parisc: disable UP-optimized flush_tlb_mm
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / ata / libata-eh.c
blob2c4ccecea691938979d4d9c155935816e9c67c32
1 /*
2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
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,
24 * USA.
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/pci.h>
37 #include <scsi/scsi.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_eh.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_cmnd.h>
42 #include "../scsi/scsi_transport_api.h"
44 #include <linux/libata.h>
46 #include "libata.h"
48 enum {
49 /* speed down verdicts */
50 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
51 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
52 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
53 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
55 /* error flags */
56 ATA_EFLAG_IS_IO = (1 << 0),
57 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
59 /* error categories */
60 ATA_ECAT_NONE = 0,
61 ATA_ECAT_ATA_BUS = 1,
62 ATA_ECAT_TOUT_HSM = 2,
63 ATA_ECAT_UNK_DEV = 3,
64 ATA_ECAT_DUBIOUS_NONE = 4,
65 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
66 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
67 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
68 ATA_ECAT_NR = 8,
70 ATA_EH_CMD_DFL_TIMEOUT = 5000,
72 /* always put at least this amount of time between resets */
73 ATA_EH_RESET_COOL_DOWN = 5000,
75 /* Waiting in ->prereset can never be reliable. It's
76 * sometimes nice to wait there but it can't be depended upon;
77 * otherwise, we wouldn't be resetting. Just give it enough
78 * time for most drives to spin up.
80 ATA_EH_PRERESET_TIMEOUT = 10000,
81 ATA_EH_FASTDRAIN_INTERVAL = 3000,
84 /* The following table determines how we sequence resets. Each entry
85 * represents timeout for that try. The first try can be soft or
86 * hardreset. All others are hardreset if available. In most cases
87 * the first reset w/ 10sec timeout should succeed. Following entries
88 * are mostly for error handling, hotplug and retarded devices.
90 static const unsigned long ata_eh_reset_timeouts[] = {
91 10000, /* most drives spin up by 10sec */
92 10000, /* > 99% working drives spin up before 20sec */
93 35000, /* give > 30 secs of idleness for retarded devices */
94 5000, /* and sweet one last chance */
95 ULONG_MAX, /* > 1 min has elapsed, give up */
98 static const unsigned long ata_eh_identify_timeouts[] = {
99 5000, /* covers > 99% of successes and not too boring on failures */
100 10000, /* combined time till here is enough even for media access */
101 30000, /* for true idiots */
102 ULONG_MAX,
105 static const unsigned long ata_eh_other_timeouts[] = {
106 5000, /* same rationale as identify timeout */
107 10000, /* ditto */
108 /* but no merciful 30sec for other commands, it just isn't worth it */
109 ULONG_MAX,
112 struct ata_eh_cmd_timeout_ent {
113 const u8 *commands;
114 const unsigned long *timeouts;
117 /* The following table determines timeouts to use for EH internal
118 * commands. Each table entry is a command class and matches the
119 * commands the entry applies to and the timeout table to use.
121 * On the retry after a command timed out, the next timeout value from
122 * the table is used. If the table doesn't contain further entries,
123 * the last value is used.
125 * ehc->cmd_timeout_idx keeps track of which timeout to use per
126 * command class, so if SET_FEATURES times out on the first try, the
127 * next try will use the second timeout value only for that class.
129 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
130 static const struct ata_eh_cmd_timeout_ent
131 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
132 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
133 .timeouts = ata_eh_identify_timeouts, },
134 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
135 .timeouts = ata_eh_other_timeouts, },
136 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
137 .timeouts = ata_eh_other_timeouts, },
138 { .commands = CMDS(ATA_CMD_SET_FEATURES),
139 .timeouts = ata_eh_other_timeouts, },
140 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
141 .timeouts = ata_eh_other_timeouts, },
143 #undef CMDS
145 static void __ata_port_freeze(struct ata_port *ap);
146 #ifdef CONFIG_PM
147 static void ata_eh_handle_port_suspend(struct ata_port *ap);
148 static void ata_eh_handle_port_resume(struct ata_port *ap);
149 #else /* CONFIG_PM */
150 static void ata_eh_handle_port_suspend(struct ata_port *ap)
153 static void ata_eh_handle_port_resume(struct ata_port *ap)
155 #endif /* CONFIG_PM */
157 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
158 va_list args)
160 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
161 ATA_EH_DESC_LEN - ehi->desc_len,
162 fmt, args);
166 * __ata_ehi_push_desc - push error description without adding separator
167 * @ehi: target EHI
168 * @fmt: printf format string
170 * Format string according to @fmt and append it to @ehi->desc.
172 * LOCKING:
173 * spin_lock_irqsave(host lock)
175 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
177 va_list args;
179 va_start(args, fmt);
180 __ata_ehi_pushv_desc(ehi, fmt, args);
181 va_end(args);
185 * ata_ehi_push_desc - push error description with separator
186 * @ehi: target EHI
187 * @fmt: printf format string
189 * Format string according to @fmt and append it to @ehi->desc.
190 * If @ehi->desc is not empty, ", " is added in-between.
192 * LOCKING:
193 * spin_lock_irqsave(host lock)
195 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
197 va_list args;
199 if (ehi->desc_len)
200 __ata_ehi_push_desc(ehi, ", ");
202 va_start(args, fmt);
203 __ata_ehi_pushv_desc(ehi, fmt, args);
204 va_end(args);
208 * ata_ehi_clear_desc - clean error description
209 * @ehi: target EHI
211 * Clear @ehi->desc.
213 * LOCKING:
214 * spin_lock_irqsave(host lock)
216 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
218 ehi->desc[0] = '\0';
219 ehi->desc_len = 0;
223 * ata_port_desc - append port description
224 * @ap: target ATA port
225 * @fmt: printf format string
227 * Format string according to @fmt and append it to port
228 * description. If port description is not empty, " " is added
229 * in-between. This function is to be used while initializing
230 * ata_host. The description is printed on host registration.
232 * LOCKING:
233 * None.
235 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
237 va_list args;
239 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
241 if (ap->link.eh_info.desc_len)
242 __ata_ehi_push_desc(&ap->link.eh_info, " ");
244 va_start(args, fmt);
245 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
246 va_end(args);
249 #ifdef CONFIG_PCI
252 * ata_port_pbar_desc - append PCI BAR description
253 * @ap: target ATA port
254 * @bar: target PCI BAR
255 * @offset: offset into PCI BAR
256 * @name: name of the area
258 * If @offset is negative, this function formats a string which
259 * contains the name, address, size and type of the BAR and
260 * appends it to the port description. If @offset is zero or
261 * positive, only name and offsetted address is appended.
263 * LOCKING:
264 * None.
266 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
267 const char *name)
269 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
270 char *type = "";
271 unsigned long long start, len;
273 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
274 type = "m";
275 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
276 type = "i";
278 start = (unsigned long long)pci_resource_start(pdev, bar);
279 len = (unsigned long long)pci_resource_len(pdev, bar);
281 if (offset < 0)
282 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
283 else
284 ata_port_desc(ap, "%s 0x%llx", name,
285 start + (unsigned long long)offset);
288 #endif /* CONFIG_PCI */
290 static int ata_lookup_timeout_table(u8 cmd)
292 int i;
294 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
295 const u8 *cur;
297 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
298 if (*cur == cmd)
299 return i;
302 return -1;
306 * ata_internal_cmd_timeout - determine timeout for an internal command
307 * @dev: target device
308 * @cmd: internal command to be issued
310 * Determine timeout for internal command @cmd for @dev.
312 * LOCKING:
313 * EH context.
315 * RETURNS:
316 * Determined timeout.
318 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
320 struct ata_eh_context *ehc = &dev->link->eh_context;
321 int ent = ata_lookup_timeout_table(cmd);
322 int idx;
324 if (ent < 0)
325 return ATA_EH_CMD_DFL_TIMEOUT;
327 idx = ehc->cmd_timeout_idx[dev->devno][ent];
328 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
332 * ata_internal_cmd_timed_out - notification for internal command timeout
333 * @dev: target device
334 * @cmd: internal command which timed out
336 * Notify EH that internal command @cmd for @dev timed out. This
337 * function should be called only for commands whose timeouts are
338 * determined using ata_internal_cmd_timeout().
340 * LOCKING:
341 * EH context.
343 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
345 struct ata_eh_context *ehc = &dev->link->eh_context;
346 int ent = ata_lookup_timeout_table(cmd);
347 int idx;
349 if (ent < 0)
350 return;
352 idx = ehc->cmd_timeout_idx[dev->devno][ent];
353 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
354 ehc->cmd_timeout_idx[dev->devno][ent]++;
357 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
358 unsigned int err_mask)
360 struct ata_ering_entry *ent;
362 WARN_ON(!err_mask);
364 ering->cursor++;
365 ering->cursor %= ATA_ERING_SIZE;
367 ent = &ering->ring[ering->cursor];
368 ent->eflags = eflags;
369 ent->err_mask = err_mask;
370 ent->timestamp = get_jiffies_64();
373 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
375 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
377 if (ent->err_mask)
378 return ent;
379 return NULL;
382 static void ata_ering_clear(struct ata_ering *ering)
384 memset(ering, 0, sizeof(*ering));
387 static int ata_ering_map(struct ata_ering *ering,
388 int (*map_fn)(struct ata_ering_entry *, void *),
389 void *arg)
391 int idx, rc = 0;
392 struct ata_ering_entry *ent;
394 idx = ering->cursor;
395 do {
396 ent = &ering->ring[idx];
397 if (!ent->err_mask)
398 break;
399 rc = map_fn(ent, arg);
400 if (rc)
401 break;
402 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
403 } while (idx != ering->cursor);
405 return rc;
408 static unsigned int ata_eh_dev_action(struct ata_device *dev)
410 struct ata_eh_context *ehc = &dev->link->eh_context;
412 return ehc->i.action | ehc->i.dev_action[dev->devno];
415 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
416 struct ata_eh_info *ehi, unsigned int action)
418 struct ata_device *tdev;
420 if (!dev) {
421 ehi->action &= ~action;
422 ata_link_for_each_dev(tdev, link)
423 ehi->dev_action[tdev->devno] &= ~action;
424 } else {
425 /* doesn't make sense for port-wide EH actions */
426 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
428 /* break ehi->action into ehi->dev_action */
429 if (ehi->action & action) {
430 ata_link_for_each_dev(tdev, link)
431 ehi->dev_action[tdev->devno] |=
432 ehi->action & action;
433 ehi->action &= ~action;
436 /* turn off the specified per-dev action */
437 ehi->dev_action[dev->devno] &= ~action;
442 * ata_scsi_timed_out - SCSI layer time out callback
443 * @cmd: timed out SCSI command
445 * Handles SCSI layer timeout. We race with normal completion of
446 * the qc for @cmd. If the qc is already gone, we lose and let
447 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
448 * timed out and EH should be invoked. Prevent ata_qc_complete()
449 * from finishing it by setting EH_SCHEDULED and return
450 * EH_NOT_HANDLED.
452 * TODO: kill this function once old EH is gone.
454 * LOCKING:
455 * Called from timer context
457 * RETURNS:
458 * EH_HANDLED or EH_NOT_HANDLED
460 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
462 struct Scsi_Host *host = cmd->device->host;
463 struct ata_port *ap = ata_shost_to_port(host);
464 unsigned long flags;
465 struct ata_queued_cmd *qc;
466 enum scsi_eh_timer_return ret;
468 DPRINTK("ENTER\n");
470 if (ap->ops->error_handler) {
471 ret = EH_NOT_HANDLED;
472 goto out;
475 ret = EH_HANDLED;
476 spin_lock_irqsave(ap->lock, flags);
477 qc = ata_qc_from_tag(ap, ap->link.active_tag);
478 if (qc) {
479 WARN_ON(qc->scsicmd != cmd);
480 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
481 qc->err_mask |= AC_ERR_TIMEOUT;
482 ret = EH_NOT_HANDLED;
484 spin_unlock_irqrestore(ap->lock, flags);
486 out:
487 DPRINTK("EXIT, ret=%d\n", ret);
488 return ret;
492 * ata_scsi_error - SCSI layer error handler callback
493 * @host: SCSI host on which error occurred
495 * Handles SCSI-layer-thrown error events.
497 * LOCKING:
498 * Inherited from SCSI layer (none, can sleep)
500 * RETURNS:
501 * Zero.
503 void ata_scsi_error(struct Scsi_Host *host)
505 struct ata_port *ap = ata_shost_to_port(host);
506 int i;
507 unsigned long flags;
509 DPRINTK("ENTER\n");
511 /* synchronize with port task */
512 ata_port_flush_task(ap);
514 /* synchronize with host lock and sort out timeouts */
516 /* For new EH, all qcs are finished in one of three ways -
517 * normal completion, error completion, and SCSI timeout.
518 * Both cmpletions can race against SCSI timeout. When normal
519 * completion wins, the qc never reaches EH. When error
520 * completion wins, the qc has ATA_QCFLAG_FAILED set.
522 * When SCSI timeout wins, things are a bit more complex.
523 * Normal or error completion can occur after the timeout but
524 * before this point. In such cases, both types of
525 * completions are honored. A scmd is determined to have
526 * timed out iff its associated qc is active and not failed.
528 if (ap->ops->error_handler) {
529 struct scsi_cmnd *scmd, *tmp;
530 int nr_timedout = 0;
532 spin_lock_irqsave(ap->lock, flags);
534 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
535 struct ata_queued_cmd *qc;
537 for (i = 0; i < ATA_MAX_QUEUE; i++) {
538 qc = __ata_qc_from_tag(ap, i);
539 if (qc->flags & ATA_QCFLAG_ACTIVE &&
540 qc->scsicmd == scmd)
541 break;
544 if (i < ATA_MAX_QUEUE) {
545 /* the scmd has an associated qc */
546 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
547 /* which hasn't failed yet, timeout */
548 qc->err_mask |= AC_ERR_TIMEOUT;
549 qc->flags |= ATA_QCFLAG_FAILED;
550 nr_timedout++;
552 } else {
553 /* Normal completion occurred after
554 * SCSI timeout but before this point.
555 * Successfully complete it.
557 scmd->retries = scmd->allowed;
558 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
562 /* If we have timed out qcs. They belong to EH from
563 * this point but the state of the controller is
564 * unknown. Freeze the port to make sure the IRQ
565 * handler doesn't diddle with those qcs. This must
566 * be done atomically w.r.t. setting QCFLAG_FAILED.
568 if (nr_timedout)
569 __ata_port_freeze(ap);
571 spin_unlock_irqrestore(ap->lock, flags);
573 /* initialize eh_tries */
574 ap->eh_tries = ATA_EH_MAX_TRIES;
575 } else
576 spin_unlock_wait(ap->lock);
578 repeat:
579 /* invoke error handler */
580 if (ap->ops->error_handler) {
581 struct ata_link *link;
583 /* kill fast drain timer */
584 del_timer_sync(&ap->fastdrain_timer);
586 /* process port resume request */
587 ata_eh_handle_port_resume(ap);
589 /* fetch & clear EH info */
590 spin_lock_irqsave(ap->lock, flags);
592 __ata_port_for_each_link(link, ap) {
593 struct ata_eh_context *ehc = &link->eh_context;
594 struct ata_device *dev;
596 memset(&link->eh_context, 0, sizeof(link->eh_context));
597 link->eh_context.i = link->eh_info;
598 memset(&link->eh_info, 0, sizeof(link->eh_info));
600 ata_link_for_each_dev(dev, link) {
601 int devno = dev->devno;
603 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
604 if (ata_ncq_enabled(dev))
605 ehc->saved_ncq_enabled |= 1 << devno;
609 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
610 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
611 ap->excl_link = NULL; /* don't maintain exclusion over EH */
613 spin_unlock_irqrestore(ap->lock, flags);
615 /* invoke EH, skip if unloading or suspended */
616 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
617 ap->ops->error_handler(ap);
618 else
619 ata_eh_finish(ap);
621 /* process port suspend request */
622 ata_eh_handle_port_suspend(ap);
624 /* Exception might have happend after ->error_handler
625 * recovered the port but before this point. Repeat
626 * EH in such case.
628 spin_lock_irqsave(ap->lock, flags);
630 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
631 if (--ap->eh_tries) {
632 spin_unlock_irqrestore(ap->lock, flags);
633 goto repeat;
635 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
636 "tries, giving up\n", ATA_EH_MAX_TRIES);
637 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
640 /* this run is complete, make sure EH info is clear */
641 __ata_port_for_each_link(link, ap)
642 memset(&link->eh_info, 0, sizeof(link->eh_info));
644 /* Clear host_eh_scheduled while holding ap->lock such
645 * that if exception occurs after this point but
646 * before EH completion, SCSI midlayer will
647 * re-initiate EH.
649 host->host_eh_scheduled = 0;
651 spin_unlock_irqrestore(ap->lock, flags);
652 } else {
653 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
654 ap->ops->eng_timeout(ap);
657 /* finish or retry handled scmd's and clean up */
658 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
660 scsi_eh_flush_done_q(&ap->eh_done_q);
662 /* clean up */
663 spin_lock_irqsave(ap->lock, flags);
665 if (ap->pflags & ATA_PFLAG_LOADING)
666 ap->pflags &= ~ATA_PFLAG_LOADING;
667 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
668 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
670 if (ap->pflags & ATA_PFLAG_RECOVERED)
671 ata_port_printk(ap, KERN_INFO, "EH complete\n");
673 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
675 /* tell wait_eh that we're done */
676 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
677 wake_up_all(&ap->eh_wait_q);
679 spin_unlock_irqrestore(ap->lock, flags);
681 DPRINTK("EXIT\n");
685 * ata_port_wait_eh - Wait for the currently pending EH to complete
686 * @ap: Port to wait EH for
688 * Wait until the currently pending EH is complete.
690 * LOCKING:
691 * Kernel thread context (may sleep).
693 void ata_port_wait_eh(struct ata_port *ap)
695 unsigned long flags;
696 DEFINE_WAIT(wait);
698 retry:
699 spin_lock_irqsave(ap->lock, flags);
701 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
702 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
703 spin_unlock_irqrestore(ap->lock, flags);
704 schedule();
705 spin_lock_irqsave(ap->lock, flags);
707 finish_wait(&ap->eh_wait_q, &wait);
709 spin_unlock_irqrestore(ap->lock, flags);
711 /* make sure SCSI EH is complete */
712 if (scsi_host_in_recovery(ap->scsi_host)) {
713 msleep(10);
714 goto retry;
718 static int ata_eh_nr_in_flight(struct ata_port *ap)
720 unsigned int tag;
721 int nr = 0;
723 /* count only non-internal commands */
724 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
725 if (ata_qc_from_tag(ap, tag))
726 nr++;
728 return nr;
731 void ata_eh_fastdrain_timerfn(unsigned long arg)
733 struct ata_port *ap = (void *)arg;
734 unsigned long flags;
735 int cnt;
737 spin_lock_irqsave(ap->lock, flags);
739 cnt = ata_eh_nr_in_flight(ap);
741 /* are we done? */
742 if (!cnt)
743 goto out_unlock;
745 if (cnt == ap->fastdrain_cnt) {
746 unsigned int tag;
748 /* No progress during the last interval, tag all
749 * in-flight qcs as timed out and freeze the port.
751 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
752 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
753 if (qc)
754 qc->err_mask |= AC_ERR_TIMEOUT;
757 ata_port_freeze(ap);
758 } else {
759 /* some qcs have finished, give it another chance */
760 ap->fastdrain_cnt = cnt;
761 ap->fastdrain_timer.expires =
762 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
763 add_timer(&ap->fastdrain_timer);
766 out_unlock:
767 spin_unlock_irqrestore(ap->lock, flags);
771 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
772 * @ap: target ATA port
773 * @fastdrain: activate fast drain
775 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
776 * is non-zero and EH wasn't pending before. Fast drain ensures
777 * that EH kicks in in timely manner.
779 * LOCKING:
780 * spin_lock_irqsave(host lock)
782 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
784 int cnt;
786 /* already scheduled? */
787 if (ap->pflags & ATA_PFLAG_EH_PENDING)
788 return;
790 ap->pflags |= ATA_PFLAG_EH_PENDING;
792 if (!fastdrain)
793 return;
795 /* do we have in-flight qcs? */
796 cnt = ata_eh_nr_in_flight(ap);
797 if (!cnt)
798 return;
800 /* activate fast drain */
801 ap->fastdrain_cnt = cnt;
802 ap->fastdrain_timer.expires =
803 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
804 add_timer(&ap->fastdrain_timer);
808 * ata_qc_schedule_eh - schedule qc for error handling
809 * @qc: command to schedule error handling for
811 * Schedule error handling for @qc. EH will kick in as soon as
812 * other commands are drained.
814 * LOCKING:
815 * spin_lock_irqsave(host lock)
817 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
819 struct ata_port *ap = qc->ap;
821 WARN_ON(!ap->ops->error_handler);
823 qc->flags |= ATA_QCFLAG_FAILED;
824 ata_eh_set_pending(ap, 1);
826 /* The following will fail if timeout has already expired.
827 * ata_scsi_error() takes care of such scmds on EH entry.
828 * Note that ATA_QCFLAG_FAILED is unconditionally set after
829 * this function completes.
831 scsi_req_abort_cmd(qc->scsicmd);
835 * ata_port_schedule_eh - schedule error handling without a qc
836 * @ap: ATA port to schedule EH for
838 * Schedule error handling for @ap. EH will kick in as soon as
839 * all commands are drained.
841 * LOCKING:
842 * spin_lock_irqsave(host lock)
844 void ata_port_schedule_eh(struct ata_port *ap)
846 WARN_ON(!ap->ops->error_handler);
848 if (ap->pflags & ATA_PFLAG_INITIALIZING)
849 return;
851 ata_eh_set_pending(ap, 1);
852 scsi_schedule_eh(ap->scsi_host);
854 DPRINTK("port EH scheduled\n");
857 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
859 int tag, nr_aborted = 0;
861 WARN_ON(!ap->ops->error_handler);
863 /* we're gonna abort all commands, no need for fast drain */
864 ata_eh_set_pending(ap, 0);
866 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
867 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
869 if (qc && (!link || qc->dev->link == link)) {
870 qc->flags |= ATA_QCFLAG_FAILED;
871 ata_qc_complete(qc);
872 nr_aborted++;
876 if (!nr_aborted)
877 ata_port_schedule_eh(ap);
879 return nr_aborted;
883 * ata_link_abort - abort all qc's on the link
884 * @link: ATA link to abort qc's for
886 * Abort all active qc's active on @link and schedule EH.
888 * LOCKING:
889 * spin_lock_irqsave(host lock)
891 * RETURNS:
892 * Number of aborted qc's.
894 int ata_link_abort(struct ata_link *link)
896 return ata_do_link_abort(link->ap, link);
900 * ata_port_abort - abort all qc's on the port
901 * @ap: ATA port to abort qc's for
903 * Abort all active qc's of @ap and schedule EH.
905 * LOCKING:
906 * spin_lock_irqsave(host_set lock)
908 * RETURNS:
909 * Number of aborted qc's.
911 int ata_port_abort(struct ata_port *ap)
913 return ata_do_link_abort(ap, NULL);
917 * __ata_port_freeze - freeze port
918 * @ap: ATA port to freeze
920 * This function is called when HSM violation or some other
921 * condition disrupts normal operation of the port. Frozen port
922 * is not allowed to perform any operation until the port is
923 * thawed, which usually follows a successful reset.
925 * ap->ops->freeze() callback can be used for freezing the port
926 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
927 * port cannot be frozen hardware-wise, the interrupt handler
928 * must ack and clear interrupts unconditionally while the port
929 * is frozen.
931 * LOCKING:
932 * spin_lock_irqsave(host lock)
934 static void __ata_port_freeze(struct ata_port *ap)
936 WARN_ON(!ap->ops->error_handler);
938 if (ap->ops->freeze)
939 ap->ops->freeze(ap);
941 ap->pflags |= ATA_PFLAG_FROZEN;
943 DPRINTK("ata%u port frozen\n", ap->print_id);
947 * ata_port_freeze - abort & freeze port
948 * @ap: ATA port to freeze
950 * Abort and freeze @ap.
952 * LOCKING:
953 * spin_lock_irqsave(host lock)
955 * RETURNS:
956 * Number of aborted commands.
958 int ata_port_freeze(struct ata_port *ap)
960 int nr_aborted;
962 WARN_ON(!ap->ops->error_handler);
964 nr_aborted = ata_port_abort(ap);
965 __ata_port_freeze(ap);
967 return nr_aborted;
971 * sata_async_notification - SATA async notification handler
972 * @ap: ATA port where async notification is received
974 * Handler to be called when async notification via SDB FIS is
975 * received. This function schedules EH if necessary.
977 * LOCKING:
978 * spin_lock_irqsave(host lock)
980 * RETURNS:
981 * 1 if EH is scheduled, 0 otherwise.
983 int sata_async_notification(struct ata_port *ap)
985 u32 sntf;
986 int rc;
988 if (!(ap->flags & ATA_FLAG_AN))
989 return 0;
991 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
992 if (rc == 0)
993 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
995 if (!sata_pmp_attached(ap) || rc) {
996 /* PMP is not attached or SNTF is not available */
997 if (!sata_pmp_attached(ap)) {
998 /* PMP is not attached. Check whether ATAPI
999 * AN is configured. If so, notify media
1000 * change.
1002 struct ata_device *dev = ap->link.device;
1004 if ((dev->class == ATA_DEV_ATAPI) &&
1005 (dev->flags & ATA_DFLAG_AN))
1006 ata_scsi_media_change_notify(dev);
1007 return 0;
1008 } else {
1009 /* PMP is attached but SNTF is not available.
1010 * ATAPI async media change notification is
1011 * not used. The PMP must be reporting PHY
1012 * status change, schedule EH.
1014 ata_port_schedule_eh(ap);
1015 return 1;
1017 } else {
1018 /* PMP is attached and SNTF is available */
1019 struct ata_link *link;
1021 /* check and notify ATAPI AN */
1022 ata_port_for_each_link(link, ap) {
1023 if (!(sntf & (1 << link->pmp)))
1024 continue;
1026 if ((link->device->class == ATA_DEV_ATAPI) &&
1027 (link->device->flags & ATA_DFLAG_AN))
1028 ata_scsi_media_change_notify(link->device);
1031 /* If PMP is reporting that PHY status of some
1032 * downstream ports has changed, schedule EH.
1034 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1035 ata_port_schedule_eh(ap);
1036 return 1;
1039 return 0;
1044 * ata_eh_freeze_port - EH helper to freeze port
1045 * @ap: ATA port to freeze
1047 * Freeze @ap.
1049 * LOCKING:
1050 * None.
1052 void ata_eh_freeze_port(struct ata_port *ap)
1054 unsigned long flags;
1056 if (!ap->ops->error_handler)
1057 return;
1059 spin_lock_irqsave(ap->lock, flags);
1060 __ata_port_freeze(ap);
1061 spin_unlock_irqrestore(ap->lock, flags);
1065 * ata_port_thaw_port - EH helper to thaw port
1066 * @ap: ATA port to thaw
1068 * Thaw frozen port @ap.
1070 * LOCKING:
1071 * None.
1073 void ata_eh_thaw_port(struct ata_port *ap)
1075 unsigned long flags;
1077 if (!ap->ops->error_handler)
1078 return;
1080 spin_lock_irqsave(ap->lock, flags);
1082 ap->pflags &= ~ATA_PFLAG_FROZEN;
1084 if (ap->ops->thaw)
1085 ap->ops->thaw(ap);
1087 spin_unlock_irqrestore(ap->lock, flags);
1089 DPRINTK("ata%u port thawed\n", ap->print_id);
1092 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1094 /* nada */
1097 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1099 struct ata_port *ap = qc->ap;
1100 struct scsi_cmnd *scmd = qc->scsicmd;
1101 unsigned long flags;
1103 spin_lock_irqsave(ap->lock, flags);
1104 qc->scsidone = ata_eh_scsidone;
1105 __ata_qc_complete(qc);
1106 WARN_ON(ata_tag_valid(qc->tag));
1107 spin_unlock_irqrestore(ap->lock, flags);
1109 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1113 * ata_eh_qc_complete - Complete an active ATA command from EH
1114 * @qc: Command to complete
1116 * Indicate to the mid and upper layers that an ATA command has
1117 * completed. To be used from EH.
1119 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1121 struct scsi_cmnd *scmd = qc->scsicmd;
1122 scmd->retries = scmd->allowed;
1123 __ata_eh_qc_complete(qc);
1127 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1128 * @qc: Command to retry
1130 * Indicate to the mid and upper layers that an ATA command
1131 * should be retried. To be used from EH.
1133 * SCSI midlayer limits the number of retries to scmd->allowed.
1134 * scmd->retries is decremented for commands which get retried
1135 * due to unrelated failures (qc->err_mask is zero).
1137 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1139 struct scsi_cmnd *scmd = qc->scsicmd;
1140 if (!qc->err_mask && scmd->retries)
1141 scmd->retries--;
1142 __ata_eh_qc_complete(qc);
1146 * ata_eh_detach_dev - detach ATA device
1147 * @dev: ATA device to detach
1149 * Detach @dev.
1151 * LOCKING:
1152 * None.
1154 void ata_eh_detach_dev(struct ata_device *dev)
1156 struct ata_link *link = dev->link;
1157 struct ata_port *ap = link->ap;
1158 unsigned long flags;
1160 ata_dev_disable(dev);
1162 spin_lock_irqsave(ap->lock, flags);
1164 dev->flags &= ~ATA_DFLAG_DETACH;
1166 if (ata_scsi_offline_dev(dev)) {
1167 dev->flags |= ATA_DFLAG_DETACHED;
1168 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1171 /* clear per-dev EH actions */
1172 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1173 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1175 spin_unlock_irqrestore(ap->lock, flags);
1179 * ata_eh_about_to_do - about to perform eh_action
1180 * @link: target ATA link
1181 * @dev: target ATA dev for per-dev action (can be NULL)
1182 * @action: action about to be performed
1184 * Called just before performing EH actions to clear related bits
1185 * in @link->eh_info such that eh actions are not unnecessarily
1186 * repeated.
1188 * LOCKING:
1189 * None.
1191 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1192 unsigned int action)
1194 struct ata_port *ap = link->ap;
1195 struct ata_eh_info *ehi = &link->eh_info;
1196 struct ata_eh_context *ehc = &link->eh_context;
1197 unsigned long flags;
1199 spin_lock_irqsave(ap->lock, flags);
1201 ata_eh_clear_action(link, dev, ehi, action);
1203 if (!(ehc->i.flags & ATA_EHI_QUIET))
1204 ap->pflags |= ATA_PFLAG_RECOVERED;
1206 spin_unlock_irqrestore(ap->lock, flags);
1210 * ata_eh_done - EH action complete
1211 * @ap: target ATA port
1212 * @dev: target ATA dev for per-dev action (can be NULL)
1213 * @action: action just completed
1215 * Called right after performing EH actions to clear related bits
1216 * in @link->eh_context.
1218 * LOCKING:
1219 * None.
1221 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1222 unsigned int action)
1224 struct ata_eh_context *ehc = &link->eh_context;
1226 ata_eh_clear_action(link, dev, &ehc->i, action);
1230 * ata_err_string - convert err_mask to descriptive string
1231 * @err_mask: error mask to convert to string
1233 * Convert @err_mask to descriptive string. Errors are
1234 * prioritized according to severity and only the most severe
1235 * error is reported.
1237 * LOCKING:
1238 * None.
1240 * RETURNS:
1241 * Descriptive string for @err_mask
1243 static const char *ata_err_string(unsigned int err_mask)
1245 if (err_mask & AC_ERR_HOST_BUS)
1246 return "host bus error";
1247 if (err_mask & AC_ERR_ATA_BUS)
1248 return "ATA bus error";
1249 if (err_mask & AC_ERR_TIMEOUT)
1250 return "timeout";
1251 if (err_mask & AC_ERR_HSM)
1252 return "HSM violation";
1253 if (err_mask & AC_ERR_SYSTEM)
1254 return "internal error";
1255 if (err_mask & AC_ERR_MEDIA)
1256 return "media error";
1257 if (err_mask & AC_ERR_INVALID)
1258 return "invalid argument";
1259 if (err_mask & AC_ERR_DEV)
1260 return "device error";
1261 return "unknown error";
1265 * ata_read_log_page - read a specific log page
1266 * @dev: target device
1267 * @page: page to read
1268 * @buf: buffer to store read page
1269 * @sectors: number of sectors to read
1271 * Read log page using READ_LOG_EXT command.
1273 * LOCKING:
1274 * Kernel thread context (may sleep).
1276 * RETURNS:
1277 * 0 on success, AC_ERR_* mask otherwise.
1279 static unsigned int ata_read_log_page(struct ata_device *dev,
1280 u8 page, void *buf, unsigned int sectors)
1282 struct ata_taskfile tf;
1283 unsigned int err_mask;
1285 DPRINTK("read log page - page %d\n", page);
1287 ata_tf_init(dev, &tf);
1288 tf.command = ATA_CMD_READ_LOG_EXT;
1289 tf.lbal = page;
1290 tf.nsect = sectors;
1291 tf.hob_nsect = sectors >> 8;
1292 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1293 tf.protocol = ATA_PROT_PIO;
1295 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1296 buf, sectors * ATA_SECT_SIZE, 0);
1298 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1299 return err_mask;
1303 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1304 * @dev: Device to read log page 10h from
1305 * @tag: Resulting tag of the failed command
1306 * @tf: Resulting taskfile registers of the failed command
1308 * Read log page 10h to obtain NCQ error details and clear error
1309 * condition.
1311 * LOCKING:
1312 * Kernel thread context (may sleep).
1314 * RETURNS:
1315 * 0 on success, -errno otherwise.
1317 static int ata_eh_read_log_10h(struct ata_device *dev,
1318 int *tag, struct ata_taskfile *tf)
1320 u8 *buf = dev->link->ap->sector_buf;
1321 unsigned int err_mask;
1322 u8 csum;
1323 int i;
1325 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1326 if (err_mask)
1327 return -EIO;
1329 csum = 0;
1330 for (i = 0; i < ATA_SECT_SIZE; i++)
1331 csum += buf[i];
1332 if (csum)
1333 ata_dev_printk(dev, KERN_WARNING,
1334 "invalid checksum 0x%x on log page 10h\n", csum);
1336 if (buf[0] & 0x80)
1337 return -ENOENT;
1339 *tag = buf[0] & 0x1f;
1341 tf->command = buf[2];
1342 tf->feature = buf[3];
1343 tf->lbal = buf[4];
1344 tf->lbam = buf[5];
1345 tf->lbah = buf[6];
1346 tf->device = buf[7];
1347 tf->hob_lbal = buf[8];
1348 tf->hob_lbam = buf[9];
1349 tf->hob_lbah = buf[10];
1350 tf->nsect = buf[12];
1351 tf->hob_nsect = buf[13];
1353 return 0;
1357 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1358 * @dev: device to perform REQUEST_SENSE to
1359 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1360 * @dfl_sense_key: default sense key to use
1362 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1363 * SENSE. This function is EH helper.
1365 * LOCKING:
1366 * Kernel thread context (may sleep).
1368 * RETURNS:
1369 * 0 on success, AC_ERR_* mask on failure
1371 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1372 u8 *sense_buf, u8 dfl_sense_key)
1374 u8 cdb[ATAPI_CDB_LEN] =
1375 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1376 struct ata_port *ap = dev->link->ap;
1377 struct ata_taskfile tf;
1379 DPRINTK("ATAPI request sense\n");
1381 /* FIXME: is this needed? */
1382 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1384 /* initialize sense_buf with the error register,
1385 * for the case where they are -not- overwritten
1387 sense_buf[0] = 0x70;
1388 sense_buf[2] = dfl_sense_key;
1390 /* some devices time out if garbage left in tf */
1391 ata_tf_init(dev, &tf);
1393 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1394 tf.command = ATA_CMD_PACKET;
1396 /* is it pointless to prefer PIO for "safety reasons"? */
1397 if (ap->flags & ATA_FLAG_PIO_DMA) {
1398 tf.protocol = ATAPI_PROT_DMA;
1399 tf.feature |= ATAPI_PKT_DMA;
1400 } else {
1401 tf.protocol = ATAPI_PROT_PIO;
1402 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1403 tf.lbah = 0;
1406 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1407 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1411 * ata_eh_analyze_serror - analyze SError for a failed port
1412 * @link: ATA link to analyze SError for
1414 * Analyze SError if available and further determine cause of
1415 * failure.
1417 * LOCKING:
1418 * None.
1420 static void ata_eh_analyze_serror(struct ata_link *link)
1422 struct ata_eh_context *ehc = &link->eh_context;
1423 u32 serror = ehc->i.serror;
1424 unsigned int err_mask = 0, action = 0;
1425 u32 hotplug_mask;
1427 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1428 err_mask |= AC_ERR_ATA_BUS;
1429 action |= ATA_EH_RESET;
1431 if (serror & SERR_PROTOCOL) {
1432 err_mask |= AC_ERR_HSM;
1433 action |= ATA_EH_RESET;
1435 if (serror & SERR_INTERNAL) {
1436 err_mask |= AC_ERR_SYSTEM;
1437 action |= ATA_EH_RESET;
1440 /* Determine whether a hotplug event has occurred. Both
1441 * SError.N/X are considered hotplug events for enabled or
1442 * host links. For disabled PMP links, only N bit is
1443 * considered as X bit is left at 1 for link plugging.
1445 hotplug_mask = 0;
1447 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1448 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1449 else
1450 hotplug_mask = SERR_PHYRDY_CHG;
1452 if (serror & hotplug_mask)
1453 ata_ehi_hotplugged(&ehc->i);
1455 ehc->i.err_mask |= err_mask;
1456 ehc->i.action |= action;
1460 * ata_eh_analyze_ncq_error - analyze NCQ error
1461 * @link: ATA link to analyze NCQ error for
1463 * Read log page 10h, determine the offending qc and acquire
1464 * error status TF. For NCQ device errors, all LLDDs have to do
1465 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1466 * care of the rest.
1468 * LOCKING:
1469 * Kernel thread context (may sleep).
1471 void ata_eh_analyze_ncq_error(struct ata_link *link)
1473 struct ata_port *ap = link->ap;
1474 struct ata_eh_context *ehc = &link->eh_context;
1475 struct ata_device *dev = link->device;
1476 struct ata_queued_cmd *qc;
1477 struct ata_taskfile tf;
1478 int tag, rc;
1480 /* if frozen, we can't do much */
1481 if (ap->pflags & ATA_PFLAG_FROZEN)
1482 return;
1484 /* is it NCQ device error? */
1485 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1486 return;
1488 /* has LLDD analyzed already? */
1489 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1490 qc = __ata_qc_from_tag(ap, tag);
1492 if (!(qc->flags & ATA_QCFLAG_FAILED))
1493 continue;
1495 if (qc->err_mask)
1496 return;
1499 /* okay, this error is ours */
1500 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1501 if (rc) {
1502 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1503 "(errno=%d)\n", rc);
1504 return;
1507 if (!(link->sactive & (1 << tag))) {
1508 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1509 "inactive tag %d\n", tag);
1510 return;
1513 /* we've got the perpetrator, condemn it */
1514 qc = __ata_qc_from_tag(ap, tag);
1515 memcpy(&qc->result_tf, &tf, sizeof(tf));
1516 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1517 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1518 ehc->i.err_mask &= ~AC_ERR_DEV;
1522 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1523 * @qc: qc to analyze
1524 * @tf: Taskfile registers to analyze
1526 * Analyze taskfile of @qc and further determine cause of
1527 * failure. This function also requests ATAPI sense data if
1528 * avaliable.
1530 * LOCKING:
1531 * Kernel thread context (may sleep).
1533 * RETURNS:
1534 * Determined recovery action
1536 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1537 const struct ata_taskfile *tf)
1539 unsigned int tmp, action = 0;
1540 u8 stat = tf->command, err = tf->feature;
1542 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1543 qc->err_mask |= AC_ERR_HSM;
1544 return ATA_EH_RESET;
1547 if (stat & (ATA_ERR | ATA_DF))
1548 qc->err_mask |= AC_ERR_DEV;
1549 else
1550 return 0;
1552 switch (qc->dev->class) {
1553 case ATA_DEV_ATA:
1554 if (err & ATA_ICRC)
1555 qc->err_mask |= AC_ERR_ATA_BUS;
1556 if (err & ATA_UNC)
1557 qc->err_mask |= AC_ERR_MEDIA;
1558 if (err & ATA_IDNF)
1559 qc->err_mask |= AC_ERR_INVALID;
1560 break;
1562 case ATA_DEV_ATAPI:
1563 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1564 tmp = atapi_eh_request_sense(qc->dev,
1565 qc->scsicmd->sense_buffer,
1566 qc->result_tf.feature >> 4);
1567 if (!tmp) {
1568 /* ATA_QCFLAG_SENSE_VALID is used to
1569 * tell atapi_qc_complete() that sense
1570 * data is already valid.
1572 * TODO: interpret sense data and set
1573 * appropriate err_mask.
1575 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1576 } else
1577 qc->err_mask |= tmp;
1581 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1582 action |= ATA_EH_RESET;
1584 return action;
1587 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1588 int *xfer_ok)
1590 int base = 0;
1592 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1593 *xfer_ok = 1;
1595 if (!*xfer_ok)
1596 base = ATA_ECAT_DUBIOUS_NONE;
1598 if (err_mask & AC_ERR_ATA_BUS)
1599 return base + ATA_ECAT_ATA_BUS;
1601 if (err_mask & AC_ERR_TIMEOUT)
1602 return base + ATA_ECAT_TOUT_HSM;
1604 if (eflags & ATA_EFLAG_IS_IO) {
1605 if (err_mask & AC_ERR_HSM)
1606 return base + ATA_ECAT_TOUT_HSM;
1607 if ((err_mask &
1608 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1609 return base + ATA_ECAT_UNK_DEV;
1612 return 0;
1615 struct speed_down_verdict_arg {
1616 u64 since;
1617 int xfer_ok;
1618 int nr_errors[ATA_ECAT_NR];
1621 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1623 struct speed_down_verdict_arg *arg = void_arg;
1624 int cat;
1626 if (ent->timestamp < arg->since)
1627 return -1;
1629 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1630 &arg->xfer_ok);
1631 arg->nr_errors[cat]++;
1633 return 0;
1637 * ata_eh_speed_down_verdict - Determine speed down verdict
1638 * @dev: Device of interest
1640 * This function examines error ring of @dev and determines
1641 * whether NCQ needs to be turned off, transfer speed should be
1642 * stepped down, or falling back to PIO is necessary.
1644 * ECAT_ATA_BUS : ATA_BUS error for any command
1646 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1647 * IO commands
1649 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1651 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1652 * data transfer hasn't been verified.
1654 * Verdicts are
1656 * NCQ_OFF : Turn off NCQ.
1658 * SPEED_DOWN : Speed down transfer speed but don't fall back
1659 * to PIO.
1661 * FALLBACK_TO_PIO : Fall back to PIO.
1663 * Even if multiple verdicts are returned, only one action is
1664 * taken per error. An action triggered by non-DUBIOUS errors
1665 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1666 * This is to expedite speed down decisions right after device is
1667 * initially configured.
1669 * The followings are speed down rules. #1 and #2 deal with
1670 * DUBIOUS errors.
1672 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1673 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1675 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1676 * occurred during last 5 mins, NCQ_OFF.
1678 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1679 * ocurred during last 5 mins, FALLBACK_TO_PIO
1681 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1682 * during last 10 mins, NCQ_OFF.
1684 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1685 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1687 * LOCKING:
1688 * Inherited from caller.
1690 * RETURNS:
1691 * OR of ATA_EH_SPDN_* flags.
1693 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1695 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1696 u64 j64 = get_jiffies_64();
1697 struct speed_down_verdict_arg arg;
1698 unsigned int verdict = 0;
1700 /* scan past 5 mins of error history */
1701 memset(&arg, 0, sizeof(arg));
1702 arg.since = j64 - min(j64, j5mins);
1703 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1705 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1706 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1707 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1708 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1710 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1711 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1712 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1714 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1715 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1716 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1717 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1719 /* scan past 10 mins of error history */
1720 memset(&arg, 0, sizeof(arg));
1721 arg.since = j64 - min(j64, j10mins);
1722 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1724 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1725 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1726 verdict |= ATA_EH_SPDN_NCQ_OFF;
1728 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1729 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1730 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1731 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1733 return verdict;
1737 * ata_eh_speed_down - record error and speed down if necessary
1738 * @dev: Failed device
1739 * @eflags: mask of ATA_EFLAG_* flags
1740 * @err_mask: err_mask of the error
1742 * Record error and examine error history to determine whether
1743 * adjusting transmission speed is necessary. It also sets
1744 * transmission limits appropriately if such adjustment is
1745 * necessary.
1747 * LOCKING:
1748 * Kernel thread context (may sleep).
1750 * RETURNS:
1751 * Determined recovery action.
1753 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1754 unsigned int eflags, unsigned int err_mask)
1756 struct ata_link *link = dev->link;
1757 int xfer_ok = 0;
1758 unsigned int verdict;
1759 unsigned int action = 0;
1761 /* don't bother if Cat-0 error */
1762 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1763 return 0;
1765 /* record error and determine whether speed down is necessary */
1766 ata_ering_record(&dev->ering, eflags, err_mask);
1767 verdict = ata_eh_speed_down_verdict(dev);
1769 /* turn off NCQ? */
1770 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1771 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1772 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1773 dev->flags |= ATA_DFLAG_NCQ_OFF;
1774 ata_dev_printk(dev, KERN_WARNING,
1775 "NCQ disabled due to excessive errors\n");
1776 goto done;
1779 /* speed down? */
1780 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1781 /* speed down SATA link speed if possible */
1782 if (sata_down_spd_limit(link) == 0) {
1783 action |= ATA_EH_RESET;
1784 goto done;
1787 /* lower transfer mode */
1788 if (dev->spdn_cnt < 2) {
1789 static const int dma_dnxfer_sel[] =
1790 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1791 static const int pio_dnxfer_sel[] =
1792 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1793 int sel;
1795 if (dev->xfer_shift != ATA_SHIFT_PIO)
1796 sel = dma_dnxfer_sel[dev->spdn_cnt];
1797 else
1798 sel = pio_dnxfer_sel[dev->spdn_cnt];
1800 dev->spdn_cnt++;
1802 if (ata_down_xfermask_limit(dev, sel) == 0) {
1803 action |= ATA_EH_RESET;
1804 goto done;
1809 /* Fall back to PIO? Slowing down to PIO is meaningless for
1810 * SATA ATA devices. Consider it only for PATA and SATAPI.
1812 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1813 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1814 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1815 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1816 dev->spdn_cnt = 0;
1817 action |= ATA_EH_RESET;
1818 goto done;
1822 return 0;
1823 done:
1824 /* device has been slowed down, blow error history */
1825 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1826 ata_ering_clear(&dev->ering);
1827 return action;
1831 * ata_eh_link_autopsy - analyze error and determine recovery action
1832 * @link: host link to perform autopsy on
1834 * Analyze why @link failed and determine which recovery actions
1835 * are needed. This function also sets more detailed AC_ERR_*
1836 * values and fills sense data for ATAPI CHECK SENSE.
1838 * LOCKING:
1839 * Kernel thread context (may sleep).
1841 static void ata_eh_link_autopsy(struct ata_link *link)
1843 struct ata_port *ap = link->ap;
1844 struct ata_eh_context *ehc = &link->eh_context;
1845 struct ata_device *dev;
1846 unsigned int all_err_mask = 0, eflags = 0;
1847 int tag;
1848 u32 serror;
1849 int rc;
1851 DPRINTK("ENTER\n");
1853 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1854 return;
1856 /* obtain and analyze SError */
1857 rc = sata_scr_read(link, SCR_ERROR, &serror);
1858 if (rc == 0) {
1859 ehc->i.serror |= serror;
1860 ata_eh_analyze_serror(link);
1861 } else if (rc != -EOPNOTSUPP) {
1862 /* SError read failed, force reset and probing */
1863 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1864 ehc->i.action |= ATA_EH_RESET;
1865 ehc->i.err_mask |= AC_ERR_OTHER;
1868 /* analyze NCQ failure */
1869 ata_eh_analyze_ncq_error(link);
1871 /* any real error trumps AC_ERR_OTHER */
1872 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1873 ehc->i.err_mask &= ~AC_ERR_OTHER;
1875 all_err_mask |= ehc->i.err_mask;
1877 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1878 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1880 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link)
1881 continue;
1883 /* inherit upper level err_mask */
1884 qc->err_mask |= ehc->i.err_mask;
1886 /* analyze TF */
1887 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1889 /* DEV errors are probably spurious in case of ATA_BUS error */
1890 if (qc->err_mask & AC_ERR_ATA_BUS)
1891 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1892 AC_ERR_INVALID);
1894 /* any real error trumps unknown error */
1895 if (qc->err_mask & ~AC_ERR_OTHER)
1896 qc->err_mask &= ~AC_ERR_OTHER;
1898 /* SENSE_VALID trumps dev/unknown error and revalidation */
1899 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1900 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1902 /* determine whether the command is worth retrying */
1903 if (!(qc->err_mask & AC_ERR_INVALID) &&
1904 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1905 qc->flags |= ATA_QCFLAG_RETRY;
1907 /* accumulate error info */
1908 ehc->i.dev = qc->dev;
1909 all_err_mask |= qc->err_mask;
1910 if (qc->flags & ATA_QCFLAG_IO)
1911 eflags |= ATA_EFLAG_IS_IO;
1914 /* enforce default EH actions */
1915 if (ap->pflags & ATA_PFLAG_FROZEN ||
1916 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1917 ehc->i.action |= ATA_EH_RESET;
1918 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1919 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
1920 ehc->i.action |= ATA_EH_REVALIDATE;
1922 /* If we have offending qcs and the associated failed device,
1923 * perform per-dev EH action only on the offending device.
1925 if (ehc->i.dev) {
1926 ehc->i.dev_action[ehc->i.dev->devno] |=
1927 ehc->i.action & ATA_EH_PERDEV_MASK;
1928 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1931 /* propagate timeout to host link */
1932 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1933 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1935 /* record error and consider speeding down */
1936 dev = ehc->i.dev;
1937 if (!dev && ((ata_link_max_devices(link) == 1 &&
1938 ata_dev_enabled(link->device))))
1939 dev = link->device;
1941 if (dev) {
1942 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1943 eflags |= ATA_EFLAG_DUBIOUS_XFER;
1944 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
1947 DPRINTK("EXIT\n");
1951 * ata_eh_autopsy - analyze error and determine recovery action
1952 * @ap: host port to perform autopsy on
1954 * Analyze all links of @ap and determine why they failed and
1955 * which recovery actions are needed.
1957 * LOCKING:
1958 * Kernel thread context (may sleep).
1960 void ata_eh_autopsy(struct ata_port *ap)
1962 struct ata_link *link;
1964 ata_port_for_each_link(link, ap)
1965 ata_eh_link_autopsy(link);
1967 /* Autopsy of fanout ports can affect host link autopsy.
1968 * Perform host link autopsy last.
1970 if (sata_pmp_attached(ap))
1971 ata_eh_link_autopsy(&ap->link);
1975 * ata_eh_link_report - report error handling to user
1976 * @link: ATA link EH is going on
1978 * Report EH to user.
1980 * LOCKING:
1981 * None.
1983 static void ata_eh_link_report(struct ata_link *link)
1985 struct ata_port *ap = link->ap;
1986 struct ata_eh_context *ehc = &link->eh_context;
1987 const char *frozen, *desc;
1988 char tries_buf[6];
1989 int tag, nr_failed = 0;
1991 if (ehc->i.flags & ATA_EHI_QUIET)
1992 return;
1994 desc = NULL;
1995 if (ehc->i.desc[0] != '\0')
1996 desc = ehc->i.desc;
1998 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1999 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2001 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link ||
2002 ((qc->flags & ATA_QCFLAG_QUIET) &&
2003 qc->err_mask == AC_ERR_DEV))
2004 continue;
2005 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2006 continue;
2008 nr_failed++;
2011 if (!nr_failed && !ehc->i.err_mask)
2012 return;
2014 frozen = "";
2015 if (ap->pflags & ATA_PFLAG_FROZEN)
2016 frozen = " frozen";
2018 memset(tries_buf, 0, sizeof(tries_buf));
2019 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2020 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2021 ap->eh_tries);
2023 if (ehc->i.dev) {
2024 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2025 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2026 ehc->i.err_mask, link->sactive, ehc->i.serror,
2027 ehc->i.action, frozen, tries_buf);
2028 if (desc)
2029 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2030 } else {
2031 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2032 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2033 ehc->i.err_mask, link->sactive, ehc->i.serror,
2034 ehc->i.action, frozen, tries_buf);
2035 if (desc)
2036 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2039 if (ehc->i.serror)
2040 ata_link_printk(link, KERN_ERR,
2041 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2042 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2043 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2044 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2045 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2046 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2047 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2048 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2049 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2050 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2051 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2052 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2053 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2054 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2055 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2056 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2057 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2058 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2060 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2061 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2062 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2063 const u8 *cdb = qc->cdb;
2064 char data_buf[20] = "";
2065 char cdb_buf[70] = "";
2067 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2068 qc->dev->link != link || !qc->err_mask)
2069 continue;
2071 if (qc->dma_dir != DMA_NONE) {
2072 static const char *dma_str[] = {
2073 [DMA_BIDIRECTIONAL] = "bidi",
2074 [DMA_TO_DEVICE] = "out",
2075 [DMA_FROM_DEVICE] = "in",
2077 static const char *prot_str[] = {
2078 [ATA_PROT_PIO] = "pio",
2079 [ATA_PROT_DMA] = "dma",
2080 [ATA_PROT_NCQ] = "ncq",
2081 [ATAPI_PROT_PIO] = "pio",
2082 [ATAPI_PROT_DMA] = "dma",
2085 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2086 prot_str[qc->tf.protocol], qc->nbytes,
2087 dma_str[qc->dma_dir]);
2090 if (ata_is_atapi(qc->tf.protocol))
2091 snprintf(cdb_buf, sizeof(cdb_buf),
2092 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2093 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2094 cdb[0], cdb[1], cdb[2], cdb[3],
2095 cdb[4], cdb[5], cdb[6], cdb[7],
2096 cdb[8], cdb[9], cdb[10], cdb[11],
2097 cdb[12], cdb[13], cdb[14], cdb[15]);
2099 ata_dev_printk(qc->dev, KERN_ERR,
2100 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2101 "tag %d%s\n %s"
2102 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2103 "Emask 0x%x (%s)%s\n",
2104 cmd->command, cmd->feature, cmd->nsect,
2105 cmd->lbal, cmd->lbam, cmd->lbah,
2106 cmd->hob_feature, cmd->hob_nsect,
2107 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2108 cmd->device, qc->tag, data_buf, cdb_buf,
2109 res->command, res->feature, res->nsect,
2110 res->lbal, res->lbam, res->lbah,
2111 res->hob_feature, res->hob_nsect,
2112 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2113 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2114 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2116 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2117 ATA_ERR)) {
2118 if (res->command & ATA_BUSY)
2119 ata_dev_printk(qc->dev, KERN_ERR,
2120 "status: { Busy }\n");
2121 else
2122 ata_dev_printk(qc->dev, KERN_ERR,
2123 "status: { %s%s%s%s}\n",
2124 res->command & ATA_DRDY ? "DRDY " : "",
2125 res->command & ATA_DF ? "DF " : "",
2126 res->command & ATA_DRQ ? "DRQ " : "",
2127 res->command & ATA_ERR ? "ERR " : "");
2130 if (cmd->command != ATA_CMD_PACKET &&
2131 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2132 ATA_ABORTED)))
2133 ata_dev_printk(qc->dev, KERN_ERR,
2134 "error: { %s%s%s%s}\n",
2135 res->feature & ATA_ICRC ? "ICRC " : "",
2136 res->feature & ATA_UNC ? "UNC " : "",
2137 res->feature & ATA_IDNF ? "IDNF " : "",
2138 res->feature & ATA_ABORTED ? "ABRT " : "");
2143 * ata_eh_report - report error handling to user
2144 * @ap: ATA port to report EH about
2146 * Report EH to user.
2148 * LOCKING:
2149 * None.
2151 void ata_eh_report(struct ata_port *ap)
2153 struct ata_link *link;
2155 __ata_port_for_each_link(link, ap)
2156 ata_eh_link_report(link);
2159 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2160 unsigned int *classes, unsigned long deadline)
2162 struct ata_device *dev;
2164 ata_link_for_each_dev(dev, link)
2165 classes[dev->devno] = ATA_DEV_UNKNOWN;
2167 return reset(link, classes, deadline);
2170 static int ata_eh_followup_srst_needed(struct ata_link *link,
2171 int rc, const unsigned int *classes)
2173 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2174 return 0;
2175 if (rc == -EAGAIN)
2176 return 1;
2177 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2178 return 1;
2179 return 0;
2182 int ata_eh_reset(struct ata_link *link, int classify,
2183 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2184 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2186 struct ata_port *ap = link->ap;
2187 struct ata_eh_context *ehc = &link->eh_context;
2188 unsigned int *classes = ehc->classes;
2189 unsigned int lflags = link->flags;
2190 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2191 int max_tries = 0, try = 0;
2192 struct ata_device *dev;
2193 unsigned long deadline, now;
2194 ata_reset_fn_t reset;
2195 unsigned long flags;
2196 u32 sstatus;
2197 int nr_known, rc;
2200 * Prepare to reset
2202 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2203 max_tries++;
2204 if (link->flags & ATA_LFLAG_NO_HRST)
2205 hardreset = NULL;
2206 if (link->flags & ATA_LFLAG_NO_SRST)
2207 softreset = NULL;
2209 /* make sure each reset attemp is at least COOL_DOWN apart */
2210 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2211 now = jiffies;
2212 WARN_ON(time_after(ehc->last_reset, now));
2213 deadline = ata_deadline(ehc->last_reset,
2214 ATA_EH_RESET_COOL_DOWN);
2215 if (time_before(now, deadline))
2216 schedule_timeout_uninterruptible(deadline - now);
2219 spin_lock_irqsave(ap->lock, flags);
2220 ap->pflags |= ATA_PFLAG_RESETTING;
2221 spin_unlock_irqrestore(ap->lock, flags);
2223 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2225 ata_link_for_each_dev(dev, link) {
2226 /* If we issue an SRST then an ATA drive (not ATAPI)
2227 * may change configuration and be in PIO0 timing. If
2228 * we do a hard reset (or are coming from power on)
2229 * this is true for ATA or ATAPI. Until we've set a
2230 * suitable controller mode we should not touch the
2231 * bus as we may be talking too fast.
2233 dev->pio_mode = XFER_PIO_0;
2235 /* If the controller has a pio mode setup function
2236 * then use it to set the chipset to rights. Don't
2237 * touch the DMA setup as that will be dealt with when
2238 * configuring devices.
2240 if (ap->ops->set_piomode)
2241 ap->ops->set_piomode(ap, dev);
2244 /* prefer hardreset */
2245 reset = NULL;
2246 ehc->i.action &= ~ATA_EH_RESET;
2247 if (hardreset) {
2248 reset = hardreset;
2249 ehc->i.action |= ATA_EH_HARDRESET;
2250 } else if (softreset) {
2251 reset = softreset;
2252 ehc->i.action |= ATA_EH_SOFTRESET;
2255 if (prereset) {
2256 rc = prereset(link,
2257 ata_deadline(jiffies, ATA_EH_PRERESET_TIMEOUT));
2258 if (rc) {
2259 if (rc == -ENOENT) {
2260 ata_link_printk(link, KERN_DEBUG,
2261 "port disabled. ignoring.\n");
2262 ehc->i.action &= ~ATA_EH_RESET;
2264 ata_link_for_each_dev(dev, link)
2265 classes[dev->devno] = ATA_DEV_NONE;
2267 rc = 0;
2268 } else
2269 ata_link_printk(link, KERN_ERR,
2270 "prereset failed (errno=%d)\n", rc);
2271 goto out;
2274 /* prereset() might have cleared ATA_EH_RESET. If so,
2275 * bang classes and return.
2277 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2278 ata_link_for_each_dev(dev, link)
2279 classes[dev->devno] = ATA_DEV_NONE;
2280 rc = 0;
2281 goto out;
2285 retry:
2287 * Perform reset
2289 if (ata_is_host_link(link))
2290 ata_eh_freeze_port(ap);
2292 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2294 if (reset) {
2295 if (verbose)
2296 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2297 reset == softreset ? "soft" : "hard");
2299 /* mark that this EH session started with reset */
2300 ehc->last_reset = jiffies;
2301 if (reset == hardreset)
2302 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2303 else
2304 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2306 rc = ata_do_reset(link, reset, classes, deadline);
2307 if (rc && rc != -EAGAIN)
2308 goto fail;
2310 if (reset == hardreset &&
2311 ata_eh_followup_srst_needed(link, rc, classes)) {
2312 /* okay, let's do follow-up softreset */
2313 reset = softreset;
2315 if (!reset) {
2316 ata_link_printk(link, KERN_ERR,
2317 "follow-up softreset required "
2318 "but no softreset avaliable\n");
2319 rc = -EINVAL;
2320 goto fail;
2323 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2324 rc = ata_do_reset(link, reset, classes, deadline);
2326 } else {
2327 if (verbose)
2328 ata_link_printk(link, KERN_INFO, "no reset method "
2329 "available, skipping reset\n");
2330 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2331 lflags |= ATA_LFLAG_ASSUME_ATA;
2335 * Post-reset processing
2337 ata_link_for_each_dev(dev, link) {
2338 /* After the reset, the device state is PIO 0 and the
2339 * controller state is undefined. Reset also wakes up
2340 * drives from sleeping mode.
2342 dev->pio_mode = XFER_PIO_0;
2343 dev->flags &= ~ATA_DFLAG_SLEEPING;
2345 if (ata_link_offline(link))
2346 continue;
2348 /* apply class override */
2349 if (lflags & ATA_LFLAG_ASSUME_ATA)
2350 classes[dev->devno] = ATA_DEV_ATA;
2351 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2352 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
2355 /* record current link speed */
2356 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2357 link->sata_spd = (sstatus >> 4) & 0xf;
2359 /* thaw the port */
2360 if (ata_is_host_link(link))
2361 ata_eh_thaw_port(ap);
2363 /* postreset() should clear hardware SError. Although SError
2364 * is cleared during link resume, clearing SError here is
2365 * necessary as some PHYs raise hotplug events after SRST.
2366 * This introduces race condition where hotplug occurs between
2367 * reset and here. This race is mediated by cross checking
2368 * link onlineness and classification result later.
2370 if (postreset)
2371 postreset(link, classes);
2373 /* clear cached SError */
2374 spin_lock_irqsave(link->ap->lock, flags);
2375 link->eh_info.serror = 0;
2376 spin_unlock_irqrestore(link->ap->lock, flags);
2378 /* Make sure onlineness and classification result correspond.
2379 * Hotplug could have happened during reset and some
2380 * controllers fail to wait while a drive is spinning up after
2381 * being hotplugged causing misdetection. By cross checking
2382 * link onlineness and classification result, those conditions
2383 * can be reliably detected and retried.
2385 nr_known = 0;
2386 ata_link_for_each_dev(dev, link) {
2387 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2388 if (classes[dev->devno] == ATA_DEV_UNKNOWN)
2389 classes[dev->devno] = ATA_DEV_NONE;
2390 else
2391 nr_known++;
2394 if (classify && !nr_known && ata_link_online(link)) {
2395 if (try < max_tries) {
2396 ata_link_printk(link, KERN_WARNING, "link online but "
2397 "device misclassified, retrying\n");
2398 rc = -EAGAIN;
2399 goto fail;
2401 ata_link_printk(link, KERN_WARNING,
2402 "link online but device misclassified, "
2403 "device detection might fail\n");
2406 /* reset successful, schedule revalidation */
2407 ata_eh_done(link, NULL, ATA_EH_RESET);
2408 ehc->last_reset = jiffies; /* update to completion time */
2409 ehc->i.action |= ATA_EH_REVALIDATE;
2411 rc = 0;
2412 out:
2413 /* clear hotplug flag */
2414 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2416 spin_lock_irqsave(ap->lock, flags);
2417 ap->pflags &= ~ATA_PFLAG_RESETTING;
2418 spin_unlock_irqrestore(ap->lock, flags);
2420 return rc;
2422 fail:
2423 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2424 if (!ata_is_host_link(link) &&
2425 sata_scr_read(link, SCR_STATUS, &sstatus))
2426 rc = -ERESTART;
2428 if (rc == -ERESTART || try >= max_tries)
2429 goto out;
2431 now = jiffies;
2432 if (time_before(now, deadline)) {
2433 unsigned long delta = deadline - now;
2435 ata_link_printk(link, KERN_WARNING,
2436 "reset failed (errno=%d), retrying in %u secs\n",
2437 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2439 while (delta)
2440 delta = schedule_timeout_uninterruptible(delta);
2443 if (rc == -EPIPE || try == max_tries - 1)
2444 sata_down_spd_limit(link);
2445 if (hardreset)
2446 reset = hardreset;
2447 goto retry;
2450 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2451 struct ata_device **r_failed_dev)
2453 struct ata_port *ap = link->ap;
2454 struct ata_eh_context *ehc = &link->eh_context;
2455 struct ata_device *dev;
2456 unsigned int new_mask = 0;
2457 unsigned long flags;
2458 int rc = 0;
2460 DPRINTK("ENTER\n");
2462 /* For PATA drive side cable detection to work, IDENTIFY must
2463 * be done backwards such that PDIAG- is released by the slave
2464 * device before the master device is identified.
2466 ata_link_for_each_dev_reverse(dev, link) {
2467 unsigned int action = ata_eh_dev_action(dev);
2468 unsigned int readid_flags = 0;
2470 if (ehc->i.flags & ATA_EHI_DID_RESET)
2471 readid_flags |= ATA_READID_POSTRESET;
2473 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2474 WARN_ON(dev->class == ATA_DEV_PMP);
2476 if (ata_link_offline(link)) {
2477 rc = -EIO;
2478 goto err;
2481 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2482 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2483 readid_flags);
2484 if (rc)
2485 goto err;
2487 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2489 /* Configuration may have changed, reconfigure
2490 * transfer mode.
2492 ehc->i.flags |= ATA_EHI_SETMODE;
2494 /* schedule the scsi_rescan_device() here */
2495 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2496 } else if (dev->class == ATA_DEV_UNKNOWN &&
2497 ehc->tries[dev->devno] &&
2498 ata_class_enabled(ehc->classes[dev->devno])) {
2499 dev->class = ehc->classes[dev->devno];
2501 if (dev->class == ATA_DEV_PMP)
2502 rc = sata_pmp_attach(dev);
2503 else
2504 rc = ata_dev_read_id(dev, &dev->class,
2505 readid_flags, dev->id);
2506 switch (rc) {
2507 case 0:
2508 new_mask |= 1 << dev->devno;
2509 break;
2510 case -ENOENT:
2511 /* IDENTIFY was issued to non-existent
2512 * device. No need to reset. Just
2513 * thaw and kill the device.
2515 ata_eh_thaw_port(ap);
2516 dev->class = ATA_DEV_UNKNOWN;
2517 break;
2518 default:
2519 dev->class = ATA_DEV_UNKNOWN;
2520 goto err;
2525 /* PDIAG- should have been released, ask cable type if post-reset */
2526 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2527 if (ap->ops->cable_detect)
2528 ap->cbl = ap->ops->cable_detect(ap);
2529 ata_force_cbl(ap);
2532 /* Configure new devices forward such that user doesn't see
2533 * device detection messages backwards.
2535 ata_link_for_each_dev(dev, link) {
2536 if (!(new_mask & (1 << dev->devno)) ||
2537 dev->class == ATA_DEV_PMP)
2538 continue;
2540 ehc->i.flags |= ATA_EHI_PRINTINFO;
2541 rc = ata_dev_configure(dev);
2542 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2543 if (rc)
2544 goto err;
2546 spin_lock_irqsave(ap->lock, flags);
2547 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2548 spin_unlock_irqrestore(ap->lock, flags);
2550 /* new device discovered, configure xfermode */
2551 ehc->i.flags |= ATA_EHI_SETMODE;
2554 return 0;
2556 err:
2557 *r_failed_dev = dev;
2558 DPRINTK("EXIT rc=%d\n", rc);
2559 return rc;
2563 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2564 * @link: link on which timings will be programmed
2565 * @r_failed_dev: out paramter for failed device
2567 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2568 * ata_set_mode() fails, pointer to the failing device is
2569 * returned in @r_failed_dev.
2571 * LOCKING:
2572 * PCI/etc. bus probe sem.
2574 * RETURNS:
2575 * 0 on success, negative errno otherwise
2577 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2579 struct ata_port *ap = link->ap;
2580 struct ata_device *dev;
2581 int rc;
2583 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2584 ata_link_for_each_dev(dev, link) {
2585 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2586 struct ata_ering_entry *ent;
2588 ent = ata_ering_top(&dev->ering);
2589 if (ent)
2590 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2594 /* has private set_mode? */
2595 if (ap->ops->set_mode)
2596 rc = ap->ops->set_mode(link, r_failed_dev);
2597 else
2598 rc = ata_do_set_mode(link, r_failed_dev);
2600 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2601 ata_link_for_each_dev(dev, link) {
2602 struct ata_eh_context *ehc = &link->eh_context;
2603 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2604 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2606 if (dev->xfer_mode != saved_xfer_mode ||
2607 ata_ncq_enabled(dev) != saved_ncq)
2608 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2611 return rc;
2614 static int ata_link_nr_enabled(struct ata_link *link)
2616 struct ata_device *dev;
2617 int cnt = 0;
2619 ata_link_for_each_dev(dev, link)
2620 if (ata_dev_enabled(dev))
2621 cnt++;
2622 return cnt;
2625 static int ata_link_nr_vacant(struct ata_link *link)
2627 struct ata_device *dev;
2628 int cnt = 0;
2630 ata_link_for_each_dev(dev, link)
2631 if (dev->class == ATA_DEV_UNKNOWN)
2632 cnt++;
2633 return cnt;
2636 static int ata_eh_skip_recovery(struct ata_link *link)
2638 struct ata_port *ap = link->ap;
2639 struct ata_eh_context *ehc = &link->eh_context;
2640 struct ata_device *dev;
2642 /* skip disabled links */
2643 if (link->flags & ATA_LFLAG_DISABLED)
2644 return 1;
2646 /* thaw frozen port and recover failed devices */
2647 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2648 return 0;
2650 /* reset at least once if reset is requested */
2651 if ((ehc->i.action & ATA_EH_RESET) &&
2652 !(ehc->i.flags & ATA_EHI_DID_RESET))
2653 return 0;
2655 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2656 ata_link_for_each_dev(dev, link) {
2657 if (dev->class == ATA_DEV_UNKNOWN &&
2658 ehc->classes[dev->devno] != ATA_DEV_NONE)
2659 return 0;
2662 return 1;
2665 static int ata_eh_schedule_probe(struct ata_device *dev)
2667 struct ata_eh_context *ehc = &dev->link->eh_context;
2669 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2670 (ehc->did_probe_mask & (1 << dev->devno)))
2671 return 0;
2673 ata_eh_detach_dev(dev);
2674 ata_dev_init(dev);
2675 ehc->did_probe_mask |= (1 << dev->devno);
2676 ehc->i.action |= ATA_EH_RESET;
2677 ehc->saved_xfer_mode[dev->devno] = 0;
2678 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
2680 return 1;
2683 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2685 struct ata_eh_context *ehc = &dev->link->eh_context;
2687 ehc->tries[dev->devno]--;
2689 switch (err) {
2690 case -ENODEV:
2691 /* device missing or wrong IDENTIFY data, schedule probing */
2692 ehc->i.probe_mask |= (1 << dev->devno);
2693 case -EINVAL:
2694 /* give it just one more chance */
2695 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2696 case -EIO:
2697 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
2698 /* This is the last chance, better to slow
2699 * down than lose it.
2701 sata_down_spd_limit(dev->link);
2702 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2706 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2707 /* disable device if it has used up all its chances */
2708 ata_dev_disable(dev);
2710 /* detach if offline */
2711 if (ata_link_offline(dev->link))
2712 ata_eh_detach_dev(dev);
2714 /* schedule probe if necessary */
2715 if (ata_eh_schedule_probe(dev)) {
2716 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2717 memset(ehc->cmd_timeout_idx[dev->devno], 0,
2718 sizeof(ehc->cmd_timeout_idx[dev->devno]));
2721 return 1;
2722 } else {
2723 ehc->i.action |= ATA_EH_RESET;
2724 return 0;
2729 * ata_eh_recover - recover host port after error
2730 * @ap: host port to recover
2731 * @prereset: prereset method (can be NULL)
2732 * @softreset: softreset method (can be NULL)
2733 * @hardreset: hardreset method (can be NULL)
2734 * @postreset: postreset method (can be NULL)
2735 * @r_failed_link: out parameter for failed link
2737 * This is the alpha and omega, eum and yang, heart and soul of
2738 * libata exception handling. On entry, actions required to
2739 * recover each link and hotplug requests are recorded in the
2740 * link's eh_context. This function executes all the operations
2741 * with appropriate retrials and fallbacks to resurrect failed
2742 * devices, detach goners and greet newcomers.
2744 * LOCKING:
2745 * Kernel thread context (may sleep).
2747 * RETURNS:
2748 * 0 on success, -errno on failure.
2750 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2751 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2752 ata_postreset_fn_t postreset,
2753 struct ata_link **r_failed_link)
2755 struct ata_link *link;
2756 struct ata_device *dev;
2757 int nr_failed_devs;
2758 int rc;
2759 unsigned long flags;
2761 DPRINTK("ENTER\n");
2763 /* prep for recovery */
2764 ata_port_for_each_link(link, ap) {
2765 struct ata_eh_context *ehc = &link->eh_context;
2767 /* re-enable link? */
2768 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
2769 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
2770 spin_lock_irqsave(ap->lock, flags);
2771 link->flags &= ~ATA_LFLAG_DISABLED;
2772 spin_unlock_irqrestore(ap->lock, flags);
2773 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
2776 ata_link_for_each_dev(dev, link) {
2777 if (link->flags & ATA_LFLAG_NO_RETRY)
2778 ehc->tries[dev->devno] = 1;
2779 else
2780 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2782 /* collect port action mask recorded in dev actions */
2783 ehc->i.action |= ehc->i.dev_action[dev->devno] &
2784 ~ATA_EH_PERDEV_MASK;
2785 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
2787 /* process hotplug request */
2788 if (dev->flags & ATA_DFLAG_DETACH)
2789 ata_eh_detach_dev(dev);
2791 /* schedule probe if necessary */
2792 if (!ata_dev_enabled(dev))
2793 ata_eh_schedule_probe(dev);
2797 retry:
2798 rc = 0;
2799 nr_failed_devs = 0;
2801 /* if UNLOADING, finish immediately */
2802 if (ap->pflags & ATA_PFLAG_UNLOADING)
2803 goto out;
2805 /* prep for EH */
2806 ata_port_for_each_link(link, ap) {
2807 struct ata_eh_context *ehc = &link->eh_context;
2809 /* skip EH if possible. */
2810 if (ata_eh_skip_recovery(link))
2811 ehc->i.action = 0;
2813 ata_link_for_each_dev(dev, link)
2814 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
2817 /* reset */
2818 ata_port_for_each_link(link, ap) {
2819 struct ata_eh_context *ehc = &link->eh_context;
2821 if (!(ehc->i.action & ATA_EH_RESET))
2822 continue;
2824 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
2825 prereset, softreset, hardreset, postreset);
2826 if (rc) {
2827 ata_link_printk(link, KERN_ERR,
2828 "reset failed, giving up\n");
2829 goto out;
2833 /* the rest */
2834 ata_port_for_each_link(link, ap) {
2835 struct ata_eh_context *ehc = &link->eh_context;
2837 /* revalidate existing devices and attach new ones */
2838 rc = ata_eh_revalidate_and_attach(link, &dev);
2839 if (rc)
2840 goto dev_fail;
2842 /* if PMP got attached, return, pmp EH will take care of it */
2843 if (link->device->class == ATA_DEV_PMP) {
2844 ehc->i.action = 0;
2845 return 0;
2848 /* configure transfer mode if necessary */
2849 if (ehc->i.flags & ATA_EHI_SETMODE) {
2850 rc = ata_set_mode(link, &dev);
2851 if (rc)
2852 goto dev_fail;
2853 ehc->i.flags &= ~ATA_EHI_SETMODE;
2856 if (ehc->i.action & ATA_EH_LPM)
2857 ata_link_for_each_dev(dev, link)
2858 ata_dev_enable_pm(dev, ap->pm_policy);
2860 /* this link is okay now */
2861 ehc->i.flags = 0;
2862 continue;
2864 dev_fail:
2865 nr_failed_devs++;
2866 ata_eh_handle_dev_fail(dev, rc);
2868 if (ap->pflags & ATA_PFLAG_FROZEN) {
2869 /* PMP reset requires working host port.
2870 * Can't retry if it's frozen.
2872 if (sata_pmp_attached(ap))
2873 goto out;
2874 break;
2878 if (nr_failed_devs)
2879 goto retry;
2881 out:
2882 if (rc && r_failed_link)
2883 *r_failed_link = link;
2885 DPRINTK("EXIT, rc=%d\n", rc);
2886 return rc;
2890 * ata_eh_finish - finish up EH
2891 * @ap: host port to finish EH for
2893 * Recovery is complete. Clean up EH states and retry or finish
2894 * failed qcs.
2896 * LOCKING:
2897 * None.
2899 void ata_eh_finish(struct ata_port *ap)
2901 int tag;
2903 /* retry or finish qcs */
2904 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2905 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2907 if (!(qc->flags & ATA_QCFLAG_FAILED))
2908 continue;
2910 if (qc->err_mask) {
2911 /* FIXME: Once EH migration is complete,
2912 * generate sense data in this function,
2913 * considering both err_mask and tf.
2915 if (qc->flags & ATA_QCFLAG_RETRY)
2916 ata_eh_qc_retry(qc);
2917 else
2918 ata_eh_qc_complete(qc);
2919 } else {
2920 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2921 ata_eh_qc_complete(qc);
2922 } else {
2923 /* feed zero TF to sense generation */
2924 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2925 ata_eh_qc_retry(qc);
2930 /* make sure nr_active_links is zero after EH */
2931 WARN_ON(ap->nr_active_links);
2932 ap->nr_active_links = 0;
2936 * ata_do_eh - do standard error handling
2937 * @ap: host port to handle error for
2939 * @prereset: prereset method (can be NULL)
2940 * @softreset: softreset method (can be NULL)
2941 * @hardreset: hardreset method (can be NULL)
2942 * @postreset: postreset method (can be NULL)
2944 * Perform standard error handling sequence.
2946 * LOCKING:
2947 * Kernel thread context (may sleep).
2949 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2950 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2951 ata_postreset_fn_t postreset)
2953 struct ata_device *dev;
2954 int rc;
2956 ata_eh_autopsy(ap);
2957 ata_eh_report(ap);
2959 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
2960 NULL);
2961 if (rc) {
2962 ata_link_for_each_dev(dev, &ap->link)
2963 ata_dev_disable(dev);
2966 ata_eh_finish(ap);
2970 * ata_std_error_handler - standard error handler
2971 * @ap: host port to handle error for
2973 * Standard error handler
2975 * LOCKING:
2976 * Kernel thread context (may sleep).
2978 void ata_std_error_handler(struct ata_port *ap)
2980 struct ata_port_operations *ops = ap->ops;
2981 ata_reset_fn_t hardreset = ops->hardreset;
2983 /* ignore built-in hardreset if SCR access is not available */
2984 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
2985 hardreset = NULL;
2987 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
2990 #ifdef CONFIG_PM
2992 * ata_eh_handle_port_suspend - perform port suspend operation
2993 * @ap: port to suspend
2995 * Suspend @ap.
2997 * LOCKING:
2998 * Kernel thread context (may sleep).
3000 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3002 unsigned long flags;
3003 int rc = 0;
3005 /* are we suspending? */
3006 spin_lock_irqsave(ap->lock, flags);
3007 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3008 ap->pm_mesg.event == PM_EVENT_ON) {
3009 spin_unlock_irqrestore(ap->lock, flags);
3010 return;
3012 spin_unlock_irqrestore(ap->lock, flags);
3014 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3016 /* tell ACPI we're suspending */
3017 rc = ata_acpi_on_suspend(ap);
3018 if (rc)
3019 goto out;
3021 /* suspend */
3022 ata_eh_freeze_port(ap);
3024 if (ap->ops->port_suspend)
3025 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3027 ata_acpi_set_state(ap, PMSG_SUSPEND);
3028 out:
3029 /* report result */
3030 spin_lock_irqsave(ap->lock, flags);
3032 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3033 if (rc == 0)
3034 ap->pflags |= ATA_PFLAG_SUSPENDED;
3035 else if (ap->pflags & ATA_PFLAG_FROZEN)
3036 ata_port_schedule_eh(ap);
3038 if (ap->pm_result) {
3039 *ap->pm_result = rc;
3040 ap->pm_result = NULL;
3043 spin_unlock_irqrestore(ap->lock, flags);
3045 return;
3049 * ata_eh_handle_port_resume - perform port resume operation
3050 * @ap: port to resume
3052 * Resume @ap.
3054 * LOCKING:
3055 * Kernel thread context (may sleep).
3057 static void ata_eh_handle_port_resume(struct ata_port *ap)
3059 unsigned long flags;
3060 int rc = 0;
3062 /* are we resuming? */
3063 spin_lock_irqsave(ap->lock, flags);
3064 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3065 ap->pm_mesg.event != PM_EVENT_ON) {
3066 spin_unlock_irqrestore(ap->lock, flags);
3067 return;
3069 spin_unlock_irqrestore(ap->lock, flags);
3071 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3073 ata_acpi_set_state(ap, PMSG_ON);
3075 if (ap->ops->port_resume)
3076 rc = ap->ops->port_resume(ap);
3078 /* tell ACPI that we're resuming */
3079 ata_acpi_on_resume(ap);
3081 /* report result */
3082 spin_lock_irqsave(ap->lock, flags);
3083 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3084 if (ap->pm_result) {
3085 *ap->pm_result = rc;
3086 ap->pm_result = NULL;
3088 spin_unlock_irqrestore(ap->lock, flags);
3090 #endif /* CONFIG_PM */