i = 0; i < ATA_SECT_SIZE; i++)
1393 ata_dev_printk(dev, KERN_WARNING,
1394 "invalid checksum 0x%x on log page 10h\n", csum);
1399 *tag = buf[0] & 0x1f;
1401 tf->command = buf[2];
1402 tf->feature = buf[3];
1406 tf->device = buf[7];
1407 tf->hob_lbal = buf[8];
1408 tf->hob_lbam = buf[9];
1409 tf->hob_lbah = buf[10];
1410 tf->nsect = buf[12];
1411 tf->hob_nsect = buf[13];
1417 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1418 * @dev: target ATAPI device
1419 * @r_sense_key: out parameter for sense_key
1421 * Perform ATAPI TEST_UNIT_READY.
1424 * EH context (may sleep).
1427 * 0 on success, AC_ERR_* mask on failure.
1429 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1431 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1432 struct ata_taskfile tf;
1433 unsigned int err_mask;
1435 ata_tf_init(dev, &tf);
1437 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1438 tf.command = ATA_CMD_PACKET;
1439 tf.protocol = ATAPI_PROT_NODATA;
1441 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1442 if (err_mask == AC_ERR_DEV)
1443 *r_sense_key = tf.feature >> 4;
1448 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1449 * @dev: device to perform REQUEST_SENSE to
1450 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1451 * @dfl_sense_key: default sense key to use
1453 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1454 * SENSE. This function is EH helper.
1457 * Kernel thread context (may sleep).
1460 * 0 on success, AC_ERR_* mask on failure
1462 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1463 u8 *sense_buf, u8 dfl_sense_key)
1465 u8 cdb[ATAPI_CDB_LEN] =
1466 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1467 struct ata_port *ap = dev->link->ap;
1468 struct ata_taskfile tf;
1470 DPRINTK("ATAPI request sense\n");
1472 /* FIXME: is this needed? */
1473 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1475 /* initialize sense_buf with the error register,
1476 * for the case where they are -not- overwritten
1478 sense_buf[0] = 0x70;
1479 sense_buf[2] = dfl_sense_key;
1481 /* some devices time out if garbage left in tf */
1482 ata_tf_init(dev, &tf);
1484 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1485 tf.command = ATA_CMD_PACKET;
1487 /* is it pointless to prefer PIO for "safety reasons"? */
1488 if (ap->flags & ATA_FLAG_PIO_DMA) {
1489 tf.protocol = ATAPI_PROT_DMA;
1490 tf.feature |= ATAPI_PKT_DMA;
1492 tf.protocol = ATAPI_PROT_PIO;
1493 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1497 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1498 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1502 * ata_eh_analyze_serror - analyze SError for a failed port
1503 * @link: ATA link to analyze SError for
1505 * Analyze SError if available and further determine cause of
1511 static void ata_eh_analyze_serror(struct ata_link *link)
1513 struct ata_eh_context *ehc = &link->eh_context;
1514 u32 serror = ehc->i.serror;
1515 unsigned int err_mask = 0, action = 0;
1518 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1519 err_mask |= AC_ERR_ATA_BUS;
1520 action |= ATA_EH_RESET;
1522 if (serror & SERR_PROTOCOL) {
1523 err_mask |= AC_ERR_HSM;
1524 action |= ATA_EH_RESET;
1526 if (serror & SERR_INTERNAL) {
1527 err_mask |= AC_ERR_SYSTEM;
1528 action |= ATA_EH_RESET;
1531 /* Determine whether a hotplug event has occurred. Both
1532 * SError.N/X are considered hotplug events for enabled or
1533 * host links. For disabled PMP links, only N bit is
1534 * considered as X bit is left at 1 for link plugging.
1538 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1539 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1541 hotplug_mask = SERR_PHYRDY_CHG;
1543 if (serror & hotplug_mask)
1544 ata_ehi_hotplugged(&ehc->i);
1546 ehc->i.err_mask |= err_mask;
1547 ehc->i.action |= action;
1551 * ata_eh_analyze_ncq_error - analyze NCQ error
1552 * @link: ATA link to analyze NCQ error for
1554 * Read log page 10h, determine the offending qc and acquire
1555 * error status TF. For NCQ device errors, all LLDDs have to do
1556 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1560 * Kernel thread context (may sleep).
1562 void ata_eh_analyze_ncq_error(struct ata_link *link)
1564 struct ata_port *ap = link->ap;
1565 struct ata_eh_context *ehc = &link->eh_context;
1566 struct ata_device *dev = link->device;
1567 struct ata_queued_cmd *qc;
1568 struct ata_taskfile tf;
1571 /* if frozen, we can't do much */
1572 if (ap->pflags & ATA_PFLAG_FROZEN)
1575 /* is it NCQ device error? */
1576 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1579 /* has LLDD analyzed already? */
1580 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1581 qc = __ata_qc_from_tag(ap, tag);
1583 if (!(qc->flags & ATA_QCFLAG_FAILED))
1590 /* okay, this error is ours */
1591 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1593 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1594 "(errno=%d)\n", rc);
1598 if (!(link->sactive & (1 << tag))) {
1599 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1600 "inactive tag %d\n", tag);
1604 /* we've got the perpetrator, condemn it */
1605 qc = __ata_qc_from_tag(ap, tag);
1606 memcpy(&qc->result_tf, &tf, sizeof(tf));
1607 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1608 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1609 ehc->i.err_mask &= ~AC_ERR_DEV;
1613 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1614 * @qc: qc to analyze
1615 * @tf: Taskfile registers to analyze
1617 * Analyze taskfile of @qc and further determine cause of
1618 * failure. This function also requests ATAPI sense data if
1622 * Kernel thread context (may sleep).
1625 * Determined recovery action
1627 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1628 const struct ata_taskfile *tf)
1630 unsigned int tmp, action = 0;
1631 u8 stat = tf->command, err = tf->feature;
1633 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1634 qc->err_mask |= AC_ERR_HSM;
1635 return ATA_EH_RESET;
1638 if (stat & (ATA_ERR | ATA_DF))
1639 qc->err_mask |= AC_ERR_DEV;
1643 switch (qc->dev->class) {
1646 qc->err_mask |= AC_ERR_ATA_BUS;
1648 qc->err_mask |= AC_ERR_MEDIA;
1650 qc->err_mask |= AC_ERR_INVALID;
1654 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1655 tmp = atapi_eh_request_sense(qc->dev,
1656 qc->scsicmd->sense_buffer,
1657 qc->result_tf.feature >> 4);
1659 /* ATA_QCFLAG_SENSE_VALID is used to
1660 * tell atapi_qc_complete() that sense
1661 * data is already valid.
1663 * TODO: interpret sense data and set
1664 * appropriate err_mask.
1666 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1668 qc->err_mask |= tmp;
1672 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1673 action |= ATA_EH_RESET;
1678 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1683 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1687 base = ATA_ECAT_DUBIOUS_NONE;
1689 if (err_mask & AC_ERR_ATA_BUS)
1690 return base + ATA_ECAT_ATA_BUS;
1692 if (err_mask & AC_ERR_TIMEOUT)
1693 return base + ATA_ECAT_TOUT_HSM;
1695 if (eflags & ATA_EFLAG_IS_IO) {
1696 if (err_mask & AC_ERR_HSM)
1697 return base + ATA_ECAT_TOUT_HSM;
1699 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1700 return base + ATA_ECAT_UNK_DEV;
1706 struct speed_down_verdict_arg {
1709 int nr_errors[ATA_ECAT_NR];
1712 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1714 struct speed_down_verdict_arg *arg = void_arg;
1717 if (ent->timestamp < arg->since)
1720 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1722 arg->nr_errors[cat]++;
1728 * ata_eh_speed_down_verdict - Determine speed down verdict
1729 * @dev: Device of interest
1731 * This function examines error ring of @dev and determines
1732 * whether NCQ needs to be turned off, transfer speed should be
1733 * stepped down, or falling back to PIO is necessary.
1735 * ECAT_ATA_BUS : ATA_BUS error for any command
1737 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1740 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1742 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1743 * data transfer hasn't been verified.
1747 * NCQ_OFF : Turn off NCQ.
1749 * SPEED_DOWN : Speed down transfer speed but don't fall back
1752 * FALLBACK_TO_PIO : Fall back to PIO.
1754 * Even if multiple verdicts are returned, only one action is
1755 * taken per error. An action triggered by non-DUBIOUS errors
1756 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1757 * This is to expedite speed down decisions right after device is
1758 * initially configured.
1760 * The followings are speed down rules. #1 and #2 deal with
1763 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1764 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1766 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1767 * occurred during last 5 mins, NCQ_OFF.
1769 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1770 * ocurred during last 5 mins, FALLBACK_TO_PIO
1772 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1773 * during last 10 mins, NCQ_OFF.
1775 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1776 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1779 * Inherited from caller.
1782 * OR of ATA_EH_SPDN_* flags.
1784 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1786 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1787 u64 j64 = get_jiffies_64();
1788 struct speed_down_verdict_arg arg;
1789 unsigned int verdict = 0;
1791 /* scan past 5 mins of error history */
1792 memset(&arg, 0, sizeof(arg));
1793 arg.since = j64 - min(j64, j5mins);
1794 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1796 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1797 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1798 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1799 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1801 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1802 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1803 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1805 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1806 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1807 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1808 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1810 /* scan past 10 mins of error history */
1811 memset(&arg, 0, sizeof(arg));
1812 arg.since = j64 - min(j64, j10mins);
1813 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1815 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1816 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1817 verdict |= ATA_EH_SPDN_NCQ_OFF;
1819 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1820 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1821 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1822 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1828 * ata_eh_speed_down - record error and speed down if necessary
1829 * @dev: Failed device
1830 * @eflags: mask of ATA_EFLAG_* flags
1831 * @err_mask: err_mask of the error
1833 * Record error and examine error history to determine whether
1834 * adjusting transmission speed is necessary. It also sets
1835 * transmission limits appropriately if such adjustment is
1839 * Kernel thread context (may sleep).
1842 * Determined recovery action.
1844 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1845 unsigned int eflags, unsigned int err_mask)
1847 struct ata_link *link = ata_dev_phys_link(dev);
1849 unsigned int verdict;
1850 unsigned int action = 0;
1852 /* don't bother if Cat-0 error */
1853 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1856 /* record error and determine whether speed down is necessary */
1857 ata_ering_record(&dev->ering, eflags, err_mask);
1858 verdict = ata_eh_speed_down_verdict(dev);
1861 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1862 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1863 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1864 dev->flags |= ATA_DFLAG_NCQ_OFF;
1865 ata_dev_printk(dev, KERN_WARNING,
1866 "NCQ disabled due to excessive errors\n");
1871 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1872 /* speed down SATA link speed if possible */
1873 if (sata_down_spd_limit(link) == 0) {
1874 action |= ATA_EH_RESET;
1878 /* lower transfer mode */
1879 if (dev->spdn_cnt < 2) {
1880 static const int dma_dnxfer_sel[] =
1881 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1882 static const int pio_dnxfer_sel[] =
1883 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1886 if (dev->xfer_shift != ATA_SHIFT_PIO)
1887 sel = dma_dnxfer_sel[dev->spdn_cnt];
1889 sel = pio_dnxfer_sel[dev->spdn_cnt];
1893 if (ata_down_xfermask_limit(dev, sel) == 0) {
1894 action |= ATA_EH_RESET;
1900 /* Fall back to PIO? Slowing down to PIO is meaningless for
1901 * SATA ATA devices. Consider it only for PATA and SATAPI.
1903 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1904 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1905 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1906 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1908 action |= ATA_EH_RESET;
1915 /* device has been slowed down, blow error history */
1916 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1917 ata_ering_clear(&dev->ering);
1922 * ata_eh_link_autopsy - analyze error and determine recovery action
1923 * @link: host link to perform autopsy on
1925 * Analyze why @link failed and determine which recovery actions
1926 * are needed. This function also sets more detailed AC_ERR_*
1927 * values and fills sense data for ATAPI CHECK SENSE.
1930 * Kernel thread context (may sleep).
1932 static void ata_eh_link_autopsy(struct ata_link *link)
1934 struct ata_port *ap = link->ap;
1935 struct ata_eh_context *ehc = &link->eh_context;
1936 struct ata_device *dev;
1937 unsigned int all_err_mask = 0, eflags = 0;
1944 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1947 /* obtain and analyze SError */
1948 rc = sata_scr_read(link, SCR_ERROR, &serror);
1950 ehc->i.serror |= serror;
1951 ata_eh_analyze_serror(link);
1952 } else if (rc != -EOPNOTSUPP) {
1953 /* SError read failed, force reset and probing */
1954 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1955 ehc->i.action |= ATA_EH_RESET;
1956 ehc->i.err_mask |= AC_ERR_OTHER;
1959 /* analyze NCQ failure */
1960 ata_eh_analyze_ncq_error(link);
1962 /* any real error trumps AC_ERR_OTHER */
1963 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1964 ehc->i.err_mask &= ~AC_ERR_OTHER;
1966 all_err_mask |= ehc->i.err_mask;
1968 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1969 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1971 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1972 ata_dev_phys_link(qc->dev) != link)
1975 /* inherit upper level err_mask */
1976 qc->err_mask |= ehc->i.err_mask;
1979 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1981 /* DEV errors are probably spurious in case of ATA_BUS error */
1982 if (qc->err_mask & AC_ERR_ATA_BUS)
1983 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1986 /* any real error trumps unknown error */
1987 if (qc->err_mask & ~AC_ERR_OTHER)
1988 qc->err_mask &= ~AC_ERR_OTHER;
1990 /* SENSE_VALID trumps dev/unknown error and revalidation */
1991 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1992 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1994 /* determine whether the command is worth retrying */
1995 if (!(qc->err_mask & AC_ERR_INVALID) &&
1996 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1997 qc->flags |= ATA_QCFLAG_RETRY;
1999 /* accumulate error info */
2000 ehc->i.dev = qc->dev;
2001 all_err_mask |= qc->err_mask;
2002 if (qc->flags & ATA_QCFLAG_IO)
2003 eflags |= ATA_EFLAG_IS_IO;
2006 /* enforce default EH actions */
2007 if (ap->pflags & ATA_PFLAG_FROZEN ||
2008 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2009 ehc->i.action |= ATA_EH_RESET;
2010 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2011 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2012 ehc->i.action |= ATA_EH_REVALIDATE;
2014 /* If we have offending qcs and the associated failed device,
2015 * perform per-dev EH action only on the offending device.
2018 ehc->i.dev_action[ehc->i.dev->devno] |=
2019 ehc->i.action & ATA_EH_PERDEV_MASK;
2020 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2023 /* propagate timeout to host link */
2024 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2025 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2027 /* record error and consider speeding down */
2029 if (!dev && ((ata_link_max_devices(link) == 1 &&
2030 ata_dev_enabled(link->device))))
2034 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2035 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2036 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2043 * ata_eh_autopsy - analyze error and determine recovery action
2044 * @ap: host port to perform autopsy on
2046 * Analyze all links of @ap and determine why they failed and
2047 * which recovery actions are needed.
2050 * Kernel thread context (may sleep).
2052 void ata_eh_autopsy(struct ata_port *ap)
2054 struct ata_link *link;
2056 ata_for_each_link(link, ap, EDGE)
2057 ata_eh_link_autopsy(link);
2059 /* Handle the frigging slave link. Autopsy is done similarly
2060 * but actions and flags are transferred over to the master
2061 * link and handled from there.
2063 if (ap->slave_link) {
2064 struct ata_eh_context *mehc = &ap->link.eh_context;
2065 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2067 /* transfer control flags from master to slave */
2068 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2070 /* perform autopsy on the slave link */
2071 ata_eh_link_autopsy(ap->slave_link);
2073 /* transfer actions from slave to master and clear slave */
2074 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2075 mehc->i.action |= sehc->i.action;
2076 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2077 mehc->i.flags |= sehc->i.flags;
2078 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2081 /* Autopsy of fanout ports can affect host link autopsy.
2082 * Perform host link autopsy last.
2084 if (sata_pmp_attached(ap))
2085 ata_eh_link_autopsy(&ap->link);
2089 * ata_eh_link_report - report error handling to user
2090 * @link: ATA link EH is going on
2092 * Report EH to user.
2097 static void ata_eh_link_report(struct ata_link *link)
2099 struct ata_port *ap = link->ap;
2100 struct ata_eh_context *ehc = &link->eh_context;
2101 const char *frozen, *desc;
2103 int tag, nr_failed = 0;
2105 if (ehc->i.flags & ATA_EHI_QUIET)
2109 if (ehc->i.desc[0] != '\0')
2112 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2113 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2115 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2116 ata_dev_phys_link(qc->dev) != link ||
2117 ((qc->flags & ATA_QCFLAG_QUIET) &&
2118 qc->err_mask == AC_ERR_DEV))
2120 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2126 if (!nr_failed && !ehc->i.err_mask)
2130 if (ap->pflags & ATA_PFLAG_FROZEN)
2133 memset(tries_buf, 0, sizeof(tries_buf));
2134 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2135 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2139 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2140 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2141 ehc->i.err_mask, link->sactive, ehc->i.serror,
2142 ehc->i.action, frozen, tries_buf);
2144 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2146 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2147 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2148 ehc->i.err_mask, link->sactive, ehc->i.serror,
2149 ehc->i.action, frozen, tries_buf);
2151 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2155 ata_link_printk(link, KERN_ERR,
2156 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2157 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2158 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2159 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2160 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2161 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2162 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2163 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2164 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2165 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2166 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2167 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2168 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2169 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2170 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2171 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2172 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2173 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2175 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2176 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2177 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2178 const u8 *cdb = qc->cdb;
2179 char data_buf[20] = "";
2180 char cdb_buf[70] = "";
2182 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2183 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2186 if (qc->dma_dir != DMA_NONE) {
2187 static const char *dma_str[] = {
2188 [DMA_BIDIRECTIONAL] = "bidi",
2189 [DMA_TO_DEVICE] = "out",
2190 [DMA_FROM_DEVICE] = "in",
2192 static const char *prot_str[] = {
2193 [ATA_PROT_PIO] = "pio",
2194 [ATA_PROT_DMA] = "dma",
2195 [ATA_PROT_NCQ] = "ncq",
2196 [ATAPI_PROT_PIO] = "pio",
2197 [ATAPI_PROT_DMA] = "dma",
2200 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2201 prot_str[qc->tf.protocol], qc->nbytes,
2202 dma_str[qc->dma_dir]);
2205 if (ata_is_atapi(qc->tf.protocol))
2206 snprintf(cdb_buf, sizeof(cdb_buf),
2207 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2208 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2209 cdb[0], cdb[1], cdb[2], cdb[3],
2210 cdb[4], cdb[5], cdb[6], cdb[7],
2211 cdb[8], cdb[9], cdb[10], cdb[11],
2212 cdb[12], cdb[13], cdb[14], cdb[15]);
2214 ata_dev_printk(qc->dev, KERN_ERR,
2215 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2217 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2218 "Emask 0x%x (%s)%s\n",
2219 cmd->command, cmd->feature, cmd->nsect,
2220 cmd->lbal, cmd->lbam, cmd->lbah,
2221 cmd->hob_feature, cmd->hob_nsect,
2222 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2223 cmd->device, qc->tag, data_buf, cdb_buf,
2224 res->command, res->feature, res->nsect,
2225 res->lbal, res->lbam, res->lbah,
2226 res->hob_feature, res->hob_nsect,
2227 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2228 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2229 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2231 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2233 if (res->command & ATA_BUSY)
2234 ata_dev_printk(qc->dev, KERN_ERR,
2235 "status: { Busy }\n");
2237 ata_dev_printk(qc->dev, KERN_ERR,
2238 "status: { %s%s%s%s}\n",
2239 res->command & ATA_DRDY ? "DRDY " : "",
2240 res->command & ATA_DF ? "DF " : "",
2241 res->command & ATA_DRQ ? "DRQ " : "",
2242 res->command & ATA_ERR ? "ERR " : "");
2245 if (cmd->command != ATA_CMD_PACKET &&
2246 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2248 ata_dev_printk(qc->dev, KERN_ERR,
2249 "error: { %s%s%s%s}\n",
2250 res->feature & ATA_ICRC ? "ICRC " : "",
2251 res->feature & ATA_UNC ? "UNC " : "",
2252 res->feature & ATA_IDNF ? "IDNF " : "",
2253 res->feature & ATA_ABORTED ? "ABRT " : "");
2258 * ata_eh_report - report error handling to user
2259 * @ap: ATA port to report EH about
2261 * Report EH to user.
2266 void ata_eh_report(struct ata_port *ap)
2268 struct ata_link *link;
2270 ata_for_each_link(link, ap, HOST_FIRST)
2271 ata_eh_link_report(link);
2274 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2275 unsigned int *classes, unsigned long deadline,
2278 struct ata_device *dev;
2281 ata_for_each_dev(dev, link, ALL)
2282 classes[dev->devno] = ATA_DEV_UNKNOWN;
2284 return reset(link, classes, deadline);
2287 static int ata_eh_followup_srst_needed(struct ata_link *link,
2288 int rc, const unsigned int *classes)
2290 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2294 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2299 int ata_eh_reset(struct ata_link *link, int classify,
2300 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2301 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2303 struct ata_port *ap = link->ap;
2304 struct ata_link *slave = ap->slave_link;
2305 struct ata_eh_context *ehc = &link->eh_context;
2306 struct ata_eh_context *sehc = &slave->eh_context;
2307 unsigned int *classes = ehc->classes;
2308 unsigned int lflags = link->flags;
2309 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2310 int max_tries = 0, try = 0;
2311 struct ata_link *failed_link;
2312 struct ata_device *dev;
2313 unsigned long deadline, now;
2314 ata_reset_fn_t reset;
2315 unsigned long flags;
2322 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2324 if (link->flags & ATA_LFLAG_NO_HRST)
2326 if (link->flags & ATA_LFLAG_NO_SRST)
2329 /* make sure each reset attemp is at least COOL_DOWN apart */
2330 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2332 WARN_ON(time_after(ehc->last_reset, now));
2333 deadline = ata_deadline(ehc->last_reset,
2334 ATA_EH_RESET_COOL_DOWN);
2335 if (time_before(now, deadline))
2336 schedule_timeout_uninterruptible(deadline - now);
2339 spin_lock_irqsave(ap->lock, flags);
2340 ap->pflags |= ATA_PFLAG_RESETTING;
2341 spin_unlock_irqrestore(ap->lock, flags);
2343 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2345 ata_for_each_dev(dev, link, ALL) {
2346 /* If we issue an SRST then an ATA drive (not ATAPI)
2347 * may change configuration and be in PIO0 timing. If
2348 * we do a hard reset (or are coming from power on)
2349 * this is true for ATA or ATAPI. Until we've set a
2350 * suitable controller mode we should not touch the
2351 * bus as we may be talking too fast.
2353 dev->pio_mode = XFER_PIO_0;
2355 /* If the controller has a pio mode setup function
2356 * then use it to set the chipset to rights. Don't
2357 * touch the DMA setup as that will be dealt with when
2358 * configuring devices.
2360 if (ap->ops->set_piomode)
2361 ap->ops->set_piomode(ap, dev);
2364 /* prefer hardreset */
2366 ehc->i.action &= ~ATA_EH_RESET;
2369 ehc->i.action |= ATA_EH_HARDRESET;
2370 } else if (softreset) {
2372 ehc->i.action |= ATA_EH_SOFTRESET;
2376 unsigned long deadline = ata_deadline(jiffies,
2377 ATA_EH_PRERESET_TIMEOUT);
2380 sehc->i.action &= ~ATA_EH_RESET;
2381 sehc->i.action |= ehc->i.action;
2384 rc = prereset(link, deadline);
2386 /* If present, do prereset on slave link too. Reset
2387 * is skipped iff both master and slave links report
2388 * -ENOENT or clear ATA_EH_RESET.
2390 if (slave && (rc == 0 || rc == -ENOENT)) {
2393 tmp = prereset(slave, deadline);
2397 ehc->i.action |= sehc->i.action;
2401 if (rc == -ENOENT) {
2402 ata_link_printk(link, KERN_DEBUG,
2403 "port disabled. ignoring.\n");
2404 ehc->i.action &= ~ATA_EH_RESET;
2406 ata_for_each_dev(dev, link, ALL)
2407 classes[dev->devno] = ATA_DEV_NONE;
2411 ata_link_printk(link, KERN_ERR,
2412 "prereset failed (errno=%d)\n", rc);
2416 /* prereset() might have cleared ATA_EH_RESET. If so,
2417 * bang classes and return.
2419 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2420 ata_for_each_dev(dev, link, ALL)
2421 classes[dev->devno] = ATA_DEV_NONE;
2431 if (ata_is_host_link(link))
2432 ata_eh_freeze_port(ap);
2434 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2438 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2439 reset == softreset ? "soft" : "hard");
2441 /* mark that this EH session started with reset */
2442 ehc->last_reset = jiffies;
2443 if (reset == hardreset)
2444 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2446 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2448 rc = ata_do_reset(link, reset, classes, deadline, true);
2449 if (rc && rc != -EAGAIN) {
2454 /* hardreset slave link if existent */
2455 if (slave && reset == hardreset) {
2459 ata_link_printk(slave, KERN_INFO,
2460 "hard resetting link\n");
2462 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2463 tmp = ata_do_reset(slave, reset, classes, deadline,
2471 failed_link = slave;
2477 /* perform follow-up SRST if necessary */
2478 if (reset == hardreset &&
2479 ata_eh_followup_srst_needed(link, rc, classes)) {
2483 ata_link_printk(link, KERN_ERR,
2484 "follow-up softreset required "
2485 "but no softreset avaliable\n");
2491 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2492 rc = ata_do_reset(link, reset, classes, deadline, true);
2496 ata_link_printk(link, KERN_INFO, "no reset method "
2497 "available, skipping reset\n");
2498 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2499 lflags |= ATA_LFLAG_ASSUME_ATA;
2503 * Post-reset processing
2505 ata_for_each_dev(dev, link, ALL) {
2506 /* After the reset, the device state is PIO 0 and the
2507 * controller state is undefined. Reset also wakes up
2508 * drives from sleeping mode.
2510 dev->pio_mode = XFER_PIO_0;
2511 dev->flags &= ~ATA_DFLAG_SLEEPING;
2513 if (!ata_phys_link_offline(ata_dev_phys_link(dev))) {
2514 /* apply class override */
2515 if (lflags & ATA_LFLAG_ASSUME_ATA)
2516 classes[dev->devno] = ATA_DEV_ATA;
2517 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2518 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2520 classes[dev->devno] = ATA_DEV_NONE;
2523 /* record current link speed */
2524 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2525 link->sata_spd = (sstatus >> 4) & 0xf;
2526 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2527 slave->sata_spd = (sstatus >> 4) & 0xf;
2530 if (ata_is_host_link(link))
2531 ata_eh_thaw_port(ap);
2533 /* postreset() should clear hardware SError. Although SError
2534 * is cleared during link resume, clearing SError here is
2535 * necessary as some PHYs raise hotplug events after SRST.
2536 * This introduces race condition where hotplug occurs between
2537 * reset and here. This race is mediated by cross checking
2538 * link onlineness and classification result later.
2541 postreset(link, classes);
2543 postreset(slave, classes);
2546 /* clear cached SError */
2547 spin_lock_irqsave(link->ap->lock, flags);
2548 link->eh_info.serror = 0;
2550 slave->eh_info.serror = 0;
2551 spin_unlock_irqrestore(link->ap->lock, flags);
2553 /* Make sure onlineness and classification result correspond.
2554 * Hotplug could have happened during reset and some
2555 * controllers fail to wait while a drive is spinning up after
2556 * being hotplugged causing misdetection. By cross checking
2557 * link onlineness and classification result, those conditions
2558 * can be reliably detected and retried.
2561 ata_for_each_dev(dev, link, ALL) {
2562 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2563 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2564 classes[dev->devno] = ATA_DEV_NONE;
2565 if (ata_phys_link_online(ata_dev_phys_link(dev)))
2570 if (classify && nr_unknown) {
2571 if (try < max_tries) {
2572 ata_link_printk(link, KERN_WARNING, "link online but "
2573 "device misclassified, retrying\n");
2578 ata_link_printk(link, KERN_WARNING,
2579 "link online but device misclassified, "
2580 "device detection might fail\n");
2583 /* reset successful, schedule revalidation */
2584 ata_eh_done(link, NULL, ATA_EH_RESET);
2586 ata_eh_done(slave, NULL, ATA_EH_RESET);
2587 ehc->last_reset = jiffies; /* update to completion time */
2588 ehc->i.action |= ATA_EH_REVALIDATE;
2592 /* clear hotplug flag */
2593 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2595 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2597 spin_lock_irqsave(ap->lock, flags);
2598 ap->pflags &= ~ATA_PFLAG_RESETTING;
2599 spin_unlock_irqrestore(ap->lock, flags);
2604 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2605 if (!ata_is_host_link(link) &&
2606 sata_scr_read(link, SCR_STATUS, &sstatus))
2609 if (rc == -ERESTART || try >= max_tries)
2613 if (time_before(now, deadline)) {
2614 unsigned long delta = deadline - now;
2616 ata_link_printk(failed_link, KERN_WARNING,
2617 "reset failed (errno=%d), retrying in %u secs\n",
2618 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2621 delta = schedule_timeout_uninterruptible(delta);
2624 if (try == max_tries - 1) {
2625 sata_down_spd_limit(link);
2627 sata_down_spd_limit(slave);
2628 } else if (rc == -EPIPE)
2629 sata_down_spd_limit(failed_link);
2636 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2638 struct ata_link *link;
2639 struct ata_device *dev;
2640 unsigned long flags;
2643 * This function can be thought of as an extended version of
2644 * ata_eh_about_to_do() specially crafted to accommodate the
2645 * requirements of ATA_EH_PARK handling. Since the EH thread
2646 * does not leave the do {} while () loop in ata_eh_recover as
2647 * long as the timeout for a park request to *one* device on
2648 * the port has not expired, and since we still want to pick
2649 * up park requests to other devices on the same port or
2650 * timeout updates for the same device, we have to pull
2651 * ATA_EH_PARK actions from eh_info into eh_context.i
2652 * ourselves at the beginning of each pass over the loop.
2654 * Additionally, all write accesses to &ap->park_req_pending
2655 * through INIT_COMPLETION() (see below) or complete_all()
2656 * (see ata_scsi_park_store()) are protected by the host lock.
2657 * As a result we have that park_req_pending.done is zero on
2658 * exit from this function, i.e. when ATA_EH_PARK actions for
2659 * *all* devices on port ap have been pulled into the
2660 * respective eh_context structs. If, and only if,
2661 * park_req_pending.done is non-zero by the time we reach
2662 * wait_for_completion_timeout(), another ATA_EH_PARK action
2663 * has been scheduled for at least one of the devices on port
2664 * ap and we have to cycle over the do {} while () loop in
2665 * ata_eh_recover() again.
2668 spin_lock_irqsave(ap->lock, flags);
2669 INIT_COMPLETION(ap->park_req_pending);
2670 ata_for_each_link(link, ap, EDGE) {
2671 ata_for_each_dev(dev, link, ALL) {
2672 struct ata_eh_info *ehi = &link->eh_info;
2674 link->eh_context.i.dev_action[dev->devno] |=
2675 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2676 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2679 spin_unlock_irqrestore(ap->lock, flags);
2682 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2684 struct ata_eh_context *ehc = &dev->link->eh_context;
2685 struct ata_taskfile tf;
2686 unsigned int err_mask;
2688 ata_tf_init(dev, &tf);
2690 ehc->unloaded_mask |= 1 << dev->devno;
2691 tf.command = ATA_CMD_IDLEIMMEDIATE;
2697 ehc->unloaded_mask &= ~(1 << dev->devno);
2698 tf.command = ATA_CMD_CHK_POWER;
2701 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2702 tf.protocol |= ATA_PROT_NODATA;
2703 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2704 if (park && (err_mask || tf.lbal != 0xc4)) {
2705 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2706 ehc->unloaded_mask &= ~(1 << dev->devno);
2710 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2711 struct ata_device **r_failed_dev)
2713 struct ata_port *ap = link->ap;
2714 struct ata_eh_context *ehc = &link->eh_context;
2715 struct ata_device *dev;
2716 unsigned int new_mask = 0;
2717 unsigned long flags;
2722 /* For PATA drive side cable detection to work, IDENTIFY must
2723 * be done backwards such that PDIAG- is released by the slave
2724 * device before the master device is identified.
2726 ata_for_each_dev(dev, link, ALL_REVERSE) {
2727 unsigned int action = ata_eh_dev_action(dev);
2728 unsigned int readid_flags = 0;
2730 if (ehc->i.flags & ATA_EHI_DID_RESET)
2731 readid_flags |= ATA_READID_POSTRESET;
2733 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2734 WARN_ON(dev->class == ATA_DEV_PMP);
2736 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2741 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2742 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2747 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2749 /* Configuration may have changed, reconfigure
2752 ehc->i.flags |= ATA_EHI_SETMODE;
2754 /* schedule the scsi_rescan_device() here */
2755 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2756 } else if (dev->class == ATA_DEV_UNKNOWN &&
2757 ehc->tries[dev->devno] &&
2758 ata_class_enabled(ehc->classes[dev->devno])) {
2759 dev->class = ehc->classes[dev->devno];
2761 if (dev->class == ATA_DEV_PMP)
2762 rc = sata_pmp_attach(dev);
2764 rc = ata_dev_read_id(dev, &dev->class,
2765 readid_flags, dev->id);
2768 new_mask |= 1 << dev->devno;
2771 /* IDENTIFY was issued to non-existent
2772 * device. No need to reset. Just
2773 * thaw and kill the device.
2775 ata_eh_thaw_port(ap);
2776 dev->class = ATA_DEV_UNKNOWN;
2779 dev->class = ATA_DEV_UNKNOWN;
2785 /* PDIAG- should have been released, ask cable type if post-reset */
2786 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2787 if (ap->ops->cable_detect)
2788 ap->cbl = ap->ops->cable_detect(ap);
2792 /* Configure new devices forward such that user doesn't see
2793 * device detection messages backwards.
2795 ata_for_each_dev(dev, link, ALL) {
2796 if (!(new_mask & (1 << dev->devno)) ||
2797 dev->class == ATA_DEV_PMP)
2800 ehc->i.flags |= ATA_EHI_PRINTINFO;
2801 rc = ata_dev_configure(dev);
2802 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2806 spin_lock_irqsave(ap->lock, flags);
2807 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2808 spin_unlock_irqrestore(ap->lock, flags);
2810 /* new device discovered, configure xfermode */
2811 ehc->i.flags |= ATA_EHI_SETMODE;
2817 *r_failed_dev = dev;
2818 DPRINTK("EXIT rc=%d\n", rc);
2823 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2824 * @link: link on which timings will be programmed
2825 * @r_failed_dev: out paramter for failed device
2827 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2828 * ata_set_mode() fails, pointer to the failing device is
2829 * returned in @r_failed_dev.
2832 * PCI/etc. bus probe sem.
2835 * 0 on success, negative errno otherwise
2837 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2839 struct ata_port *ap = link->ap;
2840 struct ata_device *dev;
2843 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2844 ata_for_each_dev(dev, link, ENABLED) {
2845 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2846 struct ata_ering_entry *ent;
2848 ent = ata_ering_top(&dev->ering);
2850 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2854 /* has private set_mode? */
2855 if (ap->ops->set_mode)
2856 rc = ap->ops->set_mode(link, r_failed_dev);
2858 rc = ata_do_set_mode(link, r_failed_dev);
2860 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2861 ata_for_each_dev(dev, link, ENABLED) {
2862 struct ata_eh_context *ehc = &link->eh_context;
2863 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2864 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2866 if (dev->xfer_mode != saved_xfer_mode ||
2867 ata_ncq_enabled(dev) != saved_ncq)
2868 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2875 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2876 * @dev: ATAPI device to clear UA for
2878 * Resets and other operations can make an ATAPI device raise
2879 * UNIT ATTENTION which causes the next operation to fail. This
2880 * function clears UA.
2883 * EH context (may sleep).
2886 * 0 on success, -errno on failure.
2888 static int atapi_eh_clear_ua(struct ata_device *dev)
2892 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
2893 u8 sense_buffer[SCSI_SENSE_BUFFERSIZE];
2895 unsigned int err_mask;
2897 err_mask = atapi_eh_tur(dev, &sense_key);
2898 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
2899 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
2900 "failed (err_mask=0x%x)\n", err_mask);
2904 if (!err_mask || sense_key != UNIT_ATTENTION)
2907 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
2909 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
2910 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
2915 ata_dev_printk(dev, KERN_WARNING,
2916 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
2921 static int ata_link_nr_enabled(struct ata_link *link)
2923 struct ata_device *dev;
2926 ata_for_each_dev(dev, link, ENABLED)
2931 static int ata_link_nr_vacant(struct ata_link *link)
2933 struct ata_device *dev;
2936 ata_for_each_dev(dev, link, ALL)
2937 if (dev->class == ATA_DEV_UNKNOWN)
2942 static int ata_eh_skip_recovery(struct ata_link *link)
2944 struct ata_port *ap = link->ap;
2945 struct ata_eh_context *ehc = &link->eh_context;
2946 struct ata_device *dev;
2948 /* skip disabled links */
2949 if (link->flags & ATA_LFLAG_DISABLED)
2952 /* thaw frozen port and recover failed devices */
2953 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2956 /* reset at least once if reset is requested */
2957 if ((ehc->i.action & ATA_EH_RESET) &&
2958 !(ehc->i.flags & ATA_EHI_DID_RESET))
2961 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2962 ata_for_each_dev(dev, link, ALL) {
2963 if (dev->class == ATA_DEV_UNKNOWN &&
2964 ehc->classes[dev->devno] != ATA_DEV_NONE)
2971 static int ata_eh_schedule_probe(struct ata_device *dev)
2973 struct ata_eh_context *ehc = &dev->link->eh_context;
2975 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2976 (ehc->did_probe_mask & (1 << dev->devno)))
2979 ata_eh_detach_dev(dev);
2981 ehc->did_probe_mask |= (1 << dev->devno);
2982 ehc->i.action |= ATA_EH_RESET;
2983 ehc->saved_xfer_mode[dev->devno] = 0;
2984 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
2989 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2991 struct ata_eh_context *ehc = &dev->link->eh_context;
2993 ehc->tries[dev->devno]--;
2997 /* device missing or wrong IDENTIFY data, schedule probing */
2998 ehc->i.probe_mask |= (1 << dev->devno);
3000 /* give it just one more chance */
3001 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3003 if (ehc->tries[dev->devno] == 1) {
3004 /* This is the last chance, better to slow
3005 * down than lose it.
3007 sata_down_spd_limit(ata_dev_phys_link(dev));
3008 if (dev->pio_mode > XFER_PIO_0)
3009 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3013 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3014 /* disable device if it has used up all its chances */
3015 ata_dev_disable(dev);
3017 /* detach if offline */
3018 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3019 ata_eh_detach_dev(dev);
3021 /* schedule probe if necessary */
3022 if (ata_eh_schedule_probe(dev)) {
3023 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3024 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3025 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3030 ehc->i.action |= ATA_EH_RESET;
3036 * ata_eh_recover - recover host port after error
3037 * @ap: host port to recover
3038 * @prereset: prereset method (can be NULL)
3039 * @softreset: softreset method (can be NULL)
3040 * @hardreset: hardreset method (can be NULL)
3041 * @postreset: postreset method (can be NULL)
3042 * @r_failed_link: out parameter for failed link
3044 * This is the alpha and omega, eum and yang, heart and soul of
3045 * libata exception handling. On entry, actions required to
3046 * recover each link and hotplug requests are recorded in the
3047 * link's eh_context. This function executes all the operations
3048 * with appropriate retrials and fallbacks to resurrect failed
3049 * devices, detach goners and greet newcomers.
3052 * Kernel thread context (may sleep).
3055 * 0 on success, -errno on failure.
3057 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3058 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3059 ata_postreset_fn_t postreset,
3060 struct ata_link **r_failed_link)
3062 struct ata_link *link;
3063 struct ata_device *dev;
3066 unsigned long flags, deadline;
3070 /* prep for recovery */
3071 ata_for_each_link(link, ap, EDGE) {
3072 struct ata_eh_context *ehc = &link->eh_context;
3074 /* re-enable link? */
3075 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3076 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3077 spin_lock_irqsave(ap->lock, flags);
3078 link->flags &= ~ATA_LFLAG_DISABLED;
3079 spin_unlock_irqrestore(ap->lock, flags);
3080 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3083 ata_for_each_dev(dev, link, ALL) {
3084 if (link->flags & ATA_LFLAG_NO_RETRY)
3085 ehc->tries[dev->devno] = 1;
3087 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3089 /* collect port action mask recorded in dev actions */
3090 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3091 ~ATA_EH_PERDEV_MASK;
3092 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3094 /* process hotplug request */
3095 if (dev->flags & ATA_DFLAG_DETACH)
3096 ata_eh_detach_dev(dev);
3098 /* schedule probe if necessary */
3099 if (!ata_dev_enabled(dev))
3100 ata_eh_schedule_probe(dev);
3108 /* if UNLOADING, finish immediately */
3109 if (ap->pflags & ATA_PFLAG_UNLOADING)
3113 ata_for_each_link(link, ap, EDGE) {
3114 struct ata_eh_context *ehc = &link->eh_context;
3116 /* skip EH if possible. */
3117 if (ata_eh_skip_recovery(link))
3120 ata_for_each_dev(dev, link, ALL)
3121 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3125 ata_for_each_link(link, ap, EDGE) {
3126 struct ata_eh_context *ehc = &link->eh_context;
3128 if (!(ehc->i.action & ATA_EH_RESET))
3131 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3132 prereset, softreset, hardreset, postreset);
3134 ata_link_printk(link, KERN_ERR,
3135 "reset failed, giving up\n");
3144 * clears ATA_EH_PARK in eh_info and resets
3145 * ap->park_req_pending
3147 ata_eh_pull_park_action(ap);
3150 ata_for_each_link(link, ap, EDGE) {
3151 ata_for_each_dev(dev, link, ALL) {
3152 struct ata_eh_context *ehc = &link->eh_context;
3155 if (dev->class != ATA_DEV_ATA)
3157 if (!(ehc->i.dev_action[dev->devno] &
3160 tmp = dev->unpark_deadline;
3161 if (time_before(deadline, tmp))
3163 else if (time_before_eq(tmp, jiffies))
3165 if (ehc->unloaded_mask & (1 << dev->devno))
3168 ata_eh_park_issue_cmd(dev, 1);
3173 if (time_before_eq(deadline, now))
3176 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3179 ata_for_each_link(link, ap, EDGE) {
3180 ata_for_each_dev(dev, link, ALL) {
3181 if (!(link->eh_context.unloaded_mask &
3185 ata_eh_park_issue_cmd(dev, 0);
3186 ata_eh_done(link, dev, ATA_EH_PARK);
3191 ata_for_each_link(link, ap, EDGE) {
3192 struct ata_eh_context *ehc = &link->eh_context;
3194 /* revalidate existing devices and attach new ones */
3195 rc = ata_eh_revalidate_and_attach(link, &dev);
3199 /* if PMP got attached, return, pmp EH will take care of it */
3200 if (link->device->class == ATA_DEV_PMP) {
3205 /* configure transfer mode if necessary */
3206 if (ehc->i.flags & ATA_EHI_SETMODE) {
3207 rc = ata_set_mode(link, &dev);
3210 ehc->i.flags &= ~ATA_EHI_SETMODE;
3213 /* If reset has been issued, clear UA to avoid
3214 * disrupting the current users of the device.
3216 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3217 ata_for_each_dev(dev, link, ALL) {
3218 if (dev->class != ATA_DEV_ATAPI)
3220 rc = atapi_eh_clear_ua(dev);
3226 /* configure link power saving */
3227 if (ehc->i.action & ATA_EH_LPM)
3228 ata_for_each_dev(dev, link, ALL)
3229 ata_dev_enable_pm(dev, ap->pm_policy);
3231 /* this link is okay now */
3237 ata_eh_handle_dev_fail(dev, rc);
3239 if (ap->pflags & ATA_PFLAG_FROZEN) {
3240 /* PMP reset requires working host port.
3241 * Can't retry if it's frozen.
3243 if (sata_pmp_attached(ap))
3253 if (rc && r_failed_link)
3254 *r_failed_link = link;
3256 DPRINTK("EXIT, rc=%d\n", rc);
3261 * ata_eh_finish - finish up EH
3262 * @ap: host port to finish EH for
3264 * Recovery is complete. Clean up EH states and retry or finish
3270 void ata_eh_finish(struct ata_port *ap)
3274 /* retry or finish qcs */
3275 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3276 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3278 if (!(qc->flags & ATA_QCFLAG_FAILED))
3282 /* FIXME: Once EH migration is complete,
3283 * generate sense data in this function,
3284 * considering both err_mask and tf.
3286 if (qc->flags & ATA_QCFLAG_RETRY)
3287 ata_eh_qc_retry(qc);
3289 ata_eh_qc_complete(qc);
3291 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3292 ata_eh_qc_complete(qc);
3294 /* feed zero TF to sense generation */
3295 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3296 ata_eh_qc_retry(qc);
3301 /* make sure nr_active_links is zero after EH */
3302 WARN_ON(ap->nr_active_links);
3303 ap->nr_active_links = 0;
3307 * ata_do_eh - do standard error handling
3308 * @ap: host port to handle error for
3310 * @prereset: prereset method (can be NULL)
3311 * @softreset: softreset method (can be NULL)
3312 * @hardreset: hardreset method (can be NULL)
3313 * @postreset: postreset method (can be NULL)
3315 * Perform standard error handling sequence.
3318 * Kernel thread context (may sleep).
3320 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3321 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3322 ata_postreset_fn_t postreset)
3324 struct ata_device *dev;
3330 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3333 ata_for_each_dev(dev, &ap->link, ALL)
3334 ata_dev_disable(dev);
3341 * ata_std_error_handler - standard error handler
3342 * @ap: host port to handle error for
3344 * Standard error handler
3347 * Kernel thread context (may sleep).
3349 void ata_std_error_handler(struct ata_port *ap)
3351 struct ata_port_operations *ops = ap->ops;
3352 ata_reset_fn_t hardreset = ops->hardreset;
3354 /* ignore built-in hardreset if SCR access is not available */
3355 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3358 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3363 * ata_eh_handle_port_suspend - perform port suspend operation
3364 * @ap: port to suspend
3369 * Kernel thread context (may sleep).
3371 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3373 unsigned long flags;
3376 /* are we suspending? */
3377 spin_lock_irqsave(ap->lock, flags);
3378 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3379 ap->pm_mesg.event == PM_EVENT_ON) {
3380 spin_unlock_irqrestore(ap->lock, flags);
3383 spin_unlock_irqrestore(ap->lock, flags);
3385 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3387 /* tell ACPI we're suspending */
3388 rc = ata_acpi_on_suspend(ap);
3393 ata_eh_freeze_port(ap);
3395 if (ap->ops->port_suspend)
3396 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3398 ata_acpi_set_state(ap, PMSG_SUSPEND);
3401 spin_lock_irqsave(ap->lock, flags);
3403 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3405 ap->pflags |= ATA_PFLAG_SUSPENDED;
3406 else if (ap->pflags & ATA_PFLAG_FROZEN)
3407 ata_port_schedule_eh(ap);
3409 if (ap->pm_result) {
3410 *ap->pm_result = rc;
3411 ap->pm_result = NULL;
3414 spin_unlock_irqrestore(ap->lock, flags);
3420 * ata_eh_handle_port_resume - perform port resume operation
3421 * @ap: port to resume
3426 * Kernel thread context (may sleep).
3428 static void ata_eh_handle_port_resume(struct ata_port *ap)
3430 unsigned long flags;
3433 /* are we resuming? */
3434 spin_lock_irqsave(ap->lock, flags);
3435 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3436 ap->pm_mesg.event != PM_EVENT_ON) {
3437 spin_unlock_irqrestore(ap->lock, flags);
3440 spin_unlock_irqrestore(ap->lock, flags);
3442 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3444 ata_acpi_set_state(ap, PMSG_ON);
3446 if (ap->ops->port_resume)
3447 rc = ap->ops->port_resume(ap);
3449 /* tell ACPI that we're resuming */
3450 ata_acpi_on_resume(ap);
3453 spin_lock_irqsave(ap->lock, flags);
3454 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3455 if (ap->pm_result) {
3456 *ap->pm_result = rc;
3457 ap->pm_result = NULL;
3459 spin_unlock_irqrestore(ap->lock, flags);
3461 #endif /* CONFIG_PM */