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_transport_api.h"
45 #include <linux/libata.h>
50 /* speed down verdicts */
51 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
54 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
57 ATA_EFLAG_IS_IO
= (1 << 0),
58 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
60 /* error categories */
63 ATA_ECAT_TOUT_HSM
= 2,
65 ATA_ECAT_DUBIOUS_NONE
= 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
71 ATA_EH_CMD_DFL_TIMEOUT
= 5000,
73 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN
= 5000,
76 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
81 ATA_EH_PRERESET_TIMEOUT
= 10000,
82 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
87 /* The following table determines how we sequence resets. Each entry
88 * represents timeout for that try. The first try can be soft or
89 * hardreset. All others are hardreset if available. In most cases
90 * the first reset w/ 10sec timeout should succeed. Following entries
91 * are mostly for error handling, hotplug and retarded devices.
93 static const unsigned long ata_eh_reset_timeouts
[] = {
94 10000, /* most drives spin up by 10sec */
95 10000, /* > 99% working drives spin up before 20sec */
96 35000, /* give > 30 secs of idleness for retarded devices */
97 5000, /* and sweet one last chance */
98 ULONG_MAX
, /* > 1 min has elapsed, give up */
101 static const unsigned long ata_eh_identify_timeouts
[] = {
102 5000, /* covers > 99% of successes and not too boring on failures */
103 10000, /* combined time till here is enough even for media access */
104 30000, /* for true idiots */
108 static const unsigned long ata_eh_other_timeouts
[] = {
109 5000, /* same rationale as identify timeout */
111 /* but no merciful 30sec for other commands, it just isn't worth it */
115 struct ata_eh_cmd_timeout_ent
{
117 const unsigned long *timeouts
;
120 /* The following table determines timeouts to use for EH internal
121 * commands. Each table entry is a command class and matches the
122 * commands the entry applies to and the timeout table to use.
124 * On the retry after a command timed out, the next timeout value from
125 * the table is used. If the table doesn't contain further entries,
126 * the last value is used.
128 * ehc->cmd_timeout_idx keeps track of which timeout to use per
129 * command class, so if SET_FEATURES times out on the first try, the
130 * next try will use the second timeout value only for that class.
132 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
133 static const struct ata_eh_cmd_timeout_ent
134 ata_eh_cmd_timeout_table
[ATA_EH_CMD_TIMEOUT_TABLE_SIZE
] = {
135 { .commands
= CMDS(ATA_CMD_ID_ATA
, ATA_CMD_ID_ATAPI
),
136 .timeouts
= ata_eh_identify_timeouts
, },
137 { .commands
= CMDS(ATA_CMD_READ_NATIVE_MAX
, ATA_CMD_READ_NATIVE_MAX_EXT
),
138 .timeouts
= ata_eh_other_timeouts
, },
139 { .commands
= CMDS(ATA_CMD_SET_MAX
, ATA_CMD_SET_MAX_EXT
),
140 .timeouts
= ata_eh_other_timeouts
, },
141 { .commands
= CMDS(ATA_CMD_SET_FEATURES
),
142 .timeouts
= ata_eh_other_timeouts
, },
143 { .commands
= CMDS(ATA_CMD_INIT_DEV_PARAMS
),
144 .timeouts
= ata_eh_other_timeouts
, },
148 static void __ata_port_freeze(struct ata_port
*ap
);
150 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
151 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
152 #else /* CONFIG_PM */
153 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
156 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
158 #endif /* CONFIG_PM */
160 static void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
, const char *fmt
,
163 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
164 ATA_EH_DESC_LEN
- ehi
->desc_len
,
169 * __ata_ehi_push_desc - push error description without adding separator
171 * @fmt: printf format string
173 * Format string according to @fmt and append it to @ehi->desc.
176 * spin_lock_irqsave(host lock)
178 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
183 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
188 * ata_ehi_push_desc - push error description with separator
190 * @fmt: printf format string
192 * Format string according to @fmt and append it to @ehi->desc.
193 * If @ehi->desc is not empty, ", " is added in-between.
196 * spin_lock_irqsave(host lock)
198 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
203 __ata_ehi_push_desc(ehi
, ", ");
206 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
211 * ata_ehi_clear_desc - clean error description
217 * spin_lock_irqsave(host lock)
219 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
226 * ata_port_desc - append port description
227 * @ap: target ATA port
228 * @fmt: printf format string
230 * Format string according to @fmt and append it to port
231 * description. If port description is not empty, " " is added
232 * in-between. This function is to be used while initializing
233 * ata_host. The description is printed on host registration.
238 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
242 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
244 if (ap
->link
.eh_info
.desc_len
)
245 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
248 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
255 * ata_port_pbar_desc - append PCI BAR description
256 * @ap: target ATA port
257 * @bar: target PCI BAR
258 * @offset: offset into PCI BAR
259 * @name: name of the area
261 * If @offset is negative, this function formats a string which
262 * contains the name, address, size and type of the BAR and
263 * appends it to the port description. If @offset is zero or
264 * positive, only name and offsetted address is appended.
269 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
272 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
274 unsigned long long start
, len
;
276 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
278 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
281 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
282 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
285 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
287 ata_port_desc(ap
, "%s 0x%llx", name
,
288 start
+ (unsigned long long)offset
);
291 #endif /* CONFIG_PCI */
293 static int ata_lookup_timeout_table(u8 cmd
)
297 for (i
= 0; i
< ATA_EH_CMD_TIMEOUT_TABLE_SIZE
; i
++) {
300 for (cur
= ata_eh_cmd_timeout_table
[i
].commands
; *cur
; cur
++)
309 * ata_internal_cmd_timeout - determine timeout for an internal command
310 * @dev: target device
311 * @cmd: internal command to be issued
313 * Determine timeout for internal command @cmd for @dev.
319 * Determined timeout.
321 unsigned long ata_internal_cmd_timeout(struct ata_device
*dev
, u8 cmd
)
323 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
324 int ent
= ata_lookup_timeout_table(cmd
);
328 return ATA_EH_CMD_DFL_TIMEOUT
;
330 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
331 return ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
];
335 * ata_internal_cmd_timed_out - notification for internal command timeout
336 * @dev: target device
337 * @cmd: internal command which timed out
339 * Notify EH that internal command @cmd for @dev timed out. This
340 * function should be called only for commands whose timeouts are
341 * determined using ata_internal_cmd_timeout().
346 void ata_internal_cmd_timed_out(struct ata_device
*dev
, u8 cmd
)
348 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
349 int ent
= ata_lookup_timeout_table(cmd
);
355 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
356 if (ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
+ 1] != ULONG_MAX
)
357 ehc
->cmd_timeout_idx
[dev
->devno
][ent
]++;
360 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
361 unsigned int err_mask
)
363 struct ata_ering_entry
*ent
;
368 ering
->cursor
%= ATA_ERING_SIZE
;
370 ent
= &ering
->ring
[ering
->cursor
];
371 ent
->eflags
= eflags
;
372 ent
->err_mask
= err_mask
;
373 ent
->timestamp
= get_jiffies_64();
376 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
378 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
385 static void ata_ering_clear(struct ata_ering
*ering
)
387 memset(ering
, 0, sizeof(*ering
));
390 static int ata_ering_map(struct ata_ering
*ering
,
391 int (*map_fn
)(struct ata_ering_entry
*, void *),
395 struct ata_ering_entry
*ent
;
399 ent
= &ering
->ring
[idx
];
402 rc
= map_fn(ent
, arg
);
405 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
406 } while (idx
!= ering
->cursor
);
411 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
413 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
415 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
418 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
419 struct ata_eh_info
*ehi
, unsigned int action
)
421 struct ata_device
*tdev
;
424 ehi
->action
&= ~action
;
425 ata_for_each_dev(tdev
, link
, ALL
)
426 ehi
->dev_action
[tdev
->devno
] &= ~action
;
428 /* doesn't make sense for port-wide EH actions */
429 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
431 /* break ehi->action into ehi->dev_action */
432 if (ehi
->action
& action
) {
433 ata_for_each_dev(tdev
, link
, ALL
)
434 ehi
->dev_action
[tdev
->devno
] |=
435 ehi
->action
& action
;
436 ehi
->action
&= ~action
;
439 /* turn off the specified per-dev action */
440 ehi
->dev_action
[dev
->devno
] &= ~action
;
445 * ata_scsi_timed_out - SCSI layer time out callback
446 * @cmd: timed out SCSI command
448 * Handles SCSI layer timeout. We race with normal completion of
449 * the qc for @cmd. If the qc is already gone, we lose and let
450 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
451 * timed out and EH should be invoked. Prevent ata_qc_complete()
452 * from finishing it by setting EH_SCHEDULED and return
455 * TODO: kill this function once old EH is gone.
458 * Called from timer context
461 * EH_HANDLED or EH_NOT_HANDLED
463 enum blk_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
465 struct Scsi_Host
*host
= cmd
->device
->host
;
466 struct ata_port
*ap
= ata_shost_to_port(host
);
468 struct ata_queued_cmd
*qc
;
469 enum blk_eh_timer_return ret
;
473 if (ap
->ops
->error_handler
) {
474 ret
= BLK_EH_NOT_HANDLED
;
478 ret
= BLK_EH_HANDLED
;
479 spin_lock_irqsave(ap
->lock
, flags
);
480 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
482 WARN_ON(qc
->scsicmd
!= cmd
);
483 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
484 qc
->err_mask
|= AC_ERR_TIMEOUT
;
485 ret
= BLK_EH_NOT_HANDLED
;
487 spin_unlock_irqrestore(ap
->lock
, flags
);
490 DPRINTK("EXIT, ret=%d\n", ret
);
494 static void ata_eh_unload(struct ata_port
*ap
)
496 struct ata_link
*link
;
497 struct ata_device
*dev
;
500 /* Restore SControl IPM and SPD for the next driver and
501 * disable attached devices.
503 ata_for_each_link(link
, ap
, PMP_FIRST
) {
504 sata_scr_write(link
, SCR_CONTROL
, link
->saved_scontrol
& 0xff0);
505 ata_for_each_dev(dev
, link
, ALL
)
506 ata_dev_disable(dev
);
509 /* freeze and set UNLOADED */
510 spin_lock_irqsave(ap
->lock
, flags
);
512 ata_port_freeze(ap
); /* won't be thawed */
513 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
; /* clear pending from freeze */
514 ap
->pflags
|= ATA_PFLAG_UNLOADED
;
516 spin_unlock_irqrestore(ap
->lock
, flags
);
520 * ata_scsi_error - SCSI layer error handler callback
521 * @host: SCSI host on which error occurred
523 * Handles SCSI-layer-thrown error events.
526 * Inherited from SCSI layer (none, can sleep)
531 void ata_scsi_error(struct Scsi_Host
*host
)
533 struct ata_port
*ap
= ata_shost_to_port(host
);
539 /* synchronize with port task */
540 ata_port_flush_task(ap
);
542 /* synchronize with host lock and sort out timeouts */
544 /* For new EH, all qcs are finished in one of three ways -
545 * normal completion, error completion, and SCSI timeout.
546 * Both cmpletions can race against SCSI timeout. When normal
547 * completion wins, the qc never reaches EH. When error
548 * completion wins, the qc has ATA_QCFLAG_FAILED set.
550 * When SCSI timeout wins, things are a bit more complex.
551 * Normal or error completion can occur after the timeout but
552 * before this point. In such cases, both types of
553 * completions are honored. A scmd is determined to have
554 * timed out iff its associated qc is active and not failed.
556 if (ap
->ops
->error_handler
) {
557 struct scsi_cmnd
*scmd
, *tmp
;
560 spin_lock_irqsave(ap
->lock
, flags
);
562 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
563 struct ata_queued_cmd
*qc
;
565 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
566 qc
= __ata_qc_from_tag(ap
, i
);
567 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
572 if (i
< ATA_MAX_QUEUE
) {
573 /* the scmd has an associated qc */
574 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
575 /* which hasn't failed yet, timeout */
576 qc
->err_mask
|= AC_ERR_TIMEOUT
;
577 qc
->flags
|= ATA_QCFLAG_FAILED
;
581 /* Normal completion occurred after
582 * SCSI timeout but before this point.
583 * Successfully complete it.
585 scmd
->retries
= scmd
->allowed
;
586 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
590 /* If we have timed out qcs. They belong to EH from
591 * this point but the state of the controller is
592 * unknown. Freeze the port to make sure the IRQ
593 * handler doesn't diddle with those qcs. This must
594 * be done atomically w.r.t. setting QCFLAG_FAILED.
597 __ata_port_freeze(ap
);
599 spin_unlock_irqrestore(ap
->lock
, flags
);
601 /* initialize eh_tries */
602 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
604 spin_unlock_wait(ap
->lock
);
607 /* invoke error handler */
608 if (ap
->ops
->error_handler
) {
609 struct ata_link
*link
;
611 /* kill fast drain timer */
612 del_timer_sync(&ap
->fastdrain_timer
);
614 /* process port resume request */
615 ata_eh_handle_port_resume(ap
);
617 /* fetch & clear EH info */
618 spin_lock_irqsave(ap
->lock
, flags
);
620 ata_for_each_link(link
, ap
, HOST_FIRST
) {
621 struct ata_eh_context
*ehc
= &link
->eh_context
;
622 struct ata_device
*dev
;
624 memset(&link
->eh_context
, 0, sizeof(link
->eh_context
));
625 link
->eh_context
.i
= link
->eh_info
;
626 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
628 ata_for_each_dev(dev
, link
, ENABLED
) {
629 int devno
= dev
->devno
;
631 ehc
->saved_xfer_mode
[devno
] = dev
->xfer_mode
;
632 if (ata_ncq_enabled(dev
))
633 ehc
->saved_ncq_enabled
|= 1 << devno
;
637 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
638 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
639 ap
->excl_link
= NULL
; /* don't maintain exclusion over EH */
641 spin_unlock_irqrestore(ap
->lock
, flags
);
643 /* invoke EH, skip if unloading or suspended */
644 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
645 ap
->ops
->error_handler(ap
);
647 /* if unloading, commence suicide */
648 if ((ap
->pflags
& ATA_PFLAG_UNLOADING
) &&
649 !(ap
->pflags
& ATA_PFLAG_UNLOADED
))
654 /* process port suspend request */
655 ata_eh_handle_port_suspend(ap
);
657 /* Exception might have happend after ->error_handler
658 * recovered the port but before this point. Repeat
661 spin_lock_irqsave(ap
->lock
, flags
);
663 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
664 if (--ap
->eh_tries
) {
665 spin_unlock_irqrestore(ap
->lock
, flags
);
668 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
669 "tries, giving up\n", ATA_EH_MAX_TRIES
);
670 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
673 /* this run is complete, make sure EH info is clear */
674 ata_for_each_link(link
, ap
, HOST_FIRST
)
675 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
677 /* Clear host_eh_scheduled while holding ap->lock such
678 * that if exception occurs after this point but
679 * before EH completion, SCSI midlayer will
682 host
->host_eh_scheduled
= 0;
684 spin_unlock_irqrestore(ap
->lock
, flags
);
686 WARN_ON(ata_qc_from_tag(ap
, ap
->link
.active_tag
) == NULL
);
687 ap
->ops
->eng_timeout(ap
);
690 /* finish or retry handled scmd's and clean up */
691 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
693 scsi_eh_flush_done_q(&ap
->eh_done_q
);
696 spin_lock_irqsave(ap
->lock
, flags
);
698 if (ap
->pflags
& ATA_PFLAG_LOADING
)
699 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
700 else if (ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
)
701 queue_delayed_work(ata_aux_wq
, &ap
->hotplug_task
, 0);
703 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
704 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
706 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
708 /* tell wait_eh that we're done */
709 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
710 wake_up_all(&ap
->eh_wait_q
);
712 spin_unlock_irqrestore(ap
->lock
, flags
);
718 * ata_port_wait_eh - Wait for the currently pending EH to complete
719 * @ap: Port to wait EH for
721 * Wait until the currently pending EH is complete.
724 * Kernel thread context (may sleep).
726 void ata_port_wait_eh(struct ata_port
*ap
)
732 spin_lock_irqsave(ap
->lock
, flags
);
734 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
735 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
736 spin_unlock_irqrestore(ap
->lock
, flags
);
738 spin_lock_irqsave(ap
->lock
, flags
);
740 finish_wait(&ap
->eh_wait_q
, &wait
);
742 spin_unlock_irqrestore(ap
->lock
, flags
);
744 /* make sure SCSI EH is complete */
745 if (scsi_host_in_recovery(ap
->scsi_host
)) {
751 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
756 /* count only non-internal commands */
757 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
758 if (ata_qc_from_tag(ap
, tag
))
764 void ata_eh_fastdrain_timerfn(unsigned long arg
)
766 struct ata_port
*ap
= (void *)arg
;
770 spin_lock_irqsave(ap
->lock
, flags
);
772 cnt
= ata_eh_nr_in_flight(ap
);
778 if (cnt
== ap
->fastdrain_cnt
) {
781 /* No progress during the last interval, tag all
782 * in-flight qcs as timed out and freeze the port.
784 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
785 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
787 qc
->err_mask
|= AC_ERR_TIMEOUT
;
792 /* some qcs have finished, give it another chance */
793 ap
->fastdrain_cnt
= cnt
;
794 ap
->fastdrain_timer
.expires
=
795 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
796 add_timer(&ap
->fastdrain_timer
);
800 spin_unlock_irqrestore(ap
->lock
, flags
);
804 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
805 * @ap: target ATA port
806 * @fastdrain: activate fast drain
808 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
809 * is non-zero and EH wasn't pending before. Fast drain ensures
810 * that EH kicks in in timely manner.
813 * spin_lock_irqsave(host lock)
815 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
819 /* already scheduled? */
820 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
823 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
828 /* do we have in-flight qcs? */
829 cnt
= ata_eh_nr_in_flight(ap
);
833 /* activate fast drain */
834 ap
->fastdrain_cnt
= cnt
;
835 ap
->fastdrain_timer
.expires
=
836 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
837 add_timer(&ap
->fastdrain_timer
);
841 * ata_qc_schedule_eh - schedule qc for error handling
842 * @qc: command to schedule error handling for
844 * Schedule error handling for @qc. EH will kick in as soon as
845 * other commands are drained.
848 * spin_lock_irqsave(host lock)
850 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
852 struct ata_port
*ap
= qc
->ap
;
854 WARN_ON(!ap
->ops
->error_handler
);
856 qc
->flags
|= ATA_QCFLAG_FAILED
;
857 ata_eh_set_pending(ap
, 1);
859 /* The following will fail if timeout has already expired.
860 * ata_scsi_error() takes care of such scmds on EH entry.
861 * Note that ATA_QCFLAG_FAILED is unconditionally set after
862 * this function completes.
864 blk_abort_request(qc
->scsicmd
->request
);
868 * ata_port_schedule_eh - schedule error handling without a qc
869 * @ap: ATA port to schedule EH for
871 * Schedule error handling for @ap. EH will kick in as soon as
872 * all commands are drained.
875 * spin_lock_irqsave(host lock)
877 void ata_port_schedule_eh(struct ata_port
*ap
)
879 WARN_ON(!ap
->ops
->error_handler
);
881 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
884 ata_eh_set_pending(ap
, 1);
885 scsi_schedule_eh(ap
->scsi_host
);
887 DPRINTK("port EH scheduled\n");
890 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
892 int tag
, nr_aborted
= 0;
894 WARN_ON(!ap
->ops
->error_handler
);
896 /* we're gonna abort all commands, no need for fast drain */
897 ata_eh_set_pending(ap
, 0);
899 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
900 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
902 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
903 qc
->flags
|= ATA_QCFLAG_FAILED
;
910 ata_port_schedule_eh(ap
);
916 * ata_link_abort - abort all qc's on the link
917 * @link: ATA link to abort qc's for
919 * Abort all active qc's active on @link and schedule EH.
922 * spin_lock_irqsave(host lock)
925 * Number of aborted qc's.
927 int ata_link_abort(struct ata_link
*link
)
929 return ata_do_link_abort(link
->ap
, link
);
933 * ata_port_abort - abort all qc's on the port
934 * @ap: ATA port to abort qc's for
936 * Abort all active qc's of @ap and schedule EH.
939 * spin_lock_irqsave(host_set lock)
942 * Number of aborted qc's.
944 int ata_port_abort(struct ata_port
*ap
)
946 return ata_do_link_abort(ap
, NULL
);
950 * __ata_port_freeze - freeze port
951 * @ap: ATA port to freeze
953 * This function is called when HSM violation or some other
954 * condition disrupts normal operation of the port. Frozen port
955 * is not allowed to perform any operation until the port is
956 * thawed, which usually follows a successful reset.
958 * ap->ops->freeze() callback can be used for freezing the port
959 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
960 * port cannot be frozen hardware-wise, the interrupt handler
961 * must ack and clear interrupts unconditionally while the port
965 * spin_lock_irqsave(host lock)
967 static void __ata_port_freeze(struct ata_port
*ap
)
969 WARN_ON(!ap
->ops
->error_handler
);
974 ap
->pflags
|= ATA_PFLAG_FROZEN
;
976 DPRINTK("ata%u port frozen\n", ap
->print_id
);
980 * ata_port_freeze - abort & freeze port
981 * @ap: ATA port to freeze
983 * Abort and freeze @ap.
986 * spin_lock_irqsave(host lock)
989 * Number of aborted commands.
991 int ata_port_freeze(struct ata_port
*ap
)
995 WARN_ON(!ap
->ops
->error_handler
);
997 nr_aborted
= ata_port_abort(ap
);
998 __ata_port_freeze(ap
);
1004 * sata_async_notification - SATA async notification handler
1005 * @ap: ATA port where async notification is received
1007 * Handler to be called when async notification via SDB FIS is
1008 * received. This function schedules EH if necessary.
1011 * spin_lock_irqsave(host lock)
1014 * 1 if EH is scheduled, 0 otherwise.
1016 int sata_async_notification(struct ata_port
*ap
)
1021 if (!(ap
->flags
& ATA_FLAG_AN
))
1024 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
1026 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
1028 if (!sata_pmp_attached(ap
) || rc
) {
1029 /* PMP is not attached or SNTF is not available */
1030 if (!sata_pmp_attached(ap
)) {
1031 /* PMP is not attached. Check whether ATAPI
1032 * AN is configured. If so, notify media
1035 struct ata_device
*dev
= ap
->link
.device
;
1037 if ((dev
->class == ATA_DEV_ATAPI
) &&
1038 (dev
->flags
& ATA_DFLAG_AN
))
1039 ata_scsi_media_change_notify(dev
);
1042 /* PMP is attached but SNTF is not available.
1043 * ATAPI async media change notification is
1044 * not used. The PMP must be reporting PHY
1045 * status change, schedule EH.
1047 ata_port_schedule_eh(ap
);
1051 /* PMP is attached and SNTF is available */
1052 struct ata_link
*link
;
1054 /* check and notify ATAPI AN */
1055 ata_for_each_link(link
, ap
, EDGE
) {
1056 if (!(sntf
& (1 << link
->pmp
)))
1059 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
1060 (link
->device
->flags
& ATA_DFLAG_AN
))
1061 ata_scsi_media_change_notify(link
->device
);
1064 /* If PMP is reporting that PHY status of some
1065 * downstream ports has changed, schedule EH.
1067 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
1068 ata_port_schedule_eh(ap
);
1077 * ata_eh_freeze_port - EH helper to freeze port
1078 * @ap: ATA port to freeze
1085 void ata_eh_freeze_port(struct ata_port
*ap
)
1087 unsigned long flags
;
1089 if (!ap
->ops
->error_handler
)
1092 spin_lock_irqsave(ap
->lock
, flags
);
1093 __ata_port_freeze(ap
);
1094 spin_unlock_irqrestore(ap
->lock
, flags
);
1098 * ata_port_thaw_port - EH helper to thaw port
1099 * @ap: ATA port to thaw
1101 * Thaw frozen port @ap.
1106 void ata_eh_thaw_port(struct ata_port
*ap
)
1108 unsigned long flags
;
1110 if (!ap
->ops
->error_handler
)
1113 spin_lock_irqsave(ap
->lock
, flags
);
1115 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
1120 spin_unlock_irqrestore(ap
->lock
, flags
);
1122 DPRINTK("ata%u port thawed\n", ap
->print_id
);
1125 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
1130 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1132 struct ata_port
*ap
= qc
->ap
;
1133 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1134 unsigned long flags
;
1136 spin_lock_irqsave(ap
->lock
, flags
);
1137 qc
->scsidone
= ata_eh_scsidone
;
1138 __ata_qc_complete(qc
);
1139 WARN_ON(ata_tag_valid(qc
->tag
));
1140 spin_unlock_irqrestore(ap
->lock
, flags
);
1142 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1146 * ata_eh_qc_complete - Complete an active ATA command from EH
1147 * @qc: Command to complete
1149 * Indicate to the mid and upper layers that an ATA command has
1150 * completed. To be used from EH.
1152 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1154 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1155 scmd
->retries
= scmd
->allowed
;
1156 __ata_eh_qc_complete(qc
);
1160 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1161 * @qc: Command to retry
1163 * Indicate to the mid and upper layers that an ATA command
1164 * should be retried. To be used from EH.
1166 * SCSI midlayer limits the number of retries to scmd->allowed.
1167 * scmd->retries is decremented for commands which get retried
1168 * due to unrelated failures (qc->err_mask is zero).
1170 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1172 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1173 if (!qc
->err_mask
&& scmd
->retries
)
1175 __ata_eh_qc_complete(qc
);
1179 * ata_eh_detach_dev - detach ATA device
1180 * @dev: ATA device to detach
1187 void ata_eh_detach_dev(struct ata_device
*dev
)
1189 struct ata_link
*link
= dev
->link
;
1190 struct ata_port
*ap
= link
->ap
;
1191 struct ata_eh_context
*ehc
= &link
->eh_context
;
1192 unsigned long flags
;
1194 ata_dev_disable(dev
);
1196 spin_lock_irqsave(ap
->lock
, flags
);
1198 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1200 if (ata_scsi_offline_dev(dev
)) {
1201 dev
->flags
|= ATA_DFLAG_DETACHED
;
1202 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1205 /* clear per-dev EH info */
1206 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1207 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1208 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
1209 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
1211 spin_unlock_irqrestore(ap
->lock
, flags
);
1215 * ata_eh_about_to_do - about to perform eh_action
1216 * @link: target ATA link
1217 * @dev: target ATA dev for per-dev action (can be NULL)
1218 * @action: action about to be performed
1220 * Called just before performing EH actions to clear related bits
1221 * in @link->eh_info such that eh actions are not unnecessarily
1227 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1228 unsigned int action
)
1230 struct ata_port
*ap
= link
->ap
;
1231 struct ata_eh_info
*ehi
= &link
->eh_info
;
1232 struct ata_eh_context
*ehc
= &link
->eh_context
;
1233 unsigned long flags
;
1235 spin_lock_irqsave(ap
->lock
, flags
);
1237 ata_eh_clear_action(link
, dev
, ehi
, action
);
1239 /* About to take EH action, set RECOVERED. Ignore actions on
1240 * slave links as master will do them again.
1242 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
) && link
!= ap
->slave_link
)
1243 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1245 spin_unlock_irqrestore(ap
->lock
, flags
);
1249 * ata_eh_done - EH action complete
1250 * @ap: target ATA port
1251 * @dev: target ATA dev for per-dev action (can be NULL)
1252 * @action: action just completed
1254 * Called right after performing EH actions to clear related bits
1255 * in @link->eh_context.
1260 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1261 unsigned int action
)
1263 struct ata_eh_context
*ehc
= &link
->eh_context
;
1265 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1269 * ata_err_string - convert err_mask to descriptive string
1270 * @err_mask: error mask to convert to string
1272 * Convert @err_mask to descriptive string. Errors are
1273 * prioritized according to severity and only the most severe
1274 * error is reported.
1280 * Descriptive string for @err_mask
1282 static const char *ata_err_string(unsigned int err_mask
)
1284 if (err_mask
& AC_ERR_HOST_BUS
)
1285 return "host bus error";
1286 if (err_mask
& AC_ERR_ATA_BUS
)
1287 return "ATA bus error";
1288 if (err_mask
& AC_ERR_TIMEOUT
)
1290 if (err_mask
& AC_ERR_HSM
)
1291 return "HSM violation";
1292 if (err_mask
& AC_ERR_SYSTEM
)
1293 return "internal error";
1294 if (err_mask
& AC_ERR_MEDIA
)
1295 return "media error";
1296 if (err_mask
& AC_ERR_INVALID
)
1297 return "invalid argument";
1298 if (err_mask
& AC_ERR_DEV
)
1299 return "device error";
1300 return "unknown error";
1304 * ata_read_log_page - read a specific log page
1305 * @dev: target device
1306 * @page: page to read
1307 * @buf: buffer to store read page
1308 * @sectors: number of sectors to read
1310 * Read log page using READ_LOG_EXT command.
1313 * Kernel thread context (may sleep).
1316 * 0 on success, AC_ERR_* mask otherwise.
1318 static unsigned int ata_read_log_page(struct ata_device
*dev
,
1319 u8 page
, void *buf
, unsigned int sectors
)
1321 struct ata_taskfile tf
;
1322 unsigned int err_mask
;
1324 DPRINTK("read log page - page %d\n", page
);
1326 ata_tf_init(dev
, &tf
);
1327 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1330 tf
.hob_nsect
= sectors
>> 8;
1331 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1332 tf
.protocol
= ATA_PROT_PIO
;
1334 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1335 buf
, sectors
* ATA_SECT_SIZE
, 0);
1337 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1342 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1343 * @dev: Device to read log page 10h from
1344 * @tag: Resulting tag of the failed command
1345 * @tf: Resulting taskfile registers of the failed command
1347 * Read log page 10h to obtain NCQ error details and clear error
1351 * Kernel thread context (may sleep).
1354 * 0 on success, -errno otherwise.
1356 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1357 int *tag
, struct ata_taskfile
*tf
)
1359 u8
*buf
= dev
->link
->ap
->sector_buf
;
1360 unsigned int err_mask
;
1364 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
1369 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1372 ata_dev_printk(dev
, KERN_WARNING
,
1373 "invalid checksum 0x%x on log page 10h\n", csum
);
1378 *tag
= buf
[0] & 0x1f;
1380 tf
->command
= buf
[2];
1381 tf
->feature
= buf
[3];
1385 tf
->device
= buf
[7];
1386 tf
->hob_lbal
= buf
[8];
1387 tf
->hob_lbam
= buf
[9];
1388 tf
->hob_lbah
= buf
[10];
1389 tf
->nsect
= buf
[12];
1390 tf
->hob_nsect
= buf
[13];
1396 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1397 * @dev: target ATAPI device
1398 * @r_sense_key: out parameter for sense_key
1400 * Perform ATAPI TEST_UNIT_READY.
1403 * EH context (may sleep).
1406 * 0 on success, AC_ERR_* mask on failure.
1408 static unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1410 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1411 struct ata_taskfile tf
;
1412 unsigned int err_mask
;
1414 ata_tf_init(dev
, &tf
);
1416 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1417 tf
.command
= ATA_CMD_PACKET
;
1418 tf
.protocol
= ATAPI_PROT_NODATA
;
1420 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1421 if (err_mask
== AC_ERR_DEV
)
1422 *r_sense_key
= tf
.feature
>> 4;
1427 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1428 * @dev: device to perform REQUEST_SENSE to
1429 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1430 * @dfl_sense_key: default sense key to use
1432 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1433 * SENSE. This function is EH helper.
1436 * Kernel thread context (may sleep).
1439 * 0 on success, AC_ERR_* mask on failure
1441 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1442 u8
*sense_buf
, u8 dfl_sense_key
)
1444 u8 cdb
[ATAPI_CDB_LEN
] =
1445 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1446 struct ata_port
*ap
= dev
->link
->ap
;
1447 struct ata_taskfile tf
;
1449 DPRINTK("ATAPI request sense\n");
1451 /* FIXME: is this needed? */
1452 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1454 /* initialize sense_buf with the error register,
1455 * for the case where they are -not- overwritten
1457 sense_buf
[0] = 0x70;
1458 sense_buf
[2] = dfl_sense_key
;
1460 /* some devices time out if garbage left in tf */
1461 ata_tf_init(dev
, &tf
);
1463 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1464 tf
.command
= ATA_CMD_PACKET
;
1466 /* is it pointless to prefer PIO for "safety reasons"? */
1467 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1468 tf
.protocol
= ATAPI_PROT_DMA
;
1469 tf
.feature
|= ATAPI_PKT_DMA
;
1471 tf
.protocol
= ATAPI_PROT_PIO
;
1472 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1476 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1477 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1481 * ata_eh_analyze_serror - analyze SError for a failed port
1482 * @link: ATA link to analyze SError for
1484 * Analyze SError if available and further determine cause of
1490 static void ata_eh_analyze_serror(struct ata_link
*link
)
1492 struct ata_eh_context
*ehc
= &link
->eh_context
;
1493 u32 serror
= ehc
->i
.serror
;
1494 unsigned int err_mask
= 0, action
= 0;
1497 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1498 err_mask
|= AC_ERR_ATA_BUS
;
1499 action
|= ATA_EH_RESET
;
1501 if (serror
& SERR_PROTOCOL
) {
1502 err_mask
|= AC_ERR_HSM
;
1503 action
|= ATA_EH_RESET
;
1505 if (serror
& SERR_INTERNAL
) {
1506 err_mask
|= AC_ERR_SYSTEM
;
1507 action
|= ATA_EH_RESET
;
1510 /* Determine whether a hotplug event has occurred. Both
1511 * SError.N/X are considered hotplug events for enabled or
1512 * host links. For disabled PMP links, only N bit is
1513 * considered as X bit is left at 1 for link plugging.
1517 if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1518 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1520 hotplug_mask
= SERR_PHYRDY_CHG
;
1522 if (serror
& hotplug_mask
)
1523 ata_ehi_hotplugged(&ehc
->i
);
1525 ehc
->i
.err_mask
|= err_mask
;
1526 ehc
->i
.action
|= action
;
1530 * ata_eh_analyze_ncq_error - analyze NCQ error
1531 * @link: ATA link to analyze NCQ error for
1533 * Read log page 10h, determine the offending qc and acquire
1534 * error status TF. For NCQ device errors, all LLDDs have to do
1535 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1539 * Kernel thread context (may sleep).
1541 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1543 struct ata_port
*ap
= link
->ap
;
1544 struct ata_eh_context
*ehc
= &link
->eh_context
;
1545 struct ata_device
*dev
= link
->device
;
1546 struct ata_queued_cmd
*qc
;
1547 struct ata_taskfile tf
;
1550 /* if frozen, we can't do much */
1551 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1554 /* is it NCQ device error? */
1555 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1558 /* has LLDD analyzed already? */
1559 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1560 qc
= __ata_qc_from_tag(ap
, tag
);
1562 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1569 /* okay, this error is ours */
1570 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1572 ata_link_printk(link
, KERN_ERR
, "failed to read log page 10h "
1573 "(errno=%d)\n", rc
);
1577 if (!(link
->sactive
& (1 << tag
))) {
1578 ata_link_printk(link
, KERN_ERR
, "log page 10h reported "
1579 "inactive tag %d\n", tag
);
1583 /* we've got the perpetrator, condemn it */
1584 qc
= __ata_qc_from_tag(ap
, tag
);
1585 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1586 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1587 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1588 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1592 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1593 * @qc: qc to analyze
1594 * @tf: Taskfile registers to analyze
1596 * Analyze taskfile of @qc and further determine cause of
1597 * failure. This function also requests ATAPI sense data if
1601 * Kernel thread context (may sleep).
1604 * Determined recovery action
1606 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1607 const struct ata_taskfile
*tf
)
1609 unsigned int tmp
, action
= 0;
1610 u8 stat
= tf
->command
, err
= tf
->feature
;
1612 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1613 qc
->err_mask
|= AC_ERR_HSM
;
1614 return ATA_EH_RESET
;
1617 if (stat
& (ATA_ERR
| ATA_DF
))
1618 qc
->err_mask
|= AC_ERR_DEV
;
1622 switch (qc
->dev
->class) {
1625 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1627 qc
->err_mask
|= AC_ERR_MEDIA
;
1629 qc
->err_mask
|= AC_ERR_INVALID
;
1633 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1634 tmp
= atapi_eh_request_sense(qc
->dev
,
1635 qc
->scsicmd
->sense_buffer
,
1636 qc
->result_tf
.feature
>> 4);
1638 /* ATA_QCFLAG_SENSE_VALID is used to
1639 * tell atapi_qc_complete() that sense
1640 * data is already valid.
1642 * TODO: interpret sense data and set
1643 * appropriate err_mask.
1645 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1647 qc
->err_mask
|= tmp
;
1651 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1652 action
|= ATA_EH_RESET
;
1657 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1662 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1666 base
= ATA_ECAT_DUBIOUS_NONE
;
1668 if (err_mask
& AC_ERR_ATA_BUS
)
1669 return base
+ ATA_ECAT_ATA_BUS
;
1671 if (err_mask
& AC_ERR_TIMEOUT
)
1672 return base
+ ATA_ECAT_TOUT_HSM
;
1674 if (eflags
& ATA_EFLAG_IS_IO
) {
1675 if (err_mask
& AC_ERR_HSM
)
1676 return base
+ ATA_ECAT_TOUT_HSM
;
1678 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1679 return base
+ ATA_ECAT_UNK_DEV
;
1685 struct speed_down_verdict_arg
{
1688 int nr_errors
[ATA_ECAT_NR
];
1691 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1693 struct speed_down_verdict_arg
*arg
= void_arg
;
1696 if (ent
->timestamp
< arg
->since
)
1699 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1701 arg
->nr_errors
[cat
]++;
1707 * ata_eh_speed_down_verdict - Determine speed down verdict
1708 * @dev: Device of interest
1710 * This function examines error ring of @dev and determines
1711 * whether NCQ needs to be turned off, transfer speed should be
1712 * stepped down, or falling back to PIO is necessary.
1714 * ECAT_ATA_BUS : ATA_BUS error for any command
1716 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1719 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1721 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1722 * data transfer hasn't been verified.
1726 * NCQ_OFF : Turn off NCQ.
1728 * SPEED_DOWN : Speed down transfer speed but don't fall back
1731 * FALLBACK_TO_PIO : Fall back to PIO.
1733 * Even if multiple verdicts are returned, only one action is
1734 * taken per error. An action triggered by non-DUBIOUS errors
1735 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1736 * This is to expedite speed down decisions right after device is
1737 * initially configured.
1739 * The followings are speed down rules. #1 and #2 deal with
1742 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1743 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1745 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1746 * occurred during last 5 mins, NCQ_OFF.
1748 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1749 * ocurred during last 5 mins, FALLBACK_TO_PIO
1751 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1752 * during last 10 mins, NCQ_OFF.
1754 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1755 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1758 * Inherited from caller.
1761 * OR of ATA_EH_SPDN_* flags.
1763 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1765 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1766 u64 j64
= get_jiffies_64();
1767 struct speed_down_verdict_arg arg
;
1768 unsigned int verdict
= 0;
1770 /* scan past 5 mins of error history */
1771 memset(&arg
, 0, sizeof(arg
));
1772 arg
.since
= j64
- min(j64
, j5mins
);
1773 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1775 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1776 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1777 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1778 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1780 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1781 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1782 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1784 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1785 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1786 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1787 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1789 /* scan past 10 mins of error history */
1790 memset(&arg
, 0, sizeof(arg
));
1791 arg
.since
= j64
- min(j64
, j10mins
);
1792 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1794 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1795 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1796 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1798 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1799 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1800 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1801 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1807 * ata_eh_speed_down - record error and speed down if necessary
1808 * @dev: Failed device
1809 * @eflags: mask of ATA_EFLAG_* flags
1810 * @err_mask: err_mask of the error
1812 * Record error and examine error history to determine whether
1813 * adjusting transmission speed is necessary. It also sets
1814 * transmission limits appropriately if such adjustment is
1818 * Kernel thread context (may sleep).
1821 * Determined recovery action.
1823 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
1824 unsigned int eflags
, unsigned int err_mask
)
1826 struct ata_link
*link
= ata_dev_phys_link(dev
);
1828 unsigned int verdict
;
1829 unsigned int action
= 0;
1831 /* don't bother if Cat-0 error */
1832 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
1835 /* record error and determine whether speed down is necessary */
1836 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
1837 verdict
= ata_eh_speed_down_verdict(dev
);
1840 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
1841 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
1842 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
1843 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
1844 ata_dev_printk(dev
, KERN_WARNING
,
1845 "NCQ disabled due to excessive errors\n");
1850 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
1851 /* speed down SATA link speed if possible */
1852 if (sata_down_spd_limit(link
) == 0) {
1853 action
|= ATA_EH_RESET
;
1857 /* lower transfer mode */
1858 if (dev
->spdn_cnt
< 2) {
1859 static const int dma_dnxfer_sel
[] =
1860 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
1861 static const int pio_dnxfer_sel
[] =
1862 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
1865 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
1866 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
1868 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
1872 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
1873 action
|= ATA_EH_RESET
;
1879 /* Fall back to PIO? Slowing down to PIO is meaningless for
1880 * SATA ATA devices. Consider it only for PATA and SATAPI.
1882 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
1883 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
1884 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
1885 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
1887 action
|= ATA_EH_RESET
;
1894 /* device has been slowed down, blow error history */
1895 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
1896 ata_ering_clear(&dev
->ering
);
1901 * ata_eh_link_autopsy - analyze error and determine recovery action
1902 * @link: host link to perform autopsy on
1904 * Analyze why @link failed and determine which recovery actions
1905 * are needed. This function also sets more detailed AC_ERR_*
1906 * values and fills sense data for ATAPI CHECK SENSE.
1909 * Kernel thread context (may sleep).
1911 static void ata_eh_link_autopsy(struct ata_link
*link
)
1913 struct ata_port
*ap
= link
->ap
;
1914 struct ata_eh_context
*ehc
= &link
->eh_context
;
1915 struct ata_device
*dev
;
1916 unsigned int all_err_mask
= 0, eflags
= 0;
1923 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
1926 /* obtain and analyze SError */
1927 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
1929 ehc
->i
.serror
|= serror
;
1930 ata_eh_analyze_serror(link
);
1931 } else if (rc
!= -EOPNOTSUPP
) {
1932 /* SError read failed, force reset and probing */
1933 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
1934 ehc
->i
.action
|= ATA_EH_RESET
;
1935 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
1938 /* analyze NCQ failure */
1939 ata_eh_analyze_ncq_error(link
);
1941 /* any real error trumps AC_ERR_OTHER */
1942 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1943 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1945 all_err_mask
|= ehc
->i
.err_mask
;
1947 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1948 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1950 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
1951 ata_dev_phys_link(qc
->dev
) != link
)
1954 /* inherit upper level err_mask */
1955 qc
->err_mask
|= ehc
->i
.err_mask
;
1958 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1960 /* DEV errors are probably spurious in case of ATA_BUS error */
1961 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1962 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1965 /* any real error trumps unknown error */
1966 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1967 qc
->err_mask
&= ~AC_ERR_OTHER
;
1969 /* SENSE_VALID trumps dev/unknown error and revalidation */
1970 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
1971 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1973 /* determine whether the command is worth retrying */
1974 if (!(qc
->err_mask
& AC_ERR_INVALID
) &&
1975 ((qc
->flags
& ATA_QCFLAG_IO
) || qc
->err_mask
!= AC_ERR_DEV
))
1976 qc
->flags
|= ATA_QCFLAG_RETRY
;
1978 /* accumulate error info */
1979 ehc
->i
.dev
= qc
->dev
;
1980 all_err_mask
|= qc
->err_mask
;
1981 if (qc
->flags
& ATA_QCFLAG_IO
)
1982 eflags
|= ATA_EFLAG_IS_IO
;
1985 /* enforce default EH actions */
1986 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
1987 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
1988 ehc
->i
.action
|= ATA_EH_RESET
;
1989 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
1990 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
1991 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1993 /* If we have offending qcs and the associated failed device,
1994 * perform per-dev EH action only on the offending device.
1997 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
1998 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
1999 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
2002 /* propagate timeout to host link */
2003 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
2004 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
2006 /* record error and consider speeding down */
2008 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
2009 ata_dev_enabled(link
->device
))))
2013 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
2014 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
2015 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
2022 * ata_eh_autopsy - analyze error and determine recovery action
2023 * @ap: host port to perform autopsy on
2025 * Analyze all links of @ap and determine why they failed and
2026 * which recovery actions are needed.
2029 * Kernel thread context (may sleep).
2031 void ata_eh_autopsy(struct ata_port
*ap
)
2033 struct ata_link
*link
;
2035 ata_for_each_link(link
, ap
, EDGE
)
2036 ata_eh_link_autopsy(link
);
2038 /* Handle the frigging slave link. Autopsy is done similarly
2039 * but actions and flags are transferred over to the master
2040 * link and handled from there.
2042 if (ap
->slave_link
) {
2043 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2044 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2046 /* transfer control flags from master to slave */
2047 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2049 /* perform autopsy on the slave link */
2050 ata_eh_link_autopsy(ap
->slave_link
);
2052 /* transfer actions from slave to master and clear slave */
2053 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2054 mehc
->i
.action
|= sehc
->i
.action
;
2055 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2056 mehc
->i
.flags
|= sehc
->i
.flags
;
2057 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2060 /* Autopsy of fanout ports can affect host link autopsy.
2061 * Perform host link autopsy last.
2063 if (sata_pmp_attached(ap
))
2064 ata_eh_link_autopsy(&ap
->link
);
2068 * ata_eh_link_report - report error handling to user
2069 * @link: ATA link EH is going on
2071 * Report EH to user.
2076 static void ata_eh_link_report(struct ata_link
*link
)
2078 struct ata_port
*ap
= link
->ap
;
2079 struct ata_eh_context
*ehc
= &link
->eh_context
;
2080 const char *frozen
, *desc
;
2082 int tag
, nr_failed
= 0;
2084 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2088 if (ehc
->i
.desc
[0] != '\0')
2091 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2092 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2094 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2095 ata_dev_phys_link(qc
->dev
) != link
||
2096 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2097 qc
->err_mask
== AC_ERR_DEV
))
2099 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2105 if (!nr_failed
&& !ehc
->i
.err_mask
)
2109 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2112 memset(tries_buf
, 0, sizeof(tries_buf
));
2113 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2114 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
2118 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
2119 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2120 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2121 ehc
->i
.action
, frozen
, tries_buf
);
2123 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "%s\n", desc
);
2125 ata_link_printk(link
, KERN_ERR
, "exception Emask 0x%x "
2126 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2127 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2128 ehc
->i
.action
, frozen
, tries_buf
);
2130 ata_link_printk(link
, KERN_ERR
, "%s\n", desc
);
2134 ata_link_printk(link
, KERN_ERR
,
2135 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2136 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2137 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2138 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2139 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2140 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2141 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2142 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2143 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2144 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2145 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2146 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2147 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2148 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2149 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2150 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2151 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2152 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2154 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2155 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2156 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2157 const u8
*cdb
= qc
->cdb
;
2158 char data_buf
[20] = "";
2159 char cdb_buf
[70] = "";
2161 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2162 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2165 if (qc
->dma_dir
!= DMA_NONE
) {
2166 static const char *dma_str
[] = {
2167 [DMA_BIDIRECTIONAL
] = "bidi",
2168 [DMA_TO_DEVICE
] = "out",
2169 [DMA_FROM_DEVICE
] = "in",
2171 static const char *prot_str
[] = {
2172 [ATA_PROT_PIO
] = "pio",
2173 [ATA_PROT_DMA
] = "dma",
2174 [ATA_PROT_NCQ
] = "ncq",
2175 [ATAPI_PROT_PIO
] = "pio",
2176 [ATAPI_PROT_DMA
] = "dma",
2179 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2180 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
2181 dma_str
[qc
->dma_dir
]);
2184 if (ata_is_atapi(qc
->tf
.protocol
))
2185 snprintf(cdb_buf
, sizeof(cdb_buf
),
2186 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2187 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2188 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
2189 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
2190 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
2191 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
2193 ata_dev_printk(qc
->dev
, KERN_ERR
,
2194 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2196 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2197 "Emask 0x%x (%s)%s\n",
2198 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2199 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2200 cmd
->hob_feature
, cmd
->hob_nsect
,
2201 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2202 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2203 res
->command
, res
->feature
, res
->nsect
,
2204 res
->lbal
, res
->lbam
, res
->lbah
,
2205 res
->hob_feature
, res
->hob_nsect
,
2206 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2207 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2208 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2210 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2212 if (res
->command
& ATA_BUSY
)
2213 ata_dev_printk(qc
->dev
, KERN_ERR
,
2214 "status: { Busy }\n");
2216 ata_dev_printk(qc
->dev
, KERN_ERR
,
2217 "status: { %s%s%s%s}\n",
2218 res
->command
& ATA_DRDY
? "DRDY " : "",
2219 res
->command
& ATA_DF
? "DF " : "",
2220 res
->command
& ATA_DRQ
? "DRQ " : "",
2221 res
->command
& ATA_ERR
? "ERR " : "");
2224 if (cmd
->command
!= ATA_CMD_PACKET
&&
2225 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2227 ata_dev_printk(qc
->dev
, KERN_ERR
,
2228 "error: { %s%s%s%s}\n",
2229 res
->feature
& ATA_ICRC
? "ICRC " : "",
2230 res
->feature
& ATA_UNC
? "UNC " : "",
2231 res
->feature
& ATA_IDNF
? "IDNF " : "",
2232 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2237 * ata_eh_report - report error handling to user
2238 * @ap: ATA port to report EH about
2240 * Report EH to user.
2245 void ata_eh_report(struct ata_port
*ap
)
2247 struct ata_link
*link
;
2249 ata_for_each_link(link
, ap
, HOST_FIRST
)
2250 ata_eh_link_report(link
);
2253 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2254 unsigned int *classes
, unsigned long deadline
,
2257 struct ata_device
*dev
;
2260 ata_for_each_dev(dev
, link
, ALL
)
2261 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2263 return reset(link
, classes
, deadline
);
2266 static int ata_eh_followup_srst_needed(struct ata_link
*link
,
2267 int rc
, const unsigned int *classes
)
2269 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2273 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2278 int ata_eh_reset(struct ata_link
*link
, int classify
,
2279 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2280 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2282 struct ata_port
*ap
= link
->ap
;
2283 struct ata_link
*slave
= ap
->slave_link
;
2284 struct ata_eh_context
*ehc
= &link
->eh_context
;
2285 struct ata_eh_context
*sehc
= &slave
->eh_context
;
2286 unsigned int *classes
= ehc
->classes
;
2287 unsigned int lflags
= link
->flags
;
2288 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2289 int max_tries
= 0, try = 0;
2290 struct ata_link
*failed_link
;
2291 struct ata_device
*dev
;
2292 unsigned long deadline
, now
;
2293 ata_reset_fn_t reset
;
2294 unsigned long flags
;
2301 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2303 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2305 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2308 /* make sure each reset attemp is at least COOL_DOWN apart */
2309 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2311 WARN_ON(time_after(ehc
->last_reset
, now
));
2312 deadline
= ata_deadline(ehc
->last_reset
,
2313 ATA_EH_RESET_COOL_DOWN
);
2314 if (time_before(now
, deadline
))
2315 schedule_timeout_uninterruptible(deadline
- now
);
2318 spin_lock_irqsave(ap
->lock
, flags
);
2319 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2320 spin_unlock_irqrestore(ap
->lock
, flags
);
2322 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2324 ata_for_each_dev(dev
, link
, ALL
) {
2325 /* If we issue an SRST then an ATA drive (not ATAPI)
2326 * may change configuration and be in PIO0 timing. If
2327 * we do a hard reset (or are coming from power on)
2328 * this is true for ATA or ATAPI. Until we've set a
2329 * suitable controller mode we should not touch the
2330 * bus as we may be talking too fast.
2332 dev
->pio_mode
= XFER_PIO_0
;
2334 /* If the controller has a pio mode setup function
2335 * then use it to set the chipset to rights. Don't
2336 * touch the DMA setup as that will be dealt with when
2337 * configuring devices.
2339 if (ap
->ops
->set_piomode
)
2340 ap
->ops
->set_piomode(ap
, dev
);
2343 /* prefer hardreset */
2345 ehc
->i
.action
&= ~ATA_EH_RESET
;
2348 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2349 } else if (softreset
) {
2351 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2355 unsigned long deadline
= ata_deadline(jiffies
,
2356 ATA_EH_PRERESET_TIMEOUT
);
2359 sehc
->i
.action
&= ~ATA_EH_RESET
;
2360 sehc
->i
.action
|= ehc
->i
.action
;
2363 rc
= prereset(link
, deadline
);
2365 /* If present, do prereset on slave link too. Reset
2366 * is skipped iff both master and slave links report
2367 * -ENOENT or clear ATA_EH_RESET.
2369 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2372 tmp
= prereset(slave
, deadline
);
2376 ehc
->i
.action
|= sehc
->i
.action
;
2380 if (rc
== -ENOENT
) {
2381 ata_link_printk(link
, KERN_DEBUG
,
2382 "port disabled. ignoring.\n");
2383 ehc
->i
.action
&= ~ATA_EH_RESET
;
2385 ata_for_each_dev(dev
, link
, ALL
)
2386 classes
[dev
->devno
] = ATA_DEV_NONE
;
2390 ata_link_printk(link
, KERN_ERR
,
2391 "prereset failed (errno=%d)\n", rc
);
2395 /* prereset() might have cleared ATA_EH_RESET. If so,
2396 * bang classes and return.
2398 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2399 ata_for_each_dev(dev
, link
, ALL
)
2400 classes
[dev
->devno
] = ATA_DEV_NONE
;
2410 if (ata_is_host_link(link
))
2411 ata_eh_freeze_port(ap
);
2413 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2417 ata_link_printk(link
, KERN_INFO
, "%s resetting link\n",
2418 reset
== softreset
? "soft" : "hard");
2420 /* mark that this EH session started with reset */
2421 ehc
->last_reset
= jiffies
;
2422 if (reset
== hardreset
)
2423 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2425 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2427 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2428 if (rc
&& rc
!= -EAGAIN
) {
2433 /* hardreset slave link if existent */
2434 if (slave
&& reset
== hardreset
) {
2438 ata_link_printk(slave
, KERN_INFO
,
2439 "hard resetting link\n");
2441 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2442 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2450 failed_link
= slave
;
2456 /* perform follow-up SRST if necessary */
2457 if (reset
== hardreset
&&
2458 ata_eh_followup_srst_needed(link
, rc
, classes
)) {
2462 ata_link_printk(link
, KERN_ERR
,
2463 "follow-up softreset required "
2464 "but no softreset avaliable\n");
2470 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2471 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2475 ata_link_printk(link
, KERN_INFO
, "no reset method "
2476 "available, skipping reset\n");
2477 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2478 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2482 * Post-reset processing
2484 ata_for_each_dev(dev
, link
, ALL
) {
2485 /* After the reset, the device state is PIO 0 and the
2486 * controller state is undefined. Reset also wakes up
2487 * drives from sleeping mode.
2489 dev
->pio_mode
= XFER_PIO_0
;
2490 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2492 if (!ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2493 /* apply class override */
2494 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2495 classes
[dev
->devno
] = ATA_DEV_ATA
;
2496 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2497 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2499 classes
[dev
->devno
] = ATA_DEV_NONE
;
2502 /* record current link speed */
2503 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2504 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2505 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2506 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2509 if (ata_is_host_link(link
))
2510 ata_eh_thaw_port(ap
);
2512 /* postreset() should clear hardware SError. Although SError
2513 * is cleared during link resume, clearing SError here is
2514 * necessary as some PHYs raise hotplug events after SRST.
2515 * This introduces race condition where hotplug occurs between
2516 * reset and here. This race is mediated by cross checking
2517 * link onlineness and classification result later.
2520 postreset(link
, classes
);
2522 postreset(slave
, classes
);
2525 /* clear cached SError */
2526 spin_lock_irqsave(link
->ap
->lock
, flags
);
2527 link
->eh_info
.serror
= 0;
2529 slave
->eh_info
.serror
= 0;
2530 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2532 /* Make sure onlineness and classification result correspond.
2533 * Hotplug could have happened during reset and some
2534 * controllers fail to wait while a drive is spinning up after
2535 * being hotplugged causing misdetection. By cross checking
2536 * link onlineness and classification result, those conditions
2537 * can be reliably detected and retried.
2540 ata_for_each_dev(dev
, link
, ALL
) {
2541 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2542 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2543 classes
[dev
->devno
] = ATA_DEV_NONE
;
2544 if (ata_phys_link_online(ata_dev_phys_link(dev
)))
2549 if (classify
&& nr_unknown
) {
2550 if (try < max_tries
) {
2551 ata_link_printk(link
, KERN_WARNING
, "link online but "
2552 "device misclassified, retrying\n");
2557 ata_link_printk(link
, KERN_WARNING
,
2558 "link online but device misclassified, "
2559 "device detection might fail\n");
2562 /* reset successful, schedule revalidation */
2563 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2565 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
2566 ehc
->last_reset
= jiffies
; /* update to completion time */
2567 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2571 /* clear hotplug flag */
2572 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2574 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2576 spin_lock_irqsave(ap
->lock
, flags
);
2577 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2578 spin_unlock_irqrestore(ap
->lock
, flags
);
2583 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2584 if (!ata_is_host_link(link
) &&
2585 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2588 if (rc
== -ERESTART
|| try >= max_tries
)
2592 if (time_before(now
, deadline
)) {
2593 unsigned long delta
= deadline
- now
;
2595 ata_link_printk(failed_link
, KERN_WARNING
,
2596 "reset failed (errno=%d), retrying in %u secs\n",
2597 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2600 delta
= schedule_timeout_uninterruptible(delta
);
2603 if (try == max_tries
- 1) {
2604 sata_down_spd_limit(link
);
2606 sata_down_spd_limit(slave
);
2607 } else if (rc
== -EPIPE
)
2608 sata_down_spd_limit(failed_link
);
2615 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
2617 struct ata_link
*link
;
2618 struct ata_device
*dev
;
2619 unsigned long flags
;
2622 * This function can be thought of as an extended version of
2623 * ata_eh_about_to_do() specially crafted to accommodate the
2624 * requirements of ATA_EH_PARK handling. Since the EH thread
2625 * does not leave the do {} while () loop in ata_eh_recover as
2626 * long as the timeout for a park request to *one* device on
2627 * the port has not expired, and since we still want to pick
2628 * up park requests to other devices on the same port or
2629 * timeout updates for the same device, we have to pull
2630 * ATA_EH_PARK actions from eh_info into eh_context.i
2631 * ourselves at the beginning of each pass over the loop.
2633 * Additionally, all write accesses to &ap->park_req_pending
2634 * through INIT_COMPLETION() (see below) or complete_all()
2635 * (see ata_scsi_park_store()) are protected by the host lock.
2636 * As a result we have that park_req_pending.done is zero on
2637 * exit from this function, i.e. when ATA_EH_PARK actions for
2638 * *all* devices on port ap have been pulled into the
2639 * respective eh_context structs. If, and only if,
2640 * park_req_pending.done is non-zero by the time we reach
2641 * wait_for_completion_timeout(), another ATA_EH_PARK action
2642 * has been scheduled for at least one of the devices on port
2643 * ap and we have to cycle over the do {} while () loop in
2644 * ata_eh_recover() again.
2647 spin_lock_irqsave(ap
->lock
, flags
);
2648 INIT_COMPLETION(ap
->park_req_pending
);
2649 ata_for_each_link(link
, ap
, EDGE
) {
2650 ata_for_each_dev(dev
, link
, ALL
) {
2651 struct ata_eh_info
*ehi
= &link
->eh_info
;
2653 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
2654 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
2655 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
2658 spin_unlock_irqrestore(ap
->lock
, flags
);
2661 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
2663 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2664 struct ata_taskfile tf
;
2665 unsigned int err_mask
;
2667 ata_tf_init(dev
, &tf
);
2669 ehc
->unloaded_mask
|= 1 << dev
->devno
;
2670 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
2676 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2677 tf
.command
= ATA_CMD_CHK_POWER
;
2680 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
2681 tf
.protocol
|= ATA_PROT_NODATA
;
2682 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
2683 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
2684 ata_dev_printk(dev
, KERN_ERR
, "head unload failed!\n");
2685 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2689 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
2690 struct ata_device
**r_failed_dev
)
2692 struct ata_port
*ap
= link
->ap
;
2693 struct ata_eh_context
*ehc
= &link
->eh_context
;
2694 struct ata_device
*dev
;
2695 unsigned int new_mask
= 0;
2696 unsigned long flags
;
2701 /* For PATA drive side cable detection to work, IDENTIFY must
2702 * be done backwards such that PDIAG- is released by the slave
2703 * device before the master device is identified.
2705 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
2706 unsigned int action
= ata_eh_dev_action(dev
);
2707 unsigned int readid_flags
= 0;
2709 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
2710 readid_flags
|= ATA_READID_POSTRESET
;
2712 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
2713 WARN_ON(dev
->class == ATA_DEV_PMP
);
2715 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2720 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
2721 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
2726 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
2728 /* Configuration may have changed, reconfigure
2731 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2733 /* schedule the scsi_rescan_device() here */
2734 queue_work(ata_aux_wq
, &(ap
->scsi_rescan_task
));
2735 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
2736 ehc
->tries
[dev
->devno
] &&
2737 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
2738 dev
->class = ehc
->classes
[dev
->devno
];
2740 if (dev
->class == ATA_DEV_PMP
)
2741 rc
= sata_pmp_attach(dev
);
2743 rc
= ata_dev_read_id(dev
, &dev
->class,
2744 readid_flags
, dev
->id
);
2747 new_mask
|= 1 << dev
->devno
;
2750 /* IDENTIFY was issued to non-existent
2751 * device. No need to reset. Just
2752 * thaw and kill the device.
2754 ata_eh_thaw_port(ap
);
2755 dev
->class = ATA_DEV_UNKNOWN
;
2758 dev
->class = ATA_DEV_UNKNOWN
;
2764 /* PDIAG- should have been released, ask cable type if post-reset */
2765 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
2766 if (ap
->ops
->cable_detect
)
2767 ap
->cbl
= ap
->ops
->cable_detect(ap
);
2771 /* Configure new devices forward such that user doesn't see
2772 * device detection messages backwards.
2774 ata_for_each_dev(dev
, link
, ALL
) {
2775 if (!(new_mask
& (1 << dev
->devno
)) ||
2776 dev
->class == ATA_DEV_PMP
)
2779 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
2780 rc
= ata_dev_configure(dev
);
2781 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
2785 spin_lock_irqsave(ap
->lock
, flags
);
2786 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
2787 spin_unlock_irqrestore(ap
->lock
, flags
);
2789 /* new device discovered, configure xfermode */
2790 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2796 *r_failed_dev
= dev
;
2797 DPRINTK("EXIT rc=%d\n", rc
);
2802 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2803 * @link: link on which timings will be programmed
2804 * @r_failed_dev: out paramter for failed device
2806 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2807 * ata_set_mode() fails, pointer to the failing device is
2808 * returned in @r_failed_dev.
2811 * PCI/etc. bus probe sem.
2814 * 0 on success, negative errno otherwise
2816 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
2818 struct ata_port
*ap
= link
->ap
;
2819 struct ata_device
*dev
;
2822 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2823 ata_for_each_dev(dev
, link
, ENABLED
) {
2824 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
2825 struct ata_ering_entry
*ent
;
2827 ent
= ata_ering_top(&dev
->ering
);
2829 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
2833 /* has private set_mode? */
2834 if (ap
->ops
->set_mode
)
2835 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
2837 rc
= ata_do_set_mode(link
, r_failed_dev
);
2839 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2840 ata_for_each_dev(dev
, link
, ENABLED
) {
2841 struct ata_eh_context
*ehc
= &link
->eh_context
;
2842 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
2843 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
2845 if (dev
->xfer_mode
!= saved_xfer_mode
||
2846 ata_ncq_enabled(dev
) != saved_ncq
)
2847 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
2854 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2855 * @dev: ATAPI device to clear UA for
2857 * Resets and other operations can make an ATAPI device raise
2858 * UNIT ATTENTION which causes the next operation to fail. This
2859 * function clears UA.
2862 * EH context (may sleep).
2865 * 0 on success, -errno on failure.
2867 static int atapi_eh_clear_ua(struct ata_device
*dev
)
2871 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
2872 u8 sense_buffer
[SCSI_SENSE_BUFFERSIZE
];
2874 unsigned int err_mask
;
2876 err_mask
= atapi_eh_tur(dev
, &sense_key
);
2877 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
2878 ata_dev_printk(dev
, KERN_WARNING
, "TEST_UNIT_READY "
2879 "failed (err_mask=0x%x)\n", err_mask
);
2883 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
2886 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
2888 ata_dev_printk(dev
, KERN_WARNING
, "failed to clear "
2889 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
2894 ata_dev_printk(dev
, KERN_WARNING
,
2895 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES
);
2900 static int ata_link_nr_enabled(struct ata_link
*link
)
2902 struct ata_device
*dev
;
2905 ata_for_each_dev(dev
, link
, ENABLED
)
2910 static int ata_link_nr_vacant(struct ata_link
*link
)
2912 struct ata_device
*dev
;
2915 ata_for_each_dev(dev
, link
, ALL
)
2916 if (dev
->class == ATA_DEV_UNKNOWN
)
2921 static int ata_eh_skip_recovery(struct ata_link
*link
)
2923 struct ata_port
*ap
= link
->ap
;
2924 struct ata_eh_context
*ehc
= &link
->eh_context
;
2925 struct ata_device
*dev
;
2927 /* skip disabled links */
2928 if (link
->flags
& ATA_LFLAG_DISABLED
)
2931 /* thaw frozen port and recover failed devices */
2932 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
2935 /* reset at least once if reset is requested */
2936 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
2937 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
2940 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2941 ata_for_each_dev(dev
, link
, ALL
) {
2942 if (dev
->class == ATA_DEV_UNKNOWN
&&
2943 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
2950 static int ata_eh_schedule_probe(struct ata_device
*dev
)
2952 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2954 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
2955 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
2958 ata_eh_detach_dev(dev
);
2960 ehc
->did_probe_mask
|= (1 << dev
->devno
);
2961 ehc
->i
.action
|= ATA_EH_RESET
;
2962 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
2963 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
2968 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
2970 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2972 ehc
->tries
[dev
->devno
]--;
2976 /* device missing or wrong IDENTIFY data, schedule probing */
2977 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
2979 /* give it just one more chance */
2980 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
2982 if (ehc
->tries
[dev
->devno
] == 1 && dev
->pio_mode
> XFER_PIO_0
) {
2983 /* This is the last chance, better to slow
2984 * down than lose it.
2986 sata_down_spd_limit(ata_dev_phys_link(dev
));
2987 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
2991 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
2992 /* disable device if it has used up all its chances */
2993 ata_dev_disable(dev
);
2995 /* detach if offline */
2996 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
2997 ata_eh_detach_dev(dev
);
2999 /* schedule probe if necessary */
3000 if (ata_eh_schedule_probe(dev
)) {
3001 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3002 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
3003 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
3008 ehc
->i
.action
|= ATA_EH_RESET
;
3014 * ata_eh_recover - recover host port after error
3015 * @ap: host port to recover
3016 * @prereset: prereset method (can be NULL)
3017 * @softreset: softreset method (can be NULL)
3018 * @hardreset: hardreset method (can be NULL)
3019 * @postreset: postreset method (can be NULL)
3020 * @r_failed_link: out parameter for failed link
3022 * This is the alpha and omega, eum and yang, heart and soul of
3023 * libata exception handling. On entry, actions required to
3024 * recover each link and hotplug requests are recorded in the
3025 * link's eh_context. This function executes all the operations
3026 * with appropriate retrials and fallbacks to resurrect failed
3027 * devices, detach goners and greet newcomers.
3030 * Kernel thread context (may sleep).
3033 * 0 on success, -errno on failure.
3035 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3036 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3037 ata_postreset_fn_t postreset
,
3038 struct ata_link
**r_failed_link
)
3040 struct ata_link
*link
;
3041 struct ata_device
*dev
;
3044 unsigned long flags
, deadline
;
3048 /* prep for recovery */
3049 ata_for_each_link(link
, ap
, EDGE
) {
3050 struct ata_eh_context
*ehc
= &link
->eh_context
;
3052 /* re-enable link? */
3053 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3054 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3055 spin_lock_irqsave(ap
->lock
, flags
);
3056 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3057 spin_unlock_irqrestore(ap
->lock
, flags
);
3058 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3061 ata_for_each_dev(dev
, link
, ALL
) {
3062 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3063 ehc
->tries
[dev
->devno
] = 1;
3065 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3067 /* collect port action mask recorded in dev actions */
3068 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3069 ~ATA_EH_PERDEV_MASK
;
3070 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3072 /* process hotplug request */
3073 if (dev
->flags
& ATA_DFLAG_DETACH
)
3074 ata_eh_detach_dev(dev
);
3076 /* schedule probe if necessary */
3077 if (!ata_dev_enabled(dev
))
3078 ata_eh_schedule_probe(dev
);
3086 /* if UNLOADING, finish immediately */
3087 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3091 ata_for_each_link(link
, ap
, EDGE
) {
3092 struct ata_eh_context
*ehc
= &link
->eh_context
;
3094 /* skip EH if possible. */
3095 if (ata_eh_skip_recovery(link
))
3098 ata_for_each_dev(dev
, link
, ALL
)
3099 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3103 ata_for_each_link(link
, ap
, EDGE
) {
3104 struct ata_eh_context
*ehc
= &link
->eh_context
;
3106 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3109 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3110 prereset
, softreset
, hardreset
, postreset
);
3112 ata_link_printk(link
, KERN_ERR
,
3113 "reset failed, giving up\n");
3122 * clears ATA_EH_PARK in eh_info and resets
3123 * ap->park_req_pending
3125 ata_eh_pull_park_action(ap
);
3128 ata_for_each_link(link
, ap
, EDGE
) {
3129 ata_for_each_dev(dev
, link
, ALL
) {
3130 struct ata_eh_context
*ehc
= &link
->eh_context
;
3133 if (dev
->class != ATA_DEV_ATA
)
3135 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3138 tmp
= dev
->unpark_deadline
;
3139 if (time_before(deadline
, tmp
))
3141 else if (time_before_eq(tmp
, jiffies
))
3143 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3146 ata_eh_park_issue_cmd(dev
, 1);
3151 if (time_before_eq(deadline
, now
))
3154 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3157 ata_for_each_link(link
, ap
, EDGE
) {
3158 ata_for_each_dev(dev
, link
, ALL
) {
3159 if (!(link
->eh_context
.unloaded_mask
&
3163 ata_eh_park_issue_cmd(dev
, 0);
3164 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3169 ata_for_each_link(link
, ap
, EDGE
) {
3170 struct ata_eh_context
*ehc
= &link
->eh_context
;
3172 /* revalidate existing devices and attach new ones */
3173 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3177 /* if PMP got attached, return, pmp EH will take care of it */
3178 if (link
->device
->class == ATA_DEV_PMP
) {
3183 /* configure transfer mode if necessary */
3184 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3185 rc
= ata_set_mode(link
, &dev
);
3188 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3191 /* If reset has been issued, clear UA to avoid
3192 * disrupting the current users of the device.
3194 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3195 ata_for_each_dev(dev
, link
, ALL
) {
3196 if (dev
->class != ATA_DEV_ATAPI
)
3198 rc
= atapi_eh_clear_ua(dev
);
3204 /* configure link power saving */
3205 if (ehc
->i
.action
& ATA_EH_LPM
)
3206 ata_for_each_dev(dev
, link
, ALL
)
3207 ata_dev_enable_pm(dev
, ap
->pm_policy
);
3209 /* this link is okay now */
3215 ata_eh_handle_dev_fail(dev
, rc
);
3217 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3218 /* PMP reset requires working host port.
3219 * Can't retry if it's frozen.
3221 if (sata_pmp_attached(ap
))
3231 if (rc
&& r_failed_link
)
3232 *r_failed_link
= link
;
3234 DPRINTK("EXIT, rc=%d\n", rc
);
3239 * ata_eh_finish - finish up EH
3240 * @ap: host port to finish EH for
3242 * Recovery is complete. Clean up EH states and retry or finish
3248 void ata_eh_finish(struct ata_port
*ap
)
3252 /* retry or finish qcs */
3253 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
3254 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
3256 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
3260 /* FIXME: Once EH migration is complete,
3261 * generate sense data in this function,
3262 * considering both err_mask and tf.
3264 if (qc
->flags
& ATA_QCFLAG_RETRY
)
3265 ata_eh_qc_retry(qc
);
3267 ata_eh_qc_complete(qc
);
3269 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
3270 ata_eh_qc_complete(qc
);
3272 /* feed zero TF to sense generation */
3273 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
3274 ata_eh_qc_retry(qc
);
3279 /* make sure nr_active_links is zero after EH */
3280 WARN_ON(ap
->nr_active_links
);
3281 ap
->nr_active_links
= 0;
3285 * ata_do_eh - do standard error handling
3286 * @ap: host port to handle error for
3288 * @prereset: prereset method (can be NULL)
3289 * @softreset: softreset method (can be NULL)
3290 * @hardreset: hardreset method (can be NULL)
3291 * @postreset: postreset method (can be NULL)
3293 * Perform standard error handling sequence.
3296 * Kernel thread context (may sleep).
3298 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3299 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3300 ata_postreset_fn_t postreset
)
3302 struct ata_device
*dev
;
3308 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
3311 ata_for_each_dev(dev
, &ap
->link
, ALL
)
3312 ata_dev_disable(dev
);
3319 * ata_std_error_handler - standard error handler
3320 * @ap: host port to handle error for
3322 * Standard error handler
3325 * Kernel thread context (may sleep).
3327 void ata_std_error_handler(struct ata_port
*ap
)
3329 struct ata_port_operations
*ops
= ap
->ops
;
3330 ata_reset_fn_t hardreset
= ops
->hardreset
;
3332 /* ignore built-in hardreset if SCR access is not available */
3333 if (ata_is_builtin_hardreset(hardreset
) && !sata_scr_valid(&ap
->link
))
3336 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
3341 * ata_eh_handle_port_suspend - perform port suspend operation
3342 * @ap: port to suspend
3347 * Kernel thread context (may sleep).
3349 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
3351 unsigned long flags
;
3354 /* are we suspending? */
3355 spin_lock_irqsave(ap
->lock
, flags
);
3356 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3357 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
3358 spin_unlock_irqrestore(ap
->lock
, flags
);
3361 spin_unlock_irqrestore(ap
->lock
, flags
);
3363 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
3365 /* tell ACPI we're suspending */
3366 rc
= ata_acpi_on_suspend(ap
);
3371 ata_eh_freeze_port(ap
);
3373 if (ap
->ops
->port_suspend
)
3374 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
3376 ata_acpi_set_state(ap
, PMSG_SUSPEND
);
3379 spin_lock_irqsave(ap
->lock
, flags
);
3381 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
3383 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
3384 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
3385 ata_port_schedule_eh(ap
);
3387 if (ap
->pm_result
) {
3388 *ap
->pm_result
= rc
;
3389 ap
->pm_result
= NULL
;
3392 spin_unlock_irqrestore(ap
->lock
, flags
);
3398 * ata_eh_handle_port_resume - perform port resume operation
3399 * @ap: port to resume
3404 * Kernel thread context (may sleep).
3406 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
3408 unsigned long flags
;
3411 /* are we resuming? */
3412 spin_lock_irqsave(ap
->lock
, flags
);
3413 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3414 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
3415 spin_unlock_irqrestore(ap
->lock
, flags
);
3418 spin_unlock_irqrestore(ap
->lock
, flags
);
3420 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
3422 ata_acpi_set_state(ap
, PMSG_ON
);
3424 if (ap
->ops
->port_resume
)
3425 rc
= ap
->ops
->port_resume(ap
);
3427 /* tell ACPI that we're resuming */
3428 ata_acpi_on_resume(ap
);
3431 spin_lock_irqsave(ap
->lock
, flags
);
3432 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
3433 if (ap
->pm_result
) {
3434 *ap
->pm_result
= rc
;
3435 ap
->pm_result
= NULL
;
3437 spin_unlock_irqrestore(ap
->lock
, flags
);
3439 #endif /* CONFIG_PM */