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>
43 #include <linux/libata.h>
47 static void __ata_port_freeze(struct ata_port
*ap
);
49 static void ata_ering_record(struct ata_ering
*ering
, int is_io
,
50 unsigned int err_mask
)
52 struct ata_ering_entry
*ent
;
57 ering
->cursor
%= ATA_ERING_SIZE
;
59 ent
= &ering
->ring
[ering
->cursor
];
61 ent
->err_mask
= err_mask
;
62 ent
->timestamp
= get_jiffies_64();
65 static struct ata_ering_entry
* ata_ering_top(struct ata_ering
*ering
)
67 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
73 static int ata_ering_map(struct ata_ering
*ering
,
74 int (*map_fn
)(struct ata_ering_entry
*, void *),
78 struct ata_ering_entry
*ent
;
82 ent
= &ering
->ring
[idx
];
85 rc
= map_fn(ent
, arg
);
88 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
89 } while (idx
!= ering
->cursor
);
95 * ata_scsi_timed_out - SCSI layer time out callback
96 * @cmd: timed out SCSI command
98 * Handles SCSI layer timeout. We race with normal completion of
99 * the qc for @cmd. If the qc is already gone, we lose and let
100 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
101 * timed out and EH should be invoked. Prevent ata_qc_complete()
102 * from finishing it by setting EH_SCHEDULED and return
105 * TODO: kill this function once old EH is gone.
108 * Called from timer context
111 * EH_HANDLED or EH_NOT_HANDLED
113 enum scsi_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
115 struct Scsi_Host
*host
= cmd
->device
->host
;
116 struct ata_port
*ap
= ata_shost_to_port(host
);
118 struct ata_queued_cmd
*qc
;
119 enum scsi_eh_timer_return ret
;
123 if (ap
->ops
->error_handler
) {
124 ret
= EH_NOT_HANDLED
;
129 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
130 qc
= ata_qc_from_tag(ap
, ap
->active_tag
);
132 WARN_ON(qc
->scsicmd
!= cmd
);
133 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
134 qc
->err_mask
|= AC_ERR_TIMEOUT
;
135 ret
= EH_NOT_HANDLED
;
137 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
140 DPRINTK("EXIT, ret=%d\n", ret
);
145 * ata_scsi_error - SCSI layer error handler callback
146 * @host: SCSI host on which error occurred
148 * Handles SCSI-layer-thrown error events.
151 * Inherited from SCSI layer (none, can sleep)
156 void ata_scsi_error(struct Scsi_Host
*host
)
158 struct ata_port
*ap
= ata_shost_to_port(host
);
159 spinlock_t
*hs_lock
= &ap
->host_set
->lock
;
160 int i
, repeat_cnt
= ATA_EH_MAX_REPEAT
;
165 /* synchronize with port task */
166 ata_port_flush_task(ap
);
168 /* synchronize with host_set lock and sort out timeouts */
170 /* For new EH, all qcs are finished in one of three ways -
171 * normal completion, error completion, and SCSI timeout.
172 * Both cmpletions can race against SCSI timeout. When normal
173 * completion wins, the qc never reaches EH. When error
174 * completion wins, the qc has ATA_QCFLAG_FAILED set.
176 * When SCSI timeout wins, things are a bit more complex.
177 * Normal or error completion can occur after the timeout but
178 * before this point. In such cases, both types of
179 * completions are honored. A scmd is determined to have
180 * timed out iff its associated qc is active and not failed.
182 if (ap
->ops
->error_handler
) {
183 struct scsi_cmnd
*scmd
, *tmp
;
186 spin_lock_irqsave(hs_lock
, flags
);
188 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
189 struct ata_queued_cmd
*qc
;
191 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
192 qc
= __ata_qc_from_tag(ap
, i
);
193 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
198 if (i
< ATA_MAX_QUEUE
) {
199 /* the scmd has an associated qc */
200 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
201 /* which hasn't failed yet, timeout */
202 qc
->err_mask
|= AC_ERR_TIMEOUT
;
203 qc
->flags
|= ATA_QCFLAG_FAILED
;
207 /* Normal completion occurred after
208 * SCSI timeout but before this point.
209 * Successfully complete it.
211 scmd
->retries
= scmd
->allowed
;
212 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
216 /* If we have timed out qcs. They belong to EH from
217 * this point but the state of the controller is
218 * unknown. Freeze the port to make sure the IRQ
219 * handler doesn't diddle with those qcs. This must
220 * be done atomically w.r.t. setting QCFLAG_FAILED.
223 __ata_port_freeze(ap
);
225 spin_unlock_irqrestore(hs_lock
, flags
);
227 spin_unlock_wait(hs_lock
);
230 /* invoke error handler */
231 if (ap
->ops
->error_handler
) {
232 /* fetch & clear EH info */
233 spin_lock_irqsave(hs_lock
, flags
);
235 memset(&ap
->eh_context
, 0, sizeof(ap
->eh_context
));
236 ap
->eh_context
.i
= ap
->eh_info
;
237 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
239 ap
->flags
&= ~ATA_FLAG_EH_PENDING
;
241 spin_unlock_irqrestore(hs_lock
, flags
);
244 ap
->ops
->error_handler(ap
);
246 /* Exception might have happend after ->error_handler
247 * recovered the port but before this point. Repeat
250 spin_lock_irqsave(hs_lock
, flags
);
252 if (ap
->flags
& ATA_FLAG_EH_PENDING
) {
254 ata_port_printk(ap
, KERN_INFO
,
255 "EH pending after completion, "
256 "repeating EH (cnt=%d)\n", repeat_cnt
);
257 spin_unlock_irqrestore(hs_lock
, flags
);
260 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
261 "tries, giving up\n", ATA_EH_MAX_REPEAT
);
264 /* this run is complete, make sure EH info is clear */
265 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
267 /* Clear host_eh_scheduled while holding hs_lock such
268 * that if exception occurs after this point but
269 * before EH completion, SCSI midlayer will
272 host
->host_eh_scheduled
= 0;
274 spin_unlock_irqrestore(hs_lock
, flags
);
276 WARN_ON(ata_qc_from_tag(ap
, ap
->active_tag
) == NULL
);
277 ap
->ops
->eng_timeout(ap
);
280 /* finish or retry handled scmd's and clean up */
281 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
283 scsi_eh_flush_done_q(&ap
->eh_done_q
);
286 spin_lock_irqsave(hs_lock
, flags
);
288 if (ap
->flags
& ATA_FLAG_RECOVERED
)
289 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
290 ap
->flags
&= ~ATA_FLAG_RECOVERED
;
292 spin_unlock_irqrestore(hs_lock
, flags
);
298 * ata_qc_timeout - Handle timeout of queued command
299 * @qc: Command that timed out
301 * Some part of the kernel (currently, only the SCSI layer)
302 * has noticed that the active command on port @ap has not
303 * completed after a specified length of time. Handle this
304 * condition by disabling DMA (if necessary) and completing
305 * transactions, with error if necessary.
307 * This also handles the case of the "lost interrupt", where
308 * for some reason (possibly hardware bug, possibly driver bug)
309 * an interrupt was not delivered to the driver, even though the
310 * transaction completed successfully.
312 * TODO: kill this function once old EH is gone.
315 * Inherited from SCSI layer (none, can sleep)
317 static void ata_qc_timeout(struct ata_queued_cmd
*qc
)
319 struct ata_port
*ap
= qc
->ap
;
320 struct ata_host_set
*host_set
= ap
->host_set
;
321 u8 host_stat
= 0, drv_stat
;
326 ap
->hsm_task_state
= HSM_ST_IDLE
;
328 spin_lock_irqsave(&host_set
->lock
, flags
);
330 switch (qc
->tf
.protocol
) {
333 case ATA_PROT_ATAPI_DMA
:
334 host_stat
= ap
->ops
->bmdma_status(ap
);
336 /* before we do anything else, clear DMA-Start bit */
337 ap
->ops
->bmdma_stop(qc
);
343 drv_stat
= ata_chk_status(ap
);
345 /* ack bmdma irq events */
346 ap
->ops
->irq_clear(ap
);
348 ata_dev_printk(qc
->dev
, KERN_ERR
, "command 0x%x timeout, "
349 "stat 0x%x host_stat 0x%x\n",
350 qc
->tf
.command
, drv_stat
, host_stat
);
352 /* complete taskfile transaction */
353 qc
->err_mask
|= AC_ERR_TIMEOUT
;
357 spin_unlock_irqrestore(&host_set
->lock
, flags
);
359 ata_eh_qc_complete(qc
);
365 * ata_eng_timeout - Handle timeout of queued command
366 * @ap: Port on which timed-out command is active
368 * Some part of the kernel (currently, only the SCSI layer)
369 * has noticed that the active command on port @ap has not
370 * completed after a specified length of time. Handle this
371 * condition by disabling DMA (if necessary) and completing
372 * transactions, with error if necessary.
374 * This also handles the case of the "lost interrupt", where
375 * for some reason (possibly hardware bug, possibly driver bug)
376 * an interrupt was not delivered to the driver, even though the
377 * transaction completed successfully.
379 * TODO: kill this function once old EH is gone.
382 * Inherited from SCSI layer (none, can sleep)
384 void ata_eng_timeout(struct ata_port
*ap
)
388 ata_qc_timeout(ata_qc_from_tag(ap
, ap
->active_tag
));
394 * ata_qc_schedule_eh - schedule qc for error handling
395 * @qc: command to schedule error handling for
397 * Schedule error handling for @qc. EH will kick in as soon as
398 * other commands are drained.
401 * spin_lock_irqsave(host_set lock)
403 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
405 struct ata_port
*ap
= qc
->ap
;
407 WARN_ON(!ap
->ops
->error_handler
);
409 qc
->flags
|= ATA_QCFLAG_FAILED
;
410 qc
->ap
->flags
|= ATA_FLAG_EH_PENDING
;
412 /* The following will fail if timeout has already expired.
413 * ata_scsi_error() takes care of such scmds on EH entry.
414 * Note that ATA_QCFLAG_FAILED is unconditionally set after
415 * this function completes.
417 scsi_req_abort_cmd(qc
->scsicmd
);
421 * ata_port_schedule_eh - schedule error handling without a qc
422 * @ap: ATA port to schedule EH for
424 * Schedule error handling for @ap. EH will kick in as soon as
425 * all commands are drained.
428 * spin_lock_irqsave(host_set lock)
430 void ata_port_schedule_eh(struct ata_port
*ap
)
432 WARN_ON(!ap
->ops
->error_handler
);
434 ap
->flags
|= ATA_FLAG_EH_PENDING
;
435 ata_schedule_scsi_eh(ap
->host
);
437 DPRINTK("port EH scheduled\n");
441 * ata_port_abort - abort all qc's on the port
442 * @ap: ATA port to abort qc's for
444 * Abort all active qc's of @ap and schedule EH.
447 * spin_lock_irqsave(host_set lock)
450 * Number of aborted qc's.
452 int ata_port_abort(struct ata_port
*ap
)
454 int tag
, nr_aborted
= 0;
456 WARN_ON(!ap
->ops
->error_handler
);
458 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
459 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
462 qc
->flags
|= ATA_QCFLAG_FAILED
;
469 ata_port_schedule_eh(ap
);
475 * __ata_port_freeze - freeze port
476 * @ap: ATA port to freeze
478 * This function is called when HSM violation or some other
479 * condition disrupts normal operation of the port. Frozen port
480 * is not allowed to perform any operation until the port is
481 * thawed, which usually follows a successful reset.
483 * ap->ops->freeze() callback can be used for freezing the port
484 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
485 * port cannot be frozen hardware-wise, the interrupt handler
486 * must ack and clear interrupts unconditionally while the port
490 * spin_lock_irqsave(host_set lock)
492 static void __ata_port_freeze(struct ata_port
*ap
)
494 WARN_ON(!ap
->ops
->error_handler
);
499 ap
->flags
|= ATA_FLAG_FROZEN
;
501 DPRINTK("ata%u port frozen\n", ap
->id
);
505 * ata_port_freeze - abort & freeze port
506 * @ap: ATA port to freeze
508 * Abort and freeze @ap.
511 * spin_lock_irqsave(host_set lock)
514 * Number of aborted commands.
516 int ata_port_freeze(struct ata_port
*ap
)
520 WARN_ON(!ap
->ops
->error_handler
);
522 nr_aborted
= ata_port_abort(ap
);
523 __ata_port_freeze(ap
);
529 * ata_eh_freeze_port - EH helper to freeze port
530 * @ap: ATA port to freeze
537 void ata_eh_freeze_port(struct ata_port
*ap
)
541 if (!ap
->ops
->error_handler
)
544 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
545 __ata_port_freeze(ap
);
546 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
550 * ata_port_thaw_port - EH helper to thaw port
551 * @ap: ATA port to thaw
553 * Thaw frozen port @ap.
558 void ata_eh_thaw_port(struct ata_port
*ap
)
562 if (!ap
->ops
->error_handler
)
565 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
567 ap
->flags
&= ~ATA_FLAG_FROZEN
;
572 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
574 DPRINTK("ata%u port thawed\n", ap
->id
);
577 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
582 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
584 struct ata_port
*ap
= qc
->ap
;
585 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
588 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
589 qc
->scsidone
= ata_eh_scsidone
;
590 __ata_qc_complete(qc
);
591 WARN_ON(ata_tag_valid(qc
->tag
));
592 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
594 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
598 * ata_eh_qc_complete - Complete an active ATA command from EH
599 * @qc: Command to complete
601 * Indicate to the mid and upper layers that an ATA command has
602 * completed. To be used from EH.
604 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
606 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
607 scmd
->retries
= scmd
->allowed
;
608 __ata_eh_qc_complete(qc
);
612 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
613 * @qc: Command to retry
615 * Indicate to the mid and upper layers that an ATA command
616 * should be retried. To be used from EH.
618 * SCSI midlayer limits the number of retries to scmd->allowed.
619 * scmd->retries is decremented for commands which get retried
620 * due to unrelated failures (qc->err_mask is zero).
622 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
624 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
625 if (!qc
->err_mask
&& scmd
->retries
)
627 __ata_eh_qc_complete(qc
);
631 * ata_eh_about_to_do - about to perform eh_action
632 * @ap: target ATA port
633 * @action: action about to be performed
635 * Called just before performing EH actions to clear related bits
636 * in @ap->eh_info such that eh actions are not unnecessarily
642 static void ata_eh_about_to_do(struct ata_port
*ap
, unsigned int action
)
646 spin_lock_irqsave(&ap
->host_set
->lock
, flags
);
647 ap
->eh_info
.action
&= ~action
;
648 ap
->flags
|= ATA_FLAG_RECOVERED
;
649 spin_unlock_irqrestore(&ap
->host_set
->lock
, flags
);
653 * ata_err_string - convert err_mask to descriptive string
654 * @err_mask: error mask to convert to string
656 * Convert @err_mask to descriptive string. Errors are
657 * prioritized according to severity and only the most severe
664 * Descriptive string for @err_mask
666 static const char * ata_err_string(unsigned int err_mask
)
668 if (err_mask
& AC_ERR_HOST_BUS
)
669 return "host bus error";
670 if (err_mask
& AC_ERR_ATA_BUS
)
671 return "ATA bus error";
672 if (err_mask
& AC_ERR_TIMEOUT
)
674 if (err_mask
& AC_ERR_HSM
)
675 return "HSM violation";
676 if (err_mask
& AC_ERR_SYSTEM
)
677 return "internal error";
678 if (err_mask
& AC_ERR_MEDIA
)
679 return "media error";
680 if (err_mask
& AC_ERR_INVALID
)
681 return "invalid argument";
682 if (err_mask
& AC_ERR_DEV
)
683 return "device error";
684 return "unknown error";
688 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
689 * @dev: device to perform REQUEST_SENSE to
690 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
692 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
693 * SENSE. This function is EH helper.
696 * Kernel thread context (may sleep).
699 * 0 on success, AC_ERR_* mask on failure
701 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
702 unsigned char *sense_buf
)
704 struct ata_port
*ap
= dev
->ap
;
705 struct ata_taskfile tf
;
706 u8 cdb
[ATAPI_CDB_LEN
];
708 DPRINTK("ATAPI request sense\n");
710 ata_tf_init(dev
, &tf
);
712 /* FIXME: is this needed? */
713 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
715 /* XXX: why tf_read here? */
716 ap
->ops
->tf_read(ap
, &tf
);
718 /* fill these in, for the case where they are -not- overwritten */
720 sense_buf
[2] = tf
.feature
>> 4;
722 memset(cdb
, 0, ATAPI_CDB_LEN
);
723 cdb
[0] = REQUEST_SENSE
;
724 cdb
[4] = SCSI_SENSE_BUFFERSIZE
;
726 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
727 tf
.command
= ATA_CMD_PACKET
;
729 /* is it pointless to prefer PIO for "safety reasons"? */
730 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
731 tf
.protocol
= ATA_PROT_ATAPI_DMA
;
732 tf
.feature
|= ATAPI_PKT_DMA
;
734 tf
.protocol
= ATA_PROT_ATAPI
;
735 tf
.lbam
= (8 * 1024) & 0xff;
736 tf
.lbah
= (8 * 1024) >> 8;
739 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
740 sense_buf
, SCSI_SENSE_BUFFERSIZE
);
744 * ata_eh_analyze_serror - analyze SError for a failed port
745 * @ap: ATA port to analyze SError for
747 * Analyze SError if available and further determine cause of
753 static void ata_eh_analyze_serror(struct ata_port
*ap
)
755 struct ata_eh_context
*ehc
= &ap
->eh_context
;
756 u32 serror
= ehc
->i
.serror
;
757 unsigned int err_mask
= 0, action
= 0;
759 if (serror
& SERR_PERSISTENT
) {
760 err_mask
|= AC_ERR_ATA_BUS
;
761 action
|= ATA_EH_HARDRESET
;
764 (SERR_DATA_RECOVERED
| SERR_COMM_RECOVERED
| SERR_DATA
)) {
765 err_mask
|= AC_ERR_ATA_BUS
;
766 action
|= ATA_EH_SOFTRESET
;
768 if (serror
& SERR_PROTOCOL
) {
769 err_mask
|= AC_ERR_HSM
;
770 action
|= ATA_EH_SOFTRESET
;
772 if (serror
& SERR_INTERNAL
) {
773 err_mask
|= AC_ERR_SYSTEM
;
774 action
|= ATA_EH_SOFTRESET
;
776 if (serror
& (SERR_PHYRDY_CHG
| SERR_DEV_XCHG
)) {
777 err_mask
|= AC_ERR_ATA_BUS
;
778 action
|= ATA_EH_HARDRESET
;
781 ehc
->i
.err_mask
|= err_mask
;
782 ehc
->i
.action
|= action
;
786 * ata_eh_analyze_tf - analyze taskfile of a failed qc
788 * @tf: Taskfile registers to analyze
790 * Analyze taskfile of @qc and further determine cause of
791 * failure. This function also requests ATAPI sense data if
795 * Kernel thread context (may sleep).
798 * Determined recovery action
800 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
801 const struct ata_taskfile
*tf
)
803 unsigned int tmp
, action
= 0;
804 u8 stat
= tf
->command
, err
= tf
->feature
;
806 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
807 qc
->err_mask
|= AC_ERR_HSM
;
808 return ATA_EH_SOFTRESET
;
811 if (!(qc
->err_mask
& AC_ERR_DEV
))
814 switch (qc
->dev
->class) {
817 qc
->err_mask
|= AC_ERR_ATA_BUS
;
819 qc
->err_mask
|= AC_ERR_MEDIA
;
821 qc
->err_mask
|= AC_ERR_INVALID
;
825 tmp
= atapi_eh_request_sense(qc
->dev
,
826 qc
->scsicmd
->sense_buffer
);
828 /* ATA_QCFLAG_SENSE_VALID is used to tell
829 * atapi_qc_complete() that sense data is
832 * TODO: interpret sense data and set
833 * appropriate err_mask.
835 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
840 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
841 action
|= ATA_EH_SOFTRESET
;
846 static int ata_eh_categorize_ering_entry(struct ata_ering_entry
*ent
)
848 if (ent
->err_mask
& (AC_ERR_ATA_BUS
| AC_ERR_TIMEOUT
))
852 if (ent
->err_mask
& AC_ERR_HSM
)
855 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
862 struct speed_down_needed_arg
{
867 static int speed_down_needed_cb(struct ata_ering_entry
*ent
, void *void_arg
)
869 struct speed_down_needed_arg
*arg
= void_arg
;
871 if (ent
->timestamp
< arg
->since
)
874 arg
->nr_errors
[ata_eh_categorize_ering_entry(ent
)]++;
879 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
880 * @dev: Device of interest
882 * This function examines error ring of @dev and determines
883 * whether speed down is necessary. Speed down is necessary if
884 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
885 * errors during last 15 minutes.
887 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
888 * violation for known supported commands.
890 * Cat-2 errors are unclassified DEV error for known supported
894 * Inherited from caller.
897 * 1 if speed down is necessary, 0 otherwise
899 static int ata_eh_speed_down_needed(struct ata_device
*dev
)
901 const u64 interval
= 15LLU * 60 * HZ
;
902 static const int err_limits
[3] = { -1, 3, 10 };
903 struct speed_down_needed_arg arg
;
904 struct ata_ering_entry
*ent
;
908 ent
= ata_ering_top(&dev
->ering
);
912 err_cat
= ata_eh_categorize_ering_entry(ent
);
916 memset(&arg
, 0, sizeof(arg
));
918 j64
= get_jiffies_64();
920 arg
.since
= j64
- interval
;
924 ata_ering_map(&dev
->ering
, speed_down_needed_cb
, &arg
);
926 return arg
.nr_errors
[err_cat
] > err_limits
[err_cat
];
930 * ata_eh_speed_down - record error and speed down if necessary
931 * @dev: Failed device
932 * @is_io: Did the device fail during normal IO?
933 * @err_mask: err_mask of the error
935 * Record error and examine error history to determine whether
936 * adjusting transmission speed is necessary. It also sets
937 * transmission limits appropriately if such adjustment is
941 * Kernel thread context (may sleep).
944 * 0 on success, -errno otherwise
946 static int ata_eh_speed_down(struct ata_device
*dev
, int is_io
,
947 unsigned int err_mask
)
952 /* record error and determine whether speed down is necessary */
953 ata_ering_record(&dev
->ering
, is_io
, err_mask
);
955 if (!ata_eh_speed_down_needed(dev
))
958 /* speed down SATA link speed if possible */
959 if (sata_down_spd_limit(dev
->ap
) == 0)
960 return ATA_EH_HARDRESET
;
962 /* lower transfer mode */
963 if (ata_down_xfermask_limit(dev
, 0) == 0)
964 return ATA_EH_SOFTRESET
;
966 ata_dev_printk(dev
, KERN_ERR
,
967 "speed down requested but no transfer mode left\n");
972 * ata_eh_autopsy - analyze error and determine recovery action
973 * @ap: ATA port to perform autopsy on
975 * Analyze why @ap failed and determine which recovery action is
976 * needed. This function also sets more detailed AC_ERR_* values
977 * and fills sense data for ATAPI CHECK SENSE.
980 * Kernel thread context (may sleep).
982 static void ata_eh_autopsy(struct ata_port
*ap
)
984 struct ata_eh_context
*ehc
= &ap
->eh_context
;
985 unsigned int action
= ehc
->i
.action
;
986 struct ata_device
*failed_dev
= NULL
;
987 unsigned int all_err_mask
= 0;
994 /* obtain and analyze SError */
995 rc
= sata_scr_read(ap
, SCR_ERROR
, &serror
);
997 ehc
->i
.serror
|= serror
;
998 ata_eh_analyze_serror(ap
);
999 } else if (rc
!= -EOPNOTSUPP
)
1000 action
|= ATA_EH_HARDRESET
;
1002 /* any real error trumps AC_ERR_OTHER */
1003 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1004 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1006 all_err_mask
|= ehc
->i
.err_mask
;
1008 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1009 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1011 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1014 /* inherit upper level err_mask */
1015 qc
->err_mask
|= ehc
->i
.err_mask
;
1017 if (qc
->err_mask
& AC_ERR_TIMEOUT
)
1018 action
|= ATA_EH_SOFTRESET
;
1021 action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1023 /* DEV errors are probably spurious in case of ATA_BUS error */
1024 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1025 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1028 /* any real error trumps unknown error */
1029 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1030 qc
->err_mask
&= ~AC_ERR_OTHER
;
1032 /* SENSE_VALID trumps dev/unknown error and revalidation */
1033 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1034 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1035 action
&= ~ATA_EH_REVALIDATE
;
1038 /* accumulate error info */
1039 failed_dev
= qc
->dev
;
1040 all_err_mask
|= qc
->err_mask
;
1041 if (qc
->flags
& ATA_QCFLAG_IO
)
1045 /* speed down iff command was in progress */
1047 action
|= ata_eh_speed_down(failed_dev
, is_io
, all_err_mask
);
1050 action
|= ATA_EH_REVALIDATE
;
1052 ehc
->i
.dev
= failed_dev
;
1053 ehc
->i
.action
= action
;
1059 * ata_eh_report - report error handling to user
1060 * @ap: ATA port EH is going on
1062 * Report EH to user.
1067 static void ata_eh_report(struct ata_port
*ap
)
1069 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1070 const char *frozen
, *desc
;
1071 int tag
, nr_failed
= 0;
1074 if (ehc
->i
.desc
[0] != '\0')
1077 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1078 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1080 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1082 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
1088 if (!nr_failed
&& !ehc
->i
.err_mask
)
1092 if (ap
->flags
& ATA_FLAG_FROZEN
)
1096 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
,
1097 "exception Emask 0x%x SErr 0x%x action 0x%x%s\n",
1098 ehc
->i
.err_mask
, ehc
->i
.serror
, ehc
->i
.action
,
1101 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "(%s)\n", desc
);
1103 ata_port_printk(ap
, KERN_ERR
,
1104 "exception Emask 0x%x SErr 0x%x action 0x%x%s\n",
1105 ehc
->i
.err_mask
, ehc
->i
.serror
, ehc
->i
.action
,
1108 ata_port_printk(ap
, KERN_ERR
, "(%s)\n", desc
);
1111 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1112 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1114 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || !qc
->err_mask
)
1117 ata_dev_printk(qc
->dev
, KERN_ERR
, "tag %d cmd 0x%x "
1118 "Emask 0x%x stat 0x%x err 0x%x (%s)\n",
1119 qc
->tag
, qc
->tf
.command
, qc
->err_mask
,
1120 qc
->result_tf
.command
, qc
->result_tf
.feature
,
1121 ata_err_string(qc
->err_mask
));
1125 static int ata_eh_reset(struct ata_port
*ap
, ata_reset_fn_t softreset
,
1126 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
1128 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1129 unsigned int classes
[ATA_MAX_DEVICES
];
1130 int tries
= ATA_EH_RESET_TRIES
;
1131 ata_reset_fn_t reset
;
1134 if (softreset
&& (!hardreset
|| (!sata_set_spd_needed(ap
) &&
1135 !(ehc
->i
.action
& ATA_EH_HARDRESET
))))
1141 ata_port_printk(ap
, KERN_INFO
, "%s resetting port\n",
1142 reset
== softreset
? "soft" : "hard");
1145 ata_eh_about_to_do(ap
, ATA_EH_RESET_MASK
);
1146 ehc
->i
.flags
|= ATA_EHI_DID_RESET
;
1148 rc
= ata_do_reset(ap
, reset
, classes
);
1150 if (rc
&& --tries
) {
1151 ata_port_printk(ap
, KERN_WARNING
,
1152 "%sreset failed, retrying in 5 secs\n",
1153 reset
== softreset
? "soft" : "hard");
1156 if (reset
== hardreset
)
1157 sata_down_spd_limit(ap
);
1165 postreset(ap
, classes
);
1167 /* reset successful, schedule revalidation */
1169 ehc
->i
.action
&= ~ATA_EH_RESET_MASK
;
1170 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1176 static int ata_eh_revalidate(struct ata_port
*ap
,
1177 struct ata_device
**r_failed_dev
)
1179 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1180 struct ata_device
*dev
;
1185 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1186 dev
= &ap
->device
[i
];
1188 if (ehc
->i
.action
& ATA_EH_REVALIDATE
&& ata_dev_enabled(dev
) &&
1189 (!ehc
->i
.dev
|| ehc
->i
.dev
== dev
)) {
1190 if (ata_port_offline(ap
)) {
1195 ata_eh_about_to_do(ap
, ATA_EH_REVALIDATE
);
1196 rc
= ata_dev_revalidate(dev
,
1197 ehc
->i
.flags
& ATA_EHI_DID_RESET
);
1201 ehc
->i
.action
&= ~ATA_EH_REVALIDATE
;
1206 *r_failed_dev
= dev
;
1212 static int ata_port_nr_enabled(struct ata_port
*ap
)
1216 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1217 if (ata_dev_enabled(&ap
->device
[i
]))
1223 * ata_eh_recover - recover host port after error
1224 * @ap: host port to recover
1225 * @softreset: softreset method (can be NULL)
1226 * @hardreset: hardreset method (can be NULL)
1227 * @postreset: postreset method (can be NULL)
1229 * This is the alpha and omega, eum and yang, heart and soul of
1230 * libata exception handling. On entry, actions required to
1231 * recover each devices are recorded in eh_context. This
1232 * function executes all the operations with appropriate retrials
1233 * and fallbacks to resurrect failed devices.
1236 * Kernel thread context (may sleep).
1239 * 0 on success, -errno on failure.
1241 static int ata_eh_recover(struct ata_port
*ap
, ata_reset_fn_t softreset
,
1242 ata_reset_fn_t hardreset
,
1243 ata_postreset_fn_t postreset
)
1245 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1246 struct ata_device
*dev
;
1247 int down_xfermask
, i
, rc
;
1251 /* prep for recovery */
1252 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1253 dev
= &ap
->device
[i
];
1255 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
1262 /* skip EH if possible. */
1263 if (!ata_port_nr_enabled(ap
) && !(ap
->flags
& ATA_FLAG_FROZEN
))
1267 if (ehc
->i
.action
& ATA_EH_RESET_MASK
) {
1268 ata_eh_freeze_port(ap
);
1270 rc
= ata_eh_reset(ap
, softreset
, hardreset
, postreset
);
1272 ata_port_printk(ap
, KERN_ERR
,
1273 "reset failed, giving up\n");
1277 ata_eh_thaw_port(ap
);
1280 /* revalidate existing devices */
1281 rc
= ata_eh_revalidate(ap
, &dev
);
1285 /* configure transfer mode if the port has been reset */
1286 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
1287 rc
= ata_set_mode(ap
, &dev
);
1300 ehc
->tries
[dev
->devno
] = 0;
1303 sata_down_spd_limit(ap
);
1305 ehc
->tries
[dev
->devno
]--;
1306 if (down_xfermask
&&
1307 ata_down_xfermask_limit(dev
, ehc
->tries
[dev
->devno
] == 1))
1308 ehc
->tries
[dev
->devno
] = 0;
1311 /* disable device if it has used up all its chances */
1312 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
])
1313 ata_dev_disable(dev
);
1315 /* soft didn't work? be haaaaard */
1316 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
1317 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1319 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1321 if (ata_port_nr_enabled(ap
)) {
1322 ata_port_printk(ap
, KERN_WARNING
, "failed to recover some "
1323 "devices, retrying in 5 secs\n");
1326 /* no device left, repeat fast */
1334 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1335 ata_dev_disable(&ap
->device
[i
]);
1338 DPRINTK("EXIT, rc=%d\n", rc
);
1343 * ata_eh_finish - finish up EH
1344 * @ap: host port to finish EH for
1346 * Recovery is complete. Clean up EH states and retry or finish
1352 static void ata_eh_finish(struct ata_port
*ap
)
1356 /* retry or finish qcs */
1357 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1358 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1360 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1364 /* FIXME: Once EH migration is complete,
1365 * generate sense data in this function,
1366 * considering both err_mask and tf.
1368 if (qc
->err_mask
& AC_ERR_INVALID
)
1369 ata_eh_qc_complete(qc
);
1371 ata_eh_qc_retry(qc
);
1373 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1374 ata_eh_qc_complete(qc
);
1376 /* feed zero TF to sense generation */
1377 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
1378 ata_eh_qc_retry(qc
);
1385 * ata_do_eh - do standard error handling
1386 * @ap: host port to handle error for
1387 * @softreset: softreset method (can be NULL)
1388 * @hardreset: hardreset method (can be NULL)
1389 * @postreset: postreset method (can be NULL)
1391 * Perform standard error handling sequence.
1394 * Kernel thread context (may sleep).
1396 void ata_do_eh(struct ata_port
*ap
, ata_reset_fn_t softreset
,
1397 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
1401 ata_eh_recover(ap
, softreset
, hardreset
, postreset
);