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/config.h>
36 #include <linux/kernel.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_transport_api.h"
44 #include <linux/libata.h>
48 static void __ata_port_freeze(struct ata_port
*ap
);
50 static void ata_ering_record(struct ata_ering
*ering
, int is_io
,
51 unsigned int err_mask
)
53 struct ata_ering_entry
*ent
;
58 ering
->cursor
%= ATA_ERING_SIZE
;
60 ent
= &ering
->ring
[ering
->cursor
];
62 ent
->err_mask
= err_mask
;
63 ent
->timestamp
= get_jiffies_64();
66 static struct ata_ering_entry
* ata_ering_top(struct ata_ering
*ering
)
68 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
74 static int ata_ering_map(struct ata_ering
*ering
,
75 int (*map_fn
)(struct ata_ering_entry
*, void *),
79 struct ata_ering_entry
*ent
;
83 ent
= &ering
->ring
[idx
];
86 rc
= map_fn(ent
, arg
);
89 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
90 } while (idx
!= ering
->cursor
);
96 * ata_scsi_timed_out - SCSI layer time out callback
97 * @cmd: timed out SCSI command
99 * Handles SCSI layer timeout. We race with normal completion of
100 * the qc for @cmd. If the qc is already gone, we lose and let
101 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
102 * timed out and EH should be invoked. Prevent ata_qc_complete()
103 * from finishing it by setting EH_SCHEDULED and return
106 * TODO: kill this function once old EH is gone.
109 * Called from timer context
112 * EH_HANDLED or EH_NOT_HANDLED
114 enum scsi_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
116 struct Scsi_Host
*host
= cmd
->device
->host
;
117 struct ata_port
*ap
= ata_shost_to_port(host
);
119 struct ata_queued_cmd
*qc
;
120 enum scsi_eh_timer_return ret
;
124 if (ap
->ops
->error_handler
) {
125 ret
= EH_NOT_HANDLED
;
130 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
131 qc
= ata_qc_from_tag(ap
, ap
->active_tag
);
133 WARN_ON(qc
->scsicmd
!= cmd
);
134 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
135 qc
->err_mask
|= AC_ERR_TIMEOUT
;
136 ret
= EH_NOT_HANDLED
;
138 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
141 DPRINTK("EXIT, ret=%d\n", ret
);
146 * ata_scsi_error - SCSI layer error handler callback
147 * @host: SCSI host on which error occurred
149 * Handles SCSI-layer-thrown error events.
152 * Inherited from SCSI layer (none, can sleep)
157 void ata_scsi_error(struct Scsi_Host
*host
)
159 struct ata_port
*ap
= ata_shost_to_port(host
);
160 spinlock_t
*hs_lock
= &ap
->host_set
->lock
;
161 int i
, repeat_cnt
= ATA_EH_MAX_REPEAT
;
166 /* synchronize with port task */
167 ata_port_flush_task(ap
);
169 /* synchronize with host_set lock and sort out timeouts */
171 /* For new EH, all qcs are finished in one of three ways -
172 * normal completion, error completion, and SCSI timeout.
173 * Both cmpletions can race against SCSI timeout. When normal
174 * completion wins, the qc never reaches EH. When error
175 * completion wins, the qc has ATA_QCFLAG_FAILED set.
177 * When SCSI timeout wins, things are a bit more complex.
178 * Normal or error completion can occur after the timeout but
179 * before this point. In such cases, both types of
180 * completions are honored. A scmd is determined to have
181 * timed out iff its associated qc is active and not failed.
183 if (ap
->ops
->error_handler
) {
184 struct scsi_cmnd
*scmd
, *tmp
;
187 spin_lock_irqsave(hs_lock
, flags
);
189 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
190 struct ata_queued_cmd
*qc
;
192 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
193 qc
= __ata_qc_from_tag(ap
, i
);
194 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
199 if (i
< ATA_MAX_QUEUE
) {
200 /* the scmd has an associated qc */
201 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
202 /* which hasn't failed yet, timeout */
203 qc
->err_mask
|= AC_ERR_TIMEOUT
;
204 qc
->flags
|= ATA_QCFLAG_FAILED
;
208 /* Normal completion occurred after
209 * SCSI timeout but before this point.
210 * Successfully complete it.
212 scmd
->retries
= scmd
->allowed
;
213 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
217 /* If we have timed out qcs. They belong to EH from
218 * this point but the state of the controller is
219 * unknown. Freeze the port to make sure the IRQ
220 * handler doesn't diddle with those qcs. This must
221 * be done atomically w.r.t. setting QCFLAG_FAILED.
224 __ata_port_freeze(ap
);
226 spin_unlock_irqrestore(hs_lock
, flags
);
228 spin_unlock_wait(hs_lock
);
231 /* invoke error handler */
232 if (ap
->ops
->error_handler
) {
233 /* fetch & clear EH info */
234 spin_lock_irqsave(hs_lock
, flags
);
236 memset(&ap
->eh_context
, 0, sizeof(ap
->eh_context
));
237 ap
->eh_context
.i
= ap
->eh_info
;
238 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
240 ap
->flags
|= ATA_FLAG_EH_IN_PROGRESS
;
241 ap
->flags
&= ~ATA_FLAG_EH_PENDING
;
243 spin_unlock_irqrestore(hs_lock
, flags
);
246 ap
->ops
->error_handler(ap
);
248 /* Exception might have happend after ->error_handler
249 * recovered the port but before this point. Repeat
252 spin_lock_irqsave(hs_lock
, flags
);
254 if (ap
->flags
& ATA_FLAG_EH_PENDING
) {
256 ata_port_printk(ap
, KERN_INFO
,
257 "EH pending after completion, "
258 "repeating EH (cnt=%d)\n", repeat_cnt
);
259 spin_unlock_irqrestore(hs_lock
, flags
);
262 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
263 "tries, giving up\n", ATA_EH_MAX_REPEAT
);
266 /* this run is complete, make sure EH info is clear */
267 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
269 /* Clear host_eh_scheduled while holding hs_lock such
270 * that if exception occurs after this point but
271 * before EH completion, SCSI midlayer will
274 host
->host_eh_scheduled
= 0;
276 spin_unlock_irqrestore(hs_lock
, flags
);
278 WARN_ON(ata_qc_from_tag(ap
, ap
->active_tag
) == NULL
);
279 ap
->ops
->eng_timeout(ap
);
282 /* finish or retry handled scmd's and clean up */
283 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
285 scsi_eh_flush_done_q(&ap
->eh_done_q
);
288 spin_lock_irqsave(hs_lock
, flags
);
290 if (ap
->flags
& ATA_FLAG_LOADING
) {
291 ap
->flags
&= ~ATA_FLAG_LOADING
;
293 if (ap
->flags
& ATA_FLAG_SCSI_HOTPLUG
)
294 queue_work(ata_aux_wq
, &ap
->hotplug_task
);
295 if (ap
->flags
& ATA_FLAG_RECOVERED
)
296 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
299 ap
->flags
&= ~(ATA_FLAG_SCSI_HOTPLUG
| ATA_FLAG_RECOVERED
);
301 /* tell wait_eh that we're done */
302 ap
->flags
&= ~ATA_FLAG_EH_IN_PROGRESS
;
303 wake_up_all(&ap
->eh_wait_q
);
305 spin_unlock_irqrestore(hs_lock
, flags
);
311 * ata_port_wait_eh - Wait for the currently pending EH to complete
312 * @ap: Port to wait EH for
314 * Wait until the currently pending EH is complete.
317 * Kernel thread context (may sleep).
319 void ata_port_wait_eh(struct ata_port
*ap
)
325 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
327 while (ap
->flags
& (ATA_FLAG_EH_PENDING
| ATA_FLAG_EH_IN_PROGRESS
)) {
328 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
329 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
331 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
334 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
336 /* make sure SCSI EH is complete */
337 if (scsi_host_in_recovery(ap
->host
)) {
344 * ata_qc_timeout - Handle timeout of queued command
345 * @qc: Command that timed out
347 * Some part of the kernel (currently, only the SCSI layer)
348 * has noticed that the active command on port @ap has not
349 * completed after a specified length of time. Handle this
350 * condition by disabling DMA (if necessary) and completing
351 * transactions, with error if necessary.
353 * This also handles the case of the "lost interrupt", where
354 * for some reason (possibly hardware bug, possibly driver bug)
355 * an interrupt was not delivered to the driver, even though the
356 * transaction completed successfully.
358 * TODO: kill this function once old EH is gone.
361 * Inherited from SCSI layer (none, can sleep)
363 static void ata_qc_timeout(struct ata_queued_cmd
*qc
)
365 struct ata_port
*ap
= qc
->ap
;
366 struct ata_host_set
*host_set
= ap
->host_set
;
367 u8 host_stat
= 0, drv_stat
;
372 ap
->hsm_task_state
= HSM_ST_IDLE
;
374 spin_lock_irqsave(&host_set
->lock
, flags
);
376 switch (qc
->tf
.protocol
) {
379 case ATA_PROT_ATAPI_DMA
:
380 host_stat
= ap
->ops
->bmdma_status(ap
);
382 /* before we do anything else, clear DMA-Start bit */
383 ap
->ops
->bmdma_stop(qc
);
389 drv_stat
= ata_chk_status(ap
);
391 /* ack bmdma irq events */
392 ap
->ops
->irq_clear(ap
);
394 ata_dev_printk(qc
->dev
, KERN_ERR
, "command 0x%x timeout, "
395 "stat 0x%x host_stat 0x%x\n",
396 qc
->tf
.command
, drv_stat
, host_stat
);
398 /* complete taskfile transaction */
399 qc
->err_mask
|= AC_ERR_TIMEOUT
;
403 spin_unlock_irqrestore(&host_set
->lock
, flags
);
405 ata_eh_qc_complete(qc
);
411 * ata_eng_timeout - Handle timeout of queued command
412 * @ap: Port on which timed-out command is active
414 * Some part of the kernel (currently, only the SCSI layer)
415 * has noticed that the active command on port @ap has not
416 * completed after a specified length of time. Handle this
417 * condition by disabling DMA (if necessary) and completing
418 * transactions, with error if necessary.
420 * This also handles the case of the "lost interrupt", where
421 * for some reason (possibly hardware bug, possibly driver bug)
422 * an interrupt was not delivered to the driver, even though the
423 * transaction completed successfully.
425 * TODO: kill this function once old EH is gone.
428 * Inherited from SCSI layer (none, can sleep)
430 void ata_eng_timeout(struct ata_port
*ap
)
434 ata_qc_timeout(ata_qc_from_tag(ap
, ap
->active_tag
));
440 * ata_qc_schedule_eh - schedule qc for error handling
441 * @qc: command to schedule error handling for
443 * Schedule error handling for @qc. EH will kick in as soon as
444 * other commands are drained.
447 * spin_lock_irqsave(host_set lock)
449 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
451 struct ata_port
*ap
= qc
->ap
;
453 WARN_ON(!ap
->ops
->error_handler
);
455 qc
->flags
|= ATA_QCFLAG_FAILED
;
456 qc
->ap
->flags
|= ATA_FLAG_EH_PENDING
;
458 /* The following will fail if timeout has already expired.
459 * ata_scsi_error() takes care of such scmds on EH entry.
460 * Note that ATA_QCFLAG_FAILED is unconditionally set after
461 * this function completes.
463 scsi_req_abort_cmd(qc
->scsicmd
);
467 * ata_port_schedule_eh - schedule error handling without a qc
468 * @ap: ATA port to schedule EH for
470 * Schedule error handling for @ap. EH will kick in as soon as
471 * all commands are drained.
474 * spin_lock_irqsave(host_set lock)
476 void ata_port_schedule_eh(struct ata_port
*ap
)
478 WARN_ON(!ap
->ops
->error_handler
);
480 ap
->flags
|= ATA_FLAG_EH_PENDING
;
481 scsi_schedule_eh(ap
->host
);
483 DPRINTK("port EH scheduled\n");
487 * ata_port_abort - abort all qc's on the port
488 * @ap: ATA port to abort qc's for
490 * Abort all active qc's of @ap and schedule EH.
493 * spin_lock_irqsave(host_set lock)
496 * Number of aborted qc's.
498 int ata_port_abort(struct ata_port
*ap
)
500 int tag
, nr_aborted
= 0;
502 WARN_ON(!ap
->ops
->error_handler
);
504 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
505 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
508 qc
->flags
|= ATA_QCFLAG_FAILED
;
515 ata_port_schedule_eh(ap
);
521 * __ata_port_freeze - freeze port
522 * @ap: ATA port to freeze
524 * This function is called when HSM violation or some other
525 * condition disrupts normal operation of the port. Frozen port
526 * is not allowed to perform any operation until the port is
527 * thawed, which usually follows a successful reset.
529 * ap->ops->freeze() callback can be used for freezing the port
530 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
531 * port cannot be frozen hardware-wise, the interrupt handler
532 * must ack and clear interrupts unconditionally while the port
536 * spin_lock_irqsave(host_set lock)
538 static void __ata_port_freeze(struct ata_port
*ap
)
540 WARN_ON(!ap
->ops
->error_handler
);
545 ap
->flags
|= ATA_FLAG_FROZEN
;
547 DPRINTK("ata%u port frozen\n", ap
->id
);
551 * ata_port_freeze - abort & freeze port
552 * @ap: ATA port to freeze
554 * Abort and freeze @ap.
557 * spin_lock_irqsave(host_set lock)
560 * Number of aborted commands.
562 int ata_port_freeze(struct ata_port
*ap
)
566 WARN_ON(!ap
->ops
->error_handler
);
568 nr_aborted
= ata_port_abort(ap
);
569 __ata_port_freeze(ap
);
575 * ata_eh_freeze_port - EH helper to freeze port
576 * @ap: ATA port to freeze
583 void ata_eh_freeze_port(struct ata_port
*ap
)
587 if (!ap
->ops
->error_handler
)
590 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
591 __ata_port_freeze(ap
);
592 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
596 * ata_port_thaw_port - EH helper to thaw port
597 * @ap: ATA port to thaw
599 * Thaw frozen port @ap.
604 void ata_eh_thaw_port(struct ata_port
*ap
)
608 if (!ap
->ops
->error_handler
)
611 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
613 ap
->flags
&= ~ATA_FLAG_FROZEN
;
618 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
620 DPRINTK("ata%u port thawed\n", ap
->id
);
623 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
628 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
630 struct ata_port
*ap
= qc
->ap
;
631 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
634 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
635 qc
->scsidone
= ata_eh_scsidone
;
636 __ata_qc_complete(qc
);
637 WARN_ON(ata_tag_valid(qc
->tag
));
638 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
640 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
644 * ata_eh_qc_complete - Complete an active ATA command from EH
645 * @qc: Command to complete
647 * Indicate to the mid and upper layers that an ATA command has
648 * completed. To be used from EH.
650 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
652 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
653 scmd
->retries
= scmd
->allowed
;
654 __ata_eh_qc_complete(qc
);
658 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
659 * @qc: Command to retry
661 * Indicate to the mid and upper layers that an ATA command
662 * should be retried. To be used from EH.
664 * SCSI midlayer limits the number of retries to scmd->allowed.
665 * scmd->retries is decremented for commands which get retried
666 * due to unrelated failures (qc->err_mask is zero).
668 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
670 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
671 if (!qc
->err_mask
&& scmd
->retries
)
673 __ata_eh_qc_complete(qc
);
677 * ata_eh_detach_dev - detach ATA device
678 * @dev: ATA device to detach
685 static void ata_eh_detach_dev(struct ata_device
*dev
)
687 struct ata_port
*ap
= dev
->ap
;
690 ata_dev_disable(dev
);
692 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
694 dev
->flags
&= ~ATA_DFLAG_DETACH
;
696 if (ata_scsi_offline_dev(dev
)) {
697 dev
->flags
|= ATA_DFLAG_DETACHED
;
698 ap
->flags
|= ATA_FLAG_SCSI_HOTPLUG
;
701 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
705 * ata_eh_about_to_do - about to perform eh_action
706 * @ap: target ATA port
707 * @action: action about to be performed
709 * Called just before performing EH actions to clear related bits
710 * in @ap->eh_info such that eh actions are not unnecessarily
716 static void ata_eh_about_to_do(struct ata_port
*ap
, unsigned int action
)
720 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
721 ap
->eh_info
.action
&= ~action
;
722 ap
->flags
|= ATA_FLAG_RECOVERED
;
723 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
727 * ata_err_string - convert err_mask to descriptive string
728 * @err_mask: error mask to convert to string
730 * Convert @err_mask to descriptive string. Errors are
731 * prioritized according to severity and only the most severe
738 * Descriptive string for @err_mask
740 static const char * ata_err_string(unsigned int err_mask
)
742 if (err_mask
& AC_ERR_HOST_BUS
)
743 return "host bus error";
744 if (err_mask
& AC_ERR_ATA_BUS
)
745 return "ATA bus error";
746 if (err_mask
& AC_ERR_TIMEOUT
)
748 if (err_mask
& AC_ERR_HSM
)
749 return "HSM violation";
750 if (err_mask
& AC_ERR_SYSTEM
)
751 return "internal error";
752 if (err_mask
& AC_ERR_MEDIA
)
753 return "media error";
754 if (err_mask
& AC_ERR_INVALID
)
755 return "invalid argument";
756 if (err_mask
& AC_ERR_DEV
)
757 return "device error";
758 return "unknown error";
762 * ata_read_log_page - read a specific log page
763 * @dev: target device
764 * @page: page to read
765 * @buf: buffer to store read page
766 * @sectors: number of sectors to read
768 * Read log page using READ_LOG_EXT command.
771 * Kernel thread context (may sleep).
774 * 0 on success, AC_ERR_* mask otherwise.
776 static unsigned int ata_read_log_page(struct ata_device
*dev
,
777 u8 page
, void *buf
, unsigned int sectors
)
779 struct ata_taskfile tf
;
780 unsigned int err_mask
;
782 DPRINTK("read log page - page %d\n", page
);
784 ata_tf_init(dev
, &tf
);
785 tf
.command
= ATA_CMD_READ_LOG_EXT
;
788 tf
.hob_nsect
= sectors
>> 8;
789 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
790 tf
.protocol
= ATA_PROT_PIO
;
792 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
793 buf
, sectors
* ATA_SECT_SIZE
);
795 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
800 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
801 * @dev: Device to read log page 10h from
802 * @tag: Resulting tag of the failed command
803 * @tf: Resulting taskfile registers of the failed command
805 * Read log page 10h to obtain NCQ error details and clear error
809 * Kernel thread context (may sleep).
812 * 0 on success, -errno otherwise.
814 static int ata_eh_read_log_10h(struct ata_device
*dev
,
815 int *tag
, struct ata_taskfile
*tf
)
817 u8
*buf
= dev
->ap
->sector_buf
;
818 unsigned int err_mask
;
822 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
827 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
830 ata_dev_printk(dev
, KERN_WARNING
,
831 "invalid checksum 0x%x on log page 10h\n", csum
);
836 *tag
= buf
[0] & 0x1f;
838 tf
->command
= buf
[2];
839 tf
->feature
= buf
[3];
844 tf
->hob_lbal
= buf
[8];
845 tf
->hob_lbam
= buf
[9];
846 tf
->hob_lbah
= buf
[10];
848 tf
->hob_nsect
= buf
[13];
854 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
855 * @dev: device to perform REQUEST_SENSE to
856 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
858 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
859 * SENSE. This function is EH helper.
862 * Kernel thread context (may sleep).
865 * 0 on success, AC_ERR_* mask on failure
867 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
868 unsigned char *sense_buf
)
870 struct ata_port
*ap
= dev
->ap
;
871 struct ata_taskfile tf
;
872 u8 cdb
[ATAPI_CDB_LEN
];
874 DPRINTK("ATAPI request sense\n");
876 ata_tf_init(dev
, &tf
);
878 /* FIXME: is this needed? */
879 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
881 /* XXX: why tf_read here? */
882 ap
->ops
->tf_read(ap
, &tf
);
884 /* fill these in, for the case where they are -not- overwritten */
886 sense_buf
[2] = tf
.feature
>> 4;
888 memset(cdb
, 0, ATAPI_CDB_LEN
);
889 cdb
[0] = REQUEST_SENSE
;
890 cdb
[4] = SCSI_SENSE_BUFFERSIZE
;
892 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
893 tf
.command
= ATA_CMD_PACKET
;
895 /* is it pointless to prefer PIO for "safety reasons"? */
896 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
897 tf
.protocol
= ATA_PROT_ATAPI_DMA
;
898 tf
.feature
|= ATAPI_PKT_DMA
;
900 tf
.protocol
= ATA_PROT_ATAPI
;
901 tf
.lbam
= (8 * 1024) & 0xff;
902 tf
.lbah
= (8 * 1024) >> 8;
905 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
906 sense_buf
, SCSI_SENSE_BUFFERSIZE
);
910 * ata_eh_analyze_serror - analyze SError for a failed port
911 * @ap: ATA port to analyze SError for
913 * Analyze SError if available and further determine cause of
919 static void ata_eh_analyze_serror(struct ata_port
*ap
)
921 struct ata_eh_context
*ehc
= &ap
->eh_context
;
922 u32 serror
= ehc
->i
.serror
;
923 unsigned int err_mask
= 0, action
= 0;
925 if (serror
& SERR_PERSISTENT
) {
926 err_mask
|= AC_ERR_ATA_BUS
;
927 action
|= ATA_EH_HARDRESET
;
930 (SERR_DATA_RECOVERED
| SERR_COMM_RECOVERED
| SERR_DATA
)) {
931 err_mask
|= AC_ERR_ATA_BUS
;
932 action
|= ATA_EH_SOFTRESET
;
934 if (serror
& SERR_PROTOCOL
) {
935 err_mask
|= AC_ERR_HSM
;
936 action
|= ATA_EH_SOFTRESET
;
938 if (serror
& SERR_INTERNAL
) {
939 err_mask
|= AC_ERR_SYSTEM
;
940 action
|= ATA_EH_SOFTRESET
;
942 if (serror
& (SERR_PHYRDY_CHG
| SERR_DEV_XCHG
))
943 ata_ehi_hotplugged(&ehc
->i
);
945 ehc
->i
.err_mask
|= err_mask
;
946 ehc
->i
.action
|= action
;
950 * ata_eh_analyze_ncq_error - analyze NCQ error
951 * @ap: ATA port to analyze NCQ error for
953 * Read log page 10h, determine the offending qc and acquire
954 * error status TF. For NCQ device errors, all LLDDs have to do
955 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
959 * Kernel thread context (may sleep).
961 static void ata_eh_analyze_ncq_error(struct ata_port
*ap
)
963 struct ata_eh_context
*ehc
= &ap
->eh_context
;
964 struct ata_device
*dev
= ap
->device
;
965 struct ata_queued_cmd
*qc
;
966 struct ata_taskfile tf
;
969 /* if frozen, we can't do much */
970 if (ap
->flags
& ATA_FLAG_FROZEN
)
973 /* is it NCQ device error? */
974 if (!ap
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
977 /* has LLDD analyzed already? */
978 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
979 qc
= __ata_qc_from_tag(ap
, tag
);
981 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
988 /* okay, this error is ours */
989 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
991 ata_port_printk(ap
, KERN_ERR
, "failed to read log page 10h "
996 if (!(ap
->sactive
& (1 << tag
))) {
997 ata_port_printk(ap
, KERN_ERR
, "log page 10h reported "
998 "inactive tag %d\n", tag
);
1002 /* we've got the perpetrator, condemn it */
1003 qc
= __ata_qc_from_tag(ap
, tag
);
1004 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1005 qc
->err_mask
|= AC_ERR_DEV
;
1006 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1010 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1011 * @qc: qc to analyze
1012 * @tf: Taskfile registers to analyze
1014 * Analyze taskfile of @qc and further determine cause of
1015 * failure. This function also requests ATAPI sense data if
1019 * Kernel thread context (may sleep).
1022 * Determined recovery action
1024 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1025 const struct ata_taskfile
*tf
)
1027 unsigned int tmp
, action
= 0;
1028 u8 stat
= tf
->command
, err
= tf
->feature
;
1030 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1031 qc
->err_mask
|= AC_ERR_HSM
;
1032 return ATA_EH_SOFTRESET
;
1035 if (!(qc
->err_mask
& AC_ERR_DEV
))
1038 switch (qc
->dev
->class) {
1041 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1043 qc
->err_mask
|= AC_ERR_MEDIA
;
1045 qc
->err_mask
|= AC_ERR_INVALID
;
1049 tmp
= atapi_eh_request_sense(qc
->dev
,
1050 qc
->scsicmd
->sense_buffer
);
1052 /* ATA_QCFLAG_SENSE_VALID is used to tell
1053 * atapi_qc_complete() that sense data is
1056 * TODO: interpret sense data and set
1057 * appropriate err_mask.
1059 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1061 qc
->err_mask
|= tmp
;
1064 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1065 action
|= ATA_EH_SOFTRESET
;
1070 static int ata_eh_categorize_ering_entry(struct ata_ering_entry
*ent
)
1072 if (ent
->err_mask
& (AC_ERR_ATA_BUS
| AC_ERR_TIMEOUT
))
1076 if (ent
->err_mask
& AC_ERR_HSM
)
1078 if ((ent
->err_mask
&
1079 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1086 struct speed_down_needed_arg
{
1091 static int speed_down_needed_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1093 struct speed_down_needed_arg
*arg
= void_arg
;
1095 if (ent
->timestamp
< arg
->since
)
1098 arg
->nr_errors
[ata_eh_categorize_ering_entry(ent
)]++;
1103 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
1104 * @dev: Device of interest
1106 * This function examines error ring of @dev and determines
1107 * whether speed down is necessary. Speed down is necessary if
1108 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
1109 * errors during last 15 minutes.
1111 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
1112 * violation for known supported commands.
1114 * Cat-2 errors are unclassified DEV error for known supported
1118 * Inherited from caller.
1121 * 1 if speed down is necessary, 0 otherwise
1123 static int ata_eh_speed_down_needed(struct ata_device
*dev
)
1125 const u64 interval
= 15LLU * 60 * HZ
;
1126 static const int err_limits
[3] = { -1, 3, 10 };
1127 struct speed_down_needed_arg arg
;
1128 struct ata_ering_entry
*ent
;
1132 ent
= ata_ering_top(&dev
->ering
);
1136 err_cat
= ata_eh_categorize_ering_entry(ent
);
1140 memset(&arg
, 0, sizeof(arg
));
1142 j64
= get_jiffies_64();
1143 if (j64
>= interval
)
1144 arg
.since
= j64
- interval
;
1148 ata_ering_map(&dev
->ering
, speed_down_needed_cb
, &arg
);
1150 return arg
.nr_errors
[err_cat
] > err_limits
[err_cat
];
1154 * ata_eh_speed_down - record error and speed down if necessary
1155 * @dev: Failed device
1156 * @is_io: Did the device fail during normal IO?
1157 * @err_mask: err_mask of the error
1159 * Record error and examine error history to determine whether
1160 * adjusting transmission speed is necessary. It also sets
1161 * transmission limits appropriately if such adjustment is
1165 * Kernel thread context (may sleep).
1168 * 0 on success, -errno otherwise
1170 static int ata_eh_speed_down(struct ata_device
*dev
, int is_io
,
1171 unsigned int err_mask
)
1176 /* record error and determine whether speed down is necessary */
1177 ata_ering_record(&dev
->ering
, is_io
, err_mask
);
1179 if (!ata_eh_speed_down_needed(dev
))
1182 /* speed down SATA link speed if possible */
1183 if (sata_down_spd_limit(dev
->ap
) == 0)
1184 return ATA_EH_HARDRESET
;
1186 /* lower transfer mode */
1187 if (ata_down_xfermask_limit(dev
, 0) == 0)
1188 return ATA_EH_SOFTRESET
;
1190 ata_dev_printk(dev
, KERN_ERR
,
1191 "speed down requested but no transfer mode left\n");
1196 * ata_eh_autopsy - analyze error and determine recovery action
1197 * @ap: ATA port to perform autopsy on
1199 * Analyze why @ap failed and determine which recovery action is
1200 * needed. This function also sets more detailed AC_ERR_* values
1201 * and fills sense data for ATAPI CHECK SENSE.
1204 * Kernel thread context (may sleep).
1206 static void ata_eh_autopsy(struct ata_port
*ap
)
1208 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1209 unsigned int action
= ehc
->i
.action
;
1210 struct ata_device
*failed_dev
= NULL
;
1211 unsigned int all_err_mask
= 0;
1218 /* obtain and analyze SError */
1219 rc
= sata_scr_read(ap
, SCR_ERROR
, &serror
);
1221 ehc
->i
.serror
|= serror
;
1222 ata_eh_analyze_serror(ap
);
1223 } else if (rc
!= -EOPNOTSUPP
)
1224 action
|= ATA_EH_HARDRESET
;
1226 /* analyze NCQ failure */
1227 ata_eh_analyze_ncq_error(ap
);
1229 /* any real error trumps AC_ERR_OTHER */
1230 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1231 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1233 all_err_mask
|= ehc
->i
.err_mask
;
1235 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1236 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1238 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1241 /* inherit upper level err_mask */
1242 qc
->err_mask
|= ehc
->i
.err_mask
;
1245 action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1247 /* DEV errors are probably spurious in case of ATA_BUS error */
1248 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1249 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1252 /* any real error trumps unknown error */
1253 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1254 qc
->err_mask
&= ~AC_ERR_OTHER
;
1256 /* SENSE_VALID trumps dev/unknown error and revalidation */
1257 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1258 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1259 action
&= ~ATA_EH_REVALIDATE
;
1262 /* accumulate error info */
1263 failed_dev
= qc
->dev
;
1264 all_err_mask
|= qc
->err_mask
;
1265 if (qc
->flags
& ATA_QCFLAG_IO
)
1269 /* speed down iff command was in progress */
1271 action
|= ata_eh_speed_down(failed_dev
, is_io
, all_err_mask
);
1273 /* enforce default EH actions */
1274 if (ap
->flags
& ATA_FLAG_FROZEN
||
1275 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
1276 action
|= ATA_EH_SOFTRESET
;
1277 else if (all_err_mask
)
1278 action
|= ATA_EH_REVALIDATE
;
1280 /* record autopsy result */
1281 ehc
->i
.dev
= failed_dev
;
1282 ehc
->i
.action
= action
;
1288 * ata_eh_report - report error handling to user
1289 * @ap: ATA port EH is going on
1291 * Report EH to user.
1296 static void ata_eh_report(struct ata_port
*ap
)
1298 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1299 const char *frozen
, *desc
;
1300 int tag
, nr_failed
= 0;
1303 if (ehc
->i
.desc
[0] != '\0')
1306 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1307 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1309 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1311 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
1317 if (!nr_failed
&& !ehc
->i
.err_mask
)
1321 if (ap
->flags
& ATA_FLAG_FROZEN
)
1325 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
1326 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1327 ehc
->i
.err_mask
, ap
->sactive
, ehc
->i
.serror
,
1328 ehc
->i
.action
, frozen
);
1330 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "(%s)\n", desc
);
1332 ata_port_printk(ap
, KERN_ERR
, "exception Emask 0x%x "
1333 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1334 ehc
->i
.err_mask
, ap
->sactive
, ehc
->i
.serror
,
1335 ehc
->i
.action
, frozen
);
1337 ata_port_printk(ap
, KERN_ERR
, "(%s)\n", desc
);
1340 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1341 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1343 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || !qc
->err_mask
)
1346 ata_dev_printk(qc
->dev
, KERN_ERR
, "tag %d cmd 0x%x "
1347 "Emask 0x%x stat 0x%x err 0x%x (%s)\n",
1348 qc
->tag
, qc
->tf
.command
, qc
->err_mask
,
1349 qc
->result_tf
.command
, qc
->result_tf
.feature
,
1350 ata_err_string(qc
->err_mask
));
1354 static int ata_eh_followup_srst_needed(int rc
, int classify
,
1355 const unsigned int *classes
)
1361 if (classify
&& classes
[0] == ATA_DEV_UNKNOWN
)
1366 static int ata_eh_reset(struct ata_port
*ap
, int classify
,
1367 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
1368 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
1370 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1371 unsigned int *classes
= ehc
->classes
;
1372 int tries
= ATA_EH_RESET_TRIES
;
1373 int verbose
= !(ap
->flags
& ATA_FLAG_LOADING
);
1374 unsigned int action
;
1375 ata_reset_fn_t reset
;
1376 int i
, did_followup_srst
, rc
;
1378 /* Determine which reset to use and record in ehc->i.action.
1379 * prereset() may examine and modify it.
1381 action
= ehc
->i
.action
;
1382 ehc
->i
.action
&= ~ATA_EH_RESET_MASK
;
1383 if (softreset
&& (!hardreset
|| (!sata_set_spd_needed(ap
) &&
1384 !(action
& ATA_EH_HARDRESET
))))
1385 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1387 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1392 ata_port_printk(ap
, KERN_ERR
,
1393 "prereset failed (errno=%d)\n", rc
);
1398 /* prereset() might have modified ehc->i.action */
1399 if (ehc
->i
.action
& ATA_EH_HARDRESET
)
1401 else if (ehc
->i
.action
& ATA_EH_SOFTRESET
)
1404 /* prereset told us not to reset, bang classes and return */
1405 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1406 classes
[i
] = ATA_DEV_NONE
;
1410 /* did prereset() screw up? if so, fix up to avoid oopsing */
1412 ata_port_printk(ap
, KERN_ERR
, "BUG: prereset() requested "
1413 "invalid reset type\n");
1421 /* shut up during boot probing */
1423 ata_port_printk(ap
, KERN_INFO
, "%s resetting port\n",
1424 reset
== softreset
? "soft" : "hard");
1427 ata_eh_about_to_do(ap
, ATA_EH_RESET_MASK
);
1428 ehc
->i
.flags
|= ATA_EHI_DID_RESET
;
1430 rc
= ata_do_reset(ap
, reset
, classes
);
1432 did_followup_srst
= 0;
1433 if (reset
== hardreset
&&
1434 ata_eh_followup_srst_needed(rc
, classify
, classes
)) {
1435 /* okay, let's do follow-up softreset */
1436 did_followup_srst
= 1;
1440 ata_port_printk(ap
, KERN_ERR
,
1441 "follow-up softreset required "
1442 "but no softreset avaliable\n");
1446 ata_eh_about_to_do(ap
, ATA_EH_RESET_MASK
);
1447 rc
= ata_do_reset(ap
, reset
, classes
);
1449 if (rc
== 0 && classify
&&
1450 classes
[0] == ATA_DEV_UNKNOWN
) {
1451 ata_port_printk(ap
, KERN_ERR
,
1452 "classification failed\n");
1457 if (rc
&& --tries
) {
1460 if (reset
== softreset
) {
1461 if (did_followup_srst
)
1462 type
= "follow-up soft";
1468 ata_port_printk(ap
, KERN_WARNING
,
1469 "%sreset failed, retrying in 5 secs\n", type
);
1472 if (reset
== hardreset
)
1473 sata_down_spd_limit(ap
);
1480 /* After the reset, the device state is PIO 0 and the
1481 * controller state is undefined. Record the mode.
1483 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1484 ap
->device
[i
].pio_mode
= XFER_PIO_0
;
1487 postreset(ap
, classes
);
1489 /* reset successful, schedule revalidation */
1491 ehc
->i
.action
&= ~ATA_EH_RESET_MASK
;
1492 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1498 static int ata_eh_revalidate_and_attach(struct ata_port
*ap
,
1499 struct ata_device
**r_failed_dev
)
1501 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1502 struct ata_device
*dev
;
1503 unsigned long flags
;
1508 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1509 dev
= &ap
->device
[i
];
1511 if (ehc
->i
.action
& ATA_EH_REVALIDATE
&& ata_dev_enabled(dev
) &&
1512 (!ehc
->i
.dev
|| ehc
->i
.dev
== dev
)) {
1513 if (ata_port_offline(ap
)) {
1518 ata_eh_about_to_do(ap
, ATA_EH_REVALIDATE
);
1519 rc
= ata_dev_revalidate(dev
,
1520 ehc
->i
.flags
& ATA_EHI_DID_RESET
);
1524 ehc
->i
.action
&= ~ATA_EH_REVALIDATE
;
1525 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
1526 ehc
->tries
[dev
->devno
] &&
1527 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
1528 dev
->class = ehc
->classes
[dev
->devno
];
1530 rc
= ata_dev_read_id(dev
, &dev
->class, 1, dev
->id
);
1532 rc
= ata_dev_configure(dev
, 1);
1535 dev
->class = ATA_DEV_UNKNOWN
;
1539 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
1540 ap
->flags
|= ATA_FLAG_SCSI_HOTPLUG
;
1541 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
1546 *r_failed_dev
= dev
;
1552 static int ata_port_nr_enabled(struct ata_port
*ap
)
1556 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1557 if (ata_dev_enabled(&ap
->device
[i
]))
1562 static int ata_port_nr_vacant(struct ata_port
*ap
)
1566 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1567 if (ap
->device
[i
].class == ATA_DEV_UNKNOWN
)
1572 static int ata_eh_skip_recovery(struct ata_port
*ap
)
1574 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1577 if (ap
->flags
& ATA_FLAG_FROZEN
|| ata_port_nr_enabled(ap
))
1580 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1581 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1582 struct ata_device
*dev
= &ap
->device
[i
];
1584 if (dev
->class == ATA_DEV_UNKNOWN
&&
1585 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
1593 * ata_eh_recover - recover host port after error
1594 * @ap: host port to recover
1595 * @prereset: prereset method (can be NULL)
1596 * @softreset: softreset method (can be NULL)
1597 * @hardreset: hardreset method (can be NULL)
1598 * @postreset: postreset method (can be NULL)
1600 * This is the alpha and omega, eum and yang, heart and soul of
1601 * libata exception handling. On entry, actions required to
1602 * recover the port and hotplug requests are recorded in
1603 * eh_context. This function executes all the operations with
1604 * appropriate retrials and fallbacks to resurrect failed
1605 * devices, detach goners and greet newcomers.
1608 * Kernel thread context (may sleep).
1611 * 0 on success, -errno on failure.
1613 static int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
1614 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
1615 ata_postreset_fn_t postreset
)
1617 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1618 struct ata_device
*dev
;
1619 int down_xfermask
, i
, rc
;
1623 /* prep for recovery */
1624 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1625 dev
= &ap
->device
[i
];
1627 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
1629 /* process hotplug request */
1630 if (dev
->flags
& ATA_DFLAG_DETACH
)
1631 ata_eh_detach_dev(dev
);
1633 if (!ata_dev_enabled(dev
) &&
1634 ((ehc
->i
.probe_mask
& (1 << dev
->devno
)) &&
1635 !(ehc
->did_probe_mask
& (1 << dev
->devno
)))) {
1636 ata_eh_detach_dev(dev
);
1638 ehc
->did_probe_mask
|= (1 << dev
->devno
);
1639 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1647 /* skip EH if possible. */
1648 if (ata_eh_skip_recovery(ap
))
1651 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1652 ehc
->classes
[i
] = ATA_DEV_UNKNOWN
;
1655 if (ehc
->i
.action
& ATA_EH_RESET_MASK
) {
1656 ata_eh_freeze_port(ap
);
1658 rc
= ata_eh_reset(ap
, ata_port_nr_vacant(ap
), prereset
,
1659 softreset
, hardreset
, postreset
);
1661 ata_port_printk(ap
, KERN_ERR
,
1662 "reset failed, giving up\n");
1666 ata_eh_thaw_port(ap
);
1669 /* revalidate existing devices and attach new ones */
1670 rc
= ata_eh_revalidate_and_attach(ap
, &dev
);
1674 /* configure transfer mode if the port has been reset */
1675 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
1676 rc
= ata_set_mode(ap
, &dev
);
1688 /* device missing, schedule probing */
1689 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
1691 ehc
->tries
[dev
->devno
] = 0;
1694 sata_down_spd_limit(ap
);
1696 ehc
->tries
[dev
->devno
]--;
1697 if (down_xfermask
&&
1698 ata_down_xfermask_limit(dev
, ehc
->tries
[dev
->devno
] == 1))
1699 ehc
->tries
[dev
->devno
] = 0;
1702 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
1703 /* disable device if it has used up all its chances */
1704 ata_dev_disable(dev
);
1706 /* detach if offline */
1707 if (ata_port_offline(ap
))
1708 ata_eh_detach_dev(dev
);
1710 /* probe if requested */
1711 if ((ehc
->i
.probe_mask
& (1 << dev
->devno
)) &&
1712 !(ehc
->did_probe_mask
& (1 << dev
->devno
))) {
1713 ata_eh_detach_dev(dev
);
1716 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
1717 ehc
->did_probe_mask
|= (1 << dev
->devno
);
1718 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1721 /* soft didn't work? be haaaaard */
1722 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
1723 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1725 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1728 if (ata_port_nr_enabled(ap
)) {
1729 ata_port_printk(ap
, KERN_WARNING
, "failed to recover some "
1730 "devices, retrying in 5 secs\n");
1733 /* no device left, repeat fast */
1741 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1742 ata_dev_disable(&ap
->device
[i
]);
1745 DPRINTK("EXIT, rc=%d\n", rc
);
1750 * ata_eh_finish - finish up EH
1751 * @ap: host port to finish EH for
1753 * Recovery is complete. Clean up EH states and retry or finish
1759 static void ata_eh_finish(struct ata_port
*ap
)
1763 /* retry or finish qcs */
1764 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1765 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1767 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1771 /* FIXME: Once EH migration is complete,
1772 * generate sense data in this function,
1773 * considering both err_mask and tf.
1775 if (qc
->err_mask
& AC_ERR_INVALID
)
1776 ata_eh_qc_complete(qc
);
1778 ata_eh_qc_retry(qc
);
1780 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1781 ata_eh_qc_complete(qc
);
1783 /* feed zero TF to sense generation */
1784 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
1785 ata_eh_qc_retry(qc
);
1792 * ata_do_eh - do standard error handling
1793 * @ap: host port to handle error for
1794 * @prereset: prereset method (can be NULL)
1795 * @softreset: softreset method (can be NULL)
1796 * @hardreset: hardreset method (can be NULL)
1797 * @postreset: postreset method (can be NULL)
1799 * Perform standard error handling sequence.
1802 * Kernel thread context (may sleep).
1804 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
1805 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
1806 ata_postreset_fn_t postreset
)
1808 if (!(ap
->flags
& ATA_FLAG_LOADING
)) {
1813 ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
);