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 <scsi/scsi.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_eh.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_cmnd.h>
41 #include "../scsi/scsi_transport_api.h"
43 #include <linux/libata.h>
48 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
49 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
50 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
53 /* Waiting in ->prereset can never be reliable. It's sometimes nice
54 * to wait there but it can't be depended upon; otherwise, we wouldn't
55 * be resetting. Just give it enough time for most drives to spin up.
58 ATA_EH_PRERESET_TIMEOUT
= 10 * HZ
,
61 /* The following table determines how we sequence resets. Each entry
62 * represents timeout for that try. The first try can be soft or
63 * hardreset. All others are hardreset if available. In most cases
64 * the first reset w/ 10sec timeout should succeed. Following entries
65 * are mostly for error handling, hotplug and retarded devices.
67 static const unsigned long ata_eh_reset_timeouts
[] = {
68 10 * HZ
, /* most drives spin up by 10sec */
69 10 * HZ
, /* > 99% working drives spin up before 20sec */
70 35 * HZ
, /* give > 30 secs of idleness for retarded devices */
71 5 * HZ
, /* and sweet one last chance */
72 /* > 1 min has elapsed, give up */
75 static void __ata_port_freeze(struct ata_port
*ap
);
76 static void ata_eh_finish(struct ata_port
*ap
);
78 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
79 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
81 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
84 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
86 #endif /* CONFIG_PM */
88 static void ata_ering_record(struct ata_ering
*ering
, int is_io
,
89 unsigned int err_mask
)
91 struct ata_ering_entry
*ent
;
96 ering
->cursor
%= ATA_ERING_SIZE
;
98 ent
= &ering
->ring
[ering
->cursor
];
100 ent
->err_mask
= err_mask
;
101 ent
->timestamp
= get_jiffies_64();
104 static void ata_ering_clear(struct ata_ering
*ering
)
106 memset(ering
, 0, sizeof(*ering
));
109 static int ata_ering_map(struct ata_ering
*ering
,
110 int (*map_fn
)(struct ata_ering_entry
*, void *),
114 struct ata_ering_entry
*ent
;
118 ent
= &ering
->ring
[idx
];
121 rc
= map_fn(ent
, arg
);
124 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
125 } while (idx
!= ering
->cursor
);
130 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
132 struct ata_eh_context
*ehc
= &dev
->ap
->eh_context
;
134 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
137 static void ata_eh_clear_action(struct ata_device
*dev
,
138 struct ata_eh_info
*ehi
, unsigned int action
)
143 ehi
->action
&= ~action
;
144 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
145 ehi
->dev_action
[i
] &= ~action
;
147 /* doesn't make sense for port-wide EH actions */
148 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
150 /* break ehi->action into ehi->dev_action */
151 if (ehi
->action
& action
) {
152 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
153 ehi
->dev_action
[i
] |= ehi
->action
& action
;
154 ehi
->action
&= ~action
;
157 /* turn off the specified per-dev action */
158 ehi
->dev_action
[dev
->devno
] &= ~action
;
163 * ata_scsi_timed_out - SCSI layer time out callback
164 * @cmd: timed out SCSI command
166 * Handles SCSI layer timeout. We race with normal completion of
167 * the qc for @cmd. If the qc is already gone, we lose and let
168 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
169 * timed out and EH should be invoked. Prevent ata_qc_complete()
170 * from finishing it by setting EH_SCHEDULED and return
173 * TODO: kill this function once old EH is gone.
176 * Called from timer context
179 * EH_HANDLED or EH_NOT_HANDLED
181 enum scsi_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
183 struct Scsi_Host
*host
= cmd
->device
->host
;
184 struct ata_port
*ap
= ata_shost_to_port(host
);
186 struct ata_queued_cmd
*qc
;
187 enum scsi_eh_timer_return ret
;
191 if (ap
->ops
->error_handler
) {
192 ret
= EH_NOT_HANDLED
;
197 spin_lock_irqsave(ap
->lock
, flags
);
198 qc
= ata_qc_from_tag(ap
, ap
->active_tag
);
200 WARN_ON(qc
->scsicmd
!= cmd
);
201 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
202 qc
->err_mask
|= AC_ERR_TIMEOUT
;
203 ret
= EH_NOT_HANDLED
;
205 spin_unlock_irqrestore(ap
->lock
, flags
);
208 DPRINTK("EXIT, ret=%d\n", ret
);
213 * ata_scsi_error - SCSI layer error handler callback
214 * @host: SCSI host on which error occurred
216 * Handles SCSI-layer-thrown error events.
219 * Inherited from SCSI layer (none, can sleep)
224 void ata_scsi_error(struct Scsi_Host
*host
)
226 struct ata_port
*ap
= ata_shost_to_port(host
);
227 int i
, repeat_cnt
= ATA_EH_MAX_REPEAT
;
232 /* synchronize with port task */
233 ata_port_flush_task(ap
);
235 /* synchronize with host lock and sort out timeouts */
237 /* For new EH, all qcs are finished in one of three ways -
238 * normal completion, error completion, and SCSI timeout.
239 * Both cmpletions can race against SCSI timeout. When normal
240 * completion wins, the qc never reaches EH. When error
241 * completion wins, the qc has ATA_QCFLAG_FAILED set.
243 * When SCSI timeout wins, things are a bit more complex.
244 * Normal or error completion can occur after the timeout but
245 * before this point. In such cases, both types of
246 * completions are honored. A scmd is determined to have
247 * timed out iff its associated qc is active and not failed.
249 if (ap
->ops
->error_handler
) {
250 struct scsi_cmnd
*scmd
, *tmp
;
253 spin_lock_irqsave(ap
->lock
, flags
);
255 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
256 struct ata_queued_cmd
*qc
;
258 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
259 qc
= __ata_qc_from_tag(ap
, i
);
260 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
265 if (i
< ATA_MAX_QUEUE
) {
266 /* the scmd has an associated qc */
267 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
268 /* which hasn't failed yet, timeout */
269 qc
->err_mask
|= AC_ERR_TIMEOUT
;
270 qc
->flags
|= ATA_QCFLAG_FAILED
;
274 /* Normal completion occurred after
275 * SCSI timeout but before this point.
276 * Successfully complete it.
278 scmd
->retries
= scmd
->allowed
;
279 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
283 /* If we have timed out qcs. They belong to EH from
284 * this point but the state of the controller is
285 * unknown. Freeze the port to make sure the IRQ
286 * handler doesn't diddle with those qcs. This must
287 * be done atomically w.r.t. setting QCFLAG_FAILED.
290 __ata_port_freeze(ap
);
292 spin_unlock_irqrestore(ap
->lock
, flags
);
294 spin_unlock_wait(ap
->lock
);
297 /* invoke error handler */
298 if (ap
->ops
->error_handler
) {
299 /* process port resume request */
300 ata_eh_handle_port_resume(ap
);
302 /* fetch & clear EH info */
303 spin_lock_irqsave(ap
->lock
, flags
);
305 memset(&ap
->eh_context
, 0, sizeof(ap
->eh_context
));
306 ap
->eh_context
.i
= ap
->eh_info
;
307 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
309 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
310 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
312 spin_unlock_irqrestore(ap
->lock
, flags
);
314 /* invoke EH, skip if unloading or suspended */
315 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
316 ap
->ops
->error_handler(ap
);
320 /* process port suspend request */
321 ata_eh_handle_port_suspend(ap
);
323 /* Exception might have happend after ->error_handler
324 * recovered the port but before this point. Repeat
327 spin_lock_irqsave(ap
->lock
, flags
);
329 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
331 ata_port_printk(ap
, KERN_INFO
,
332 "EH pending after completion, "
333 "repeating EH (cnt=%d)\n", repeat_cnt
);
334 spin_unlock_irqrestore(ap
->lock
, flags
);
337 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
338 "tries, giving up\n", ATA_EH_MAX_REPEAT
);
339 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
342 /* this run is complete, make sure EH info is clear */
343 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
345 /* Clear host_eh_scheduled while holding ap->lock such
346 * that if exception occurs after this point but
347 * before EH completion, SCSI midlayer will
350 host
->host_eh_scheduled
= 0;
352 spin_unlock_irqrestore(ap
->lock
, flags
);
354 WARN_ON(ata_qc_from_tag(ap
, ap
->active_tag
) == NULL
);
355 ap
->ops
->eng_timeout(ap
);
358 /* finish or retry handled scmd's and clean up */
359 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
361 scsi_eh_flush_done_q(&ap
->eh_done_q
);
364 spin_lock_irqsave(ap
->lock
, flags
);
366 if (ap
->pflags
& ATA_PFLAG_LOADING
)
367 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
368 else if (ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
)
369 queue_delayed_work(ata_aux_wq
, &ap
->hotplug_task
, 0);
371 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
372 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
374 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
376 /* tell wait_eh that we're done */
377 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
378 wake_up_all(&ap
->eh_wait_q
);
380 spin_unlock_irqrestore(ap
->lock
, flags
);
386 * ata_port_wait_eh - Wait for the currently pending EH to complete
387 * @ap: Port to wait EH for
389 * Wait until the currently pending EH is complete.
392 * Kernel thread context (may sleep).
394 void ata_port_wait_eh(struct ata_port
*ap
)
400 spin_lock_irqsave(ap
->lock
, flags
);
402 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
403 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
404 spin_unlock_irqrestore(ap
->lock
, flags
);
406 spin_lock_irqsave(ap
->lock
, flags
);
408 finish_wait(&ap
->eh_wait_q
, &wait
);
410 spin_unlock_irqrestore(ap
->lock
, flags
);
412 /* make sure SCSI EH is complete */
413 if (scsi_host_in_recovery(ap
->scsi_host
)) {
420 * ata_qc_timeout - Handle timeout of queued command
421 * @qc: Command that timed out
423 * Some part of the kernel (currently, only the SCSI layer)
424 * has noticed that the active command on port @ap has not
425 * completed after a specified length of time. Handle this
426 * condition by disabling DMA (if necessary) and completing
427 * transactions, with error if necessary.
429 * This also handles the case of the "lost interrupt", where
430 * for some reason (possibly hardware bug, possibly driver bug)
431 * an interrupt was not delivered to the driver, even though the
432 * transaction completed successfully.
434 * TODO: kill this function once old EH is gone.
437 * Inherited from SCSI layer (none, can sleep)
439 static void ata_qc_timeout(struct ata_queued_cmd
*qc
)
441 struct ata_port
*ap
= qc
->ap
;
442 u8 host_stat
= 0, drv_stat
;
447 ap
->hsm_task_state
= HSM_ST_IDLE
;
449 spin_lock_irqsave(ap
->lock
, flags
);
451 switch (qc
->tf
.protocol
) {
454 case ATA_PROT_ATAPI_DMA
:
455 host_stat
= ap
->ops
->bmdma_status(ap
);
457 /* before we do anything else, clear DMA-Start bit */
458 ap
->ops
->bmdma_stop(qc
);
464 drv_stat
= ata_chk_status(ap
);
466 /* ack bmdma irq events */
467 ap
->ops
->irq_clear(ap
);
469 ata_dev_printk(qc
->dev
, KERN_ERR
, "command 0x%x timeout, "
470 "stat 0x%x host_stat 0x%x\n",
471 qc
->tf
.command
, drv_stat
, host_stat
);
473 /* complete taskfile transaction */
474 qc
->err_mask
|= AC_ERR_TIMEOUT
;
478 spin_unlock_irqrestore(ap
->lock
, flags
);
480 ata_eh_qc_complete(qc
);
486 * ata_eng_timeout - Handle timeout of queued command
487 * @ap: Port on which timed-out command is active
489 * Some part of the kernel (currently, only the SCSI layer)
490 * has noticed that the active command on port @ap has not
491 * completed after a specified length of time. Handle this
492 * condition by disabling DMA (if necessary) and completing
493 * transactions, with error if necessary.
495 * This also handles the case of the "lost interrupt", where
496 * for some reason (possibly hardware bug, possibly driver bug)
497 * an interrupt was not delivered to the driver, even though the
498 * transaction completed successfully.
500 * TODO: kill this function once old EH is gone.
503 * Inherited from SCSI layer (none, can sleep)
505 void ata_eng_timeout(struct ata_port
*ap
)
509 ata_qc_timeout(ata_qc_from_tag(ap
, ap
->active_tag
));
515 * ata_qc_schedule_eh - schedule qc for error handling
516 * @qc: command to schedule error handling for
518 * Schedule error handling for @qc. EH will kick in as soon as
519 * other commands are drained.
522 * spin_lock_irqsave(host lock)
524 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
526 struct ata_port
*ap
= qc
->ap
;
528 WARN_ON(!ap
->ops
->error_handler
);
530 qc
->flags
|= ATA_QCFLAG_FAILED
;
531 qc
->ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
533 /* The following will fail if timeout has already expired.
534 * ata_scsi_error() takes care of such scmds on EH entry.
535 * Note that ATA_QCFLAG_FAILED is unconditionally set after
536 * this function completes.
538 scsi_req_abort_cmd(qc
->scsicmd
);
542 * ata_port_schedule_eh - schedule error handling without a qc
543 * @ap: ATA port to schedule EH for
545 * Schedule error handling for @ap. EH will kick in as soon as
546 * all commands are drained.
549 * spin_lock_irqsave(host lock)
551 void ata_port_schedule_eh(struct ata_port
*ap
)
553 WARN_ON(!ap
->ops
->error_handler
);
555 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
558 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
559 scsi_schedule_eh(ap
->scsi_host
);
561 DPRINTK("port EH scheduled\n");
565 * ata_port_abort - abort all qc's on the port
566 * @ap: ATA port to abort qc's for
568 * Abort all active qc's of @ap and schedule EH.
571 * spin_lock_irqsave(host lock)
574 * Number of aborted qc's.
576 int ata_port_abort(struct ata_port
*ap
)
578 int tag
, nr_aborted
= 0;
580 WARN_ON(!ap
->ops
->error_handler
);
582 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
583 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
586 qc
->flags
|= ATA_QCFLAG_FAILED
;
593 ata_port_schedule_eh(ap
);
599 * __ata_port_freeze - freeze port
600 * @ap: ATA port to freeze
602 * This function is called when HSM violation or some other
603 * condition disrupts normal operation of the port. Frozen port
604 * is not allowed to perform any operation until the port is
605 * thawed, which usually follows a successful reset.
607 * ap->ops->freeze() callback can be used for freezing the port
608 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
609 * port cannot be frozen hardware-wise, the interrupt handler
610 * must ack and clear interrupts unconditionally while the port
614 * spin_lock_irqsave(host lock)
616 static void __ata_port_freeze(struct ata_port
*ap
)
618 WARN_ON(!ap
->ops
->error_handler
);
623 ap
->pflags
|= ATA_PFLAG_FROZEN
;
625 DPRINTK("ata%u port frozen\n", ap
->print_id
);
629 * ata_port_freeze - abort & freeze port
630 * @ap: ATA port to freeze
632 * Abort and freeze @ap.
635 * spin_lock_irqsave(host lock)
638 * Number of aborted commands.
640 int ata_port_freeze(struct ata_port
*ap
)
644 WARN_ON(!ap
->ops
->error_handler
);
646 nr_aborted
= ata_port_abort(ap
);
647 __ata_port_freeze(ap
);
653 * ata_eh_freeze_port - EH helper to freeze port
654 * @ap: ATA port to freeze
661 void ata_eh_freeze_port(struct ata_port
*ap
)
665 if (!ap
->ops
->error_handler
)
668 spin_lock_irqsave(ap
->lock
, flags
);
669 __ata_port_freeze(ap
);
670 spin_unlock_irqrestore(ap
->lock
, flags
);
674 * ata_port_thaw_port - EH helper to thaw port
675 * @ap: ATA port to thaw
677 * Thaw frozen port @ap.
682 void ata_eh_thaw_port(struct ata_port
*ap
)
686 if (!ap
->ops
->error_handler
)
689 spin_lock_irqsave(ap
->lock
, flags
);
691 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
696 spin_unlock_irqrestore(ap
->lock
, flags
);
698 DPRINTK("ata%u port thawed\n", ap
->print_id
);
701 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
706 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
708 struct ata_port
*ap
= qc
->ap
;
709 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
712 spin_lock_irqsave(ap
->lock
, flags
);
713 qc
->scsidone
= ata_eh_scsidone
;
714 __ata_qc_complete(qc
);
715 WARN_ON(ata_tag_valid(qc
->tag
));
716 spin_unlock_irqrestore(ap
->lock
, flags
);
718 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
722 * ata_eh_qc_complete - Complete an active ATA command from EH
723 * @qc: Command to complete
725 * Indicate to the mid and upper layers that an ATA command has
726 * completed. To be used from EH.
728 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
730 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
731 scmd
->retries
= scmd
->allowed
;
732 __ata_eh_qc_complete(qc
);
736 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
737 * @qc: Command to retry
739 * Indicate to the mid and upper layers that an ATA command
740 * should be retried. To be used from EH.
742 * SCSI midlayer limits the number of retries to scmd->allowed.
743 * scmd->retries is decremented for commands which get retried
744 * due to unrelated failures (qc->err_mask is zero).
746 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
748 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
749 if (!qc
->err_mask
&& scmd
->retries
)
751 __ata_eh_qc_complete(qc
);
755 * ata_eh_detach_dev - detach ATA device
756 * @dev: ATA device to detach
763 static void ata_eh_detach_dev(struct ata_device
*dev
)
765 struct ata_port
*ap
= dev
->ap
;
768 ata_dev_disable(dev
);
770 spin_lock_irqsave(ap
->lock
, flags
);
772 dev
->flags
&= ~ATA_DFLAG_DETACH
;
774 if (ata_scsi_offline_dev(dev
)) {
775 dev
->flags
|= ATA_DFLAG_DETACHED
;
776 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
779 /* clear per-dev EH actions */
780 ata_eh_clear_action(dev
, &ap
->eh_info
, ATA_EH_PERDEV_MASK
);
781 ata_eh_clear_action(dev
, &ap
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
783 spin_unlock_irqrestore(ap
->lock
, flags
);
787 * ata_eh_about_to_do - about to perform eh_action
788 * @ap: target ATA port
789 * @dev: target ATA dev for per-dev action (can be NULL)
790 * @action: action about to be performed
792 * Called just before performing EH actions to clear related bits
793 * in @ap->eh_info such that eh actions are not unnecessarily
799 static void ata_eh_about_to_do(struct ata_port
*ap
, struct ata_device
*dev
,
803 struct ata_eh_info
*ehi
= &ap
->eh_info
;
804 struct ata_eh_context
*ehc
= &ap
->eh_context
;
806 spin_lock_irqsave(ap
->lock
, flags
);
808 /* Reset is represented by combination of actions and EHI
809 * flags. Suck in all related bits before clearing eh_info to
810 * avoid losing requested action.
812 if (action
& ATA_EH_RESET_MASK
) {
813 ehc
->i
.action
|= ehi
->action
& ATA_EH_RESET_MASK
;
814 ehc
->i
.flags
|= ehi
->flags
& ATA_EHI_RESET_MODIFIER_MASK
;
816 /* make sure all reset actions are cleared & clear EHI flags */
817 action
|= ATA_EH_RESET_MASK
;
818 ehi
->flags
&= ~ATA_EHI_RESET_MODIFIER_MASK
;
821 ata_eh_clear_action(dev
, ehi
, action
);
823 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
))
824 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
826 spin_unlock_irqrestore(ap
->lock
, flags
);
830 * ata_eh_done - EH action complete
831 * @ap: target ATA port
832 * @dev: target ATA dev for per-dev action (can be NULL)
833 * @action: action just completed
835 * Called right after performing EH actions to clear related bits
836 * in @ap->eh_context.
841 static void ata_eh_done(struct ata_port
*ap
, struct ata_device
*dev
,
844 /* if reset is complete, clear all reset actions & reset modifier */
845 if (action
& ATA_EH_RESET_MASK
) {
846 action
|= ATA_EH_RESET_MASK
;
847 ap
->eh_context
.i
.flags
&= ~ATA_EHI_RESET_MODIFIER_MASK
;
850 ata_eh_clear_action(dev
, &ap
->eh_context
.i
, action
);
854 * ata_err_string - convert err_mask to descriptive string
855 * @err_mask: error mask to convert to string
857 * Convert @err_mask to descriptive string. Errors are
858 * prioritized according to severity and only the most severe
865 * Descriptive string for @err_mask
867 static const char * ata_err_string(unsigned int err_mask
)
869 if (err_mask
& AC_ERR_HOST_BUS
)
870 return "host bus error";
871 if (err_mask
& AC_ERR_ATA_BUS
)
872 return "ATA bus error";
873 if (err_mask
& AC_ERR_TIMEOUT
)
875 if (err_mask
& AC_ERR_HSM
)
876 return "HSM violation";
877 if (err_mask
& AC_ERR_SYSTEM
)
878 return "internal error";
879 if (err_mask
& AC_ERR_MEDIA
)
880 return "media error";
881 if (err_mask
& AC_ERR_INVALID
)
882 return "invalid argument";
883 if (err_mask
& AC_ERR_DEV
)
884 return "device error";
885 return "unknown error";
889 * ata_read_log_page - read a specific log page
890 * @dev: target device
891 * @page: page to read
892 * @buf: buffer to store read page
893 * @sectors: number of sectors to read
895 * Read log page using READ_LOG_EXT command.
898 * Kernel thread context (may sleep).
901 * 0 on success, AC_ERR_* mask otherwise.
903 static unsigned int ata_read_log_page(struct ata_device
*dev
,
904 u8 page
, void *buf
, unsigned int sectors
)
906 struct ata_taskfile tf
;
907 unsigned int err_mask
;
909 DPRINTK("read log page - page %d\n", page
);
911 ata_tf_init(dev
, &tf
);
912 tf
.command
= ATA_CMD_READ_LOG_EXT
;
915 tf
.hob_nsect
= sectors
>> 8;
916 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
917 tf
.protocol
= ATA_PROT_PIO
;
919 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
920 buf
, sectors
* ATA_SECT_SIZE
);
922 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
927 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
928 * @dev: Device to read log page 10h from
929 * @tag: Resulting tag of the failed command
930 * @tf: Resulting taskfile registers of the failed command
932 * Read log page 10h to obtain NCQ error details and clear error
936 * Kernel thread context (may sleep).
939 * 0 on success, -errno otherwise.
941 static int ata_eh_read_log_10h(struct ata_device
*dev
,
942 int *tag
, struct ata_taskfile
*tf
)
944 u8
*buf
= dev
->ap
->sector_buf
;
945 unsigned int err_mask
;
949 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
954 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
957 ata_dev_printk(dev
, KERN_WARNING
,
958 "invalid checksum 0x%x on log page 10h\n", csum
);
963 *tag
= buf
[0] & 0x1f;
965 tf
->command
= buf
[2];
966 tf
->feature
= buf
[3];
971 tf
->hob_lbal
= buf
[8];
972 tf
->hob_lbam
= buf
[9];
973 tf
->hob_lbah
= buf
[10];
975 tf
->hob_nsect
= buf
[13];
981 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
982 * @dev: device to perform REQUEST_SENSE to
983 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
985 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
986 * SENSE. This function is EH helper.
989 * Kernel thread context (may sleep).
992 * 0 on success, AC_ERR_* mask on failure
994 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd
*qc
)
996 struct ata_device
*dev
= qc
->dev
;
997 unsigned char *sense_buf
= qc
->scsicmd
->sense_buffer
;
998 struct ata_port
*ap
= dev
->ap
;
999 struct ata_taskfile tf
;
1000 u8 cdb
[ATAPI_CDB_LEN
];
1002 DPRINTK("ATAPI request sense\n");
1004 /* FIXME: is this needed? */
1005 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1007 /* initialize sense_buf with the error register,
1008 * for the case where they are -not- overwritten
1010 sense_buf
[0] = 0x70;
1011 sense_buf
[2] = qc
->result_tf
.feature
>> 4;
1013 /* some devices time out if garbage left in tf */
1014 ata_tf_init(dev
, &tf
);
1016 memset(cdb
, 0, ATAPI_CDB_LEN
);
1017 cdb
[0] = REQUEST_SENSE
;
1018 cdb
[4] = SCSI_SENSE_BUFFERSIZE
;
1020 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1021 tf
.command
= ATA_CMD_PACKET
;
1023 /* is it pointless to prefer PIO for "safety reasons"? */
1024 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1025 tf
.protocol
= ATA_PROT_ATAPI_DMA
;
1026 tf
.feature
|= ATAPI_PKT_DMA
;
1028 tf
.protocol
= ATA_PROT_ATAPI
;
1029 tf
.lbam
= (8 * 1024) & 0xff;
1030 tf
.lbah
= (8 * 1024) >> 8;
1033 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1034 sense_buf
, SCSI_SENSE_BUFFERSIZE
);
1038 * ata_eh_analyze_serror - analyze SError for a failed port
1039 * @ap: ATA port to analyze SError for
1041 * Analyze SError if available and further determine cause of
1047 static void ata_eh_analyze_serror(struct ata_port
*ap
)
1049 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1050 u32 serror
= ehc
->i
.serror
;
1051 unsigned int err_mask
= 0, action
= 0;
1053 if (serror
& SERR_PERSISTENT
) {
1054 err_mask
|= AC_ERR_ATA_BUS
;
1055 action
|= ATA_EH_HARDRESET
;
1058 (SERR_DATA_RECOVERED
| SERR_COMM_RECOVERED
| SERR_DATA
)) {
1059 err_mask
|= AC_ERR_ATA_BUS
;
1060 action
|= ATA_EH_SOFTRESET
;
1062 if (serror
& SERR_PROTOCOL
) {
1063 err_mask
|= AC_ERR_HSM
;
1064 action
|= ATA_EH_SOFTRESET
;
1066 if (serror
& SERR_INTERNAL
) {
1067 err_mask
|= AC_ERR_SYSTEM
;
1068 action
|= ATA_EH_HARDRESET
;
1070 if (serror
& (SERR_PHYRDY_CHG
| SERR_DEV_XCHG
))
1071 ata_ehi_hotplugged(&ehc
->i
);
1073 ehc
->i
.err_mask
|= err_mask
;
1074 ehc
->i
.action
|= action
;
1078 * ata_eh_analyze_ncq_error - analyze NCQ error
1079 * @ap: ATA port to analyze NCQ error for
1081 * Read log page 10h, determine the offending qc and acquire
1082 * error status TF. For NCQ device errors, all LLDDs have to do
1083 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1087 * Kernel thread context (may sleep).
1089 static void ata_eh_analyze_ncq_error(struct ata_port
*ap
)
1091 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1092 struct ata_device
*dev
= ap
->device
;
1093 struct ata_queued_cmd
*qc
;
1094 struct ata_taskfile tf
;
1097 /* if frozen, we can't do much */
1098 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1101 /* is it NCQ device error? */
1102 if (!ap
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1105 /* has LLDD analyzed already? */
1106 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1107 qc
= __ata_qc_from_tag(ap
, tag
);
1109 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1116 /* okay, this error is ours */
1117 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1119 ata_port_printk(ap
, KERN_ERR
, "failed to read log page 10h "
1120 "(errno=%d)\n", rc
);
1124 if (!(ap
->sactive
& (1 << tag
))) {
1125 ata_port_printk(ap
, KERN_ERR
, "log page 10h reported "
1126 "inactive tag %d\n", tag
);
1130 /* we've got the perpetrator, condemn it */
1131 qc
= __ata_qc_from_tag(ap
, tag
);
1132 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1133 qc
->err_mask
|= AC_ERR_DEV
;
1134 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1138 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1139 * @qc: qc to analyze
1140 * @tf: Taskfile registers to analyze
1142 * Analyze taskfile of @qc and further determine cause of
1143 * failure. This function also requests ATAPI sense data if
1147 * Kernel thread context (may sleep).
1150 * Determined recovery action
1152 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1153 const struct ata_taskfile
*tf
)
1155 unsigned int tmp
, action
= 0;
1156 u8 stat
= tf
->command
, err
= tf
->feature
;
1158 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1159 qc
->err_mask
|= AC_ERR_HSM
;
1160 return ATA_EH_SOFTRESET
;
1163 if (stat
& (ATA_ERR
| ATA_DF
))
1164 qc
->err_mask
|= AC_ERR_DEV
;
1168 switch (qc
->dev
->class) {
1171 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1173 qc
->err_mask
|= AC_ERR_MEDIA
;
1175 qc
->err_mask
|= AC_ERR_INVALID
;
1179 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1180 tmp
= atapi_eh_request_sense(qc
);
1182 /* ATA_QCFLAG_SENSE_VALID is used to
1183 * tell atapi_qc_complete() that sense
1184 * data is already valid.
1186 * TODO: interpret sense data and set
1187 * appropriate err_mask.
1189 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1191 qc
->err_mask
|= tmp
;
1195 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1196 action
|= ATA_EH_SOFTRESET
;
1201 static int ata_eh_categorize_error(int is_io
, unsigned int err_mask
)
1203 if (err_mask
& AC_ERR_ATA_BUS
)
1206 if (err_mask
& AC_ERR_TIMEOUT
)
1210 if (err_mask
& AC_ERR_HSM
)
1213 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1220 struct speed_down_verdict_arg
{
1225 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1227 struct speed_down_verdict_arg
*arg
= void_arg
;
1228 int cat
= ata_eh_categorize_error(ent
->is_io
, ent
->err_mask
);
1230 if (ent
->timestamp
< arg
->since
)
1233 arg
->nr_errors
[cat
]++;
1238 * ata_eh_speed_down_verdict - Determine speed down verdict
1239 * @dev: Device of interest
1241 * This function examines error ring of @dev and determines
1242 * whether NCQ needs to be turned off, transfer speed should be
1243 * stepped down, or falling back to PIO is necessary.
1245 * Cat-1 is ATA_BUS error for any command.
1247 * Cat-2 is TIMEOUT for any command or HSM violation for known
1248 * supported commands.
1250 * Cat-3 is is unclassified DEV error for known supported
1253 * NCQ needs to be turned off if there have been more than 3
1254 * Cat-2 + Cat-3 errors during last 10 minutes.
1256 * Speed down is necessary if there have been more than 3 Cat-1 +
1257 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1259 * Falling back to PIO mode is necessary if there have been more
1260 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1263 * Inherited from caller.
1266 * OR of ATA_EH_SPDN_* flags.
1268 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1270 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1271 u64 j64
= get_jiffies_64();
1272 struct speed_down_verdict_arg arg
;
1273 unsigned int verdict
= 0;
1275 /* scan past 10 mins of error history */
1276 memset(&arg
, 0, sizeof(arg
));
1277 arg
.since
= j64
- min(j64
, j10mins
);
1278 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1280 if (arg
.nr_errors
[2] + arg
.nr_errors
[3] > 3)
1281 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1282 if (arg
.nr_errors
[1] + arg
.nr_errors
[2] > 3 || arg
.nr_errors
[3] > 10)
1283 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1285 /* scan past 3 mins of error history */
1286 memset(&arg
, 0, sizeof(arg
));
1287 arg
.since
= j64
- min(j64
, j5mins
);
1288 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1290 if (arg
.nr_errors
[1] + arg
.nr_errors
[2] + arg
.nr_errors
[3] > 10)
1291 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1297 * ata_eh_speed_down - record error and speed down if necessary
1298 * @dev: Failed device
1299 * @is_io: Did the device fail during normal IO?
1300 * @err_mask: err_mask of the error
1302 * Record error and examine error history to determine whether
1303 * adjusting transmission speed is necessary. It also sets
1304 * transmission limits appropriately if such adjustment is
1308 * Kernel thread context (may sleep).
1311 * Determined recovery action.
1313 static unsigned int ata_eh_speed_down(struct ata_device
*dev
, int is_io
,
1314 unsigned int err_mask
)
1316 unsigned int verdict
;
1317 unsigned int action
= 0;
1319 /* don't bother if Cat-0 error */
1320 if (ata_eh_categorize_error(is_io
, err_mask
) == 0)
1323 /* record error and determine whether speed down is necessary */
1324 ata_ering_record(&dev
->ering
, is_io
, err_mask
);
1325 verdict
= ata_eh_speed_down_verdict(dev
);
1328 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
1329 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
1330 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
1331 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
1332 ata_dev_printk(dev
, KERN_WARNING
,
1333 "NCQ disabled due to excessive errors\n");
1338 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
1339 /* speed down SATA link speed if possible */
1340 if (sata_down_spd_limit(dev
->ap
) == 0) {
1341 action
|= ATA_EH_HARDRESET
;
1345 /* lower transfer mode */
1346 if (dev
->spdn_cnt
< 2) {
1347 static const int dma_dnxfer_sel
[] =
1348 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
1349 static const int pio_dnxfer_sel
[] =
1350 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
1353 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
1354 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
1356 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
1360 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
1361 action
|= ATA_EH_SOFTRESET
;
1367 /* Fall back to PIO? Slowing down to PIO is meaningless for
1368 * SATA. Consider it only for PATA.
1370 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
1371 (dev
->ap
->cbl
!= ATA_CBL_SATA
) &&
1372 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
1373 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
1375 action
|= ATA_EH_SOFTRESET
;
1382 /* device has been slowed down, blow error history */
1383 ata_ering_clear(&dev
->ering
);
1388 * ata_eh_autopsy - analyze error and determine recovery action
1389 * @ap: ATA port to perform autopsy on
1391 * Analyze why @ap failed and determine which recovery action is
1392 * needed. This function also sets more detailed AC_ERR_* values
1393 * and fills sense data for ATAPI CHECK SENSE.
1396 * Kernel thread context (may sleep).
1398 static void ata_eh_autopsy(struct ata_port
*ap
)
1400 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1401 unsigned int all_err_mask
= 0;
1408 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
1411 /* obtain and analyze SError */
1412 rc
= sata_scr_read(ap
, SCR_ERROR
, &serror
);
1414 ehc
->i
.serror
|= serror
;
1415 ata_eh_analyze_serror(ap
);
1416 } else if (rc
!= -EOPNOTSUPP
)
1417 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1419 /* analyze NCQ failure */
1420 ata_eh_analyze_ncq_error(ap
);
1422 /* any real error trumps AC_ERR_OTHER */
1423 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1424 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1426 all_err_mask
|= ehc
->i
.err_mask
;
1428 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1429 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1431 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1434 /* inherit upper level err_mask */
1435 qc
->err_mask
|= ehc
->i
.err_mask
;
1438 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1440 /* DEV errors are probably spurious in case of ATA_BUS error */
1441 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1442 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1445 /* any real error trumps unknown error */
1446 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1447 qc
->err_mask
&= ~AC_ERR_OTHER
;
1449 /* SENSE_VALID trumps dev/unknown error and revalidation */
1450 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1451 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1452 ehc
->i
.action
&= ~ATA_EH_REVALIDATE
;
1455 /* accumulate error info */
1456 ehc
->i
.dev
= qc
->dev
;
1457 all_err_mask
|= qc
->err_mask
;
1458 if (qc
->flags
& ATA_QCFLAG_IO
)
1462 /* enforce default EH actions */
1463 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
1464 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
1465 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1466 else if (all_err_mask
)
1467 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1469 /* if we have offending qcs and the associated failed device */
1472 ehc
->i
.action
|= ata_eh_speed_down(ehc
->i
.dev
, is_io
,
1475 /* perform per-dev EH action only on the offending device */
1476 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
1477 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
1478 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
1485 * ata_eh_report - report error handling to user
1486 * @ap: ATA port EH is going on
1488 * Report EH to user.
1493 static void ata_eh_report(struct ata_port
*ap
)
1495 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1496 const char *frozen
, *desc
;
1497 int tag
, nr_failed
= 0;
1500 if (ehc
->i
.desc
[0] != '\0')
1503 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1504 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1506 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1508 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
1514 if (!nr_failed
&& !ehc
->i
.err_mask
)
1518 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1522 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
1523 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1524 ehc
->i
.err_mask
, ap
->sactive
, ehc
->i
.serror
,
1525 ehc
->i
.action
, frozen
);
1527 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "(%s)\n", desc
);
1529 ata_port_printk(ap
, KERN_ERR
, "exception Emask 0x%x "
1530 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1531 ehc
->i
.err_mask
, ap
->sactive
, ehc
->i
.serror
,
1532 ehc
->i
.action
, frozen
);
1534 ata_port_printk(ap
, KERN_ERR
, "(%s)\n", desc
);
1537 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1538 static const char *dma_str
[] = {
1539 [DMA_BIDIRECTIONAL
] = "bidi",
1540 [DMA_TO_DEVICE
] = "out",
1541 [DMA_FROM_DEVICE
] = "in",
1544 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1545 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
1547 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || !qc
->err_mask
)
1550 ata_dev_printk(qc
->dev
, KERN_ERR
,
1551 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1552 "tag %d cdb 0x%x data %u %s\n "
1553 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1554 "Emask 0x%x (%s)\n",
1555 cmd
->command
, cmd
->feature
, cmd
->nsect
,
1556 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
1557 cmd
->hob_feature
, cmd
->hob_nsect
,
1558 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
1559 cmd
->device
, qc
->tag
, qc
->cdb
[0], qc
->nbytes
,
1560 dma_str
[qc
->dma_dir
],
1561 res
->command
, res
->feature
, res
->nsect
,
1562 res
->lbal
, res
->lbam
, res
->lbah
,
1563 res
->hob_feature
, res
->hob_nsect
,
1564 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
1565 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
));
1569 static int ata_do_reset(struct ata_port
*ap
, ata_reset_fn_t reset
,
1570 unsigned int *classes
, unsigned long deadline
)
1574 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1575 classes
[i
] = ATA_DEV_UNKNOWN
;
1577 rc
= reset(ap
, classes
, deadline
);
1581 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1582 * is complete and convert all ATA_DEV_UNKNOWN to
1585 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1586 if (classes
[i
] != ATA_DEV_UNKNOWN
)
1589 if (i
< ATA_MAX_DEVICES
)
1590 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1591 if (classes
[i
] == ATA_DEV_UNKNOWN
)
1592 classes
[i
] = ATA_DEV_NONE
;
1597 static int ata_eh_followup_srst_needed(int rc
, int classify
,
1598 const unsigned int *classes
)
1604 if (classify
&& classes
[0] == ATA_DEV_UNKNOWN
)
1609 static int ata_eh_reset(struct ata_port
*ap
, int classify
,
1610 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
1611 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
1613 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1614 unsigned int *classes
= ehc
->classes
;
1615 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
1617 unsigned long deadline
;
1618 unsigned int action
;
1619 ata_reset_fn_t reset
;
1622 /* about to reset */
1623 ata_eh_about_to_do(ap
, NULL
, ehc
->i
.action
& ATA_EH_RESET_MASK
);
1625 /* Determine which reset to use and record in ehc->i.action.
1626 * prereset() may examine and modify it.
1628 action
= ehc
->i
.action
;
1629 ehc
->i
.action
&= ~ATA_EH_RESET_MASK
;
1630 if (softreset
&& (!hardreset
|| (!sata_set_spd_needed(ap
) &&
1631 !(action
& ATA_EH_HARDRESET
))))
1632 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1634 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1637 rc
= prereset(ap
, jiffies
+ ATA_EH_PRERESET_TIMEOUT
);
1639 if (rc
== -ENOENT
) {
1640 ata_port_printk(ap
, KERN_DEBUG
,
1641 "port disabled. ignoring.\n");
1642 ap
->eh_context
.i
.action
&= ~ATA_EH_RESET_MASK
;
1644 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1645 classes
[i
] = ATA_DEV_NONE
;
1649 ata_port_printk(ap
, KERN_ERR
,
1650 "prereset failed (errno=%d)\n", rc
);
1655 /* prereset() might have modified ehc->i.action */
1656 if (ehc
->i
.action
& ATA_EH_HARDRESET
)
1658 else if (ehc
->i
.action
& ATA_EH_SOFTRESET
)
1661 /* prereset told us not to reset, bang classes and return */
1662 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1663 classes
[i
] = ATA_DEV_NONE
;
1667 /* did prereset() screw up? if so, fix up to avoid oopsing */
1676 deadline
= jiffies
+ ata_eh_reset_timeouts
[try++];
1678 /* shut up during boot probing */
1680 ata_port_printk(ap
, KERN_INFO
, "%s resetting port\n",
1681 reset
== softreset
? "soft" : "hard");
1683 /* mark that this EH session started with reset */
1684 if (reset
== hardreset
)
1685 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
1687 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
1689 rc
= ata_do_reset(ap
, reset
, classes
, deadline
);
1691 if (reset
== hardreset
&&
1692 ata_eh_followup_srst_needed(rc
, classify
, classes
)) {
1693 /* okay, let's do follow-up softreset */
1697 ata_port_printk(ap
, KERN_ERR
,
1698 "follow-up softreset required "
1699 "but no softreset avaliable\n");
1703 ata_eh_about_to_do(ap
, NULL
, ATA_EH_RESET_MASK
);
1704 rc
= ata_do_reset(ap
, reset
, classes
, deadline
);
1706 if (rc
== 0 && classify
&&
1707 classes
[0] == ATA_DEV_UNKNOWN
) {
1708 ata_port_printk(ap
, KERN_ERR
,
1709 "classification failed\n");
1714 if (rc
&& try < ARRAY_SIZE(ata_eh_reset_timeouts
)) {
1715 unsigned long now
= jiffies
;
1717 if (time_before(now
, deadline
)) {
1718 unsigned long delta
= deadline
- jiffies
;
1720 ata_port_printk(ap
, KERN_WARNING
, "reset failed "
1721 "(errno=%d), retrying in %u secs\n",
1722 rc
, (jiffies_to_msecs(delta
) + 999) / 1000);
1724 schedule_timeout_uninterruptible(delta
);
1727 if (reset
== hardreset
&&
1728 try == ARRAY_SIZE(ata_eh_reset_timeouts
) - 1)
1729 sata_down_spd_limit(ap
);
1736 /* After the reset, the device state is PIO 0 and the
1737 * controller state is undefined. Record the mode.
1739 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1740 ap
->device
[i
].pio_mode
= XFER_PIO_0
;
1743 postreset(ap
, classes
);
1745 /* reset successful, schedule revalidation */
1746 ata_eh_done(ap
, NULL
, ehc
->i
.action
& ATA_EH_RESET_MASK
);
1747 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1753 static int ata_eh_revalidate_and_attach(struct ata_port
*ap
,
1754 struct ata_device
**r_failed_dev
)
1756 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1757 struct ata_device
*dev
;
1758 unsigned int new_mask
= 0;
1759 unsigned long flags
;
1764 /* For PATA drive side cable detection to work, IDENTIFY must
1765 * be done backwards such that PDIAG- is released by the slave
1766 * device before the master device is identified.
1768 for (i
= ATA_MAX_DEVICES
- 1; i
>= 0; i
--) {
1769 unsigned int action
, readid_flags
= 0;
1771 dev
= &ap
->device
[i
];
1772 action
= ata_eh_dev_action(dev
);
1774 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
1775 readid_flags
|= ATA_READID_POSTRESET
;
1777 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
1778 if (ata_port_offline(ap
)) {
1783 ata_eh_about_to_do(ap
, dev
, ATA_EH_REVALIDATE
);
1784 rc
= ata_dev_revalidate(dev
, readid_flags
);
1788 ata_eh_done(ap
, dev
, ATA_EH_REVALIDATE
);
1790 /* Configuration may have changed, reconfigure
1793 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
1795 /* schedule the scsi_rescan_device() here */
1796 queue_work(ata_aux_wq
, &(ap
->scsi_rescan_task
));
1797 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
1798 ehc
->tries
[dev
->devno
] &&
1799 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
1800 dev
->class = ehc
->classes
[dev
->devno
];
1802 rc
= ata_dev_read_id(dev
, &dev
->class, readid_flags
,
1809 /* IDENTIFY was issued to non-existent
1810 * device. No need to reset. Just
1811 * thaw and kill the device.
1813 ata_eh_thaw_port(ap
);
1814 dev
->class = ATA_DEV_UNKNOWN
;
1817 dev
->class = ATA_DEV_UNKNOWN
;
1823 /* PDIAG- should have been released, ask cable type if post-reset */
1824 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ap
->ops
->cable_detect
)
1825 ap
->cbl
= ap
->ops
->cable_detect(ap
);
1827 /* Configure new devices forward such that user doesn't see
1828 * device detection messages backwards.
1830 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1831 dev
= &ap
->device
[i
];
1833 if (!(new_mask
& (1 << i
)))
1836 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
1837 rc
= ata_dev_configure(dev
);
1838 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
1842 spin_lock_irqsave(ap
->lock
, flags
);
1843 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1844 spin_unlock_irqrestore(ap
->lock
, flags
);
1846 /* new device discovered, configure xfermode */
1847 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
1853 *r_failed_dev
= dev
;
1854 DPRINTK("EXIT rc=%d\n", rc
);
1858 static int ata_port_nr_enabled(struct ata_port
*ap
)
1862 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1863 if (ata_dev_enabled(&ap
->device
[i
]))
1868 static int ata_port_nr_vacant(struct ata_port
*ap
)
1872 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1873 if (ap
->device
[i
].class == ATA_DEV_UNKNOWN
)
1878 static int ata_eh_skip_recovery(struct ata_port
*ap
)
1880 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1883 /* thaw frozen port, resume link and recover failed devices */
1884 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) ||
1885 (ehc
->i
.flags
& ATA_EHI_RESUME_LINK
) || ata_port_nr_enabled(ap
))
1888 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1889 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1890 struct ata_device
*dev
= &ap
->device
[i
];
1892 if (dev
->class == ATA_DEV_UNKNOWN
&&
1893 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
1900 static void ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
1902 struct ata_port
*ap
= dev
->ap
;
1903 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1905 ehc
->tries
[dev
->devno
]--;
1909 /* device missing or wrong IDENTIFY data, schedule probing */
1910 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
1912 /* give it just one more chance */
1913 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
1915 if (ehc
->tries
[dev
->devno
] == 1) {
1916 /* This is the last chance, better to slow
1917 * down than lose it.
1919 sata_down_spd_limit(ap
);
1920 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
1924 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
1925 /* disable device if it has used up all its chances */
1926 ata_dev_disable(dev
);
1928 /* detach if offline */
1929 if (ata_port_offline(ap
))
1930 ata_eh_detach_dev(dev
);
1932 /* probe if requested */
1933 if ((ehc
->i
.probe_mask
& (1 << dev
->devno
)) &&
1934 !(ehc
->did_probe_mask
& (1 << dev
->devno
))) {
1935 ata_eh_detach_dev(dev
);
1938 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
1939 ehc
->did_probe_mask
|= (1 << dev
->devno
);
1940 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1943 /* soft didn't work? be haaaaard */
1944 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
1945 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1947 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1952 * ata_eh_recover - recover host port after error
1953 * @ap: host port to recover
1954 * @prereset: prereset method (can be NULL)
1955 * @softreset: softreset method (can be NULL)
1956 * @hardreset: hardreset method (can be NULL)
1957 * @postreset: postreset method (can be NULL)
1959 * This is the alpha and omega, eum and yang, heart and soul of
1960 * libata exception handling. On entry, actions required to
1961 * recover the port and hotplug requests are recorded in
1962 * eh_context. This function executes all the operations with
1963 * appropriate retrials and fallbacks to resurrect failed
1964 * devices, detach goners and greet newcomers.
1967 * Kernel thread context (may sleep).
1970 * 0 on success, -errno on failure.
1972 static int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
1973 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
1974 ata_postreset_fn_t postreset
)
1976 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1977 struct ata_device
*dev
;
1982 /* prep for recovery */
1983 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1984 dev
= &ap
->device
[i
];
1986 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
1988 /* collect port action mask recorded in dev actions */
1989 ehc
->i
.action
|= ehc
->i
.dev_action
[i
] & ~ATA_EH_PERDEV_MASK
;
1990 ehc
->i
.dev_action
[i
] &= ATA_EH_PERDEV_MASK
;
1992 /* process hotplug request */
1993 if (dev
->flags
& ATA_DFLAG_DETACH
)
1994 ata_eh_detach_dev(dev
);
1996 if (!ata_dev_enabled(dev
) &&
1997 ((ehc
->i
.probe_mask
& (1 << dev
->devno
)) &&
1998 !(ehc
->did_probe_mask
& (1 << dev
->devno
)))) {
1999 ata_eh_detach_dev(dev
);
2001 ehc
->did_probe_mask
|= (1 << dev
->devno
);
2002 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2009 /* if UNLOADING, finish immediately */
2010 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
2013 /* skip EH if possible. */
2014 if (ata_eh_skip_recovery(ap
))
2017 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
2018 ehc
->classes
[i
] = ATA_DEV_UNKNOWN
;
2021 if (ehc
->i
.action
& ATA_EH_RESET_MASK
) {
2022 ata_eh_freeze_port(ap
);
2024 rc
= ata_eh_reset(ap
, ata_port_nr_vacant(ap
), prereset
,
2025 softreset
, hardreset
, postreset
);
2027 ata_port_printk(ap
, KERN_ERR
,
2028 "reset failed, giving up\n");
2032 ata_eh_thaw_port(ap
);
2035 /* revalidate existing devices and attach new ones */
2036 rc
= ata_eh_revalidate_and_attach(ap
, &dev
);
2040 /* configure transfer mode if necessary */
2041 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
2042 rc
= ata_set_mode(ap
, &dev
);
2045 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
2051 ata_eh_handle_dev_fail(dev
, rc
);
2053 if (ata_port_nr_enabled(ap
)) {
2054 ata_port_printk(ap
, KERN_WARNING
, "failed to recover some "
2055 "devices, retrying in 5 secs\n");
2058 /* no device left, repeat fast */
2066 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
2067 ata_dev_disable(&ap
->device
[i
]);
2070 DPRINTK("EXIT, rc=%d\n", rc
);
2075 * ata_eh_finish - finish up EH
2076 * @ap: host port to finish EH for
2078 * Recovery is complete. Clean up EH states and retry or finish
2084 static void ata_eh_finish(struct ata_port
*ap
)
2088 /* retry or finish qcs */
2089 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2090 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2092 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
2096 /* FIXME: Once EH migration is complete,
2097 * generate sense data in this function,
2098 * considering both err_mask and tf.
2100 if (qc
->err_mask
& AC_ERR_INVALID
)
2101 ata_eh_qc_complete(qc
);
2103 ata_eh_qc_retry(qc
);
2105 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
2106 ata_eh_qc_complete(qc
);
2108 /* feed zero TF to sense generation */
2109 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
2110 ata_eh_qc_retry(qc
);
2117 * ata_do_eh - do standard error handling
2118 * @ap: host port to handle error for
2119 * @prereset: prereset method (can be NULL)
2120 * @softreset: softreset method (can be NULL)
2121 * @hardreset: hardreset method (can be NULL)
2122 * @postreset: postreset method (can be NULL)
2124 * Perform standard error handling sequence.
2127 * Kernel thread context (may sleep).
2129 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
2130 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
2131 ata_postreset_fn_t postreset
)
2135 ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
);
2141 * ata_eh_handle_port_suspend - perform port suspend operation
2142 * @ap: port to suspend
2147 * Kernel thread context (may sleep).
2149 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
2151 unsigned long flags
;
2154 /* are we suspending? */
2155 spin_lock_irqsave(ap
->lock
, flags
);
2156 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
2157 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
2158 spin_unlock_irqrestore(ap
->lock
, flags
);
2161 spin_unlock_irqrestore(ap
->lock
, flags
);
2163 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
2165 /* tell ACPI we're suspending */
2166 rc
= ata_acpi_on_suspend(ap
);
2171 ata_eh_freeze_port(ap
);
2173 if (ap
->ops
->port_suspend
)
2174 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
2178 spin_lock_irqsave(ap
->lock
, flags
);
2180 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
2182 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
2183 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2184 ata_port_schedule_eh(ap
);
2186 if (ap
->pm_result
) {
2187 *ap
->pm_result
= rc
;
2188 ap
->pm_result
= NULL
;
2191 spin_unlock_irqrestore(ap
->lock
, flags
);
2197 * ata_eh_handle_port_resume - perform port resume operation
2198 * @ap: port to resume
2203 * Kernel thread context (may sleep).
2205 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
2207 unsigned long flags
;
2210 /* are we resuming? */
2211 spin_lock_irqsave(ap
->lock
, flags
);
2212 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
2213 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
2214 spin_unlock_irqrestore(ap
->lock
, flags
);
2217 spin_unlock_irqrestore(ap
->lock
, flags
);
2219 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
2221 if (ap
->ops
->port_resume
)
2222 rc
= ap
->ops
->port_resume(ap
);
2224 /* tell ACPI that we're resuming */
2225 ata_acpi_on_resume(ap
);
2228 spin_lock_irqsave(ap
->lock
, flags
);
2229 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
2230 if (ap
->pm_result
) {
2231 *ap
->pm_result
= rc
;
2232 ap
->pm_result
= NULL
;
2234 spin_unlock_irqrestore(ap
->lock
, flags
);
2236 #endif /* CONFIG_PM */