libata: transfer EHI control flags to slave ehc.i
[linux-2.6/mini2440.git] / drivers / ata / libata-eh.c
blobd2409a8acecec54321630b82f9d1b4d8445ad0d3
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/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include "../scsi/scsi_transport_api.h"
45 #include <linux/libata.h>
47 #include "libata.h"
49 enum {
50 /* speed down verdicts */
51 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
54 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
56 /* error flags */
57 ATA_EFLAG_IS_IO = (1 << 0),
58 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
60 /* error categories */
61 ATA_ECAT_NONE = 0,
62 ATA_ECAT_ATA_BUS = 1,
63 ATA_ECAT_TOUT_HSM = 2,
64 ATA_ECAT_UNK_DEV = 3,
65 ATA_ECAT_DUBIOUS_NONE = 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
69 ATA_ECAT_NR = 8,
71 ATA_EH_CMD_DFL_TIMEOUT = 5000,
73 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN = 5000,
76 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
81 ATA_EH_PRERESET_TIMEOUT = 10000,
82 ATA_EH_FASTDRAIN_INTERVAL = 3000,
84 ATA_EH_UA_TRIES = 5,
87 /* The following table determines how we sequence resets. Each entry
88 * represents timeout for that try. The first try can be soft or
89 * hardreset. All others are hardreset if available. In most cases
90 * the first reset w/ 10sec timeout should succeed. Following entries
91 * are mostly for error handling, hotplug and retarded devices.
93 static const unsigned long ata_eh_reset_timeouts[] = {
94 10000, /* most drives spin up by 10sec */
95 10000, /* > 99% working drives spin up before 20sec */
96 35000, /* give > 30 secs of idleness for retarded devices */
97 5000, /* and sweet one last chance */
98 ULONG_MAX, /* > 1 min has elapsed, give up */
101 static const unsigned long ata_eh_identify_timeouts[] = {
102 5000, /* covers > 99% of successes and not too boring on failures */
103 10000, /* combined time till here is enough even for media access */
104 30000, /* for true idiots */
105 ULONG_MAX,
108 static const unsigned long ata_eh_other_timeouts[] = {
109 5000, /* same rationale as identify timeout */
110 10000, /* ditto */
111 /* but no merciful 30sec for other commands, it just isn't worth it */
112 ULONG_MAX,
115 struct ata_eh_cmd_timeout_ent {
116 const u8 *commands;
117 const unsigned long *timeouts;
120 /* The following table determines timeouts to use for EH internal
121 * commands. Each table entry is a command class and matches the
122 * commands the entry applies to and the timeout table to use.
124 * On the retry after a command timed out, the next timeout value from
125 * the table is used. If the table doesn't contain further entries,
126 * the last value is used.
128 * ehc->cmd_timeout_idx keeps track of which timeout to use per
129 * command class, so if SET_FEATURES times out on the first try, the
130 * next try will use the second timeout value only for that class.
132 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
133 static const struct ata_eh_cmd_timeout_ent
134 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
135 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
136 .timeouts = ata_eh_identify_timeouts, },
137 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
138 .timeouts = ata_eh_other_timeouts, },
139 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
140 .timeouts = ata_eh_other_timeouts, },
141 { .commands = CMDS(ATA_CMD_SET_FEATURES),
142 .timeouts = ata_eh_other_timeouts, },
143 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
144 .timeouts = ata_eh_other_timeouts, },
146 #undef CMDS
148 static void __ata_port_freeze(struct ata_port *ap);
149 #ifdef CONFIG_PM
150 static void ata_eh_handle_port_suspend(struct ata_port *ap);
151 static void ata_eh_handle_port_resume(struct ata_port *ap);
152 #else /* CONFIG_PM */
153 static void ata_eh_handle_port_suspend(struct ata_port *ap)
156 static void ata_eh_handle_port_resume(struct ata_port *ap)
158 #endif /* CONFIG_PM */
160 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
161 va_list args)
163 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
164 ATA_EH_DESC_LEN - ehi->desc_len,
165 fmt, args);
169 * __ata_ehi_push_desc - push error description without adding separator
170 * @ehi: target EHI
171 * @fmt: printf format string
173 * Format string according to @fmt and append it to @ehi->desc.
175 * LOCKING:
176 * spin_lock_irqsave(host lock)
178 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
180 va_list args;
182 va_start(args, fmt);
183 __ata_ehi_pushv_desc(ehi, fmt, args);
184 va_end(args);
188 * ata_ehi_push_desc - push error description with separator
189 * @ehi: target EHI
190 * @fmt: printf format string
192 * Format string according to @fmt and append it to @ehi->desc.
193 * If @ehi->desc is not empty, ", " is added in-between.
195 * LOCKING:
196 * spin_lock_irqsave(host lock)
198 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
200 va_list args;
202 if (ehi->desc_len)
203 __ata_ehi_push_desc(ehi, ", ");
205 va_start(args, fmt);
206 __ata_ehi_pushv_desc(ehi, fmt, args);
207 va_end(args);
211 * ata_ehi_clear_desc - clean error description
212 * @ehi: target EHI
214 * Clear @ehi->desc.
216 * LOCKING:
217 * spin_lock_irqsave(host lock)
219 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
221 ehi->desc[0] = '\0';
222 ehi->desc_len = 0;
226 * ata_port_desc - append port description
227 * @ap: target ATA port
228 * @fmt: printf format string
230 * Format string according to @fmt and append it to port
231 * description. If port description is not empty, " " is added
232 * in-between. This function is to be used while initializing
233 * ata_host. The description is printed on host registration.
235 * LOCKING:
236 * None.
238 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
240 va_list args;
242 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
244 if (ap->link.eh_info.desc_len)
245 __ata_ehi_push_desc(&ap->link.eh_info, " ");
247 va_start(args, fmt);
248 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
249 va_end(args);
252 #ifdef CONFIG_PCI
255 * ata_port_pbar_desc - append PCI BAR description
256 * @ap: target ATA port
257 * @bar: target PCI BAR
258 * @offset: offset into PCI BAR
259 * @name: name of the area
261 * If @offset is negative, this function formats a string which
262 * contains the name, address, size and type of the BAR and
263 * appends it to the port description. If @offset is zero or
264 * positive, only name and offsetted address is appended.
266 * LOCKING:
267 * None.
269 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
270 const char *name)
272 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
273 char *type = "";
274 unsigned long long start, len;
276 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
277 type = "m";
278 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
279 type = "i";
281 start = (unsigned long long)pci_resource_start(pdev, bar);
282 len = (unsigned long long)pci_resource_len(pdev, bar);
284 if (offset < 0)
285 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
286 else
287 ata_port_desc(ap, "%s 0x%llx", name,
288 start + (unsigned long long)offset);
291 #endif /* CONFIG_PCI */
293 static int ata_lookup_timeout_table(u8 cmd)
295 int i;
297 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
298 const u8 *cur;
300 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
301 if (*cur == cmd)
302 return i;
305 return -1;
309 * ata_internal_cmd_timeout - determine timeout for an internal command
310 * @dev: target device
311 * @cmd: internal command to be issued
313 * Determine timeout for internal command @cmd for @dev.
315 * LOCKING:
316 * EH context.
318 * RETURNS:
319 * Determined timeout.
321 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
323 struct ata_eh_context *ehc = &dev->link->eh_context;
324 int ent = ata_lookup_timeout_table(cmd);
325 int idx;
327 if (ent < 0)
328 return ATA_EH_CMD_DFL_TIMEOUT;
330 idx = ehc->cmd_timeout_idx[dev->devno][ent];
331 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
335 * ata_internal_cmd_timed_out - notification for internal command timeout
336 * @dev: target device
337 * @cmd: internal command which timed out
339 * Notify EH that internal command @cmd for @dev timed out. This
340 * function should be called only for commands whose timeouts are
341 * determined using ata_internal_cmd_timeout().
343 * LOCKING:
344 * EH context.
346 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
348 struct ata_eh_context *ehc = &dev->link->eh_context;
349 int ent = ata_lookup_timeout_table(cmd);
350 int idx;
352 if (ent < 0)
353 return;
355 idx = ehc->cmd_timeout_idx[dev->devno][ent];
356 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
357 ehc->cmd_timeout_idx[dev->devno][ent]++;
360 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
361 unsigned int err_mask)
363 struct ata_ering_entry *ent;
365 WARN_ON(!err_mask);
367 ering->cursor++;
368 ering->cursor %= ATA_ERING_SIZE;
370 ent = &ering->ring[ering->cursor];
371 ent->eflags = eflags;
372 ent->err_mask = err_mask;
373 ent->timestamp = get_jiffies_64();
376 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
378 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
380 if (ent->err_mask)
381 return ent;
382 return NULL;
385 static void ata_ering_clear(struct ata_ering *ering)
387 memset(ering, 0, sizeof(*ering));
390 static int ata_ering_map(struct ata_ering *ering,
391 int (*map_fn)(struct ata_ering_entry *, void *),
392 void *arg)
394 int idx, rc = 0;
395 struct ata_ering_entry *ent;
397 idx = ering->cursor;
398 do {
399 ent = &ering->ring[idx];
400 if (!ent->err_mask)
401 break;
402 rc = map_fn(ent, arg);
403 if (rc)
404 break;
405 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
406 } while (idx != ering->cursor);
408 return rc;
411 static unsigned int ata_eh_dev_action(struct ata_device *dev)
413 struct ata_eh_context *ehc = &dev->link->eh_context;
415 return ehc->i.action | ehc->i.dev_action[dev->devno];
418 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
419 struct ata_eh_info *ehi, unsigned int action)
421 struct ata_device *tdev;
423 if (!dev) {
424 ehi->action &= ~action;
425 ata_link_for_each_dev(tdev, link)
426 ehi->dev_action[tdev->devno] &= ~action;
427 } else {
428 /* doesn't make sense for port-wide EH actions */
429 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
431 /* break ehi->action into ehi->dev_action */
432 if (ehi->action & action) {
433 ata_link_for_each_dev(tdev, link)
434 ehi->dev_action[tdev->devno] |=
435 ehi->action & action;
436 ehi->action &= ~action;
439 /* turn off the specified per-dev action */
440 ehi->dev_action[dev->devno] &= ~action;
445 * ata_scsi_timed_out - SCSI layer time out callback
446 * @cmd: timed out SCSI command
448 * Handles SCSI layer timeout. We race with normal completion of
449 * the qc for @cmd. If the qc is already gone, we lose and let
450 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
451 * timed out and EH should be invoked. Prevent ata_qc_complete()
452 * from finishing it by setting EH_SCHEDULED and return
453 * EH_NOT_HANDLED.
455 * TODO: kill this function once old EH is gone.
457 * LOCKING:
458 * Called from timer context
460 * RETURNS:
461 * EH_HANDLED or EH_NOT_HANDLED
463 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
465 struct Scsi_Host *host = cmd->device->host;
466 struct ata_port *ap = ata_shost_to_port(host);
467 unsigned long flags;
468 struct ata_queued_cmd *qc;
469 enum blk_eh_timer_return ret;
471 DPRINTK("ENTER\n");
473 if (ap->ops->error_handler) {
474 ret = BLK_EH_NOT_HANDLED;
475 goto out;
478 ret = BLK_EH_HANDLED;
479 spin_lock_irqsave(ap->lock, flags);
480 qc = ata_qc_from_tag(ap, ap->link.active_tag);
481 if (qc) {
482 WARN_ON(qc->scsicmd != cmd);
483 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
484 qc->err_mask |= AC_ERR_TIMEOUT;
485 ret = BLK_EH_NOT_HANDLED;
487 spin_unlock_irqrestore(ap->lock, flags);
489 out:
490 DPRINTK("EXIT, ret=%d\n", ret);
491 return ret;
495 * ata_scsi_error - SCSI layer error handler callback
496 * @host: SCSI host on which error occurred
498 * Handles SCSI-layer-thrown error events.
500 * LOCKING:
501 * Inherited from SCSI layer (none, can sleep)
503 * RETURNS:
504 * Zero.
506 void ata_scsi_error(struct Scsi_Host *host)
508 struct ata_port *ap = ata_shost_to_port(host);
509 int i;
510 unsigned long flags;
512 DPRINTK("ENTER\n");
514 /* synchronize with port task */
515 ata_port_flush_task(ap);
517 /* synchronize with host lock and sort out timeouts */
519 /* For new EH, all qcs are finished in one of three ways -
520 * normal completion, error completion, and SCSI timeout.
521 * Both cmpletions can race against SCSI timeout. When normal
522 * completion wins, the qc never reaches EH. When error
523 * completion wins, the qc has ATA_QCFLAG_FAILED set.
525 * When SCSI timeout wins, things are a bit more complex.
526 * Normal or error completion can occur after the timeout but
527 * before this point. In such cases, both types of
528 * completions are honored. A scmd is determined to have
529 * timed out iff its associated qc is active and not failed.
531 if (ap->ops->error_handler) {
532 struct scsi_cmnd *scmd, *tmp;
533 int nr_timedout = 0;
535 spin_lock_irqsave(ap->lock, flags);
537 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
538 struct ata_queued_cmd *qc;
540 for (i = 0; i < ATA_MAX_QUEUE; i++) {
541 qc = __ata_qc_from_tag(ap, i);
542 if (qc->flags & ATA_QCFLAG_ACTIVE &&
543 qc->scsicmd == scmd)
544 break;
547 if (i < ATA_MAX_QUEUE) {
548 /* the scmd has an associated qc */
549 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
550 /* which hasn't failed yet, timeout */
551 qc->err_mask |= AC_ERR_TIMEOUT;
552 qc->flags |= ATA_QCFLAG_FAILED;
553 nr_timedout++;
555 } else {
556 /* Normal completion occurred after
557 * SCSI timeout but before this point.
558 * Successfully complete it.
560 scmd->retries = scmd->allowed;
561 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
565 /* If we have timed out qcs. They belong to EH from
566 * this point but the state of the controller is
567 * unknown. Freeze the port to make sure the IRQ
568 * handler doesn't diddle with those qcs. This must
569 * be done atomically w.r.t. setting QCFLAG_FAILED.
571 if (nr_timedout)
572 __ata_port_freeze(ap);
574 spin_unlock_irqrestore(ap->lock, flags);
576 /* initialize eh_tries */
577 ap->eh_tries = ATA_EH_MAX_TRIES;
578 } else
579 spin_unlock_wait(ap->lock);
581 repeat:
582 /* invoke error handler */
583 if (ap->ops->error_handler) {
584 struct ata_link *link;
586 /* kill fast drain timer */
587 del_timer_sync(&ap->fastdrain_timer);
589 /* process port resume request */
590 ata_eh_handle_port_resume(ap);
592 /* fetch & clear EH info */
593 spin_lock_irqsave(ap->lock, flags);
595 __ata_port_for_each_link(link, ap) {
596 struct ata_eh_context *ehc = &link->eh_context;
597 struct ata_device *dev;
599 memset(&link->eh_context, 0, sizeof(link->eh_context));
600 link->eh_context.i = link->eh_info;
601 memset(&link->eh_info, 0, sizeof(link->eh_info));
603 ata_link_for_each_dev(dev, link) {
604 int devno = dev->devno;
606 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
607 if (ata_ncq_enabled(dev))
608 ehc->saved_ncq_enabled |= 1 << devno;
611 /* set last reset timestamp to some time in the past */
612 ehc->last_reset = jiffies - 60 * HZ;
615 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
616 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
617 ap->excl_link = NULL; /* don't maintain exclusion over EH */
619 spin_unlock_irqrestore(ap->lock, flags);
621 /* invoke EH, skip if unloading or suspended */
622 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
623 ap->ops->error_handler(ap);
624 else
625 ata_eh_finish(ap);
627 /* process port suspend request */
628 ata_eh_handle_port_suspend(ap);
630 /* Exception might have happend after ->error_handler
631 * recovered the port but before this point. Repeat
632 * EH in such case.
634 spin_lock_irqsave(ap->lock, flags);
636 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
637 if (--ap->eh_tries) {
638 spin_unlock_irqrestore(ap->lock, flags);
639 goto repeat;
641 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
642 "tries, giving up\n", ATA_EH_MAX_TRIES);
643 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
646 /* this run is complete, make sure EH info is clear */
647 __ata_port_for_each_link(link, ap)
648 memset(&link->eh_info, 0, sizeof(link->eh_info));
650 /* Clear host_eh_scheduled while holding ap->lock such
651 * that if exception occurs after this point but
652 * before EH completion, SCSI midlayer will
653 * re-initiate EH.
655 host->host_eh_scheduled = 0;
657 spin_unlock_irqrestore(ap->lock, flags);
658 } else {
659 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
660 ap->ops->eng_timeout(ap);
663 /* finish or retry handled scmd's and clean up */
664 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
666 scsi_eh_flush_done_q(&ap->eh_done_q);
668 /* clean up */
669 spin_lock_irqsave(ap->lock, flags);
671 if (ap->pflags & ATA_PFLAG_LOADING)
672 ap->pflags &= ~ATA_PFLAG_LOADING;
673 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
674 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
676 if (ap->pflags & ATA_PFLAG_RECOVERED)
677 ata_port_printk(ap, KERN_INFO, "EH complete\n");
679 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
681 /* tell wait_eh that we're done */
682 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
683 wake_up_all(&ap->eh_wait_q);
685 spin_unlock_irqrestore(ap->lock, flags);
687 DPRINTK("EXIT\n");
691 * ata_port_wait_eh - Wait for the currently pending EH to complete
692 * @ap: Port to wait EH for
694 * Wait until the currently pending EH is complete.
696 * LOCKING:
697 * Kernel thread context (may sleep).
699 void ata_port_wait_eh(struct ata_port *ap)
701 unsigned long flags;
702 DEFINE_WAIT(wait);
704 retry:
705 spin_lock_irqsave(ap->lock, flags);
707 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
708 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
709 spin_unlock_irqrestore(ap->lock, flags);
710 schedule();
711 spin_lock_irqsave(ap->lock, flags);
713 finish_wait(&ap->eh_wait_q, &wait);
715 spin_unlock_irqrestore(ap->lock, flags);
717 /* make sure SCSI EH is complete */
718 if (scsi_host_in_recovery(ap->scsi_host)) {
719 msleep(10);
720 goto retry;
724 static int ata_eh_nr_in_flight(struct ata_port *ap)
726 unsigned int tag;
727 int nr = 0;
729 /* count only non-internal commands */
730 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
731 if (ata_qc_from_tag(ap, tag))
732 nr++;
734 return nr;
737 void ata_eh_fastdrain_timerfn(unsigned long arg)
739 struct ata_port *ap = (void *)arg;
740 unsigned long flags;
741 int cnt;
743 spin_lock_irqsave(ap->lock, flags);
745 cnt = ata_eh_nr_in_flight(ap);
747 /* are we done? */
748 if (!cnt)
749 goto out_unlock;
751 if (cnt == ap->fastdrain_cnt) {
752 unsigned int tag;
754 /* No progress during the last interval, tag all
755 * in-flight qcs as timed out and freeze the port.
757 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
758 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
759 if (qc)
760 qc->err_mask |= AC_ERR_TIMEOUT;
763 ata_port_freeze(ap);
764 } else {
765 /* some qcs have finished, give it another chance */
766 ap->fastdrain_cnt = cnt;
767 ap->fastdrain_timer.expires =
768 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
769 add_timer(&ap->fastdrain_timer);
772 out_unlock:
773 spin_unlock_irqrestore(ap->lock, flags);
777 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
778 * @ap: target ATA port
779 * @fastdrain: activate fast drain
781 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
782 * is non-zero and EH wasn't pending before. Fast drain ensures
783 * that EH kicks in in timely manner.
785 * LOCKING:
786 * spin_lock_irqsave(host lock)
788 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
790 int cnt;
792 /* already scheduled? */
793 if (ap->pflags & ATA_PFLAG_EH_PENDING)
794 return;
796 ap->pflags |= ATA_PFLAG_EH_PENDING;
798 if (!fastdrain)
799 return;
801 /* do we have in-flight qcs? */
802 cnt = ata_eh_nr_in_flight(ap);
803 if (!cnt)
804 return;
806 /* activate fast drain */
807 ap->fastdrain_cnt = cnt;
808 ap->fastdrain_timer.expires =
809 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
810 add_timer(&ap->fastdrain_timer);
814 * ata_qc_schedule_eh - schedule qc for error handling
815 * @qc: command to schedule error handling for
817 * Schedule error handling for @qc. EH will kick in as soon as
818 * other commands are drained.
820 * LOCKING:
821 * spin_lock_irqsave(host lock)
823 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
825 struct ata_port *ap = qc->ap;
827 WARN_ON(!ap->ops->error_handler);
829 qc->flags |= ATA_QCFLAG_FAILED;
830 ata_eh_set_pending(ap, 1);
832 /* The following will fail if timeout has already expired.
833 * ata_scsi_error() takes care of such scmds on EH entry.
834 * Note that ATA_QCFLAG_FAILED is unconditionally set after
835 * this function completes.
837 blk_abort_request(qc->scsicmd->request);
841 * ata_port_schedule_eh - schedule error handling without a qc
842 * @ap: ATA port to schedule EH for
844 * Schedule error handling for @ap. EH will kick in as soon as
845 * all commands are drained.
847 * LOCKING:
848 * spin_lock_irqsave(host lock)
850 void ata_port_schedule_eh(struct ata_port *ap)
852 WARN_ON(!ap->ops->error_handler);
854 if (ap->pflags & ATA_PFLAG_INITIALIZING)
855 return;
857 ata_eh_set_pending(ap, 1);
858 scsi_schedule_eh(ap->scsi_host);
860 DPRINTK("port EH scheduled\n");
863 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
865 int tag, nr_aborted = 0;
867 WARN_ON(!ap->ops->error_handler);
869 /* we're gonna abort all commands, no need for fast drain */
870 ata_eh_set_pending(ap, 0);
872 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
873 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
875 if (qc && (!link || qc->dev->link == link)) {
876 qc->flags |= ATA_QCFLAG_FAILED;
877 ata_qc_complete(qc);
878 nr_aborted++;
882 if (!nr_aborted)
883 ata_port_schedule_eh(ap);
885 return nr_aborted;
889 * ata_link_abort - abort all qc's on the link
890 * @link: ATA link to abort qc's for
892 * Abort all active qc's active on @link and schedule EH.
894 * LOCKING:
895 * spin_lock_irqsave(host lock)
897 * RETURNS:
898 * Number of aborted qc's.
900 int ata_link_abort(struct ata_link *link)
902 return ata_do_link_abort(link->ap, link);
906 * ata_port_abort - abort all qc's on the port
907 * @ap: ATA port to abort qc's for
909 * Abort all active qc's of @ap and schedule EH.
911 * LOCKING:
912 * spin_lock_irqsave(host_set lock)
914 * RETURNS:
915 * Number of aborted qc's.
917 int ata_port_abort(struct ata_port *ap)
919 return ata_do_link_abort(ap, NULL);
923 * __ata_port_freeze - freeze port
924 * @ap: ATA port to freeze
926 * This function is called when HSM violation or some other
927 * condition disrupts normal operation of the port. Frozen port
928 * is not allowed to perform any operation until the port is
929 * thawed, which usually follows a successful reset.
931 * ap->ops->freeze() callback can be used for freezing the port
932 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
933 * port cannot be frozen hardware-wise, the interrupt handler
934 * must ack and clear interrupts unconditionally while the port
935 * is frozen.
937 * LOCKING:
938 * spin_lock_irqsave(host lock)
940 static void __ata_port_freeze(struct ata_port *ap)
942 WARN_ON(!ap->ops->error_handler);
944 if (ap->ops->freeze)
945 ap->ops->freeze(ap);
947 ap->pflags |= ATA_PFLAG_FROZEN;
949 DPRINTK("ata%u port frozen\n", ap->print_id);
953 * ata_port_freeze - abort & freeze port
954 * @ap: ATA port to freeze
956 * Abort and freeze @ap.
958 * LOCKING:
959 * spin_lock_irqsave(host lock)
961 * RETURNS:
962 * Number of aborted commands.
964 int ata_port_freeze(struct ata_port *ap)
966 int nr_aborted;
968 WARN_ON(!ap->ops->error_handler);
970 nr_aborted = ata_port_abort(ap);
971 __ata_port_freeze(ap);
973 return nr_aborted;
977 * sata_async_notification - SATA async notification handler
978 * @ap: ATA port where async notification is received
980 * Handler to be called when async notification via SDB FIS is
981 * received. This function schedules EH if necessary.
983 * LOCKING:
984 * spin_lock_irqsave(host lock)
986 * RETURNS:
987 * 1 if EH is scheduled, 0 otherwise.
989 int sata_async_notification(struct ata_port *ap)
991 u32 sntf;
992 int rc;
994 if (!(ap->flags & ATA_FLAG_AN))
995 return 0;
997 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
998 if (rc == 0)
999 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1001 if (!sata_pmp_attached(ap) || rc) {
1002 /* PMP is not attached or SNTF is not available */
1003 if (!sata_pmp_attached(ap)) {
1004 /* PMP is not attached. Check whether ATAPI
1005 * AN is configured. If so, notify media
1006 * change.
1008 struct ata_device *dev = ap->link.device;
1010 if ((dev->class == ATA_DEV_ATAPI) &&
1011 (dev->flags & ATA_DFLAG_AN))
1012 ata_scsi_media_change_notify(dev);
1013 return 0;
1014 } else {
1015 /* PMP is attached but SNTF is not available.
1016 * ATAPI async media change notification is
1017 * not used. The PMP must be reporting PHY
1018 * status change, schedule EH.
1020 ata_port_schedule_eh(ap);
1021 return 1;
1023 } else {
1024 /* PMP is attached and SNTF is available */
1025 struct ata_link *link;
1027 /* check and notify ATAPI AN */
1028 ata_port_for_each_link(link, ap) {
1029 if (!(sntf & (1 << link->pmp)))
1030 continue;
1032 if ((link->device->class == ATA_DEV_ATAPI) &&
1033 (link->device->flags & ATA_DFLAG_AN))
1034 ata_scsi_media_change_notify(link->device);
1037 /* If PMP is reporting that PHY status of some
1038 * downstream ports has changed, schedule EH.
1040 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1041 ata_port_schedule_eh(ap);
1042 return 1;
1045 return 0;
1050 * ata_eh_freeze_port - EH helper to freeze port
1051 * @ap: ATA port to freeze
1053 * Freeze @ap.
1055 * LOCKING:
1056 * None.
1058 void ata_eh_freeze_port(struct ata_port *ap)
1060 unsigned long flags;
1062 if (!ap->ops->error_handler)
1063 return;
1065 spin_lock_irqsave(ap->lock, flags);
1066 __ata_port_freeze(ap);
1067 spin_unlock_irqrestore(ap->lock, flags);
1071 * ata_port_thaw_port - EH helper to thaw port
1072 * @ap: ATA port to thaw
1074 * Thaw frozen port @ap.
1076 * LOCKING:
1077 * None.
1079 void ata_eh_thaw_port(struct ata_port *ap)
1081 unsigned long flags;
1083 if (!ap->ops->error_handler)
1084 return;
1086 spin_lock_irqsave(ap->lock, flags);
1088 ap->pflags &= ~ATA_PFLAG_FROZEN;
1090 if (ap->ops->thaw)
1091 ap->ops->thaw(ap);
1093 spin_unlock_irqrestore(ap->lock, flags);
1095 DPRINTK("ata%u port thawed\n", ap->print_id);
1098 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1100 /* nada */
1103 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1105 struct ata_port *ap = qc->ap;
1106 struct scsi_cmnd *scmd = qc->scsicmd;
1107 unsigned long flags;
1109 spin_lock_irqsave(ap->lock, flags);
1110 qc->scsidone = ata_eh_scsidone;
1111 __ata_qc_complete(qc);
1112 WARN_ON(ata_tag_valid(qc->tag));
1113 spin_unlock_irqrestore(ap->lock, flags);
1115 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1119 * ata_eh_qc_complete - Complete an active ATA command from EH
1120 * @qc: Command to complete
1122 * Indicate to the mid and upper layers that an ATA command has
1123 * completed. To be used from EH.
1125 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1127 struct scsi_cmnd *scmd = qc->scsicmd;
1128 scmd->retries = scmd->allowed;
1129 __ata_eh_qc_complete(qc);
1133 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1134 * @qc: Command to retry
1136 * Indicate to the mid and upper layers that an ATA command
1137 * should be retried. To be used from EH.
1139 * SCSI midlayer limits the number of retries to scmd->allowed.
1140 * scmd->retries is decremented for commands which get retried
1141 * due to unrelated failures (qc->err_mask is zero).
1143 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1145 struct scsi_cmnd *scmd = qc->scsicmd;
1146 if (!qc->err_mask && scmd->retries)
1147 scmd->retries--;
1148 __ata_eh_qc_complete(qc);
1152 * ata_eh_detach_dev - detach ATA device
1153 * @dev: ATA device to detach
1155 * Detach @dev.
1157 * LOCKING:
1158 * None.
1160 void ata_eh_detach_dev(struct ata_device *dev)
1162 struct ata_link *link = dev->link;
1163 struct ata_port *ap = link->ap;
1164 unsigned long flags;
1166 ata_dev_disable(dev);
1168 spin_lock_irqsave(ap->lock, flags);
1170 dev->flags &= ~ATA_DFLAG_DETACH;
1172 if (ata_scsi_offline_dev(dev)) {
1173 dev->flags |= ATA_DFLAG_DETACHED;
1174 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1177 /* clear per-dev EH actions */
1178 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1179 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1181 spin_unlock_irqrestore(ap->lock, flags);
1185 * ata_eh_about_to_do - about to perform eh_action
1186 * @link: target ATA link
1187 * @dev: target ATA dev for per-dev action (can be NULL)
1188 * @action: action about to be performed
1190 * Called just before performing EH actions to clear related bits
1191 * in @link->eh_info such that eh actions are not unnecessarily
1192 * repeated.
1194 * LOCKING:
1195 * None.
1197 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1198 unsigned int action)
1200 struct ata_port *ap = link->ap;
1201 struct ata_eh_info *ehi = &link->eh_info;
1202 struct ata_eh_context *ehc = &link->eh_context;
1203 unsigned long flags;
1205 spin_lock_irqsave(ap->lock, flags);
1207 ata_eh_clear_action(link, dev, ehi, action);
1209 if (!(ehc->i.flags & ATA_EHI_QUIET))
1210 ap->pflags |= ATA_PFLAG_RECOVERED;
1212 spin_unlock_irqrestore(ap->lock, flags);
1216 * ata_eh_done - EH action complete
1217 * @ap: target ATA port
1218 * @dev: target ATA dev for per-dev action (can be NULL)
1219 * @action: action just completed
1221 * Called right after performing EH actions to clear related bits
1222 * in @link->eh_context.
1224 * LOCKING:
1225 * None.
1227 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1228 unsigned int action)
1230 struct ata_eh_context *ehc = &link->eh_context;
1232 ata_eh_clear_action(link, dev, &ehc->i, action);
1236 * ata_err_string - convert err_mask to descriptive string
1237 * @err_mask: error mask to convert to string
1239 * Convert @err_mask to descriptive string. Errors are
1240 * prioritized according to severity and only the most severe
1241 * error is reported.
1243 * LOCKING:
1244 * None.
1246 * RETURNS:
1247 * Descriptive string for @err_mask
1249 static const char *ata_err_string(unsigned int err_mask)
1251 if (err_mask & AC_ERR_HOST_BUS)
1252 return "host bus error";
1253 if (err_mask & AC_ERR_ATA_BUS)
1254 return "ATA bus error";
1255 if (err_mask & AC_ERR_TIMEOUT)
1256 return "timeout";
1257 if (err_mask & AC_ERR_HSM)
1258 return "HSM violation";
1259 if (err_mask & AC_ERR_SYSTEM)
1260 return "internal error";
1261 if (err_mask & AC_ERR_MEDIA)
1262 return "media error";
1263 if (err_mask & AC_ERR_INVALID)
1264 return "invalid argument";
1265 if (err_mask & AC_ERR_DEV)
1266 return "device error";
1267 return "unknown error";
1271 * ata_read_log_page - read a specific log page
1272 * @dev: target device
1273 * @page: page to read
1274 * @buf: buffer to store read page
1275 * @sectors: number of sectors to read
1277 * Read log page using READ_LOG_EXT command.
1279 * LOCKING:
1280 * Kernel thread context (may sleep).
1282 * RETURNS:
1283 * 0 on success, AC_ERR_* mask otherwise.
1285 static unsigned int ata_read_log_page(struct ata_device *dev,
1286 u8 page, void *buf, unsigned int sectors)
1288 struct ata_taskfile tf;
1289 unsigned int err_mask;
1291 DPRINTK("read log page - page %d\n", page);
1293 ata_tf_init(dev, &tf);
1294 tf.command = ATA_CMD_READ_LOG_EXT;
1295 tf.lbal = page;
1296 tf.nsect = sectors;
1297 tf.hob_nsect = sectors >> 8;
1298 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1299 tf.protocol = ATA_PROT_PIO;
1301 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1302 buf, sectors * ATA_SECT_SIZE, 0);
1304 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1305 return err_mask;
1309 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1310 * @dev: Device to read log page 10h from
1311 * @tag: Resulting tag of the failed command
1312 * @tf: Resulting taskfile registers of the failed command
1314 * Read log page 10h to obtain NCQ error details and clear error
1315 * condition.
1317 * LOCKING:
1318 * Kernel thread context (may sleep).
1320 * RETURNS:
1321 * 0 on success, -errno otherwise.
1323 static int ata_eh_read_log_10h(struct ata_device *dev,
1324 int *tag, struct ata_taskfile *tf)
1326 u8 *buf = dev->link->ap->sector_buf;
1327 unsigned int err_mask;
1328 u8 csum;
1329 int i;
1331 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1332 if (err_mask)
1333 return -EIO;
1335 csum = 0;
1336 for (i = 0; i < ATA_SECT_SIZE; i++)
1337 csum += buf[i];
1338 if (csum)
1339 ata_dev_printk(dev, KERN_WARNING,
1340 "invalid checksum 0x%x on log page 10h\n", csum);
1342 if (buf[0] & 0x80)
1343 return -ENOENT;
1345 *tag = buf[0] & 0x1f;
1347 tf->command = buf[2];
1348 tf->feature = buf[3];
1349 tf->lbal = buf[4];
1350 tf->lbam = buf[5];
1351 tf->lbah = buf[6];
1352 tf->device = buf[7];
1353 tf->hob_lbal = buf[8];
1354 tf->hob_lbam = buf[9];
1355 tf->hob_lbah = buf[10];
1356 tf->nsect = buf[12];
1357 tf->hob_nsect = buf[13];
1359 return 0;
1363 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1364 * @dev: target ATAPI device
1365 * @r_sense_key: out parameter for sense_key
1367 * Perform ATAPI TEST_UNIT_READY.
1369 * LOCKING:
1370 * EH context (may sleep).
1372 * RETURNS:
1373 * 0 on success, AC_ERR_* mask on failure.
1375 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1377 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1378 struct ata_taskfile tf;
1379 unsigned int err_mask;
1381 ata_tf_init(dev, &tf);
1383 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1384 tf.command = ATA_CMD_PACKET;
1385 tf.protocol = ATAPI_PROT_NODATA;
1387 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1388 if (err_mask == AC_ERR_DEV)
1389 *r_sense_key = tf.feature >> 4;
1390 return err_mask;
1394 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1395 * @dev: device to perform REQUEST_SENSE to
1396 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1397 * @dfl_sense_key: default sense key to use
1399 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1400 * SENSE. This function is EH helper.
1402 * LOCKING:
1403 * Kernel thread context (may sleep).
1405 * RETURNS:
1406 * 0 on success, AC_ERR_* mask on failure
1408 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1409 u8 *sense_buf, u8 dfl_sense_key)
1411 u8 cdb[ATAPI_CDB_LEN] =
1412 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1413 struct ata_port *ap = dev->link->ap;
1414 struct ata_taskfile tf;
1416 DPRINTK("ATAPI request sense\n");
1418 /* FIXME: is this needed? */
1419 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1421 /* initialize sense_buf with the error register,
1422 * for the case where they are -not- overwritten
1424 sense_buf[0] = 0x70;
1425 sense_buf[2] = dfl_sense_key;
1427 /* some devices time out if garbage left in tf */
1428 ata_tf_init(dev, &tf);
1430 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1431 tf.command = ATA_CMD_PACKET;
1433 /* is it pointless to prefer PIO for "safety reasons"? */
1434 if (ap->flags & ATA_FLAG_PIO_DMA) {
1435 tf.protocol = ATAPI_PROT_DMA;
1436 tf.feature |= ATAPI_PKT_DMA;
1437 } else {
1438 tf.protocol = ATAPI_PROT_PIO;
1439 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1440 tf.lbah = 0;
1443 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1444 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1448 * ata_eh_analyze_serror - analyze SError for a failed port
1449 * @link: ATA link to analyze SError for
1451 * Analyze SError if available and further determine cause of
1452 * failure.
1454 * LOCKING:
1455 * None.
1457 static void ata_eh_analyze_serror(struct ata_link *link)
1459 struct ata_eh_context *ehc = &link->eh_context;
1460 u32 serror = ehc->i.serror;
1461 unsigned int err_mask = 0, action = 0;
1462 u32 hotplug_mask;
1464 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1465 err_mask |= AC_ERR_ATA_BUS;
1466 action |= ATA_EH_RESET;
1468 if (serror & SERR_PROTOCOL) {
1469 err_mask |= AC_ERR_HSM;
1470 action |= ATA_EH_RESET;
1472 if (serror & SERR_INTERNAL) {
1473 err_mask |= AC_ERR_SYSTEM;
1474 action |= ATA_EH_RESET;
1477 /* Determine whether a hotplug event has occurred. Both
1478 * SError.N/X are considered hotplug events for enabled or
1479 * host links. For disabled PMP links, only N bit is
1480 * considered as X bit is left at 1 for link plugging.
1482 hotplug_mask = 0;
1484 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1485 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1486 else
1487 hotplug_mask = SERR_PHYRDY_CHG;
1489 if (serror & hotplug_mask)
1490 ata_ehi_hotplugged(&ehc->i);
1492 ehc->i.err_mask |= err_mask;
1493 ehc->i.action |= action;
1497 * ata_eh_analyze_ncq_error - analyze NCQ error
1498 * @link: ATA link to analyze NCQ error for
1500 * Read log page 10h, determine the offending qc and acquire
1501 * error status TF. For NCQ device errors, all LLDDs have to do
1502 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1503 * care of the rest.
1505 * LOCKING:
1506 * Kernel thread context (may sleep).
1508 void ata_eh_analyze_ncq_error(struct ata_link *link)
1510 struct ata_port *ap = link->ap;
1511 struct ata_eh_context *ehc = &link->eh_context;
1512 struct ata_device *dev = link->device;
1513 struct ata_queued_cmd *qc;
1514 struct ata_taskfile tf;
1515 int tag, rc;
1517 /* if frozen, we can't do much */
1518 if (ap->pflags & ATA_PFLAG_FROZEN)
1519 return;
1521 /* is it NCQ device error? */
1522 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1523 return;
1525 /* has LLDD analyzed already? */
1526 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1527 qc = __ata_qc_from_tag(ap, tag);
1529 if (!(qc->flags & ATA_QCFLAG_FAILED))
1530 continue;
1532 if (qc->err_mask)
1533 return;
1536 /* okay, this error is ours */
1537 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1538 if (rc) {
1539 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1540 "(errno=%d)\n", rc);
1541 return;
1544 if (!(link->sactive & (1 << tag))) {
1545 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1546 "inactive tag %d\n", tag);
1547 return;
1550 /* we've got the perpetrator, condemn it */
1551 qc = __ata_qc_from_tag(ap, tag);
1552 memcpy(&qc->result_tf, &tf, sizeof(tf));
1553 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1554 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1555 ehc->i.err_mask &= ~AC_ERR_DEV;
1559 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1560 * @qc: qc to analyze
1561 * @tf: Taskfile registers to analyze
1563 * Analyze taskfile of @qc and further determine cause of
1564 * failure. This function also requests ATAPI sense data if
1565 * avaliable.
1567 * LOCKING:
1568 * Kernel thread context (may sleep).
1570 * RETURNS:
1571 * Determined recovery action
1573 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1574 const struct ata_taskfile *tf)
1576 unsigned int tmp, action = 0;
1577 u8 stat = tf->command, err = tf->feature;
1579 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1580 qc->err_mask |= AC_ERR_HSM;
1581 return ATA_EH_RESET;
1584 if (stat & (ATA_ERR | ATA_DF))
1585 qc->err_mask |= AC_ERR_DEV;
1586 else
1587 return 0;
1589 switch (qc->dev->class) {
1590 case ATA_DEV_ATA:
1591 if (err & ATA_ICRC)
1592 qc->err_mask |= AC_ERR_ATA_BUS;
1593 if (err & ATA_UNC)
1594 qc->err_mask |= AC_ERR_MEDIA;
1595 if (err & ATA_IDNF)
1596 qc->err_mask |= AC_ERR_INVALID;
1597 break;
1599 case ATA_DEV_ATAPI:
1600 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1601 tmp = atapi_eh_request_sense(qc->dev,
1602 qc->scsicmd->sense_buffer,
1603 qc->result_tf.feature >> 4);
1604 if (!tmp) {
1605 /* ATA_QCFLAG_SENSE_VALID is used to
1606 * tell atapi_qc_complete() that sense
1607 * data is already valid.
1609 * TODO: interpret sense data and set
1610 * appropriate err_mask.
1612 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1613 } else
1614 qc->err_mask |= tmp;
1618 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1619 action |= ATA_EH_RESET;
1621 return action;
1624 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1625 int *xfer_ok)
1627 int base = 0;
1629 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1630 *xfer_ok = 1;
1632 if (!*xfer_ok)
1633 base = ATA_ECAT_DUBIOUS_NONE;
1635 if (err_mask & AC_ERR_ATA_BUS)
1636 return base + ATA_ECAT_ATA_BUS;
1638 if (err_mask & AC_ERR_TIMEOUT)
1639 return base + ATA_ECAT_TOUT_HSM;
1641 if (eflags & ATA_EFLAG_IS_IO) {
1642 if (err_mask & AC_ERR_HSM)
1643 return base + ATA_ECAT_TOUT_HSM;
1644 if ((err_mask &
1645 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1646 return base + ATA_ECAT_UNK_DEV;
1649 return 0;
1652 struct speed_down_verdict_arg {
1653 u64 since;
1654 int xfer_ok;
1655 int nr_errors[ATA_ECAT_NR];
1658 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1660 struct speed_down_verdict_arg *arg = void_arg;
1661 int cat;
1663 if (ent->timestamp < arg->since)
1664 return -1;
1666 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1667 &arg->xfer_ok);
1668 arg->nr_errors[cat]++;
1670 return 0;
1674 * ata_eh_speed_down_verdict - Determine speed down verdict
1675 * @dev: Device of interest
1677 * This function examines error ring of @dev and determines
1678 * whether NCQ needs to be turned off, transfer speed should be
1679 * stepped down, or falling back to PIO is necessary.
1681 * ECAT_ATA_BUS : ATA_BUS error for any command
1683 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1684 * IO commands
1686 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1688 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1689 * data transfer hasn't been verified.
1691 * Verdicts are
1693 * NCQ_OFF : Turn off NCQ.
1695 * SPEED_DOWN : Speed down transfer speed but don't fall back
1696 * to PIO.
1698 * FALLBACK_TO_PIO : Fall back to PIO.
1700 * Even if multiple verdicts are returned, only one action is
1701 * taken per error. An action triggered by non-DUBIOUS errors
1702 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1703 * This is to expedite speed down decisions right after device is
1704 * initially configured.
1706 * The followings are speed down rules. #1 and #2 deal with
1707 * DUBIOUS errors.
1709 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1710 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1712 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1713 * occurred during last 5 mins, NCQ_OFF.
1715 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1716 * ocurred during last 5 mins, FALLBACK_TO_PIO
1718 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1719 * during last 10 mins, NCQ_OFF.
1721 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1722 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1724 * LOCKING:
1725 * Inherited from caller.
1727 * RETURNS:
1728 * OR of ATA_EH_SPDN_* flags.
1730 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1732 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1733 u64 j64 = get_jiffies_64();
1734 struct speed_down_verdict_arg arg;
1735 unsigned int verdict = 0;
1737 /* scan past 5 mins of error history */
1738 memset(&arg, 0, sizeof(arg));
1739 arg.since = j64 - min(j64, j5mins);
1740 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1742 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1743 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1744 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1745 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1747 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1748 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1749 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1751 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1752 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1753 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1754 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1756 /* scan past 10 mins of error history */
1757 memset(&arg, 0, sizeof(arg));
1758 arg.since = j64 - min(j64, j10mins);
1759 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1761 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1762 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1763 verdict |= ATA_EH_SPDN_NCQ_OFF;
1765 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1766 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1767 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1768 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1770 return verdict;
1774 * ata_eh_speed_down - record error and speed down if necessary
1775 * @dev: Failed device
1776 * @eflags: mask of ATA_EFLAG_* flags
1777 * @err_mask: err_mask of the error
1779 * Record error and examine error history to determine whether
1780 * adjusting transmission speed is necessary. It also sets
1781 * transmission limits appropriately if such adjustment is
1782 * necessary.
1784 * LOCKING:
1785 * Kernel thread context (may sleep).
1787 * RETURNS:
1788 * Determined recovery action.
1790 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1791 unsigned int eflags, unsigned int err_mask)
1793 struct ata_link *link = ata_dev_phys_link(dev);
1794 int xfer_ok = 0;
1795 unsigned int verdict;
1796 unsigned int action = 0;
1798 /* don't bother if Cat-0 error */
1799 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1800 return 0;
1802 /* record error and determine whether speed down is necessary */
1803 ata_ering_record(&dev->ering, eflags, err_mask);
1804 verdict = ata_eh_speed_down_verdict(dev);
1806 /* turn off NCQ? */
1807 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1808 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1809 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1810 dev->flags |= ATA_DFLAG_NCQ_OFF;
1811 ata_dev_printk(dev, KERN_WARNING,
1812 "NCQ disabled due to excessive errors\n");
1813 goto done;
1816 /* speed down? */
1817 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1818 /* speed down SATA link speed if possible */
1819 if (sata_down_spd_limit(link) == 0) {
1820 action |= ATA_EH_RESET;
1821 goto done;
1824 /* lower transfer mode */
1825 if (dev->spdn_cnt < 2) {
1826 static const int dma_dnxfer_sel[] =
1827 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1828 static const int pio_dnxfer_sel[] =
1829 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1830 int sel;
1832 if (dev->xfer_shift != ATA_SHIFT_PIO)
1833 sel = dma_dnxfer_sel[dev->spdn_cnt];
1834 else
1835 sel = pio_dnxfer_sel[dev->spdn_cnt];
1837 dev->spdn_cnt++;
1839 if (ata_down_xfermask_limit(dev, sel) == 0) {
1840 action |= ATA_EH_RESET;
1841 goto done;
1846 /* Fall back to PIO? Slowing down to PIO is meaningless for
1847 * SATA ATA devices. Consider it only for PATA and SATAPI.
1849 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1850 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1851 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1852 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1853 dev->spdn_cnt = 0;
1854 action |= ATA_EH_RESET;
1855 goto done;
1859 return 0;
1860 done:
1861 /* device has been slowed down, blow error history */
1862 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1863 ata_ering_clear(&dev->ering);
1864 return action;
1868 * ata_eh_link_autopsy - analyze error and determine recovery action
1869 * @link: host link to perform autopsy on
1871 * Analyze why @link failed and determine which recovery actions
1872 * are needed. This function also sets more detailed AC_ERR_*
1873 * values and fills sense data for ATAPI CHECK SENSE.
1875 * LOCKING:
1876 * Kernel thread context (may sleep).
1878 static void ata_eh_link_autopsy(struct ata_link *link)
1880 struct ata_port *ap = link->ap;
1881 struct ata_eh_context *ehc = &link->eh_context;
1882 struct ata_device *dev;
1883 unsigned int all_err_mask = 0, eflags = 0;
1884 int tag;
1885 u32 serror;
1886 int rc;
1888 DPRINTK("ENTER\n");
1890 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1891 return;
1893 /* obtain and analyze SError */
1894 rc = sata_scr_read(link, SCR_ERROR, &serror);
1895 if (rc == 0) {
1896 ehc->i.serror |= serror;
1897 ata_eh_analyze_serror(link);
1898 } else if (rc != -EOPNOTSUPP) {
1899 /* SError read failed, force reset and probing */
1900 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1901 ehc->i.action |= ATA_EH_RESET;
1902 ehc->i.err_mask |= AC_ERR_OTHER;
1905 /* analyze NCQ failure */
1906 ata_eh_analyze_ncq_error(link);
1908 /* any real error trumps AC_ERR_OTHER */
1909 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1910 ehc->i.err_mask &= ~AC_ERR_OTHER;
1912 all_err_mask |= ehc->i.err_mask;
1914 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1915 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1917 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1918 ata_dev_phys_link(qc->dev) != link)
1919 continue;
1921 /* inherit upper level err_mask */
1922 qc->err_mask |= ehc->i.err_mask;
1924 /* analyze TF */
1925 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1927 /* DEV errors are probably spurious in case of ATA_BUS error */
1928 if (qc->err_mask & AC_ERR_ATA_BUS)
1929 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1930 AC_ERR_INVALID);
1932 /* any real error trumps unknown error */
1933 if (qc->err_mask & ~AC_ERR_OTHER)
1934 qc->err_mask &= ~AC_ERR_OTHER;
1936 /* SENSE_VALID trumps dev/unknown error and revalidation */
1937 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1938 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1940 /* determine whether the command is worth retrying */
1941 if (!(qc->err_mask & AC_ERR_INVALID) &&
1942 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1943 qc->flags |= ATA_QCFLAG_RETRY;
1945 /* accumulate error info */
1946 ehc->i.dev = qc->dev;
1947 all_err_mask |= qc->err_mask;
1948 if (qc->flags & ATA_QCFLAG_IO)
1949 eflags |= ATA_EFLAG_IS_IO;
1952 /* enforce default EH actions */
1953 if (ap->pflags & ATA_PFLAG_FROZEN ||
1954 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1955 ehc->i.action |= ATA_EH_RESET;
1956 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1957 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
1958 ehc->i.action |= ATA_EH_REVALIDATE;
1960 /* If we have offending qcs and the associated failed device,
1961 * perform per-dev EH action only on the offending device.
1963 if (ehc->i.dev) {
1964 ehc->i.dev_action[ehc->i.dev->devno] |=
1965 ehc->i.action & ATA_EH_PERDEV_MASK;
1966 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1969 /* propagate timeout to host link */
1970 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1971 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1973 /* record error and consider speeding down */
1974 dev = ehc->i.dev;
1975 if (!dev && ((ata_link_max_devices(link) == 1 &&
1976 ata_dev_enabled(link->device))))
1977 dev = link->device;
1979 if (dev) {
1980 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1981 eflags |= ATA_EFLAG_DUBIOUS_XFER;
1982 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
1985 DPRINTK("EXIT\n");
1989 * ata_eh_autopsy - analyze error and determine recovery action
1990 * @ap: host port to perform autopsy on
1992 * Analyze all links of @ap and determine why they failed and
1993 * which recovery actions are needed.
1995 * LOCKING:
1996 * Kernel thread context (may sleep).
1998 void ata_eh_autopsy(struct ata_port *ap)
2000 struct ata_link *link;
2002 ata_port_for_each_link(link, ap)
2003 ata_eh_link_autopsy(link);
2005 /* Handle the frigging slave link. Autopsy is done similarly
2006 * but actions and flags are transferred over to the master
2007 * link and handled from there.
2009 if (ap->slave_link) {
2010 struct ata_eh_context *mehc = &ap->link.eh_context;
2011 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2013 /* transfer control flags from master to slave */
2014 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2016 /* perform autopsy on the slave link */
2017 ata_eh_link_autopsy(ap->slave_link);
2019 /* transfer actions from slave to master and clear slave */
2020 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2021 mehc->i.action |= sehc->i.action;
2022 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2023 mehc->i.flags |= sehc->i.flags;
2024 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2027 /* Autopsy of fanout ports can affect host link autopsy.
2028 * Perform host link autopsy last.
2030 if (sata_pmp_attached(ap))
2031 ata_eh_link_autopsy(&ap->link);
2035 * ata_eh_link_report - report error handling to user
2036 * @link: ATA link EH is going on
2038 * Report EH to user.
2040 * LOCKING:
2041 * None.
2043 static void ata_eh_link_report(struct ata_link *link)
2045 struct ata_port *ap = link->ap;
2046 struct ata_eh_context *ehc = &link->eh_context;
2047 const char *frozen, *desc;
2048 char tries_buf[6];
2049 int tag, nr_failed = 0;
2051 if (ehc->i.flags & ATA_EHI_QUIET)
2052 return;
2054 desc = NULL;
2055 if (ehc->i.desc[0] != '\0')
2056 desc = ehc->i.desc;
2058 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2059 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2061 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2062 ata_dev_phys_link(qc->dev) != link ||
2063 ((qc->flags & ATA_QCFLAG_QUIET) &&
2064 qc->err_mask == AC_ERR_DEV))
2065 continue;
2066 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2067 continue;
2069 nr_failed++;
2072 if (!nr_failed && !ehc->i.err_mask)
2073 return;
2075 frozen = "";
2076 if (ap->pflags & ATA_PFLAG_FROZEN)
2077 frozen = " frozen";
2079 memset(tries_buf, 0, sizeof(tries_buf));
2080 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2081 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2082 ap->eh_tries);
2084 if (ehc->i.dev) {
2085 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2086 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2087 ehc->i.err_mask, link->sactive, ehc->i.serror,
2088 ehc->i.action, frozen, tries_buf);
2089 if (desc)
2090 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2091 } else {
2092 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2093 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2094 ehc->i.err_mask, link->sactive, ehc->i.serror,
2095 ehc->i.action, frozen, tries_buf);
2096 if (desc)
2097 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2100 if (ehc->i.serror)
2101 ata_link_printk(link, KERN_ERR,
2102 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2103 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2104 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2105 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2106 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2107 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2108 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2109 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2110 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2111 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2112 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2113 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2114 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2115 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2116 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2117 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2118 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2119 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2121 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2122 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2123 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2124 const u8 *cdb = qc->cdb;
2125 char data_buf[20] = "";
2126 char cdb_buf[70] = "";
2128 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2129 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2130 continue;
2132 if (qc->dma_dir != DMA_NONE) {
2133 static const char *dma_str[] = {
2134 [DMA_BIDIRECTIONAL] = "bidi",
2135 [DMA_TO_DEVICE] = "out",
2136 [DMA_FROM_DEVICE] = "in",
2138 static const char *prot_str[] = {
2139 [ATA_PROT_PIO] = "pio",
2140 [ATA_PROT_DMA] = "dma",
2141 [ATA_PROT_NCQ] = "ncq",
2142 [ATAPI_PROT_PIO] = "pio",
2143 [ATAPI_PROT_DMA] = "dma",
2146 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2147 prot_str[qc->tf.protocol], qc->nbytes,
2148 dma_str[qc->dma_dir]);
2151 if (ata_is_atapi(qc->tf.protocol))
2152 snprintf(cdb_buf, sizeof(cdb_buf),
2153 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2154 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2155 cdb[0], cdb[1], cdb[2], cdb[3],
2156 cdb[4], cdb[5], cdb[6], cdb[7],
2157 cdb[8], cdb[9], cdb[10], cdb[11],
2158 cdb[12], cdb[13], cdb[14], cdb[15]);
2160 ata_dev_printk(qc->dev, KERN_ERR,
2161 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2162 "tag %d%s\n %s"
2163 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2164 "Emask 0x%x (%s)%s\n",
2165 cmd->command, cmd->feature, cmd->nsect,
2166 cmd->lbal, cmd->lbam, cmd->lbah,
2167 cmd->hob_feature, cmd->hob_nsect,
2168 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2169 cmd->device, qc->tag, data_buf, cdb_buf,
2170 res->command, res->feature, res->nsect,
2171 res->lbal, res->lbam, res->lbah,
2172 res->hob_feature, res->hob_nsect,
2173 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2174 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2175 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2177 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2178 ATA_ERR)) {
2179 if (res->command & ATA_BUSY)
2180 ata_dev_printk(qc->dev, KERN_ERR,
2181 "status: { Busy }\n");
2182 else
2183 ata_dev_printk(qc->dev, KERN_ERR,
2184 "status: { %s%s%s%s}\n",
2185 res->command & ATA_DRDY ? "DRDY " : "",
2186 res->command & ATA_DF ? "DF " : "",
2187 res->command & ATA_DRQ ? "DRQ " : "",
2188 res->command & ATA_ERR ? "ERR " : "");
2191 if (cmd->command != ATA_CMD_PACKET &&
2192 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2193 ATA_ABORTED)))
2194 ata_dev_printk(qc->dev, KERN_ERR,
2195 "error: { %s%s%s%s}\n",
2196 res->feature & ATA_ICRC ? "ICRC " : "",
2197 res->feature & ATA_UNC ? "UNC " : "",
2198 res->feature & ATA_IDNF ? "IDNF " : "",
2199 res->feature & ATA_ABORTED ? "ABRT " : "");
2204 * ata_eh_report - report error handling to user
2205 * @ap: ATA port to report EH about
2207 * Report EH to user.
2209 * LOCKING:
2210 * None.
2212 void ata_eh_report(struct ata_port *ap)
2214 struct ata_link *link;
2216 __ata_port_for_each_link(link, ap)
2217 ata_eh_link_report(link);
2220 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2221 unsigned int *classes, unsigned long deadline,
2222 bool clear_classes)
2224 struct ata_device *dev;
2226 if (clear_classes)
2227 ata_link_for_each_dev(dev, link)
2228 classes[dev->devno] = ATA_DEV_UNKNOWN;
2230 return reset(link, classes, deadline);
2233 static int ata_eh_followup_srst_needed(struct ata_link *link,
2234 int rc, const unsigned int *classes)
2236 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2237 return 0;
2238 if (rc == -EAGAIN)
2239 return 1;
2240 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2241 return 1;
2242 return 0;
2245 int ata_eh_reset(struct ata_link *link, int classify,
2246 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2247 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2249 struct ata_port *ap = link->ap;
2250 struct ata_link *slave = ap->slave_link;
2251 struct ata_eh_context *ehc = &link->eh_context;
2252 struct ata_eh_context *sehc = &slave->eh_context;
2253 unsigned int *classes = ehc->classes;
2254 unsigned int lflags = link->flags;
2255 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2256 int max_tries = 0, try = 0;
2257 struct ata_link *failed_link;
2258 struct ata_device *dev;
2259 unsigned long deadline, now;
2260 ata_reset_fn_t reset;
2261 unsigned long flags;
2262 u32 sstatus;
2263 int nr_unknown, rc;
2266 * Prepare to reset
2268 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2269 max_tries++;
2270 if (link->flags & ATA_LFLAG_NO_HRST)
2271 hardreset = NULL;
2272 if (link->flags & ATA_LFLAG_NO_SRST)
2273 softreset = NULL;
2275 now = jiffies;
2276 deadline = ata_deadline(ehc->last_reset, ATA_EH_RESET_COOL_DOWN);
2277 if (time_before(now, deadline))
2278 schedule_timeout_uninterruptible(deadline - now);
2280 spin_lock_irqsave(ap->lock, flags);
2281 ap->pflags |= ATA_PFLAG_RESETTING;
2282 spin_unlock_irqrestore(ap->lock, flags);
2284 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2285 ehc->last_reset = jiffies;
2287 ata_link_for_each_dev(dev, link) {
2288 /* If we issue an SRST then an ATA drive (not ATAPI)
2289 * may change configuration and be in PIO0 timing. If
2290 * we do a hard reset (or are coming from power on)
2291 * this is true for ATA or ATAPI. Until we've set a
2292 * suitable controller mode we should not touch the
2293 * bus as we may be talking too fast.
2295 dev->pio_mode = XFER_PIO_0;
2297 /* If the controller has a pio mode setup function
2298 * then use it to set the chipset to rights. Don't
2299 * touch the DMA setup as that will be dealt with when
2300 * configuring devices.
2302 if (ap->ops->set_piomode)
2303 ap->ops->set_piomode(ap, dev);
2306 /* prefer hardreset */
2307 reset = NULL;
2308 ehc->i.action &= ~ATA_EH_RESET;
2309 if (hardreset) {
2310 reset = hardreset;
2311 ehc->i.action |= ATA_EH_HARDRESET;
2312 } else if (softreset) {
2313 reset = softreset;
2314 ehc->i.action |= ATA_EH_SOFTRESET;
2317 if (prereset) {
2318 unsigned long deadline = ata_deadline(jiffies,
2319 ATA_EH_PRERESET_TIMEOUT);
2321 if (slave) {
2322 sehc->i.action &= ~ATA_EH_RESET;
2323 sehc->i.action |= ehc->i.action;
2326 rc = prereset(link, deadline);
2328 /* If present, do prereset on slave link too. Reset
2329 * is skipped iff both master and slave links report
2330 * -ENOENT or clear ATA_EH_RESET.
2332 if (slave && (rc == 0 || rc == -ENOENT)) {
2333 int tmp;
2335 tmp = prereset(slave, deadline);
2336 if (tmp != -ENOENT)
2337 rc = tmp;
2339 ehc->i.action |= sehc->i.action;
2342 if (rc) {
2343 if (rc == -ENOENT) {
2344 ata_link_printk(link, KERN_DEBUG,
2345 "port disabled. ignoring.\n");
2346 ehc->i.action &= ~ATA_EH_RESET;
2348 ata_link_for_each_dev(dev, link)
2349 classes[dev->devno] = ATA_DEV_NONE;
2351 rc = 0;
2352 } else
2353 ata_link_printk(link, KERN_ERR,
2354 "prereset failed (errno=%d)\n", rc);
2355 goto out;
2358 /* prereset() might have cleared ATA_EH_RESET. If so,
2359 * bang classes and return.
2361 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2362 ata_link_for_each_dev(dev, link)
2363 classes[dev->devno] = ATA_DEV_NONE;
2364 rc = 0;
2365 goto out;
2369 retry:
2371 * Perform reset
2373 ehc->last_reset = jiffies;
2374 if (ata_is_host_link(link))
2375 ata_eh_freeze_port(ap);
2377 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2379 if (reset) {
2380 if (verbose)
2381 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2382 reset == softreset ? "soft" : "hard");
2384 /* mark that this EH session started with reset */
2385 if (reset == hardreset)
2386 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2387 else
2388 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2390 rc = ata_do_reset(link, reset, classes, deadline, true);
2391 if (rc && rc != -EAGAIN) {
2392 failed_link = link;
2393 goto fail;
2396 /* hardreset slave link if existent */
2397 if (slave && reset == hardreset) {
2398 int tmp;
2400 if (verbose)
2401 ata_link_printk(slave, KERN_INFO,
2402 "hard resetting link\n");
2404 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2405 tmp = ata_do_reset(slave, reset, classes, deadline,
2406 false);
2407 switch (tmp) {
2408 case -EAGAIN:
2409 rc = -EAGAIN;
2410 case 0:
2411 break;
2412 default:
2413 failed_link = slave;
2414 rc = tmp;
2415 goto fail;
2419 /* perform follow-up SRST if necessary */
2420 if (reset == hardreset &&
2421 ata_eh_followup_srst_needed(link, rc, classes)) {
2422 reset = softreset;
2424 if (!reset) {
2425 ata_link_printk(link, KERN_ERR,
2426 "follow-up softreset required "
2427 "but no softreset avaliable\n");
2428 failed_link = link;
2429 rc = -EINVAL;
2430 goto fail;
2433 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2434 rc = ata_do_reset(link, reset, classes, deadline, true);
2436 } else {
2437 if (verbose)
2438 ata_link_printk(link, KERN_INFO, "no reset method "
2439 "available, skipping reset\n");
2440 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2441 lflags |= ATA_LFLAG_ASSUME_ATA;
2445 * Post-reset processing
2447 ata_link_for_each_dev(dev, link) {
2448 /* After the reset, the device state is PIO 0 and the
2449 * controller state is undefined. Reset also wakes up
2450 * drives from sleeping mode.
2452 dev->pio_mode = XFER_PIO_0;
2453 dev->flags &= ~ATA_DFLAG_SLEEPING;
2455 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2456 continue;
2458 /* apply class override */
2459 if (lflags & ATA_LFLAG_ASSUME_ATA)
2460 classes[dev->devno] = ATA_DEV_ATA;
2461 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2462 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
2465 /* record current link speed */
2466 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2467 link->sata_spd = (sstatus >> 4) & 0xf;
2468 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2469 slave->sata_spd = (sstatus >> 4) & 0xf;
2471 /* thaw the port */
2472 if (ata_is_host_link(link))
2473 ata_eh_thaw_port(ap);
2475 /* postreset() should clear hardware SError. Although SError
2476 * is cleared during link resume, clearing SError here is
2477 * necessary as some PHYs raise hotplug events after SRST.
2478 * This introduces race condition where hotplug occurs between
2479 * reset and here. This race is mediated by cross checking
2480 * link onlineness and classification result later.
2482 if (postreset) {
2483 postreset(link, classes);
2484 if (slave)
2485 postreset(slave, classes);
2488 /* clear cached SError */
2489 spin_lock_irqsave(link->ap->lock, flags);
2490 link->eh_info.serror = 0;
2491 if (slave)
2492 slave->eh_info.serror = 0;
2493 spin_unlock_irqrestore(link->ap->lock, flags);
2495 /* Make sure onlineness and classification result correspond.
2496 * Hotplug could have happened during reset and some
2497 * controllers fail to wait while a drive is spinning up after
2498 * being hotplugged causing misdetection. By cross checking
2499 * link onlineness and classification result, those conditions
2500 * can be reliably detected and retried.
2502 nr_unknown = 0;
2503 ata_link_for_each_dev(dev, link) {
2504 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2505 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2506 classes[dev->devno] = ATA_DEV_NONE;
2507 if (ata_phys_link_online(ata_dev_phys_link(dev)))
2508 nr_unknown++;
2512 if (classify && nr_unknown) {
2513 if (try < max_tries) {
2514 ata_link_printk(link, KERN_WARNING, "link online but "
2515 "device misclassified, retrying\n");
2516 failed_link = link;
2517 rc = -EAGAIN;
2518 goto fail;
2520 ata_link_printk(link, KERN_WARNING,
2521 "link online but device misclassified, "
2522 "device detection might fail\n");
2525 /* reset successful, schedule revalidation */
2526 ata_eh_done(link, NULL, ATA_EH_RESET);
2527 if (slave)
2528 ata_eh_done(slave, NULL, ATA_EH_RESET);
2529 ehc->last_reset = jiffies;
2530 ehc->i.action |= ATA_EH_REVALIDATE;
2532 rc = 0;
2533 out:
2534 /* clear hotplug flag */
2535 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2536 if (slave)
2537 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2539 spin_lock_irqsave(ap->lock, flags);
2540 ap->pflags &= ~ATA_PFLAG_RESETTING;
2541 spin_unlock_irqrestore(ap->lock, flags);
2543 return rc;
2545 fail:
2546 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2547 if (!ata_is_host_link(link) &&
2548 sata_scr_read(link, SCR_STATUS, &sstatus))
2549 rc = -ERESTART;
2551 if (rc == -ERESTART || try >= max_tries)
2552 goto out;
2554 now = jiffies;
2555 if (time_before(now, deadline)) {
2556 unsigned long delta = deadline - now;
2558 ata_link_printk(failed_link, KERN_WARNING,
2559 "reset failed (errno=%d), retrying in %u secs\n",
2560 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2562 while (delta)
2563 delta = schedule_timeout_uninterruptible(delta);
2566 if (try == max_tries - 1) {
2567 sata_down_spd_limit(link);
2568 if (slave)
2569 sata_down_spd_limit(slave);
2570 } else if (rc == -EPIPE)
2571 sata_down_spd_limit(failed_link);
2573 if (hardreset)
2574 reset = hardreset;
2575 goto retry;
2578 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2580 struct ata_link *link;
2581 struct ata_device *dev;
2582 unsigned long flags;
2585 * This function can be thought of as an extended version of
2586 * ata_eh_about_to_do() specially crafted to accommodate the
2587 * requirements of ATA_EH_PARK handling. Since the EH thread
2588 * does not leave the do {} while () loop in ata_eh_recover as
2589 * long as the timeout for a park request to *one* device on
2590 * the port has not expired, and since we still want to pick
2591 * up park requests to other devices on the same port or
2592 * timeout updates for the same device, we have to pull
2593 * ATA_EH_PARK actions from eh_info into eh_context.i
2594 * ourselves at the beginning of each pass over the loop.
2596 * Additionally, all write accesses to &ap->park_req_pending
2597 * through INIT_COMPLETION() (see below) or complete_all()
2598 * (see ata_scsi_park_store()) are protected by the host lock.
2599 * As a result we have that park_req_pending.done is zero on
2600 * exit from this function, i.e. when ATA_EH_PARK actions for
2601 * *all* devices on port ap have been pulled into the
2602 * respective eh_context structs. If, and only if,
2603 * park_req_pending.done is non-zero by the time we reach
2604 * wait_for_completion_timeout(), another ATA_EH_PARK action
2605 * has been scheduled for at least one of the devices on port
2606 * ap and we have to cycle over the do {} while () loop in
2607 * ata_eh_recover() again.
2610 spin_lock_irqsave(ap->lock, flags);
2611 INIT_COMPLETION(ap->park_req_pending);
2612 ata_port_for_each_link(link, ap) {
2613 ata_link_for_each_dev(dev, link) {
2614 struct ata_eh_info *ehi = &link->eh_info;
2616 link->eh_context.i.dev_action[dev->devno] |=
2617 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2618 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2621 spin_unlock_irqrestore(ap->lock, flags);
2624 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2626 struct ata_eh_context *ehc = &dev->link->eh_context;
2627 struct ata_taskfile tf;
2628 unsigned int err_mask;
2630 ata_tf_init(dev, &tf);
2631 if (park) {
2632 ehc->unloaded_mask |= 1 << dev->devno;
2633 tf.command = ATA_CMD_IDLEIMMEDIATE;
2634 tf.feature = 0x44;
2635 tf.lbal = 0x4c;
2636 tf.lbam = 0x4e;
2637 tf.lbah = 0x55;
2638 } else {
2639 ehc->unloaded_mask &= ~(1 << dev->devno);
2640 tf.command = ATA_CMD_CHK_POWER;
2643 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2644 tf.protocol |= ATA_PROT_NODATA;
2645 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2646 if (park && (err_mask || tf.lbal != 0xc4)) {
2647 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2648 ehc->unloaded_mask &= ~(1 << dev->devno);
2652 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2653 struct ata_device **r_failed_dev)
2655 struct ata_port *ap = link->ap;
2656 struct ata_eh_context *ehc = &link->eh_context;
2657 struct ata_device *dev;
2658 unsigned int new_mask = 0;
2659 unsigned long flags;
2660 int rc = 0;
2662 DPRINTK("ENTER\n");
2664 /* For PATA drive side cable detection to work, IDENTIFY must
2665 * be done backwards such that PDIAG- is released by the slave
2666 * device before the master device is identified.
2668 ata_link_for_each_dev_reverse(dev, link) {
2669 unsigned int action = ata_eh_dev_action(dev);
2670 unsigned int readid_flags = 0;
2672 if (ehc->i.flags & ATA_EHI_DID_RESET)
2673 readid_flags |= ATA_READID_POSTRESET;
2675 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2676 WARN_ON(dev->class == ATA_DEV_PMP);
2678 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2679 rc = -EIO;
2680 goto err;
2683 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2684 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2685 readid_flags);
2686 if (rc)
2687 goto err;
2689 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2691 /* Configuration may have changed, reconfigure
2692 * transfer mode.
2694 ehc->i.flags |= ATA_EHI_SETMODE;
2696 /* schedule the scsi_rescan_device() here */
2697 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2698 } else if (dev->class == ATA_DEV_UNKNOWN &&
2699 ehc->tries[dev->devno] &&
2700 ata_class_enabled(ehc->classes[dev->devno])) {
2701 dev->class = ehc->classes[dev->devno];
2703 if (dev->class == ATA_DEV_PMP)
2704 rc = sata_pmp_attach(dev);
2705 else
2706 rc = ata_dev_read_id(dev, &dev->class,
2707 readid_flags, dev->id);
2708 switch (rc) {
2709 case 0:
2710 new_mask |= 1 << dev->devno;
2711 break;
2712 case -ENOENT:
2713 /* IDENTIFY was issued to non-existent
2714 * device. No need to reset. Just
2715 * thaw and kill the device.
2717 ata_eh_thaw_port(ap);
2718 dev->class = ATA_DEV_UNKNOWN;
2719 break;
2720 default:
2721 dev->class = ATA_DEV_UNKNOWN;
2722 goto err;
2727 /* PDIAG- should have been released, ask cable type if post-reset */
2728 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2729 if (ap->ops->cable_detect)
2730 ap->cbl = ap->ops->cable_detect(ap);
2731 ata_force_cbl(ap);
2734 /* Configure new devices forward such that user doesn't see
2735 * device detection messages backwards.
2737 ata_link_for_each_dev(dev, link) {
2738 if (!(new_mask & (1 << dev->devno)) ||
2739 dev->class == ATA_DEV_PMP)
2740 continue;
2742 ehc->i.flags |= ATA_EHI_PRINTINFO;
2743 rc = ata_dev_configure(dev);
2744 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2745 if (rc)
2746 goto err;
2748 spin_lock_irqsave(ap->lock, flags);
2749 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2750 spin_unlock_irqrestore(ap->lock, flags);
2752 /* new device discovered, configure xfermode */
2753 ehc->i.flags |= ATA_EHI_SETMODE;
2756 return 0;
2758 err:
2759 *r_failed_dev = dev;
2760 DPRINTK("EXIT rc=%d\n", rc);
2761 return rc;
2765 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2766 * @link: link on which timings will be programmed
2767 * @r_failed_dev: out paramter for failed device
2769 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2770 * ata_set_mode() fails, pointer to the failing device is
2771 * returned in @r_failed_dev.
2773 * LOCKING:
2774 * PCI/etc. bus probe sem.
2776 * RETURNS:
2777 * 0 on success, negative errno otherwise
2779 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2781 struct ata_port *ap = link->ap;
2782 struct ata_device *dev;
2783 int rc;
2785 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2786 ata_link_for_each_dev(dev, link) {
2787 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2788 struct ata_ering_entry *ent;
2790 ent = ata_ering_top(&dev->ering);
2791 if (ent)
2792 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2796 /* has private set_mode? */
2797 if (ap->ops->set_mode)
2798 rc = ap->ops->set_mode(link, r_failed_dev);
2799 else
2800 rc = ata_do_set_mode(link, r_failed_dev);
2802 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2803 ata_link_for_each_dev(dev, link) {
2804 struct ata_eh_context *ehc = &link->eh_context;
2805 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2806 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2808 if (dev->xfer_mode != saved_xfer_mode ||
2809 ata_ncq_enabled(dev) != saved_ncq)
2810 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2813 return rc;
2817 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2818 * @dev: ATAPI device to clear UA for
2820 * Resets and other operations can make an ATAPI device raise
2821 * UNIT ATTENTION which causes the next operation to fail. This
2822 * function clears UA.
2824 * LOCKING:
2825 * EH context (may sleep).
2827 * RETURNS:
2828 * 0 on success, -errno on failure.
2830 static int atapi_eh_clear_ua(struct ata_device *dev)
2832 int i;
2834 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
2835 u8 sense_buffer[SCSI_SENSE_BUFFERSIZE];
2836 u8 sense_key = 0;
2837 unsigned int err_mask;
2839 err_mask = atapi_eh_tur(dev, &sense_key);
2840 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
2841 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
2842 "failed (err_mask=0x%x)\n", err_mask);
2843 return -EIO;
2846 if (!err_mask || sense_key != UNIT_ATTENTION)
2847 return 0;
2849 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
2850 if (err_mask) {
2851 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
2852 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
2853 return -EIO;
2857 ata_dev_printk(dev, KERN_WARNING,
2858 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
2860 return 0;
2863 static int ata_link_nr_enabled(struct ata_link *link)
2865 struct ata_device *dev;
2866 int cnt = 0;
2868 ata_link_for_each_dev(dev, link)
2869 if (ata_dev_enabled(dev))
2870 cnt++;
2871 return cnt;
2874 static int ata_link_nr_vacant(struct ata_link *link)
2876 struct ata_device *dev;
2877 int cnt = 0;
2879 ata_link_for_each_dev(dev, link)
2880 if (dev->class == ATA_DEV_UNKNOWN)
2881 cnt++;
2882 return cnt;
2885 static int ata_eh_skip_recovery(struct ata_link *link)
2887 struct ata_port *ap = link->ap;
2888 struct ata_eh_context *ehc = &link->eh_context;
2889 struct ata_device *dev;
2891 /* skip disabled links */
2892 if (link->flags & ATA_LFLAG_DISABLED)
2893 return 1;
2895 /* thaw frozen port and recover failed devices */
2896 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2897 return 0;
2899 /* reset at least once if reset is requested */
2900 if ((ehc->i.action & ATA_EH_RESET) &&
2901 !(ehc->i.flags & ATA_EHI_DID_RESET))
2902 return 0;
2904 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2905 ata_link_for_each_dev(dev, link) {
2906 if (dev->class == ATA_DEV_UNKNOWN &&
2907 ehc->classes[dev->devno] != ATA_DEV_NONE)
2908 return 0;
2911 return 1;
2914 static int ata_eh_schedule_probe(struct ata_device *dev)
2916 struct ata_eh_context *ehc = &dev->link->eh_context;
2918 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2919 (ehc->did_probe_mask & (1 << dev->devno)))
2920 return 0;
2922 ata_eh_detach_dev(dev);
2923 ata_dev_init(dev);
2924 ehc->did_probe_mask |= (1 << dev->devno);
2925 ehc->i.action |= ATA_EH_RESET;
2926 ehc->saved_xfer_mode[dev->devno] = 0;
2927 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
2929 return 1;
2932 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2934 struct ata_eh_context *ehc = &dev->link->eh_context;
2936 ehc->tries[dev->devno]--;
2938 switch (err) {
2939 case -ENODEV:
2940 /* device missing or wrong IDENTIFY data, schedule probing */
2941 ehc->i.probe_mask |= (1 << dev->devno);
2942 case -EINVAL:
2943 /* give it just one more chance */
2944 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2945 case -EIO:
2946 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
2947 /* This is the last chance, better to slow
2948 * down than lose it.
2950 sata_down_spd_limit(ata_dev_phys_link(dev));
2951 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2955 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2956 /* disable device if it has used up all its chances */
2957 ata_dev_disable(dev);
2959 /* detach if offline */
2960 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2961 ata_eh_detach_dev(dev);
2963 /* schedule probe if necessary */
2964 if (ata_eh_schedule_probe(dev)) {
2965 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2966 memset(ehc->cmd_timeout_idx[dev->devno], 0,
2967 sizeof(ehc->cmd_timeout_idx[dev->devno]));
2970 return 1;
2971 } else {
2972 ehc->i.action |= ATA_EH_RESET;
2973 return 0;
2978 * ata_eh_recover - recover host port after error
2979 * @ap: host port to recover
2980 * @prereset: prereset method (can be NULL)
2981 * @softreset: softreset method (can be NULL)
2982 * @hardreset: hardreset method (can be NULL)
2983 * @postreset: postreset method (can be NULL)
2984 * @r_failed_link: out parameter for failed link
2986 * This is the alpha and omega, eum and yang, heart and soul of
2987 * libata exception handling. On entry, actions required to
2988 * recover each link and hotplug requests are recorded in the
2989 * link's eh_context. This function executes all the operations
2990 * with appropriate retrials and fallbacks to resurrect failed
2991 * devices, detach goners and greet newcomers.
2993 * LOCKING:
2994 * Kernel thread context (may sleep).
2996 * RETURNS:
2997 * 0 on success, -errno on failure.
2999 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3000 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3001 ata_postreset_fn_t postreset,
3002 struct ata_link **r_failed_link)
3004 struct ata_link *link;
3005 struct ata_device *dev;
3006 int nr_failed_devs;
3007 int rc;
3008 unsigned long flags, deadline;
3010 DPRINTK("ENTER\n");
3012 /* prep for recovery */
3013 ata_port_for_each_link(link, ap) {
3014 struct ata_eh_context *ehc = &link->eh_context;
3016 /* re-enable link? */
3017 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3018 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3019 spin_lock_irqsave(ap->lock, flags);
3020 link->flags &= ~ATA_LFLAG_DISABLED;
3021 spin_unlock_irqrestore(ap->lock, flags);
3022 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3025 ata_link_for_each_dev(dev, link) {
3026 if (link->flags & ATA_LFLAG_NO_RETRY)
3027 ehc->tries[dev->devno] = 1;
3028 else
3029 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3031 /* collect port action mask recorded in dev actions */
3032 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3033 ~ATA_EH_PERDEV_MASK;
3034 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3036 /* process hotplug request */
3037 if (dev->flags & ATA_DFLAG_DETACH)
3038 ata_eh_detach_dev(dev);
3040 /* schedule probe if necessary */
3041 if (!ata_dev_enabled(dev))
3042 ata_eh_schedule_probe(dev);
3046 retry:
3047 rc = 0;
3048 nr_failed_devs = 0;
3050 /* if UNLOADING, finish immediately */
3051 if (ap->pflags & ATA_PFLAG_UNLOADING)
3052 goto out;
3054 /* prep for EH */
3055 ata_port_for_each_link(link, ap) {
3056 struct ata_eh_context *ehc = &link->eh_context;
3058 /* skip EH if possible. */
3059 if (ata_eh_skip_recovery(link))
3060 ehc->i.action = 0;
3062 ata_link_for_each_dev(dev, link)
3063 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3066 /* reset */
3067 ata_port_for_each_link(link, ap) {
3068 struct ata_eh_context *ehc = &link->eh_context;
3070 if (!(ehc->i.action & ATA_EH_RESET))
3071 continue;
3073 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3074 prereset, softreset, hardreset, postreset);
3075 if (rc) {
3076 ata_link_printk(link, KERN_ERR,
3077 "reset failed, giving up\n");
3078 goto out;
3082 do {
3083 unsigned long now;
3086 * clears ATA_EH_PARK in eh_info and resets
3087 * ap->park_req_pending
3089 ata_eh_pull_park_action(ap);
3091 deadline = jiffies;
3092 ata_port_for_each_link(link, ap) {
3093 ata_link_for_each_dev(dev, link) {
3094 struct ata_eh_context *ehc = &link->eh_context;
3095 unsigned long tmp;
3097 if (dev->class != ATA_DEV_ATA)
3098 continue;
3099 if (!(ehc->i.dev_action[dev->devno] &
3100 ATA_EH_PARK))
3101 continue;
3102 tmp = dev->unpark_deadline;
3103 if (time_before(deadline, tmp))
3104 deadline = tmp;
3105 else if (time_before_eq(tmp, jiffies))
3106 continue;
3107 if (ehc->unloaded_mask & (1 << dev->devno))
3108 continue;
3110 ata_eh_park_issue_cmd(dev, 1);
3114 now = jiffies;
3115 if (time_before_eq(deadline, now))
3116 break;
3118 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3119 deadline - now);
3120 } while (deadline);
3121 ata_port_for_each_link(link, ap) {
3122 ata_link_for_each_dev(dev, link) {
3123 if (!(link->eh_context.unloaded_mask &
3124 (1 << dev->devno)))
3125 continue;
3127 ata_eh_park_issue_cmd(dev, 0);
3128 ata_eh_done(link, dev, ATA_EH_PARK);
3132 /* the rest */
3133 ata_port_for_each_link(link, ap) {
3134 struct ata_eh_context *ehc = &link->eh_context;
3136 /* revalidate existing devices and attach new ones */
3137 rc = ata_eh_revalidate_and_attach(link, &dev);
3138 if (rc)
3139 goto dev_fail;
3141 /* if PMP got attached, return, pmp EH will take care of it */
3142 if (link->device->class == ATA_DEV_PMP) {
3143 ehc->i.action = 0;
3144 return 0;
3147 /* configure transfer mode if necessary */
3148 if (ehc->i.flags & ATA_EHI_SETMODE) {
3149 rc = ata_set_mode(link, &dev);
3150 if (rc)
3151 goto dev_fail;
3152 ehc->i.flags &= ~ATA_EHI_SETMODE;
3155 /* If reset has been issued, clear UA to avoid
3156 * disrupting the current users of the device.
3158 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3159 ata_link_for_each_dev(dev, link) {
3160 if (dev->class != ATA_DEV_ATAPI)
3161 continue;
3162 rc = atapi_eh_clear_ua(dev);
3163 if (rc)
3164 goto dev_fail;
3168 /* configure link power saving */
3169 if (ehc->i.action & ATA_EH_LPM)
3170 ata_link_for_each_dev(dev, link)
3171 ata_dev_enable_pm(dev, ap->pm_policy);
3173 /* this link is okay now */
3174 ehc->i.flags = 0;
3175 continue;
3177 dev_fail:
3178 nr_failed_devs++;
3179 ata_eh_handle_dev_fail(dev, rc);
3181 if (ap->pflags & ATA_PFLAG_FROZEN) {
3182 /* PMP reset requires working host port.
3183 * Can't retry if it's frozen.
3185 if (sata_pmp_attached(ap))
3186 goto out;
3187 break;
3191 if (nr_failed_devs)
3192 goto retry;
3194 out:
3195 if (rc && r_failed_link)
3196 *r_failed_link = link;
3198 DPRINTK("EXIT, rc=%d\n", rc);
3199 return rc;
3203 * ata_eh_finish - finish up EH
3204 * @ap: host port to finish EH for
3206 * Recovery is complete. Clean up EH states and retry or finish
3207 * failed qcs.
3209 * LOCKING:
3210 * None.
3212 void ata_eh_finish(struct ata_port *ap)
3214 int tag;
3216 /* retry or finish qcs */
3217 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3218 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3220 if (!(qc->flags & ATA_QCFLAG_FAILED))
3221 continue;
3223 if (qc->err_mask) {
3224 /* FIXME: Once EH migration is complete,
3225 * generate sense data in this function,
3226 * considering both err_mask and tf.
3228 if (qc->flags & ATA_QCFLAG_RETRY)
3229 ata_eh_qc_retry(qc);
3230 else
3231 ata_eh_qc_complete(qc);
3232 } else {
3233 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3234 ata_eh_qc_complete(qc);
3235 } else {
3236 /* feed zero TF to sense generation */
3237 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3238 ata_eh_qc_retry(qc);
3243 /* make sure nr_active_links is zero after EH */
3244 WARN_ON(ap->nr_active_links);
3245 ap->nr_active_links = 0;
3249 * ata_do_eh - do standard error handling
3250 * @ap: host port to handle error for
3252 * @prereset: prereset method (can be NULL)
3253 * @softreset: softreset method (can be NULL)
3254 * @hardreset: hardreset method (can be NULL)
3255 * @postreset: postreset method (can be NULL)
3257 * Perform standard error handling sequence.
3259 * LOCKING:
3260 * Kernel thread context (may sleep).
3262 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3263 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3264 ata_postreset_fn_t postreset)
3266 struct ata_device *dev;
3267 int rc;
3269 ata_eh_autopsy(ap);
3270 ata_eh_report(ap);
3272 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3273 NULL);
3274 if (rc) {
3275 ata_link_for_each_dev(dev, &ap->link)
3276 ata_dev_disable(dev);
3279 ata_eh_finish(ap);
3283 * ata_std_error_handler - standard error handler
3284 * @ap: host port to handle error for
3286 * Standard error handler
3288 * LOCKING:
3289 * Kernel thread context (may sleep).
3291 void ata_std_error_handler(struct ata_port *ap)
3293 struct ata_port_operations *ops = ap->ops;
3294 ata_reset_fn_t hardreset = ops->hardreset;
3296 /* ignore built-in hardreset if SCR access is not available */
3297 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3298 hardreset = NULL;
3300 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3303 #ifdef CONFIG_PM
3305 * ata_eh_handle_port_suspend - perform port suspend operation
3306 * @ap: port to suspend
3308 * Suspend @ap.
3310 * LOCKING:
3311 * Kernel thread context (may sleep).
3313 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3315 unsigned long flags;
3316 int rc = 0;
3318 /* are we suspending? */
3319 spin_lock_irqsave(ap->lock, flags);
3320 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3321 ap->pm_mesg.event == PM_EVENT_ON) {
3322 spin_unlock_irqrestore(ap->lock, flags);
3323 return;
3325 spin_unlock_irqrestore(ap->lock, flags);
3327 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3329 /* tell ACPI we're suspending */
3330 rc = ata_acpi_on_suspend(ap);
3331 if (rc)
3332 goto out;
3334 /* suspend */
3335 ata_eh_freeze_port(ap);
3337 if (ap->ops->port_suspend)
3338 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3340 ata_acpi_set_state(ap, PMSG_SUSPEND);
3341 out:
3342 /* report result */
3343 spin_lock_irqsave(ap->lock, flags);
3345 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3346 if (rc == 0)
3347 ap->pflags |= ATA_PFLAG_SUSPENDED;
3348 else if (ap->pflags & ATA_PFLAG_FROZEN)
3349 ata_port_schedule_eh(ap);
3351 if (ap->pm_result) {
3352 *ap->pm_result = rc;
3353 ap->pm_result = NULL;
3356 spin_unlock_irqrestore(ap->lock, flags);
3358 return;
3362 * ata_eh_handle_port_resume - perform port resume operation
3363 * @ap: port to resume
3365 * Resume @ap.
3367 * LOCKING:
3368 * Kernel thread context (may sleep).
3370 static void ata_eh_handle_port_resume(struct ata_port *ap)
3372 unsigned long flags;
3373 int rc = 0;
3375 /* are we resuming? */
3376 spin_lock_irqsave(ap->lock, flags);
3377 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3378 ap->pm_mesg.event != PM_EVENT_ON) {
3379 spin_unlock_irqrestore(ap->lock, flags);
3380 return;
3382 spin_unlock_irqrestore(ap->lock, flags);
3384 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3386 ata_acpi_set_state(ap, PMSG_ON);
3388 if (ap->ops->port_resume)
3389 rc = ap->ops->port_resume(ap);
3391 /* tell ACPI that we're resuming */
3392 ata_acpi_on_resume(ap);
3394 /* report result */
3395 spin_lock_irqsave(ap->lock, flags);
3396 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3397 if (ap->pm_result) {
3398 *ap->pm_result = rc;
3399 ap->pm_result = NULL;
3401 spin_unlock_irqrestore(ap->lock, flags);
3403 #endif /* CONFIG_PM */