[IA64] Fix build for CONFIG_SMP=n
[linux-2.6/mini2440.git] / drivers / ata / libata-eh.c
blob2eaa39fc65d031d5f44519e3ec43e9a80d3cb34b
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 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
50 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
51 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
54 /* Waiting in ->prereset can never be reliable. It's sometimes nice
55 * to wait there but it can't be depended upon; otherwise, we wouldn't
56 * be resetting. Just give it enough time for most drives to spin up.
58 enum {
59 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
60 ATA_EH_FASTDRAIN_INTERVAL = 3 * HZ,
63 /* The following table determines how we sequence resets. Each entry
64 * represents timeout for that try. The first try can be soft or
65 * hardreset. All others are hardreset if available. In most cases
66 * the first reset w/ 10sec timeout should succeed. Following entries
67 * are mostly for error handling, hotplug and retarded devices.
69 static const unsigned long ata_eh_reset_timeouts[] = {
70 10 * HZ, /* most drives spin up by 10sec */
71 10 * HZ, /* > 99% working drives spin up before 20sec */
72 35 * HZ, /* give > 30 secs of idleness for retarded devices */
73 5 * HZ, /* and sweet one last chance */
74 /* > 1 min has elapsed, give up */
77 static void __ata_port_freeze(struct ata_port *ap);
78 #ifdef CONFIG_PM
79 static void ata_eh_handle_port_suspend(struct ata_port *ap);
80 static void ata_eh_handle_port_resume(struct ata_port *ap);
81 #else /* CONFIG_PM */
82 static void ata_eh_handle_port_suspend(struct ata_port *ap)
83 { }
85 static void ata_eh_handle_port_resume(struct ata_port *ap)
86 { }
87 #endif /* CONFIG_PM */
89 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
90 va_list args)
92 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
93 ATA_EH_DESC_LEN - ehi->desc_len,
94 fmt, args);
97 /**
98 * __ata_ehi_push_desc - push error description without adding separator
99 * @ehi: target EHI
100 * @fmt: printf format string
102 * Format string according to @fmt and append it to @ehi->desc.
104 * LOCKING:
105 * spin_lock_irqsave(host lock)
107 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
109 va_list args;
111 va_start(args, fmt);
112 __ata_ehi_pushv_desc(ehi, fmt, args);
113 va_end(args);
117 * ata_ehi_push_desc - push error description with separator
118 * @ehi: target EHI
119 * @fmt: printf format string
121 * Format string according to @fmt and append it to @ehi->desc.
122 * If @ehi->desc is not empty, ", " is added in-between.
124 * LOCKING:
125 * spin_lock_irqsave(host lock)
127 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
129 va_list args;
131 if (ehi->desc_len)
132 __ata_ehi_push_desc(ehi, ", ");
134 va_start(args, fmt);
135 __ata_ehi_pushv_desc(ehi, fmt, args);
136 va_end(args);
140 * ata_ehi_clear_desc - clean error description
141 * @ehi: target EHI
143 * Clear @ehi->desc.
145 * LOCKING:
146 * spin_lock_irqsave(host lock)
148 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
150 ehi->desc[0] = '\0';
151 ehi->desc_len = 0;
155 * ata_port_desc - append port description
156 * @ap: target ATA port
157 * @fmt: printf format string
159 * Format string according to @fmt and append it to port
160 * description. If port description is not empty, " " is added
161 * in-between. This function is to be used while initializing
162 * ata_host. The description is printed on host registration.
164 * LOCKING:
165 * None.
167 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
169 va_list args;
171 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
173 if (ap->link.eh_info.desc_len)
174 __ata_ehi_push_desc(&ap->link.eh_info, " ");
176 va_start(args, fmt);
177 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
178 va_end(args);
181 #ifdef CONFIG_PCI
184 * ata_port_pbar_desc - append PCI BAR description
185 * @ap: target ATA port
186 * @bar: target PCI BAR
187 * @offset: offset into PCI BAR
188 * @name: name of the area
190 * If @offset is negative, this function formats a string which
191 * contains the name, address, size and type of the BAR and
192 * appends it to the port description. If @offset is zero or
193 * positive, only name and offsetted address is appended.
195 * LOCKING:
196 * None.
198 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
199 const char *name)
201 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
202 char *type = "";
203 unsigned long long start, len;
205 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
206 type = "m";
207 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
208 type = "i";
210 start = (unsigned long long)pci_resource_start(pdev, bar);
211 len = (unsigned long long)pci_resource_len(pdev, bar);
213 if (offset < 0)
214 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
215 else
216 ata_port_desc(ap, "%s 0x%llx", name, start + offset);
219 #endif /* CONFIG_PCI */
221 static void ata_ering_record(struct ata_ering *ering, int is_io,
222 unsigned int err_mask)
224 struct ata_ering_entry *ent;
226 WARN_ON(!err_mask);
228 ering->cursor++;
229 ering->cursor %= ATA_ERING_SIZE;
231 ent = &ering->ring[ering->cursor];
232 ent->is_io = is_io;
233 ent->err_mask = err_mask;
234 ent->timestamp = get_jiffies_64();
237 static void ata_ering_clear(struct ata_ering *ering)
239 memset(ering, 0, sizeof(*ering));
242 static int ata_ering_map(struct ata_ering *ering,
243 int (*map_fn)(struct ata_ering_entry *, void *),
244 void *arg)
246 int idx, rc = 0;
247 struct ata_ering_entry *ent;
249 idx = ering->cursor;
250 do {
251 ent = &ering->ring[idx];
252 if (!ent->err_mask)
253 break;
254 rc = map_fn(ent, arg);
255 if (rc)
256 break;
257 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
258 } while (idx != ering->cursor);
260 return rc;
263 static unsigned int ata_eh_dev_action(struct ata_device *dev)
265 struct ata_eh_context *ehc = &dev->link->eh_context;
267 return ehc->i.action | ehc->i.dev_action[dev->devno];
270 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
271 struct ata_eh_info *ehi, unsigned int action)
273 struct ata_device *tdev;
275 if (!dev) {
276 ehi->action &= ~action;
277 ata_link_for_each_dev(tdev, link)
278 ehi->dev_action[tdev->devno] &= ~action;
279 } else {
280 /* doesn't make sense for port-wide EH actions */
281 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
283 /* break ehi->action into ehi->dev_action */
284 if (ehi->action & action) {
285 ata_link_for_each_dev(tdev, link)
286 ehi->dev_action[tdev->devno] |=
287 ehi->action & action;
288 ehi->action &= ~action;
291 /* turn off the specified per-dev action */
292 ehi->dev_action[dev->devno] &= ~action;
297 * ata_scsi_timed_out - SCSI layer time out callback
298 * @cmd: timed out SCSI command
300 * Handles SCSI layer timeout. We race with normal completion of
301 * the qc for @cmd. If the qc is already gone, we lose and let
302 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
303 * timed out and EH should be invoked. Prevent ata_qc_complete()
304 * from finishing it by setting EH_SCHEDULED and return
305 * EH_NOT_HANDLED.
307 * TODO: kill this function once old EH is gone.
309 * LOCKING:
310 * Called from timer context
312 * RETURNS:
313 * EH_HANDLED or EH_NOT_HANDLED
315 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
317 struct Scsi_Host *host = cmd->device->host;
318 struct ata_port *ap = ata_shost_to_port(host);
319 unsigned long flags;
320 struct ata_queued_cmd *qc;
321 enum scsi_eh_timer_return ret;
323 DPRINTK("ENTER\n");
325 if (ap->ops->error_handler) {
326 ret = EH_NOT_HANDLED;
327 goto out;
330 ret = EH_HANDLED;
331 spin_lock_irqsave(ap->lock, flags);
332 qc = ata_qc_from_tag(ap, ap->link.active_tag);
333 if (qc) {
334 WARN_ON(qc->scsicmd != cmd);
335 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
336 qc->err_mask |= AC_ERR_TIMEOUT;
337 ret = EH_NOT_HANDLED;
339 spin_unlock_irqrestore(ap->lock, flags);
341 out:
342 DPRINTK("EXIT, ret=%d\n", ret);
343 return ret;
347 * ata_scsi_error - SCSI layer error handler callback
348 * @host: SCSI host on which error occurred
350 * Handles SCSI-layer-thrown error events.
352 * LOCKING:
353 * Inherited from SCSI layer (none, can sleep)
355 * RETURNS:
356 * Zero.
358 void ata_scsi_error(struct Scsi_Host *host)
360 struct ata_port *ap = ata_shost_to_port(host);
361 int i;
362 unsigned long flags;
364 DPRINTK("ENTER\n");
366 /* synchronize with port task */
367 ata_port_flush_task(ap);
369 /* synchronize with host lock and sort out timeouts */
371 /* For new EH, all qcs are finished in one of three ways -
372 * normal completion, error completion, and SCSI timeout.
373 * Both cmpletions can race against SCSI timeout. When normal
374 * completion wins, the qc never reaches EH. When error
375 * completion wins, the qc has ATA_QCFLAG_FAILED set.
377 * When SCSI timeout wins, things are a bit more complex.
378 * Normal or error completion can occur after the timeout but
379 * before this point. In such cases, both types of
380 * completions are honored. A scmd is determined to have
381 * timed out iff its associated qc is active and not failed.
383 if (ap->ops->error_handler) {
384 struct scsi_cmnd *scmd, *tmp;
385 int nr_timedout = 0;
387 spin_lock_irqsave(ap->lock, flags);
389 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
390 struct ata_queued_cmd *qc;
392 for (i = 0; i < ATA_MAX_QUEUE; i++) {
393 qc = __ata_qc_from_tag(ap, i);
394 if (qc->flags & ATA_QCFLAG_ACTIVE &&
395 qc->scsicmd == scmd)
396 break;
399 if (i < ATA_MAX_QUEUE) {
400 /* the scmd has an associated qc */
401 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
402 /* which hasn't failed yet, timeout */
403 qc->err_mask |= AC_ERR_TIMEOUT;
404 qc->flags |= ATA_QCFLAG_FAILED;
405 nr_timedout++;
407 } else {
408 /* Normal completion occurred after
409 * SCSI timeout but before this point.
410 * Successfully complete it.
412 scmd->retries = scmd->allowed;
413 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
417 /* If we have timed out qcs. They belong to EH from
418 * this point but the state of the controller is
419 * unknown. Freeze the port to make sure the IRQ
420 * handler doesn't diddle with those qcs. This must
421 * be done atomically w.r.t. setting QCFLAG_FAILED.
423 if (nr_timedout)
424 __ata_port_freeze(ap);
426 spin_unlock_irqrestore(ap->lock, flags);
428 /* initialize eh_tries */
429 ap->eh_tries = ATA_EH_MAX_TRIES;
430 } else
431 spin_unlock_wait(ap->lock);
433 repeat:
434 /* invoke error handler */
435 if (ap->ops->error_handler) {
436 struct ata_link *link;
438 /* kill fast drain timer */
439 del_timer_sync(&ap->fastdrain_timer);
441 /* process port resume request */
442 ata_eh_handle_port_resume(ap);
444 /* fetch & clear EH info */
445 spin_lock_irqsave(ap->lock, flags);
447 __ata_port_for_each_link(link, ap) {
448 memset(&link->eh_context, 0, sizeof(link->eh_context));
449 link->eh_context.i = link->eh_info;
450 memset(&link->eh_info, 0, sizeof(link->eh_info));
453 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
454 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
455 ap->excl_link = NULL; /* don't maintain exclusion over EH */
457 spin_unlock_irqrestore(ap->lock, flags);
459 /* invoke EH, skip if unloading or suspended */
460 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
461 ap->ops->error_handler(ap);
462 else
463 ata_eh_finish(ap);
465 /* process port suspend request */
466 ata_eh_handle_port_suspend(ap);
468 /* Exception might have happend after ->error_handler
469 * recovered the port but before this point. Repeat
470 * EH in such case.
472 spin_lock_irqsave(ap->lock, flags);
474 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
475 if (--ap->eh_tries) {
476 spin_unlock_irqrestore(ap->lock, flags);
477 goto repeat;
479 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
480 "tries, giving up\n", ATA_EH_MAX_TRIES);
481 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
484 /* this run is complete, make sure EH info is clear */
485 __ata_port_for_each_link(link, ap)
486 memset(&link->eh_info, 0, sizeof(link->eh_info));
488 /* Clear host_eh_scheduled while holding ap->lock such
489 * that if exception occurs after this point but
490 * before EH completion, SCSI midlayer will
491 * re-initiate EH.
493 host->host_eh_scheduled = 0;
495 spin_unlock_irqrestore(ap->lock, flags);
496 } else {
497 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
498 ap->ops->eng_timeout(ap);
501 /* finish or retry handled scmd's and clean up */
502 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
504 scsi_eh_flush_done_q(&ap->eh_done_q);
506 /* clean up */
507 spin_lock_irqsave(ap->lock, flags);
509 if (ap->pflags & ATA_PFLAG_LOADING)
510 ap->pflags &= ~ATA_PFLAG_LOADING;
511 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
512 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
514 if (ap->pflags & ATA_PFLAG_RECOVERED)
515 ata_port_printk(ap, KERN_INFO, "EH complete\n");
517 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
519 /* tell wait_eh that we're done */
520 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
521 wake_up_all(&ap->eh_wait_q);
523 spin_unlock_irqrestore(ap->lock, flags);
525 DPRINTK("EXIT\n");
529 * ata_port_wait_eh - Wait for the currently pending EH to complete
530 * @ap: Port to wait EH for
532 * Wait until the currently pending EH is complete.
534 * LOCKING:
535 * Kernel thread context (may sleep).
537 void ata_port_wait_eh(struct ata_port *ap)
539 unsigned long flags;
540 DEFINE_WAIT(wait);
542 retry:
543 spin_lock_irqsave(ap->lock, flags);
545 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
546 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
547 spin_unlock_irqrestore(ap->lock, flags);
548 schedule();
549 spin_lock_irqsave(ap->lock, flags);
551 finish_wait(&ap->eh_wait_q, &wait);
553 spin_unlock_irqrestore(ap->lock, flags);
555 /* make sure SCSI EH is complete */
556 if (scsi_host_in_recovery(ap->scsi_host)) {
557 msleep(10);
558 goto retry;
563 * ata_qc_timeout - Handle timeout of queued command
564 * @qc: Command that timed out
566 * Some part of the kernel (currently, only the SCSI layer)
567 * has noticed that the active command on port @ap has not
568 * completed after a specified length of time. Handle this
569 * condition by disabling DMA (if necessary) and completing
570 * transactions, with error if necessary.
572 * This also handles the case of the "lost interrupt", where
573 * for some reason (possibly hardware bug, possibly driver bug)
574 * an interrupt was not delivered to the driver, even though the
575 * transaction completed successfully.
577 * TODO: kill this function once old EH is gone.
579 * LOCKING:
580 * Inherited from SCSI layer (none, can sleep)
582 static void ata_qc_timeout(struct ata_queued_cmd *qc)
584 struct ata_port *ap = qc->ap;
585 u8 host_stat = 0, drv_stat;
586 unsigned long flags;
588 DPRINTK("ENTER\n");
590 ap->hsm_task_state = HSM_ST_IDLE;
592 spin_lock_irqsave(ap->lock, flags);
594 switch (qc->tf.protocol) {
596 case ATA_PROT_DMA:
597 case ATA_PROT_ATAPI_DMA:
598 host_stat = ap->ops->bmdma_status(ap);
600 /* before we do anything else, clear DMA-Start bit */
601 ap->ops->bmdma_stop(qc);
603 /* fall through */
605 default:
606 ata_altstatus(ap);
607 drv_stat = ata_chk_status(ap);
609 /* ack bmdma irq events */
610 ap->ops->irq_clear(ap);
612 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
613 "stat 0x%x host_stat 0x%x\n",
614 qc->tf.command, drv_stat, host_stat);
616 /* complete taskfile transaction */
617 qc->err_mask |= AC_ERR_TIMEOUT;
618 break;
621 spin_unlock_irqrestore(ap->lock, flags);
623 ata_eh_qc_complete(qc);
625 DPRINTK("EXIT\n");
629 * ata_eng_timeout - Handle timeout of queued command
630 * @ap: Port on which timed-out command is active
632 * Some part of the kernel (currently, only the SCSI layer)
633 * has noticed that the active command on port @ap has not
634 * completed after a specified length of time. Handle this
635 * condition by disabling DMA (if necessary) and completing
636 * transactions, with error if necessary.
638 * This also handles the case of the "lost interrupt", where
639 * for some reason (possibly hardware bug, possibly driver bug)
640 * an interrupt was not delivered to the driver, even though the
641 * transaction completed successfully.
643 * TODO: kill this function once old EH is gone.
645 * LOCKING:
646 * Inherited from SCSI layer (none, can sleep)
648 void ata_eng_timeout(struct ata_port *ap)
650 DPRINTK("ENTER\n");
652 ata_qc_timeout(ata_qc_from_tag(ap, ap->link.active_tag));
654 DPRINTK("EXIT\n");
657 static int ata_eh_nr_in_flight(struct ata_port *ap)
659 unsigned int tag;
660 int nr = 0;
662 /* count only non-internal commands */
663 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
664 if (ata_qc_from_tag(ap, tag))
665 nr++;
667 return nr;
670 void ata_eh_fastdrain_timerfn(unsigned long arg)
672 struct ata_port *ap = (void *)arg;
673 unsigned long flags;
674 int cnt;
676 spin_lock_irqsave(ap->lock, flags);
678 cnt = ata_eh_nr_in_flight(ap);
680 /* are we done? */
681 if (!cnt)
682 goto out_unlock;
684 if (cnt == ap->fastdrain_cnt) {
685 unsigned int tag;
687 /* No progress during the last interval, tag all
688 * in-flight qcs as timed out and freeze the port.
690 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
691 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
692 if (qc)
693 qc->err_mask |= AC_ERR_TIMEOUT;
696 ata_port_freeze(ap);
697 } else {
698 /* some qcs have finished, give it another chance */
699 ap->fastdrain_cnt = cnt;
700 ap->fastdrain_timer.expires =
701 jiffies + ATA_EH_FASTDRAIN_INTERVAL;
702 add_timer(&ap->fastdrain_timer);
705 out_unlock:
706 spin_unlock_irqrestore(ap->lock, flags);
710 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
711 * @ap: target ATA port
712 * @fastdrain: activate fast drain
714 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
715 * is non-zero and EH wasn't pending before. Fast drain ensures
716 * that EH kicks in in timely manner.
718 * LOCKING:
719 * spin_lock_irqsave(host lock)
721 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
723 int cnt;
725 /* already scheduled? */
726 if (ap->pflags & ATA_PFLAG_EH_PENDING)
727 return;
729 ap->pflags |= ATA_PFLAG_EH_PENDING;
731 if (!fastdrain)
732 return;
734 /* do we have in-flight qcs? */
735 cnt = ata_eh_nr_in_flight(ap);
736 if (!cnt)
737 return;
739 /* activate fast drain */
740 ap->fastdrain_cnt = cnt;
741 ap->fastdrain_timer.expires = jiffies + ATA_EH_FASTDRAIN_INTERVAL;
742 add_timer(&ap->fastdrain_timer);
746 * ata_qc_schedule_eh - schedule qc for error handling
747 * @qc: command to schedule error handling for
749 * Schedule error handling for @qc. EH will kick in as soon as
750 * other commands are drained.
752 * LOCKING:
753 * spin_lock_irqsave(host lock)
755 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
757 struct ata_port *ap = qc->ap;
759 WARN_ON(!ap->ops->error_handler);
761 qc->flags |= ATA_QCFLAG_FAILED;
762 ata_eh_set_pending(ap, 1);
764 /* The following will fail if timeout has already expired.
765 * ata_scsi_error() takes care of such scmds on EH entry.
766 * Note that ATA_QCFLAG_FAILED is unconditionally set after
767 * this function completes.
769 scsi_req_abort_cmd(qc->scsicmd);
773 * ata_port_schedule_eh - schedule error handling without a qc
774 * @ap: ATA port to schedule EH for
776 * Schedule error handling for @ap. EH will kick in as soon as
777 * all commands are drained.
779 * LOCKING:
780 * spin_lock_irqsave(host lock)
782 void ata_port_schedule_eh(struct ata_port *ap)
784 WARN_ON(!ap->ops->error_handler);
786 if (ap->pflags & ATA_PFLAG_INITIALIZING)
787 return;
789 ata_eh_set_pending(ap, 1);
790 scsi_schedule_eh(ap->scsi_host);
792 DPRINTK("port EH scheduled\n");
795 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
797 int tag, nr_aborted = 0;
799 WARN_ON(!ap->ops->error_handler);
801 /* we're gonna abort all commands, no need for fast drain */
802 ata_eh_set_pending(ap, 0);
804 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
805 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
807 if (qc && (!link || qc->dev->link == link)) {
808 qc->flags |= ATA_QCFLAG_FAILED;
809 ata_qc_complete(qc);
810 nr_aborted++;
814 if (!nr_aborted)
815 ata_port_schedule_eh(ap);
817 return nr_aborted;
821 * ata_link_abort - abort all qc's on the link
822 * @link: ATA link to abort qc's for
824 * Abort all active qc's active on @link and schedule EH.
826 * LOCKING:
827 * spin_lock_irqsave(host lock)
829 * RETURNS:
830 * Number of aborted qc's.
832 int ata_link_abort(struct ata_link *link)
834 return ata_do_link_abort(link->ap, link);
838 * ata_port_abort - abort all qc's on the port
839 * @ap: ATA port to abort qc's for
841 * Abort all active qc's of @ap and schedule EH.
843 * LOCKING:
844 * spin_lock_irqsave(host_set lock)
846 * RETURNS:
847 * Number of aborted qc's.
849 int ata_port_abort(struct ata_port *ap)
851 return ata_do_link_abort(ap, NULL);
855 * __ata_port_freeze - freeze port
856 * @ap: ATA port to freeze
858 * This function is called when HSM violation or some other
859 * condition disrupts normal operation of the port. Frozen port
860 * is not allowed to perform any operation until the port is
861 * thawed, which usually follows a successful reset.
863 * ap->ops->freeze() callback can be used for freezing the port
864 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
865 * port cannot be frozen hardware-wise, the interrupt handler
866 * must ack and clear interrupts unconditionally while the port
867 * is frozen.
869 * LOCKING:
870 * spin_lock_irqsave(host lock)
872 static void __ata_port_freeze(struct ata_port *ap)
874 WARN_ON(!ap->ops->error_handler);
876 if (ap->ops->freeze)
877 ap->ops->freeze(ap);
879 ap->pflags |= ATA_PFLAG_FROZEN;
881 DPRINTK("ata%u port frozen\n", ap->print_id);
885 * ata_port_freeze - abort & freeze port
886 * @ap: ATA port to freeze
888 * Abort and freeze @ap.
890 * LOCKING:
891 * spin_lock_irqsave(host lock)
893 * RETURNS:
894 * Number of aborted commands.
896 int ata_port_freeze(struct ata_port *ap)
898 int nr_aborted;
900 WARN_ON(!ap->ops->error_handler);
902 nr_aborted = ata_port_abort(ap);
903 __ata_port_freeze(ap);
905 return nr_aborted;
909 * sata_async_notification - SATA async notification handler
910 * @ap: ATA port where async notification is received
912 * Handler to be called when async notification via SDB FIS is
913 * received. This function schedules EH if necessary.
915 * LOCKING:
916 * spin_lock_irqsave(host lock)
918 * RETURNS:
919 * 1 if EH is scheduled, 0 otherwise.
921 int sata_async_notification(struct ata_port *ap)
923 u32 sntf;
924 int rc;
926 if (!(ap->flags & ATA_FLAG_AN))
927 return 0;
929 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
930 if (rc == 0)
931 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
933 if (!ap->nr_pmp_links || rc) {
934 /* PMP is not attached or SNTF is not available */
935 if (!ap->nr_pmp_links) {
936 /* PMP is not attached. Check whether ATAPI
937 * AN is configured. If so, notify media
938 * change.
940 struct ata_device *dev = ap->link.device;
942 if ((dev->class == ATA_DEV_ATAPI) &&
943 (dev->flags & ATA_DFLAG_AN))
944 ata_scsi_media_change_notify(dev);
945 return 0;
946 } else {
947 /* PMP is attached but SNTF is not available.
948 * ATAPI async media change notification is
949 * not used. The PMP must be reporting PHY
950 * status change, schedule EH.
952 ata_port_schedule_eh(ap);
953 return 1;
955 } else {
956 /* PMP is attached and SNTF is available */
957 struct ata_link *link;
959 /* check and notify ATAPI AN */
960 ata_port_for_each_link(link, ap) {
961 if (!(sntf & (1 << link->pmp)))
962 continue;
964 if ((link->device->class == ATA_DEV_ATAPI) &&
965 (link->device->flags & ATA_DFLAG_AN))
966 ata_scsi_media_change_notify(link->device);
969 /* If PMP is reporting that PHY status of some
970 * downstream ports has changed, schedule EH.
972 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
973 ata_port_schedule_eh(ap);
974 return 1;
977 return 0;
982 * ata_eh_freeze_port - EH helper to freeze port
983 * @ap: ATA port to freeze
985 * Freeze @ap.
987 * LOCKING:
988 * None.
990 void ata_eh_freeze_port(struct ata_port *ap)
992 unsigned long flags;
994 if (!ap->ops->error_handler)
995 return;
997 spin_lock_irqsave(ap->lock, flags);
998 __ata_port_freeze(ap);
999 spin_unlock_irqrestore(ap->lock, flags);
1003 * ata_port_thaw_port - EH helper to thaw port
1004 * @ap: ATA port to thaw
1006 * Thaw frozen port @ap.
1008 * LOCKING:
1009 * None.
1011 void ata_eh_thaw_port(struct ata_port *ap)
1013 unsigned long flags;
1015 if (!ap->ops->error_handler)
1016 return;
1018 spin_lock_irqsave(ap->lock, flags);
1020 ap->pflags &= ~ATA_PFLAG_FROZEN;
1022 if (ap->ops->thaw)
1023 ap->ops->thaw(ap);
1025 spin_unlock_irqrestore(ap->lock, flags);
1027 DPRINTK("ata%u port thawed\n", ap->print_id);
1030 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1032 /* nada */
1035 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1037 struct ata_port *ap = qc->ap;
1038 struct scsi_cmnd *scmd = qc->scsicmd;
1039 unsigned long flags;
1041 spin_lock_irqsave(ap->lock, flags);
1042 qc->scsidone = ata_eh_scsidone;
1043 __ata_qc_complete(qc);
1044 WARN_ON(ata_tag_valid(qc->tag));
1045 spin_unlock_irqrestore(ap->lock, flags);
1047 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1051 * ata_eh_qc_complete - Complete an active ATA command from EH
1052 * @qc: Command to complete
1054 * Indicate to the mid and upper layers that an ATA command has
1055 * completed. To be used from EH.
1057 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1059 struct scsi_cmnd *scmd = qc->scsicmd;
1060 scmd->retries = scmd->allowed;
1061 __ata_eh_qc_complete(qc);
1065 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1066 * @qc: Command to retry
1068 * Indicate to the mid and upper layers that an ATA command
1069 * should be retried. To be used from EH.
1071 * SCSI midlayer limits the number of retries to scmd->allowed.
1072 * scmd->retries is decremented for commands which get retried
1073 * due to unrelated failures (qc->err_mask is zero).
1075 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1077 struct scsi_cmnd *scmd = qc->scsicmd;
1078 if (!qc->err_mask && scmd->retries)
1079 scmd->retries--;
1080 __ata_eh_qc_complete(qc);
1084 * ata_eh_detach_dev - detach ATA device
1085 * @dev: ATA device to detach
1087 * Detach @dev.
1089 * LOCKING:
1090 * None.
1092 void ata_eh_detach_dev(struct ata_device *dev)
1094 struct ata_link *link = dev->link;
1095 struct ata_port *ap = link->ap;
1096 unsigned long flags;
1098 ata_dev_disable(dev);
1100 spin_lock_irqsave(ap->lock, flags);
1102 dev->flags &= ~ATA_DFLAG_DETACH;
1104 if (ata_scsi_offline_dev(dev)) {
1105 dev->flags |= ATA_DFLAG_DETACHED;
1106 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1109 /* clear per-dev EH actions */
1110 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1111 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1113 spin_unlock_irqrestore(ap->lock, flags);
1117 * ata_eh_about_to_do - about to perform eh_action
1118 * @link: target ATA link
1119 * @dev: target ATA dev for per-dev action (can be NULL)
1120 * @action: action about to be performed
1122 * Called just before performing EH actions to clear related bits
1123 * in @link->eh_info such that eh actions are not unnecessarily
1124 * repeated.
1126 * LOCKING:
1127 * None.
1129 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1130 unsigned int action)
1132 struct ata_port *ap = link->ap;
1133 struct ata_eh_info *ehi = &link->eh_info;
1134 struct ata_eh_context *ehc = &link->eh_context;
1135 unsigned long flags;
1137 spin_lock_irqsave(ap->lock, flags);
1139 /* Reset is represented by combination of actions and EHI
1140 * flags. Suck in all related bits before clearing eh_info to
1141 * avoid losing requested action.
1143 if (action & ATA_EH_RESET_MASK) {
1144 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
1145 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
1147 /* make sure all reset actions are cleared & clear EHI flags */
1148 action |= ATA_EH_RESET_MASK;
1149 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
1152 ata_eh_clear_action(link, dev, ehi, action);
1154 if (!(ehc->i.flags & ATA_EHI_QUIET))
1155 ap->pflags |= ATA_PFLAG_RECOVERED;
1157 spin_unlock_irqrestore(ap->lock, flags);
1161 * ata_eh_done - EH action complete
1162 * @ap: target ATA port
1163 * @dev: target ATA dev for per-dev action (can be NULL)
1164 * @action: action just completed
1166 * Called right after performing EH actions to clear related bits
1167 * in @link->eh_context.
1169 * LOCKING:
1170 * None.
1172 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1173 unsigned int action)
1175 struct ata_eh_context *ehc = &link->eh_context;
1177 /* if reset is complete, clear all reset actions & reset modifier */
1178 if (action & ATA_EH_RESET_MASK) {
1179 action |= ATA_EH_RESET_MASK;
1180 ehc->i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
1183 ata_eh_clear_action(link, dev, &ehc->i, action);
1187 * ata_err_string - convert err_mask to descriptive string
1188 * @err_mask: error mask to convert to string
1190 * Convert @err_mask to descriptive string. Errors are
1191 * prioritized according to severity and only the most severe
1192 * error is reported.
1194 * LOCKING:
1195 * None.
1197 * RETURNS:
1198 * Descriptive string for @err_mask
1200 static const char * ata_err_string(unsigned int err_mask)
1202 if (err_mask & AC_ERR_HOST_BUS)
1203 return "host bus error";
1204 if (err_mask & AC_ERR_ATA_BUS)
1205 return "ATA bus error";
1206 if (err_mask & AC_ERR_TIMEOUT)
1207 return "timeout";
1208 if (err_mask & AC_ERR_HSM)
1209 return "HSM violation";
1210 if (err_mask & AC_ERR_SYSTEM)
1211 return "internal error";
1212 if (err_mask & AC_ERR_MEDIA)
1213 return "media error";
1214 if (err_mask & AC_ERR_INVALID)
1215 return "invalid argument";
1216 if (err_mask & AC_ERR_DEV)
1217 return "device error";
1218 return "unknown error";
1222 * ata_read_log_page - read a specific log page
1223 * @dev: target device
1224 * @page: page to read
1225 * @buf: buffer to store read page
1226 * @sectors: number of sectors to read
1228 * Read log page using READ_LOG_EXT command.
1230 * LOCKING:
1231 * Kernel thread context (may sleep).
1233 * RETURNS:
1234 * 0 on success, AC_ERR_* mask otherwise.
1236 static unsigned int ata_read_log_page(struct ata_device *dev,
1237 u8 page, void *buf, unsigned int sectors)
1239 struct ata_taskfile tf;
1240 unsigned int err_mask;
1242 DPRINTK("read log page - page %d\n", page);
1244 ata_tf_init(dev, &tf);
1245 tf.command = ATA_CMD_READ_LOG_EXT;
1246 tf.lbal = page;
1247 tf.nsect = sectors;
1248 tf.hob_nsect = sectors >> 8;
1249 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1250 tf.protocol = ATA_PROT_PIO;
1252 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1253 buf, sectors * ATA_SECT_SIZE, 0);
1255 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1256 return err_mask;
1260 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1261 * @dev: Device to read log page 10h from
1262 * @tag: Resulting tag of the failed command
1263 * @tf: Resulting taskfile registers of the failed command
1265 * Read log page 10h to obtain NCQ error details and clear error
1266 * condition.
1268 * LOCKING:
1269 * Kernel thread context (may sleep).
1271 * RETURNS:
1272 * 0 on success, -errno otherwise.
1274 static int ata_eh_read_log_10h(struct ata_device *dev,
1275 int *tag, struct ata_taskfile *tf)
1277 u8 *buf = dev->link->ap->sector_buf;
1278 unsigned int err_mask;
1279 u8 csum;
1280 int i;
1282 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1283 if (err_mask)
1284 return -EIO;
1286 csum = 0;
1287 for (i = 0; i < ATA_SECT_SIZE; i++)
1288 csum += buf[i];
1289 if (csum)
1290 ata_dev_printk(dev, KERN_WARNING,
1291 "invalid checksum 0x%x on log page 10h\n", csum);
1293 if (buf[0] & 0x80)
1294 return -ENOENT;
1296 *tag = buf[0] & 0x1f;
1298 tf->command = buf[2];
1299 tf->feature = buf[3];
1300 tf->lbal = buf[4];
1301 tf->lbam = buf[5];
1302 tf->lbah = buf[6];
1303 tf->device = buf[7];
1304 tf->hob_lbal = buf[8];
1305 tf->hob_lbam = buf[9];
1306 tf->hob_lbah = buf[10];
1307 tf->nsect = buf[12];
1308 tf->hob_nsect = buf[13];
1310 return 0;
1314 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1315 * @dev: device to perform REQUEST_SENSE to
1316 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1318 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1319 * SENSE. This function is EH helper.
1321 * LOCKING:
1322 * Kernel thread context (may sleep).
1324 * RETURNS:
1325 * 0 on success, AC_ERR_* mask on failure
1327 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
1329 struct ata_device *dev = qc->dev;
1330 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
1331 struct ata_port *ap = dev->link->ap;
1332 struct ata_taskfile tf;
1333 u8 cdb[ATAPI_CDB_LEN];
1335 DPRINTK("ATAPI request sense\n");
1337 /* FIXME: is this needed? */
1338 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1340 /* initialize sense_buf with the error register,
1341 * for the case where they are -not- overwritten
1343 sense_buf[0] = 0x70;
1344 sense_buf[2] = qc->result_tf.feature >> 4;
1346 /* some devices time out if garbage left in tf */
1347 ata_tf_init(dev, &tf);
1349 memset(cdb, 0, ATAPI_CDB_LEN);
1350 cdb[0] = REQUEST_SENSE;
1351 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1353 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1354 tf.command = ATA_CMD_PACKET;
1356 /* is it pointless to prefer PIO for "safety reasons"? */
1357 if (ap->flags & ATA_FLAG_PIO_DMA) {
1358 tf.protocol = ATA_PROT_ATAPI_DMA;
1359 tf.feature |= ATAPI_PKT_DMA;
1360 } else {
1361 tf.protocol = ATA_PROT_ATAPI;
1362 tf.lbam = (8 * 1024) & 0xff;
1363 tf.lbah = (8 * 1024) >> 8;
1366 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1367 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1371 * ata_eh_analyze_serror - analyze SError for a failed port
1372 * @link: ATA link to analyze SError for
1374 * Analyze SError if available and further determine cause of
1375 * failure.
1377 * LOCKING:
1378 * None.
1380 static void ata_eh_analyze_serror(struct ata_link *link)
1382 struct ata_eh_context *ehc = &link->eh_context;
1383 u32 serror = ehc->i.serror;
1384 unsigned int err_mask = 0, action = 0;
1385 u32 hotplug_mask;
1387 if (serror & SERR_PERSISTENT) {
1388 err_mask |= AC_ERR_ATA_BUS;
1389 action |= ATA_EH_HARDRESET;
1391 if (serror &
1392 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1393 err_mask |= AC_ERR_ATA_BUS;
1394 action |= ATA_EH_SOFTRESET;
1396 if (serror & SERR_PROTOCOL) {
1397 err_mask |= AC_ERR_HSM;
1398 action |= ATA_EH_SOFTRESET;
1400 if (serror & SERR_INTERNAL) {
1401 err_mask |= AC_ERR_SYSTEM;
1402 action |= ATA_EH_HARDRESET;
1405 /* Determine whether a hotplug event has occurred. Both
1406 * SError.N/X are considered hotplug events for enabled or
1407 * host links. For disabled PMP links, only N bit is
1408 * considered as X bit is left at 1 for link plugging.
1410 hotplug_mask = 0;
1412 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1413 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1414 else
1415 hotplug_mask = SERR_PHYRDY_CHG;
1417 if (serror & hotplug_mask)
1418 ata_ehi_hotplugged(&ehc->i);
1420 ehc->i.err_mask |= err_mask;
1421 ehc->i.action |= action;
1425 * ata_eh_analyze_ncq_error - analyze NCQ error
1426 * @link: ATA link to analyze NCQ error for
1428 * Read log page 10h, determine the offending qc and acquire
1429 * error status TF. For NCQ device errors, all LLDDs have to do
1430 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1431 * care of the rest.
1433 * LOCKING:
1434 * Kernel thread context (may sleep).
1436 static void ata_eh_analyze_ncq_error(struct ata_link *link)
1438 struct ata_port *ap = link->ap;
1439 struct ata_eh_context *ehc = &link->eh_context;
1440 struct ata_device *dev = link->device;
1441 struct ata_queued_cmd *qc;
1442 struct ata_taskfile tf;
1443 int tag, rc;
1445 /* if frozen, we can't do much */
1446 if (ap->pflags & ATA_PFLAG_FROZEN)
1447 return;
1449 /* is it NCQ device error? */
1450 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1451 return;
1453 /* has LLDD analyzed already? */
1454 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1455 qc = __ata_qc_from_tag(ap, tag);
1457 if (!(qc->flags & ATA_QCFLAG_FAILED))
1458 continue;
1460 if (qc->err_mask)
1461 return;
1464 /* okay, this error is ours */
1465 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1466 if (rc) {
1467 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1468 "(errno=%d)\n", rc);
1469 return;
1472 if (!(link->sactive & (1 << tag))) {
1473 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1474 "inactive tag %d\n", tag);
1475 return;
1478 /* we've got the perpetrator, condemn it */
1479 qc = __ata_qc_from_tag(ap, tag);
1480 memcpy(&qc->result_tf, &tf, sizeof(tf));
1481 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1482 ehc->i.err_mask &= ~AC_ERR_DEV;
1486 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1487 * @qc: qc to analyze
1488 * @tf: Taskfile registers to analyze
1490 * Analyze taskfile of @qc and further determine cause of
1491 * failure. This function also requests ATAPI sense data if
1492 * avaliable.
1494 * LOCKING:
1495 * Kernel thread context (may sleep).
1497 * RETURNS:
1498 * Determined recovery action
1500 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1501 const struct ata_taskfile *tf)
1503 unsigned int tmp, action = 0;
1504 u8 stat = tf->command, err = tf->feature;
1506 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1507 qc->err_mask |= AC_ERR_HSM;
1508 return ATA_EH_SOFTRESET;
1511 if (stat & (ATA_ERR | ATA_DF))
1512 qc->err_mask |= AC_ERR_DEV;
1513 else
1514 return 0;
1516 switch (qc->dev->class) {
1517 case ATA_DEV_ATA:
1518 if (err & ATA_ICRC)
1519 qc->err_mask |= AC_ERR_ATA_BUS;
1520 if (err & ATA_UNC)
1521 qc->err_mask |= AC_ERR_MEDIA;
1522 if (err & ATA_IDNF)
1523 qc->err_mask |= AC_ERR_INVALID;
1524 break;
1526 case ATA_DEV_ATAPI:
1527 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1528 tmp = atapi_eh_request_sense(qc);
1529 if (!tmp) {
1530 /* ATA_QCFLAG_SENSE_VALID is used to
1531 * tell atapi_qc_complete() that sense
1532 * data is already valid.
1534 * TODO: interpret sense data and set
1535 * appropriate err_mask.
1537 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1538 } else
1539 qc->err_mask |= tmp;
1543 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1544 action |= ATA_EH_SOFTRESET;
1546 return action;
1549 static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
1551 if (err_mask & AC_ERR_ATA_BUS)
1552 return 1;
1554 if (err_mask & AC_ERR_TIMEOUT)
1555 return 2;
1557 if (is_io) {
1558 if (err_mask & AC_ERR_HSM)
1559 return 2;
1560 if ((err_mask &
1561 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1562 return 3;
1565 return 0;
1568 struct speed_down_verdict_arg {
1569 u64 since;
1570 int nr_errors[4];
1573 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1575 struct speed_down_verdict_arg *arg = void_arg;
1576 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
1578 if (ent->timestamp < arg->since)
1579 return -1;
1581 arg->nr_errors[cat]++;
1582 return 0;
1586 * ata_eh_speed_down_verdict - Determine speed down verdict
1587 * @dev: Device of interest
1589 * This function examines error ring of @dev and determines
1590 * whether NCQ needs to be turned off, transfer speed should be
1591 * stepped down, or falling back to PIO is necessary.
1593 * Cat-1 is ATA_BUS error for any command.
1595 * Cat-2 is TIMEOUT for any command or HSM violation for known
1596 * supported commands.
1598 * Cat-3 is is unclassified DEV error for known supported
1599 * command.
1601 * NCQ needs to be turned off if there have been more than 3
1602 * Cat-2 + Cat-3 errors during last 10 minutes.
1604 * Speed down is necessary if there have been more than 3 Cat-1 +
1605 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1607 * Falling back to PIO mode is necessary if there have been more
1608 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1610 * LOCKING:
1611 * Inherited from caller.
1613 * RETURNS:
1614 * OR of ATA_EH_SPDN_* flags.
1616 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1618 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1619 u64 j64 = get_jiffies_64();
1620 struct speed_down_verdict_arg arg;
1621 unsigned int verdict = 0;
1623 /* scan past 10 mins of error history */
1624 memset(&arg, 0, sizeof(arg));
1625 arg.since = j64 - min(j64, j10mins);
1626 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1628 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1629 verdict |= ATA_EH_SPDN_NCQ_OFF;
1630 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1631 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1633 /* scan past 3 mins of error history */
1634 memset(&arg, 0, sizeof(arg));
1635 arg.since = j64 - min(j64, j5mins);
1636 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1638 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1639 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1641 return verdict;
1645 * ata_eh_speed_down - record error and speed down if necessary
1646 * @dev: Failed device
1647 * @is_io: Did the device fail during normal IO?
1648 * @err_mask: err_mask of the error
1650 * Record error and examine error history to determine whether
1651 * adjusting transmission speed is necessary. It also sets
1652 * transmission limits appropriately if such adjustment is
1653 * necessary.
1655 * LOCKING:
1656 * Kernel thread context (may sleep).
1658 * RETURNS:
1659 * Determined recovery action.
1661 static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1662 unsigned int err_mask)
1664 unsigned int verdict;
1665 unsigned int action = 0;
1667 /* don't bother if Cat-0 error */
1668 if (ata_eh_categorize_error(is_io, err_mask) == 0)
1669 return 0;
1671 /* record error and determine whether speed down is necessary */
1672 ata_ering_record(&dev->ering, is_io, err_mask);
1673 verdict = ata_eh_speed_down_verdict(dev);
1675 /* turn off NCQ? */
1676 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1677 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1678 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1679 dev->flags |= ATA_DFLAG_NCQ_OFF;
1680 ata_dev_printk(dev, KERN_WARNING,
1681 "NCQ disabled due to excessive errors\n");
1682 goto done;
1685 /* speed down? */
1686 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1687 /* speed down SATA link speed if possible */
1688 if (sata_down_spd_limit(dev->link) == 0) {
1689 action |= ATA_EH_HARDRESET;
1690 goto done;
1693 /* lower transfer mode */
1694 if (dev->spdn_cnt < 2) {
1695 static const int dma_dnxfer_sel[] =
1696 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1697 static const int pio_dnxfer_sel[] =
1698 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1699 int sel;
1701 if (dev->xfer_shift != ATA_SHIFT_PIO)
1702 sel = dma_dnxfer_sel[dev->spdn_cnt];
1703 else
1704 sel = pio_dnxfer_sel[dev->spdn_cnt];
1706 dev->spdn_cnt++;
1708 if (ata_down_xfermask_limit(dev, sel) == 0) {
1709 action |= ATA_EH_SOFTRESET;
1710 goto done;
1715 /* Fall back to PIO? Slowing down to PIO is meaningless for
1716 * SATA. Consider it only for PATA.
1718 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1719 (dev->link->ap->cbl != ATA_CBL_SATA) &&
1720 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1721 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1722 dev->spdn_cnt = 0;
1723 action |= ATA_EH_SOFTRESET;
1724 goto done;
1728 return 0;
1729 done:
1730 /* device has been slowed down, blow error history */
1731 ata_ering_clear(&dev->ering);
1732 return action;
1736 * ata_eh_link_autopsy - analyze error and determine recovery action
1737 * @link: host link to perform autopsy on
1739 * Analyze why @link failed and determine which recovery actions
1740 * are needed. This function also sets more detailed AC_ERR_*
1741 * values and fills sense data for ATAPI CHECK SENSE.
1743 * LOCKING:
1744 * Kernel thread context (may sleep).
1746 static void ata_eh_link_autopsy(struct ata_link *link)
1748 struct ata_port *ap = link->ap;
1749 struct ata_eh_context *ehc = &link->eh_context;
1750 unsigned int all_err_mask = 0;
1751 int tag, is_io = 0;
1752 u32 serror;
1753 int rc;
1755 DPRINTK("ENTER\n");
1757 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1758 return;
1760 /* obtain and analyze SError */
1761 rc = sata_scr_read(link, SCR_ERROR, &serror);
1762 if (rc == 0) {
1763 ehc->i.serror |= serror;
1764 ata_eh_analyze_serror(link);
1765 } else if (rc != -EOPNOTSUPP) {
1766 /* SError read failed, force hardreset and probing */
1767 ata_ehi_schedule_probe(&ehc->i);
1768 ehc->i.action |= ATA_EH_HARDRESET;
1769 ehc->i.err_mask |= AC_ERR_OTHER;
1772 /* analyze NCQ failure */
1773 ata_eh_analyze_ncq_error(link);
1775 /* any real error trumps AC_ERR_OTHER */
1776 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1777 ehc->i.err_mask &= ~AC_ERR_OTHER;
1779 all_err_mask |= ehc->i.err_mask;
1781 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1782 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1784 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link)
1785 continue;
1787 /* inherit upper level err_mask */
1788 qc->err_mask |= ehc->i.err_mask;
1790 /* analyze TF */
1791 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1793 /* DEV errors are probably spurious in case of ATA_BUS error */
1794 if (qc->err_mask & AC_ERR_ATA_BUS)
1795 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1796 AC_ERR_INVALID);
1798 /* any real error trumps unknown error */
1799 if (qc->err_mask & ~AC_ERR_OTHER)
1800 qc->err_mask &= ~AC_ERR_OTHER;
1802 /* SENSE_VALID trumps dev/unknown error and revalidation */
1803 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1804 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1805 ehc->i.action &= ~ATA_EH_REVALIDATE;
1808 /* accumulate error info */
1809 ehc->i.dev = qc->dev;
1810 all_err_mask |= qc->err_mask;
1811 if (qc->flags & ATA_QCFLAG_IO)
1812 is_io = 1;
1815 /* enforce default EH actions */
1816 if (ap->pflags & ATA_PFLAG_FROZEN ||
1817 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1818 ehc->i.action |= ATA_EH_SOFTRESET;
1819 else if (all_err_mask)
1820 ehc->i.action |= ATA_EH_REVALIDATE;
1822 /* if we have offending qcs and the associated failed device */
1823 if (ehc->i.dev) {
1824 /* speed down */
1825 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1826 all_err_mask);
1828 /* perform per-dev EH action only on the offending device */
1829 ehc->i.dev_action[ehc->i.dev->devno] |=
1830 ehc->i.action & ATA_EH_PERDEV_MASK;
1831 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1834 DPRINTK("EXIT\n");
1838 * ata_eh_autopsy - analyze error and determine recovery action
1839 * @ap: host port to perform autopsy on
1841 * Analyze all links of @ap and determine why they failed and
1842 * which recovery actions are needed.
1844 * LOCKING:
1845 * Kernel thread context (may sleep).
1847 void ata_eh_autopsy(struct ata_port *ap)
1849 struct ata_link *link;
1851 __ata_port_for_each_link(link, ap)
1852 ata_eh_link_autopsy(link);
1856 * ata_eh_link_report - report error handling to user
1857 * @link: ATA link EH is going on
1859 * Report EH to user.
1861 * LOCKING:
1862 * None.
1864 static void ata_eh_link_report(struct ata_link *link)
1866 struct ata_port *ap = link->ap;
1867 struct ata_eh_context *ehc = &link->eh_context;
1868 const char *frozen, *desc;
1869 char tries_buf[6];
1870 int tag, nr_failed = 0;
1872 if (ehc->i.flags & ATA_EHI_QUIET)
1873 return;
1875 desc = NULL;
1876 if (ehc->i.desc[0] != '\0')
1877 desc = ehc->i.desc;
1879 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1880 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1882 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link)
1883 continue;
1884 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1885 continue;
1887 nr_failed++;
1890 if (!nr_failed && !ehc->i.err_mask)
1891 return;
1893 frozen = "";
1894 if (ap->pflags & ATA_PFLAG_FROZEN)
1895 frozen = " frozen";
1897 memset(tries_buf, 0, sizeof(tries_buf));
1898 if (ap->eh_tries < ATA_EH_MAX_TRIES)
1899 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
1900 ap->eh_tries);
1902 if (ehc->i.dev) {
1903 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1904 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1905 ehc->i.err_mask, link->sactive, ehc->i.serror,
1906 ehc->i.action, frozen, tries_buf);
1907 if (desc)
1908 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
1909 } else {
1910 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
1911 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1912 ehc->i.err_mask, link->sactive, ehc->i.serror,
1913 ehc->i.action, frozen, tries_buf);
1914 if (desc)
1915 ata_link_printk(link, KERN_ERR, "%s\n", desc);
1918 if (ehc->i.serror)
1919 ata_port_printk(ap, KERN_ERR,
1920 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
1921 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
1922 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
1923 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
1924 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
1925 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
1926 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
1927 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
1928 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
1929 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
1930 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
1931 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
1932 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
1933 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
1934 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
1935 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
1936 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
1937 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "" );
1939 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1940 static const char *dma_str[] = {
1941 [DMA_BIDIRECTIONAL] = "bidi",
1942 [DMA_TO_DEVICE] = "out",
1943 [DMA_FROM_DEVICE] = "in",
1944 [DMA_NONE] = "",
1946 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1947 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
1949 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1950 qc->dev->link != link || !qc->err_mask)
1951 continue;
1953 ata_dev_printk(qc->dev, KERN_ERR,
1954 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1955 "tag %d cdb 0x%x data %u %s\n "
1956 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1957 "Emask 0x%x (%s)%s\n",
1958 cmd->command, cmd->feature, cmd->nsect,
1959 cmd->lbal, cmd->lbam, cmd->lbah,
1960 cmd->hob_feature, cmd->hob_nsect,
1961 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
1962 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
1963 dma_str[qc->dma_dir],
1964 res->command, res->feature, res->nsect,
1965 res->lbal, res->lbam, res->lbah,
1966 res->hob_feature, res->hob_nsect,
1967 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1968 res->device, qc->err_mask, ata_err_string(qc->err_mask),
1969 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
1971 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
1972 ATA_ERR) ) {
1973 if (res->command & ATA_BUSY)
1974 ata_dev_printk(qc->dev, KERN_ERR,
1975 "status: { Busy }\n" );
1976 else
1977 ata_dev_printk(qc->dev, KERN_ERR,
1978 "status: { %s%s%s%s}\n",
1979 res->command & ATA_DRDY ? "DRDY " : "",
1980 res->command & ATA_DF ? "DF " : "",
1981 res->command & ATA_DRQ ? "DRQ " : "",
1982 res->command & ATA_ERR ? "ERR " : "" );
1985 if (cmd->command != ATA_CMD_PACKET &&
1986 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
1987 ATA_ABORTED)))
1988 ata_dev_printk(qc->dev, KERN_ERR,
1989 "error: { %s%s%s%s}\n",
1990 res->feature & ATA_ICRC ? "ICRC " : "",
1991 res->feature & ATA_UNC ? "UNC " : "",
1992 res->feature & ATA_IDNF ? "IDNF " : "",
1993 res->feature & ATA_ABORTED ? "ABRT " : "" );
1998 * ata_eh_report - report error handling to user
1999 * @ap: ATA port to report EH about
2001 * Report EH to user.
2003 * LOCKING:
2004 * None.
2006 void ata_eh_report(struct ata_port *ap)
2008 struct ata_link *link;
2010 __ata_port_for_each_link(link, ap)
2011 ata_eh_link_report(link);
2014 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2015 unsigned int *classes, unsigned long deadline)
2017 struct ata_device *dev;
2018 int rc;
2020 ata_link_for_each_dev(dev, link)
2021 classes[dev->devno] = ATA_DEV_UNKNOWN;
2023 rc = reset(link, classes, deadline);
2024 if (rc)
2025 return rc;
2027 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
2028 * is complete and convert all ATA_DEV_UNKNOWN to
2029 * ATA_DEV_NONE.
2031 ata_link_for_each_dev(dev, link)
2032 if (classes[dev->devno] != ATA_DEV_UNKNOWN)
2033 break;
2035 if (dev) {
2036 ata_link_for_each_dev(dev, link) {
2037 if (classes[dev->devno] == ATA_DEV_UNKNOWN)
2038 classes[dev->devno] = ATA_DEV_NONE;
2042 return 0;
2045 static int ata_eh_followup_srst_needed(struct ata_link *link,
2046 int rc, int classify,
2047 const unsigned int *classes)
2049 if (link->flags & ATA_LFLAG_NO_SRST)
2050 return 0;
2051 if (rc == -EAGAIN)
2052 return 1;
2053 if (rc != 0)
2054 return 0;
2055 if ((link->ap->flags & ATA_FLAG_PMP) && ata_is_host_link(link))
2056 return 1;
2057 if (classify && !(link->flags & ATA_LFLAG_ASSUME_CLASS) &&
2058 classes[0] == ATA_DEV_UNKNOWN)
2059 return 1;
2060 return 0;
2063 int ata_eh_reset(struct ata_link *link, int classify,
2064 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2065 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2067 struct ata_port *ap = link->ap;
2068 struct ata_eh_context *ehc = &link->eh_context;
2069 unsigned int *classes = ehc->classes;
2070 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2071 int try = 0;
2072 struct ata_device *dev;
2073 unsigned long deadline;
2074 unsigned int action;
2075 ata_reset_fn_t reset;
2076 unsigned long flags;
2077 int rc;
2079 /* about to reset */
2080 spin_lock_irqsave(ap->lock, flags);
2081 ap->pflags |= ATA_PFLAG_RESETTING;
2082 spin_unlock_irqrestore(ap->lock, flags);
2084 ata_eh_about_to_do(link, NULL, ehc->i.action & ATA_EH_RESET_MASK);
2086 /* Determine which reset to use and record in ehc->i.action.
2087 * prereset() may examine and modify it.
2089 action = ehc->i.action;
2090 ehc->i.action &= ~ATA_EH_RESET_MASK;
2091 if (softreset && (!hardreset || (!(link->flags & ATA_LFLAG_NO_SRST) &&
2092 !sata_set_spd_needed(link) &&
2093 !(action & ATA_EH_HARDRESET))))
2094 ehc->i.action |= ATA_EH_SOFTRESET;
2095 else
2096 ehc->i.action |= ATA_EH_HARDRESET;
2098 if (prereset) {
2099 rc = prereset(link, jiffies + ATA_EH_PRERESET_TIMEOUT);
2100 if (rc) {
2101 if (rc == -ENOENT) {
2102 ata_link_printk(link, KERN_DEBUG,
2103 "port disabled. ignoring.\n");
2104 ehc->i.action &= ~ATA_EH_RESET_MASK;
2106 ata_link_for_each_dev(dev, link)
2107 classes[dev->devno] = ATA_DEV_NONE;
2109 rc = 0;
2110 } else
2111 ata_link_printk(link, KERN_ERR,
2112 "prereset failed (errno=%d)\n", rc);
2113 goto out;
2117 /* prereset() might have modified ehc->i.action */
2118 if (ehc->i.action & ATA_EH_HARDRESET)
2119 reset = hardreset;
2120 else if (ehc->i.action & ATA_EH_SOFTRESET)
2121 reset = softreset;
2122 else {
2123 /* prereset told us not to reset, bang classes and return */
2124 ata_link_for_each_dev(dev, link)
2125 classes[dev->devno] = ATA_DEV_NONE;
2126 rc = 0;
2127 goto out;
2130 /* did prereset() screw up? if so, fix up to avoid oopsing */
2131 if (!reset) {
2132 if (softreset)
2133 reset = softreset;
2134 else
2135 reset = hardreset;
2138 retry:
2139 deadline = jiffies + ata_eh_reset_timeouts[try++];
2141 /* shut up during boot probing */
2142 if (verbose)
2143 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2144 reset == softreset ? "soft" : "hard");
2146 /* mark that this EH session started with reset */
2147 if (reset == hardreset)
2148 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2149 else
2150 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2152 rc = ata_do_reset(link, reset, classes, deadline);
2154 if (reset == hardreset &&
2155 ata_eh_followup_srst_needed(link, rc, classify, classes)) {
2156 /* okay, let's do follow-up softreset */
2157 reset = softreset;
2159 if (!reset) {
2160 ata_link_printk(link, KERN_ERR,
2161 "follow-up softreset required "
2162 "but no softreset avaliable\n");
2163 rc = -EINVAL;
2164 goto out;
2167 ata_eh_about_to_do(link, NULL, ATA_EH_RESET_MASK);
2168 rc = ata_do_reset(link, reset, classes, deadline);
2170 if (rc == 0 && classify && classes[0] == ATA_DEV_UNKNOWN &&
2171 !(link->flags & ATA_LFLAG_ASSUME_CLASS)) {
2172 ata_link_printk(link, KERN_ERR,
2173 "classification failed\n");
2174 rc = -EINVAL;
2175 goto out;
2179 /* if we skipped follow-up srst, clear rc */
2180 if (rc == -EAGAIN)
2181 rc = 0;
2183 if (rc && rc != -ERESTART && try < ARRAY_SIZE(ata_eh_reset_timeouts)) {
2184 unsigned long now = jiffies;
2186 if (time_before(now, deadline)) {
2187 unsigned long delta = deadline - jiffies;
2189 ata_link_printk(link, KERN_WARNING, "reset failed "
2190 "(errno=%d), retrying in %u secs\n",
2191 rc, (jiffies_to_msecs(delta) + 999) / 1000);
2193 while (delta)
2194 delta = schedule_timeout_uninterruptible(delta);
2197 if (rc == -EPIPE ||
2198 try == ARRAY_SIZE(ata_eh_reset_timeouts) - 1)
2199 sata_down_spd_limit(link);
2200 if (hardreset)
2201 reset = hardreset;
2202 goto retry;
2205 if (rc == 0) {
2206 u32 sstatus;
2208 ata_link_for_each_dev(dev, link) {
2209 /* After the reset, the device state is PIO 0
2210 * and the controller state is undefined.
2211 * Record the mode.
2213 dev->pio_mode = XFER_PIO_0;
2215 if (ata_link_offline(link))
2216 continue;
2218 /* apply class override and convert UNKNOWN to NONE */
2219 if (link->flags & ATA_LFLAG_ASSUME_ATA)
2220 classes[dev->devno] = ATA_DEV_ATA;
2221 else if (link->flags & ATA_LFLAG_ASSUME_SEMB)
2222 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
2223 else if (classes[dev->devno] == ATA_DEV_UNKNOWN)
2224 classes[dev->devno] = ATA_DEV_NONE;
2227 /* record current link speed */
2228 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2229 link->sata_spd = (sstatus >> 4) & 0xf;
2231 if (postreset)
2232 postreset(link, classes);
2234 /* reset successful, schedule revalidation */
2235 ata_eh_done(link, NULL, ehc->i.action & ATA_EH_RESET_MASK);
2236 ehc->i.action |= ATA_EH_REVALIDATE;
2238 out:
2239 /* clear hotplug flag */
2240 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2242 spin_lock_irqsave(ap->lock, flags);
2243 ap->pflags &= ~ATA_PFLAG_RESETTING;
2244 spin_unlock_irqrestore(ap->lock, flags);
2246 return rc;
2249 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2250 struct ata_device **r_failed_dev)
2252 struct ata_port *ap = link->ap;
2253 struct ata_eh_context *ehc = &link->eh_context;
2254 struct ata_device *dev;
2255 unsigned int new_mask = 0;
2256 unsigned long flags;
2257 int rc = 0;
2259 DPRINTK("ENTER\n");
2261 /* For PATA drive side cable detection to work, IDENTIFY must
2262 * be done backwards such that PDIAG- is released by the slave
2263 * device before the master device is identified.
2265 ata_link_for_each_dev_reverse(dev, link) {
2266 unsigned int action = ata_eh_dev_action(dev);
2267 unsigned int readid_flags = 0;
2269 if (ehc->i.flags & ATA_EHI_DID_RESET)
2270 readid_flags |= ATA_READID_POSTRESET;
2272 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2273 WARN_ON(dev->class == ATA_DEV_PMP);
2275 if (ata_link_offline(link)) {
2276 rc = -EIO;
2277 goto err;
2280 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2281 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2282 readid_flags);
2283 if (rc)
2284 goto err;
2286 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2288 /* Configuration may have changed, reconfigure
2289 * transfer mode.
2291 ehc->i.flags |= ATA_EHI_SETMODE;
2293 /* schedule the scsi_rescan_device() here */
2294 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2295 } else if (dev->class == ATA_DEV_UNKNOWN &&
2296 ehc->tries[dev->devno] &&
2297 ata_class_enabled(ehc->classes[dev->devno])) {
2298 dev->class = ehc->classes[dev->devno];
2300 if (dev->class == ATA_DEV_PMP)
2301 rc = sata_pmp_attach(dev);
2302 else
2303 rc = ata_dev_read_id(dev, &dev->class,
2304 readid_flags, dev->id);
2305 switch (rc) {
2306 case 0:
2307 new_mask |= 1 << dev->devno;
2308 break;
2309 case -ENOENT:
2310 /* IDENTIFY was issued to non-existent
2311 * device. No need to reset. Just
2312 * thaw and kill the device.
2314 ata_eh_thaw_port(ap);
2315 dev->class = ATA_DEV_UNKNOWN;
2316 break;
2317 default:
2318 dev->class = ATA_DEV_UNKNOWN;
2319 goto err;
2324 /* PDIAG- should have been released, ask cable type if post-reset */
2325 if (ata_is_host_link(link) && ap->ops->cable_detect &&
2326 (ehc->i.flags & ATA_EHI_DID_RESET))
2327 ap->cbl = ap->ops->cable_detect(ap);
2329 /* Configure new devices forward such that user doesn't see
2330 * device detection messages backwards.
2332 ata_link_for_each_dev(dev, link) {
2333 if (!(new_mask & (1 << dev->devno)) ||
2334 dev->class == ATA_DEV_PMP)
2335 continue;
2337 ehc->i.flags |= ATA_EHI_PRINTINFO;
2338 rc = ata_dev_configure(dev);
2339 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2340 if (rc)
2341 goto err;
2343 spin_lock_irqsave(ap->lock, flags);
2344 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2345 spin_unlock_irqrestore(ap->lock, flags);
2347 /* new device discovered, configure xfermode */
2348 ehc->i.flags |= ATA_EHI_SETMODE;
2351 return 0;
2353 err:
2354 *r_failed_dev = dev;
2355 DPRINTK("EXIT rc=%d\n", rc);
2356 return rc;
2359 static int ata_link_nr_enabled(struct ata_link *link)
2361 struct ata_device *dev;
2362 int cnt = 0;
2364 ata_link_for_each_dev(dev, link)
2365 if (ata_dev_enabled(dev))
2366 cnt++;
2367 return cnt;
2370 static int ata_link_nr_vacant(struct ata_link *link)
2372 struct ata_device *dev;
2373 int cnt = 0;
2375 ata_link_for_each_dev(dev, link)
2376 if (dev->class == ATA_DEV_UNKNOWN)
2377 cnt++;
2378 return cnt;
2381 static int ata_eh_skip_recovery(struct ata_link *link)
2383 struct ata_eh_context *ehc = &link->eh_context;
2384 struct ata_device *dev;
2386 /* skip disabled links */
2387 if (link->flags & ATA_LFLAG_DISABLED)
2388 return 1;
2390 /* thaw frozen port, resume link and recover failed devices */
2391 if ((link->ap->pflags & ATA_PFLAG_FROZEN) ||
2392 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_link_nr_enabled(link))
2393 return 0;
2395 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2396 ata_link_for_each_dev(dev, link) {
2397 if (dev->class == ATA_DEV_UNKNOWN &&
2398 ehc->classes[dev->devno] != ATA_DEV_NONE)
2399 return 0;
2402 return 1;
2405 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2407 struct ata_eh_context *ehc = &dev->link->eh_context;
2409 ehc->tries[dev->devno]--;
2411 switch (err) {
2412 case -ENODEV:
2413 /* device missing or wrong IDENTIFY data, schedule probing */
2414 ehc->i.probe_mask |= (1 << dev->devno);
2415 case -EINVAL:
2416 /* give it just one more chance */
2417 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2418 case -EIO:
2419 if (ehc->tries[dev->devno] == 1) {
2420 /* This is the last chance, better to slow
2421 * down than lose it.
2423 sata_down_spd_limit(dev->link);
2424 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2428 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2429 /* disable device if it has used up all its chances */
2430 ata_dev_disable(dev);
2432 /* detach if offline */
2433 if (ata_link_offline(dev->link))
2434 ata_eh_detach_dev(dev);
2436 /* probe if requested */
2437 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2438 !(ehc->did_probe_mask & (1 << dev->devno))) {
2439 ata_eh_detach_dev(dev);
2440 ata_dev_init(dev);
2442 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2443 ehc->did_probe_mask |= (1 << dev->devno);
2444 ehc->i.action |= ATA_EH_SOFTRESET;
2447 return 1;
2448 } else {
2449 /* soft didn't work? be haaaaard */
2450 if (ehc->i.flags & ATA_EHI_DID_RESET)
2451 ehc->i.action |= ATA_EH_HARDRESET;
2452 else
2453 ehc->i.action |= ATA_EH_SOFTRESET;
2455 return 0;
2460 * ata_eh_recover - recover host port after error
2461 * @ap: host port to recover
2462 * @prereset: prereset method (can be NULL)
2463 * @softreset: softreset method (can be NULL)
2464 * @hardreset: hardreset method (can be NULL)
2465 * @postreset: postreset method (can be NULL)
2466 * @r_failed_link: out parameter for failed link
2468 * This is the alpha and omega, eum and yang, heart and soul of
2469 * libata exception handling. On entry, actions required to
2470 * recover each link and hotplug requests are recorded in the
2471 * link's eh_context. This function executes all the operations
2472 * with appropriate retrials and fallbacks to resurrect failed
2473 * devices, detach goners and greet newcomers.
2475 * LOCKING:
2476 * Kernel thread context (may sleep).
2478 * RETURNS:
2479 * 0 on success, -errno on failure.
2481 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2482 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2483 ata_postreset_fn_t postreset,
2484 struct ata_link **r_failed_link)
2486 struct ata_link *link;
2487 struct ata_device *dev;
2488 int nr_failed_devs, nr_disabled_devs;
2489 int reset, rc;
2490 unsigned long flags;
2492 DPRINTK("ENTER\n");
2494 /* prep for recovery */
2495 ata_port_for_each_link(link, ap) {
2496 struct ata_eh_context *ehc = &link->eh_context;
2498 /* re-enable link? */
2499 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
2500 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
2501 spin_lock_irqsave(ap->lock, flags);
2502 link->flags &= ~ATA_LFLAG_DISABLED;
2503 spin_unlock_irqrestore(ap->lock, flags);
2504 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
2507 ata_link_for_each_dev(dev, link) {
2508 if (link->flags & ATA_LFLAG_NO_RETRY)
2509 ehc->tries[dev->devno] = 1;
2510 else
2511 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2513 /* collect port action mask recorded in dev actions */
2514 ehc->i.action |= ehc->i.dev_action[dev->devno] &
2515 ~ATA_EH_PERDEV_MASK;
2516 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
2518 /* process hotplug request */
2519 if (dev->flags & ATA_DFLAG_DETACH)
2520 ata_eh_detach_dev(dev);
2522 if (!ata_dev_enabled(dev) &&
2523 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2524 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2525 ata_eh_detach_dev(dev);
2526 ata_dev_init(dev);
2527 ehc->did_probe_mask |= (1 << dev->devno);
2528 ehc->i.action |= ATA_EH_SOFTRESET;
2533 retry:
2534 rc = 0;
2535 nr_failed_devs = 0;
2536 nr_disabled_devs = 0;
2537 reset = 0;
2539 /* if UNLOADING, finish immediately */
2540 if (ap->pflags & ATA_PFLAG_UNLOADING)
2541 goto out;
2543 /* prep for EH */
2544 ata_port_for_each_link(link, ap) {
2545 struct ata_eh_context *ehc = &link->eh_context;
2547 /* skip EH if possible. */
2548 if (ata_eh_skip_recovery(link))
2549 ehc->i.action = 0;
2551 /* do we need to reset? */
2552 if (ehc->i.action & ATA_EH_RESET_MASK)
2553 reset = 1;
2555 ata_link_for_each_dev(dev, link)
2556 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
2559 /* reset */
2560 if (reset) {
2561 /* if PMP is attached, this function only deals with
2562 * downstream links, port should stay thawed.
2564 if (!ap->nr_pmp_links)
2565 ata_eh_freeze_port(ap);
2567 ata_port_for_each_link(link, ap) {
2568 struct ata_eh_context *ehc = &link->eh_context;
2570 if (!(ehc->i.action & ATA_EH_RESET_MASK))
2571 continue;
2573 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
2574 prereset, softreset, hardreset,
2575 postreset);
2576 if (rc) {
2577 ata_link_printk(link, KERN_ERR,
2578 "reset failed, giving up\n");
2579 goto out;
2583 if (!ap->nr_pmp_links)
2584 ata_eh_thaw_port(ap);
2587 /* the rest */
2588 ata_port_for_each_link(link, ap) {
2589 struct ata_eh_context *ehc = &link->eh_context;
2591 /* revalidate existing devices and attach new ones */
2592 rc = ata_eh_revalidate_and_attach(link, &dev);
2593 if (rc)
2594 goto dev_fail;
2596 /* if PMP got attached, return, pmp EH will take care of it */
2597 if (link->device->class == ATA_DEV_PMP) {
2598 ehc->i.action = 0;
2599 return 0;
2602 /* configure transfer mode if necessary */
2603 if (ehc->i.flags & ATA_EHI_SETMODE) {
2604 rc = ata_set_mode(link, &dev);
2605 if (rc)
2606 goto dev_fail;
2607 ehc->i.flags &= ~ATA_EHI_SETMODE;
2610 /* this link is okay now */
2611 ehc->i.flags = 0;
2612 continue;
2614 dev_fail:
2615 nr_failed_devs++;
2616 if (ata_eh_handle_dev_fail(dev, rc))
2617 nr_disabled_devs++;
2619 if (ap->pflags & ATA_PFLAG_FROZEN) {
2620 /* PMP reset requires working host port.
2621 * Can't retry if it's frozen.
2623 if (ap->nr_pmp_links)
2624 goto out;
2625 break;
2629 if (nr_failed_devs) {
2630 if (nr_failed_devs != nr_disabled_devs) {
2631 ata_port_printk(ap, KERN_WARNING, "failed to recover "
2632 "some devices, retrying in 5 secs\n");
2633 ssleep(5);
2634 } else {
2635 /* no device left to recover, repeat fast */
2636 msleep(500);
2639 goto retry;
2642 out:
2643 if (rc && r_failed_link)
2644 *r_failed_link = link;
2646 DPRINTK("EXIT, rc=%d\n", rc);
2647 return rc;
2651 * ata_eh_finish - finish up EH
2652 * @ap: host port to finish EH for
2654 * Recovery is complete. Clean up EH states and retry or finish
2655 * failed qcs.
2657 * LOCKING:
2658 * None.
2660 void ata_eh_finish(struct ata_port *ap)
2662 int tag;
2664 /* retry or finish qcs */
2665 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2666 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2668 if (!(qc->flags & ATA_QCFLAG_FAILED))
2669 continue;
2671 if (qc->err_mask) {
2672 /* FIXME: Once EH migration is complete,
2673 * generate sense data in this function,
2674 * considering both err_mask and tf.
2676 if (qc->err_mask & AC_ERR_INVALID)
2677 ata_eh_qc_complete(qc);
2678 else
2679 ata_eh_qc_retry(qc);
2680 } else {
2681 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2682 ata_eh_qc_complete(qc);
2683 } else {
2684 /* feed zero TF to sense generation */
2685 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2686 ata_eh_qc_retry(qc);
2691 /* make sure nr_active_links is zero after EH */
2692 WARN_ON(ap->nr_active_links);
2693 ap->nr_active_links = 0;
2697 * ata_do_eh - do standard error handling
2698 * @ap: host port to handle error for
2699 * @prereset: prereset method (can be NULL)
2700 * @softreset: softreset method (can be NULL)
2701 * @hardreset: hardreset method (can be NULL)
2702 * @postreset: postreset method (can be NULL)
2704 * Perform standard error handling sequence.
2706 * LOCKING:
2707 * Kernel thread context (may sleep).
2709 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2710 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2711 ata_postreset_fn_t postreset)
2713 struct ata_device *dev;
2714 int rc;
2716 ata_eh_autopsy(ap);
2717 ata_eh_report(ap);
2719 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
2720 NULL);
2721 if (rc) {
2722 ata_link_for_each_dev(dev, &ap->link)
2723 ata_dev_disable(dev);
2726 ata_eh_finish(ap);
2729 #ifdef CONFIG_PM
2731 * ata_eh_handle_port_suspend - perform port suspend operation
2732 * @ap: port to suspend
2734 * Suspend @ap.
2736 * LOCKING:
2737 * Kernel thread context (may sleep).
2739 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2741 unsigned long flags;
2742 int rc = 0;
2744 /* are we suspending? */
2745 spin_lock_irqsave(ap->lock, flags);
2746 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2747 ap->pm_mesg.event == PM_EVENT_ON) {
2748 spin_unlock_irqrestore(ap->lock, flags);
2749 return;
2751 spin_unlock_irqrestore(ap->lock, flags);
2753 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2755 /* tell ACPI we're suspending */
2756 rc = ata_acpi_on_suspend(ap);
2757 if (rc)
2758 goto out;
2760 /* suspend */
2761 ata_eh_freeze_port(ap);
2763 if (ap->ops->port_suspend)
2764 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2766 out:
2767 /* report result */
2768 spin_lock_irqsave(ap->lock, flags);
2770 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2771 if (rc == 0)
2772 ap->pflags |= ATA_PFLAG_SUSPENDED;
2773 else if (ap->pflags & ATA_PFLAG_FROZEN)
2774 ata_port_schedule_eh(ap);
2776 if (ap->pm_result) {
2777 *ap->pm_result = rc;
2778 ap->pm_result = NULL;
2781 spin_unlock_irqrestore(ap->lock, flags);
2783 return;
2787 * ata_eh_handle_port_resume - perform port resume operation
2788 * @ap: port to resume
2790 * Resume @ap.
2792 * LOCKING:
2793 * Kernel thread context (may sleep).
2795 static void ata_eh_handle_port_resume(struct ata_port *ap)
2797 unsigned long flags;
2798 int rc = 0;
2800 /* are we resuming? */
2801 spin_lock_irqsave(ap->lock, flags);
2802 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2803 ap->pm_mesg.event != PM_EVENT_ON) {
2804 spin_unlock_irqrestore(ap->lock, flags);
2805 return;
2807 spin_unlock_irqrestore(ap->lock, flags);
2809 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
2811 if (ap->ops->port_resume)
2812 rc = ap->ops->port_resume(ap);
2814 /* tell ACPI that we're resuming */
2815 ata_acpi_on_resume(ap);
2817 /* report result */
2818 spin_lock_irqsave(ap->lock, flags);
2819 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2820 if (ap->pm_result) {
2821 *ap->pm_result = rc;
2822 ap->pm_result = NULL;
2824 spin_unlock_irqrestore(ap->lock, flags);
2826 #endif /* CONFIG_PM */