2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include <scsi/scsi_dbg.h>
44 #include "../scsi/scsi_transport_api.h"
46 #include <linux/libata.h>
51 /* speed down verdicts */
52 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
55 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
58 ATA_EFLAG_IS_IO
= (1 << 0),
59 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
60 ATA_EFLAG_OLD_ER
= (1 << 31),
62 /* error categories */
65 ATA_ECAT_TOUT_HSM
= 2,
67 ATA_ECAT_DUBIOUS_NONE
= 4,
68 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
69 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
70 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
73 ATA_EH_CMD_DFL_TIMEOUT
= 5000,
75 /* always put at least this amount of time between resets */
76 ATA_EH_RESET_COOL_DOWN
= 5000,
78 /* Waiting in ->prereset can never be reliable. It's
79 * sometimes nice to wait there but it can't be depended upon;
80 * otherwise, we wouldn't be resetting. Just give it enough
81 * time for most drives to spin up.
83 ATA_EH_PRERESET_TIMEOUT
= 10000,
84 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
88 /* probe speed down parameters, see ata_eh_schedule_probe() */
89 ATA_EH_PROBE_TRIAL_INTERVAL
= 60000, /* 1 min */
90 ATA_EH_PROBE_TRIALS
= 2,
93 /* The following table determines how we sequence resets. Each entry
94 * represents timeout for that try. The first try can be soft or
95 * hardreset. All others are hardreset if available. In most cases
96 * the first reset w/ 10sec timeout should succeed. Following entries
97 * are mostly for error handling, hotplug and retarded devices.
99 static const unsigned long ata_eh_reset_timeouts
[] = {
100 10000, /* most drives spin up by 10sec */
101 10000, /* > 99% working drives spin up before 20sec */
102 35000, /* give > 30 secs of idleness for retarded devices */
103 5000, /* and sweet one last chance */
104 ULONG_MAX
, /* > 1 min has elapsed, give up */
107 static const unsigned long ata_eh_identify_timeouts
[] = {
108 5000, /* covers > 99% of successes and not too boring on failures */
109 10000, /* combined time till here is enough even for media access */
110 30000, /* for true idiots */
114 static const unsigned long ata_eh_flush_timeouts
[] = {
115 15000, /* be generous with flush */
117 30000, /* and even more generous */
121 static const unsigned long ata_eh_other_timeouts
[] = {
122 5000, /* same rationale as identify timeout */
124 /* but no merciful 30sec for other commands, it just isn't worth it */
128 struct ata_eh_cmd_timeout_ent
{
130 const unsigned long *timeouts
;
133 /* The following table determines timeouts to use for EH internal
134 * commands. Each table entry is a command class and matches the
135 * commands the entry applies to and the timeout table to use.
137 * On the retry after a command timed out, the next timeout value from
138 * the table is used. If the table doesn't contain further entries,
139 * the last value is used.
141 * ehc->cmd_timeout_idx keeps track of which timeout to use per
142 * command class, so if SET_FEATURES times out on the first try, the
143 * next try will use the second timeout value only for that class.
145 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
146 static const struct ata_eh_cmd_timeout_ent
147 ata_eh_cmd_timeout_table
[ATA_EH_CMD_TIMEOUT_TABLE_SIZE
] = {
148 { .commands
= CMDS(ATA_CMD_ID_ATA
, ATA_CMD_ID_ATAPI
),
149 .timeouts
= ata_eh_identify_timeouts
, },
150 { .commands
= CMDS(ATA_CMD_READ_NATIVE_MAX
, ATA_CMD_READ_NATIVE_MAX_EXT
),
151 .timeouts
= ata_eh_other_timeouts
, },
152 { .commands
= CMDS(ATA_CMD_SET_MAX
, ATA_CMD_SET_MAX_EXT
),
153 .timeouts
= ata_eh_other_timeouts
, },
154 { .commands
= CMDS(ATA_CMD_SET_FEATURES
),
155 .timeouts
= ata_eh_other_timeouts
, },
156 { .commands
= CMDS(ATA_CMD_INIT_DEV_PARAMS
),
157 .timeouts
= ata_eh_other_timeouts
, },
158 { .commands
= CMDS(ATA_CMD_FLUSH
, ATA_CMD_FLUSH_EXT
),
159 .timeouts
= ata_eh_flush_timeouts
},
163 static void __ata_port_freeze(struct ata_port
*ap
);
165 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
166 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
167 #else /* CONFIG_PM */
168 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
171 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
173 #endif /* CONFIG_PM */
175 static void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
, const char *fmt
,
178 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
179 ATA_EH_DESC_LEN
- ehi
->desc_len
,
184 * __ata_ehi_push_desc - push error description without adding separator
186 * @fmt: printf format string
188 * Format string according to @fmt and append it to @ehi->desc.
191 * spin_lock_irqsave(host lock)
193 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
198 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
203 * ata_ehi_push_desc - push error description with separator
205 * @fmt: printf format string
207 * Format string according to @fmt and append it to @ehi->desc.
208 * If @ehi->desc is not empty, ", " is added in-between.
211 * spin_lock_irqsave(host lock)
213 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
218 __ata_ehi_push_desc(ehi
, ", ");
221 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
226 * ata_ehi_clear_desc - clean error description
232 * spin_lock_irqsave(host lock)
234 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
241 * ata_port_desc - append port description
242 * @ap: target ATA port
243 * @fmt: printf format string
245 * Format string according to @fmt and append it to port
246 * description. If port description is not empty, " " is added
247 * in-between. This function is to be used while initializing
248 * ata_host. The description is printed on host registration.
253 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
257 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
259 if (ap
->link
.eh_info
.desc_len
)
260 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
263 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
270 * ata_port_pbar_desc - append PCI BAR description
271 * @ap: target ATA port
272 * @bar: target PCI BAR
273 * @offset: offset into PCI BAR
274 * @name: name of the area
276 * If @offset is negative, this function formats a string which
277 * contains the name, address, size and type of the BAR and
278 * appends it to the port description. If @offset is zero or
279 * positive, only name and offsetted address is appended.
284 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
287 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
289 unsigned long long start
, len
;
291 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
293 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
296 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
297 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
300 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
302 ata_port_desc(ap
, "%s 0x%llx", name
,
303 start
+ (unsigned long long)offset
);
306 #endif /* CONFIG_PCI */
308 static int ata_lookup_timeout_table(u8 cmd
)
312 for (i
= 0; i
< ATA_EH_CMD_TIMEOUT_TABLE_SIZE
; i
++) {
315 for (cur
= ata_eh_cmd_timeout_table
[i
].commands
; *cur
; cur
++)
324 * ata_internal_cmd_timeout - determine timeout for an internal command
325 * @dev: target device
326 * @cmd: internal command to be issued
328 * Determine timeout for internal command @cmd for @dev.
334 * Determined timeout.
336 unsigned long ata_internal_cmd_timeout(struct ata_device
*dev
, u8 cmd
)
338 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
339 int ent
= ata_lookup_timeout_table(cmd
);
343 return ATA_EH_CMD_DFL_TIMEOUT
;
345 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
346 return ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
];
350 * ata_internal_cmd_timed_out - notification for internal command timeout
351 * @dev: target device
352 * @cmd: internal command which timed out
354 * Notify EH that internal command @cmd for @dev timed out. This
355 * function should be called only for commands whose timeouts are
356 * determined using ata_internal_cmd_timeout().
361 void ata_internal_cmd_timed_out(struct ata_device
*dev
, u8 cmd
)
363 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
364 int ent
= ata_lookup_timeout_table(cmd
);
370 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
371 if (ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
+ 1] != ULONG_MAX
)
372 ehc
->cmd_timeout_idx
[dev
->devno
][ent
]++;
375 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
376 unsigned int err_mask
)
378 struct ata_ering_entry
*ent
;
383 ering
->cursor
%= ATA_ERING_SIZE
;
385 ent
= &ering
->ring
[ering
->cursor
];
386 ent
->eflags
= eflags
;
387 ent
->err_mask
= err_mask
;
388 ent
->timestamp
= get_jiffies_64();
391 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
393 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
400 int ata_ering_map(struct ata_ering
*ering
,
401 int (*map_fn
)(struct ata_ering_entry
*, void *),
405 struct ata_ering_entry
*ent
;
409 ent
= &ering
->ring
[idx
];
412 rc
= map_fn(ent
, arg
);
415 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
416 } while (idx
!= ering
->cursor
);
421 int ata_ering_clear_cb(struct ata_ering_entry
*ent
, void *void_arg
)
423 ent
->eflags
|= ATA_EFLAG_OLD_ER
;
427 static void ata_ering_clear(struct ata_ering
*ering
)
429 ata_ering_map(ering
, ata_ering_clear_cb
, NULL
);
432 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
434 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
436 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
439 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
440 struct ata_eh_info
*ehi
, unsigned int action
)
442 struct ata_device
*tdev
;
445 ehi
->action
&= ~action
;
446 ata_for_each_dev(tdev
, link
, ALL
)
447 ehi
->dev_action
[tdev
->devno
] &= ~action
;
449 /* doesn't make sense for port-wide EH actions */
450 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
452 /* break ehi->action into ehi->dev_action */
453 if (ehi
->action
& action
) {
454 ata_for_each_dev(tdev
, link
, ALL
)
455 ehi
->dev_action
[tdev
->devno
] |=
456 ehi
->action
& action
;
457 ehi
->action
&= ~action
;
460 /* turn off the specified per-dev action */
461 ehi
->dev_action
[dev
->devno
] &= ~action
;
466 * ata_scsi_timed_out - SCSI layer time out callback
467 * @cmd: timed out SCSI command
469 * Handles SCSI layer timeout. We race with normal completion of
470 * the qc for @cmd. If the qc is already gone, we lose and let
471 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
472 * timed out and EH should be invoked. Prevent ata_qc_complete()
473 * from finishing it by setting EH_SCHEDULED and return
476 * TODO: kill this function once old EH is gone.
479 * Called from timer context
482 * EH_HANDLED or EH_NOT_HANDLED
484 enum blk_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
486 struct Scsi_Host
*host
= cmd
->device
->host
;
487 struct ata_port
*ap
= ata_shost_to_port(host
);
489 struct ata_queued_cmd
*qc
;
490 enum blk_eh_timer_return ret
;
494 if (ap
->ops
->error_handler
) {
495 ret
= BLK_EH_NOT_HANDLED
;
499 ret
= BLK_EH_HANDLED
;
500 spin_lock_irqsave(ap
->lock
, flags
);
501 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
503 WARN_ON(qc
->scsicmd
!= cmd
);
504 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
505 qc
->err_mask
|= AC_ERR_TIMEOUT
;
506 ret
= BLK_EH_NOT_HANDLED
;
508 spin_unlock_irqrestore(ap
->lock
, flags
);
511 DPRINTK("EXIT, ret=%d\n", ret
);
515 static void ata_eh_unload(struct ata_port
*ap
)
517 struct ata_link
*link
;
518 struct ata_device
*dev
;
521 /* Restore SControl IPM and SPD for the next driver and
522 * disable attached devices.
524 ata_for_each_link(link
, ap
, PMP_FIRST
) {
525 sata_scr_write(link
, SCR_CONTROL
, link
->saved_scontrol
& 0xff0);
526 ata_for_each_dev(dev
, link
, ALL
)
527 ata_dev_disable(dev
);
530 /* freeze and set UNLOADED */
531 spin_lock_irqsave(ap
->lock
, flags
);
533 ata_port_freeze(ap
); /* won't be thawed */
534 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
; /* clear pending from freeze */
535 ap
->pflags
|= ATA_PFLAG_UNLOADED
;
537 spin_unlock_irqrestore(ap
->lock
, flags
);
541 * ata_scsi_error - SCSI layer error handler callback
542 * @host: SCSI host on which error occurred
544 * Handles SCSI-layer-thrown error events.
547 * Inherited from SCSI layer (none, can sleep)
552 void ata_scsi_error(struct Scsi_Host
*host
)
554 struct ata_port
*ap
= ata_shost_to_port(host
);
560 /* make sure sff pio task is not running */
561 ata_sff_flush_pio_task(ap
);
563 /* synchronize with host lock and sort out timeouts */
565 /* For new EH, all qcs are finished in one of three ways -
566 * normal completion, error completion, and SCSI timeout.
567 * Both completions can race against SCSI timeout. When normal
568 * completion wins, the qc never reaches EH. When error
569 * completion wins, the qc has ATA_QCFLAG_FAILED set.
571 * When SCSI timeout wins, things are a bit more complex.
572 * Normal or error completion can occur after the timeout but
573 * before this point. In such cases, both types of
574 * completions are honored. A scmd is determined to have
575 * timed out iff its associated qc is active and not failed.
577 if (ap
->ops
->error_handler
) {
578 struct scsi_cmnd
*scmd
, *tmp
;
581 spin_lock_irqsave(ap
->lock
, flags
);
583 /* This must occur under the ap->lock as we don't want
584 a polled recovery to race the real interrupt handler
586 The lost_interrupt handler checks for any completed but
587 non-notified command and completes much like an IRQ handler.
589 We then fall into the error recovery code which will treat
590 this as if normal completion won the race */
592 if (ap
->ops
->lost_interrupt
)
593 ap
->ops
->lost_interrupt(ap
);
595 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
596 struct ata_queued_cmd
*qc
;
598 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
599 qc
= __ata_qc_from_tag(ap
, i
);
600 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
605 if (i
< ATA_MAX_QUEUE
) {
606 /* the scmd has an associated qc */
607 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
608 /* which hasn't failed yet, timeout */
609 qc
->err_mask
|= AC_ERR_TIMEOUT
;
610 qc
->flags
|= ATA_QCFLAG_FAILED
;
614 /* Normal completion occurred after
615 * SCSI timeout but before this point.
616 * Successfully complete it.
618 scmd
->retries
= scmd
->allowed
;
619 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
623 /* If we have timed out qcs. They belong to EH from
624 * this point but the state of the controller is
625 * unknown. Freeze the port to make sure the IRQ
626 * handler doesn't diddle with those qcs. This must
627 * be done atomically w.r.t. setting QCFLAG_FAILED.
630 __ata_port_freeze(ap
);
632 spin_unlock_irqrestore(ap
->lock
, flags
);
634 /* initialize eh_tries */
635 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
637 spin_unlock_wait(ap
->lock
);
639 /* If we timed raced normal completion and there is nothing to
640 recover nr_timedout == 0 why exactly are we doing error recovery ? */
643 /* invoke error handler */
644 if (ap
->ops
->error_handler
) {
645 struct ata_link
*link
;
647 /* kill fast drain timer */
648 del_timer_sync(&ap
->fastdrain_timer
);
650 /* process port resume request */
651 ata_eh_handle_port_resume(ap
);
653 /* fetch & clear EH info */
654 spin_lock_irqsave(ap
->lock
, flags
);
656 ata_for_each_link(link
, ap
, HOST_FIRST
) {
657 struct ata_eh_context
*ehc
= &link
->eh_context
;
658 struct ata_device
*dev
;
660 memset(&link
->eh_context
, 0, sizeof(link
->eh_context
));
661 link
->eh_context
.i
= link
->eh_info
;
662 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
664 ata_for_each_dev(dev
, link
, ENABLED
) {
665 int devno
= dev
->devno
;
667 ehc
->saved_xfer_mode
[devno
] = dev
->xfer_mode
;
668 if (ata_ncq_enabled(dev
))
669 ehc
->saved_ncq_enabled
|= 1 << devno
;
673 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
674 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
675 ap
->excl_link
= NULL
; /* don't maintain exclusion over EH */
677 spin_unlock_irqrestore(ap
->lock
, flags
);
679 /* invoke EH, skip if unloading or suspended */
680 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
681 ap
->ops
->error_handler(ap
);
683 /* if unloading, commence suicide */
684 if ((ap
->pflags
& ATA_PFLAG_UNLOADING
) &&
685 !(ap
->pflags
& ATA_PFLAG_UNLOADED
))
690 /* process port suspend request */
691 ata_eh_handle_port_suspend(ap
);
693 /* Exception might have happend after ->error_handler
694 * recovered the port but before this point. Repeat
697 spin_lock_irqsave(ap
->lock
, flags
);
699 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
700 if (--ap
->eh_tries
) {
701 spin_unlock_irqrestore(ap
->lock
, flags
);
704 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
705 "tries, giving up\n", ATA_EH_MAX_TRIES
);
706 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
709 /* this run is complete, make sure EH info is clear */
710 ata_for_each_link(link
, ap
, HOST_FIRST
)
711 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
713 /* Clear host_eh_scheduled while holding ap->lock such
714 * that if exception occurs after this point but
715 * before EH completion, SCSI midlayer will
718 host
->host_eh_scheduled
= 0;
720 spin_unlock_irqrestore(ap
->lock
, flags
);
722 WARN_ON(ata_qc_from_tag(ap
, ap
->link
.active_tag
) == NULL
);
723 ap
->ops
->eng_timeout(ap
);
726 /* finish or retry handled scmd's and clean up */
727 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
729 scsi_eh_flush_done_q(&ap
->eh_done_q
);
732 spin_lock_irqsave(ap
->lock
, flags
);
734 if (ap
->pflags
& ATA_PFLAG_LOADING
)
735 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
736 else if (ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
)
737 schedule_delayed_work(&ap
->hotplug_task
, 0);
739 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
740 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
742 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
744 /* tell wait_eh that we're done */
745 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
746 wake_up_all(&ap
->eh_wait_q
);
748 spin_unlock_irqrestore(ap
->lock
, flags
);
754 * ata_port_wait_eh - Wait for the currently pending EH to complete
755 * @ap: Port to wait EH for
757 * Wait until the currently pending EH is complete.
760 * Kernel thread context (may sleep).
762 void ata_port_wait_eh(struct ata_port
*ap
)
768 spin_lock_irqsave(ap
->lock
, flags
);
770 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
771 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
772 spin_unlock_irqrestore(ap
->lock
, flags
);
774 spin_lock_irqsave(ap
->lock
, flags
);
776 finish_wait(&ap
->eh_wait_q
, &wait
);
778 spin_unlock_irqrestore(ap
->lock
, flags
);
780 /* make sure SCSI EH is complete */
781 if (scsi_host_in_recovery(ap
->scsi_host
)) {
787 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
792 /* count only non-internal commands */
793 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
794 if (ata_qc_from_tag(ap
, tag
))
800 void ata_eh_fastdrain_timerfn(unsigned long arg
)
802 struct ata_port
*ap
= (void *)arg
;
806 spin_lock_irqsave(ap
->lock
, flags
);
808 cnt
= ata_eh_nr_in_flight(ap
);
814 if (cnt
== ap
->fastdrain_cnt
) {
817 /* No progress during the last interval, tag all
818 * in-flight qcs as timed out and freeze the port.
820 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
821 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
823 qc
->err_mask
|= AC_ERR_TIMEOUT
;
828 /* some qcs have finished, give it another chance */
829 ap
->fastdrain_cnt
= cnt
;
830 ap
->fastdrain_timer
.expires
=
831 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
832 add_timer(&ap
->fastdrain_timer
);
836 spin_unlock_irqrestore(ap
->lock
, flags
);
840 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
841 * @ap: target ATA port
842 * @fastdrain: activate fast drain
844 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
845 * is non-zero and EH wasn't pending before. Fast drain ensures
846 * that EH kicks in in timely manner.
849 * spin_lock_irqsave(host lock)
851 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
855 /* already scheduled? */
856 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
859 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
864 /* do we have in-flight qcs? */
865 cnt
= ata_eh_nr_in_flight(ap
);
869 /* activate fast drain */
870 ap
->fastdrain_cnt
= cnt
;
871 ap
->fastdrain_timer
.expires
=
872 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
873 add_timer(&ap
->fastdrain_timer
);
877 * ata_qc_schedule_eh - schedule qc for error handling
878 * @qc: command to schedule error handling for
880 * Schedule error handling for @qc. EH will kick in as soon as
881 * other commands are drained.
884 * spin_lock_irqsave(host lock)
886 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
888 struct ata_port
*ap
= qc
->ap
;
889 struct request_queue
*q
= qc
->scsicmd
->device
->request_queue
;
892 WARN_ON(!ap
->ops
->error_handler
);
894 qc
->flags
|= ATA_QCFLAG_FAILED
;
895 ata_eh_set_pending(ap
, 1);
897 /* The following will fail if timeout has already expired.
898 * ata_scsi_error() takes care of such scmds on EH entry.
899 * Note that ATA_QCFLAG_FAILED is unconditionally set after
900 * this function completes.
902 spin_lock_irqsave(q
->queue_lock
, flags
);
903 blk_abort_request(qc
->scsicmd
->request
);
904 spin_unlock_irqrestore(q
->queue_lock
, flags
);
908 * ata_port_schedule_eh - schedule error handling without a qc
909 * @ap: ATA port to schedule EH for
911 * Schedule error handling for @ap. EH will kick in as soon as
912 * all commands are drained.
915 * spin_lock_irqsave(host lock)
917 void ata_port_schedule_eh(struct ata_port
*ap
)
919 WARN_ON(!ap
->ops
->error_handler
);
921 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
924 ata_eh_set_pending(ap
, 1);
925 scsi_schedule_eh(ap
->scsi_host
);
927 DPRINTK("port EH scheduled\n");
930 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
932 int tag
, nr_aborted
= 0;
934 WARN_ON(!ap
->ops
->error_handler
);
936 /* we're gonna abort all commands, no need for fast drain */
937 ata_eh_set_pending(ap
, 0);
939 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
940 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
942 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
943 qc
->flags
|= ATA_QCFLAG_FAILED
;
950 ata_port_schedule_eh(ap
);
956 * ata_link_abort - abort all qc's on the link
957 * @link: ATA link to abort qc's for
959 * Abort all active qc's active on @link and schedule EH.
962 * spin_lock_irqsave(host lock)
965 * Number of aborted qc's.
967 int ata_link_abort(struct ata_link
*link
)
969 return ata_do_link_abort(link
->ap
, link
);
973 * ata_port_abort - abort all qc's on the port
974 * @ap: ATA port to abort qc's for
976 * Abort all active qc's of @ap and schedule EH.
979 * spin_lock_irqsave(host_set lock)
982 * Number of aborted qc's.
984 int ata_port_abort(struct ata_port
*ap
)
986 return ata_do_link_abort(ap
, NULL
);
990 * __ata_port_freeze - freeze port
991 * @ap: ATA port to freeze
993 * This function is called when HSM violation or some other
994 * condition disrupts normal operation of the port. Frozen port
995 * is not allowed to perform any operation until the port is
996 * thawed, which usually follows a successful reset.
998 * ap->ops->freeze() callback can be used for freezing the port
999 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1000 * port cannot be frozen hardware-wise, the interrupt handler
1001 * must ack and clear interrupts unconditionally while the port
1005 * spin_lock_irqsave(host lock)
1007 static void __ata_port_freeze(struct ata_port
*ap
)
1009 WARN_ON(!ap
->ops
->error_handler
);
1011 if (ap
->ops
->freeze
)
1012 ap
->ops
->freeze(ap
);
1014 ap
->pflags
|= ATA_PFLAG_FROZEN
;
1016 DPRINTK("ata%u port frozen\n", ap
->print_id
);
1020 * ata_port_freeze - abort & freeze port
1021 * @ap: ATA port to freeze
1023 * Abort and freeze @ap. The freeze operation must be called
1024 * first, because some hardware requires special operations
1025 * before the taskfile registers are accessible.
1028 * spin_lock_irqsave(host lock)
1031 * Number of aborted commands.
1033 int ata_port_freeze(struct ata_port
*ap
)
1037 WARN_ON(!ap
->ops
->error_handler
);
1039 __ata_port_freeze(ap
);
1040 nr_aborted
= ata_port_abort(ap
);
1046 * sata_async_notification - SATA async notification handler
1047 * @ap: ATA port where async notification is received
1049 * Handler to be called when async notification via SDB FIS is
1050 * received. This function schedules EH if necessary.
1053 * spin_lock_irqsave(host lock)
1056 * 1 if EH is scheduled, 0 otherwise.
1058 int sata_async_notification(struct ata_port
*ap
)
1063 if (!(ap
->flags
& ATA_FLAG_AN
))
1066 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
1068 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
1070 if (!sata_pmp_attached(ap
) || rc
) {
1071 /* PMP is not attached or SNTF is not available */
1072 if (!sata_pmp_attached(ap
)) {
1073 /* PMP is not attached. Check whether ATAPI
1074 * AN is configured. If so, notify media
1077 struct ata_device
*dev
= ap
->link
.device
;
1079 if ((dev
->class == ATA_DEV_ATAPI
) &&
1080 (dev
->flags
& ATA_DFLAG_AN
))
1081 ata_scsi_media_change_notify(dev
);
1084 /* PMP is attached but SNTF is not available.
1085 * ATAPI async media change notification is
1086 * not used. The PMP must be reporting PHY
1087 * status change, schedule EH.
1089 ata_port_schedule_eh(ap
);
1093 /* PMP is attached and SNTF is available */
1094 struct ata_link
*link
;
1096 /* check and notify ATAPI AN */
1097 ata_for_each_link(link
, ap
, EDGE
) {
1098 if (!(sntf
& (1 << link
->pmp
)))
1101 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
1102 (link
->device
->flags
& ATA_DFLAG_AN
))
1103 ata_scsi_media_change_notify(link
->device
);
1106 /* If PMP is reporting that PHY status of some
1107 * downstream ports has changed, schedule EH.
1109 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
1110 ata_port_schedule_eh(ap
);
1119 * ata_eh_freeze_port - EH helper to freeze port
1120 * @ap: ATA port to freeze
1127 void ata_eh_freeze_port(struct ata_port
*ap
)
1129 unsigned long flags
;
1131 if (!ap
->ops
->error_handler
)
1134 spin_lock_irqsave(ap
->lock
, flags
);
1135 __ata_port_freeze(ap
);
1136 spin_unlock_irqrestore(ap
->lock
, flags
);
1140 * ata_port_thaw_port - EH helper to thaw port
1141 * @ap: ATA port to thaw
1143 * Thaw frozen port @ap.
1148 void ata_eh_thaw_port(struct ata_port
*ap
)
1150 unsigned long flags
;
1152 if (!ap
->ops
->error_handler
)
1155 spin_lock_irqsave(ap
->lock
, flags
);
1157 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
1162 spin_unlock_irqrestore(ap
->lock
, flags
);
1164 DPRINTK("ata%u port thawed\n", ap
->print_id
);
1167 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
1172 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1174 struct ata_port
*ap
= qc
->ap
;
1175 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1176 unsigned long flags
;
1178 spin_lock_irqsave(ap
->lock
, flags
);
1179 qc
->scsidone
= ata_eh_scsidone
;
1180 __ata_qc_complete(qc
);
1181 WARN_ON(ata_tag_valid(qc
->tag
));
1182 spin_unlock_irqrestore(ap
->lock
, flags
);
1184 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1188 * ata_eh_qc_complete - Complete an active ATA command from EH
1189 * @qc: Command to complete
1191 * Indicate to the mid and upper layers that an ATA command has
1192 * completed. To be used from EH.
1194 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1196 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1197 scmd
->retries
= scmd
->allowed
;
1198 __ata_eh_qc_complete(qc
);
1202 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1203 * @qc: Command to retry
1205 * Indicate to the mid and upper layers that an ATA command
1206 * should be retried. To be used from EH.
1208 * SCSI midlayer limits the number of retries to scmd->allowed.
1209 * scmd->retries is decremented for commands which get retried
1210 * due to unrelated failures (qc->err_mask is zero).
1212 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1214 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1215 if (!qc
->err_mask
&& scmd
->retries
)
1217 __ata_eh_qc_complete(qc
);
1221 * ata_dev_disable - disable ATA device
1222 * @dev: ATA device to disable
1229 void ata_dev_disable(struct ata_device
*dev
)
1231 if (!ata_dev_enabled(dev
))
1234 if (ata_msg_drv(dev
->link
->ap
))
1235 ata_dev_printk(dev
, KERN_WARNING
, "disabled\n");
1236 ata_acpi_on_disable(dev
);
1237 ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO0
| ATA_DNXFER_QUIET
);
1240 /* From now till the next successful probe, ering is used to
1241 * track probe failures. Clear accumulated device error info.
1243 ata_ering_clear(&dev
->ering
);
1247 * ata_eh_detach_dev - detach ATA device
1248 * @dev: ATA device to detach
1255 void ata_eh_detach_dev(struct ata_device
*dev
)
1257 struct ata_link
*link
= dev
->link
;
1258 struct ata_port
*ap
= link
->ap
;
1259 struct ata_eh_context
*ehc
= &link
->eh_context
;
1260 unsigned long flags
;
1262 ata_dev_disable(dev
);
1264 spin_lock_irqsave(ap
->lock
, flags
);
1266 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1268 if (ata_scsi_offline_dev(dev
)) {
1269 dev
->flags
|= ATA_DFLAG_DETACHED
;
1270 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1273 /* clear per-dev EH info */
1274 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1275 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1276 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
1277 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
1279 spin_unlock_irqrestore(ap
->lock
, flags
);
1283 * ata_eh_about_to_do - about to perform eh_action
1284 * @link: target ATA link
1285 * @dev: target ATA dev for per-dev action (can be NULL)
1286 * @action: action about to be performed
1288 * Called just before performing EH actions to clear related bits
1289 * in @link->eh_info such that eh actions are not unnecessarily
1295 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1296 unsigned int action
)
1298 struct ata_port
*ap
= link
->ap
;
1299 struct ata_eh_info
*ehi
= &link
->eh_info
;
1300 struct ata_eh_context
*ehc
= &link
->eh_context
;
1301 unsigned long flags
;
1303 spin_lock_irqsave(ap
->lock
, flags
);
1305 ata_eh_clear_action(link
, dev
, ehi
, action
);
1307 /* About to take EH action, set RECOVERED. Ignore actions on
1308 * slave links as master will do them again.
1310 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
) && link
!= ap
->slave_link
)
1311 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1313 spin_unlock_irqrestore(ap
->lock
, flags
);
1317 * ata_eh_done - EH action complete
1318 * @ap: target ATA port
1319 * @dev: target ATA dev for per-dev action (can be NULL)
1320 * @action: action just completed
1322 * Called right after performing EH actions to clear related bits
1323 * in @link->eh_context.
1328 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1329 unsigned int action
)
1331 struct ata_eh_context
*ehc
= &link
->eh_context
;
1333 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1337 * ata_err_string - convert err_mask to descriptive string
1338 * @err_mask: error mask to convert to string
1340 * Convert @err_mask to descriptive string. Errors are
1341 * prioritized according to severity and only the most severe
1342 * error is reported.
1348 * Descriptive string for @err_mask
1350 static const char *ata_err_string(unsigned int err_mask
)
1352 if (err_mask
& AC_ERR_HOST_BUS
)
1353 return "host bus error";
1354 if (err_mask
& AC_ERR_ATA_BUS
)
1355 return "ATA bus error";
1356 if (err_mask
& AC_ERR_TIMEOUT
)
1358 if (err_mask
& AC_ERR_HSM
)
1359 return "HSM violation";
1360 if (err_mask
& AC_ERR_SYSTEM
)
1361 return "internal error";
1362 if (err_mask
& AC_ERR_MEDIA
)
1363 return "media error";
1364 if (err_mask
& AC_ERR_INVALID
)
1365 return "invalid argument";
1366 if (err_mask
& AC_ERR_DEV
)
1367 return "device error";
1368 return "unknown error";
1372 * ata_read_log_page - read a specific log page
1373 * @dev: target device
1374 * @page: page to read
1375 * @buf: buffer to store read page
1376 * @sectors: number of sectors to read
1378 * Read log page using READ_LOG_EXT command.
1381 * Kernel thread context (may sleep).
1384 * 0 on success, AC_ERR_* mask otherwise.
1386 static unsigned int ata_read_log_page(struct ata_device
*dev
,
1387 u8 page
, void *buf
, unsigned int sectors
)
1389 struct ata_taskfile tf
;
1390 unsigned int err_mask
;
1392 DPRINTK("read log page - page %d\n", page
);
1394 ata_tf_init(dev
, &tf
);
1395 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1398 tf
.hob_nsect
= sectors
>> 8;
1399 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1400 tf
.protocol
= ATA_PROT_PIO
;
1402 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1403 buf
, sectors
* ATA_SECT_SIZE
, 0);
1405 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1410 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1411 * @dev: Device to read log page 10h from
1412 * @tag: Resulting tag of the failed command
1413 * @tf: Resulting taskfile registers of the failed command
1415 * Read log page 10h to obtain NCQ error details and clear error
1419 * Kernel thread context (may sleep).
1422 * 0 on success, -errno otherwise.
1424 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1425 int *tag
, struct ata_taskfile
*tf
)
1427 u8
*buf
= dev
->link
->ap
->sector_buf
;
1428 unsigned int err_mask
;
1432 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
1437 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1440 ata_dev_printk(dev
, KERN_WARNING
,
1441 "invalid checksum 0x%x on log page 10h\n", csum
);
1446 *tag
= buf
[0] & 0x1f;
1448 tf
->command
= buf
[2];
1449 tf
->feature
= buf
[3];
1453 tf
->device
= buf
[7];
1454 tf
->hob_lbal
= buf
[8];
1455 tf
->hob_lbam
= buf
[9];
1456 tf
->hob_lbah
= buf
[10];
1457 tf
->nsect
= buf
[12];
1458 tf
->hob_nsect
= buf
[13];
1464 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1465 * @dev: target ATAPI device
1466 * @r_sense_key: out parameter for sense_key
1468 * Perform ATAPI TEST_UNIT_READY.
1471 * EH context (may sleep).
1474 * 0 on success, AC_ERR_* mask on failure.
1476 static unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1478 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1479 struct ata_taskfile tf
;
1480 unsigned int err_mask
;
1482 ata_tf_init(dev
, &tf
);
1484 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1485 tf
.command
= ATA_CMD_PACKET
;
1486 tf
.protocol
= ATAPI_PROT_NODATA
;
1488 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1489 if (err_mask
== AC_ERR_DEV
)
1490 *r_sense_key
= tf
.feature
>> 4;
1495 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1496 * @dev: device to perform REQUEST_SENSE to
1497 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1498 * @dfl_sense_key: default sense key to use
1500 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1501 * SENSE. This function is EH helper.
1504 * Kernel thread context (may sleep).
1507 * 0 on success, AC_ERR_* mask on failure
1509 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1510 u8
*sense_buf
, u8 dfl_sense_key
)
1512 u8 cdb
[ATAPI_CDB_LEN
] =
1513 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1514 struct ata_port
*ap
= dev
->link
->ap
;
1515 struct ata_taskfile tf
;
1517 DPRINTK("ATAPI request sense\n");
1519 /* FIXME: is this needed? */
1520 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1522 /* initialize sense_buf with the error register,
1523 * for the case where they are -not- overwritten
1525 sense_buf
[0] = 0x70;
1526 sense_buf
[2] = dfl_sense_key
;
1528 /* some devices time out if garbage left in tf */
1529 ata_tf_init(dev
, &tf
);
1531 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1532 tf
.command
= ATA_CMD_PACKET
;
1534 /* is it pointless to prefer PIO for "safety reasons"? */
1535 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1536 tf
.protocol
= ATAPI_PROT_DMA
;
1537 tf
.feature
|= ATAPI_PKT_DMA
;
1539 tf
.protocol
= ATAPI_PROT_PIO
;
1540 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1544 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1545 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1549 * ata_eh_analyze_serror - analyze SError for a failed port
1550 * @link: ATA link to analyze SError for
1552 * Analyze SError if available and further determine cause of
1558 static void ata_eh_analyze_serror(struct ata_link
*link
)
1560 struct ata_eh_context
*ehc
= &link
->eh_context
;
1561 u32 serror
= ehc
->i
.serror
;
1562 unsigned int err_mask
= 0, action
= 0;
1565 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1566 err_mask
|= AC_ERR_ATA_BUS
;
1567 action
|= ATA_EH_RESET
;
1569 if (serror
& SERR_PROTOCOL
) {
1570 err_mask
|= AC_ERR_HSM
;
1571 action
|= ATA_EH_RESET
;
1573 if (serror
& SERR_INTERNAL
) {
1574 err_mask
|= AC_ERR_SYSTEM
;
1575 action
|= ATA_EH_RESET
;
1578 /* Determine whether a hotplug event has occurred. Both
1579 * SError.N/X are considered hotplug events for enabled or
1580 * host links. For disabled PMP links, only N bit is
1581 * considered as X bit is left at 1 for link plugging.
1585 if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1586 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1588 hotplug_mask
= SERR_PHYRDY_CHG
;
1590 if (serror
& hotplug_mask
)
1591 ata_ehi_hotplugged(&ehc
->i
);
1593 ehc
->i
.err_mask
|= err_mask
;
1594 ehc
->i
.action
|= action
;
1598 * ata_eh_analyze_ncq_error - analyze NCQ error
1599 * @link: ATA link to analyze NCQ error for
1601 * Read log page 10h, determine the offending qc and acquire
1602 * error status TF. For NCQ device errors, all LLDDs have to do
1603 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1607 * Kernel thread context (may sleep).
1609 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1611 struct ata_port
*ap
= link
->ap
;
1612 struct ata_eh_context
*ehc
= &link
->eh_context
;
1613 struct ata_device
*dev
= link
->device
;
1614 struct ata_queued_cmd
*qc
;
1615 struct ata_taskfile tf
;
1618 /* if frozen, we can't do much */
1619 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1622 /* is it NCQ device error? */
1623 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1626 /* has LLDD analyzed already? */
1627 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1628 qc
= __ata_qc_from_tag(ap
, tag
);
1630 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1637 /* okay, this error is ours */
1638 memset(&tf
, 0, sizeof(tf
));
1639 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1641 ata_link_printk(link
, KERN_ERR
, "failed to read log page 10h "
1642 "(errno=%d)\n", rc
);
1646 if (!(link
->sactive
& (1 << tag
))) {
1647 ata_link_printk(link
, KERN_ERR
, "log page 10h reported "
1648 "inactive tag %d\n", tag
);
1652 /* we've got the perpetrator, condemn it */
1653 qc
= __ata_qc_from_tag(ap
, tag
);
1654 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1655 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1656 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1657 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1661 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1662 * @qc: qc to analyze
1663 * @tf: Taskfile registers to analyze
1665 * Analyze taskfile of @qc and further determine cause of
1666 * failure. This function also requests ATAPI sense data if
1670 * Kernel thread context (may sleep).
1673 * Determined recovery action
1675 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1676 const struct ata_taskfile
*tf
)
1678 unsigned int tmp
, action
= 0;
1679 u8 stat
= tf
->command
, err
= tf
->feature
;
1681 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1682 qc
->err_mask
|= AC_ERR_HSM
;
1683 return ATA_EH_RESET
;
1686 if (stat
& (ATA_ERR
| ATA_DF
))
1687 qc
->err_mask
|= AC_ERR_DEV
;
1691 switch (qc
->dev
->class) {
1694 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1696 qc
->err_mask
|= AC_ERR_MEDIA
;
1698 qc
->err_mask
|= AC_ERR_INVALID
;
1702 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1703 tmp
= atapi_eh_request_sense(qc
->dev
,
1704 qc
->scsicmd
->sense_buffer
,
1705 qc
->result_tf
.feature
>> 4);
1707 /* ATA_QCFLAG_SENSE_VALID is used to
1708 * tell atapi_qc_complete() that sense
1709 * data is already valid.
1711 * TODO: interpret sense data and set
1712 * appropriate err_mask.
1714 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1716 qc
->err_mask
|= tmp
;
1720 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1721 action
|= ATA_EH_RESET
;
1726 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1731 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1735 base
= ATA_ECAT_DUBIOUS_NONE
;
1737 if (err_mask
& AC_ERR_ATA_BUS
)
1738 return base
+ ATA_ECAT_ATA_BUS
;
1740 if (err_mask
& AC_ERR_TIMEOUT
)
1741 return base
+ ATA_ECAT_TOUT_HSM
;
1743 if (eflags
& ATA_EFLAG_IS_IO
) {
1744 if (err_mask
& AC_ERR_HSM
)
1745 return base
+ ATA_ECAT_TOUT_HSM
;
1747 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1748 return base
+ ATA_ECAT_UNK_DEV
;
1754 struct speed_down_verdict_arg
{
1757 int nr_errors
[ATA_ECAT_NR
];
1760 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1762 struct speed_down_verdict_arg
*arg
= void_arg
;
1765 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) || (ent
->timestamp
< arg
->since
))
1768 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1770 arg
->nr_errors
[cat
]++;
1776 * ata_eh_speed_down_verdict - Determine speed down verdict
1777 * @dev: Device of interest
1779 * This function examines error ring of @dev and determines
1780 * whether NCQ needs to be turned off, transfer speed should be
1781 * stepped down, or falling back to PIO is necessary.
1783 * ECAT_ATA_BUS : ATA_BUS error for any command
1785 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1788 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1790 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1791 * data transfer hasn't been verified.
1795 * NCQ_OFF : Turn off NCQ.
1797 * SPEED_DOWN : Speed down transfer speed but don't fall back
1800 * FALLBACK_TO_PIO : Fall back to PIO.
1802 * Even if multiple verdicts are returned, only one action is
1803 * taken per error. An action triggered by non-DUBIOUS errors
1804 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1805 * This is to expedite speed down decisions right after device is
1806 * initially configured.
1808 * The followings are speed down rules. #1 and #2 deal with
1811 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1812 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1814 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1815 * occurred during last 5 mins, NCQ_OFF.
1817 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1818 * ocurred during last 5 mins, FALLBACK_TO_PIO
1820 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1821 * during last 10 mins, NCQ_OFF.
1823 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1824 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1827 * Inherited from caller.
1830 * OR of ATA_EH_SPDN_* flags.
1832 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1834 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1835 u64 j64
= get_jiffies_64();
1836 struct speed_down_verdict_arg arg
;
1837 unsigned int verdict
= 0;
1839 /* scan past 5 mins of error history */
1840 memset(&arg
, 0, sizeof(arg
));
1841 arg
.since
= j64
- min(j64
, j5mins
);
1842 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1844 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1845 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1846 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1847 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1849 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1850 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1851 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1853 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1854 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1855 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1856 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1858 /* scan past 10 mins of error history */
1859 memset(&arg
, 0, sizeof(arg
));
1860 arg
.since
= j64
- min(j64
, j10mins
);
1861 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1863 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1864 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1865 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1867 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1868 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1869 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1870 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1876 * ata_eh_speed_down - record error and speed down if necessary
1877 * @dev: Failed device
1878 * @eflags: mask of ATA_EFLAG_* flags
1879 * @err_mask: err_mask of the error
1881 * Record error and examine error history to determine whether
1882 * adjusting transmission speed is necessary. It also sets
1883 * transmission limits appropriately if such adjustment is
1887 * Kernel thread context (may sleep).
1890 * Determined recovery action.
1892 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
1893 unsigned int eflags
, unsigned int err_mask
)
1895 struct ata_link
*link
= ata_dev_phys_link(dev
);
1897 unsigned int verdict
;
1898 unsigned int action
= 0;
1900 /* don't bother if Cat-0 error */
1901 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
1904 /* record error and determine whether speed down is necessary */
1905 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
1906 verdict
= ata_eh_speed_down_verdict(dev
);
1909 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
1910 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
1911 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
1912 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
1913 ata_dev_printk(dev
, KERN_WARNING
,
1914 "NCQ disabled due to excessive errors\n");
1919 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
1920 /* speed down SATA link speed if possible */
1921 if (sata_down_spd_limit(link
, 0) == 0) {
1922 action
|= ATA_EH_RESET
;
1926 /* lower transfer mode */
1927 if (dev
->spdn_cnt
< 2) {
1928 static const int dma_dnxfer_sel
[] =
1929 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
1930 static const int pio_dnxfer_sel
[] =
1931 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
1934 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
1935 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
1937 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
1941 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
1942 action
|= ATA_EH_RESET
;
1948 /* Fall back to PIO? Slowing down to PIO is meaningless for
1949 * SATA ATA devices. Consider it only for PATA and SATAPI.
1951 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
1952 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
1953 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
1954 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
1956 action
|= ATA_EH_RESET
;
1963 /* device has been slowed down, blow error history */
1964 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
1965 ata_ering_clear(&dev
->ering
);
1970 * ata_eh_link_autopsy - analyze error and determine recovery action
1971 * @link: host link to perform autopsy on
1973 * Analyze why @link failed and determine which recovery actions
1974 * are needed. This function also sets more detailed AC_ERR_*
1975 * values and fills sense data for ATAPI CHECK SENSE.
1978 * Kernel thread context (may sleep).
1980 static void ata_eh_link_autopsy(struct ata_link
*link
)
1982 struct ata_port
*ap
= link
->ap
;
1983 struct ata_eh_context
*ehc
= &link
->eh_context
;
1984 struct ata_device
*dev
;
1985 unsigned int all_err_mask
= 0, eflags
= 0;
1992 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
1995 /* obtain and analyze SError */
1996 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
1998 ehc
->i
.serror
|= serror
;
1999 ata_eh_analyze_serror(link
);
2000 } else if (rc
!= -EOPNOTSUPP
) {
2001 /* SError read failed, force reset and probing */
2002 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
2003 ehc
->i
.action
|= ATA_EH_RESET
;
2004 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
2007 /* analyze NCQ failure */
2008 ata_eh_analyze_ncq_error(link
);
2010 /* any real error trumps AC_ERR_OTHER */
2011 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
2012 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
2014 all_err_mask
|= ehc
->i
.err_mask
;
2016 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2017 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2019 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2020 ata_dev_phys_link(qc
->dev
) != link
)
2023 /* inherit upper level err_mask */
2024 qc
->err_mask
|= ehc
->i
.err_mask
;
2027 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
2029 /* DEV errors are probably spurious in case of ATA_BUS error */
2030 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
2031 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
2034 /* any real error trumps unknown error */
2035 if (qc
->err_mask
& ~AC_ERR_OTHER
)
2036 qc
->err_mask
&= ~AC_ERR_OTHER
;
2038 /* SENSE_VALID trumps dev/unknown error and revalidation */
2039 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
2040 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
2042 /* determine whether the command is worth retrying */
2043 if (qc
->flags
& ATA_QCFLAG_IO
||
2044 (!(qc
->err_mask
& AC_ERR_INVALID
) &&
2045 qc
->err_mask
!= AC_ERR_DEV
))
2046 qc
->flags
|= ATA_QCFLAG_RETRY
;
2048 /* accumulate error info */
2049 ehc
->i
.dev
= qc
->dev
;
2050 all_err_mask
|= qc
->err_mask
;
2051 if (qc
->flags
& ATA_QCFLAG_IO
)
2052 eflags
|= ATA_EFLAG_IS_IO
;
2055 /* enforce default EH actions */
2056 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
2057 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
2058 ehc
->i
.action
|= ATA_EH_RESET
;
2059 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
2060 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
2061 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2063 /* If we have offending qcs and the associated failed device,
2064 * perform per-dev EH action only on the offending device.
2067 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
2068 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
2069 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
2072 /* propagate timeout to host link */
2073 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
2074 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
2076 /* record error and consider speeding down */
2078 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
2079 ata_dev_enabled(link
->device
))))
2083 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
2084 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
2085 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
2092 * ata_eh_autopsy - analyze error and determine recovery action
2093 * @ap: host port to perform autopsy on
2095 * Analyze all links of @ap and determine why they failed and
2096 * which recovery actions are needed.
2099 * Kernel thread context (may sleep).
2101 void ata_eh_autopsy(struct ata_port
*ap
)
2103 struct ata_link
*link
;
2105 ata_for_each_link(link
, ap
, EDGE
)
2106 ata_eh_link_autopsy(link
);
2108 /* Handle the frigging slave link. Autopsy is done similarly
2109 * but actions and flags are transferred over to the master
2110 * link and handled from there.
2112 if (ap
->slave_link
) {
2113 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2114 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2116 /* transfer control flags from master to slave */
2117 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2119 /* perform autopsy on the slave link */
2120 ata_eh_link_autopsy(ap
->slave_link
);
2122 /* transfer actions from slave to master and clear slave */
2123 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2124 mehc
->i
.action
|= sehc
->i
.action
;
2125 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2126 mehc
->i
.flags
|= sehc
->i
.flags
;
2127 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2130 /* Autopsy of fanout ports can affect host link autopsy.
2131 * Perform host link autopsy last.
2133 if (sata_pmp_attached(ap
))
2134 ata_eh_link_autopsy(&ap
->link
);
2138 * ata_get_cmd_descript - get description for ATA command
2139 * @command: ATA command code to get description for
2141 * Return a textual description of the given command, or NULL if the
2142 * command is not known.
2147 const char *ata_get_cmd_descript(u8 command
)
2149 #ifdef CONFIG_ATA_VERBOSE_ERROR
2155 { ATA_CMD_DEV_RESET
, "DEVICE RESET" },
2156 { ATA_CMD_CHK_POWER
, "CHECK POWER MODE" },
2157 { ATA_CMD_STANDBY
, "STANDBY" },
2158 { ATA_CMD_IDLE
, "IDLE" },
2159 { ATA_CMD_EDD
, "EXECUTE DEVICE DIAGNOSTIC" },
2160 { ATA_CMD_DOWNLOAD_MICRO
, "DOWNLOAD MICROCODE" },
2161 { ATA_CMD_NOP
, "NOP" },
2162 { ATA_CMD_FLUSH
, "FLUSH CACHE" },
2163 { ATA_CMD_FLUSH_EXT
, "FLUSH CACHE EXT" },
2164 { ATA_CMD_ID_ATA
, "IDENTIFY DEVICE" },
2165 { ATA_CMD_ID_ATAPI
, "IDENTIFY PACKET DEVICE" },
2166 { ATA_CMD_SERVICE
, "SERVICE" },
2167 { ATA_CMD_READ
, "READ DMA" },
2168 { ATA_CMD_READ_EXT
, "READ DMA EXT" },
2169 { ATA_CMD_READ_QUEUED
, "READ DMA QUEUED" },
2170 { ATA_CMD_READ_STREAM_EXT
, "READ STREAM EXT" },
2171 { ATA_CMD_READ_STREAM_DMA_EXT
, "READ STREAM DMA EXT" },
2172 { ATA_CMD_WRITE
, "WRITE DMA" },
2173 { ATA_CMD_WRITE_EXT
, "WRITE DMA EXT" },
2174 { ATA_CMD_WRITE_QUEUED
, "WRITE DMA QUEUED EXT" },
2175 { ATA_CMD_WRITE_STREAM_EXT
, "WRITE STREAM EXT" },
2176 { ATA_CMD_WRITE_STREAM_DMA_EXT
, "WRITE STREAM DMA EXT" },
2177 { ATA_CMD_WRITE_FUA_EXT
, "WRITE DMA FUA EXT" },
2178 { ATA_CMD_WRITE_QUEUED_FUA_EXT
, "WRITE DMA QUEUED FUA EXT" },
2179 { ATA_CMD_FPDMA_READ
, "READ FPDMA QUEUED" },
2180 { ATA_CMD_FPDMA_WRITE
, "WRITE FPDMA QUEUED" },
2181 { ATA_CMD_PIO_READ
, "READ SECTOR(S)" },
2182 { ATA_CMD_PIO_READ_EXT
, "READ SECTOR(S) EXT" },
2183 { ATA_CMD_PIO_WRITE
, "WRITE SECTOR(S)" },
2184 { ATA_CMD_PIO_WRITE_EXT
, "WRITE SECTOR(S) EXT" },
2185 { ATA_CMD_READ_MULTI
, "READ MULTIPLE" },
2186 { ATA_CMD_READ_MULTI_EXT
, "READ MULTIPLE EXT" },
2187 { ATA_CMD_WRITE_MULTI
, "WRITE MULTIPLE" },
2188 { ATA_CMD_WRITE_MULTI_EXT
, "WRITE MULTIPLE EXT" },
2189 { ATA_CMD_WRITE_MULTI_FUA_EXT
, "WRITE MULTIPLE FUA EXT" },
2190 { ATA_CMD_SET_FEATURES
, "SET FEATURES" },
2191 { ATA_CMD_SET_MULTI
, "SET MULTIPLE MODE" },
2192 { ATA_CMD_VERIFY
, "READ VERIFY SECTOR(S)" },
2193 { ATA_CMD_VERIFY_EXT
, "READ VERIFY SECTOR(S) EXT" },
2194 { ATA_CMD_WRITE_UNCORR_EXT
, "WRITE UNCORRECTABLE EXT" },
2195 { ATA_CMD_STANDBYNOW1
, "STANDBY IMMEDIATE" },
2196 { ATA_CMD_IDLEIMMEDIATE
, "IDLE IMMEDIATE" },
2197 { ATA_CMD_SLEEP
, "SLEEP" },
2198 { ATA_CMD_INIT_DEV_PARAMS
, "INITIALIZE DEVICE PARAMETERS" },
2199 { ATA_CMD_READ_NATIVE_MAX
, "READ NATIVE MAX ADDRESS" },
2200 { ATA_CMD_READ_NATIVE_MAX_EXT
, "READ NATIVE MAX ADDRESS EXT" },
2201 { ATA_CMD_SET_MAX
, "SET MAX ADDRESS" },
2202 { ATA_CMD_SET_MAX_EXT
, "SET MAX ADDRESS EXT" },
2203 { ATA_CMD_READ_LOG_EXT
, "READ LOG EXT" },
2204 { ATA_CMD_WRITE_LOG_EXT
, "WRITE LOG EXT" },
2205 { ATA_CMD_READ_LOG_DMA_EXT
, "READ LOG DMA EXT" },
2206 { ATA_CMD_WRITE_LOG_DMA_EXT
, "WRITE LOG DMA EXT" },
2207 { ATA_CMD_TRUSTED_RCV
, "TRUSTED RECEIVE" },
2208 { ATA_CMD_TRUSTED_RCV_DMA
, "TRUSTED RECEIVE DMA" },
2209 { ATA_CMD_TRUSTED_SND
, "TRUSTED SEND" },
2210 { ATA_CMD_TRUSTED_SND_DMA
, "TRUSTED SEND DMA" },
2211 { ATA_CMD_PMP_READ
, "READ BUFFER" },
2212 { ATA_CMD_PMP_WRITE
, "WRITE BUFFER" },
2213 { ATA_CMD_CONF_OVERLAY
, "DEVICE CONFIGURATION OVERLAY" },
2214 { ATA_CMD_SEC_SET_PASS
, "SECURITY SET PASSWORD" },
2215 { ATA_CMD_SEC_UNLOCK
, "SECURITY UNLOCK" },
2216 { ATA_CMD_SEC_ERASE_PREP
, "SECURITY ERASE PREPARE" },
2217 { ATA_CMD_SEC_ERASE_UNIT
, "SECURITY ERASE UNIT" },
2218 { ATA_CMD_SEC_FREEZE_LOCK
, "SECURITY FREEZE LOCK" },
2219 { ATA_CMD_SEC_DISABLE_PASS
, "SECURITY DISABLE PASSWORD" },
2220 { ATA_CMD_CONFIG_STREAM
, "CONFIGURE STREAM" },
2221 { ATA_CMD_SMART
, "SMART" },
2222 { ATA_CMD_MEDIA_LOCK
, "DOOR LOCK" },
2223 { ATA_CMD_MEDIA_UNLOCK
, "DOOR UNLOCK" },
2224 { ATA_CMD_DSM
, "DATA SET MANAGEMENT" },
2225 { ATA_CMD_CHK_MED_CRD_TYP
, "CHECK MEDIA CARD TYPE" },
2226 { ATA_CMD_CFA_REQ_EXT_ERR
, "CFA REQUEST EXTENDED ERROR" },
2227 { ATA_CMD_CFA_WRITE_NE
, "CFA WRITE SECTORS WITHOUT ERASE" },
2228 { ATA_CMD_CFA_TRANS_SECT
, "CFA TRANSLATE SECTOR" },
2229 { ATA_CMD_CFA_ERASE
, "CFA ERASE SECTORS" },
2230 { ATA_CMD_CFA_WRITE_MULT_NE
, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2231 { ATA_CMD_READ_LONG
, "READ LONG (with retries)" },
2232 { ATA_CMD_READ_LONG_ONCE
, "READ LONG (without retries)" },
2233 { ATA_CMD_WRITE_LONG
, "WRITE LONG (with retries)" },
2234 { ATA_CMD_WRITE_LONG_ONCE
, "WRITE LONG (without retries)" },
2235 { ATA_CMD_RESTORE
, "RECALIBRATE" },
2236 { 0, NULL
} /* terminate list */
2240 for (i
= 0; cmd_descr
[i
].text
; i
++)
2241 if (cmd_descr
[i
].command
== command
)
2242 return cmd_descr
[i
].text
;
2249 * ata_eh_link_report - report error handling to user
2250 * @link: ATA link EH is going on
2252 * Report EH to user.
2257 static void ata_eh_link_report(struct ata_link
*link
)
2259 struct ata_port
*ap
= link
->ap
;
2260 struct ata_eh_context
*ehc
= &link
->eh_context
;
2261 const char *frozen
, *desc
;
2263 int tag
, nr_failed
= 0;
2265 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2269 if (ehc
->i
.desc
[0] != '\0')
2272 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2273 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2275 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2276 ata_dev_phys_link(qc
->dev
) != link
||
2277 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2278 qc
->err_mask
== AC_ERR_DEV
))
2280 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2286 if (!nr_failed
&& !ehc
->i
.err_mask
)
2290 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2293 memset(tries_buf
, 0, sizeof(tries_buf
));
2294 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2295 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
2299 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
2300 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2301 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2302 ehc
->i
.action
, frozen
, tries_buf
);
2304 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "%s\n", desc
);
2306 ata_link_printk(link
, KERN_ERR
, "exception Emask 0x%x "
2307 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2308 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2309 ehc
->i
.action
, frozen
, tries_buf
);
2311 ata_link_printk(link
, KERN_ERR
, "%s\n", desc
);
2314 #ifdef CONFIG_ATA_VERBOSE_ERROR
2316 ata_link_printk(link
, KERN_ERR
,
2317 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2318 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2319 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2320 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2321 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2322 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2323 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2324 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2325 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2326 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2327 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2328 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2329 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2330 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2331 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2332 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2333 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2334 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2337 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2338 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2339 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2340 const u8
*cdb
= qc
->cdb
;
2341 char data_buf
[20] = "";
2342 char cdb_buf
[70] = "";
2344 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2345 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2348 if (qc
->dma_dir
!= DMA_NONE
) {
2349 static const char *dma_str
[] = {
2350 [DMA_BIDIRECTIONAL
] = "bidi",
2351 [DMA_TO_DEVICE
] = "out",
2352 [DMA_FROM_DEVICE
] = "in",
2354 static const char *prot_str
[] = {
2355 [ATA_PROT_PIO
] = "pio",
2356 [ATA_PROT_DMA
] = "dma",
2357 [ATA_PROT_NCQ
] = "ncq",
2358 [ATAPI_PROT_PIO
] = "pio",
2359 [ATAPI_PROT_DMA
] = "dma",
2362 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2363 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
2364 dma_str
[qc
->dma_dir
]);
2367 if (ata_is_atapi(qc
->tf
.protocol
)) {
2369 scsi_print_command(qc
->scsicmd
);
2371 snprintf(cdb_buf
, sizeof(cdb_buf
),
2372 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2373 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2374 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
2375 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
2376 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
2377 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
2379 const char *descr
= ata_get_cmd_descript(cmd
->command
);
2381 ata_dev_printk(qc
->dev
, KERN_ERR
,
2382 "failed command: %s\n", descr
);
2385 ata_dev_printk(qc
->dev
, KERN_ERR
,
2386 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2388 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2389 "Emask 0x%x (%s)%s\n",
2390 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2391 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2392 cmd
->hob_feature
, cmd
->hob_nsect
,
2393 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2394 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2395 res
->command
, res
->feature
, res
->nsect
,
2396 res
->lbal
, res
->lbam
, res
->lbah
,
2397 res
->hob_feature
, res
->hob_nsect
,
2398 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2399 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2400 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2402 #ifdef CONFIG_ATA_VERBOSE_ERROR
2403 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2405 if (res
->command
& ATA_BUSY
)
2406 ata_dev_printk(qc
->dev
, KERN_ERR
,
2407 "status: { Busy }\n");
2409 ata_dev_printk(qc
->dev
, KERN_ERR
,
2410 "status: { %s%s%s%s}\n",
2411 res
->command
& ATA_DRDY
? "DRDY " : "",
2412 res
->command
& ATA_DF
? "DF " : "",
2413 res
->command
& ATA_DRQ
? "DRQ " : "",
2414 res
->command
& ATA_ERR
? "ERR " : "");
2417 if (cmd
->command
!= ATA_CMD_PACKET
&&
2418 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2420 ata_dev_printk(qc
->dev
, KERN_ERR
,
2421 "error: { %s%s%s%s}\n",
2422 res
->feature
& ATA_ICRC
? "ICRC " : "",
2423 res
->feature
& ATA_UNC
? "UNC " : "",
2424 res
->feature
& ATA_IDNF
? "IDNF " : "",
2425 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2431 * ata_eh_report - report error handling to user
2432 * @ap: ATA port to report EH about
2434 * Report EH to user.
2439 void ata_eh_report(struct ata_port
*ap
)
2441 struct ata_link
*link
;
2443 ata_for_each_link(link
, ap
, HOST_FIRST
)
2444 ata_eh_link_report(link
);
2447 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2448 unsigned int *classes
, unsigned long deadline
,
2451 struct ata_device
*dev
;
2454 ata_for_each_dev(dev
, link
, ALL
)
2455 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2457 return reset(link
, classes
, deadline
);
2460 static int ata_eh_followup_srst_needed(struct ata_link
*link
,
2461 int rc
, const unsigned int *classes
)
2463 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2467 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2472 int ata_eh_reset(struct ata_link
*link
, int classify
,
2473 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2474 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2476 struct ata_port
*ap
= link
->ap
;
2477 struct ata_link
*slave
= ap
->slave_link
;
2478 struct ata_eh_context
*ehc
= &link
->eh_context
;
2479 struct ata_eh_context
*sehc
= slave
? &slave
->eh_context
: NULL
;
2480 unsigned int *classes
= ehc
->classes
;
2481 unsigned int lflags
= link
->flags
;
2482 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2483 int max_tries
= 0, try = 0;
2484 struct ata_link
*failed_link
;
2485 struct ata_device
*dev
;
2486 unsigned long deadline
, now
;
2487 ata_reset_fn_t reset
;
2488 unsigned long flags
;
2495 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2497 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2499 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2502 /* make sure each reset attemp is at least COOL_DOWN apart */
2503 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2505 WARN_ON(time_after(ehc
->last_reset
, now
));
2506 deadline
= ata_deadline(ehc
->last_reset
,
2507 ATA_EH_RESET_COOL_DOWN
);
2508 if (time_before(now
, deadline
))
2509 schedule_timeout_uninterruptible(deadline
- now
);
2512 spin_lock_irqsave(ap
->lock
, flags
);
2513 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2514 spin_unlock_irqrestore(ap
->lock
, flags
);
2516 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2518 ata_for_each_dev(dev
, link
, ALL
) {
2519 /* If we issue an SRST then an ATA drive (not ATAPI)
2520 * may change configuration and be in PIO0 timing. If
2521 * we do a hard reset (or are coming from power on)
2522 * this is true for ATA or ATAPI. Until we've set a
2523 * suitable controller mode we should not touch the
2524 * bus as we may be talking too fast.
2526 dev
->pio_mode
= XFER_PIO_0
;
2528 /* If the controller has a pio mode setup function
2529 * then use it to set the chipset to rights. Don't
2530 * touch the DMA setup as that will be dealt with when
2531 * configuring devices.
2533 if (ap
->ops
->set_piomode
)
2534 ap
->ops
->set_piomode(ap
, dev
);
2537 /* prefer hardreset */
2539 ehc
->i
.action
&= ~ATA_EH_RESET
;
2542 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2543 } else if (softreset
) {
2545 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2549 unsigned long deadline
= ata_deadline(jiffies
,
2550 ATA_EH_PRERESET_TIMEOUT
);
2553 sehc
->i
.action
&= ~ATA_EH_RESET
;
2554 sehc
->i
.action
|= ehc
->i
.action
;
2557 rc
= prereset(link
, deadline
);
2559 /* If present, do prereset on slave link too. Reset
2560 * is skipped iff both master and slave links report
2561 * -ENOENT or clear ATA_EH_RESET.
2563 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2566 tmp
= prereset(slave
, deadline
);
2570 ehc
->i
.action
|= sehc
->i
.action
;
2574 if (rc
== -ENOENT
) {
2575 ata_link_printk(link
, KERN_DEBUG
,
2576 "port disabled. ignoring.\n");
2577 ehc
->i
.action
&= ~ATA_EH_RESET
;
2579 ata_for_each_dev(dev
, link
, ALL
)
2580 classes
[dev
->devno
] = ATA_DEV_NONE
;
2584 ata_link_printk(link
, KERN_ERR
,
2585 "prereset failed (errno=%d)\n", rc
);
2589 /* prereset() might have cleared ATA_EH_RESET. If so,
2590 * bang classes, thaw and return.
2592 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2593 ata_for_each_dev(dev
, link
, ALL
)
2594 classes
[dev
->devno
] = ATA_DEV_NONE
;
2595 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) &&
2596 ata_is_host_link(link
))
2597 ata_eh_thaw_port(ap
);
2607 if (ata_is_host_link(link
))
2608 ata_eh_freeze_port(ap
);
2610 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2614 ata_link_printk(link
, KERN_INFO
, "%s resetting link\n",
2615 reset
== softreset
? "soft" : "hard");
2617 /* mark that this EH session started with reset */
2618 ehc
->last_reset
= jiffies
;
2619 if (reset
== hardreset
)
2620 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2622 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2624 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2625 if (rc
&& rc
!= -EAGAIN
) {
2630 /* hardreset slave link if existent */
2631 if (slave
&& reset
== hardreset
) {
2635 ata_link_printk(slave
, KERN_INFO
,
2636 "hard resetting link\n");
2638 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2639 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2647 failed_link
= slave
;
2653 /* perform follow-up SRST if necessary */
2654 if (reset
== hardreset
&&
2655 ata_eh_followup_srst_needed(link
, rc
, classes
)) {
2659 ata_link_printk(link
, KERN_ERR
,
2660 "follow-up softreset required "
2661 "but no softreset avaliable\n");
2667 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2668 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2676 ata_link_printk(link
, KERN_INFO
, "no reset method "
2677 "available, skipping reset\n");
2678 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2679 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2683 * Post-reset processing
2685 ata_for_each_dev(dev
, link
, ALL
) {
2686 /* After the reset, the device state is PIO 0 and the
2687 * controller state is undefined. Reset also wakes up
2688 * drives from sleeping mode.
2690 dev
->pio_mode
= XFER_PIO_0
;
2691 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2693 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
2696 /* apply class override */
2697 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2698 classes
[dev
->devno
] = ATA_DEV_ATA
;
2699 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2700 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2703 /* record current link speed */
2704 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2705 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2706 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2707 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2710 if (ata_is_host_link(link
))
2711 ata_eh_thaw_port(ap
);
2713 /* postreset() should clear hardware SError. Although SError
2714 * is cleared during link resume, clearing SError here is
2715 * necessary as some PHYs raise hotplug events after SRST.
2716 * This introduces race condition where hotplug occurs between
2717 * reset and here. This race is mediated by cross checking
2718 * link onlineness and classification result later.
2721 postreset(link
, classes
);
2723 postreset(slave
, classes
);
2727 * Some controllers can't be frozen very well and may set
2728 * spuruious error conditions during reset. Clear accumulated
2729 * error information. As reset is the final recovery action,
2730 * nothing is lost by doing this.
2732 spin_lock_irqsave(link
->ap
->lock
, flags
);
2733 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
2735 memset(&slave
->eh_info
, 0, sizeof(link
->eh_info
));
2736 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
2737 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2740 * Make sure onlineness and classification result correspond.
2741 * Hotplug could have happened during reset and some
2742 * controllers fail to wait while a drive is spinning up after
2743 * being hotplugged causing misdetection. By cross checking
2744 * link on/offlineness and classification result, those
2745 * conditions can be reliably detected and retried.
2748 ata_for_each_dev(dev
, link
, ALL
) {
2749 if (ata_phys_link_online(ata_dev_phys_link(dev
))) {
2750 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2751 ata_dev_printk(dev
, KERN_DEBUG
, "link online "
2752 "but device misclassifed\n");
2753 classes
[dev
->devno
] = ATA_DEV_NONE
;
2756 } else if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2757 if (ata_class_enabled(classes
[dev
->devno
]))
2758 ata_dev_printk(dev
, KERN_DEBUG
, "link offline, "
2759 "clearing class %d to NONE\n",
2760 classes
[dev
->devno
]);
2761 classes
[dev
->devno
] = ATA_DEV_NONE
;
2762 } else if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2763 ata_dev_printk(dev
, KERN_DEBUG
, "link status unknown, "
2764 "clearing UNKNOWN to NONE\n");
2765 classes
[dev
->devno
] = ATA_DEV_NONE
;
2769 if (classify
&& nr_unknown
) {
2770 if (try < max_tries
) {
2771 ata_link_printk(link
, KERN_WARNING
, "link online but "
2772 "%d devices misclassified, retrying\n",
2778 ata_link_printk(link
, KERN_WARNING
,
2779 "link online but %d devices misclassified, "
2780 "device detection might fail\n", nr_unknown
);
2783 /* reset successful, schedule revalidation */
2784 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2786 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
2787 ehc
->last_reset
= jiffies
; /* update to completion time */
2788 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2792 /* clear hotplug flag */
2793 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2795 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2797 spin_lock_irqsave(ap
->lock
, flags
);
2798 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2799 spin_unlock_irqrestore(ap
->lock
, flags
);
2804 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2805 if (!ata_is_host_link(link
) &&
2806 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2809 if (rc
== -ERESTART
|| try >= max_tries
)
2813 if (time_before(now
, deadline
)) {
2814 unsigned long delta
= deadline
- now
;
2816 ata_link_printk(failed_link
, KERN_WARNING
,
2817 "reset failed (errno=%d), retrying in %u secs\n",
2818 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2821 delta
= schedule_timeout_uninterruptible(delta
);
2824 if (try == max_tries
- 1) {
2825 sata_down_spd_limit(link
, 0);
2827 sata_down_spd_limit(slave
, 0);
2828 } else if (rc
== -EPIPE
)
2829 sata_down_spd_limit(failed_link
, 0);
2836 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
2838 struct ata_link
*link
;
2839 struct ata_device
*dev
;
2840 unsigned long flags
;
2843 * This function can be thought of as an extended version of
2844 * ata_eh_about_to_do() specially crafted to accommodate the
2845 * requirements of ATA_EH_PARK handling. Since the EH thread
2846 * does not leave the do {} while () loop in ata_eh_recover as
2847 * long as the timeout for a park request to *one* device on
2848 * the port has not expired, and since we still want to pick
2849 * up park requests to other devices on the same port or
2850 * timeout updates for the same device, we have to pull
2851 * ATA_EH_PARK actions from eh_info into eh_context.i
2852 * ourselves at the beginning of each pass over the loop.
2854 * Additionally, all write accesses to &ap->park_req_pending
2855 * through INIT_COMPLETION() (see below) or complete_all()
2856 * (see ata_scsi_park_store()) are protected by the host lock.
2857 * As a result we have that park_req_pending.done is zero on
2858 * exit from this function, i.e. when ATA_EH_PARK actions for
2859 * *all* devices on port ap have been pulled into the
2860 * respective eh_context structs. If, and only if,
2861 * park_req_pending.done is non-zero by the time we reach
2862 * wait_for_completion_timeout(), another ATA_EH_PARK action
2863 * has been scheduled for at least one of the devices on port
2864 * ap and we have to cycle over the do {} while () loop in
2865 * ata_eh_recover() again.
2868 spin_lock_irqsave(ap
->lock
, flags
);
2869 INIT_COMPLETION(ap
->park_req_pending
);
2870 ata_for_each_link(link
, ap
, EDGE
) {
2871 ata_for_each_dev(dev
, link
, ALL
) {
2872 struct ata_eh_info
*ehi
= &link
->eh_info
;
2874 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
2875 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
2876 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
2879 spin_unlock_irqrestore(ap
->lock
, flags
);
2882 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
2884 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2885 struct ata_taskfile tf
;
2886 unsigned int err_mask
;
2888 ata_tf_init(dev
, &tf
);
2890 ehc
->unloaded_mask
|= 1 << dev
->devno
;
2891 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
2897 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2898 tf
.command
= ATA_CMD_CHK_POWER
;
2901 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
2902 tf
.protocol
|= ATA_PROT_NODATA
;
2903 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
2904 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
2905 ata_dev_printk(dev
, KERN_ERR
, "head unload failed!\n");
2906 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2910 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
2911 struct ata_device
**r_failed_dev
)
2913 struct ata_port
*ap
= link
->ap
;
2914 struct ata_eh_context
*ehc
= &link
->eh_context
;
2915 struct ata_device
*dev
;
2916 unsigned int new_mask
= 0;
2917 unsigned long flags
;
2922 /* For PATA drive side cable detection to work, IDENTIFY must
2923 * be done backwards such that PDIAG- is released by the slave
2924 * device before the master device is identified.
2926 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
2927 unsigned int action
= ata_eh_dev_action(dev
);
2928 unsigned int readid_flags
= 0;
2930 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
2931 readid_flags
|= ATA_READID_POSTRESET
;
2933 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
2934 WARN_ON(dev
->class == ATA_DEV_PMP
);
2936 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2941 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
2942 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
2947 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
2949 /* Configuration may have changed, reconfigure
2952 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2954 /* schedule the scsi_rescan_device() here */
2955 schedule_work(&(ap
->scsi_rescan_task
));
2956 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
2957 ehc
->tries
[dev
->devno
] &&
2958 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
2959 /* Temporarily set dev->class, it will be
2960 * permanently set once all configurations are
2961 * complete. This is necessary because new
2962 * device configuration is done in two
2965 dev
->class = ehc
->classes
[dev
->devno
];
2967 if (dev
->class == ATA_DEV_PMP
)
2968 rc
= sata_pmp_attach(dev
);
2970 rc
= ata_dev_read_id(dev
, &dev
->class,
2971 readid_flags
, dev
->id
);
2973 /* read_id might have changed class, store and reset */
2974 ehc
->classes
[dev
->devno
] = dev
->class;
2975 dev
->class = ATA_DEV_UNKNOWN
;
2979 /* clear error info accumulated during probe */
2980 ata_ering_clear(&dev
->ering
);
2981 new_mask
|= 1 << dev
->devno
;
2984 /* IDENTIFY was issued to non-existent
2985 * device. No need to reset. Just
2986 * thaw and ignore the device.
2988 ata_eh_thaw_port(ap
);
2996 /* PDIAG- should have been released, ask cable type if post-reset */
2997 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
2998 if (ap
->ops
->cable_detect
)
2999 ap
->cbl
= ap
->ops
->cable_detect(ap
);
3003 /* Configure new devices forward such that user doesn't see
3004 * device detection messages backwards.
3006 ata_for_each_dev(dev
, link
, ALL
) {
3007 if (!(new_mask
& (1 << dev
->devno
)))
3010 dev
->class = ehc
->classes
[dev
->devno
];
3012 if (dev
->class == ATA_DEV_PMP
)
3015 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
3016 rc
= ata_dev_configure(dev
);
3017 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
3019 dev
->class = ATA_DEV_UNKNOWN
;
3023 spin_lock_irqsave(ap
->lock
, flags
);
3024 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
3025 spin_unlock_irqrestore(ap
->lock
, flags
);
3027 /* new device discovered, configure xfermode */
3028 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3034 *r_failed_dev
= dev
;
3035 DPRINTK("EXIT rc=%d\n", rc
);
3040 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3041 * @link: link on which timings will be programmed
3042 * @r_failed_dev: out parameter for failed device
3044 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3045 * ata_set_mode() fails, pointer to the failing device is
3046 * returned in @r_failed_dev.
3049 * PCI/etc. bus probe sem.
3052 * 0 on success, negative errno otherwise
3054 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
3056 struct ata_port
*ap
= link
->ap
;
3057 struct ata_device
*dev
;
3060 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3061 ata_for_each_dev(dev
, link
, ENABLED
) {
3062 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
3063 struct ata_ering_entry
*ent
;
3065 ent
= ata_ering_top(&dev
->ering
);
3067 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
3071 /* has private set_mode? */
3072 if (ap
->ops
->set_mode
)
3073 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
3075 rc
= ata_do_set_mode(link
, r_failed_dev
);
3077 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3078 ata_for_each_dev(dev
, link
, ENABLED
) {
3079 struct ata_eh_context
*ehc
= &link
->eh_context
;
3080 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
3081 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
3083 if (dev
->xfer_mode
!= saved_xfer_mode
||
3084 ata_ncq_enabled(dev
) != saved_ncq
)
3085 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
3092 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3093 * @dev: ATAPI device to clear UA for
3095 * Resets and other operations can make an ATAPI device raise
3096 * UNIT ATTENTION which causes the next operation to fail. This
3097 * function clears UA.
3100 * EH context (may sleep).
3103 * 0 on success, -errno on failure.
3105 static int atapi_eh_clear_ua(struct ata_device
*dev
)
3109 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
3110 u8
*sense_buffer
= dev
->link
->ap
->sector_buf
;
3112 unsigned int err_mask
;
3114 err_mask
= atapi_eh_tur(dev
, &sense_key
);
3115 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
3116 ata_dev_printk(dev
, KERN_WARNING
, "TEST_UNIT_READY "
3117 "failed (err_mask=0x%x)\n", err_mask
);
3121 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
3124 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
3126 ata_dev_printk(dev
, KERN_WARNING
, "failed to clear "
3127 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
3132 ata_dev_printk(dev
, KERN_WARNING
,
3133 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES
);
3139 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3140 * @dev: ATA device which may need FLUSH retry
3142 * If @dev failed FLUSH, it needs to be reported upper layer
3143 * immediately as it means that @dev failed to remap and already
3144 * lost at least a sector and further FLUSH retrials won't make
3145 * any difference to the lost sector. However, if FLUSH failed
3146 * for other reasons, for example transmission error, FLUSH needs
3149 * This function determines whether FLUSH failure retry is
3150 * necessary and performs it if so.
3153 * 0 if EH can continue, -errno if EH needs to be repeated.
3155 static int ata_eh_maybe_retry_flush(struct ata_device
*dev
)
3157 struct ata_link
*link
= dev
->link
;
3158 struct ata_port
*ap
= link
->ap
;
3159 struct ata_queued_cmd
*qc
;
3160 struct ata_taskfile tf
;
3161 unsigned int err_mask
;
3164 /* did flush fail for this device? */
3165 if (!ata_tag_valid(link
->active_tag
))
3168 qc
= __ata_qc_from_tag(ap
, link
->active_tag
);
3169 if (qc
->dev
!= dev
|| (qc
->tf
.command
!= ATA_CMD_FLUSH_EXT
&&
3170 qc
->tf
.command
!= ATA_CMD_FLUSH
))
3173 /* if the device failed it, it should be reported to upper layers */
3174 if (qc
->err_mask
& AC_ERR_DEV
)
3177 /* flush failed for some other reason, give it another shot */
3178 ata_tf_init(dev
, &tf
);
3180 tf
.command
= qc
->tf
.command
;
3181 tf
.flags
|= ATA_TFLAG_DEVICE
;
3182 tf
.protocol
= ATA_PROT_NODATA
;
3184 ata_dev_printk(dev
, KERN_WARNING
, "retrying FLUSH 0x%x Emask 0x%x\n",
3185 tf
.command
, qc
->err_mask
);
3187 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3190 * FLUSH is complete but there's no way to
3191 * successfully complete a failed command from EH.
3192 * Making sure retry is allowed at least once and
3193 * retrying it should do the trick - whatever was in
3194 * the cache is already on the platter and this won't
3195 * cause infinite loop.
3197 qc
->scsicmd
->allowed
= max(qc
->scsicmd
->allowed
, 1);
3199 ata_dev_printk(dev
, KERN_WARNING
, "FLUSH failed Emask 0x%x\n",
3203 /* if device failed it, report it to upper layers */
3204 if (err_mask
& AC_ERR_DEV
) {
3205 qc
->err_mask
|= AC_ERR_DEV
;
3207 if (!(ap
->pflags
& ATA_PFLAG_FROZEN
))
3214 static int ata_link_nr_enabled(struct ata_link
*link
)
3216 struct ata_device
*dev
;
3219 ata_for_each_dev(dev
, link
, ENABLED
)
3224 static int ata_link_nr_vacant(struct ata_link
*link
)
3226 struct ata_device
*dev
;
3229 ata_for_each_dev(dev
, link
, ALL
)
3230 if (dev
->class == ATA_DEV_UNKNOWN
)
3235 static int ata_eh_skip_recovery(struct ata_link
*link
)
3237 struct ata_port
*ap
= link
->ap
;
3238 struct ata_eh_context
*ehc
= &link
->eh_context
;
3239 struct ata_device
*dev
;
3241 /* skip disabled links */
3242 if (link
->flags
& ATA_LFLAG_DISABLED
)
3245 /* skip if explicitly requested */
3246 if (ehc
->i
.flags
& ATA_EHI_NO_RECOVERY
)
3249 /* thaw frozen port and recover failed devices */
3250 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
3253 /* reset at least once if reset is requested */
3254 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
3255 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
3258 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3259 ata_for_each_dev(dev
, link
, ALL
) {
3260 if (dev
->class == ATA_DEV_UNKNOWN
&&
3261 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
3268 static int ata_count_probe_trials_cb(struct ata_ering_entry
*ent
, void *void_arg
)
3270 u64 interval
= msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL
);
3271 u64 now
= get_jiffies_64();
3272 int *trials
= void_arg
;
3274 if (ent
->timestamp
< now
- min(now
, interval
))
3281 static int ata_eh_schedule_probe(struct ata_device
*dev
)
3283 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3284 struct ata_link
*link
= ata_dev_phys_link(dev
);
3287 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
3288 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
3291 ata_eh_detach_dev(dev
);
3293 ehc
->did_probe_mask
|= (1 << dev
->devno
);
3294 ehc
->i
.action
|= ATA_EH_RESET
;
3295 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
3296 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
3298 /* Record and count probe trials on the ering. The specific
3299 * error mask used is irrelevant. Because a successful device
3300 * detection clears the ering, this count accumulates only if
3301 * there are consecutive failed probes.
3303 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3304 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3305 * forced to 1.5Gbps.
3307 * This is to work around cases where failed link speed
3308 * negotiation results in device misdetection leading to
3309 * infinite DEVXCHG or PHRDY CHG events.
3311 ata_ering_record(&dev
->ering
, 0, AC_ERR_OTHER
);
3312 ata_ering_map(&dev
->ering
, ata_count_probe_trials_cb
, &trials
);
3314 if (trials
> ATA_EH_PROBE_TRIALS
)
3315 sata_down_spd_limit(link
, 1);
3320 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
3322 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3324 /* -EAGAIN from EH routine indicates retry without prejudice.
3325 * The requester is responsible for ensuring forward progress.
3328 ehc
->tries
[dev
->devno
]--;
3332 /* device missing or wrong IDENTIFY data, schedule probing */
3333 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
3335 /* give it just one more chance */
3336 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
3338 if (ehc
->tries
[dev
->devno
] == 1) {
3339 /* This is the last chance, better to slow
3340 * down than lose it.
3342 sata_down_spd_limit(ata_dev_phys_link(dev
), 0);
3343 if (dev
->pio_mode
> XFER_PIO_0
)
3344 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
3348 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
3349 /* disable device if it has used up all its chances */
3350 ata_dev_disable(dev
);
3352 /* detach if offline */
3353 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
3354 ata_eh_detach_dev(dev
);
3356 /* schedule probe if necessary */
3357 if (ata_eh_schedule_probe(dev
)) {
3358 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3359 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
3360 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
3365 ehc
->i
.action
|= ATA_EH_RESET
;
3371 * ata_eh_recover - recover host port after error
3372 * @ap: host port to recover
3373 * @prereset: prereset method (can be NULL)
3374 * @softreset: softreset method (can be NULL)
3375 * @hardreset: hardreset method (can be NULL)
3376 * @postreset: postreset method (can be NULL)
3377 * @r_failed_link: out parameter for failed link
3379 * This is the alpha and omega, eum and yang, heart and soul of
3380 * libata exception handling. On entry, actions required to
3381 * recover each link and hotplug requests are recorded in the
3382 * link's eh_context. This function executes all the operations
3383 * with appropriate retrials and fallbacks to resurrect failed
3384 * devices, detach goners and greet newcomers.
3387 * Kernel thread context (may sleep).
3390 * 0 on success, -errno on failure.
3392 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3393 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3394 ata_postreset_fn_t postreset
,
3395 struct ata_link
**r_failed_link
)
3397 struct ata_link
*link
;
3398 struct ata_device
*dev
;
3401 unsigned long flags
, deadline
;
3405 /* prep for recovery */
3406 ata_for_each_link(link
, ap
, EDGE
) {
3407 struct ata_eh_context
*ehc
= &link
->eh_context
;
3409 /* re-enable link? */
3410 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3411 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3412 spin_lock_irqsave(ap
->lock
, flags
);
3413 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3414 spin_unlock_irqrestore(ap
->lock
, flags
);
3415 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3418 ata_for_each_dev(dev
, link
, ALL
) {
3419 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3420 ehc
->tries
[dev
->devno
] = 1;
3422 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3424 /* collect port action mask recorded in dev actions */
3425 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3426 ~ATA_EH_PERDEV_MASK
;
3427 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3429 /* process hotplug request */
3430 if (dev
->flags
& ATA_DFLAG_DETACH
)
3431 ata_eh_detach_dev(dev
);
3433 /* schedule probe if necessary */
3434 if (!ata_dev_enabled(dev
))
3435 ata_eh_schedule_probe(dev
);
3443 /* if UNLOADING, finish immediately */
3444 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3448 ata_for_each_link(link
, ap
, EDGE
) {
3449 struct ata_eh_context
*ehc
= &link
->eh_context
;
3451 /* skip EH if possible. */
3452 if (ata_eh_skip_recovery(link
))
3455 ata_for_each_dev(dev
, link
, ALL
)
3456 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3460 ata_for_each_link(link
, ap
, EDGE
) {
3461 struct ata_eh_context
*ehc
= &link
->eh_context
;
3463 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3466 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3467 prereset
, softreset
, hardreset
, postreset
);
3469 ata_link_printk(link
, KERN_ERR
,
3470 "reset failed, giving up\n");
3479 * clears ATA_EH_PARK in eh_info and resets
3480 * ap->park_req_pending
3482 ata_eh_pull_park_action(ap
);
3485 ata_for_each_link(link
, ap
, EDGE
) {
3486 ata_for_each_dev(dev
, link
, ALL
) {
3487 struct ata_eh_context
*ehc
= &link
->eh_context
;
3490 if (dev
->class != ATA_DEV_ATA
)
3492 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3495 tmp
= dev
->unpark_deadline
;
3496 if (time_before(deadline
, tmp
))
3498 else if (time_before_eq(tmp
, jiffies
))
3500 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3503 ata_eh_park_issue_cmd(dev
, 1);
3508 if (time_before_eq(deadline
, now
))
3511 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3514 ata_for_each_link(link
, ap
, EDGE
) {
3515 ata_for_each_dev(dev
, link
, ALL
) {
3516 if (!(link
->eh_context
.unloaded_mask
&
3520 ata_eh_park_issue_cmd(dev
, 0);
3521 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3526 ata_for_each_link(link
, ap
, EDGE
) {
3527 struct ata_eh_context
*ehc
= &link
->eh_context
;
3529 /* revalidate existing devices and attach new ones */
3530 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3534 /* if PMP got attached, return, pmp EH will take care of it */
3535 if (link
->device
->class == ATA_DEV_PMP
) {
3540 /* configure transfer mode if necessary */
3541 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3542 rc
= ata_set_mode(link
, &dev
);
3545 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3548 /* If reset has been issued, clear UA to avoid
3549 * disrupting the current users of the device.
3551 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3552 ata_for_each_dev(dev
, link
, ALL
) {
3553 if (dev
->class != ATA_DEV_ATAPI
)
3555 rc
= atapi_eh_clear_ua(dev
);
3561 /* retry flush if necessary */
3562 ata_for_each_dev(dev
, link
, ALL
) {
3563 if (dev
->class != ATA_DEV_ATA
)
3565 rc
= ata_eh_maybe_retry_flush(dev
);
3570 /* configure link power saving */
3571 if (ehc
->i
.action
& ATA_EH_LPM
)
3572 ata_for_each_dev(dev
, link
, ALL
)
3573 ata_dev_enable_pm(dev
, ap
->pm_policy
);
3575 /* this link is okay now */
3581 ata_eh_handle_dev_fail(dev
, rc
);
3583 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3584 /* PMP reset requires working host port.
3585 * Can't retry if it's frozen.
3587 if (sata_pmp_attached(ap
))
3597 if (rc
&& r_failed_link
)
3598 *r_failed_link
= link
;
3600 DPRINTK("EXIT, rc=%d\n", rc
);
3605 * ata_eh_finish - finish up EH
3606 * @ap: host port to finish EH for
3608 * Recovery is complete. Clean up EH states and retry or finish
3614 void ata_eh_finish(struct ata_port
*ap
)
3618 /* retry or finish qcs */
3619 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
3620 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
3622 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
3626 /* FIXME: Once EH migration is complete,
3627 * generate sense data in this function,
3628 * considering both err_mask and tf.
3630 if (qc
->flags
& ATA_QCFLAG_RETRY
)
3631 ata_eh_qc_retry(qc
);
3633 ata_eh_qc_complete(qc
);
3635 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
3636 ata_eh_qc_complete(qc
);
3638 /* feed zero TF to sense generation */
3639 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
3640 ata_eh_qc_retry(qc
);
3645 /* make sure nr_active_links is zero after EH */
3646 WARN_ON(ap
->nr_active_links
);
3647 ap
->nr_active_links
= 0;
3651 * ata_do_eh - do standard error handling
3652 * @ap: host port to handle error for
3654 * @prereset: prereset method (can be NULL)
3655 * @softreset: softreset method (can be NULL)
3656 * @hardreset: hardreset method (can be NULL)
3657 * @postreset: postreset method (can be NULL)
3659 * Perform standard error handling sequence.
3662 * Kernel thread context (may sleep).
3664 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3665 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3666 ata_postreset_fn_t postreset
)
3668 struct ata_device
*dev
;
3674 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
3677 ata_for_each_dev(dev
, &ap
->link
, ALL
)
3678 ata_dev_disable(dev
);
3685 * ata_std_error_handler - standard error handler
3686 * @ap: host port to handle error for
3688 * Standard error handler
3691 * Kernel thread context (may sleep).
3693 void ata_std_error_handler(struct ata_port
*ap
)
3695 struct ata_port_operations
*ops
= ap
->ops
;
3696 ata_reset_fn_t hardreset
= ops
->hardreset
;
3698 /* ignore built-in hardreset if SCR access is not available */
3699 if (hardreset
== sata_std_hardreset
&& !sata_scr_valid(&ap
->link
))
3702 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
3707 * ata_eh_handle_port_suspend - perform port suspend operation
3708 * @ap: port to suspend
3713 * Kernel thread context (may sleep).
3715 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
3717 unsigned long flags
;
3720 /* are we suspending? */
3721 spin_lock_irqsave(ap
->lock
, flags
);
3722 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3723 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
3724 spin_unlock_irqrestore(ap
->lock
, flags
);
3727 spin_unlock_irqrestore(ap
->lock
, flags
);
3729 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
3731 /* tell ACPI we're suspending */
3732 rc
= ata_acpi_on_suspend(ap
);
3737 ata_eh_freeze_port(ap
);
3739 if (ap
->ops
->port_suspend
)
3740 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
3742 ata_acpi_set_state(ap
, PMSG_SUSPEND
);
3745 spin_lock_irqsave(ap
->lock
, flags
);
3747 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
3749 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
3750 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
3751 ata_port_schedule_eh(ap
);
3753 if (ap
->pm_result
) {
3754 *ap
->pm_result
= rc
;
3755 ap
->pm_result
= NULL
;
3758 spin_unlock_irqrestore(ap
->lock
, flags
);
3764 * ata_eh_handle_port_resume - perform port resume operation
3765 * @ap: port to resume
3770 * Kernel thread context (may sleep).
3772 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
3774 struct ata_link
*link
;
3775 struct ata_device
*dev
;
3776 unsigned long flags
;
3779 /* are we resuming? */
3780 spin_lock_irqsave(ap
->lock
, flags
);
3781 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3782 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
3783 spin_unlock_irqrestore(ap
->lock
, flags
);
3786 spin_unlock_irqrestore(ap
->lock
, flags
);
3788 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
3791 * Error timestamps are in jiffies which doesn't run while
3792 * suspended and PHY events during resume isn't too uncommon.
3793 * When the two are combined, it can lead to unnecessary speed
3794 * downs if the machine is suspended and resumed repeatedly.
3795 * Clear error history.
3797 ata_for_each_link(link
, ap
, HOST_FIRST
)
3798 ata_for_each_dev(dev
, link
, ALL
)
3799 ata_ering_clear(&dev
->ering
);
3801 ata_acpi_set_state(ap
, PMSG_ON
);
3803 if (ap
->ops
->port_resume
)
3804 rc
= ap
->ops
->port_resume(ap
);
3806 /* tell ACPI that we're resuming */
3807 ata_acpi_on_resume(ap
);
3810 spin_lock_irqsave(ap
->lock
, flags
);
3811 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
3812 if (ap
->pm_result
) {
3813 *ap
->pm_result
= rc
;
3814 ap
->pm_result
= NULL
;
3816 spin_unlock_irqrestore(ap
->lock
, flags
);
3818 #endif /* CONFIG_PM */