Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / drivers / ata / libata-eh.c
blob58db8344d04d9faf5527f50bac986ce9888e5902
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/export.h>
38 #include <linux/pci.h>
39 #include <scsi/scsi.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_cmnd.h>
44 #include <scsi/scsi_dbg.h>
45 #include "../scsi/scsi_transport_api.h"
47 #include <linux/libata.h>
49 #include "libata.h"
51 enum {
52 /* speed down verdicts */
53 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
54 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
55 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
56 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
58 /* error flags */
59 ATA_EFLAG_IS_IO = (1 << 0),
60 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
61 ATA_EFLAG_OLD_ER = (1 << 31),
63 /* error categories */
64 ATA_ECAT_NONE = 0,
65 ATA_ECAT_ATA_BUS = 1,
66 ATA_ECAT_TOUT_HSM = 2,
67 ATA_ECAT_UNK_DEV = 3,
68 ATA_ECAT_DUBIOUS_NONE = 4,
69 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
70 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
71 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
72 ATA_ECAT_NR = 8,
74 ATA_EH_CMD_DFL_TIMEOUT = 5000,
76 /* always put at least this amount of time between resets */
77 ATA_EH_RESET_COOL_DOWN = 5000,
79 /* Waiting in ->prereset can never be reliable. It's
80 * sometimes nice to wait there but it can't be depended upon;
81 * otherwise, we wouldn't be resetting. Just give it enough
82 * time for most drives to spin up.
84 ATA_EH_PRERESET_TIMEOUT = 10000,
85 ATA_EH_FASTDRAIN_INTERVAL = 3000,
87 ATA_EH_UA_TRIES = 5,
89 /* probe speed down parameters, see ata_eh_schedule_probe() */
90 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
91 ATA_EH_PROBE_TRIALS = 2,
94 /* The following table determines how we sequence resets. Each entry
95 * represents timeout for that try. The first try can be soft or
96 * hardreset. All others are hardreset if available. In most cases
97 * the first reset w/ 10sec timeout should succeed. Following entries
98 * are mostly for error handling, hotplug and retarded devices.
100 static const unsigned long ata_eh_reset_timeouts[] = {
101 10000, /* most drives spin up by 10sec */
102 10000, /* > 99% working drives spin up before 20sec */
103 35000, /* give > 30 secs of idleness for retarded devices */
104 5000, /* and sweet one last chance */
105 ULONG_MAX, /* > 1 min has elapsed, give up */
108 static const unsigned long ata_eh_identify_timeouts[] = {
109 5000, /* covers > 99% of successes and not too boring on failures */
110 10000, /* combined time till here is enough even for media access */
111 30000, /* for true idiots */
112 ULONG_MAX,
115 static const unsigned long ata_eh_flush_timeouts[] = {
116 15000, /* be generous with flush */
117 15000, /* ditto */
118 30000, /* and even more generous */
119 ULONG_MAX,
122 static const unsigned long ata_eh_other_timeouts[] = {
123 5000, /* same rationale as identify timeout */
124 10000, /* ditto */
125 /* but no merciful 30sec for other commands, it just isn't worth it */
126 ULONG_MAX,
129 struct ata_eh_cmd_timeout_ent {
130 const u8 *commands;
131 const unsigned long *timeouts;
134 /* The following table determines timeouts to use for EH internal
135 * commands. Each table entry is a command class and matches the
136 * commands the entry applies to and the timeout table to use.
138 * On the retry after a command timed out, the next timeout value from
139 * the table is used. If the table doesn't contain further entries,
140 * the last value is used.
142 * ehc->cmd_timeout_idx keeps track of which timeout to use per
143 * command class, so if SET_FEATURES times out on the first try, the
144 * next try will use the second timeout value only for that class.
146 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
147 static const struct ata_eh_cmd_timeout_ent
148 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
149 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
150 .timeouts = ata_eh_identify_timeouts, },
151 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
152 .timeouts = ata_eh_other_timeouts, },
153 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_FEATURES),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
158 .timeouts = ata_eh_other_timeouts, },
159 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
160 .timeouts = ata_eh_flush_timeouts },
162 #undef CMDS
164 static void __ata_port_freeze(struct ata_port *ap);
165 #ifdef CONFIG_PM
166 static void ata_eh_handle_port_suspend(struct ata_port *ap);
167 static void ata_eh_handle_port_resume(struct ata_port *ap);
168 #else /* CONFIG_PM */
169 static void ata_eh_handle_port_suspend(struct ata_port *ap)
172 static void ata_eh_handle_port_resume(struct ata_port *ap)
174 #endif /* CONFIG_PM */
176 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
177 va_list args)
179 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
180 ATA_EH_DESC_LEN - ehi->desc_len,
181 fmt, args);
185 * __ata_ehi_push_desc - push error description without adding separator
186 * @ehi: target EHI
187 * @fmt: printf format string
189 * Format string according to @fmt and append it to @ehi->desc.
191 * LOCKING:
192 * spin_lock_irqsave(host lock)
194 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
196 va_list args;
198 va_start(args, fmt);
199 __ata_ehi_pushv_desc(ehi, fmt, args);
200 va_end(args);
204 * ata_ehi_push_desc - push error description with separator
205 * @ehi: target EHI
206 * @fmt: printf format string
208 * Format string according to @fmt and append it to @ehi->desc.
209 * If @ehi->desc is not empty, ", " is added in-between.
211 * LOCKING:
212 * spin_lock_irqsave(host lock)
214 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
216 va_list args;
218 if (ehi->desc_len)
219 __ata_ehi_push_desc(ehi, ", ");
221 va_start(args, fmt);
222 __ata_ehi_pushv_desc(ehi, fmt, args);
223 va_end(args);
227 * ata_ehi_clear_desc - clean error description
228 * @ehi: target EHI
230 * Clear @ehi->desc.
232 * LOCKING:
233 * spin_lock_irqsave(host lock)
235 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
237 ehi->desc[0] = '\0';
238 ehi->desc_len = 0;
242 * ata_port_desc - append port description
243 * @ap: target ATA port
244 * @fmt: printf format string
246 * Format string according to @fmt and append it to port
247 * description. If port description is not empty, " " is added
248 * in-between. This function is to be used while initializing
249 * ata_host. The description is printed on host registration.
251 * LOCKING:
252 * None.
254 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
256 va_list args;
258 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
260 if (ap->link.eh_info.desc_len)
261 __ata_ehi_push_desc(&ap->link.eh_info, " ");
263 va_start(args, fmt);
264 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
265 va_end(args);
268 #ifdef CONFIG_PCI
271 * ata_port_pbar_desc - append PCI BAR description
272 * @ap: target ATA port
273 * @bar: target PCI BAR
274 * @offset: offset into PCI BAR
275 * @name: name of the area
277 * If @offset is negative, this function formats a string which
278 * contains the name, address, size and type of the BAR and
279 * appends it to the port description. If @offset is zero or
280 * positive, only name and offsetted address is appended.
282 * LOCKING:
283 * None.
285 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
286 const char *name)
288 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
289 char *type = "";
290 unsigned long long start, len;
292 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
293 type = "m";
294 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
295 type = "i";
297 start = (unsigned long long)pci_resource_start(pdev, bar);
298 len = (unsigned long long)pci_resource_len(pdev, bar);
300 if (offset < 0)
301 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
302 else
303 ata_port_desc(ap, "%s 0x%llx", name,
304 start + (unsigned long long)offset);
307 #endif /* CONFIG_PCI */
309 static int ata_lookup_timeout_table(u8 cmd)
311 int i;
313 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
314 const u8 *cur;
316 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
317 if (*cur == cmd)
318 return i;
321 return -1;
325 * ata_internal_cmd_timeout - determine timeout for an internal command
326 * @dev: target device
327 * @cmd: internal command to be issued
329 * Determine timeout for internal command @cmd for @dev.
331 * LOCKING:
332 * EH context.
334 * RETURNS:
335 * Determined timeout.
337 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
339 struct ata_eh_context *ehc = &dev->link->eh_context;
340 int ent = ata_lookup_timeout_table(cmd);
341 int idx;
343 if (ent < 0)
344 return ATA_EH_CMD_DFL_TIMEOUT;
346 idx = ehc->cmd_timeout_idx[dev->devno][ent];
347 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
351 * ata_internal_cmd_timed_out - notification for internal command timeout
352 * @dev: target device
353 * @cmd: internal command which timed out
355 * Notify EH that internal command @cmd for @dev timed out. This
356 * function should be called only for commands whose timeouts are
357 * determined using ata_internal_cmd_timeout().
359 * LOCKING:
360 * EH context.
362 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
364 struct ata_eh_context *ehc = &dev->link->eh_context;
365 int ent = ata_lookup_timeout_table(cmd);
366 int idx;
368 if (ent < 0)
369 return;
371 idx = ehc->cmd_timeout_idx[dev->devno][ent];
372 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
373 ehc->cmd_timeout_idx[dev->devno][ent]++;
376 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
377 unsigned int err_mask)
379 struct ata_ering_entry *ent;
381 WARN_ON(!err_mask);
383 ering->cursor++;
384 ering->cursor %= ATA_ERING_SIZE;
386 ent = &ering->ring[ering->cursor];
387 ent->eflags = eflags;
388 ent->err_mask = err_mask;
389 ent->timestamp = get_jiffies_64();
392 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
394 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
396 if (ent->err_mask)
397 return ent;
398 return NULL;
401 int ata_ering_map(struct ata_ering *ering,
402 int (*map_fn)(struct ata_ering_entry *, void *),
403 void *arg)
405 int idx, rc = 0;
406 struct ata_ering_entry *ent;
408 idx = ering->cursor;
409 do {
410 ent = &ering->ring[idx];
411 if (!ent->err_mask)
412 break;
413 rc = map_fn(ent, arg);
414 if (rc)
415 break;
416 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
417 } while (idx != ering->cursor);
419 return rc;
422 int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
424 ent->eflags |= ATA_EFLAG_OLD_ER;
425 return 0;
428 static void ata_ering_clear(struct ata_ering *ering)
430 ata_ering_map(ering, ata_ering_clear_cb, NULL);
433 static unsigned int ata_eh_dev_action(struct ata_device *dev)
435 struct ata_eh_context *ehc = &dev->link->eh_context;
437 return ehc->i.action | ehc->i.dev_action[dev->devno];
440 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
441 struct ata_eh_info *ehi, unsigned int action)
443 struct ata_device *tdev;
445 if (!dev) {
446 ehi->action &= ~action;
447 ata_for_each_dev(tdev, link, ALL)
448 ehi->dev_action[tdev->devno] &= ~action;
449 } else {
450 /* doesn't make sense for port-wide EH actions */
451 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
453 /* break ehi->action into ehi->dev_action */
454 if (ehi->action & action) {
455 ata_for_each_dev(tdev, link, ALL)
456 ehi->dev_action[tdev->devno] |=
457 ehi->action & action;
458 ehi->action &= ~action;
461 /* turn off the specified per-dev action */
462 ehi->dev_action[dev->devno] &= ~action;
467 * ata_eh_acquire - acquire EH ownership
468 * @ap: ATA port to acquire EH ownership for
470 * Acquire EH ownership for @ap. This is the basic exclusion
471 * mechanism for ports sharing a host. Only one port hanging off
472 * the same host can claim the ownership of EH.
474 * LOCKING:
475 * EH context.
477 void ata_eh_acquire(struct ata_port *ap)
479 mutex_lock(&ap->host->eh_mutex);
480 WARN_ON_ONCE(ap->host->eh_owner);
481 ap->host->eh_owner = current;
485 * ata_eh_release - release EH ownership
486 * @ap: ATA port to release EH ownership for
488 * Release EH ownership for @ap if the caller. The caller must
489 * have acquired EH ownership using ata_eh_acquire() previously.
491 * LOCKING:
492 * EH context.
494 void ata_eh_release(struct ata_port *ap)
496 WARN_ON_ONCE(ap->host->eh_owner != current);
497 ap->host->eh_owner = NULL;
498 mutex_unlock(&ap->host->eh_mutex);
502 * ata_scsi_timed_out - SCSI layer time out callback
503 * @cmd: timed out SCSI command
505 * Handles SCSI layer timeout. We race with normal completion of
506 * the qc for @cmd. If the qc is already gone, we lose and let
507 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
508 * timed out and EH should be invoked. Prevent ata_qc_complete()
509 * from finishing it by setting EH_SCHEDULED and return
510 * EH_NOT_HANDLED.
512 * TODO: kill this function once old EH is gone.
514 * LOCKING:
515 * Called from timer context
517 * RETURNS:
518 * EH_HANDLED or EH_NOT_HANDLED
520 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
522 struct Scsi_Host *host = cmd->device->host;
523 struct ata_port *ap = ata_shost_to_port(host);
524 unsigned long flags;
525 struct ata_queued_cmd *qc;
526 enum blk_eh_timer_return ret;
528 DPRINTK("ENTER\n");
530 if (ap->ops->error_handler) {
531 ret = BLK_EH_NOT_HANDLED;
532 goto out;
535 ret = BLK_EH_HANDLED;
536 spin_lock_irqsave(ap->lock, flags);
537 qc = ata_qc_from_tag(ap, ap->link.active_tag);
538 if (qc) {
539 WARN_ON(qc->scsicmd != cmd);
540 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
541 qc->err_mask |= AC_ERR_TIMEOUT;
542 ret = BLK_EH_NOT_HANDLED;
544 spin_unlock_irqrestore(ap->lock, flags);
546 out:
547 DPRINTK("EXIT, ret=%d\n", ret);
548 return ret;
551 static void ata_eh_unload(struct ata_port *ap)
553 struct ata_link *link;
554 struct ata_device *dev;
555 unsigned long flags;
557 /* Restore SControl IPM and SPD for the next driver and
558 * disable attached devices.
560 ata_for_each_link(link, ap, PMP_FIRST) {
561 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
562 ata_for_each_dev(dev, link, ALL)
563 ata_dev_disable(dev);
566 /* freeze and set UNLOADED */
567 spin_lock_irqsave(ap->lock, flags);
569 ata_port_freeze(ap); /* won't be thawed */
570 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
571 ap->pflags |= ATA_PFLAG_UNLOADED;
573 spin_unlock_irqrestore(ap->lock, flags);
577 * ata_scsi_error - SCSI layer error handler callback
578 * @host: SCSI host on which error occurred
580 * Handles SCSI-layer-thrown error events.
582 * LOCKING:
583 * Inherited from SCSI layer (none, can sleep)
585 * RETURNS:
586 * Zero.
588 void ata_scsi_error(struct Scsi_Host *host)
590 struct ata_port *ap = ata_shost_to_port(host);
591 unsigned long flags;
592 LIST_HEAD(eh_work_q);
594 DPRINTK("ENTER\n");
596 spin_lock_irqsave(host->host_lock, flags);
597 list_splice_init(&host->eh_cmd_q, &eh_work_q);
598 spin_unlock_irqrestore(host->host_lock, flags);
600 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
602 /* If we timed raced normal completion and there is nothing to
603 recover nr_timedout == 0 why exactly are we doing error recovery ? */
604 ata_scsi_port_error_handler(host, ap);
606 /* finish or retry handled scmd's and clean up */
607 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
609 DPRINTK("EXIT\n");
613 * ata_scsi_cmd_error_handler - error callback for a list of commands
614 * @host: scsi host containing the port
615 * @ap: ATA port within the host
616 * @eh_work_q: list of commands to process
618 * process the given list of commands and return those finished to the
619 * ap->eh_done_q. This function is the first part of the libata error
620 * handler which processes a given list of failed commands.
622 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
623 struct list_head *eh_work_q)
625 int i;
626 unsigned long flags;
628 /* make sure sff pio task is not running */
629 ata_sff_flush_pio_task(ap);
631 /* synchronize with host lock and sort out timeouts */
633 /* For new EH, all qcs are finished in one of three ways -
634 * normal completion, error completion, and SCSI timeout.
635 * Both completions can race against SCSI timeout. When normal
636 * completion wins, the qc never reaches EH. When error
637 * completion wins, the qc has ATA_QCFLAG_FAILED set.
639 * When SCSI timeout wins, things are a bit more complex.
640 * Normal or error completion can occur after the timeout but
641 * before this point. In such cases, both types of
642 * completions are honored. A scmd is determined to have
643 * timed out iff its associated qc is active and not failed.
645 if (ap->ops->error_handler) {
646 struct scsi_cmnd *scmd, *tmp;
647 int nr_timedout = 0;
649 spin_lock_irqsave(ap->lock, flags);
651 /* This must occur under the ap->lock as we don't want
652 a polled recovery to race the real interrupt handler
654 The lost_interrupt handler checks for any completed but
655 non-notified command and completes much like an IRQ handler.
657 We then fall into the error recovery code which will treat
658 this as if normal completion won the race */
660 if (ap->ops->lost_interrupt)
661 ap->ops->lost_interrupt(ap);
663 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
664 struct ata_queued_cmd *qc;
666 for (i = 0; i < ATA_MAX_QUEUE; i++) {
667 qc = __ata_qc_from_tag(ap, i);
668 if (qc->flags & ATA_QCFLAG_ACTIVE &&
669 qc->scsicmd == scmd)
670 break;
673 if (i < ATA_MAX_QUEUE) {
674 /* the scmd has an associated qc */
675 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
676 /* which hasn't failed yet, timeout */
677 qc->err_mask |= AC_ERR_TIMEOUT;
678 qc->flags |= ATA_QCFLAG_FAILED;
679 nr_timedout++;
681 } else {
682 /* Normal completion occurred after
683 * SCSI timeout but before this point.
684 * Successfully complete it.
686 scmd->retries = scmd->allowed;
687 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
691 /* If we have timed out qcs. They belong to EH from
692 * this point but the state of the controller is
693 * unknown. Freeze the port to make sure the IRQ
694 * handler doesn't diddle with those qcs. This must
695 * be done atomically w.r.t. setting QCFLAG_FAILED.
697 if (nr_timedout)
698 __ata_port_freeze(ap);
700 spin_unlock_irqrestore(ap->lock, flags);
702 /* initialize eh_tries */
703 ap->eh_tries = ATA_EH_MAX_TRIES;
704 } else
705 spin_unlock_wait(ap->lock);
708 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
711 * ata_scsi_port_error_handler - recover the port after the commands
712 * @host: SCSI host containing the port
713 * @ap: the ATA port
715 * Handle the recovery of the port @ap after all the commands
716 * have been recovered.
718 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
720 unsigned long flags;
722 /* invoke error handler */
723 if (ap->ops->error_handler) {
724 struct ata_link *link;
726 /* acquire EH ownership */
727 ata_eh_acquire(ap);
728 repeat:
729 /* kill fast drain timer */
730 del_timer_sync(&ap->fastdrain_timer);
732 /* process port resume request */
733 ata_eh_handle_port_resume(ap);
735 /* fetch & clear EH info */
736 spin_lock_irqsave(ap->lock, flags);
738 ata_for_each_link(link, ap, HOST_FIRST) {
739 struct ata_eh_context *ehc = &link->eh_context;
740 struct ata_device *dev;
742 memset(&link->eh_context, 0, sizeof(link->eh_context));
743 link->eh_context.i = link->eh_info;
744 memset(&link->eh_info, 0, sizeof(link->eh_info));
746 ata_for_each_dev(dev, link, ENABLED) {
747 int devno = dev->devno;
749 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
750 if (ata_ncq_enabled(dev))
751 ehc->saved_ncq_enabled |= 1 << devno;
755 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
756 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
757 ap->excl_link = NULL; /* don't maintain exclusion over EH */
759 spin_unlock_irqrestore(ap->lock, flags);
761 /* invoke EH, skip if unloading or suspended */
762 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
763 ap->ops->error_handler(ap);
764 else {
765 /* if unloading, commence suicide */
766 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
767 !(ap->pflags & ATA_PFLAG_UNLOADED))
768 ata_eh_unload(ap);
769 ata_eh_finish(ap);
772 /* process port suspend request */
773 ata_eh_handle_port_suspend(ap);
775 /* Exception might have happened after ->error_handler
776 * recovered the port but before this point. Repeat
777 * EH in such case.
779 spin_lock_irqsave(ap->lock, flags);
781 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
782 if (--ap->eh_tries) {
783 spin_unlock_irqrestore(ap->lock, flags);
784 goto repeat;
786 ata_port_err(ap,
787 "EH pending after %d tries, giving up\n",
788 ATA_EH_MAX_TRIES);
789 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
792 /* this run is complete, make sure EH info is clear */
793 ata_for_each_link(link, ap, HOST_FIRST)
794 memset(&link->eh_info, 0, sizeof(link->eh_info));
796 /* Clear host_eh_scheduled while holding ap->lock such
797 * that if exception occurs after this point but
798 * before EH completion, SCSI midlayer will
799 * re-initiate EH.
801 host->host_eh_scheduled = 0;
803 spin_unlock_irqrestore(ap->lock, flags);
804 ata_eh_release(ap);
805 } else {
806 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
807 ap->ops->eng_timeout(ap);
810 scsi_eh_flush_done_q(&ap->eh_done_q);
812 /* clean up */
813 spin_lock_irqsave(ap->lock, flags);
815 if (ap->pflags & ATA_PFLAG_LOADING)
816 ap->pflags &= ~ATA_PFLAG_LOADING;
817 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
818 schedule_delayed_work(&ap->hotplug_task, 0);
820 if (ap->pflags & ATA_PFLAG_RECOVERED)
821 ata_port_info(ap, "EH complete\n");
823 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
825 /* tell wait_eh that we're done */
826 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
827 wake_up_all(&ap->eh_wait_q);
829 spin_unlock_irqrestore(ap->lock, flags);
831 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
834 * ata_port_wait_eh - Wait for the currently pending EH to complete
835 * @ap: Port to wait EH for
837 * Wait until the currently pending EH is complete.
839 * LOCKING:
840 * Kernel thread context (may sleep).
842 void ata_port_wait_eh(struct ata_port *ap)
844 unsigned long flags;
845 DEFINE_WAIT(wait);
847 retry:
848 spin_lock_irqsave(ap->lock, flags);
850 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
851 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
852 spin_unlock_irqrestore(ap->lock, flags);
853 schedule();
854 spin_lock_irqsave(ap->lock, flags);
856 finish_wait(&ap->eh_wait_q, &wait);
858 spin_unlock_irqrestore(ap->lock, flags);
860 /* make sure SCSI EH is complete */
861 if (scsi_host_in_recovery(ap->scsi_host)) {
862 ata_msleep(ap, 10);
863 goto retry;
867 static int ata_eh_nr_in_flight(struct ata_port *ap)
869 unsigned int tag;
870 int nr = 0;
872 /* count only non-internal commands */
873 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
874 if (ata_qc_from_tag(ap, tag))
875 nr++;
877 return nr;
880 void ata_eh_fastdrain_timerfn(unsigned long arg)
882 struct ata_port *ap = (void *)arg;
883 unsigned long flags;
884 int cnt;
886 spin_lock_irqsave(ap->lock, flags);
888 cnt = ata_eh_nr_in_flight(ap);
890 /* are we done? */
891 if (!cnt)
892 goto out_unlock;
894 if (cnt == ap->fastdrain_cnt) {
895 unsigned int tag;
897 /* No progress during the last interval, tag all
898 * in-flight qcs as timed out and freeze the port.
900 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
901 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
902 if (qc)
903 qc->err_mask |= AC_ERR_TIMEOUT;
906 ata_port_freeze(ap);
907 } else {
908 /* some qcs have finished, give it another chance */
909 ap->fastdrain_cnt = cnt;
910 ap->fastdrain_timer.expires =
911 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
912 add_timer(&ap->fastdrain_timer);
915 out_unlock:
916 spin_unlock_irqrestore(ap->lock, flags);
920 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
921 * @ap: target ATA port
922 * @fastdrain: activate fast drain
924 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
925 * is non-zero and EH wasn't pending before. Fast drain ensures
926 * that EH kicks in in timely manner.
928 * LOCKING:
929 * spin_lock_irqsave(host lock)
931 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
933 int cnt;
935 /* already scheduled? */
936 if (ap->pflags & ATA_PFLAG_EH_PENDING)
937 return;
939 ap->pflags |= ATA_PFLAG_EH_PENDING;
941 if (!fastdrain)
942 return;
944 /* do we have in-flight qcs? */
945 cnt = ata_eh_nr_in_flight(ap);
946 if (!cnt)
947 return;
949 /* activate fast drain */
950 ap->fastdrain_cnt = cnt;
951 ap->fastdrain_timer.expires =
952 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
953 add_timer(&ap->fastdrain_timer);
957 * ata_qc_schedule_eh - schedule qc for error handling
958 * @qc: command to schedule error handling for
960 * Schedule error handling for @qc. EH will kick in as soon as
961 * other commands are drained.
963 * LOCKING:
964 * spin_lock_irqsave(host lock)
966 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
968 struct ata_port *ap = qc->ap;
969 struct request_queue *q = qc->scsicmd->device->request_queue;
970 unsigned long flags;
972 WARN_ON(!ap->ops->error_handler);
974 qc->flags |= ATA_QCFLAG_FAILED;
975 ata_eh_set_pending(ap, 1);
977 /* The following will fail if timeout has already expired.
978 * ata_scsi_error() takes care of such scmds on EH entry.
979 * Note that ATA_QCFLAG_FAILED is unconditionally set after
980 * this function completes.
982 spin_lock_irqsave(q->queue_lock, flags);
983 blk_abort_request(qc->scsicmd->request);
984 spin_unlock_irqrestore(q->queue_lock, flags);
988 * ata_port_schedule_eh - schedule error handling without a qc
989 * @ap: ATA port to schedule EH for
991 * Schedule error handling for @ap. EH will kick in as soon as
992 * all commands are drained.
994 * LOCKING:
995 * spin_lock_irqsave(host lock)
997 void ata_port_schedule_eh(struct ata_port *ap)
999 WARN_ON(!ap->ops->error_handler);
1001 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1002 return;
1004 ata_eh_set_pending(ap, 1);
1005 scsi_schedule_eh(ap->scsi_host);
1007 DPRINTK("port EH scheduled\n");
1010 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1012 int tag, nr_aborted = 0;
1014 WARN_ON(!ap->ops->error_handler);
1016 /* we're gonna abort all commands, no need for fast drain */
1017 ata_eh_set_pending(ap, 0);
1019 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1020 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1022 if (qc && (!link || qc->dev->link == link)) {
1023 qc->flags |= ATA_QCFLAG_FAILED;
1024 ata_qc_complete(qc);
1025 nr_aborted++;
1029 if (!nr_aborted)
1030 ata_port_schedule_eh(ap);
1032 return nr_aborted;
1036 * ata_link_abort - abort all qc's on the link
1037 * @link: ATA link to abort qc's for
1039 * Abort all active qc's active on @link and schedule EH.
1041 * LOCKING:
1042 * spin_lock_irqsave(host lock)
1044 * RETURNS:
1045 * Number of aborted qc's.
1047 int ata_link_abort(struct ata_link *link)
1049 return ata_do_link_abort(link->ap, link);
1053 * ata_port_abort - abort all qc's on the port
1054 * @ap: ATA port to abort qc's for
1056 * Abort all active qc's of @ap and schedule EH.
1058 * LOCKING:
1059 * spin_lock_irqsave(host_set lock)
1061 * RETURNS:
1062 * Number of aborted qc's.
1064 int ata_port_abort(struct ata_port *ap)
1066 return ata_do_link_abort(ap, NULL);
1070 * __ata_port_freeze - freeze port
1071 * @ap: ATA port to freeze
1073 * This function is called when HSM violation or some other
1074 * condition disrupts normal operation of the port. Frozen port
1075 * is not allowed to perform any operation until the port is
1076 * thawed, which usually follows a successful reset.
1078 * ap->ops->freeze() callback can be used for freezing the port
1079 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1080 * port cannot be frozen hardware-wise, the interrupt handler
1081 * must ack and clear interrupts unconditionally while the port
1082 * is frozen.
1084 * LOCKING:
1085 * spin_lock_irqsave(host lock)
1087 static void __ata_port_freeze(struct ata_port *ap)
1089 WARN_ON(!ap->ops->error_handler);
1091 if (ap->ops->freeze)
1092 ap->ops->freeze(ap);
1094 ap->pflags |= ATA_PFLAG_FROZEN;
1096 DPRINTK("ata%u port frozen\n", ap->print_id);
1100 * ata_port_freeze - abort & freeze port
1101 * @ap: ATA port to freeze
1103 * Abort and freeze @ap. The freeze operation must be called
1104 * first, because some hardware requires special operations
1105 * before the taskfile registers are accessible.
1107 * LOCKING:
1108 * spin_lock_irqsave(host lock)
1110 * RETURNS:
1111 * Number of aborted commands.
1113 int ata_port_freeze(struct ata_port *ap)
1115 int nr_aborted;
1117 WARN_ON(!ap->ops->error_handler);
1119 __ata_port_freeze(ap);
1120 nr_aborted = ata_port_abort(ap);
1122 return nr_aborted;
1126 * sata_async_notification - SATA async notification handler
1127 * @ap: ATA port where async notification is received
1129 * Handler to be called when async notification via SDB FIS is
1130 * received. This function schedules EH if necessary.
1132 * LOCKING:
1133 * spin_lock_irqsave(host lock)
1135 * RETURNS:
1136 * 1 if EH is scheduled, 0 otherwise.
1138 int sata_async_notification(struct ata_port *ap)
1140 u32 sntf;
1141 int rc;
1143 if (!(ap->flags & ATA_FLAG_AN))
1144 return 0;
1146 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1147 if (rc == 0)
1148 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1150 if (!sata_pmp_attached(ap) || rc) {
1151 /* PMP is not attached or SNTF is not available */
1152 if (!sata_pmp_attached(ap)) {
1153 /* PMP is not attached. Check whether ATAPI
1154 * AN is configured. If so, notify media
1155 * change.
1157 struct ata_device *dev = ap->link.device;
1159 if ((dev->class == ATA_DEV_ATAPI) &&
1160 (dev->flags & ATA_DFLAG_AN))
1161 ata_scsi_media_change_notify(dev);
1162 return 0;
1163 } else {
1164 /* PMP is attached but SNTF is not available.
1165 * ATAPI async media change notification is
1166 * not used. The PMP must be reporting PHY
1167 * status change, schedule EH.
1169 ata_port_schedule_eh(ap);
1170 return 1;
1172 } else {
1173 /* PMP is attached and SNTF is available */
1174 struct ata_link *link;
1176 /* check and notify ATAPI AN */
1177 ata_for_each_link(link, ap, EDGE) {
1178 if (!(sntf & (1 << link->pmp)))
1179 continue;
1181 if ((link->device->class == ATA_DEV_ATAPI) &&
1182 (link->device->flags & ATA_DFLAG_AN))
1183 ata_scsi_media_change_notify(link->device);
1186 /* If PMP is reporting that PHY status of some
1187 * downstream ports has changed, schedule EH.
1189 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1190 ata_port_schedule_eh(ap);
1191 return 1;
1194 return 0;
1199 * ata_eh_freeze_port - EH helper to freeze port
1200 * @ap: ATA port to freeze
1202 * Freeze @ap.
1204 * LOCKING:
1205 * None.
1207 void ata_eh_freeze_port(struct ata_port *ap)
1209 unsigned long flags;
1211 if (!ap->ops->error_handler)
1212 return;
1214 spin_lock_irqsave(ap->lock, flags);
1215 __ata_port_freeze(ap);
1216 spin_unlock_irqrestore(ap->lock, flags);
1220 * ata_port_thaw_port - EH helper to thaw port
1221 * @ap: ATA port to thaw
1223 * Thaw frozen port @ap.
1225 * LOCKING:
1226 * None.
1228 void ata_eh_thaw_port(struct ata_port *ap)
1230 unsigned long flags;
1232 if (!ap->ops->error_handler)
1233 return;
1235 spin_lock_irqsave(ap->lock, flags);
1237 ap->pflags &= ~ATA_PFLAG_FROZEN;
1239 if (ap->ops->thaw)
1240 ap->ops->thaw(ap);
1242 spin_unlock_irqrestore(ap->lock, flags);
1244 DPRINTK("ata%u port thawed\n", ap->print_id);
1247 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1249 /* nada */
1252 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1254 struct ata_port *ap = qc->ap;
1255 struct scsi_cmnd *scmd = qc->scsicmd;
1256 unsigned long flags;
1258 spin_lock_irqsave(ap->lock, flags);
1259 qc->scsidone = ata_eh_scsidone;
1260 __ata_qc_complete(qc);
1261 WARN_ON(ata_tag_valid(qc->tag));
1262 spin_unlock_irqrestore(ap->lock, flags);
1264 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1268 * ata_eh_qc_complete - Complete an active ATA command from EH
1269 * @qc: Command to complete
1271 * Indicate to the mid and upper layers that an ATA command has
1272 * completed. To be used from EH.
1274 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1276 struct scsi_cmnd *scmd = qc->scsicmd;
1277 scmd->retries = scmd->allowed;
1278 __ata_eh_qc_complete(qc);
1282 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1283 * @qc: Command to retry
1285 * Indicate to the mid and upper layers that an ATA command
1286 * should be retried. To be used from EH.
1288 * SCSI midlayer limits the number of retries to scmd->allowed.
1289 * scmd->retries is decremented for commands which get retried
1290 * due to unrelated failures (qc->err_mask is zero).
1292 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1294 struct scsi_cmnd *scmd = qc->scsicmd;
1295 if (!qc->err_mask && scmd->retries)
1296 scmd->retries--;
1297 __ata_eh_qc_complete(qc);
1301 * ata_dev_disable - disable ATA device
1302 * @dev: ATA device to disable
1304 * Disable @dev.
1306 * Locking:
1307 * EH context.
1309 void ata_dev_disable(struct ata_device *dev)
1311 if (!ata_dev_enabled(dev))
1312 return;
1314 if (ata_msg_drv(dev->link->ap))
1315 ata_dev_warn(dev, "disabled\n");
1316 ata_acpi_on_disable(dev);
1317 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1318 dev->class++;
1320 /* From now till the next successful probe, ering is used to
1321 * track probe failures. Clear accumulated device error info.
1323 ata_ering_clear(&dev->ering);
1327 * ata_eh_detach_dev - detach ATA device
1328 * @dev: ATA device to detach
1330 * Detach @dev.
1332 * LOCKING:
1333 * None.
1335 void ata_eh_detach_dev(struct ata_device *dev)
1337 struct ata_link *link = dev->link;
1338 struct ata_port *ap = link->ap;
1339 struct ata_eh_context *ehc = &link->eh_context;
1340 unsigned long flags;
1342 ata_dev_disable(dev);
1344 spin_lock_irqsave(ap->lock, flags);
1346 dev->flags &= ~ATA_DFLAG_DETACH;
1348 if (ata_scsi_offline_dev(dev)) {
1349 dev->flags |= ATA_DFLAG_DETACHED;
1350 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1353 /* clear per-dev EH info */
1354 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1355 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1356 ehc->saved_xfer_mode[dev->devno] = 0;
1357 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1359 spin_unlock_irqrestore(ap->lock, flags);
1363 * ata_eh_about_to_do - about to perform eh_action
1364 * @link: target ATA link
1365 * @dev: target ATA dev for per-dev action (can be NULL)
1366 * @action: action about to be performed
1368 * Called just before performing EH actions to clear related bits
1369 * in @link->eh_info such that eh actions are not unnecessarily
1370 * repeated.
1372 * LOCKING:
1373 * None.
1375 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1376 unsigned int action)
1378 struct ata_port *ap = link->ap;
1379 struct ata_eh_info *ehi = &link->eh_info;
1380 struct ata_eh_context *ehc = &link->eh_context;
1381 unsigned long flags;
1383 spin_lock_irqsave(ap->lock, flags);
1385 ata_eh_clear_action(link, dev, ehi, action);
1387 /* About to take EH action, set RECOVERED. Ignore actions on
1388 * slave links as master will do them again.
1390 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1391 ap->pflags |= ATA_PFLAG_RECOVERED;
1393 spin_unlock_irqrestore(ap->lock, flags);
1397 * ata_eh_done - EH action complete
1398 * @ap: target ATA port
1399 * @dev: target ATA dev for per-dev action (can be NULL)
1400 * @action: action just completed
1402 * Called right after performing EH actions to clear related bits
1403 * in @link->eh_context.
1405 * LOCKING:
1406 * None.
1408 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1409 unsigned int action)
1411 struct ata_eh_context *ehc = &link->eh_context;
1413 ata_eh_clear_action(link, dev, &ehc->i, action);
1417 * ata_err_string - convert err_mask to descriptive string
1418 * @err_mask: error mask to convert to string
1420 * Convert @err_mask to descriptive string. Errors are
1421 * prioritized according to severity and only the most severe
1422 * error is reported.
1424 * LOCKING:
1425 * None.
1427 * RETURNS:
1428 * Descriptive string for @err_mask
1430 static const char *ata_err_string(unsigned int err_mask)
1432 if (err_mask & AC_ERR_HOST_BUS)
1433 return "host bus error";
1434 if (err_mask & AC_ERR_ATA_BUS)
1435 return "ATA bus error";
1436 if (err_mask & AC_ERR_TIMEOUT)
1437 return "timeout";
1438 if (err_mask & AC_ERR_HSM)
1439 return "HSM violation";
1440 if (err_mask & AC_ERR_SYSTEM)
1441 return "internal error";
1442 if (err_mask & AC_ERR_MEDIA)
1443 return "media error";
1444 if (err_mask & AC_ERR_INVALID)
1445 return "invalid argument";
1446 if (err_mask & AC_ERR_DEV)
1447 return "device error";
1448 return "unknown error";
1452 * ata_read_log_page - read a specific log page
1453 * @dev: target device
1454 * @page: page to read
1455 * @buf: buffer to store read page
1456 * @sectors: number of sectors to read
1458 * Read log page using READ_LOG_EXT command.
1460 * LOCKING:
1461 * Kernel thread context (may sleep).
1463 * RETURNS:
1464 * 0 on success, AC_ERR_* mask otherwise.
1466 static unsigned int ata_read_log_page(struct ata_device *dev,
1467 u8 page, void *buf, unsigned int sectors)
1469 struct ata_taskfile tf;
1470 unsigned int err_mask;
1472 DPRINTK("read log page - page %d\n", page);
1474 ata_tf_init(dev, &tf);
1475 tf.command = ATA_CMD_READ_LOG_EXT;
1476 tf.lbal = page;
1477 tf.nsect = sectors;
1478 tf.hob_nsect = sectors >> 8;
1479 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1480 tf.protocol = ATA_PROT_PIO;
1482 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1483 buf, sectors * ATA_SECT_SIZE, 0);
1485 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1486 return err_mask;
1490 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1491 * @dev: Device to read log page 10h from
1492 * @tag: Resulting tag of the failed command
1493 * @tf: Resulting taskfile registers of the failed command
1495 * Read log page 10h to obtain NCQ error details and clear error
1496 * condition.
1498 * LOCKING:
1499 * Kernel thread context (may sleep).
1501 * RETURNS:
1502 * 0 on success, -errno otherwise.
1504 static int ata_eh_read_log_10h(struct ata_device *dev,
1505 int *tag, struct ata_taskfile *tf)
1507 u8 *buf = dev->link->ap->sector_buf;
1508 unsigned int err_mask;
1509 u8 csum;
1510 int i;
1512 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1513 if (err_mask)
1514 return -EIO;
1516 csum = 0;
1517 for (i = 0; i < ATA_SECT_SIZE; i++)
1518 csum += buf[i];
1519 if (csum)
1520 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1521 csum);
1523 if (buf[0] & 0x80)
1524 return -ENOENT;
1526 *tag = buf[0] & 0x1f;
1528 tf->command = buf[2];
1529 tf->feature = buf[3];
1530 tf->lbal = buf[4];
1531 tf->lbam = buf[5];
1532 tf->lbah = buf[6];
1533 tf->device = buf[7];
1534 tf->hob_lbal = buf[8];
1535 tf->hob_lbam = buf[9];
1536 tf->hob_lbah = buf[10];
1537 tf->nsect = buf[12];
1538 tf->hob_nsect = buf[13];
1540 return 0;
1544 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1545 * @dev: target ATAPI device
1546 * @r_sense_key: out parameter for sense_key
1548 * Perform ATAPI TEST_UNIT_READY.
1550 * LOCKING:
1551 * EH context (may sleep).
1553 * RETURNS:
1554 * 0 on success, AC_ERR_* mask on failure.
1556 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1558 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1559 struct ata_taskfile tf;
1560 unsigned int err_mask;
1562 ata_tf_init(dev, &tf);
1564 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1565 tf.command = ATA_CMD_PACKET;
1566 tf.protocol = ATAPI_PROT_NODATA;
1568 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1569 if (err_mask == AC_ERR_DEV)
1570 *r_sense_key = tf.feature >> 4;
1571 return err_mask;
1575 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1576 * @dev: device to perform REQUEST_SENSE to
1577 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1578 * @dfl_sense_key: default sense key to use
1580 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1581 * SENSE. This function is EH helper.
1583 * LOCKING:
1584 * Kernel thread context (may sleep).
1586 * RETURNS:
1587 * 0 on success, AC_ERR_* mask on failure
1589 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1590 u8 *sense_buf, u8 dfl_sense_key)
1592 u8 cdb[ATAPI_CDB_LEN] =
1593 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1594 struct ata_port *ap = dev->link->ap;
1595 struct ata_taskfile tf;
1597 DPRINTK("ATAPI request sense\n");
1599 /* FIXME: is this needed? */
1600 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1602 /* initialize sense_buf with the error register,
1603 * for the case where they are -not- overwritten
1605 sense_buf[0] = 0x70;
1606 sense_buf[2] = dfl_sense_key;
1608 /* some devices time out if garbage left in tf */
1609 ata_tf_init(dev, &tf);
1611 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1612 tf.command = ATA_CMD_PACKET;
1614 /* is it pointless to prefer PIO for "safety reasons"? */
1615 if (ap->flags & ATA_FLAG_PIO_DMA) {
1616 tf.protocol = ATAPI_PROT_DMA;
1617 tf.feature |= ATAPI_PKT_DMA;
1618 } else {
1619 tf.protocol = ATAPI_PROT_PIO;
1620 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1621 tf.lbah = 0;
1624 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1625 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1629 * ata_eh_analyze_serror - analyze SError for a failed port
1630 * @link: ATA link to analyze SError for
1632 * Analyze SError if available and further determine cause of
1633 * failure.
1635 * LOCKING:
1636 * None.
1638 static void ata_eh_analyze_serror(struct ata_link *link)
1640 struct ata_eh_context *ehc = &link->eh_context;
1641 u32 serror = ehc->i.serror;
1642 unsigned int err_mask = 0, action = 0;
1643 u32 hotplug_mask;
1645 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1646 err_mask |= AC_ERR_ATA_BUS;
1647 action |= ATA_EH_RESET;
1649 if (serror & SERR_PROTOCOL) {
1650 err_mask |= AC_ERR_HSM;
1651 action |= ATA_EH_RESET;
1653 if (serror & SERR_INTERNAL) {
1654 err_mask |= AC_ERR_SYSTEM;
1655 action |= ATA_EH_RESET;
1658 /* Determine whether a hotplug event has occurred. Both
1659 * SError.N/X are considered hotplug events for enabled or
1660 * host links. For disabled PMP links, only N bit is
1661 * considered as X bit is left at 1 for link plugging.
1663 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1664 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1665 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1666 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1667 else
1668 hotplug_mask = SERR_PHYRDY_CHG;
1670 if (serror & hotplug_mask)
1671 ata_ehi_hotplugged(&ehc->i);
1673 ehc->i.err_mask |= err_mask;
1674 ehc->i.action |= action;
1678 * ata_eh_analyze_ncq_error - analyze NCQ error
1679 * @link: ATA link to analyze NCQ error for
1681 * Read log page 10h, determine the offending qc and acquire
1682 * error status TF. For NCQ device errors, all LLDDs have to do
1683 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1684 * care of the rest.
1686 * LOCKING:
1687 * Kernel thread context (may sleep).
1689 void ata_eh_analyze_ncq_error(struct ata_link *link)
1691 struct ata_port *ap = link->ap;
1692 struct ata_eh_context *ehc = &link->eh_context;
1693 struct ata_device *dev = link->device;
1694 struct ata_queued_cmd *qc;
1695 struct ata_taskfile tf;
1696 int tag, rc;
1698 /* if frozen, we can't do much */
1699 if (ap->pflags & ATA_PFLAG_FROZEN)
1700 return;
1702 /* is it NCQ device error? */
1703 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1704 return;
1706 /* has LLDD analyzed already? */
1707 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1708 qc = __ata_qc_from_tag(ap, tag);
1710 if (!(qc->flags & ATA_QCFLAG_FAILED))
1711 continue;
1713 if (qc->err_mask)
1714 return;
1717 /* okay, this error is ours */
1718 memset(&tf, 0, sizeof(tf));
1719 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1720 if (rc) {
1721 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1722 rc);
1723 return;
1726 if (!(link->sactive & (1 << tag))) {
1727 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1728 tag);
1729 return;
1732 /* we've got the perpetrator, condemn it */
1733 qc = __ata_qc_from_tag(ap, tag);
1734 memcpy(&qc->result_tf, &tf, sizeof(tf));
1735 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1736 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1737 ehc->i.err_mask &= ~AC_ERR_DEV;
1741 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1742 * @qc: qc to analyze
1743 * @tf: Taskfile registers to analyze
1745 * Analyze taskfile of @qc and further determine cause of
1746 * failure. This function also requests ATAPI sense data if
1747 * available.
1749 * LOCKING:
1750 * Kernel thread context (may sleep).
1752 * RETURNS:
1753 * Determined recovery action
1755 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1756 const struct ata_taskfile *tf)
1758 unsigned int tmp, action = 0;
1759 u8 stat = tf->command, err = tf->feature;
1761 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1762 qc->err_mask |= AC_ERR_HSM;
1763 return ATA_EH_RESET;
1766 if (stat & (ATA_ERR | ATA_DF))
1767 qc->err_mask |= AC_ERR_DEV;
1768 else
1769 return 0;
1771 switch (qc->dev->class) {
1772 case ATA_DEV_ATA:
1773 if (err & ATA_ICRC)
1774 qc->err_mask |= AC_ERR_ATA_BUS;
1775 if (err & ATA_UNC)
1776 qc->err_mask |= AC_ERR_MEDIA;
1777 if (err & ATA_IDNF)
1778 qc->err_mask |= AC_ERR_INVALID;
1779 break;
1781 case ATA_DEV_ATAPI:
1782 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1783 tmp = atapi_eh_request_sense(qc->dev,
1784 qc->scsicmd->sense_buffer,
1785 qc->result_tf.feature >> 4);
1786 if (!tmp) {
1787 /* ATA_QCFLAG_SENSE_VALID is used to
1788 * tell atapi_qc_complete() that sense
1789 * data is already valid.
1791 * TODO: interpret sense data and set
1792 * appropriate err_mask.
1794 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1795 } else
1796 qc->err_mask |= tmp;
1800 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1801 action |= ATA_EH_RESET;
1803 return action;
1806 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1807 int *xfer_ok)
1809 int base = 0;
1811 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1812 *xfer_ok = 1;
1814 if (!*xfer_ok)
1815 base = ATA_ECAT_DUBIOUS_NONE;
1817 if (err_mask & AC_ERR_ATA_BUS)
1818 return base + ATA_ECAT_ATA_BUS;
1820 if (err_mask & AC_ERR_TIMEOUT)
1821 return base + ATA_ECAT_TOUT_HSM;
1823 if (eflags & ATA_EFLAG_IS_IO) {
1824 if (err_mask & AC_ERR_HSM)
1825 return base + ATA_ECAT_TOUT_HSM;
1826 if ((err_mask &
1827 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1828 return base + ATA_ECAT_UNK_DEV;
1831 return 0;
1834 struct speed_down_verdict_arg {
1835 u64 since;
1836 int xfer_ok;
1837 int nr_errors[ATA_ECAT_NR];
1840 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1842 struct speed_down_verdict_arg *arg = void_arg;
1843 int cat;
1845 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1846 return -1;
1848 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1849 &arg->xfer_ok);
1850 arg->nr_errors[cat]++;
1852 return 0;
1856 * ata_eh_speed_down_verdict - Determine speed down verdict
1857 * @dev: Device of interest
1859 * This function examines error ring of @dev and determines
1860 * whether NCQ needs to be turned off, transfer speed should be
1861 * stepped down, or falling back to PIO is necessary.
1863 * ECAT_ATA_BUS : ATA_BUS error for any command
1865 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1866 * IO commands
1868 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1870 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1871 * data transfer hasn't been verified.
1873 * Verdicts are
1875 * NCQ_OFF : Turn off NCQ.
1877 * SPEED_DOWN : Speed down transfer speed but don't fall back
1878 * to PIO.
1880 * FALLBACK_TO_PIO : Fall back to PIO.
1882 * Even if multiple verdicts are returned, only one action is
1883 * taken per error. An action triggered by non-DUBIOUS errors
1884 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1885 * This is to expedite speed down decisions right after device is
1886 * initially configured.
1888 * The followings are speed down rules. #1 and #2 deal with
1889 * DUBIOUS errors.
1891 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1892 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1894 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1895 * occurred during last 5 mins, NCQ_OFF.
1897 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1898 * occurred during last 5 mins, FALLBACK_TO_PIO
1900 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1901 * during last 10 mins, NCQ_OFF.
1903 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1904 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1906 * LOCKING:
1907 * Inherited from caller.
1909 * RETURNS:
1910 * OR of ATA_EH_SPDN_* flags.
1912 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1914 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1915 u64 j64 = get_jiffies_64();
1916 struct speed_down_verdict_arg arg;
1917 unsigned int verdict = 0;
1919 /* scan past 5 mins of error history */
1920 memset(&arg, 0, sizeof(arg));
1921 arg.since = j64 - min(j64, j5mins);
1922 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1924 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1925 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1926 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1927 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1929 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1930 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1931 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1933 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1934 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1935 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1936 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1938 /* scan past 10 mins of error history */
1939 memset(&arg, 0, sizeof(arg));
1940 arg.since = j64 - min(j64, j10mins);
1941 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1943 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1944 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1945 verdict |= ATA_EH_SPDN_NCQ_OFF;
1947 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1948 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1949 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1950 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1952 return verdict;
1956 * ata_eh_speed_down - record error and speed down if necessary
1957 * @dev: Failed device
1958 * @eflags: mask of ATA_EFLAG_* flags
1959 * @err_mask: err_mask of the error
1961 * Record error and examine error history to determine whether
1962 * adjusting transmission speed is necessary. It also sets
1963 * transmission limits appropriately if such adjustment is
1964 * necessary.
1966 * LOCKING:
1967 * Kernel thread context (may sleep).
1969 * RETURNS:
1970 * Determined recovery action.
1972 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1973 unsigned int eflags, unsigned int err_mask)
1975 struct ata_link *link = ata_dev_phys_link(dev);
1976 int xfer_ok = 0;
1977 unsigned int verdict;
1978 unsigned int action = 0;
1980 /* don't bother if Cat-0 error */
1981 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1982 return 0;
1984 /* record error and determine whether speed down is necessary */
1985 ata_ering_record(&dev->ering, eflags, err_mask);
1986 verdict = ata_eh_speed_down_verdict(dev);
1988 /* turn off NCQ? */
1989 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1990 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1991 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1992 dev->flags |= ATA_DFLAG_NCQ_OFF;
1993 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1994 goto done;
1997 /* speed down? */
1998 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1999 /* speed down SATA link speed if possible */
2000 if (sata_down_spd_limit(link, 0) == 0) {
2001 action |= ATA_EH_RESET;
2002 goto done;
2005 /* lower transfer mode */
2006 if (dev->spdn_cnt < 2) {
2007 static const int dma_dnxfer_sel[] =
2008 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2009 static const int pio_dnxfer_sel[] =
2010 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2011 int sel;
2013 if (dev->xfer_shift != ATA_SHIFT_PIO)
2014 sel = dma_dnxfer_sel[dev->spdn_cnt];
2015 else
2016 sel = pio_dnxfer_sel[dev->spdn_cnt];
2018 dev->spdn_cnt++;
2020 if (ata_down_xfermask_limit(dev, sel) == 0) {
2021 action |= ATA_EH_RESET;
2022 goto done;
2027 /* Fall back to PIO? Slowing down to PIO is meaningless for
2028 * SATA ATA devices. Consider it only for PATA and SATAPI.
2030 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2031 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2032 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2033 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2034 dev->spdn_cnt = 0;
2035 action |= ATA_EH_RESET;
2036 goto done;
2040 return 0;
2041 done:
2042 /* device has been slowed down, blow error history */
2043 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2044 ata_ering_clear(&dev->ering);
2045 return action;
2049 * ata_eh_link_autopsy - analyze error and determine recovery action
2050 * @link: host link to perform autopsy on
2052 * Analyze why @link failed and determine which recovery actions
2053 * are needed. This function also sets more detailed AC_ERR_*
2054 * values and fills sense data for ATAPI CHECK SENSE.
2056 * LOCKING:
2057 * Kernel thread context (may sleep).
2059 static void ata_eh_link_autopsy(struct ata_link *link)
2061 struct ata_port *ap = link->ap;
2062 struct ata_eh_context *ehc = &link->eh_context;
2063 struct ata_device *dev;
2064 unsigned int all_err_mask = 0, eflags = 0;
2065 int tag;
2066 u32 serror;
2067 int rc;
2069 DPRINTK("ENTER\n");
2071 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2072 return;
2074 /* obtain and analyze SError */
2075 rc = sata_scr_read(link, SCR_ERROR, &serror);
2076 if (rc == 0) {
2077 ehc->i.serror |= serror;
2078 ata_eh_analyze_serror(link);
2079 } else if (rc != -EOPNOTSUPP) {
2080 /* SError read failed, force reset and probing */
2081 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2082 ehc->i.action |= ATA_EH_RESET;
2083 ehc->i.err_mask |= AC_ERR_OTHER;
2086 /* analyze NCQ failure */
2087 ata_eh_analyze_ncq_error(link);
2089 /* any real error trumps AC_ERR_OTHER */
2090 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2091 ehc->i.err_mask &= ~AC_ERR_OTHER;
2093 all_err_mask |= ehc->i.err_mask;
2095 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2096 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2098 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2099 ata_dev_phys_link(qc->dev) != link)
2100 continue;
2102 /* inherit upper level err_mask */
2103 qc->err_mask |= ehc->i.err_mask;
2105 /* analyze TF */
2106 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2108 /* DEV errors are probably spurious in case of ATA_BUS error */
2109 if (qc->err_mask & AC_ERR_ATA_BUS)
2110 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2111 AC_ERR_INVALID);
2113 /* any real error trumps unknown error */
2114 if (qc->err_mask & ~AC_ERR_OTHER)
2115 qc->err_mask &= ~AC_ERR_OTHER;
2117 /* SENSE_VALID trumps dev/unknown error and revalidation */
2118 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2119 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2121 /* determine whether the command is worth retrying */
2122 if (qc->flags & ATA_QCFLAG_IO ||
2123 (!(qc->err_mask & AC_ERR_INVALID) &&
2124 qc->err_mask != AC_ERR_DEV))
2125 qc->flags |= ATA_QCFLAG_RETRY;
2127 /* accumulate error info */
2128 ehc->i.dev = qc->dev;
2129 all_err_mask |= qc->err_mask;
2130 if (qc->flags & ATA_QCFLAG_IO)
2131 eflags |= ATA_EFLAG_IS_IO;
2134 /* enforce default EH actions */
2135 if (ap->pflags & ATA_PFLAG_FROZEN ||
2136 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2137 ehc->i.action |= ATA_EH_RESET;
2138 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2139 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2140 ehc->i.action |= ATA_EH_REVALIDATE;
2142 /* If we have offending qcs and the associated failed device,
2143 * perform per-dev EH action only on the offending device.
2145 if (ehc->i.dev) {
2146 ehc->i.dev_action[ehc->i.dev->devno] |=
2147 ehc->i.action & ATA_EH_PERDEV_MASK;
2148 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2151 /* propagate timeout to host link */
2152 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2153 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2155 /* record error and consider speeding down */
2156 dev = ehc->i.dev;
2157 if (!dev && ((ata_link_max_devices(link) == 1 &&
2158 ata_dev_enabled(link->device))))
2159 dev = link->device;
2161 if (dev) {
2162 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2163 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2164 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2167 DPRINTK("EXIT\n");
2171 * ata_eh_autopsy - analyze error and determine recovery action
2172 * @ap: host port to perform autopsy on
2174 * Analyze all links of @ap and determine why they failed and
2175 * which recovery actions are needed.
2177 * LOCKING:
2178 * Kernel thread context (may sleep).
2180 void ata_eh_autopsy(struct ata_port *ap)
2182 struct ata_link *link;
2184 ata_for_each_link(link, ap, EDGE)
2185 ata_eh_link_autopsy(link);
2187 /* Handle the frigging slave link. Autopsy is done similarly
2188 * but actions and flags are transferred over to the master
2189 * link and handled from there.
2191 if (ap->slave_link) {
2192 struct ata_eh_context *mehc = &ap->link.eh_context;
2193 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2195 /* transfer control flags from master to slave */
2196 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2198 /* perform autopsy on the slave link */
2199 ata_eh_link_autopsy(ap->slave_link);
2201 /* transfer actions from slave to master and clear slave */
2202 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2203 mehc->i.action |= sehc->i.action;
2204 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2205 mehc->i.flags |= sehc->i.flags;
2206 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2209 /* Autopsy of fanout ports can affect host link autopsy.
2210 * Perform host link autopsy last.
2212 if (sata_pmp_attached(ap))
2213 ata_eh_link_autopsy(&ap->link);
2217 * ata_get_cmd_descript - get description for ATA command
2218 * @command: ATA command code to get description for
2220 * Return a textual description of the given command, or NULL if the
2221 * command is not known.
2223 * LOCKING:
2224 * None
2226 const char *ata_get_cmd_descript(u8 command)
2228 #ifdef CONFIG_ATA_VERBOSE_ERROR
2229 static const struct
2231 u8 command;
2232 const char *text;
2233 } cmd_descr[] = {
2234 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2235 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2236 { ATA_CMD_STANDBY, "STANDBY" },
2237 { ATA_CMD_IDLE, "IDLE" },
2238 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2239 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2240 { ATA_CMD_NOP, "NOP" },
2241 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2242 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2243 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2244 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2245 { ATA_CMD_SERVICE, "SERVICE" },
2246 { ATA_CMD_READ, "READ DMA" },
2247 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2248 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2249 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2250 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2251 { ATA_CMD_WRITE, "WRITE DMA" },
2252 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2253 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2254 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2255 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2256 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2257 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2258 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2259 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2260 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2261 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2262 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2263 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2264 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2265 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2266 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2267 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2268 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2269 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2270 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2271 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2272 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2273 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2274 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2275 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2276 { ATA_CMD_SLEEP, "SLEEP" },
2277 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2278 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2279 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2280 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2281 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2282 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2283 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2284 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2285 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2286 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2287 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2288 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2289 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2290 { ATA_CMD_PMP_READ, "READ BUFFER" },
2291 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2292 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2293 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2294 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2295 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2296 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2297 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2298 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2299 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2300 { ATA_CMD_SMART, "SMART" },
2301 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2302 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2303 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2304 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2305 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2306 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2307 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2308 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2309 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2310 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2311 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2312 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2313 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2314 { ATA_CMD_RESTORE, "RECALIBRATE" },
2315 { 0, NULL } /* terminate list */
2318 unsigned int i;
2319 for (i = 0; cmd_descr[i].text; i++)
2320 if (cmd_descr[i].command == command)
2321 return cmd_descr[i].text;
2322 #endif
2324 return NULL;
2328 * ata_eh_link_report - report error handling to user
2329 * @link: ATA link EH is going on
2331 * Report EH to user.
2333 * LOCKING:
2334 * None.
2336 static void ata_eh_link_report(struct ata_link *link)
2338 struct ata_port *ap = link->ap;
2339 struct ata_eh_context *ehc = &link->eh_context;
2340 const char *frozen, *desc;
2341 char tries_buf[6];
2342 int tag, nr_failed = 0;
2344 if (ehc->i.flags & ATA_EHI_QUIET)
2345 return;
2347 desc = NULL;
2348 if (ehc->i.desc[0] != '\0')
2349 desc = ehc->i.desc;
2351 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2352 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2354 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2355 ata_dev_phys_link(qc->dev) != link ||
2356 ((qc->flags & ATA_QCFLAG_QUIET) &&
2357 qc->err_mask == AC_ERR_DEV))
2358 continue;
2359 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2360 continue;
2362 nr_failed++;
2365 if (!nr_failed && !ehc->i.err_mask)
2366 return;
2368 frozen = "";
2369 if (ap->pflags & ATA_PFLAG_FROZEN)
2370 frozen = " frozen";
2372 memset(tries_buf, 0, sizeof(tries_buf));
2373 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2374 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2375 ap->eh_tries);
2377 if (ehc->i.dev) {
2378 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2379 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2380 ehc->i.err_mask, link->sactive, ehc->i.serror,
2381 ehc->i.action, frozen, tries_buf);
2382 if (desc)
2383 ata_dev_err(ehc->i.dev, "%s\n", desc);
2384 } else {
2385 ata_link_err(link, "exception Emask 0x%x "
2386 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2387 ehc->i.err_mask, link->sactive, ehc->i.serror,
2388 ehc->i.action, frozen, tries_buf);
2389 if (desc)
2390 ata_link_err(link, "%s\n", desc);
2393 #ifdef CONFIG_ATA_VERBOSE_ERROR
2394 if (ehc->i.serror)
2395 ata_link_err(link,
2396 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2397 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2398 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2399 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2400 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2401 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2402 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2403 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2404 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2405 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2406 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2407 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2408 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2409 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2410 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2411 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2412 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2413 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2414 #endif
2416 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2417 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2418 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2419 const u8 *cdb = qc->cdb;
2420 char data_buf[20] = "";
2421 char cdb_buf[70] = "";
2423 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2424 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2425 continue;
2427 if (qc->dma_dir != DMA_NONE) {
2428 static const char *dma_str[] = {
2429 [DMA_BIDIRECTIONAL] = "bidi",
2430 [DMA_TO_DEVICE] = "out",
2431 [DMA_FROM_DEVICE] = "in",
2433 static const char *prot_str[] = {
2434 [ATA_PROT_PIO] = "pio",
2435 [ATA_PROT_DMA] = "dma",
2436 [ATA_PROT_NCQ] = "ncq",
2437 [ATAPI_PROT_PIO] = "pio",
2438 [ATAPI_PROT_DMA] = "dma",
2441 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2442 prot_str[qc->tf.protocol], qc->nbytes,
2443 dma_str[qc->dma_dir]);
2446 if (ata_is_atapi(qc->tf.protocol)) {
2447 if (qc->scsicmd)
2448 scsi_print_command(qc->scsicmd);
2449 else
2450 snprintf(cdb_buf, sizeof(cdb_buf),
2451 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2452 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2453 cdb[0], cdb[1], cdb[2], cdb[3],
2454 cdb[4], cdb[5], cdb[6], cdb[7],
2455 cdb[8], cdb[9], cdb[10], cdb[11],
2456 cdb[12], cdb[13], cdb[14], cdb[15]);
2457 } else {
2458 const char *descr = ata_get_cmd_descript(cmd->command);
2459 if (descr)
2460 ata_dev_err(qc->dev, "failed command: %s\n",
2461 descr);
2464 ata_dev_err(qc->dev,
2465 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2466 "tag %d%s\n %s"
2467 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2468 "Emask 0x%x (%s)%s\n",
2469 cmd->command, cmd->feature, cmd->nsect,
2470 cmd->lbal, cmd->lbam, cmd->lbah,
2471 cmd->hob_feature, cmd->hob_nsect,
2472 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2473 cmd->device, qc->tag, data_buf, cdb_buf,
2474 res->command, res->feature, res->nsect,
2475 res->lbal, res->lbam, res->lbah,
2476 res->hob_feature, res->hob_nsect,
2477 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2478 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2479 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2481 #ifdef CONFIG_ATA_VERBOSE_ERROR
2482 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2483 ATA_ERR)) {
2484 if (res->command & ATA_BUSY)
2485 ata_dev_err(qc->dev, "status: { Busy }\n");
2486 else
2487 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
2488 res->command & ATA_DRDY ? "DRDY " : "",
2489 res->command & ATA_DF ? "DF " : "",
2490 res->command & ATA_DRQ ? "DRQ " : "",
2491 res->command & ATA_ERR ? "ERR " : "");
2494 if (cmd->command != ATA_CMD_PACKET &&
2495 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2496 ATA_ABORTED)))
2497 ata_dev_err(qc->dev, "error: { %s%s%s%s}\n",
2498 res->feature & ATA_ICRC ? "ICRC " : "",
2499 res->feature & ATA_UNC ? "UNC " : "",
2500 res->feature & ATA_IDNF ? "IDNF " : "",
2501 res->feature & ATA_ABORTED ? "ABRT " : "");
2502 #endif
2507 * ata_eh_report - report error handling to user
2508 * @ap: ATA port to report EH about
2510 * Report EH to user.
2512 * LOCKING:
2513 * None.
2515 void ata_eh_report(struct ata_port *ap)
2517 struct ata_link *link;
2519 ata_for_each_link(link, ap, HOST_FIRST)
2520 ata_eh_link_report(link);
2523 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2524 unsigned int *classes, unsigned long deadline,
2525 bool clear_classes)
2527 struct ata_device *dev;
2529 if (clear_classes)
2530 ata_for_each_dev(dev, link, ALL)
2531 classes[dev->devno] = ATA_DEV_UNKNOWN;
2533 return reset(link, classes, deadline);
2536 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2538 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2539 return 0;
2540 if (rc == -EAGAIN)
2541 return 1;
2542 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2543 return 1;
2544 return 0;
2547 int ata_eh_reset(struct ata_link *link, int classify,
2548 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2549 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2551 struct ata_port *ap = link->ap;
2552 struct ata_link *slave = ap->slave_link;
2553 struct ata_eh_context *ehc = &link->eh_context;
2554 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2555 unsigned int *classes = ehc->classes;
2556 unsigned int lflags = link->flags;
2557 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2558 int max_tries = 0, try = 0;
2559 struct ata_link *failed_link;
2560 struct ata_device *dev;
2561 unsigned long deadline, now;
2562 ata_reset_fn_t reset;
2563 unsigned long flags;
2564 u32 sstatus;
2565 int nr_unknown, rc;
2568 * Prepare to reset
2570 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2571 max_tries++;
2572 if (link->flags & ATA_LFLAG_NO_HRST)
2573 hardreset = NULL;
2574 if (link->flags & ATA_LFLAG_NO_SRST)
2575 softreset = NULL;
2577 /* make sure each reset attempt is at least COOL_DOWN apart */
2578 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2579 now = jiffies;
2580 WARN_ON(time_after(ehc->last_reset, now));
2581 deadline = ata_deadline(ehc->last_reset,
2582 ATA_EH_RESET_COOL_DOWN);
2583 if (time_before(now, deadline))
2584 schedule_timeout_uninterruptible(deadline - now);
2587 spin_lock_irqsave(ap->lock, flags);
2588 ap->pflags |= ATA_PFLAG_RESETTING;
2589 spin_unlock_irqrestore(ap->lock, flags);
2591 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2593 ata_for_each_dev(dev, link, ALL) {
2594 /* If we issue an SRST then an ATA drive (not ATAPI)
2595 * may change configuration and be in PIO0 timing. If
2596 * we do a hard reset (or are coming from power on)
2597 * this is true for ATA or ATAPI. Until we've set a
2598 * suitable controller mode we should not touch the
2599 * bus as we may be talking too fast.
2601 dev->pio_mode = XFER_PIO_0;
2603 /* If the controller has a pio mode setup function
2604 * then use it to set the chipset to rights. Don't
2605 * touch the DMA setup as that will be dealt with when
2606 * configuring devices.
2608 if (ap->ops->set_piomode)
2609 ap->ops->set_piomode(ap, dev);
2612 /* prefer hardreset */
2613 reset = NULL;
2614 ehc->i.action &= ~ATA_EH_RESET;
2615 if (hardreset) {
2616 reset = hardreset;
2617 ehc->i.action |= ATA_EH_HARDRESET;
2618 } else if (softreset) {
2619 reset = softreset;
2620 ehc->i.action |= ATA_EH_SOFTRESET;
2623 if (prereset) {
2624 unsigned long deadline = ata_deadline(jiffies,
2625 ATA_EH_PRERESET_TIMEOUT);
2627 if (slave) {
2628 sehc->i.action &= ~ATA_EH_RESET;
2629 sehc->i.action |= ehc->i.action;
2632 rc = prereset(link, deadline);
2634 /* If present, do prereset on slave link too. Reset
2635 * is skipped iff both master and slave links report
2636 * -ENOENT or clear ATA_EH_RESET.
2638 if (slave && (rc == 0 || rc == -ENOENT)) {
2639 int tmp;
2641 tmp = prereset(slave, deadline);
2642 if (tmp != -ENOENT)
2643 rc = tmp;
2645 ehc->i.action |= sehc->i.action;
2648 if (rc) {
2649 if (rc == -ENOENT) {
2650 ata_link_dbg(link, "port disabled--ignoring\n");
2651 ehc->i.action &= ~ATA_EH_RESET;
2653 ata_for_each_dev(dev, link, ALL)
2654 classes[dev->devno] = ATA_DEV_NONE;
2656 rc = 0;
2657 } else
2658 ata_link_err(link,
2659 "prereset failed (errno=%d)\n",
2660 rc);
2661 goto out;
2664 /* prereset() might have cleared ATA_EH_RESET. If so,
2665 * bang classes, thaw and return.
2667 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2668 ata_for_each_dev(dev, link, ALL)
2669 classes[dev->devno] = ATA_DEV_NONE;
2670 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2671 ata_is_host_link(link))
2672 ata_eh_thaw_port(ap);
2673 rc = 0;
2674 goto out;
2678 retry:
2680 * Perform reset
2682 if (ata_is_host_link(link))
2683 ata_eh_freeze_port(ap);
2685 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2687 if (reset) {
2688 if (verbose)
2689 ata_link_info(link, "%s resetting link\n",
2690 reset == softreset ? "soft" : "hard");
2692 /* mark that this EH session started with reset */
2693 ehc->last_reset = jiffies;
2694 if (reset == hardreset)
2695 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2696 else
2697 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2699 rc = ata_do_reset(link, reset, classes, deadline, true);
2700 if (rc && rc != -EAGAIN) {
2701 failed_link = link;
2702 goto fail;
2705 /* hardreset slave link if existent */
2706 if (slave && reset == hardreset) {
2707 int tmp;
2709 if (verbose)
2710 ata_link_info(slave, "hard resetting link\n");
2712 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2713 tmp = ata_do_reset(slave, reset, classes, deadline,
2714 false);
2715 switch (tmp) {
2716 case -EAGAIN:
2717 rc = -EAGAIN;
2718 case 0:
2719 break;
2720 default:
2721 failed_link = slave;
2722 rc = tmp;
2723 goto fail;
2727 /* perform follow-up SRST if necessary */
2728 if (reset == hardreset &&
2729 ata_eh_followup_srst_needed(link, rc)) {
2730 reset = softreset;
2732 if (!reset) {
2733 ata_link_err(link,
2734 "follow-up softreset required but no softreset available\n");
2735 failed_link = link;
2736 rc = -EINVAL;
2737 goto fail;
2740 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2741 rc = ata_do_reset(link, reset, classes, deadline, true);
2742 if (rc) {
2743 failed_link = link;
2744 goto fail;
2747 } else {
2748 if (verbose)
2749 ata_link_info(link,
2750 "no reset method available, skipping reset\n");
2751 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2752 lflags |= ATA_LFLAG_ASSUME_ATA;
2756 * Post-reset processing
2758 ata_for_each_dev(dev, link, ALL) {
2759 /* After the reset, the device state is PIO 0 and the
2760 * controller state is undefined. Reset also wakes up
2761 * drives from sleeping mode.
2763 dev->pio_mode = XFER_PIO_0;
2764 dev->flags &= ~ATA_DFLAG_SLEEPING;
2766 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2767 continue;
2769 /* apply class override */
2770 if (lflags & ATA_LFLAG_ASSUME_ATA)
2771 classes[dev->devno] = ATA_DEV_ATA;
2772 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2773 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2776 /* record current link speed */
2777 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2778 link->sata_spd = (sstatus >> 4) & 0xf;
2779 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2780 slave->sata_spd = (sstatus >> 4) & 0xf;
2782 /* thaw the port */
2783 if (ata_is_host_link(link))
2784 ata_eh_thaw_port(ap);
2786 /* postreset() should clear hardware SError. Although SError
2787 * is cleared during link resume, clearing SError here is
2788 * necessary as some PHYs raise hotplug events after SRST.
2789 * This introduces race condition where hotplug occurs between
2790 * reset and here. This race is mediated by cross checking
2791 * link onlineness and classification result later.
2793 if (postreset) {
2794 postreset(link, classes);
2795 if (slave)
2796 postreset(slave, classes);
2800 * Some controllers can't be frozen very well and may set spurious
2801 * error conditions during reset. Clear accumulated error
2802 * information and re-thaw the port if frozen. As reset is the
2803 * final recovery action and we cross check link onlineness against
2804 * device classification later, no hotplug event is lost by this.
2806 spin_lock_irqsave(link->ap->lock, flags);
2807 memset(&link->eh_info, 0, sizeof(link->eh_info));
2808 if (slave)
2809 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2810 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2811 spin_unlock_irqrestore(link->ap->lock, flags);
2813 if (ap->pflags & ATA_PFLAG_FROZEN)
2814 ata_eh_thaw_port(ap);
2817 * Make sure onlineness and classification result correspond.
2818 * Hotplug could have happened during reset and some
2819 * controllers fail to wait while a drive is spinning up after
2820 * being hotplugged causing misdetection. By cross checking
2821 * link on/offlineness and classification result, those
2822 * conditions can be reliably detected and retried.
2824 nr_unknown = 0;
2825 ata_for_each_dev(dev, link, ALL) {
2826 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2827 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2828 ata_dev_dbg(dev, "link online but device misclassified\n");
2829 classes[dev->devno] = ATA_DEV_NONE;
2830 nr_unknown++;
2832 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2833 if (ata_class_enabled(classes[dev->devno]))
2834 ata_dev_dbg(dev,
2835 "link offline, clearing class %d to NONE\n",
2836 classes[dev->devno]);
2837 classes[dev->devno] = ATA_DEV_NONE;
2838 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2839 ata_dev_dbg(dev,
2840 "link status unknown, clearing UNKNOWN to NONE\n");
2841 classes[dev->devno] = ATA_DEV_NONE;
2845 if (classify && nr_unknown) {
2846 if (try < max_tries) {
2847 ata_link_warn(link,
2848 "link online but %d devices misclassified, retrying\n",
2849 nr_unknown);
2850 failed_link = link;
2851 rc = -EAGAIN;
2852 goto fail;
2854 ata_link_warn(link,
2855 "link online but %d devices misclassified, "
2856 "device detection might fail\n", nr_unknown);
2859 /* reset successful, schedule revalidation */
2860 ata_eh_done(link, NULL, ATA_EH_RESET);
2861 if (slave)
2862 ata_eh_done(slave, NULL, ATA_EH_RESET);
2863 ehc->last_reset = jiffies; /* update to completion time */
2864 ehc->i.action |= ATA_EH_REVALIDATE;
2865 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2867 rc = 0;
2868 out:
2869 /* clear hotplug flag */
2870 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2871 if (slave)
2872 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2874 spin_lock_irqsave(ap->lock, flags);
2875 ap->pflags &= ~ATA_PFLAG_RESETTING;
2876 spin_unlock_irqrestore(ap->lock, flags);
2878 return rc;
2880 fail:
2881 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2882 if (!ata_is_host_link(link) &&
2883 sata_scr_read(link, SCR_STATUS, &sstatus))
2884 rc = -ERESTART;
2886 if (try >= max_tries) {
2888 * Thaw host port even if reset failed, so that the port
2889 * can be retried on the next phy event. This risks
2890 * repeated EH runs but seems to be a better tradeoff than
2891 * shutting down a port after a botched hotplug attempt.
2893 if (ata_is_host_link(link))
2894 ata_eh_thaw_port(ap);
2895 goto out;
2898 now = jiffies;
2899 if (time_before(now, deadline)) {
2900 unsigned long delta = deadline - now;
2902 ata_link_warn(failed_link,
2903 "reset failed (errno=%d), retrying in %u secs\n",
2904 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2906 ata_eh_release(ap);
2907 while (delta)
2908 delta = schedule_timeout_uninterruptible(delta);
2909 ata_eh_acquire(ap);
2913 * While disks spinup behind PMP, some controllers fail sending SRST.
2914 * They need to be reset - as well as the PMP - before retrying.
2916 if (rc == -ERESTART) {
2917 if (ata_is_host_link(link))
2918 ata_eh_thaw_port(ap);
2919 goto out;
2922 if (try == max_tries - 1) {
2923 sata_down_spd_limit(link, 0);
2924 if (slave)
2925 sata_down_spd_limit(slave, 0);
2926 } else if (rc == -EPIPE)
2927 sata_down_spd_limit(failed_link, 0);
2929 if (hardreset)
2930 reset = hardreset;
2931 goto retry;
2934 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2936 struct ata_link *link;
2937 struct ata_device *dev;
2938 unsigned long flags;
2941 * This function can be thought of as an extended version of
2942 * ata_eh_about_to_do() specially crafted to accommodate the
2943 * requirements of ATA_EH_PARK handling. Since the EH thread
2944 * does not leave the do {} while () loop in ata_eh_recover as
2945 * long as the timeout for a park request to *one* device on
2946 * the port has not expired, and since we still want to pick
2947 * up park requests to other devices on the same port or
2948 * timeout updates for the same device, we have to pull
2949 * ATA_EH_PARK actions from eh_info into eh_context.i
2950 * ourselves at the beginning of each pass over the loop.
2952 * Additionally, all write accesses to &ap->park_req_pending
2953 * through INIT_COMPLETION() (see below) or complete_all()
2954 * (see ata_scsi_park_store()) are protected by the host lock.
2955 * As a result we have that park_req_pending.done is zero on
2956 * exit from this function, i.e. when ATA_EH_PARK actions for
2957 * *all* devices on port ap have been pulled into the
2958 * respective eh_context structs. If, and only if,
2959 * park_req_pending.done is non-zero by the time we reach
2960 * wait_for_completion_timeout(), another ATA_EH_PARK action
2961 * has been scheduled for at least one of the devices on port
2962 * ap and we have to cycle over the do {} while () loop in
2963 * ata_eh_recover() again.
2966 spin_lock_irqsave(ap->lock, flags);
2967 INIT_COMPLETION(ap->park_req_pending);
2968 ata_for_each_link(link, ap, EDGE) {
2969 ata_for_each_dev(dev, link, ALL) {
2970 struct ata_eh_info *ehi = &link->eh_info;
2972 link->eh_context.i.dev_action[dev->devno] |=
2973 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2974 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2977 spin_unlock_irqrestore(ap->lock, flags);
2980 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2982 struct ata_eh_context *ehc = &dev->link->eh_context;
2983 struct ata_taskfile tf;
2984 unsigned int err_mask;
2986 ata_tf_init(dev, &tf);
2987 if (park) {
2988 ehc->unloaded_mask |= 1 << dev->devno;
2989 tf.command = ATA_CMD_IDLEIMMEDIATE;
2990 tf.feature = 0x44;
2991 tf.lbal = 0x4c;
2992 tf.lbam = 0x4e;
2993 tf.lbah = 0x55;
2994 } else {
2995 ehc->unloaded_mask &= ~(1 << dev->devno);
2996 tf.command = ATA_CMD_CHK_POWER;
2999 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3000 tf.protocol |= ATA_PROT_NODATA;
3001 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3002 if (park && (err_mask || tf.lbal != 0xc4)) {
3003 ata_dev_err(dev, "head unload failed!\n");
3004 ehc->unloaded_mask &= ~(1 << dev->devno);
3008 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3009 struct ata_device **r_failed_dev)
3011 struct ata_port *ap = link->ap;
3012 struct ata_eh_context *ehc = &link->eh_context;
3013 struct ata_device *dev;
3014 unsigned int new_mask = 0;
3015 unsigned long flags;
3016 int rc = 0;
3018 DPRINTK("ENTER\n");
3020 /* For PATA drive side cable detection to work, IDENTIFY must
3021 * be done backwards such that PDIAG- is released by the slave
3022 * device before the master device is identified.
3024 ata_for_each_dev(dev, link, ALL_REVERSE) {
3025 unsigned int action = ata_eh_dev_action(dev);
3026 unsigned int readid_flags = 0;
3028 if (ehc->i.flags & ATA_EHI_DID_RESET)
3029 readid_flags |= ATA_READID_POSTRESET;
3031 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3032 WARN_ON(dev->class == ATA_DEV_PMP);
3034 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3035 rc = -EIO;
3036 goto err;
3039 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3040 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3041 readid_flags);
3042 if (rc)
3043 goto err;
3045 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3047 /* Configuration may have changed, reconfigure
3048 * transfer mode.
3050 ehc->i.flags |= ATA_EHI_SETMODE;
3052 /* schedule the scsi_rescan_device() here */
3053 schedule_work(&(ap->scsi_rescan_task));
3054 } else if (dev->class == ATA_DEV_UNKNOWN &&
3055 ehc->tries[dev->devno] &&
3056 ata_class_enabled(ehc->classes[dev->devno])) {
3057 /* Temporarily set dev->class, it will be
3058 * permanently set once all configurations are
3059 * complete. This is necessary because new
3060 * device configuration is done in two
3061 * separate loops.
3063 dev->class = ehc->classes[dev->devno];
3065 if (dev->class == ATA_DEV_PMP)
3066 rc = sata_pmp_attach(dev);
3067 else
3068 rc = ata_dev_read_id(dev, &dev->class,
3069 readid_flags, dev->id);
3071 /* read_id might have changed class, store and reset */
3072 ehc->classes[dev->devno] = dev->class;
3073 dev->class = ATA_DEV_UNKNOWN;
3075 switch (rc) {
3076 case 0:
3077 /* clear error info accumulated during probe */
3078 ata_ering_clear(&dev->ering);
3079 new_mask |= 1 << dev->devno;
3080 break;
3081 case -ENOENT:
3082 /* IDENTIFY was issued to non-existent
3083 * device. No need to reset. Just
3084 * thaw and ignore the device.
3086 ata_eh_thaw_port(ap);
3087 break;
3088 default:
3089 goto err;
3094 /* PDIAG- should have been released, ask cable type if post-reset */
3095 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3096 if (ap->ops->cable_detect)
3097 ap->cbl = ap->ops->cable_detect(ap);
3098 ata_force_cbl(ap);
3101 /* Configure new devices forward such that user doesn't see
3102 * device detection messages backwards.
3104 ata_for_each_dev(dev, link, ALL) {
3105 if (!(new_mask & (1 << dev->devno)))
3106 continue;
3108 dev->class = ehc->classes[dev->devno];
3110 if (dev->class == ATA_DEV_PMP)
3111 continue;
3113 ehc->i.flags |= ATA_EHI_PRINTINFO;
3114 rc = ata_dev_configure(dev);
3115 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3116 if (rc) {
3117 dev->class = ATA_DEV_UNKNOWN;
3118 goto err;
3121 spin_lock_irqsave(ap->lock, flags);
3122 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3123 spin_unlock_irqrestore(ap->lock, flags);
3125 /* new device discovered, configure xfermode */
3126 ehc->i.flags |= ATA_EHI_SETMODE;
3129 return 0;
3131 err:
3132 *r_failed_dev = dev;
3133 DPRINTK("EXIT rc=%d\n", rc);
3134 return rc;
3138 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3139 * @link: link on which timings will be programmed
3140 * @r_failed_dev: out parameter for failed device
3142 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3143 * ata_set_mode() fails, pointer to the failing device is
3144 * returned in @r_failed_dev.
3146 * LOCKING:
3147 * PCI/etc. bus probe sem.
3149 * RETURNS:
3150 * 0 on success, negative errno otherwise
3152 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3154 struct ata_port *ap = link->ap;
3155 struct ata_device *dev;
3156 int rc;
3158 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3159 ata_for_each_dev(dev, link, ENABLED) {
3160 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3161 struct ata_ering_entry *ent;
3163 ent = ata_ering_top(&dev->ering);
3164 if (ent)
3165 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3169 /* has private set_mode? */
3170 if (ap->ops->set_mode)
3171 rc = ap->ops->set_mode(link, r_failed_dev);
3172 else
3173 rc = ata_do_set_mode(link, r_failed_dev);
3175 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3176 ata_for_each_dev(dev, link, ENABLED) {
3177 struct ata_eh_context *ehc = &link->eh_context;
3178 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3179 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3181 if (dev->xfer_mode != saved_xfer_mode ||
3182 ata_ncq_enabled(dev) != saved_ncq)
3183 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3186 return rc;
3190 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3191 * @dev: ATAPI device to clear UA for
3193 * Resets and other operations can make an ATAPI device raise
3194 * UNIT ATTENTION which causes the next operation to fail. This
3195 * function clears UA.
3197 * LOCKING:
3198 * EH context (may sleep).
3200 * RETURNS:
3201 * 0 on success, -errno on failure.
3203 static int atapi_eh_clear_ua(struct ata_device *dev)
3205 int i;
3207 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3208 u8 *sense_buffer = dev->link->ap->sector_buf;
3209 u8 sense_key = 0;
3210 unsigned int err_mask;
3212 err_mask = atapi_eh_tur(dev, &sense_key);
3213 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3214 ata_dev_warn(dev,
3215 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3216 err_mask);
3217 return -EIO;
3220 if (!err_mask || sense_key != UNIT_ATTENTION)
3221 return 0;
3223 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3224 if (err_mask) {
3225 ata_dev_warn(dev, "failed to clear "
3226 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3227 return -EIO;
3231 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3232 ATA_EH_UA_TRIES);
3234 return 0;
3238 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3239 * @dev: ATA device which may need FLUSH retry
3241 * If @dev failed FLUSH, it needs to be reported upper layer
3242 * immediately as it means that @dev failed to remap and already
3243 * lost at least a sector and further FLUSH retrials won't make
3244 * any difference to the lost sector. However, if FLUSH failed
3245 * for other reasons, for example transmission error, FLUSH needs
3246 * to be retried.
3248 * This function determines whether FLUSH failure retry is
3249 * necessary and performs it if so.
3251 * RETURNS:
3252 * 0 if EH can continue, -errno if EH needs to be repeated.
3254 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3256 struct ata_link *link = dev->link;
3257 struct ata_port *ap = link->ap;
3258 struct ata_queued_cmd *qc;
3259 struct ata_taskfile tf;
3260 unsigned int err_mask;
3261 int rc = 0;
3263 /* did flush fail for this device? */
3264 if (!ata_tag_valid(link->active_tag))
3265 return 0;
3267 qc = __ata_qc_from_tag(ap, link->active_tag);
3268 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3269 qc->tf.command != ATA_CMD_FLUSH))
3270 return 0;
3272 /* if the device failed it, it should be reported to upper layers */
3273 if (qc->err_mask & AC_ERR_DEV)
3274 return 0;
3276 /* flush failed for some other reason, give it another shot */
3277 ata_tf_init(dev, &tf);
3279 tf.command = qc->tf.command;
3280 tf.flags |= ATA_TFLAG_DEVICE;
3281 tf.protocol = ATA_PROT_NODATA;
3283 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3284 tf.command, qc->err_mask);
3286 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3287 if (!err_mask) {
3289 * FLUSH is complete but there's no way to
3290 * successfully complete a failed command from EH.
3291 * Making sure retry is allowed at least once and
3292 * retrying it should do the trick - whatever was in
3293 * the cache is already on the platter and this won't
3294 * cause infinite loop.
3296 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3297 } else {
3298 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3299 err_mask);
3300 rc = -EIO;
3302 /* if device failed it, report it to upper layers */
3303 if (err_mask & AC_ERR_DEV) {
3304 qc->err_mask |= AC_ERR_DEV;
3305 qc->result_tf = tf;
3306 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3307 rc = 0;
3310 return rc;
3314 * ata_eh_set_lpm - configure SATA interface power management
3315 * @link: link to configure power management
3316 * @policy: the link power management policy
3317 * @r_failed_dev: out parameter for failed device
3319 * Enable SATA Interface power management. This will enable
3320 * Device Interface Power Management (DIPM) for min_power
3321 * policy, and then call driver specific callbacks for
3322 * enabling Host Initiated Power management.
3324 * LOCKING:
3325 * EH context.
3327 * RETURNS:
3328 * 0 on success, -errno on failure.
3330 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3331 struct ata_device **r_failed_dev)
3333 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3334 struct ata_eh_context *ehc = &link->eh_context;
3335 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3336 enum ata_lpm_policy old_policy = link->lpm_policy;
3337 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3338 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3339 unsigned int err_mask;
3340 int rc;
3342 /* if the link or host doesn't do LPM, noop */
3343 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3344 return 0;
3347 * DIPM is enabled only for MIN_POWER as some devices
3348 * misbehave when the host NACKs transition to SLUMBER. Order
3349 * device and link configurations such that the host always
3350 * allows DIPM requests.
3352 ata_for_each_dev(dev, link, ENABLED) {
3353 bool hipm = ata_id_has_hipm(dev->id);
3354 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3356 /* find the first enabled and LPM enabled devices */
3357 if (!link_dev)
3358 link_dev = dev;
3360 if (!lpm_dev && (hipm || dipm))
3361 lpm_dev = dev;
3363 hints &= ~ATA_LPM_EMPTY;
3364 if (!hipm)
3365 hints &= ~ATA_LPM_HIPM;
3367 /* disable DIPM before changing link config */
3368 if (policy != ATA_LPM_MIN_POWER && dipm) {
3369 err_mask = ata_dev_set_feature(dev,
3370 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3371 if (err_mask && err_mask != AC_ERR_DEV) {
3372 ata_dev_warn(dev,
3373 "failed to disable DIPM, Emask 0x%x\n",
3374 err_mask);
3375 rc = -EIO;
3376 goto fail;
3381 if (ap) {
3382 rc = ap->ops->set_lpm(link, policy, hints);
3383 if (!rc && ap->slave_link)
3384 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3385 } else
3386 rc = sata_pmp_set_lpm(link, policy, hints);
3389 * Attribute link config failure to the first (LPM) enabled
3390 * device on the link.
3392 if (rc) {
3393 if (rc == -EOPNOTSUPP) {
3394 link->flags |= ATA_LFLAG_NO_LPM;
3395 return 0;
3397 dev = lpm_dev ? lpm_dev : link_dev;
3398 goto fail;
3402 * Low level driver acked the transition. Issue DIPM command
3403 * with the new policy set.
3405 link->lpm_policy = policy;
3406 if (ap && ap->slave_link)
3407 ap->slave_link->lpm_policy = policy;
3409 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3410 ata_for_each_dev(dev, link, ENABLED) {
3411 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3412 ata_id_has_dipm(dev->id)) {
3413 err_mask = ata_dev_set_feature(dev,
3414 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3415 if (err_mask && err_mask != AC_ERR_DEV) {
3416 ata_dev_warn(dev,
3417 "failed to enable DIPM, Emask 0x%x\n",
3418 err_mask);
3419 rc = -EIO;
3420 goto fail;
3425 return 0;
3427 fail:
3428 /* restore the old policy */
3429 link->lpm_policy = old_policy;
3430 if (ap && ap->slave_link)
3431 ap->slave_link->lpm_policy = old_policy;
3433 /* if no device or only one more chance is left, disable LPM */
3434 if (!dev || ehc->tries[dev->devno] <= 2) {
3435 ata_link_warn(link, "disabling LPM on the link\n");
3436 link->flags |= ATA_LFLAG_NO_LPM;
3438 if (r_failed_dev)
3439 *r_failed_dev = dev;
3440 return rc;
3443 int ata_link_nr_enabled(struct ata_link *link)
3445 struct ata_device *dev;
3446 int cnt = 0;
3448 ata_for_each_dev(dev, link, ENABLED)
3449 cnt++;
3450 return cnt;
3453 static int ata_link_nr_vacant(struct ata_link *link)
3455 struct ata_device *dev;
3456 int cnt = 0;
3458 ata_for_each_dev(dev, link, ALL)
3459 if (dev->class == ATA_DEV_UNKNOWN)
3460 cnt++;
3461 return cnt;
3464 static int ata_eh_skip_recovery(struct ata_link *link)
3466 struct ata_port *ap = link->ap;
3467 struct ata_eh_context *ehc = &link->eh_context;
3468 struct ata_device *dev;
3470 /* skip disabled links */
3471 if (link->flags & ATA_LFLAG_DISABLED)
3472 return 1;
3474 /* skip if explicitly requested */
3475 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3476 return 1;
3478 /* thaw frozen port and recover failed devices */
3479 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3480 return 0;
3482 /* reset at least once if reset is requested */
3483 if ((ehc->i.action & ATA_EH_RESET) &&
3484 !(ehc->i.flags & ATA_EHI_DID_RESET))
3485 return 0;
3487 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3488 ata_for_each_dev(dev, link, ALL) {
3489 if (dev->class == ATA_DEV_UNKNOWN &&
3490 ehc->classes[dev->devno] != ATA_DEV_NONE)
3491 return 0;
3494 return 1;
3497 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3499 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3500 u64 now = get_jiffies_64();
3501 int *trials = void_arg;
3503 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3504 (ent->timestamp < now - min(now, interval)))
3505 return -1;
3507 (*trials)++;
3508 return 0;
3511 static int ata_eh_schedule_probe(struct ata_device *dev)
3513 struct ata_eh_context *ehc = &dev->link->eh_context;
3514 struct ata_link *link = ata_dev_phys_link(dev);
3515 int trials = 0;
3517 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3518 (ehc->did_probe_mask & (1 << dev->devno)))
3519 return 0;
3521 ata_eh_detach_dev(dev);
3522 ata_dev_init(dev);
3523 ehc->did_probe_mask |= (1 << dev->devno);
3524 ehc->i.action |= ATA_EH_RESET;
3525 ehc->saved_xfer_mode[dev->devno] = 0;
3526 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3528 /* the link maybe in a deep sleep, wake it up */
3529 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3530 if (ata_is_host_link(link))
3531 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3532 ATA_LPM_EMPTY);
3533 else
3534 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3535 ATA_LPM_EMPTY);
3538 /* Record and count probe trials on the ering. The specific
3539 * error mask used is irrelevant. Because a successful device
3540 * detection clears the ering, this count accumulates only if
3541 * there are consecutive failed probes.
3543 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3544 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3545 * forced to 1.5Gbps.
3547 * This is to work around cases where failed link speed
3548 * negotiation results in device misdetection leading to
3549 * infinite DEVXCHG or PHRDY CHG events.
3551 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3552 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3554 if (trials > ATA_EH_PROBE_TRIALS)
3555 sata_down_spd_limit(link, 1);
3557 return 1;
3560 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3562 struct ata_eh_context *ehc = &dev->link->eh_context;
3564 /* -EAGAIN from EH routine indicates retry without prejudice.
3565 * The requester is responsible for ensuring forward progress.
3567 if (err != -EAGAIN)
3568 ehc->tries[dev->devno]--;
3570 switch (err) {
3571 case -ENODEV:
3572 /* device missing or wrong IDENTIFY data, schedule probing */
3573 ehc->i.probe_mask |= (1 << dev->devno);
3574 case -EINVAL:
3575 /* give it just one more chance */
3576 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3577 case -EIO:
3578 if (ehc->tries[dev->devno] == 1) {
3579 /* This is the last chance, better to slow
3580 * down than lose it.
3582 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3583 if (dev->pio_mode > XFER_PIO_0)
3584 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3588 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3589 /* disable device if it has used up all its chances */
3590 ata_dev_disable(dev);
3592 /* detach if offline */
3593 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3594 ata_eh_detach_dev(dev);
3596 /* schedule probe if necessary */
3597 if (ata_eh_schedule_probe(dev)) {
3598 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3599 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3600 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3603 return 1;
3604 } else {
3605 ehc->i.action |= ATA_EH_RESET;
3606 return 0;
3611 * ata_eh_recover - recover host port after error
3612 * @ap: host port to recover
3613 * @prereset: prereset method (can be NULL)
3614 * @softreset: softreset method (can be NULL)
3615 * @hardreset: hardreset method (can be NULL)
3616 * @postreset: postreset method (can be NULL)
3617 * @r_failed_link: out parameter for failed link
3619 * This is the alpha and omega, eum and yang, heart and soul of
3620 * libata exception handling. On entry, actions required to
3621 * recover each link and hotplug requests are recorded in the
3622 * link's eh_context. This function executes all the operations
3623 * with appropriate retrials and fallbacks to resurrect failed
3624 * devices, detach goners and greet newcomers.
3626 * LOCKING:
3627 * Kernel thread context (may sleep).
3629 * RETURNS:
3630 * 0 on success, -errno on failure.
3632 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3633 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3634 ata_postreset_fn_t postreset,
3635 struct ata_link **r_failed_link)
3637 struct ata_link *link;
3638 struct ata_device *dev;
3639 int rc, nr_fails;
3640 unsigned long flags, deadline;
3642 DPRINTK("ENTER\n");
3644 /* prep for recovery */
3645 ata_for_each_link(link, ap, EDGE) {
3646 struct ata_eh_context *ehc = &link->eh_context;
3648 /* re-enable link? */
3649 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3650 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3651 spin_lock_irqsave(ap->lock, flags);
3652 link->flags &= ~ATA_LFLAG_DISABLED;
3653 spin_unlock_irqrestore(ap->lock, flags);
3654 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3657 ata_for_each_dev(dev, link, ALL) {
3658 if (link->flags & ATA_LFLAG_NO_RETRY)
3659 ehc->tries[dev->devno] = 1;
3660 else
3661 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3663 /* collect port action mask recorded in dev actions */
3664 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3665 ~ATA_EH_PERDEV_MASK;
3666 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3668 /* process hotplug request */
3669 if (dev->flags & ATA_DFLAG_DETACH)
3670 ata_eh_detach_dev(dev);
3672 /* schedule probe if necessary */
3673 if (!ata_dev_enabled(dev))
3674 ata_eh_schedule_probe(dev);
3678 retry:
3679 rc = 0;
3681 /* if UNLOADING, finish immediately */
3682 if (ap->pflags & ATA_PFLAG_UNLOADING)
3683 goto out;
3685 /* prep for EH */
3686 ata_for_each_link(link, ap, EDGE) {
3687 struct ata_eh_context *ehc = &link->eh_context;
3689 /* skip EH if possible. */
3690 if (ata_eh_skip_recovery(link))
3691 ehc->i.action = 0;
3693 ata_for_each_dev(dev, link, ALL)
3694 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3697 /* reset */
3698 ata_for_each_link(link, ap, EDGE) {
3699 struct ata_eh_context *ehc = &link->eh_context;
3701 if (!(ehc->i.action & ATA_EH_RESET))
3702 continue;
3704 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3705 prereset, softreset, hardreset, postreset);
3706 if (rc) {
3707 ata_link_err(link, "reset failed, giving up\n");
3708 goto out;
3712 do {
3713 unsigned long now;
3716 * clears ATA_EH_PARK in eh_info and resets
3717 * ap->park_req_pending
3719 ata_eh_pull_park_action(ap);
3721 deadline = jiffies;
3722 ata_for_each_link(link, ap, EDGE) {
3723 ata_for_each_dev(dev, link, ALL) {
3724 struct ata_eh_context *ehc = &link->eh_context;
3725 unsigned long tmp;
3727 if (dev->class != ATA_DEV_ATA)
3728 continue;
3729 if (!(ehc->i.dev_action[dev->devno] &
3730 ATA_EH_PARK))
3731 continue;
3732 tmp = dev->unpark_deadline;
3733 if (time_before(deadline, tmp))
3734 deadline = tmp;
3735 else if (time_before_eq(tmp, jiffies))
3736 continue;
3737 if (ehc->unloaded_mask & (1 << dev->devno))
3738 continue;
3740 ata_eh_park_issue_cmd(dev, 1);
3744 now = jiffies;
3745 if (time_before_eq(deadline, now))
3746 break;
3748 ata_eh_release(ap);
3749 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3750 deadline - now);
3751 ata_eh_acquire(ap);
3752 } while (deadline);
3753 ata_for_each_link(link, ap, EDGE) {
3754 ata_for_each_dev(dev, link, ALL) {
3755 if (!(link->eh_context.unloaded_mask &
3756 (1 << dev->devno)))
3757 continue;
3759 ata_eh_park_issue_cmd(dev, 0);
3760 ata_eh_done(link, dev, ATA_EH_PARK);
3764 /* the rest */
3765 nr_fails = 0;
3766 ata_for_each_link(link, ap, PMP_FIRST) {
3767 struct ata_eh_context *ehc = &link->eh_context;
3769 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3770 goto config_lpm;
3772 /* revalidate existing devices and attach new ones */
3773 rc = ata_eh_revalidate_and_attach(link, &dev);
3774 if (rc)
3775 goto rest_fail;
3777 /* if PMP got attached, return, pmp EH will take care of it */
3778 if (link->device->class == ATA_DEV_PMP) {
3779 ehc->i.action = 0;
3780 return 0;
3783 /* configure transfer mode if necessary */
3784 if (ehc->i.flags & ATA_EHI_SETMODE) {
3785 rc = ata_set_mode(link, &dev);
3786 if (rc)
3787 goto rest_fail;
3788 ehc->i.flags &= ~ATA_EHI_SETMODE;
3791 /* If reset has been issued, clear UA to avoid
3792 * disrupting the current users of the device.
3794 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3795 ata_for_each_dev(dev, link, ALL) {
3796 if (dev->class != ATA_DEV_ATAPI)
3797 continue;
3798 rc = atapi_eh_clear_ua(dev);
3799 if (rc)
3800 goto rest_fail;
3804 /* retry flush if necessary */
3805 ata_for_each_dev(dev, link, ALL) {
3806 if (dev->class != ATA_DEV_ATA)
3807 continue;
3808 rc = ata_eh_maybe_retry_flush(dev);
3809 if (rc)
3810 goto rest_fail;
3813 config_lpm:
3814 /* configure link power saving */
3815 if (link->lpm_policy != ap->target_lpm_policy) {
3816 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3817 if (rc)
3818 goto rest_fail;
3821 /* this link is okay now */
3822 ehc->i.flags = 0;
3823 continue;
3825 rest_fail:
3826 nr_fails++;
3827 if (dev)
3828 ata_eh_handle_dev_fail(dev, rc);
3830 if (ap->pflags & ATA_PFLAG_FROZEN) {
3831 /* PMP reset requires working host port.
3832 * Can't retry if it's frozen.
3834 if (sata_pmp_attached(ap))
3835 goto out;
3836 break;
3840 if (nr_fails)
3841 goto retry;
3843 out:
3844 if (rc && r_failed_link)
3845 *r_failed_link = link;
3847 DPRINTK("EXIT, rc=%d\n", rc);
3848 return rc;
3852 * ata_eh_finish - finish up EH
3853 * @ap: host port to finish EH for
3855 * Recovery is complete. Clean up EH states and retry or finish
3856 * failed qcs.
3858 * LOCKING:
3859 * None.
3861 void ata_eh_finish(struct ata_port *ap)
3863 int tag;
3865 /* retry or finish qcs */
3866 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3867 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3869 if (!(qc->flags & ATA_QCFLAG_FAILED))
3870 continue;
3872 if (qc->err_mask) {
3873 /* FIXME: Once EH migration is complete,
3874 * generate sense data in this function,
3875 * considering both err_mask and tf.
3877 if (qc->flags & ATA_QCFLAG_RETRY)
3878 ata_eh_qc_retry(qc);
3879 else
3880 ata_eh_qc_complete(qc);
3881 } else {
3882 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3883 ata_eh_qc_complete(qc);
3884 } else {
3885 /* feed zero TF to sense generation */
3886 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3887 ata_eh_qc_retry(qc);
3892 /* make sure nr_active_links is zero after EH */
3893 WARN_ON(ap->nr_active_links);
3894 ap->nr_active_links = 0;
3898 * ata_do_eh - do standard error handling
3899 * @ap: host port to handle error for
3901 * @prereset: prereset method (can be NULL)
3902 * @softreset: softreset method (can be NULL)
3903 * @hardreset: hardreset method (can be NULL)
3904 * @postreset: postreset method (can be NULL)
3906 * Perform standard error handling sequence.
3908 * LOCKING:
3909 * Kernel thread context (may sleep).
3911 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3912 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3913 ata_postreset_fn_t postreset)
3915 struct ata_device *dev;
3916 int rc;
3918 ata_eh_autopsy(ap);
3919 ata_eh_report(ap);
3921 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3922 NULL);
3923 if (rc) {
3924 ata_for_each_dev(dev, &ap->link, ALL)
3925 ata_dev_disable(dev);
3928 ata_eh_finish(ap);
3932 * ata_std_error_handler - standard error handler
3933 * @ap: host port to handle error for
3935 * Standard error handler
3937 * LOCKING:
3938 * Kernel thread context (may sleep).
3940 void ata_std_error_handler(struct ata_port *ap)
3942 struct ata_port_operations *ops = ap->ops;
3943 ata_reset_fn_t hardreset = ops->hardreset;
3945 /* ignore built-in hardreset if SCR access is not available */
3946 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3947 hardreset = NULL;
3949 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3952 #ifdef CONFIG_PM
3954 * ata_eh_handle_port_suspend - perform port suspend operation
3955 * @ap: port to suspend
3957 * Suspend @ap.
3959 * LOCKING:
3960 * Kernel thread context (may sleep).
3962 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3964 unsigned long flags;
3965 int rc = 0;
3967 /* are we suspending? */
3968 spin_lock_irqsave(ap->lock, flags);
3969 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3970 ap->pm_mesg.event == PM_EVENT_ON) {
3971 spin_unlock_irqrestore(ap->lock, flags);
3972 return;
3974 spin_unlock_irqrestore(ap->lock, flags);
3976 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3978 /* tell ACPI we're suspending */
3979 rc = ata_acpi_on_suspend(ap);
3980 if (rc)
3981 goto out;
3983 /* suspend */
3984 ata_eh_freeze_port(ap);
3986 if (ap->ops->port_suspend)
3987 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3989 ata_acpi_set_state(ap, PMSG_SUSPEND);
3990 out:
3991 /* report result */
3992 spin_lock_irqsave(ap->lock, flags);
3994 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3995 if (rc == 0)
3996 ap->pflags |= ATA_PFLAG_SUSPENDED;
3997 else if (ap->pflags & ATA_PFLAG_FROZEN)
3998 ata_port_schedule_eh(ap);
4000 if (ap->pm_result) {
4001 *ap->pm_result = rc;
4002 ap->pm_result = NULL;
4005 spin_unlock_irqrestore(ap->lock, flags);
4007 return;
4011 * ata_eh_handle_port_resume - perform port resume operation
4012 * @ap: port to resume
4014 * Resume @ap.
4016 * LOCKING:
4017 * Kernel thread context (may sleep).
4019 static void ata_eh_handle_port_resume(struct ata_port *ap)
4021 struct ata_link *link;
4022 struct ata_device *dev;
4023 unsigned long flags;
4024 int rc = 0;
4026 /* are we resuming? */
4027 spin_lock_irqsave(ap->lock, flags);
4028 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4029 ap->pm_mesg.event != PM_EVENT_ON) {
4030 spin_unlock_irqrestore(ap->lock, flags);
4031 return;
4033 spin_unlock_irqrestore(ap->lock, flags);
4035 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4038 * Error timestamps are in jiffies which doesn't run while
4039 * suspended and PHY events during resume isn't too uncommon.
4040 * When the two are combined, it can lead to unnecessary speed
4041 * downs if the machine is suspended and resumed repeatedly.
4042 * Clear error history.
4044 ata_for_each_link(link, ap, HOST_FIRST)
4045 ata_for_each_dev(dev, link, ALL)
4046 ata_ering_clear(&dev->ering);
4048 ata_acpi_set_state(ap, PMSG_ON);
4050 if (ap->ops->port_resume)
4051 rc = ap->ops->port_resume(ap);
4053 /* tell ACPI that we're resuming */
4054 ata_acpi_on_resume(ap);
4056 /* report result */
4057 spin_lock_irqsave(ap->lock, flags);
4058 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4059 if (ap->pm_result) {
4060 *ap->pm_result = rc;
4061 ap->pm_result = NULL;
4063 spin_unlock_irqrestore(ap->lock, flags);
4065 #endif /* CONFIG_PM */