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/pci.h>
37 #include <scsi/scsi.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_eh.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_cmnd.h>
42 #include "../scsi/scsi_transport_api.h"
44 #include <linux/libata.h>
49 /* speed down verdicts */
50 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
51 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
52 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
53 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
56 ATA_EFLAG_IS_IO
= (1 << 0),
57 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
59 /* error categories */
62 ATA_ECAT_TOUT_HSM
= 2,
64 ATA_ECAT_DUBIOUS_NONE
= 4,
65 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
66 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
67 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
70 /* always put at least this amount of time between resets */
71 ATA_EH_RESET_COOL_DOWN
= 5000,
73 /* Waiting in ->prereset can never be reliable. It's
74 * sometimes nice to wait there but it can't be depended upon;
75 * otherwise, we wouldn't be resetting. Just give it enough
76 * time for most drives to spin up.
78 ATA_EH_PRERESET_TIMEOUT
= 10000,
79 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
82 /* The following table determines how we sequence resets. Each entry
83 * represents timeout for that try. The first try can be soft or
84 * hardreset. All others are hardreset if available. In most cases
85 * the first reset w/ 10sec timeout should succeed. Following entries
86 * are mostly for error handling, hotplug and retarded devices.
88 static const unsigned long ata_eh_reset_timeouts
[] = {
89 10000, /* most drives spin up by 10sec */
90 10000, /* > 99% working drives spin up before 20sec */
91 35000, /* give > 30 secs of idleness for retarded devices */
92 5000, /* and sweet one last chance */
93 ULONG_MAX
, /* > 1 min has elapsed, give up */
96 static void __ata_port_freeze(struct ata_port
*ap
);
98 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
99 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
100 #else /* CONFIG_PM */
101 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
104 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
106 #endif /* CONFIG_PM */
108 static void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
, const char *fmt
,
111 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
112 ATA_EH_DESC_LEN
- ehi
->desc_len
,
117 * __ata_ehi_push_desc - push error description without adding separator
119 * @fmt: printf format string
121 * Format string according to @fmt and append it to @ehi->desc.
124 * spin_lock_irqsave(host lock)
126 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
131 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
136 * ata_ehi_push_desc - push error description with separator
138 * @fmt: printf format string
140 * Format string according to @fmt and append it to @ehi->desc.
141 * If @ehi->desc is not empty, ", " is added in-between.
144 * spin_lock_irqsave(host lock)
146 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
151 __ata_ehi_push_desc(ehi
, ", ");
154 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
159 * ata_ehi_clear_desc - clean error description
165 * spin_lock_irqsave(host lock)
167 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
174 * ata_port_desc - append port description
175 * @ap: target ATA port
176 * @fmt: printf format string
178 * Format string according to @fmt and append it to port
179 * description. If port description is not empty, " " is added
180 * in-between. This function is to be used while initializing
181 * ata_host. The description is printed on host registration.
186 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
190 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
192 if (ap
->link
.eh_info
.desc_len
)
193 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
196 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
203 * ata_port_pbar_desc - append PCI BAR description
204 * @ap: target ATA port
205 * @bar: target PCI BAR
206 * @offset: offset into PCI BAR
207 * @name: name of the area
209 * If @offset is negative, this function formats a string which
210 * contains the name, address, size and type of the BAR and
211 * appends it to the port description. If @offset is zero or
212 * positive, only name and offsetted address is appended.
217 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
220 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
222 unsigned long long start
, len
;
224 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
226 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
229 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
230 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
233 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
235 ata_port_desc(ap
, "%s 0x%llx", name
,
236 start
+ (unsigned long long)offset
);
239 #endif /* CONFIG_PCI */
241 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
242 unsigned int err_mask
)
244 struct ata_ering_entry
*ent
;
249 ering
->cursor
%= ATA_ERING_SIZE
;
251 ent
= &ering
->ring
[ering
->cursor
];
252 ent
->eflags
= eflags
;
253 ent
->err_mask
= err_mask
;
254 ent
->timestamp
= get_jiffies_64();
257 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
259 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
266 static void ata_ering_clear(struct ata_ering
*ering
)
268 memset(ering
, 0, sizeof(*ering
));
271 static int ata_ering_map(struct ata_ering
*ering
,
272 int (*map_fn
)(struct ata_ering_entry
*, void *),
276 struct ata_ering_entry
*ent
;
280 ent
= &ering
->ring
[idx
];
283 rc
= map_fn(ent
, arg
);
286 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
287 } while (idx
!= ering
->cursor
);
292 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
294 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
296 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
299 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
300 struct ata_eh_info
*ehi
, unsigned int action
)
302 struct ata_device
*tdev
;
305 ehi
->action
&= ~action
;
306 ata_link_for_each_dev(tdev
, link
)
307 ehi
->dev_action
[tdev
->devno
] &= ~action
;
309 /* doesn't make sense for port-wide EH actions */
310 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
312 /* break ehi->action into ehi->dev_action */
313 if (ehi
->action
& action
) {
314 ata_link_for_each_dev(tdev
, link
)
315 ehi
->dev_action
[tdev
->devno
] |=
316 ehi
->action
& action
;
317 ehi
->action
&= ~action
;
320 /* turn off the specified per-dev action */
321 ehi
->dev_action
[dev
->devno
] &= ~action
;
326 * ata_scsi_timed_out - SCSI layer time out callback
327 * @cmd: timed out SCSI command
329 * Handles SCSI layer timeout. We race with normal completion of
330 * the qc for @cmd. If the qc is already gone, we lose and let
331 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
332 * timed out and EH should be invoked. Prevent ata_qc_complete()
333 * from finishing it by setting EH_SCHEDULED and return
336 * TODO: kill this function once old EH is gone.
339 * Called from timer context
342 * EH_HANDLED or EH_NOT_HANDLED
344 enum scsi_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
346 struct Scsi_Host
*host
= cmd
->device
->host
;
347 struct ata_port
*ap
= ata_shost_to_port(host
);
349 struct ata_queued_cmd
*qc
;
350 enum scsi_eh_timer_return ret
;
354 if (ap
->ops
->error_handler
) {
355 ret
= EH_NOT_HANDLED
;
360 spin_lock_irqsave(ap
->lock
, flags
);
361 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
363 WARN_ON(qc
->scsicmd
!= cmd
);
364 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
365 qc
->err_mask
|= AC_ERR_TIMEOUT
;
366 ret
= EH_NOT_HANDLED
;
368 spin_unlock_irqrestore(ap
->lock
, flags
);
371 DPRINTK("EXIT, ret=%d\n", ret
);
376 * ata_scsi_error - SCSI layer error handler callback
377 * @host: SCSI host on which error occurred
379 * Handles SCSI-layer-thrown error events.
382 * Inherited from SCSI layer (none, can sleep)
387 void ata_scsi_error(struct Scsi_Host
*host
)
389 struct ata_port
*ap
= ata_shost_to_port(host
);
395 /* synchronize with port task */
396 ata_port_flush_task(ap
);
398 /* synchronize with host lock and sort out timeouts */
400 /* For new EH, all qcs are finished in one of three ways -
401 * normal completion, error completion, and SCSI timeout.
402 * Both cmpletions can race against SCSI timeout. When normal
403 * completion wins, the qc never reaches EH. When error
404 * completion wins, the qc has ATA_QCFLAG_FAILED set.
406 * When SCSI timeout wins, things are a bit more complex.
407 * Normal or error completion can occur after the timeout but
408 * before this point. In such cases, both types of
409 * completions are honored. A scmd is determined to have
410 * timed out iff its associated qc is active and not failed.
412 if (ap
->ops
->error_handler
) {
413 struct scsi_cmnd
*scmd
, *tmp
;
416 spin_lock_irqsave(ap
->lock
, flags
);
418 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
419 struct ata_queued_cmd
*qc
;
421 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
422 qc
= __ata_qc_from_tag(ap
, i
);
423 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
428 if (i
< ATA_MAX_QUEUE
) {
429 /* the scmd has an associated qc */
430 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
431 /* which hasn't failed yet, timeout */
432 qc
->err_mask
|= AC_ERR_TIMEOUT
;
433 qc
->flags
|= ATA_QCFLAG_FAILED
;
437 /* Normal completion occurred after
438 * SCSI timeout but before this point.
439 * Successfully complete it.
441 scmd
->retries
= scmd
->allowed
;
442 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
446 /* If we have timed out qcs. They belong to EH from
447 * this point but the state of the controller is
448 * unknown. Freeze the port to make sure the IRQ
449 * handler doesn't diddle with those qcs. This must
450 * be done atomically w.r.t. setting QCFLAG_FAILED.
453 __ata_port_freeze(ap
);
455 spin_unlock_irqrestore(ap
->lock
, flags
);
457 /* initialize eh_tries */
458 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
460 spin_unlock_wait(ap
->lock
);
463 /* invoke error handler */
464 if (ap
->ops
->error_handler
) {
465 struct ata_link
*link
;
467 /* kill fast drain timer */
468 del_timer_sync(&ap
->fastdrain_timer
);
470 /* process port resume request */
471 ata_eh_handle_port_resume(ap
);
473 /* fetch & clear EH info */
474 spin_lock_irqsave(ap
->lock
, flags
);
476 __ata_port_for_each_link(link
, ap
) {
477 struct ata_eh_context
*ehc
= &link
->eh_context
;
478 struct ata_device
*dev
;
480 memset(&link
->eh_context
, 0, sizeof(link
->eh_context
));
481 link
->eh_context
.i
= link
->eh_info
;
482 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
484 ata_link_for_each_dev(dev
, link
) {
485 int devno
= dev
->devno
;
487 ehc
->saved_xfer_mode
[devno
] = dev
->xfer_mode
;
488 if (ata_ncq_enabled(dev
))
489 ehc
->saved_ncq_enabled
|= 1 << devno
;
492 /* set last reset timestamp to some time in the past */
493 ehc
->last_reset
= jiffies
- 60 * HZ
;
496 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
497 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
498 ap
->excl_link
= NULL
; /* don't maintain exclusion over EH */
500 spin_unlock_irqrestore(ap
->lock
, flags
);
502 /* invoke EH, skip if unloading or suspended */
503 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
504 ap
->ops
->error_handler(ap
);
508 /* process port suspend request */
509 ata_eh_handle_port_suspend(ap
);
511 /* Exception might have happend after ->error_handler
512 * recovered the port but before this point. Repeat
515 spin_lock_irqsave(ap
->lock
, flags
);
517 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
518 if (--ap
->eh_tries
) {
519 spin_unlock_irqrestore(ap
->lock
, flags
);
522 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
523 "tries, giving up\n", ATA_EH_MAX_TRIES
);
524 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
527 /* this run is complete, make sure EH info is clear */
528 __ata_port_for_each_link(link
, ap
)
529 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
531 /* Clear host_eh_scheduled while holding ap->lock such
532 * that if exception occurs after this point but
533 * before EH completion, SCSI midlayer will
536 host
->host_eh_scheduled
= 0;
538 spin_unlock_irqrestore(ap
->lock
, flags
);
540 WARN_ON(ata_qc_from_tag(ap
, ap
->link
.active_tag
) == NULL
);
541 ap
->ops
->eng_timeout(ap
);
544 /* finish or retry handled scmd's and clean up */
545 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
547 scsi_eh_flush_done_q(&ap
->eh_done_q
);
550 spin_lock_irqsave(ap
->lock
, flags
);
552 if (ap
->pflags
& ATA_PFLAG_LOADING
)
553 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
554 else if (ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
)
555 queue_delayed_work(ata_aux_wq
, &ap
->hotplug_task
, 0);
557 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
558 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
560 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
562 /* tell wait_eh that we're done */
563 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
564 wake_up_all(&ap
->eh_wait_q
);
566 spin_unlock_irqrestore(ap
->lock
, flags
);
572 * ata_port_wait_eh - Wait for the currently pending EH to complete
573 * @ap: Port to wait EH for
575 * Wait until the currently pending EH is complete.
578 * Kernel thread context (may sleep).
580 void ata_port_wait_eh(struct ata_port
*ap
)
586 spin_lock_irqsave(ap
->lock
, flags
);
588 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
589 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
590 spin_unlock_irqrestore(ap
->lock
, flags
);
592 spin_lock_irqsave(ap
->lock
, flags
);
594 finish_wait(&ap
->eh_wait_q
, &wait
);
596 spin_unlock_irqrestore(ap
->lock
, flags
);
598 /* make sure SCSI EH is complete */
599 if (scsi_host_in_recovery(ap
->scsi_host
)) {
605 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
610 /* count only non-internal commands */
611 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
612 if (ata_qc_from_tag(ap
, tag
))
618 void ata_eh_fastdrain_timerfn(unsigned long arg
)
620 struct ata_port
*ap
= (void *)arg
;
624 spin_lock_irqsave(ap
->lock
, flags
);
626 cnt
= ata_eh_nr_in_flight(ap
);
632 if (cnt
== ap
->fastdrain_cnt
) {
635 /* No progress during the last interval, tag all
636 * in-flight qcs as timed out and freeze the port.
638 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
639 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
641 qc
->err_mask
|= AC_ERR_TIMEOUT
;
646 /* some qcs have finished, give it another chance */
647 ap
->fastdrain_cnt
= cnt
;
648 ap
->fastdrain_timer
.expires
=
649 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
650 add_timer(&ap
->fastdrain_timer
);
654 spin_unlock_irqrestore(ap
->lock
, flags
);
658 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
659 * @ap: target ATA port
660 * @fastdrain: activate fast drain
662 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
663 * is non-zero and EH wasn't pending before. Fast drain ensures
664 * that EH kicks in in timely manner.
667 * spin_lock_irqsave(host lock)
669 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
673 /* already scheduled? */
674 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
677 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
682 /* do we have in-flight qcs? */
683 cnt
= ata_eh_nr_in_flight(ap
);
687 /* activate fast drain */
688 ap
->fastdrain_cnt
= cnt
;
689 ap
->fastdrain_timer
.expires
=
690 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
691 add_timer(&ap
->fastdrain_timer
);
695 * ata_qc_schedule_eh - schedule qc for error handling
696 * @qc: command to schedule error handling for
698 * Schedule error handling for @qc. EH will kick in as soon as
699 * other commands are drained.
702 * spin_lock_irqsave(host lock)
704 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
706 struct ata_port
*ap
= qc
->ap
;
708 WARN_ON(!ap
->ops
->error_handler
);
710 qc
->flags
|= ATA_QCFLAG_FAILED
;
711 ata_eh_set_pending(ap
, 1);
713 /* The following will fail if timeout has already expired.
714 * ata_scsi_error() takes care of such scmds on EH entry.
715 * Note that ATA_QCFLAG_FAILED is unconditionally set after
716 * this function completes.
718 scsi_req_abort_cmd(qc
->scsicmd
);
722 * ata_port_schedule_eh - schedule error handling without a qc
723 * @ap: ATA port to schedule EH for
725 * Schedule error handling for @ap. EH will kick in as soon as
726 * all commands are drained.
729 * spin_lock_irqsave(host lock)
731 void ata_port_schedule_eh(struct ata_port
*ap
)
733 WARN_ON(!ap
->ops
->error_handler
);
735 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
738 ata_eh_set_pending(ap
, 1);
739 scsi_schedule_eh(ap
->scsi_host
);
741 DPRINTK("port EH scheduled\n");
744 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
746 int tag
, nr_aborted
= 0;
748 WARN_ON(!ap
->ops
->error_handler
);
750 /* we're gonna abort all commands, no need for fast drain */
751 ata_eh_set_pending(ap
, 0);
753 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
754 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
756 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
757 qc
->flags
|= ATA_QCFLAG_FAILED
;
764 ata_port_schedule_eh(ap
);
770 * ata_link_abort - abort all qc's on the link
771 * @link: ATA link to abort qc's for
773 * Abort all active qc's active on @link and schedule EH.
776 * spin_lock_irqsave(host lock)
779 * Number of aborted qc's.
781 int ata_link_abort(struct ata_link
*link
)
783 return ata_do_link_abort(link
->ap
, link
);
787 * ata_port_abort - abort all qc's on the port
788 * @ap: ATA port to abort qc's for
790 * Abort all active qc's of @ap and schedule EH.
793 * spin_lock_irqsave(host_set lock)
796 * Number of aborted qc's.
798 int ata_port_abort(struct ata_port
*ap
)
800 return ata_do_link_abort(ap
, NULL
);
804 * __ata_port_freeze - freeze port
805 * @ap: ATA port to freeze
807 * This function is called when HSM violation or some other
808 * condition disrupts normal operation of the port. Frozen port
809 * is not allowed to perform any operation until the port is
810 * thawed, which usually follows a successful reset.
812 * ap->ops->freeze() callback can be used for freezing the port
813 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
814 * port cannot be frozen hardware-wise, the interrupt handler
815 * must ack and clear interrupts unconditionally while the port
819 * spin_lock_irqsave(host lock)
821 static void __ata_port_freeze(struct ata_port
*ap
)
823 WARN_ON(!ap
->ops
->error_handler
);
828 ap
->pflags
|= ATA_PFLAG_FROZEN
;
830 DPRINTK("ata%u port frozen\n", ap
->print_id
);
834 * ata_port_freeze - abort & freeze port
835 * @ap: ATA port to freeze
837 * Abort and freeze @ap.
840 * spin_lock_irqsave(host lock)
843 * Number of aborted commands.
845 int ata_port_freeze(struct ata_port
*ap
)
849 WARN_ON(!ap
->ops
->error_handler
);
851 nr_aborted
= ata_port_abort(ap
);
852 __ata_port_freeze(ap
);
858 * sata_async_notification - SATA async notification handler
859 * @ap: ATA port where async notification is received
861 * Handler to be called when async notification via SDB FIS is
862 * received. This function schedules EH if necessary.
865 * spin_lock_irqsave(host lock)
868 * 1 if EH is scheduled, 0 otherwise.
870 int sata_async_notification(struct ata_port
*ap
)
875 if (!(ap
->flags
& ATA_FLAG_AN
))
878 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
880 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
882 if (!sata_pmp_attached(ap
) || rc
) {
883 /* PMP is not attached or SNTF is not available */
884 if (!sata_pmp_attached(ap
)) {
885 /* PMP is not attached. Check whether ATAPI
886 * AN is configured. If so, notify media
889 struct ata_device
*dev
= ap
->link
.device
;
891 if ((dev
->class == ATA_DEV_ATAPI
) &&
892 (dev
->flags
& ATA_DFLAG_AN
))
893 ata_scsi_media_change_notify(dev
);
896 /* PMP is attached but SNTF is not available.
897 * ATAPI async media change notification is
898 * not used. The PMP must be reporting PHY
899 * status change, schedule EH.
901 ata_port_schedule_eh(ap
);
905 /* PMP is attached and SNTF is available */
906 struct ata_link
*link
;
908 /* check and notify ATAPI AN */
909 ata_port_for_each_link(link
, ap
) {
910 if (!(sntf
& (1 << link
->pmp
)))
913 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
914 (link
->device
->flags
& ATA_DFLAG_AN
))
915 ata_scsi_media_change_notify(link
->device
);
918 /* If PMP is reporting that PHY status of some
919 * downstream ports has changed, schedule EH.
921 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
922 ata_port_schedule_eh(ap
);
931 * ata_eh_freeze_port - EH helper to freeze port
932 * @ap: ATA port to freeze
939 void ata_eh_freeze_port(struct ata_port
*ap
)
943 if (!ap
->ops
->error_handler
)
946 spin_lock_irqsave(ap
->lock
, flags
);
947 __ata_port_freeze(ap
);
948 spin_unlock_irqrestore(ap
->lock
, flags
);
952 * ata_port_thaw_port - EH helper to thaw port
953 * @ap: ATA port to thaw
955 * Thaw frozen port @ap.
960 void ata_eh_thaw_port(struct ata_port
*ap
)
964 if (!ap
->ops
->error_handler
)
967 spin_lock_irqsave(ap
->lock
, flags
);
969 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
974 spin_unlock_irqrestore(ap
->lock
, flags
);
976 DPRINTK("ata%u port thawed\n", ap
->print_id
);
979 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
984 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
986 struct ata_port
*ap
= qc
->ap
;
987 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
990 spin_lock_irqsave(ap
->lock
, flags
);
991 qc
->scsidone
= ata_eh_scsidone
;
992 __ata_qc_complete(qc
);
993 WARN_ON(ata_tag_valid(qc
->tag
));
994 spin_unlock_irqrestore(ap
->lock
, flags
);
996 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1000 * ata_eh_qc_complete - Complete an active ATA command from EH
1001 * @qc: Command to complete
1003 * Indicate to the mid and upper layers that an ATA command has
1004 * completed. To be used from EH.
1006 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1008 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1009 scmd
->retries
= scmd
->allowed
;
1010 __ata_eh_qc_complete(qc
);
1014 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1015 * @qc: Command to retry
1017 * Indicate to the mid and upper layers that an ATA command
1018 * should be retried. To be used from EH.
1020 * SCSI midlayer limits the number of retries to scmd->allowed.
1021 * scmd->retries is decremented for commands which get retried
1022 * due to unrelated failures (qc->err_mask is zero).
1024 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1026 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1027 if (!qc
->err_mask
&& scmd
->retries
)
1029 __ata_eh_qc_complete(qc
);
1033 * ata_eh_detach_dev - detach ATA device
1034 * @dev: ATA device to detach
1041 void ata_eh_detach_dev(struct ata_device
*dev
)
1043 struct ata_link
*link
= dev
->link
;
1044 struct ata_port
*ap
= link
->ap
;
1045 unsigned long flags
;
1047 ata_dev_disable(dev
);
1049 spin_lock_irqsave(ap
->lock
, flags
);
1051 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1053 if (ata_scsi_offline_dev(dev
)) {
1054 dev
->flags
|= ATA_DFLAG_DETACHED
;
1055 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1058 /* clear per-dev EH actions */
1059 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1060 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1062 spin_unlock_irqrestore(ap
->lock
, flags
);
1066 * ata_eh_about_to_do - about to perform eh_action
1067 * @link: target ATA link
1068 * @dev: target ATA dev for per-dev action (can be NULL)
1069 * @action: action about to be performed
1071 * Called just before performing EH actions to clear related bits
1072 * in @link->eh_info such that eh actions are not unnecessarily
1078 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1079 unsigned int action
)
1081 struct ata_port
*ap
= link
->ap
;
1082 struct ata_eh_info
*ehi
= &link
->eh_info
;
1083 struct ata_eh_context
*ehc
= &link
->eh_context
;
1084 unsigned long flags
;
1086 spin_lock_irqsave(ap
->lock
, flags
);
1088 ata_eh_clear_action(link
, dev
, ehi
, action
);
1090 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
))
1091 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1093 spin_unlock_irqrestore(ap
->lock
, flags
);
1097 * ata_eh_done - EH action complete
1098 * @ap: target ATA port
1099 * @dev: target ATA dev for per-dev action (can be NULL)
1100 * @action: action just completed
1102 * Called right after performing EH actions to clear related bits
1103 * in @link->eh_context.
1108 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1109 unsigned int action
)
1111 struct ata_eh_context
*ehc
= &link
->eh_context
;
1113 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1117 * ata_err_string - convert err_mask to descriptive string
1118 * @err_mask: error mask to convert to string
1120 * Convert @err_mask to descriptive string. Errors are
1121 * prioritized according to severity and only the most severe
1122 * error is reported.
1128 * Descriptive string for @err_mask
1130 static const char *ata_err_string(unsigned int err_mask
)
1132 if (err_mask
& AC_ERR_HOST_BUS
)
1133 return "host bus error";
1134 if (err_mask
& AC_ERR_ATA_BUS
)
1135 return "ATA bus error";
1136 if (err_mask
& AC_ERR_TIMEOUT
)
1138 if (err_mask
& AC_ERR_HSM
)
1139 return "HSM violation";
1140 if (err_mask
& AC_ERR_SYSTEM
)
1141 return "internal error";
1142 if (err_mask
& AC_ERR_MEDIA
)
1143 return "media error";
1144 if (err_mask
& AC_ERR_INVALID
)
1145 return "invalid argument";
1146 if (err_mask
& AC_ERR_DEV
)
1147 return "device error";
1148 return "unknown error";
1152 * ata_read_log_page - read a specific log page
1153 * @dev: target device
1154 * @page: page to read
1155 * @buf: buffer to store read page
1156 * @sectors: number of sectors to read
1158 * Read log page using READ_LOG_EXT command.
1161 * Kernel thread context (may sleep).
1164 * 0 on success, AC_ERR_* mask otherwise.
1166 static unsigned int ata_read_log_page(struct ata_device
*dev
,
1167 u8 page
, void *buf
, unsigned int sectors
)
1169 struct ata_taskfile tf
;
1170 unsigned int err_mask
;
1172 DPRINTK("read log page - page %d\n", page
);
1174 ata_tf_init(dev
, &tf
);
1175 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1178 tf
.hob_nsect
= sectors
>> 8;
1179 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1180 tf
.protocol
= ATA_PROT_PIO
;
1182 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1183 buf
, sectors
* ATA_SECT_SIZE
, 0);
1185 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1190 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1191 * @dev: Device to read log page 10h from
1192 * @tag: Resulting tag of the failed command
1193 * @tf: Resulting taskfile registers of the failed command
1195 * Read log page 10h to obtain NCQ error details and clear error
1199 * Kernel thread context (may sleep).
1202 * 0 on success, -errno otherwise.
1204 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1205 int *tag
, struct ata_taskfile
*tf
)
1207 u8
*buf
= dev
->link
->ap
->sector_buf
;
1208 unsigned int err_mask
;
1212 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
1217 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1220 ata_dev_printk(dev
, KERN_WARNING
,
1221 "invalid checksum 0x%x on log page 10h\n", csum
);
1226 *tag
= buf
[0] & 0x1f;
1228 tf
->command
= buf
[2];
1229 tf
->feature
= buf
[3];
1233 tf
->device
= buf
[7];
1234 tf
->hob_lbal
= buf
[8];
1235 tf
->hob_lbam
= buf
[9];
1236 tf
->hob_lbah
= buf
[10];
1237 tf
->nsect
= buf
[12];
1238 tf
->hob_nsect
= buf
[13];
1244 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1245 * @dev: device to perform REQUEST_SENSE to
1246 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1248 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1249 * SENSE. This function is EH helper.
1252 * Kernel thread context (may sleep).
1255 * 0 on success, AC_ERR_* mask on failure
1257 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd
*qc
)
1259 struct ata_device
*dev
= qc
->dev
;
1260 unsigned char *sense_buf
= qc
->scsicmd
->sense_buffer
;
1261 struct ata_port
*ap
= dev
->link
->ap
;
1262 struct ata_taskfile tf
;
1263 u8 cdb
[ATAPI_CDB_LEN
];
1265 DPRINTK("ATAPI request sense\n");
1267 /* FIXME: is this needed? */
1268 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1270 /* initialize sense_buf with the error register,
1271 * for the case where they are -not- overwritten
1273 sense_buf
[0] = 0x70;
1274 sense_buf
[2] = qc
->result_tf
.feature
>> 4;
1276 /* some devices time out if garbage left in tf */
1277 ata_tf_init(dev
, &tf
);
1279 memset(cdb
, 0, ATAPI_CDB_LEN
);
1280 cdb
[0] = REQUEST_SENSE
;
1281 cdb
[4] = SCSI_SENSE_BUFFERSIZE
;
1283 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1284 tf
.command
= ATA_CMD_PACKET
;
1286 /* is it pointless to prefer PIO for "safety reasons"? */
1287 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1288 tf
.protocol
= ATAPI_PROT_DMA
;
1289 tf
.feature
|= ATAPI_PKT_DMA
;
1291 tf
.protocol
= ATAPI_PROT_PIO
;
1292 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1296 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1297 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1301 * ata_eh_analyze_serror - analyze SError for a failed port
1302 * @link: ATA link to analyze SError for
1304 * Analyze SError if available and further determine cause of
1310 static void ata_eh_analyze_serror(struct ata_link
*link
)
1312 struct ata_eh_context
*ehc
= &link
->eh_context
;
1313 u32 serror
= ehc
->i
.serror
;
1314 unsigned int err_mask
= 0, action
= 0;
1317 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1318 err_mask
|= AC_ERR_ATA_BUS
;
1319 action
|= ATA_EH_RESET
;
1321 if (serror
& SERR_PROTOCOL
) {
1322 err_mask
|= AC_ERR_HSM
;
1323 action
|= ATA_EH_RESET
;
1325 if (serror
& SERR_INTERNAL
) {
1326 err_mask
|= AC_ERR_SYSTEM
;
1327 action
|= ATA_EH_RESET
;
1330 /* Determine whether a hotplug event has occurred. Both
1331 * SError.N/X are considered hotplug events for enabled or
1332 * host links. For disabled PMP links, only N bit is
1333 * considered as X bit is left at 1 for link plugging.
1337 if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1338 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1340 hotplug_mask
= SERR_PHYRDY_CHG
;
1342 if (serror
& hotplug_mask
)
1343 ata_ehi_hotplugged(&ehc
->i
);
1345 ehc
->i
.err_mask
|= err_mask
;
1346 ehc
->i
.action
|= action
;
1350 * ata_eh_analyze_ncq_error - analyze NCQ error
1351 * @link: ATA link to analyze NCQ error for
1353 * Read log page 10h, determine the offending qc and acquire
1354 * error status TF. For NCQ device errors, all LLDDs have to do
1355 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1359 * Kernel thread context (may sleep).
1361 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1363 struct ata_port
*ap
= link
->ap
;
1364 struct ata_eh_context
*ehc
= &link
->eh_context
;
1365 struct ata_device
*dev
= link
->device
;
1366 struct ata_queued_cmd
*qc
;
1367 struct ata_taskfile tf
;
1370 /* if frozen, we can't do much */
1371 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1374 /* is it NCQ device error? */
1375 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1378 /* has LLDD analyzed already? */
1379 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1380 qc
= __ata_qc_from_tag(ap
, tag
);
1382 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1389 /* okay, this error is ours */
1390 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1392 ata_link_printk(link
, KERN_ERR
, "failed to read log page 10h "
1393 "(errno=%d)\n", rc
);
1397 if (!(link
->sactive
& (1 << tag
))) {
1398 ata_link_printk(link
, KERN_ERR
, "log page 10h reported "
1399 "inactive tag %d\n", tag
);
1403 /* we've got the perpetrator, condemn it */
1404 qc
= __ata_qc_from_tag(ap
, tag
);
1405 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1406 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1407 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1408 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1412 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1413 * @qc: qc to analyze
1414 * @tf: Taskfile registers to analyze
1416 * Analyze taskfile of @qc and further determine cause of
1417 * failure. This function also requests ATAPI sense data if
1421 * Kernel thread context (may sleep).
1424 * Determined recovery action
1426 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1427 const struct ata_taskfile
*tf
)
1429 unsigned int tmp
, action
= 0;
1430 u8 stat
= tf
->command
, err
= tf
->feature
;
1432 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1433 qc
->err_mask
|= AC_ERR_HSM
;
1434 return ATA_EH_RESET
;
1437 if (stat
& (ATA_ERR
| ATA_DF
))
1438 qc
->err_mask
|= AC_ERR_DEV
;
1442 switch (qc
->dev
->class) {
1445 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1447 qc
->err_mask
|= AC_ERR_MEDIA
;
1449 qc
->err_mask
|= AC_ERR_INVALID
;
1453 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1454 tmp
= atapi_eh_request_sense(qc
);
1456 /* ATA_QCFLAG_SENSE_VALID is used to
1457 * tell atapi_qc_complete() that sense
1458 * data is already valid.
1460 * TODO: interpret sense data and set
1461 * appropriate err_mask.
1463 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1465 qc
->err_mask
|= tmp
;
1469 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1470 action
|= ATA_EH_RESET
;
1475 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1480 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1484 base
= ATA_ECAT_DUBIOUS_NONE
;
1486 if (err_mask
& AC_ERR_ATA_BUS
)
1487 return base
+ ATA_ECAT_ATA_BUS
;
1489 if (err_mask
& AC_ERR_TIMEOUT
)
1490 return base
+ ATA_ECAT_TOUT_HSM
;
1492 if (eflags
& ATA_EFLAG_IS_IO
) {
1493 if (err_mask
& AC_ERR_HSM
)
1494 return base
+ ATA_ECAT_TOUT_HSM
;
1496 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1497 return base
+ ATA_ECAT_UNK_DEV
;
1503 struct speed_down_verdict_arg
{
1506 int nr_errors
[ATA_ECAT_NR
];
1509 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1511 struct speed_down_verdict_arg
*arg
= void_arg
;
1514 if (ent
->timestamp
< arg
->since
)
1517 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1519 arg
->nr_errors
[cat
]++;
1525 * ata_eh_speed_down_verdict - Determine speed down verdict
1526 * @dev: Device of interest
1528 * This function examines error ring of @dev and determines
1529 * whether NCQ needs to be turned off, transfer speed should be
1530 * stepped down, or falling back to PIO is necessary.
1532 * ECAT_ATA_BUS : ATA_BUS error for any command
1534 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1537 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1539 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1540 * data transfer hasn't been verified.
1544 * NCQ_OFF : Turn off NCQ.
1546 * SPEED_DOWN : Speed down transfer speed but don't fall back
1549 * FALLBACK_TO_PIO : Fall back to PIO.
1551 * Even if multiple verdicts are returned, only one action is
1552 * taken per error. An action triggered by non-DUBIOUS errors
1553 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1554 * This is to expedite speed down decisions right after device is
1555 * initially configured.
1557 * The followings are speed down rules. #1 and #2 deal with
1560 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1561 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1563 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1564 * occurred during last 5 mins, NCQ_OFF.
1566 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1567 * ocurred during last 5 mins, FALLBACK_TO_PIO
1569 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1570 * during last 10 mins, NCQ_OFF.
1572 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1573 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1576 * Inherited from caller.
1579 * OR of ATA_EH_SPDN_* flags.
1581 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1583 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1584 u64 j64
= get_jiffies_64();
1585 struct speed_down_verdict_arg arg
;
1586 unsigned int verdict
= 0;
1588 /* scan past 5 mins of error history */
1589 memset(&arg
, 0, sizeof(arg
));
1590 arg
.since
= j64
- min(j64
, j5mins
);
1591 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1593 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1594 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1595 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1596 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1598 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1599 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1600 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1602 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1603 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1604 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1605 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1607 /* scan past 10 mins of error history */
1608 memset(&arg
, 0, sizeof(arg
));
1609 arg
.since
= j64
- min(j64
, j10mins
);
1610 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1612 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1613 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1614 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1616 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1617 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1618 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1619 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1625 * ata_eh_speed_down - record error and speed down if necessary
1626 * @dev: Failed device
1627 * @eflags: mask of ATA_EFLAG_* flags
1628 * @err_mask: err_mask of the error
1630 * Record error and examine error history to determine whether
1631 * adjusting transmission speed is necessary. It also sets
1632 * transmission limits appropriately if such adjustment is
1636 * Kernel thread context (may sleep).
1639 * Determined recovery action.
1641 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
1642 unsigned int eflags
, unsigned int err_mask
)
1644 struct ata_link
*link
= dev
->link
;
1646 unsigned int verdict
;
1647 unsigned int action
= 0;
1649 /* don't bother if Cat-0 error */
1650 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
1653 /* record error and determine whether speed down is necessary */
1654 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
1655 verdict
= ata_eh_speed_down_verdict(dev
);
1658 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
1659 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
1660 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
1661 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
1662 ata_dev_printk(dev
, KERN_WARNING
,
1663 "NCQ disabled due to excessive errors\n");
1668 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
1669 /* speed down SATA link speed if possible */
1670 if (sata_down_spd_limit(link
) == 0) {
1671 action
|= ATA_EH_RESET
;
1675 /* lower transfer mode */
1676 if (dev
->spdn_cnt
< 2) {
1677 static const int dma_dnxfer_sel
[] =
1678 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
1679 static const int pio_dnxfer_sel
[] =
1680 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
1683 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
1684 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
1686 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
1690 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
1691 action
|= ATA_EH_RESET
;
1697 /* Fall back to PIO? Slowing down to PIO is meaningless for
1698 * SATA ATA devices. Consider it only for PATA and SATAPI.
1700 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
1701 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
1702 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
1703 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
1705 action
|= ATA_EH_RESET
;
1712 /* device has been slowed down, blow error history */
1713 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
1714 ata_ering_clear(&dev
->ering
);
1719 * ata_eh_link_autopsy - analyze error and determine recovery action
1720 * @link: host link to perform autopsy on
1722 * Analyze why @link failed and determine which recovery actions
1723 * are needed. This function also sets more detailed AC_ERR_*
1724 * values and fills sense data for ATAPI CHECK SENSE.
1727 * Kernel thread context (may sleep).
1729 static void ata_eh_link_autopsy(struct ata_link
*link
)
1731 struct ata_port
*ap
= link
->ap
;
1732 struct ata_eh_context
*ehc
= &link
->eh_context
;
1733 struct ata_device
*dev
;
1734 unsigned int all_err_mask
= 0, eflags
= 0;
1741 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
1744 /* obtain and analyze SError */
1745 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
1747 ehc
->i
.serror
|= serror
;
1748 ata_eh_analyze_serror(link
);
1749 } else if (rc
!= -EOPNOTSUPP
) {
1750 /* SError read failed, force reset and probing */
1751 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
1752 ehc
->i
.action
|= ATA_EH_RESET
;
1753 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
1756 /* analyze NCQ failure */
1757 ata_eh_analyze_ncq_error(link
);
1759 /* any real error trumps AC_ERR_OTHER */
1760 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1761 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1763 all_err_mask
|= ehc
->i
.err_mask
;
1765 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1766 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1768 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || qc
->dev
->link
!= link
)
1771 /* inherit upper level err_mask */
1772 qc
->err_mask
|= ehc
->i
.err_mask
;
1775 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1777 /* DEV errors are probably spurious in case of ATA_BUS error */
1778 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1779 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1782 /* any real error trumps unknown error */
1783 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1784 qc
->err_mask
&= ~AC_ERR_OTHER
;
1786 /* SENSE_VALID trumps dev/unknown error and revalidation */
1787 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
1788 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1790 /* determine whether the command is worth retrying */
1791 if (!(qc
->err_mask
& AC_ERR_INVALID
) &&
1792 ((qc
->flags
& ATA_QCFLAG_IO
) || qc
->err_mask
!= AC_ERR_DEV
))
1793 qc
->flags
|= ATA_QCFLAG_RETRY
;
1795 /* accumulate error info */
1796 ehc
->i
.dev
= qc
->dev
;
1797 all_err_mask
|= qc
->err_mask
;
1798 if (qc
->flags
& ATA_QCFLAG_IO
)
1799 eflags
|= ATA_EFLAG_IS_IO
;
1802 /* enforce default EH actions */
1803 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
1804 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
1805 ehc
->i
.action
|= ATA_EH_RESET
;
1806 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
1807 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
1808 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1810 /* If we have offending qcs and the associated failed device,
1811 * perform per-dev EH action only on the offending device.
1814 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
1815 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
1816 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
1819 /* propagate timeout to host link */
1820 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
1821 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
1823 /* record error and consider speeding down */
1825 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
1826 ata_dev_enabled(link
->device
))))
1830 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
1831 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
1832 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
1839 * ata_eh_autopsy - analyze error and determine recovery action
1840 * @ap: host port to perform autopsy on
1842 * Analyze all links of @ap and determine why they failed and
1843 * which recovery actions are needed.
1846 * Kernel thread context (may sleep).
1848 void ata_eh_autopsy(struct ata_port
*ap
)
1850 struct ata_link
*link
;
1852 ata_port_for_each_link(link
, ap
)
1853 ata_eh_link_autopsy(link
);
1855 /* Autopsy of fanout ports can affect host link autopsy.
1856 * Perform host link autopsy last.
1858 if (sata_pmp_attached(ap
))
1859 ata_eh_link_autopsy(&ap
->link
);
1863 * ata_eh_link_report - report error handling to user
1864 * @link: ATA link EH is going on
1866 * Report EH to user.
1871 static void ata_eh_link_report(struct ata_link
*link
)
1873 struct ata_port
*ap
= link
->ap
;
1874 struct ata_eh_context
*ehc
= &link
->eh_context
;
1875 const char *frozen
, *desc
;
1877 int tag
, nr_failed
= 0;
1879 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
1883 if (ehc
->i
.desc
[0] != '\0')
1886 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1887 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1889 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || qc
->dev
->link
!= link
||
1890 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
1891 qc
->err_mask
== AC_ERR_DEV
))
1893 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
1899 if (!nr_failed
&& !ehc
->i
.err_mask
)
1903 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1906 memset(tries_buf
, 0, sizeof(tries_buf
));
1907 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
1908 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
1912 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
1913 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1914 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
1915 ehc
->i
.action
, frozen
, tries_buf
);
1917 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "%s\n", desc
);
1919 ata_link_printk(link
, KERN_ERR
, "exception Emask 0x%x "
1920 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1921 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
1922 ehc
->i
.action
, frozen
, tries_buf
);
1924 ata_link_printk(link
, KERN_ERR
, "%s\n", desc
);
1928 ata_port_printk(ap
, KERN_ERR
,
1929 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
1930 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
1931 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
1932 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
1933 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
1934 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
1935 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
1936 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
1937 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
1938 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
1939 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
1940 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
1941 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
1942 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
1943 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
1944 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
1945 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
1946 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
1948 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1949 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1950 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
1951 const u8
*cdb
= qc
->cdb
;
1952 char data_buf
[20] = "";
1953 char cdb_buf
[70] = "";
1955 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
1956 qc
->dev
->link
!= link
|| !qc
->err_mask
)
1959 if (qc
->dma_dir
!= DMA_NONE
) {
1960 static const char *dma_str
[] = {
1961 [DMA_BIDIRECTIONAL
] = "bidi",
1962 [DMA_TO_DEVICE
] = "out",
1963 [DMA_FROM_DEVICE
] = "in",
1965 static const char *prot_str
[] = {
1966 [ATA_PROT_PIO
] = "pio",
1967 [ATA_PROT_DMA
] = "dma",
1968 [ATA_PROT_NCQ
] = "ncq",
1969 [ATAPI_PROT_PIO
] = "pio",
1970 [ATAPI_PROT_DMA
] = "dma",
1973 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
1974 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
1975 dma_str
[qc
->dma_dir
]);
1978 if (ata_is_atapi(qc
->tf
.protocol
))
1979 snprintf(cdb_buf
, sizeof(cdb_buf
),
1980 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
1981 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
1982 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
1983 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
1984 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
1985 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
1987 ata_dev_printk(qc
->dev
, KERN_ERR
,
1988 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1990 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1991 "Emask 0x%x (%s)%s\n",
1992 cmd
->command
, cmd
->feature
, cmd
->nsect
,
1993 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
1994 cmd
->hob_feature
, cmd
->hob_nsect
,
1995 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
1996 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
1997 res
->command
, res
->feature
, res
->nsect
,
1998 res
->lbal
, res
->lbam
, res
->lbah
,
1999 res
->hob_feature
, res
->hob_nsect
,
2000 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2001 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2002 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2004 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2006 if (res
->command
& ATA_BUSY
)
2007 ata_dev_printk(qc
->dev
, KERN_ERR
,
2008 "status: { Busy }\n");
2010 ata_dev_printk(qc
->dev
, KERN_ERR
,
2011 "status: { %s%s%s%s}\n",
2012 res
->command
& ATA_DRDY
? "DRDY " : "",
2013 res
->command
& ATA_DF
? "DF " : "",
2014 res
->command
& ATA_DRQ
? "DRQ " : "",
2015 res
->command
& ATA_ERR
? "ERR " : "");
2018 if (cmd
->command
!= ATA_CMD_PACKET
&&
2019 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2021 ata_dev_printk(qc
->dev
, KERN_ERR
,
2022 "error: { %s%s%s%s}\n",
2023 res
->feature
& ATA_ICRC
? "ICRC " : "",
2024 res
->feature
& ATA_UNC
? "UNC " : "",
2025 res
->feature
& ATA_IDNF
? "IDNF " : "",
2026 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2031 * ata_eh_report - report error handling to user
2032 * @ap: ATA port to report EH about
2034 * Report EH to user.
2039 void ata_eh_report(struct ata_port
*ap
)
2041 struct ata_link
*link
;
2043 __ata_port_for_each_link(link
, ap
)
2044 ata_eh_link_report(link
);
2047 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2048 unsigned int *classes
, unsigned long deadline
)
2050 struct ata_device
*dev
;
2052 ata_link_for_each_dev(dev
, link
)
2053 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2055 return reset(link
, classes
, deadline
);
2058 static int ata_eh_followup_srst_needed(struct ata_link
*link
,
2059 int rc
, int classify
,
2060 const unsigned int *classes
)
2062 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2064 if (rc
== -EAGAIN
) {
2071 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2076 int ata_eh_reset(struct ata_link
*link
, int classify
,
2077 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2078 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2080 struct ata_port
*ap
= link
->ap
;
2081 struct ata_eh_context
*ehc
= &link
->eh_context
;
2082 unsigned int *classes
= ehc
->classes
;
2083 unsigned int lflags
= link
->flags
;
2084 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2085 int max_tries
= 0, try = 0;
2086 struct ata_device
*dev
;
2087 unsigned long deadline
, now
;
2088 ata_reset_fn_t reset
;
2089 unsigned long flags
;
2096 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2100 deadline
= ata_deadline(ehc
->last_reset
, ATA_EH_RESET_COOL_DOWN
);
2101 if (time_before(now
, deadline
))
2102 schedule_timeout_uninterruptible(deadline
- now
);
2104 spin_lock_irqsave(ap
->lock
, flags
);
2105 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2106 spin_unlock_irqrestore(ap
->lock
, flags
);
2108 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2109 ehc
->last_reset
= jiffies
;
2111 ata_link_for_each_dev(dev
, link
) {
2112 /* If we issue an SRST then an ATA drive (not ATAPI)
2113 * may change configuration and be in PIO0 timing. If
2114 * we do a hard reset (or are coming from power on)
2115 * this is true for ATA or ATAPI. Until we've set a
2116 * suitable controller mode we should not touch the
2117 * bus as we may be talking too fast.
2119 dev
->pio_mode
= XFER_PIO_0
;
2121 /* If the controller has a pio mode setup function
2122 * then use it to set the chipset to rights. Don't
2123 * touch the DMA setup as that will be dealt with when
2124 * configuring devices.
2126 if (ap
->ops
->set_piomode
)
2127 ap
->ops
->set_piomode(ap
, dev
);
2130 /* prefer hardreset */
2132 ehc
->i
.action
&= ~ATA_EH_RESET
;
2135 ehc
->i
.action
= ATA_EH_HARDRESET
;
2136 } else if (softreset
) {
2138 ehc
->i
.action
= ATA_EH_SOFTRESET
;
2143 ata_deadline(jiffies
, ATA_EH_PRERESET_TIMEOUT
));
2145 if (rc
== -ENOENT
) {
2146 ata_link_printk(link
, KERN_DEBUG
,
2147 "port disabled. ignoring.\n");
2148 ehc
->i
.action
&= ~ATA_EH_RESET
;
2150 ata_link_for_each_dev(dev
, link
)
2151 classes
[dev
->devno
] = ATA_DEV_NONE
;
2155 ata_link_printk(link
, KERN_ERR
,
2156 "prereset failed (errno=%d)\n", rc
);
2160 /* prereset() might have cleared ATA_EH_RESET. If so,
2161 * bang classes and return.
2163 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2164 ata_link_for_each_dev(dev
, link
)
2165 classes
[dev
->devno
] = ATA_DEV_NONE
;
2175 ehc
->last_reset
= jiffies
;
2176 if (ata_is_host_link(link
))
2177 ata_eh_freeze_port(ap
);
2179 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2183 ata_link_printk(link
, KERN_INFO
, "%s resetting link\n",
2184 reset
== softreset
? "soft" : "hard");
2186 /* mark that this EH session started with reset */
2187 if (reset
== hardreset
)
2188 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2190 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2192 rc
= ata_do_reset(link
, reset
, classes
, deadline
);
2194 if (reset
== hardreset
&&
2195 ata_eh_followup_srst_needed(link
, rc
, classify
, classes
)) {
2196 /* okay, let's do follow-up softreset */
2200 ata_link_printk(link
, KERN_ERR
,
2201 "follow-up softreset required "
2202 "but no softreset avaliable\n");
2207 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2208 rc
= ata_do_reset(link
, reset
, classes
, deadline
);
2211 /* -EAGAIN can happen if we skipped followup SRST */
2212 if (rc
&& rc
!= -EAGAIN
)
2216 ata_link_printk(link
, KERN_INFO
, "no reset method "
2217 "available, skipping reset\n");
2218 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2219 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2223 * Post-reset processing
2225 ata_link_for_each_dev(dev
, link
) {
2226 /* After the reset, the device state is PIO 0 and the
2227 * controller state is undefined. Reset also wakes up
2228 * drives from sleeping mode.
2230 dev
->pio_mode
= XFER_PIO_0
;
2231 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2233 if (ata_link_offline(link
))
2236 /* apply class override */
2237 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2238 classes
[dev
->devno
] = ATA_DEV_ATA
;
2239 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2240 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
; /* not yet */
2243 /* record current link speed */
2244 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2245 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2248 if (ata_is_host_link(link
))
2249 ata_eh_thaw_port(ap
);
2251 /* postreset() should clear hardware SError. Although SError
2252 * is cleared during link resume, clearing SError here is
2253 * necessary as some PHYs raise hotplug events after SRST.
2254 * This introduces race condition where hotplug occurs between
2255 * reset and here. This race is mediated by cross checking
2256 * link onlineness and classification result later.
2259 postreset(link
, classes
);
2261 /* clear cached SError */
2262 spin_lock_irqsave(link
->ap
->lock
, flags
);
2263 link
->eh_info
.serror
= 0;
2264 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2266 /* Make sure onlineness and classification result correspond.
2267 * Hotplug could have happened during reset and some
2268 * controllers fail to wait while a drive is spinning up after
2269 * being hotplugged causing misdetection. By cross checking
2270 * link onlineness and classification result, those conditions
2271 * can be reliably detected and retried.
2274 ata_link_for_each_dev(dev
, link
) {
2275 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2276 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
)
2277 classes
[dev
->devno
] = ATA_DEV_NONE
;
2282 if (classify
&& !nr_known
&& ata_link_online(link
)) {
2283 if (try < max_tries
) {
2284 ata_link_printk(link
, KERN_WARNING
, "link online but "
2285 "device misclassified, retrying\n");
2289 ata_link_printk(link
, KERN_WARNING
,
2290 "link online but device misclassified, "
2291 "device detection might fail\n");
2294 /* reset successful, schedule revalidation */
2295 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2296 ehc
->last_reset
= jiffies
;
2297 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2301 /* clear hotplug flag */
2302 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2304 spin_lock_irqsave(ap
->lock
, flags
);
2305 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2306 spin_unlock_irqrestore(ap
->lock
, flags
);
2311 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2312 if (!ata_is_host_link(link
) &&
2313 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2316 if (rc
== -ERESTART
|| try >= max_tries
)
2320 if (time_before(now
, deadline
)) {
2321 unsigned long delta
= deadline
- now
;
2323 ata_link_printk(link
, KERN_WARNING
,
2324 "reset failed (errno=%d), retrying in %u secs\n",
2325 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2328 delta
= schedule_timeout_uninterruptible(delta
);
2331 if (rc
== -EPIPE
|| try == max_tries
- 1)
2332 sata_down_spd_limit(link
);
2338 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
2339 struct ata_device
**r_failed_dev
)
2341 struct ata_port
*ap
= link
->ap
;
2342 struct ata_eh_context
*ehc
= &link
->eh_context
;
2343 struct ata_device
*dev
;
2344 unsigned int new_mask
= 0;
2345 unsigned long flags
;
2350 /* For PATA drive side cable detection to work, IDENTIFY must
2351 * be done backwards such that PDIAG- is released by the slave
2352 * device before the master device is identified.
2354 ata_link_for_each_dev_reverse(dev
, link
) {
2355 unsigned int action
= ata_eh_dev_action(dev
);
2356 unsigned int readid_flags
= 0;
2358 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
2359 readid_flags
|= ATA_READID_POSTRESET
;
2361 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
2362 WARN_ON(dev
->class == ATA_DEV_PMP
);
2364 if (ata_link_offline(link
)) {
2369 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
2370 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
2375 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
2377 /* Configuration may have changed, reconfigure
2380 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2382 /* schedule the scsi_rescan_device() here */
2383 queue_work(ata_aux_wq
, &(ap
->scsi_rescan_task
));
2384 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
2385 ehc
->tries
[dev
->devno
] &&
2386 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
2387 dev
->class = ehc
->classes
[dev
->devno
];
2389 if (dev
->class == ATA_DEV_PMP
)
2390 rc
= sata_pmp_attach(dev
);
2392 rc
= ata_dev_read_id(dev
, &dev
->class,
2393 readid_flags
, dev
->id
);
2396 new_mask
|= 1 << dev
->devno
;
2399 /* IDENTIFY was issued to non-existent
2400 * device. No need to reset. Just
2401 * thaw and kill the device.
2403 ata_eh_thaw_port(ap
);
2404 dev
->class = ATA_DEV_UNKNOWN
;
2407 dev
->class = ATA_DEV_UNKNOWN
;
2413 /* PDIAG- should have been released, ask cable type if post-reset */
2414 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
2415 if (ap
->ops
->cable_detect
)
2416 ap
->cbl
= ap
->ops
->cable_detect(ap
);
2420 /* Configure new devices forward such that user doesn't see
2421 * device detection messages backwards.
2423 ata_link_for_each_dev(dev
, link
) {
2424 if (!(new_mask
& (1 << dev
->devno
)) ||
2425 dev
->class == ATA_DEV_PMP
)
2428 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
2429 rc
= ata_dev_configure(dev
);
2430 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
2434 spin_lock_irqsave(ap
->lock
, flags
);
2435 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
2436 spin_unlock_irqrestore(ap
->lock
, flags
);
2438 /* new device discovered, configure xfermode */
2439 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2445 *r_failed_dev
= dev
;
2446 DPRINTK("EXIT rc=%d\n", rc
);
2451 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2452 * @link: link on which timings will be programmed
2453 * @r_failed_dev: out paramter for failed device
2455 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2456 * ata_set_mode() fails, pointer to the failing device is
2457 * returned in @r_failed_dev.
2460 * PCI/etc. bus probe sem.
2463 * 0 on success, negative errno otherwise
2465 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
2467 struct ata_port
*ap
= link
->ap
;
2468 struct ata_device
*dev
;
2471 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2472 ata_link_for_each_dev(dev
, link
) {
2473 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
2474 struct ata_ering_entry
*ent
;
2476 ent
= ata_ering_top(&dev
->ering
);
2478 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
2482 /* has private set_mode? */
2483 if (ap
->ops
->set_mode
)
2484 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
2486 rc
= ata_do_set_mode(link
, r_failed_dev
);
2488 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2489 ata_link_for_each_dev(dev
, link
) {
2490 struct ata_eh_context
*ehc
= &link
->eh_context
;
2491 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
2492 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
2494 if (dev
->xfer_mode
!= saved_xfer_mode
||
2495 ata_ncq_enabled(dev
) != saved_ncq
)
2496 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
2502 static int ata_link_nr_enabled(struct ata_link
*link
)
2504 struct ata_device
*dev
;
2507 ata_link_for_each_dev(dev
, link
)
2508 if (ata_dev_enabled(dev
))
2513 static int ata_link_nr_vacant(struct ata_link
*link
)
2515 struct ata_device
*dev
;
2518 ata_link_for_each_dev(dev
, link
)
2519 if (dev
->class == ATA_DEV_UNKNOWN
)
2524 static int ata_eh_skip_recovery(struct ata_link
*link
)
2526 struct ata_port
*ap
= link
->ap
;
2527 struct ata_eh_context
*ehc
= &link
->eh_context
;
2528 struct ata_device
*dev
;
2530 /* skip disabled links */
2531 if (link
->flags
& ATA_LFLAG_DISABLED
)
2534 /* thaw frozen port and recover failed devices */
2535 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
2538 /* reset at least once if reset is requested */
2539 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
2540 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
2543 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2544 ata_link_for_each_dev(dev
, link
) {
2545 if (dev
->class == ATA_DEV_UNKNOWN
&&
2546 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
2553 static int ata_eh_schedule_probe(struct ata_device
*dev
)
2555 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2557 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
2558 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
2561 ata_eh_detach_dev(dev
);
2563 ehc
->did_probe_mask
|= (1 << dev
->devno
);
2564 ehc
->i
.action
|= ATA_EH_RESET
;
2565 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
2566 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
2571 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
2573 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2575 ehc
->tries
[dev
->devno
]--;
2579 /* device missing or wrong IDENTIFY data, schedule probing */
2580 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
2582 /* give it just one more chance */
2583 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
2585 if (ehc
->tries
[dev
->devno
] == 1 && dev
->pio_mode
> XFER_PIO_0
) {
2586 /* This is the last chance, better to slow
2587 * down than lose it.
2589 sata_down_spd_limit(dev
->link
);
2590 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
2594 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
2595 /* disable device if it has used up all its chances */
2596 ata_dev_disable(dev
);
2598 /* detach if offline */
2599 if (ata_link_offline(dev
->link
))
2600 ata_eh_detach_dev(dev
);
2602 /* schedule probe if necessary */
2603 if (ata_eh_schedule_probe(dev
))
2604 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
2608 ehc
->i
.action
|= ATA_EH_RESET
;
2614 * ata_eh_recover - recover host port after error
2615 * @ap: host port to recover
2616 * @prereset: prereset method (can be NULL)
2617 * @softreset: softreset method (can be NULL)
2618 * @hardreset: hardreset method (can be NULL)
2619 * @postreset: postreset method (can be NULL)
2620 * @r_failed_link: out parameter for failed link
2622 * This is the alpha and omega, eum and yang, heart and soul of
2623 * libata exception handling. On entry, actions required to
2624 * recover each link and hotplug requests are recorded in the
2625 * link's eh_context. This function executes all the operations
2626 * with appropriate retrials and fallbacks to resurrect failed
2627 * devices, detach goners and greet newcomers.
2630 * Kernel thread context (may sleep).
2633 * 0 on success, -errno on failure.
2635 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
2636 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
2637 ata_postreset_fn_t postreset
,
2638 struct ata_link
**r_failed_link
)
2640 struct ata_link
*link
;
2641 struct ata_device
*dev
;
2644 unsigned long flags
;
2648 /* prep for recovery */
2649 ata_port_for_each_link(link
, ap
) {
2650 struct ata_eh_context
*ehc
= &link
->eh_context
;
2652 /* re-enable link? */
2653 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
2654 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
2655 spin_lock_irqsave(ap
->lock
, flags
);
2656 link
->flags
&= ~ATA_LFLAG_DISABLED
;
2657 spin_unlock_irqrestore(ap
->lock
, flags
);
2658 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
2661 ata_link_for_each_dev(dev
, link
) {
2662 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
2663 ehc
->tries
[dev
->devno
] = 1;
2665 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
2667 /* collect port action mask recorded in dev actions */
2668 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
2669 ~ATA_EH_PERDEV_MASK
;
2670 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
2672 /* process hotplug request */
2673 if (dev
->flags
& ATA_DFLAG_DETACH
)
2674 ata_eh_detach_dev(dev
);
2676 /* schedule probe if necessary */
2677 if (!ata_dev_enabled(dev
))
2678 ata_eh_schedule_probe(dev
);
2686 /* if UNLOADING, finish immediately */
2687 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
2691 ata_port_for_each_link(link
, ap
) {
2692 struct ata_eh_context
*ehc
= &link
->eh_context
;
2694 /* skip EH if possible. */
2695 if (ata_eh_skip_recovery(link
))
2698 ata_link_for_each_dev(dev
, link
)
2699 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2703 ata_port_for_each_link(link
, ap
) {
2704 struct ata_eh_context
*ehc
= &link
->eh_context
;
2706 if (!(ehc
->i
.action
& ATA_EH_RESET
))
2709 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
2710 prereset
, softreset
, hardreset
, postreset
);
2712 ata_link_printk(link
, KERN_ERR
,
2713 "reset failed, giving up\n");
2719 ata_port_for_each_link(link
, ap
) {
2720 struct ata_eh_context
*ehc
= &link
->eh_context
;
2722 /* revalidate existing devices and attach new ones */
2723 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
2727 /* if PMP got attached, return, pmp EH will take care of it */
2728 if (link
->device
->class == ATA_DEV_PMP
) {
2733 /* configure transfer mode if necessary */
2734 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
2735 rc
= ata_set_mode(link
, &dev
);
2738 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
2741 if (ehc
->i
.action
& ATA_EH_LPM
)
2742 ata_link_for_each_dev(dev
, link
)
2743 ata_dev_enable_pm(dev
, ap
->pm_policy
);
2745 /* this link is okay now */
2751 ata_eh_handle_dev_fail(dev
, rc
);
2753 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
2754 /* PMP reset requires working host port.
2755 * Can't retry if it's frozen.
2757 if (sata_pmp_attached(ap
))
2767 if (rc
&& r_failed_link
)
2768 *r_failed_link
= link
;
2770 DPRINTK("EXIT, rc=%d\n", rc
);
2775 * ata_eh_finish - finish up EH
2776 * @ap: host port to finish EH for
2778 * Recovery is complete. Clean up EH states and retry or finish
2784 void ata_eh_finish(struct ata_port
*ap
)
2788 /* retry or finish qcs */
2789 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2790 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2792 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
2796 /* FIXME: Once EH migration is complete,
2797 * generate sense data in this function,
2798 * considering both err_mask and tf.
2800 if (qc
->flags
& ATA_QCFLAG_RETRY
)
2801 ata_eh_qc_retry(qc
);
2803 ata_eh_qc_complete(qc
);
2805 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
2806 ata_eh_qc_complete(qc
);
2808 /* feed zero TF to sense generation */
2809 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
2810 ata_eh_qc_retry(qc
);
2815 /* make sure nr_active_links is zero after EH */
2816 WARN_ON(ap
->nr_active_links
);
2817 ap
->nr_active_links
= 0;
2821 * ata_do_eh - do standard error handling
2822 * @ap: host port to handle error for
2824 * @prereset: prereset method (can be NULL)
2825 * @softreset: softreset method (can be NULL)
2826 * @hardreset: hardreset method (can be NULL)
2827 * @postreset: postreset method (can be NULL)
2829 * Perform standard error handling sequence.
2832 * Kernel thread context (may sleep).
2834 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
2835 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
2836 ata_postreset_fn_t postreset
)
2838 struct ata_device
*dev
;
2844 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
2847 ata_link_for_each_dev(dev
, &ap
->link
)
2848 ata_dev_disable(dev
);
2855 * ata_std_error_handler - standard error handler
2856 * @ap: host port to handle error for
2858 * Standard error handler
2861 * Kernel thread context (may sleep).
2863 void ata_std_error_handler(struct ata_port
*ap
)
2865 struct ata_port_operations
*ops
= ap
->ops
;
2866 ata_reset_fn_t hardreset
= ops
->hardreset
;
2868 /* ignore built-in hardreset if SCR access is not available */
2869 if (ata_is_builtin_hardreset(hardreset
) && !sata_scr_valid(&ap
->link
))
2872 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
2877 * ata_eh_handle_port_suspend - perform port suspend operation
2878 * @ap: port to suspend
2883 * Kernel thread context (may sleep).
2885 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
2887 unsigned long flags
;
2890 /* are we suspending? */
2891 spin_lock_irqsave(ap
->lock
, flags
);
2892 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
2893 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
2894 spin_unlock_irqrestore(ap
->lock
, flags
);
2897 spin_unlock_irqrestore(ap
->lock
, flags
);
2899 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
2901 /* tell ACPI we're suspending */
2902 rc
= ata_acpi_on_suspend(ap
);
2907 ata_eh_freeze_port(ap
);
2909 if (ap
->ops
->port_suspend
)
2910 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
2912 ata_acpi_set_state(ap
, PMSG_SUSPEND
);
2915 spin_lock_irqsave(ap
->lock
, flags
);
2917 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
2919 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
2920 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2921 ata_port_schedule_eh(ap
);
2923 if (ap
->pm_result
) {
2924 *ap
->pm_result
= rc
;
2925 ap
->pm_result
= NULL
;
2928 spin_unlock_irqrestore(ap
->lock
, flags
);
2934 * ata_eh_handle_port_resume - perform port resume operation
2935 * @ap: port to resume
2940 * Kernel thread context (may sleep).
2942 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
2944 unsigned long flags
;
2947 /* are we resuming? */
2948 spin_lock_irqsave(ap
->lock
, flags
);
2949 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
2950 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
2951 spin_unlock_irqrestore(ap
->lock
, flags
);
2954 spin_unlock_irqrestore(ap
->lock
, flags
);
2956 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
2958 ata_acpi_set_state(ap
, PMSG_ON
);
2960 if (ap
->ops
->port_resume
)
2961 rc
= ap
->ops
->port_resume(ap
);
2963 /* tell ACPI that we're resuming */
2964 ata_acpi_on_resume(ap
);
2967 spin_lock_irqsave(ap
->lock
, flags
);
2968 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
2969 if (ap
->pm_result
) {
2970 *ap
->pm_result
= rc
;
2971 ap
->pm_result
= NULL
;
2973 spin_unlock_irqrestore(ap
->lock
, flags
);
2975 #endif /* CONFIG_PM */