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
);
49 static void ata_eh_finish(struct ata_port
*ap
);
51 static void ata_ering_record(struct ata_ering
*ering
, int is_io
,
52 unsigned int err_mask
)
54 struct ata_ering_entry
*ent
;
59 ering
->cursor
%= ATA_ERING_SIZE
;
61 ent
= &ering
->ring
[ering
->cursor
];
63 ent
->err_mask
= err_mask
;
64 ent
->timestamp
= get_jiffies_64();
67 static struct ata_ering_entry
* ata_ering_top(struct ata_ering
*ering
)
69 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
75 static int ata_ering_map(struct ata_ering
*ering
,
76 int (*map_fn
)(struct ata_ering_entry
*, void *),
80 struct ata_ering_entry
*ent
;
84 ent
= &ering
->ring
[idx
];
87 rc
= map_fn(ent
, arg
);
90 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
91 } while (idx
!= ering
->cursor
);
96 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
98 struct ata_eh_context
*ehc
= &dev
->ap
->eh_context
;
100 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
103 static void ata_eh_clear_action(struct ata_device
*dev
,
104 struct ata_eh_info
*ehi
, unsigned int action
)
109 ehi
->action
&= ~action
;
110 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
111 ehi
->dev_action
[i
] &= ~action
;
113 /* doesn't make sense for port-wide EH actions */
114 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
116 /* break ehi->action into ehi->dev_action */
117 if (ehi
->action
& action
) {
118 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
119 ehi
->dev_action
[i
] |= ehi
->action
& action
;
120 ehi
->action
&= ~action
;
123 /* turn off the specified per-dev action */
124 ehi
->dev_action
[dev
->devno
] &= ~action
;
129 * ata_scsi_timed_out - SCSI layer time out callback
130 * @cmd: timed out SCSI command
132 * Handles SCSI layer timeout. We race with normal completion of
133 * the qc for @cmd. If the qc is already gone, we lose and let
134 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
135 * timed out and EH should be invoked. Prevent ata_qc_complete()
136 * from finishing it by setting EH_SCHEDULED and return
139 * TODO: kill this function once old EH is gone.
142 * Called from timer context
145 * EH_HANDLED or EH_NOT_HANDLED
147 enum scsi_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
149 struct Scsi_Host
*host
= cmd
->device
->host
;
150 struct ata_port
*ap
= ata_shost_to_port(host
);
152 struct ata_queued_cmd
*qc
;
153 enum scsi_eh_timer_return ret
;
157 if (ap
->ops
->error_handler
) {
158 ret
= EH_NOT_HANDLED
;
163 spin_lock_irqsave(ap
->lock
, flags
);
164 qc
= ata_qc_from_tag(ap
, ap
->active_tag
);
166 WARN_ON(qc
->scsicmd
!= cmd
);
167 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
168 qc
->err_mask
|= AC_ERR_TIMEOUT
;
169 ret
= EH_NOT_HANDLED
;
171 spin_unlock_irqrestore(ap
->lock
, flags
);
174 DPRINTK("EXIT, ret=%d\n", ret
);
179 * ata_scsi_error - SCSI layer error handler callback
180 * @host: SCSI host on which error occurred
182 * Handles SCSI-layer-thrown error events.
185 * Inherited from SCSI layer (none, can sleep)
190 void ata_scsi_error(struct Scsi_Host
*host
)
192 struct ata_port
*ap
= ata_shost_to_port(host
);
193 spinlock_t
*ap_lock
= ap
->lock
;
194 int i
, repeat_cnt
= ATA_EH_MAX_REPEAT
;
199 /* synchronize with port task */
200 ata_port_flush_task(ap
);
202 /* synchronize with host_set lock and sort out timeouts */
204 /* For new EH, all qcs are finished in one of three ways -
205 * normal completion, error completion, and SCSI timeout.
206 * Both cmpletions can race against SCSI timeout. When normal
207 * completion wins, the qc never reaches EH. When error
208 * completion wins, the qc has ATA_QCFLAG_FAILED set.
210 * When SCSI timeout wins, things are a bit more complex.
211 * Normal or error completion can occur after the timeout but
212 * before this point. In such cases, both types of
213 * completions are honored. A scmd is determined to have
214 * timed out iff its associated qc is active and not failed.
216 if (ap
->ops
->error_handler
) {
217 struct scsi_cmnd
*scmd
, *tmp
;
220 spin_lock_irqsave(ap_lock
, flags
);
222 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
223 struct ata_queued_cmd
*qc
;
225 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
226 qc
= __ata_qc_from_tag(ap
, i
);
227 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
232 if (i
< ATA_MAX_QUEUE
) {
233 /* the scmd has an associated qc */
234 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
235 /* which hasn't failed yet, timeout */
236 qc
->err_mask
|= AC_ERR_TIMEOUT
;
237 qc
->flags
|= ATA_QCFLAG_FAILED
;
241 /* Normal completion occurred after
242 * SCSI timeout but before this point.
243 * Successfully complete it.
245 scmd
->retries
= scmd
->allowed
;
246 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
250 /* If we have timed out qcs. They belong to EH from
251 * this point but the state of the controller is
252 * unknown. Freeze the port to make sure the IRQ
253 * handler doesn't diddle with those qcs. This must
254 * be done atomically w.r.t. setting QCFLAG_FAILED.
257 __ata_port_freeze(ap
);
259 spin_unlock_irqrestore(ap_lock
, flags
);
261 spin_unlock_wait(ap_lock
);
264 /* invoke error handler */
265 if (ap
->ops
->error_handler
) {
266 /* fetch & clear EH info */
267 spin_lock_irqsave(ap_lock
, flags
);
269 memset(&ap
->eh_context
, 0, sizeof(ap
->eh_context
));
270 ap
->eh_context
.i
= ap
->eh_info
;
271 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
273 ap
->flags
|= ATA_FLAG_EH_IN_PROGRESS
;
274 ap
->flags
&= ~ATA_FLAG_EH_PENDING
;
276 spin_unlock_irqrestore(ap_lock
, flags
);
278 /* invoke EH. if unloading, just finish failed qcs */
279 if (!(ap
->flags
& ATA_FLAG_UNLOADING
))
280 ap
->ops
->error_handler(ap
);
284 /* Exception might have happend after ->error_handler
285 * recovered the port but before this point. Repeat
288 spin_lock_irqsave(ap_lock
, flags
);
290 if (ap
->flags
& ATA_FLAG_EH_PENDING
) {
292 ata_port_printk(ap
, KERN_INFO
,
293 "EH pending after completion, "
294 "repeating EH (cnt=%d)\n", repeat_cnt
);
295 spin_unlock_irqrestore(ap_lock
, flags
);
298 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
299 "tries, giving up\n", ATA_EH_MAX_REPEAT
);
302 /* this run is complete, make sure EH info is clear */
303 memset(&ap
->eh_info
, 0, sizeof(ap
->eh_info
));
305 /* Clear host_eh_scheduled while holding ap_lock such
306 * that if exception occurs after this point but
307 * before EH completion, SCSI midlayer will
310 host
->host_eh_scheduled
= 0;
312 spin_unlock_irqrestore(ap_lock
, flags
);
314 WARN_ON(ata_qc_from_tag(ap
, ap
->active_tag
) == NULL
);
315 ap
->ops
->eng_timeout(ap
);
318 /* finish or retry handled scmd's and clean up */
319 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
321 scsi_eh_flush_done_q(&ap
->eh_done_q
);
324 spin_lock_irqsave(ap_lock
, flags
);
326 if (ap
->flags
& ATA_FLAG_LOADING
) {
327 ap
->flags
&= ~ATA_FLAG_LOADING
;
329 if (ap
->flags
& ATA_FLAG_SCSI_HOTPLUG
)
330 queue_work(ata_aux_wq
, &ap
->hotplug_task
);
331 if (ap
->flags
& ATA_FLAG_RECOVERED
)
332 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
335 ap
->flags
&= ~(ATA_FLAG_SCSI_HOTPLUG
| ATA_FLAG_RECOVERED
);
337 /* tell wait_eh that we're done */
338 ap
->flags
&= ~ATA_FLAG_EH_IN_PROGRESS
;
339 wake_up_all(&ap
->eh_wait_q
);
341 spin_unlock_irqrestore(ap_lock
, flags
);
347 * ata_port_wait_eh - Wait for the currently pending EH to complete
348 * @ap: Port to wait EH for
350 * Wait until the currently pending EH is complete.
353 * Kernel thread context (may sleep).
355 void ata_port_wait_eh(struct ata_port
*ap
)
361 spin_lock_irqsave(ap
->lock
, flags
);
363 while (ap
->flags
& (ATA_FLAG_EH_PENDING
| ATA_FLAG_EH_IN_PROGRESS
)) {
364 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
365 spin_unlock_irqrestore(ap
->lock
, flags
);
367 spin_lock_irqsave(ap
->lock
, flags
);
369 finish_wait(&ap
->eh_wait_q
, &wait
);
371 spin_unlock_irqrestore(ap
->lock
, flags
);
373 /* make sure SCSI EH is complete */
374 if (scsi_host_in_recovery(ap
->host
)) {
381 * ata_qc_timeout - Handle timeout of queued command
382 * @qc: Command that timed out
384 * Some part of the kernel (currently, only the SCSI layer)
385 * has noticed that the active command on port @ap has not
386 * completed after a specified length of time. Handle this
387 * condition by disabling DMA (if necessary) and completing
388 * transactions, with error if necessary.
390 * This also handles the case of the "lost interrupt", where
391 * for some reason (possibly hardware bug, possibly driver bug)
392 * an interrupt was not delivered to the driver, even though the
393 * transaction completed successfully.
395 * TODO: kill this function once old EH is gone.
398 * Inherited from SCSI layer (none, can sleep)
400 static void ata_qc_timeout(struct ata_queued_cmd
*qc
)
402 struct ata_port
*ap
= qc
->ap
;
403 u8 host_stat
= 0, drv_stat
;
408 ap
->hsm_task_state
= HSM_ST_IDLE
;
410 spin_lock_irqsave(ap
->lock
, flags
);
412 switch (qc
->tf
.protocol
) {
415 case ATA_PROT_ATAPI_DMA
:
416 host_stat
= ap
->ops
->bmdma_status(ap
);
418 /* before we do anything else, clear DMA-Start bit */
419 ap
->ops
->bmdma_stop(qc
);
425 drv_stat
= ata_chk_status(ap
);
427 /* ack bmdma irq events */
428 ap
->ops
->irq_clear(ap
);
430 ata_dev_printk(qc
->dev
, KERN_ERR
, "command 0x%x timeout, "
431 "stat 0x%x host_stat 0x%x\n",
432 qc
->tf
.command
, drv_stat
, host_stat
);
434 /* complete taskfile transaction */
435 qc
->err_mask
|= AC_ERR_TIMEOUT
;
439 spin_unlock_irqrestore(ap
->lock
, flags
);
441 ata_eh_qc_complete(qc
);
447 * ata_eng_timeout - Handle timeout of queued command
448 * @ap: Port on which timed-out command is active
450 * Some part of the kernel (currently, only the SCSI layer)
451 * has noticed that the active command on port @ap has not
452 * completed after a specified length of time. Handle this
453 * condition by disabling DMA (if necessary) and completing
454 * transactions, with error if necessary.
456 * This also handles the case of the "lost interrupt", where
457 * for some reason (possibly hardware bug, possibly driver bug)
458 * an interrupt was not delivered to the driver, even though the
459 * transaction completed successfully.
461 * TODO: kill this function once old EH is gone.
464 * Inherited from SCSI layer (none, can sleep)
466 void ata_eng_timeout(struct ata_port
*ap
)
470 ata_qc_timeout(ata_qc_from_tag(ap
, ap
->active_tag
));
476 * ata_qc_schedule_eh - schedule qc for error handling
477 * @qc: command to schedule error handling for
479 * Schedule error handling for @qc. EH will kick in as soon as
480 * other commands are drained.
483 * spin_lock_irqsave(host_set lock)
485 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
487 struct ata_port
*ap
= qc
->ap
;
489 WARN_ON(!ap
->ops
->error_handler
);
491 qc
->flags
|= ATA_QCFLAG_FAILED
;
492 qc
->ap
->flags
|= ATA_FLAG_EH_PENDING
;
494 /* The following will fail if timeout has already expired.
495 * ata_scsi_error() takes care of such scmds on EH entry.
496 * Note that ATA_QCFLAG_FAILED is unconditionally set after
497 * this function completes.
499 scsi_req_abort_cmd(qc
->scsicmd
);
503 * ata_port_schedule_eh - schedule error handling without a qc
504 * @ap: ATA port to schedule EH for
506 * Schedule error handling for @ap. EH will kick in as soon as
507 * all commands are drained.
510 * spin_lock_irqsave(host_set lock)
512 void ata_port_schedule_eh(struct ata_port
*ap
)
514 WARN_ON(!ap
->ops
->error_handler
);
516 ap
->flags
|= ATA_FLAG_EH_PENDING
;
517 scsi_schedule_eh(ap
->host
);
519 DPRINTK("port EH scheduled\n");
523 * ata_port_abort - abort all qc's on the port
524 * @ap: ATA port to abort qc's for
526 * Abort all active qc's of @ap and schedule EH.
529 * spin_lock_irqsave(host_set lock)
532 * Number of aborted qc's.
534 int ata_port_abort(struct ata_port
*ap
)
536 int tag
, nr_aborted
= 0;
538 WARN_ON(!ap
->ops
->error_handler
);
540 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
541 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
544 qc
->flags
|= ATA_QCFLAG_FAILED
;
551 ata_port_schedule_eh(ap
);
557 * __ata_port_freeze - freeze port
558 * @ap: ATA port to freeze
560 * This function is called when HSM violation or some other
561 * condition disrupts normal operation of the port. Frozen port
562 * is not allowed to perform any operation until the port is
563 * thawed, which usually follows a successful reset.
565 * ap->ops->freeze() callback can be used for freezing the port
566 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
567 * port cannot be frozen hardware-wise, the interrupt handler
568 * must ack and clear interrupts unconditionally while the port
572 * spin_lock_irqsave(host_set lock)
574 static void __ata_port_freeze(struct ata_port
*ap
)
576 WARN_ON(!ap
->ops
->error_handler
);
581 ap
->flags
|= ATA_FLAG_FROZEN
;
583 DPRINTK("ata%u port frozen\n", ap
->id
);
587 * ata_port_freeze - abort & freeze port
588 * @ap: ATA port to freeze
590 * Abort and freeze @ap.
593 * spin_lock_irqsave(host_set lock)
596 * Number of aborted commands.
598 int ata_port_freeze(struct ata_port
*ap
)
602 WARN_ON(!ap
->ops
->error_handler
);
604 nr_aborted
= ata_port_abort(ap
);
605 __ata_port_freeze(ap
);
611 * ata_eh_freeze_port - EH helper to freeze port
612 * @ap: ATA port to freeze
619 void ata_eh_freeze_port(struct ata_port
*ap
)
623 if (!ap
->ops
->error_handler
)
626 spin_lock_irqsave(ap
->lock
, flags
);
627 __ata_port_freeze(ap
);
628 spin_unlock_irqrestore(ap
->lock
, flags
);
632 * ata_port_thaw_port - EH helper to thaw port
633 * @ap: ATA port to thaw
635 * Thaw frozen port @ap.
640 void ata_eh_thaw_port(struct ata_port
*ap
)
644 if (!ap
->ops
->error_handler
)
647 spin_lock_irqsave(ap
->lock
, flags
);
649 ap
->flags
&= ~ATA_FLAG_FROZEN
;
654 spin_unlock_irqrestore(ap
->lock
, flags
);
656 DPRINTK("ata%u port thawed\n", ap
->id
);
659 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
664 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
666 struct ata_port
*ap
= qc
->ap
;
667 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
670 spin_lock_irqsave(ap
->lock
, flags
);
671 qc
->scsidone
= ata_eh_scsidone
;
672 __ata_qc_complete(qc
);
673 WARN_ON(ata_tag_valid(qc
->tag
));
674 spin_unlock_irqrestore(ap
->lock
, flags
);
676 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
680 * ata_eh_qc_complete - Complete an active ATA command from EH
681 * @qc: Command to complete
683 * Indicate to the mid and upper layers that an ATA command has
684 * completed. To be used from EH.
686 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
688 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
689 scmd
->retries
= scmd
->allowed
;
690 __ata_eh_qc_complete(qc
);
694 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
695 * @qc: Command to retry
697 * Indicate to the mid and upper layers that an ATA command
698 * should be retried. To be used from EH.
700 * SCSI midlayer limits the number of retries to scmd->allowed.
701 * scmd->retries is decremented for commands which get retried
702 * due to unrelated failures (qc->err_mask is zero).
704 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
706 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
707 if (!qc
->err_mask
&& scmd
->retries
)
709 __ata_eh_qc_complete(qc
);
713 * ata_eh_detach_dev - detach ATA device
714 * @dev: ATA device to detach
721 static void ata_eh_detach_dev(struct ata_device
*dev
)
723 struct ata_port
*ap
= dev
->ap
;
726 ata_dev_disable(dev
);
728 spin_lock_irqsave(ap
->lock
, flags
);
730 dev
->flags
&= ~ATA_DFLAG_DETACH
;
732 if (ata_scsi_offline_dev(dev
)) {
733 dev
->flags
|= ATA_DFLAG_DETACHED
;
734 ap
->flags
|= ATA_FLAG_SCSI_HOTPLUG
;
737 spin_unlock_irqrestore(ap
->lock
, flags
);
741 * ata_eh_about_to_do - about to perform eh_action
742 * @ap: target ATA port
743 * @dev: target ATA dev for per-dev action (can be NULL)
744 * @action: action about to be performed
746 * Called just before performing EH actions to clear related bits
747 * in @ap->eh_info such that eh actions are not unnecessarily
753 static void ata_eh_about_to_do(struct ata_port
*ap
, struct ata_device
*dev
,
758 spin_lock_irqsave(ap
->lock
, flags
);
759 ata_eh_clear_action(dev
, &ap
->eh_info
, action
);
760 ap
->flags
|= ATA_FLAG_RECOVERED
;
761 spin_unlock_irqrestore(ap
->lock
, flags
);
765 * ata_eh_done - EH action complete
766 * @ap: target ATA port
767 * @dev: target ATA dev for per-dev action (can be NULL)
768 * @action: action just completed
770 * Called right after performing EH actions to clear related bits
771 * in @ap->eh_context.
776 static void ata_eh_done(struct ata_port
*ap
, struct ata_device
*dev
,
779 ata_eh_clear_action(dev
, &ap
->eh_context
.i
, action
);
783 * ata_err_string - convert err_mask to descriptive string
784 * @err_mask: error mask to convert to string
786 * Convert @err_mask to descriptive string. Errors are
787 * prioritized according to severity and only the most severe
794 * Descriptive string for @err_mask
796 static const char * ata_err_string(unsigned int err_mask
)
798 if (err_mask
& AC_ERR_HOST_BUS
)
799 return "host bus error";
800 if (err_mask
& AC_ERR_ATA_BUS
)
801 return "ATA bus error";
802 if (err_mask
& AC_ERR_TIMEOUT
)
804 if (err_mask
& AC_ERR_HSM
)
805 return "HSM violation";
806 if (err_mask
& AC_ERR_SYSTEM
)
807 return "internal error";
808 if (err_mask
& AC_ERR_MEDIA
)
809 return "media error";
810 if (err_mask
& AC_ERR_INVALID
)
811 return "invalid argument";
812 if (err_mask
& AC_ERR_DEV
)
813 return "device error";
814 return "unknown error";
818 * ata_read_log_page - read a specific log page
819 * @dev: target device
820 * @page: page to read
821 * @buf: buffer to store read page
822 * @sectors: number of sectors to read
824 * Read log page using READ_LOG_EXT command.
827 * Kernel thread context (may sleep).
830 * 0 on success, AC_ERR_* mask otherwise.
832 static unsigned int ata_read_log_page(struct ata_device
*dev
,
833 u8 page
, void *buf
, unsigned int sectors
)
835 struct ata_taskfile tf
;
836 unsigned int err_mask
;
838 DPRINTK("read log page - page %d\n", page
);
840 ata_tf_init(dev
, &tf
);
841 tf
.command
= ATA_CMD_READ_LOG_EXT
;
844 tf
.hob_nsect
= sectors
>> 8;
845 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
846 tf
.protocol
= ATA_PROT_PIO
;
848 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
849 buf
, sectors
* ATA_SECT_SIZE
);
851 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
856 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
857 * @dev: Device to read log page 10h from
858 * @tag: Resulting tag of the failed command
859 * @tf: Resulting taskfile registers of the failed command
861 * Read log page 10h to obtain NCQ error details and clear error
865 * Kernel thread context (may sleep).
868 * 0 on success, -errno otherwise.
870 static int ata_eh_read_log_10h(struct ata_device
*dev
,
871 int *tag
, struct ata_taskfile
*tf
)
873 u8
*buf
= dev
->ap
->sector_buf
;
874 unsigned int err_mask
;
878 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
883 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
886 ata_dev_printk(dev
, KERN_WARNING
,
887 "invalid checksum 0x%x on log page 10h\n", csum
);
892 *tag
= buf
[0] & 0x1f;
894 tf
->command
= buf
[2];
895 tf
->feature
= buf
[3];
900 tf
->hob_lbal
= buf
[8];
901 tf
->hob_lbam
= buf
[9];
902 tf
->hob_lbah
= buf
[10];
904 tf
->hob_nsect
= buf
[13];
910 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
911 * @dev: device to perform REQUEST_SENSE to
912 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
914 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
915 * SENSE. This function is EH helper.
918 * Kernel thread context (may sleep).
921 * 0 on success, AC_ERR_* mask on failure
923 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
924 unsigned char *sense_buf
)
926 struct ata_port
*ap
= dev
->ap
;
927 struct ata_taskfile tf
;
928 u8 cdb
[ATAPI_CDB_LEN
];
930 DPRINTK("ATAPI request sense\n");
932 ata_tf_init(dev
, &tf
);
934 /* FIXME: is this needed? */
935 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
937 /* XXX: why tf_read here? */
938 ap
->ops
->tf_read(ap
, &tf
);
940 /* fill these in, for the case where they are -not- overwritten */
942 sense_buf
[2] = tf
.feature
>> 4;
944 memset(cdb
, 0, ATAPI_CDB_LEN
);
945 cdb
[0] = REQUEST_SENSE
;
946 cdb
[4] = SCSI_SENSE_BUFFERSIZE
;
948 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
949 tf
.command
= ATA_CMD_PACKET
;
951 /* is it pointless to prefer PIO for "safety reasons"? */
952 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
953 tf
.protocol
= ATA_PROT_ATAPI_DMA
;
954 tf
.feature
|= ATAPI_PKT_DMA
;
956 tf
.protocol
= ATA_PROT_ATAPI
;
957 tf
.lbam
= (8 * 1024) & 0xff;
958 tf
.lbah
= (8 * 1024) >> 8;
961 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
962 sense_buf
, SCSI_SENSE_BUFFERSIZE
);
966 * ata_eh_analyze_serror - analyze SError for a failed port
967 * @ap: ATA port to analyze SError for
969 * Analyze SError if available and further determine cause of
975 static void ata_eh_analyze_serror(struct ata_port
*ap
)
977 struct ata_eh_context
*ehc
= &ap
->eh_context
;
978 u32 serror
= ehc
->i
.serror
;
979 unsigned int err_mask
= 0, action
= 0;
981 if (serror
& SERR_PERSISTENT
) {
982 err_mask
|= AC_ERR_ATA_BUS
;
983 action
|= ATA_EH_HARDRESET
;
986 (SERR_DATA_RECOVERED
| SERR_COMM_RECOVERED
| SERR_DATA
)) {
987 err_mask
|= AC_ERR_ATA_BUS
;
988 action
|= ATA_EH_SOFTRESET
;
990 if (serror
& SERR_PROTOCOL
) {
991 err_mask
|= AC_ERR_HSM
;
992 action
|= ATA_EH_SOFTRESET
;
994 if (serror
& SERR_INTERNAL
) {
995 err_mask
|= AC_ERR_SYSTEM
;
996 action
|= ATA_EH_SOFTRESET
;
998 if (serror
& (SERR_PHYRDY_CHG
| SERR_DEV_XCHG
))
999 ata_ehi_hotplugged(&ehc
->i
);
1001 ehc
->i
.err_mask
|= err_mask
;
1002 ehc
->i
.action
|= action
;
1006 * ata_eh_analyze_ncq_error - analyze NCQ error
1007 * @ap: ATA port to analyze NCQ error for
1009 * Read log page 10h, determine the offending qc and acquire
1010 * error status TF. For NCQ device errors, all LLDDs have to do
1011 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1015 * Kernel thread context (may sleep).
1017 static void ata_eh_analyze_ncq_error(struct ata_port
*ap
)
1019 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1020 struct ata_device
*dev
= ap
->device
;
1021 struct ata_queued_cmd
*qc
;
1022 struct ata_taskfile tf
;
1025 /* if frozen, we can't do much */
1026 if (ap
->flags
& ATA_FLAG_FROZEN
)
1029 /* is it NCQ device error? */
1030 if (!ap
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1033 /* has LLDD analyzed already? */
1034 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1035 qc
= __ata_qc_from_tag(ap
, tag
);
1037 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1044 /* okay, this error is ours */
1045 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1047 ata_port_printk(ap
, KERN_ERR
, "failed to read log page 10h "
1048 "(errno=%d)\n", rc
);
1052 if (!(ap
->sactive
& (1 << tag
))) {
1053 ata_port_printk(ap
, KERN_ERR
, "log page 10h reported "
1054 "inactive tag %d\n", tag
);
1058 /* we've got the perpetrator, condemn it */
1059 qc
= __ata_qc_from_tag(ap
, tag
);
1060 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1061 qc
->err_mask
|= AC_ERR_DEV
;
1062 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1066 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1067 * @qc: qc to analyze
1068 * @tf: Taskfile registers to analyze
1070 * Analyze taskfile of @qc and further determine cause of
1071 * failure. This function also requests ATAPI sense data if
1075 * Kernel thread context (may sleep).
1078 * Determined recovery action
1080 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1081 const struct ata_taskfile
*tf
)
1083 unsigned int tmp
, action
= 0;
1084 u8 stat
= tf
->command
, err
= tf
->feature
;
1086 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1087 qc
->err_mask
|= AC_ERR_HSM
;
1088 return ATA_EH_SOFTRESET
;
1091 if (!(qc
->err_mask
& AC_ERR_DEV
))
1094 switch (qc
->dev
->class) {
1097 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1099 qc
->err_mask
|= AC_ERR_MEDIA
;
1101 qc
->err_mask
|= AC_ERR_INVALID
;
1105 tmp
= atapi_eh_request_sense(qc
->dev
,
1106 qc
->scsicmd
->sense_buffer
);
1108 /* ATA_QCFLAG_SENSE_VALID is used to tell
1109 * atapi_qc_complete() that sense data is
1112 * TODO: interpret sense data and set
1113 * appropriate err_mask.
1115 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1117 qc
->err_mask
|= tmp
;
1120 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1121 action
|= ATA_EH_SOFTRESET
;
1126 static int ata_eh_categorize_ering_entry(struct ata_ering_entry
*ent
)
1128 if (ent
->err_mask
& (AC_ERR_ATA_BUS
| AC_ERR_TIMEOUT
))
1132 if (ent
->err_mask
& AC_ERR_HSM
)
1134 if ((ent
->err_mask
&
1135 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1142 struct speed_down_needed_arg
{
1147 static int speed_down_needed_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1149 struct speed_down_needed_arg
*arg
= void_arg
;
1151 if (ent
->timestamp
< arg
->since
)
1154 arg
->nr_errors
[ata_eh_categorize_ering_entry(ent
)]++;
1159 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
1160 * @dev: Device of interest
1162 * This function examines error ring of @dev and determines
1163 * whether speed down is necessary. Speed down is necessary if
1164 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
1165 * errors during last 15 minutes.
1167 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
1168 * violation for known supported commands.
1170 * Cat-2 errors are unclassified DEV error for known supported
1174 * Inherited from caller.
1177 * 1 if speed down is necessary, 0 otherwise
1179 static int ata_eh_speed_down_needed(struct ata_device
*dev
)
1181 const u64 interval
= 15LLU * 60 * HZ
;
1182 static const int err_limits
[3] = { -1, 3, 10 };
1183 struct speed_down_needed_arg arg
;
1184 struct ata_ering_entry
*ent
;
1188 ent
= ata_ering_top(&dev
->ering
);
1192 err_cat
= ata_eh_categorize_ering_entry(ent
);
1196 memset(&arg
, 0, sizeof(arg
));
1198 j64
= get_jiffies_64();
1199 if (j64
>= interval
)
1200 arg
.since
= j64
- interval
;
1204 ata_ering_map(&dev
->ering
, speed_down_needed_cb
, &arg
);
1206 return arg
.nr_errors
[err_cat
] > err_limits
[err_cat
];
1210 * ata_eh_speed_down - record error and speed down if necessary
1211 * @dev: Failed device
1212 * @is_io: Did the device fail during normal IO?
1213 * @err_mask: err_mask of the error
1215 * Record error and examine error history to determine whether
1216 * adjusting transmission speed is necessary. It also sets
1217 * transmission limits appropriately if such adjustment is
1221 * Kernel thread context (may sleep).
1224 * 0 on success, -errno otherwise
1226 static int ata_eh_speed_down(struct ata_device
*dev
, int is_io
,
1227 unsigned int err_mask
)
1232 /* record error and determine whether speed down is necessary */
1233 ata_ering_record(&dev
->ering
, is_io
, err_mask
);
1235 if (!ata_eh_speed_down_needed(dev
))
1238 /* speed down SATA link speed if possible */
1239 if (sata_down_spd_limit(dev
->ap
) == 0)
1240 return ATA_EH_HARDRESET
;
1242 /* lower transfer mode */
1243 if (ata_down_xfermask_limit(dev
, 0) == 0)
1244 return ATA_EH_SOFTRESET
;
1246 ata_dev_printk(dev
, KERN_ERR
,
1247 "speed down requested but no transfer mode left\n");
1252 * ata_eh_autopsy - analyze error and determine recovery action
1253 * @ap: ATA port to perform autopsy on
1255 * Analyze why @ap failed and determine which recovery action is
1256 * needed. This function also sets more detailed AC_ERR_* values
1257 * and fills sense data for ATAPI CHECK SENSE.
1260 * Kernel thread context (may sleep).
1262 static void ata_eh_autopsy(struct ata_port
*ap
)
1264 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1265 unsigned int action
= ehc
->i
.action
;
1266 struct ata_device
*failed_dev
= NULL
;
1267 unsigned int all_err_mask
= 0;
1274 /* obtain and analyze SError */
1275 rc
= sata_scr_read(ap
, SCR_ERROR
, &serror
);
1277 ehc
->i
.serror
|= serror
;
1278 ata_eh_analyze_serror(ap
);
1279 } else if (rc
!= -EOPNOTSUPP
)
1280 action
|= ATA_EH_HARDRESET
;
1282 /* analyze NCQ failure */
1283 ata_eh_analyze_ncq_error(ap
);
1285 /* any real error trumps AC_ERR_OTHER */
1286 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1287 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1289 all_err_mask
|= ehc
->i
.err_mask
;
1291 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1292 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1294 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1297 /* inherit upper level err_mask */
1298 qc
->err_mask
|= ehc
->i
.err_mask
;
1301 action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1303 /* DEV errors are probably spurious in case of ATA_BUS error */
1304 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1305 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1308 /* any real error trumps unknown error */
1309 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1310 qc
->err_mask
&= ~AC_ERR_OTHER
;
1312 /* SENSE_VALID trumps dev/unknown error and revalidation */
1313 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1314 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1315 action
&= ~ATA_EH_REVALIDATE
;
1318 /* accumulate error info */
1319 failed_dev
= qc
->dev
;
1320 all_err_mask
|= qc
->err_mask
;
1321 if (qc
->flags
& ATA_QCFLAG_IO
)
1325 /* enforce default EH actions */
1326 if (ap
->flags
& ATA_FLAG_FROZEN
||
1327 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
1328 action
|= ATA_EH_SOFTRESET
;
1329 else if (all_err_mask
)
1330 action
|= ATA_EH_REVALIDATE
;
1332 /* if we have offending qcs and the associated failed device */
1335 action
|= ata_eh_speed_down(failed_dev
, is_io
, all_err_mask
);
1337 /* perform per-dev EH action only on the offending device */
1338 ehc
->i
.dev_action
[failed_dev
->devno
] |=
1339 action
& ATA_EH_PERDEV_MASK
;
1340 action
&= ~ATA_EH_PERDEV_MASK
;
1343 /* record autopsy result */
1344 ehc
->i
.dev
= failed_dev
;
1345 ehc
->i
.action
= action
;
1351 * ata_eh_report - report error handling to user
1352 * @ap: ATA port EH is going on
1354 * Report EH to user.
1359 static void ata_eh_report(struct ata_port
*ap
)
1361 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1362 const char *frozen
, *desc
;
1363 int tag
, nr_failed
= 0;
1366 if (ehc
->i
.desc
[0] != '\0')
1369 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1370 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1372 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1374 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
1380 if (!nr_failed
&& !ehc
->i
.err_mask
)
1384 if (ap
->flags
& ATA_FLAG_FROZEN
)
1388 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
1389 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1390 ehc
->i
.err_mask
, ap
->sactive
, ehc
->i
.serror
,
1391 ehc
->i
.action
, frozen
);
1393 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "(%s)\n", desc
);
1395 ata_port_printk(ap
, KERN_ERR
, "exception Emask 0x%x "
1396 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1397 ehc
->i
.err_mask
, ap
->sactive
, ehc
->i
.serror
,
1398 ehc
->i
.action
, frozen
);
1400 ata_port_printk(ap
, KERN_ERR
, "(%s)\n", desc
);
1403 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1404 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1406 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || !qc
->err_mask
)
1409 ata_dev_printk(qc
->dev
, KERN_ERR
, "tag %d cmd 0x%x "
1410 "Emask 0x%x stat 0x%x err 0x%x (%s)\n",
1411 qc
->tag
, qc
->tf
.command
, qc
->err_mask
,
1412 qc
->result_tf
.command
, qc
->result_tf
.feature
,
1413 ata_err_string(qc
->err_mask
));
1417 static int ata_do_reset(struct ata_port
*ap
, ata_reset_fn_t reset
,
1418 unsigned int *classes
)
1422 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1423 classes
[i
] = ATA_DEV_UNKNOWN
;
1425 rc
= reset(ap
, classes
);
1429 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1430 * is complete and convert all ATA_DEV_UNKNOWN to
1433 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1434 if (classes
[i
] != ATA_DEV_UNKNOWN
)
1437 if (i
< ATA_MAX_DEVICES
)
1438 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1439 if (classes
[i
] == ATA_DEV_UNKNOWN
)
1440 classes
[i
] = ATA_DEV_NONE
;
1445 static int ata_eh_followup_srst_needed(int rc
, int classify
,
1446 const unsigned int *classes
)
1452 if (classify
&& classes
[0] == ATA_DEV_UNKNOWN
)
1457 static int ata_eh_reset(struct ata_port
*ap
, int classify
,
1458 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
1459 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
1461 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1462 unsigned int *classes
= ehc
->classes
;
1463 int tries
= ATA_EH_RESET_TRIES
;
1464 int verbose
= !(ap
->flags
& ATA_FLAG_LOADING
);
1465 unsigned int action
;
1466 ata_reset_fn_t reset
;
1467 int i
, did_followup_srst
, rc
;
1469 /* Determine which reset to use and record in ehc->i.action.
1470 * prereset() may examine and modify it.
1472 action
= ehc
->i
.action
;
1473 ehc
->i
.action
&= ~ATA_EH_RESET_MASK
;
1474 if (softreset
&& (!hardreset
|| (!sata_set_spd_needed(ap
) &&
1475 !(action
& ATA_EH_HARDRESET
))))
1476 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1478 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1483 ata_port_printk(ap
, KERN_ERR
,
1484 "prereset failed (errno=%d)\n", rc
);
1489 /* prereset() might have modified ehc->i.action */
1490 if (ehc
->i
.action
& ATA_EH_HARDRESET
)
1492 else if (ehc
->i
.action
& ATA_EH_SOFTRESET
)
1495 /* prereset told us not to reset, bang classes and return */
1496 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1497 classes
[i
] = ATA_DEV_NONE
;
1501 /* did prereset() screw up? if so, fix up to avoid oopsing */
1503 ata_port_printk(ap
, KERN_ERR
, "BUG: prereset() requested "
1504 "invalid reset type\n");
1512 /* shut up during boot probing */
1514 ata_port_printk(ap
, KERN_INFO
, "%s resetting port\n",
1515 reset
== softreset
? "soft" : "hard");
1518 ata_eh_about_to_do(ap
, NULL
, ATA_EH_RESET_MASK
);
1519 ehc
->i
.flags
|= ATA_EHI_DID_RESET
;
1521 rc
= ata_do_reset(ap
, reset
, classes
);
1523 did_followup_srst
= 0;
1524 if (reset
== hardreset
&&
1525 ata_eh_followup_srst_needed(rc
, classify
, classes
)) {
1526 /* okay, let's do follow-up softreset */
1527 did_followup_srst
= 1;
1531 ata_port_printk(ap
, KERN_ERR
,
1532 "follow-up softreset required "
1533 "but no softreset avaliable\n");
1537 ata_eh_about_to_do(ap
, NULL
, ATA_EH_RESET_MASK
);
1538 rc
= ata_do_reset(ap
, reset
, classes
);
1540 if (rc
== 0 && classify
&&
1541 classes
[0] == ATA_DEV_UNKNOWN
) {
1542 ata_port_printk(ap
, KERN_ERR
,
1543 "classification failed\n");
1548 if (rc
&& --tries
) {
1551 if (reset
== softreset
) {
1552 if (did_followup_srst
)
1553 type
= "follow-up soft";
1559 ata_port_printk(ap
, KERN_WARNING
,
1560 "%sreset failed, retrying in 5 secs\n", type
);
1563 if (reset
== hardreset
)
1564 sata_down_spd_limit(ap
);
1571 /* After the reset, the device state is PIO 0 and the
1572 * controller state is undefined. Record the mode.
1574 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1575 ap
->device
[i
].pio_mode
= XFER_PIO_0
;
1578 postreset(ap
, classes
);
1580 /* reset successful, schedule revalidation */
1581 ata_eh_done(ap
, NULL
, ATA_EH_RESET_MASK
);
1582 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1588 static int ata_eh_revalidate_and_attach(struct ata_port
*ap
,
1589 struct ata_device
**r_failed_dev
)
1591 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1592 struct ata_device
*dev
;
1593 unsigned long flags
;
1598 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1599 unsigned int action
;
1601 dev
= &ap
->device
[i
];
1602 action
= ata_eh_dev_action(dev
);
1604 if (action
& ATA_EH_REVALIDATE
&& ata_dev_enabled(dev
)) {
1605 if (ata_port_offline(ap
)) {
1610 ata_eh_about_to_do(ap
, dev
, ATA_EH_REVALIDATE
);
1611 rc
= ata_dev_revalidate(dev
,
1612 ehc
->i
.flags
& ATA_EHI_DID_RESET
);
1616 ata_eh_done(ap
, dev
, ATA_EH_REVALIDATE
);
1618 /* schedule the scsi_rescan_device() here */
1619 queue_work(ata_aux_wq
, &(ap
->scsi_rescan_task
));
1620 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
1621 ehc
->tries
[dev
->devno
] &&
1622 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
1623 dev
->class = ehc
->classes
[dev
->devno
];
1625 rc
= ata_dev_read_id(dev
, &dev
->class, 1, dev
->id
);
1627 rc
= ata_dev_configure(dev
, 1);
1630 dev
->class = ATA_DEV_UNKNOWN
;
1634 spin_lock_irqsave(ap
->lock
, flags
);
1635 ap
->flags
|= ATA_FLAG_SCSI_HOTPLUG
;
1636 spin_unlock_irqrestore(ap
->lock
, flags
);
1641 *r_failed_dev
= dev
;
1647 static int ata_port_nr_enabled(struct ata_port
*ap
)
1651 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1652 if (ata_dev_enabled(&ap
->device
[i
]))
1657 static int ata_port_nr_vacant(struct ata_port
*ap
)
1661 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1662 if (ap
->device
[i
].class == ATA_DEV_UNKNOWN
)
1667 static int ata_eh_skip_recovery(struct ata_port
*ap
)
1669 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1672 if (ap
->flags
& ATA_FLAG_FROZEN
|| ata_port_nr_enabled(ap
))
1675 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1676 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1677 struct ata_device
*dev
= &ap
->device
[i
];
1679 if (dev
->class == ATA_DEV_UNKNOWN
&&
1680 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
1688 * ata_eh_recover - recover host port after error
1689 * @ap: host port to recover
1690 * @prereset: prereset method (can be NULL)
1691 * @softreset: softreset method (can be NULL)
1692 * @hardreset: hardreset method (can be NULL)
1693 * @postreset: postreset method (can be NULL)
1695 * This is the alpha and omega, eum and yang, heart and soul of
1696 * libata exception handling. On entry, actions required to
1697 * recover the port and hotplug requests are recorded in
1698 * eh_context. This function executes all the operations with
1699 * appropriate retrials and fallbacks to resurrect failed
1700 * devices, detach goners and greet newcomers.
1703 * Kernel thread context (may sleep).
1706 * 0 on success, -errno on failure.
1708 static int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
1709 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
1710 ata_postreset_fn_t postreset
)
1712 struct ata_eh_context
*ehc
= &ap
->eh_context
;
1713 struct ata_device
*dev
;
1714 int down_xfermask
, i
, rc
;
1718 /* prep for recovery */
1719 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++) {
1720 dev
= &ap
->device
[i
];
1722 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
1724 /* process hotplug request */
1725 if (dev
->flags
& ATA_DFLAG_DETACH
)
1726 ata_eh_detach_dev(dev
);
1728 if (!ata_dev_enabled(dev
) &&
1729 ((ehc
->i
.probe_mask
& (1 << dev
->devno
)) &&
1730 !(ehc
->did_probe_mask
& (1 << dev
->devno
)))) {
1731 ata_eh_detach_dev(dev
);
1733 ehc
->did_probe_mask
|= (1 << dev
->devno
);
1734 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1742 /* if UNLOADING, finish immediately */
1743 if (ap
->flags
& ATA_FLAG_UNLOADING
)
1746 /* skip EH if possible. */
1747 if (ata_eh_skip_recovery(ap
))
1750 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1751 ehc
->classes
[i
] = ATA_DEV_UNKNOWN
;
1754 if (ehc
->i
.action
& ATA_EH_RESET_MASK
) {
1755 ata_eh_freeze_port(ap
);
1757 rc
= ata_eh_reset(ap
, ata_port_nr_vacant(ap
), prereset
,
1758 softreset
, hardreset
, postreset
);
1760 ata_port_printk(ap
, KERN_ERR
,
1761 "reset failed, giving up\n");
1765 ata_eh_thaw_port(ap
);
1768 /* revalidate existing devices and attach new ones */
1769 rc
= ata_eh_revalidate_and_attach(ap
, &dev
);
1773 /* configure transfer mode if the port has been reset */
1774 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
1775 rc
= ata_set_mode(ap
, &dev
);
1787 /* device missing, schedule probing */
1788 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
1790 ehc
->tries
[dev
->devno
] = 0;
1793 sata_down_spd_limit(ap
);
1795 ehc
->tries
[dev
->devno
]--;
1796 if (down_xfermask
&&
1797 ata_down_xfermask_limit(dev
, ehc
->tries
[dev
->devno
] == 1))
1798 ehc
->tries
[dev
->devno
] = 0;
1801 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
1802 /* disable device if it has used up all its chances */
1803 ata_dev_disable(dev
);
1805 /* detach if offline */
1806 if (ata_port_offline(ap
))
1807 ata_eh_detach_dev(dev
);
1809 /* probe if requested */
1810 if ((ehc
->i
.probe_mask
& (1 << dev
->devno
)) &&
1811 !(ehc
->did_probe_mask
& (1 << dev
->devno
))) {
1812 ata_eh_detach_dev(dev
);
1815 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
1816 ehc
->did_probe_mask
|= (1 << dev
->devno
);
1817 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1820 /* soft didn't work? be haaaaard */
1821 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
1822 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1824 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1827 if (ata_port_nr_enabled(ap
)) {
1828 ata_port_printk(ap
, KERN_WARNING
, "failed to recover some "
1829 "devices, retrying in 5 secs\n");
1832 /* no device left, repeat fast */
1840 for (i
= 0; i
< ATA_MAX_DEVICES
; i
++)
1841 ata_dev_disable(&ap
->device
[i
]);
1844 DPRINTK("EXIT, rc=%d\n", rc
);
1849 * ata_eh_finish - finish up EH
1850 * @ap: host port to finish EH for
1852 * Recovery is complete. Clean up EH states and retry or finish
1858 static void ata_eh_finish(struct ata_port
*ap
)
1862 /* retry or finish qcs */
1863 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1864 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1866 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1870 /* FIXME: Once EH migration is complete,
1871 * generate sense data in this function,
1872 * considering both err_mask and tf.
1874 if (qc
->err_mask
& AC_ERR_INVALID
)
1875 ata_eh_qc_complete(qc
);
1877 ata_eh_qc_retry(qc
);
1879 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1880 ata_eh_qc_complete(qc
);
1882 /* feed zero TF to sense generation */
1883 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
1884 ata_eh_qc_retry(qc
);
1891 * ata_do_eh - do standard error handling
1892 * @ap: host port to handle error for
1893 * @prereset: prereset method (can be NULL)
1894 * @softreset: softreset method (can be NULL)
1895 * @hardreset: hardreset method (can be NULL)
1896 * @postreset: postreset method (can be NULL)
1898 * Perform standard error handling sequence.
1901 * Kernel thread context (may sleep).
1903 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
1904 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
1905 ata_postreset_fn_t postreset
)
1907 if (!(ap
->flags
& ATA_FLAG_LOADING
)) {
1912 ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
);