Revert "libata: Implement support for sense data reporting"
[deliverable/linux.git] / drivers / ata / libata-eh.c
1 /*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
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.
15 *
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.
20 *
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.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
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"
46
47 #include <linux/libata.h>
48
49 #include <trace/events/libata.h>
50 #include "libata.h"
51
52 enum {
53 /* speed down verdicts */
54 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
55 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
56 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
57 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
58
59 /* error flags */
60 ATA_EFLAG_IS_IO = (1 << 0),
61 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
62 ATA_EFLAG_OLD_ER = (1 << 31),
63
64 /* error categories */
65 ATA_ECAT_NONE = 0,
66 ATA_ECAT_ATA_BUS = 1,
67 ATA_ECAT_TOUT_HSM = 2,
68 ATA_ECAT_UNK_DEV = 3,
69 ATA_ECAT_DUBIOUS_NONE = 4,
70 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
71 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
72 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
73 ATA_ECAT_NR = 8,
74
75 ATA_EH_CMD_DFL_TIMEOUT = 5000,
76
77 /* always put at least this amount of time between resets */
78 ATA_EH_RESET_COOL_DOWN = 5000,
79
80 /* Waiting in ->prereset can never be reliable. It's
81 * sometimes nice to wait there but it can't be depended upon;
82 * otherwise, we wouldn't be resetting. Just give it enough
83 * time for most drives to spin up.
84 */
85 ATA_EH_PRERESET_TIMEOUT = 10000,
86 ATA_EH_FASTDRAIN_INTERVAL = 3000,
87
88 ATA_EH_UA_TRIES = 5,
89
90 /* probe speed down parameters, see ata_eh_schedule_probe() */
91 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
92 ATA_EH_PROBE_TRIALS = 2,
93 };
94
95 /* The following table determines how we sequence resets. Each entry
96 * represents timeout for that try. The first try can be soft or
97 * hardreset. All others are hardreset if available. In most cases
98 * the first reset w/ 10sec timeout should succeed. Following entries
99 * are mostly for error handling, hotplug and those outlier devices that
100 * take an exceptionally long time to recover from reset.
101 */
102 static const unsigned long ata_eh_reset_timeouts[] = {
103 10000, /* most drives spin up by 10sec */
104 10000, /* > 99% working drives spin up before 20sec */
105 35000, /* give > 30 secs of idleness for outlier devices */
106 5000, /* and sweet one last chance */
107 ULONG_MAX, /* > 1 min has elapsed, give up */
108 };
109
110 static const unsigned long ata_eh_identify_timeouts[] = {
111 5000, /* covers > 99% of successes and not too boring on failures */
112 10000, /* combined time till here is enough even for media access */
113 30000, /* for true idiots */
114 ULONG_MAX,
115 };
116
117 static const unsigned long ata_eh_flush_timeouts[] = {
118 15000, /* be generous with flush */
119 15000, /* ditto */
120 30000, /* and even more generous */
121 ULONG_MAX,
122 };
123
124 static const unsigned long ata_eh_other_timeouts[] = {
125 5000, /* same rationale as identify timeout */
126 10000, /* ditto */
127 /* but no merciful 30sec for other commands, it just isn't worth it */
128 ULONG_MAX,
129 };
130
131 struct ata_eh_cmd_timeout_ent {
132 const u8 *commands;
133 const unsigned long *timeouts;
134 };
135
136 /* The following table determines timeouts to use for EH internal
137 * commands. Each table entry is a command class and matches the
138 * commands the entry applies to and the timeout table to use.
139 *
140 * On the retry after a command timed out, the next timeout value from
141 * the table is used. If the table doesn't contain further entries,
142 * the last value is used.
143 *
144 * ehc->cmd_timeout_idx keeps track of which timeout to use per
145 * command class, so if SET_FEATURES times out on the first try, the
146 * next try will use the second timeout value only for that class.
147 */
148 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
149 static const struct ata_eh_cmd_timeout_ent
150 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
151 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
152 .timeouts = ata_eh_identify_timeouts, },
153 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_SET_FEATURES),
158 .timeouts = ata_eh_other_timeouts, },
159 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
160 .timeouts = ata_eh_other_timeouts, },
161 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
162 .timeouts = ata_eh_flush_timeouts },
163 };
164 #undef CMDS
165
166 static void __ata_port_freeze(struct ata_port *ap);
167 #ifdef CONFIG_PM
168 static void ata_eh_handle_port_suspend(struct ata_port *ap);
169 static void ata_eh_handle_port_resume(struct ata_port *ap);
170 #else /* CONFIG_PM */
171 static void ata_eh_handle_port_suspend(struct ata_port *ap)
172 { }
173
174 static void ata_eh_handle_port_resume(struct ata_port *ap)
175 { }
176 #endif /* CONFIG_PM */
177
178 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
179 va_list args)
180 {
181 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
182 ATA_EH_DESC_LEN - ehi->desc_len,
183 fmt, args);
184 }
185
186 /**
187 * __ata_ehi_push_desc - push error description without adding separator
188 * @ehi: target EHI
189 * @fmt: printf format string
190 *
191 * Format string according to @fmt and append it to @ehi->desc.
192 *
193 * LOCKING:
194 * spin_lock_irqsave(host lock)
195 */
196 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
197 {
198 va_list args;
199
200 va_start(args, fmt);
201 __ata_ehi_pushv_desc(ehi, fmt, args);
202 va_end(args);
203 }
204
205 /**
206 * ata_ehi_push_desc - push error description with separator
207 * @ehi: target EHI
208 * @fmt: printf format string
209 *
210 * Format string according to @fmt and append it to @ehi->desc.
211 * If @ehi->desc is not empty, ", " is added in-between.
212 *
213 * LOCKING:
214 * spin_lock_irqsave(host lock)
215 */
216 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
217 {
218 va_list args;
219
220 if (ehi->desc_len)
221 __ata_ehi_push_desc(ehi, ", ");
222
223 va_start(args, fmt);
224 __ata_ehi_pushv_desc(ehi, fmt, args);
225 va_end(args);
226 }
227
228 /**
229 * ata_ehi_clear_desc - clean error description
230 * @ehi: target EHI
231 *
232 * Clear @ehi->desc.
233 *
234 * LOCKING:
235 * spin_lock_irqsave(host lock)
236 */
237 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
238 {
239 ehi->desc[0] = '\0';
240 ehi->desc_len = 0;
241 }
242
243 /**
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
247 *
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
252 *
253 * LOCKING:
254 * None.
255 */
256 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
257 {
258 va_list args;
259
260 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
261
262 if (ap->link.eh_info.desc_len)
263 __ata_ehi_push_desc(&ap->link.eh_info, " ");
264
265 va_start(args, fmt);
266 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
267 va_end(args);
268 }
269
270 #ifdef CONFIG_PCI
271
272 /**
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
278 *
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
283 *
284 * LOCKING:
285 * None.
286 */
287 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
288 const char *name)
289 {
290 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
291 char *type = "";
292 unsigned long long start, len;
293
294 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
295 type = "m";
296 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
297 type = "i";
298
299 start = (unsigned long long)pci_resource_start(pdev, bar);
300 len = (unsigned long long)pci_resource_len(pdev, bar);
301
302 if (offset < 0)
303 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
304 else
305 ata_port_desc(ap, "%s 0x%llx", name,
306 start + (unsigned long long)offset);
307 }
308
309 #endif /* CONFIG_PCI */
310
311 static int ata_lookup_timeout_table(u8 cmd)
312 {
313 int i;
314
315 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
316 const u8 *cur;
317
318 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
319 if (*cur == cmd)
320 return i;
321 }
322
323 return -1;
324 }
325
326 /**
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
330 *
331 * Determine timeout for internal command @cmd for @dev.
332 *
333 * LOCKING:
334 * EH context.
335 *
336 * RETURNS:
337 * Determined timeout.
338 */
339 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
340 {
341 struct ata_eh_context *ehc = &dev->link->eh_context;
342 int ent = ata_lookup_timeout_table(cmd);
343 int idx;
344
345 if (ent < 0)
346 return ATA_EH_CMD_DFL_TIMEOUT;
347
348 idx = ehc->cmd_timeout_idx[dev->devno][ent];
349 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
350 }
351
352 /**
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
356 *
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
360 *
361 * LOCKING:
362 * EH context.
363 */
364 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
365 {
366 struct ata_eh_context *ehc = &dev->link->eh_context;
367 int ent = ata_lookup_timeout_table(cmd);
368 int idx;
369
370 if (ent < 0)
371 return;
372
373 idx = ehc->cmd_timeout_idx[dev->devno][ent];
374 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
375 ehc->cmd_timeout_idx[dev->devno][ent]++;
376 }
377
378 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
379 unsigned int err_mask)
380 {
381 struct ata_ering_entry *ent;
382
383 WARN_ON(!err_mask);
384
385 ering->cursor++;
386 ering->cursor %= ATA_ERING_SIZE;
387
388 ent = &ering->ring[ering->cursor];
389 ent->eflags = eflags;
390 ent->err_mask = err_mask;
391 ent->timestamp = get_jiffies_64();
392 }
393
394 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
395 {
396 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
397
398 if (ent->err_mask)
399 return ent;
400 return NULL;
401 }
402
403 int ata_ering_map(struct ata_ering *ering,
404 int (*map_fn)(struct ata_ering_entry *, void *),
405 void *arg)
406 {
407 int idx, rc = 0;
408 struct ata_ering_entry *ent;
409
410 idx = ering->cursor;
411 do {
412 ent = &ering->ring[idx];
413 if (!ent->err_mask)
414 break;
415 rc = map_fn(ent, arg);
416 if (rc)
417 break;
418 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
419 } while (idx != ering->cursor);
420
421 return rc;
422 }
423
424 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
425 {
426 ent->eflags |= ATA_EFLAG_OLD_ER;
427 return 0;
428 }
429
430 static void ata_ering_clear(struct ata_ering *ering)
431 {
432 ata_ering_map(ering, ata_ering_clear_cb, NULL);
433 }
434
435 static unsigned int ata_eh_dev_action(struct ata_device *dev)
436 {
437 struct ata_eh_context *ehc = &dev->link->eh_context;
438
439 return ehc->i.action | ehc->i.dev_action[dev->devno];
440 }
441
442 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
443 struct ata_eh_info *ehi, unsigned int action)
444 {
445 struct ata_device *tdev;
446
447 if (!dev) {
448 ehi->action &= ~action;
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] &= ~action;
451 } else {
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
454
455 /* break ehi->action into ehi->dev_action */
456 if (ehi->action & action) {
457 ata_for_each_dev(tdev, link, ALL)
458 ehi->dev_action[tdev->devno] |=
459 ehi->action & action;
460 ehi->action &= ~action;
461 }
462
463 /* turn off the specified per-dev action */
464 ehi->dev_action[dev->devno] &= ~action;
465 }
466 }
467
468 /**
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
471 *
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
475 *
476 * LOCKING:
477 * EH context.
478 */
479 void ata_eh_acquire(struct ata_port *ap)
480 {
481 mutex_lock(&ap->host->eh_mutex);
482 WARN_ON_ONCE(ap->host->eh_owner);
483 ap->host->eh_owner = current;
484 }
485
486 /**
487 * ata_eh_release - release EH ownership
488 * @ap: ATA port to release EH ownership for
489 *
490 * Release EH ownership for @ap if the caller. The caller must
491 * have acquired EH ownership using ata_eh_acquire() previously.
492 *
493 * LOCKING:
494 * EH context.
495 */
496 void ata_eh_release(struct ata_port *ap)
497 {
498 WARN_ON_ONCE(ap->host->eh_owner != current);
499 ap->host->eh_owner = NULL;
500 mutex_unlock(&ap->host->eh_mutex);
501 }
502
503 /**
504 * ata_scsi_timed_out - SCSI layer time out callback
505 * @cmd: timed out SCSI command
506 *
507 * Handles SCSI layer timeout. We race with normal completion of
508 * the qc for @cmd. If the qc is already gone, we lose and let
509 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
510 * timed out and EH should be invoked. Prevent ata_qc_complete()
511 * from finishing it by setting EH_SCHEDULED and return
512 * EH_NOT_HANDLED.
513 *
514 * TODO: kill this function once old EH is gone.
515 *
516 * LOCKING:
517 * Called from timer context
518 *
519 * RETURNS:
520 * EH_HANDLED or EH_NOT_HANDLED
521 */
522 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
523 {
524 struct Scsi_Host *host = cmd->device->host;
525 struct ata_port *ap = ata_shost_to_port(host);
526 unsigned long flags;
527 struct ata_queued_cmd *qc;
528 enum blk_eh_timer_return ret;
529
530 DPRINTK("ENTER\n");
531
532 if (ap->ops->error_handler) {
533 ret = BLK_EH_NOT_HANDLED;
534 goto out;
535 }
536
537 ret = BLK_EH_HANDLED;
538 spin_lock_irqsave(ap->lock, flags);
539 qc = ata_qc_from_tag(ap, ap->link.active_tag);
540 if (qc) {
541 WARN_ON(qc->scsicmd != cmd);
542 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
543 qc->err_mask |= AC_ERR_TIMEOUT;
544 ret = BLK_EH_NOT_HANDLED;
545 }
546 spin_unlock_irqrestore(ap->lock, flags);
547
548 out:
549 DPRINTK("EXIT, ret=%d\n", ret);
550 return ret;
551 }
552
553 static void ata_eh_unload(struct ata_port *ap)
554 {
555 struct ata_link *link;
556 struct ata_device *dev;
557 unsigned long flags;
558
559 /* Restore SControl IPM and SPD for the next driver and
560 * disable attached devices.
561 */
562 ata_for_each_link(link, ap, PMP_FIRST) {
563 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
564 ata_for_each_dev(dev, link, ALL)
565 ata_dev_disable(dev);
566 }
567
568 /* freeze and set UNLOADED */
569 spin_lock_irqsave(ap->lock, flags);
570
571 ata_port_freeze(ap); /* won't be thawed */
572 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
573 ap->pflags |= ATA_PFLAG_UNLOADED;
574
575 spin_unlock_irqrestore(ap->lock, flags);
576 }
577
578 /**
579 * ata_scsi_error - SCSI layer error handler callback
580 * @host: SCSI host on which error occurred
581 *
582 * Handles SCSI-layer-thrown error events.
583 *
584 * LOCKING:
585 * Inherited from SCSI layer (none, can sleep)
586 *
587 * RETURNS:
588 * Zero.
589 */
590 void ata_scsi_error(struct Scsi_Host *host)
591 {
592 struct ata_port *ap = ata_shost_to_port(host);
593 unsigned long flags;
594 LIST_HEAD(eh_work_q);
595
596 DPRINTK("ENTER\n");
597
598 spin_lock_irqsave(host->host_lock, flags);
599 list_splice_init(&host->eh_cmd_q, &eh_work_q);
600 spin_unlock_irqrestore(host->host_lock, flags);
601
602 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
603
604 /* If we timed raced normal completion and there is nothing to
605 recover nr_timedout == 0 why exactly are we doing error recovery ? */
606 ata_scsi_port_error_handler(host, ap);
607
608 /* finish or retry handled scmd's and clean up */
609 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
610
611 DPRINTK("EXIT\n");
612 }
613
614 /**
615 * ata_scsi_cmd_error_handler - error callback for a list of commands
616 * @host: scsi host containing the port
617 * @ap: ATA port within the host
618 * @eh_work_q: list of commands to process
619 *
620 * process the given list of commands and return those finished to the
621 * ap->eh_done_q. This function is the first part of the libata error
622 * handler which processes a given list of failed commands.
623 */
624 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
625 struct list_head *eh_work_q)
626 {
627 int i;
628 unsigned long flags;
629
630 /* make sure sff pio task is not running */
631 ata_sff_flush_pio_task(ap);
632
633 /* synchronize with host lock and sort out timeouts */
634
635 /* For new EH, all qcs are finished in one of three ways -
636 * normal completion, error completion, and SCSI timeout.
637 * Both completions can race against SCSI timeout. When normal
638 * completion wins, the qc never reaches EH. When error
639 * completion wins, the qc has ATA_QCFLAG_FAILED set.
640 *
641 * When SCSI timeout wins, things are a bit more complex.
642 * Normal or error completion can occur after the timeout but
643 * before this point. In such cases, both types of
644 * completions are honored. A scmd is determined to have
645 * timed out iff its associated qc is active and not failed.
646 */
647 if (ap->ops->error_handler) {
648 struct scsi_cmnd *scmd, *tmp;
649 int nr_timedout = 0;
650
651 spin_lock_irqsave(ap->lock, flags);
652
653 /* This must occur under the ap->lock as we don't want
654 a polled recovery to race the real interrupt handler
655
656 The lost_interrupt handler checks for any completed but
657 non-notified command and completes much like an IRQ handler.
658
659 We then fall into the error recovery code which will treat
660 this as if normal completion won the race */
661
662 if (ap->ops->lost_interrupt)
663 ap->ops->lost_interrupt(ap);
664
665 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
666 struct ata_queued_cmd *qc;
667
668 for (i = 0; i < ATA_MAX_QUEUE; i++) {
669 qc = __ata_qc_from_tag(ap, i);
670 if (qc->flags & ATA_QCFLAG_ACTIVE &&
671 qc->scsicmd == scmd)
672 break;
673 }
674
675 if (i < ATA_MAX_QUEUE) {
676 /* the scmd has an associated qc */
677 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
678 /* which hasn't failed yet, timeout */
679 qc->err_mask |= AC_ERR_TIMEOUT;
680 qc->flags |= ATA_QCFLAG_FAILED;
681 nr_timedout++;
682 }
683 } else {
684 /* Normal completion occurred after
685 * SCSI timeout but before this point.
686 * Successfully complete it.
687 */
688 scmd->retries = scmd->allowed;
689 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
690 }
691 }
692
693 /* If we have timed out qcs. They belong to EH from
694 * this point but the state of the controller is
695 * unknown. Freeze the port to make sure the IRQ
696 * handler doesn't diddle with those qcs. This must
697 * be done atomically w.r.t. setting QCFLAG_FAILED.
698 */
699 if (nr_timedout)
700 __ata_port_freeze(ap);
701
702 spin_unlock_irqrestore(ap->lock, flags);
703
704 /* initialize eh_tries */
705 ap->eh_tries = ATA_EH_MAX_TRIES;
706 } else
707 spin_unlock_wait(ap->lock);
708
709 }
710 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
711
712 /**
713 * ata_scsi_port_error_handler - recover the port after the commands
714 * @host: SCSI host containing the port
715 * @ap: the ATA port
716 *
717 * Handle the recovery of the port @ap after all the commands
718 * have been recovered.
719 */
720 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
721 {
722 unsigned long flags;
723
724 /* invoke error handler */
725 if (ap->ops->error_handler) {
726 struct ata_link *link;
727
728 /* acquire EH ownership */
729 ata_eh_acquire(ap);
730 repeat:
731 /* kill fast drain timer */
732 del_timer_sync(&ap->fastdrain_timer);
733
734 /* process port resume request */
735 ata_eh_handle_port_resume(ap);
736
737 /* fetch & clear EH info */
738 spin_lock_irqsave(ap->lock, flags);
739
740 ata_for_each_link(link, ap, HOST_FIRST) {
741 struct ata_eh_context *ehc = &link->eh_context;
742 struct ata_device *dev;
743
744 memset(&link->eh_context, 0, sizeof(link->eh_context));
745 link->eh_context.i = link->eh_info;
746 memset(&link->eh_info, 0, sizeof(link->eh_info));
747
748 ata_for_each_dev(dev, link, ENABLED) {
749 int devno = dev->devno;
750
751 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
752 if (ata_ncq_enabled(dev))
753 ehc->saved_ncq_enabled |= 1 << devno;
754 }
755 }
756
757 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
758 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
759 ap->excl_link = NULL; /* don't maintain exclusion over EH */
760
761 spin_unlock_irqrestore(ap->lock, flags);
762
763 /* invoke EH, skip if unloading or suspended */
764 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
765 ap->ops->error_handler(ap);
766 else {
767 /* if unloading, commence suicide */
768 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
769 !(ap->pflags & ATA_PFLAG_UNLOADED))
770 ata_eh_unload(ap);
771 ata_eh_finish(ap);
772 }
773
774 /* process port suspend request */
775 ata_eh_handle_port_suspend(ap);
776
777 /* Exception might have happened after ->error_handler
778 * recovered the port but before this point. Repeat
779 * EH in such case.
780 */
781 spin_lock_irqsave(ap->lock, flags);
782
783 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
784 if (--ap->eh_tries) {
785 spin_unlock_irqrestore(ap->lock, flags);
786 goto repeat;
787 }
788 ata_port_err(ap,
789 "EH pending after %d tries, giving up\n",
790 ATA_EH_MAX_TRIES);
791 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
792 }
793
794 /* this run is complete, make sure EH info is clear */
795 ata_for_each_link(link, ap, HOST_FIRST)
796 memset(&link->eh_info, 0, sizeof(link->eh_info));
797
798 /* end eh (clear host_eh_scheduled) while holding
799 * ap->lock such that if exception occurs after this
800 * point but before EH completion, SCSI midlayer will
801 * re-initiate EH.
802 */
803 ap->ops->end_eh(ap);
804
805 spin_unlock_irqrestore(ap->lock, flags);
806 ata_eh_release(ap);
807 } else {
808 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
809 ap->ops->eng_timeout(ap);
810 }
811
812 scsi_eh_flush_done_q(&ap->eh_done_q);
813
814 /* clean up */
815 spin_lock_irqsave(ap->lock, flags);
816
817 if (ap->pflags & ATA_PFLAG_LOADING)
818 ap->pflags &= ~ATA_PFLAG_LOADING;
819 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
820 schedule_delayed_work(&ap->hotplug_task, 0);
821
822 if (ap->pflags & ATA_PFLAG_RECOVERED)
823 ata_port_info(ap, "EH complete\n");
824
825 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
826
827 /* tell wait_eh that we're done */
828 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
829 wake_up_all(&ap->eh_wait_q);
830
831 spin_unlock_irqrestore(ap->lock, flags);
832 }
833 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
834
835 /**
836 * ata_port_wait_eh - Wait for the currently pending EH to complete
837 * @ap: Port to wait EH for
838 *
839 * Wait until the currently pending EH is complete.
840 *
841 * LOCKING:
842 * Kernel thread context (may sleep).
843 */
844 void ata_port_wait_eh(struct ata_port *ap)
845 {
846 unsigned long flags;
847 DEFINE_WAIT(wait);
848
849 retry:
850 spin_lock_irqsave(ap->lock, flags);
851
852 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
853 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
854 spin_unlock_irqrestore(ap->lock, flags);
855 schedule();
856 spin_lock_irqsave(ap->lock, flags);
857 }
858 finish_wait(&ap->eh_wait_q, &wait);
859
860 spin_unlock_irqrestore(ap->lock, flags);
861
862 /* make sure SCSI EH is complete */
863 if (scsi_host_in_recovery(ap->scsi_host)) {
864 ata_msleep(ap, 10);
865 goto retry;
866 }
867 }
868 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
869
870 static int ata_eh_nr_in_flight(struct ata_port *ap)
871 {
872 unsigned int tag;
873 int nr = 0;
874
875 /* count only non-internal commands */
876 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
877 if (ata_qc_from_tag(ap, tag))
878 nr++;
879
880 return nr;
881 }
882
883 void ata_eh_fastdrain_timerfn(unsigned long arg)
884 {
885 struct ata_port *ap = (void *)arg;
886 unsigned long flags;
887 int cnt;
888
889 spin_lock_irqsave(ap->lock, flags);
890
891 cnt = ata_eh_nr_in_flight(ap);
892
893 /* are we done? */
894 if (!cnt)
895 goto out_unlock;
896
897 if (cnt == ap->fastdrain_cnt) {
898 unsigned int tag;
899
900 /* No progress during the last interval, tag all
901 * in-flight qcs as timed out and freeze the port.
902 */
903 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
904 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
905 if (qc)
906 qc->err_mask |= AC_ERR_TIMEOUT;
907 }
908
909 ata_port_freeze(ap);
910 } else {
911 /* some qcs have finished, give it another chance */
912 ap->fastdrain_cnt = cnt;
913 ap->fastdrain_timer.expires =
914 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
915 add_timer(&ap->fastdrain_timer);
916 }
917
918 out_unlock:
919 spin_unlock_irqrestore(ap->lock, flags);
920 }
921
922 /**
923 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
924 * @ap: target ATA port
925 * @fastdrain: activate fast drain
926 *
927 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
928 * is non-zero and EH wasn't pending before. Fast drain ensures
929 * that EH kicks in in timely manner.
930 *
931 * LOCKING:
932 * spin_lock_irqsave(host lock)
933 */
934 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
935 {
936 int cnt;
937
938 /* already scheduled? */
939 if (ap->pflags & ATA_PFLAG_EH_PENDING)
940 return;
941
942 ap->pflags |= ATA_PFLAG_EH_PENDING;
943
944 if (!fastdrain)
945 return;
946
947 /* do we have in-flight qcs? */
948 cnt = ata_eh_nr_in_flight(ap);
949 if (!cnt)
950 return;
951
952 /* activate fast drain */
953 ap->fastdrain_cnt = cnt;
954 ap->fastdrain_timer.expires =
955 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
956 add_timer(&ap->fastdrain_timer);
957 }
958
959 /**
960 * ata_qc_schedule_eh - schedule qc for error handling
961 * @qc: command to schedule error handling for
962 *
963 * Schedule error handling for @qc. EH will kick in as soon as
964 * other commands are drained.
965 *
966 * LOCKING:
967 * spin_lock_irqsave(host lock)
968 */
969 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
970 {
971 struct ata_port *ap = qc->ap;
972 struct request_queue *q = qc->scsicmd->device->request_queue;
973 unsigned long flags;
974
975 WARN_ON(!ap->ops->error_handler);
976
977 qc->flags |= ATA_QCFLAG_FAILED;
978 ata_eh_set_pending(ap, 1);
979
980 /* The following will fail if timeout has already expired.
981 * ata_scsi_error() takes care of such scmds on EH entry.
982 * Note that ATA_QCFLAG_FAILED is unconditionally set after
983 * this function completes.
984 */
985 spin_lock_irqsave(q->queue_lock, flags);
986 blk_abort_request(qc->scsicmd->request);
987 spin_unlock_irqrestore(q->queue_lock, flags);
988 }
989
990 /**
991 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
992 * @ap: ATA port to schedule EH for
993 *
994 * LOCKING: inherited from ata_port_schedule_eh
995 * spin_lock_irqsave(host lock)
996 */
997 void ata_std_sched_eh(struct ata_port *ap)
998 {
999 WARN_ON(!ap->ops->error_handler);
1000
1001 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1002 return;
1003
1004 ata_eh_set_pending(ap, 1);
1005 scsi_schedule_eh(ap->scsi_host);
1006
1007 DPRINTK("port EH scheduled\n");
1008 }
1009 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
1010
1011 /**
1012 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1013 * @ap: ATA port to end EH for
1014 *
1015 * In the libata object model there is a 1:1 mapping of ata_port to
1016 * shost, so host fields can be directly manipulated under ap->lock, in
1017 * the libsas case we need to hold a lock at the ha->level to coordinate
1018 * these events.
1019 *
1020 * LOCKING:
1021 * spin_lock_irqsave(host lock)
1022 */
1023 void ata_std_end_eh(struct ata_port *ap)
1024 {
1025 struct Scsi_Host *host = ap->scsi_host;
1026
1027 host->host_eh_scheduled = 0;
1028 }
1029 EXPORT_SYMBOL(ata_std_end_eh);
1030
1031
1032 /**
1033 * ata_port_schedule_eh - schedule error handling without a qc
1034 * @ap: ATA port to schedule EH for
1035 *
1036 * Schedule error handling for @ap. EH will kick in as soon as
1037 * all commands are drained.
1038 *
1039 * LOCKING:
1040 * spin_lock_irqsave(host lock)
1041 */
1042 void ata_port_schedule_eh(struct ata_port *ap)
1043 {
1044 /* see: ata_std_sched_eh, unless you know better */
1045 ap->ops->sched_eh(ap);
1046 }
1047
1048 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1049 {
1050 int tag, nr_aborted = 0;
1051
1052 WARN_ON(!ap->ops->error_handler);
1053
1054 /* we're gonna abort all commands, no need for fast drain */
1055 ata_eh_set_pending(ap, 0);
1056
1057 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1058 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1059
1060 if (qc && (!link || qc->dev->link == link)) {
1061 qc->flags |= ATA_QCFLAG_FAILED;
1062 ata_qc_complete(qc);
1063 nr_aborted++;
1064 }
1065 }
1066
1067 if (!nr_aborted)
1068 ata_port_schedule_eh(ap);
1069
1070 return nr_aborted;
1071 }
1072
1073 /**
1074 * ata_link_abort - abort all qc's on the link
1075 * @link: ATA link to abort qc's for
1076 *
1077 * Abort all active qc's active on @link and schedule EH.
1078 *
1079 * LOCKING:
1080 * spin_lock_irqsave(host lock)
1081 *
1082 * RETURNS:
1083 * Number of aborted qc's.
1084 */
1085 int ata_link_abort(struct ata_link *link)
1086 {
1087 return ata_do_link_abort(link->ap, link);
1088 }
1089
1090 /**
1091 * ata_port_abort - abort all qc's on the port
1092 * @ap: ATA port to abort qc's for
1093 *
1094 * Abort all active qc's of @ap and schedule EH.
1095 *
1096 * LOCKING:
1097 * spin_lock_irqsave(host_set lock)
1098 *
1099 * RETURNS:
1100 * Number of aborted qc's.
1101 */
1102 int ata_port_abort(struct ata_port *ap)
1103 {
1104 return ata_do_link_abort(ap, NULL);
1105 }
1106
1107 /**
1108 * __ata_port_freeze - freeze port
1109 * @ap: ATA port to freeze
1110 *
1111 * This function is called when HSM violation or some other
1112 * condition disrupts normal operation of the port. Frozen port
1113 * is not allowed to perform any operation until the port is
1114 * thawed, which usually follows a successful reset.
1115 *
1116 * ap->ops->freeze() callback can be used for freezing the port
1117 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1118 * port cannot be frozen hardware-wise, the interrupt handler
1119 * must ack and clear interrupts unconditionally while the port
1120 * is frozen.
1121 *
1122 * LOCKING:
1123 * spin_lock_irqsave(host lock)
1124 */
1125 static void __ata_port_freeze(struct ata_port *ap)
1126 {
1127 WARN_ON(!ap->ops->error_handler);
1128
1129 if (ap->ops->freeze)
1130 ap->ops->freeze(ap);
1131
1132 ap->pflags |= ATA_PFLAG_FROZEN;
1133
1134 DPRINTK("ata%u port frozen\n", ap->print_id);
1135 }
1136
1137 /**
1138 * ata_port_freeze - abort & freeze port
1139 * @ap: ATA port to freeze
1140 *
1141 * Abort and freeze @ap. The freeze operation must be called
1142 * first, because some hardware requires special operations
1143 * before the taskfile registers are accessible.
1144 *
1145 * LOCKING:
1146 * spin_lock_irqsave(host lock)
1147 *
1148 * RETURNS:
1149 * Number of aborted commands.
1150 */
1151 int ata_port_freeze(struct ata_port *ap)
1152 {
1153 int nr_aborted;
1154
1155 WARN_ON(!ap->ops->error_handler);
1156
1157 __ata_port_freeze(ap);
1158 nr_aborted = ata_port_abort(ap);
1159
1160 return nr_aborted;
1161 }
1162
1163 /**
1164 * sata_async_notification - SATA async notification handler
1165 * @ap: ATA port where async notification is received
1166 *
1167 * Handler to be called when async notification via SDB FIS is
1168 * received. This function schedules EH if necessary.
1169 *
1170 * LOCKING:
1171 * spin_lock_irqsave(host lock)
1172 *
1173 * RETURNS:
1174 * 1 if EH is scheduled, 0 otherwise.
1175 */
1176 int sata_async_notification(struct ata_port *ap)
1177 {
1178 u32 sntf;
1179 int rc;
1180
1181 if (!(ap->flags & ATA_FLAG_AN))
1182 return 0;
1183
1184 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1185 if (rc == 0)
1186 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1187
1188 if (!sata_pmp_attached(ap) || rc) {
1189 /* PMP is not attached or SNTF is not available */
1190 if (!sata_pmp_attached(ap)) {
1191 /* PMP is not attached. Check whether ATAPI
1192 * AN is configured. If so, notify media
1193 * change.
1194 */
1195 struct ata_device *dev = ap->link.device;
1196
1197 if ((dev->class == ATA_DEV_ATAPI) &&
1198 (dev->flags & ATA_DFLAG_AN))
1199 ata_scsi_media_change_notify(dev);
1200 return 0;
1201 } else {
1202 /* PMP is attached but SNTF is not available.
1203 * ATAPI async media change notification is
1204 * not used. The PMP must be reporting PHY
1205 * status change, schedule EH.
1206 */
1207 ata_port_schedule_eh(ap);
1208 return 1;
1209 }
1210 } else {
1211 /* PMP is attached and SNTF is available */
1212 struct ata_link *link;
1213
1214 /* check and notify ATAPI AN */
1215 ata_for_each_link(link, ap, EDGE) {
1216 if (!(sntf & (1 << link->pmp)))
1217 continue;
1218
1219 if ((link->device->class == ATA_DEV_ATAPI) &&
1220 (link->device->flags & ATA_DFLAG_AN))
1221 ata_scsi_media_change_notify(link->device);
1222 }
1223
1224 /* If PMP is reporting that PHY status of some
1225 * downstream ports has changed, schedule EH.
1226 */
1227 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1228 ata_port_schedule_eh(ap);
1229 return 1;
1230 }
1231
1232 return 0;
1233 }
1234 }
1235
1236 /**
1237 * ata_eh_freeze_port - EH helper to freeze port
1238 * @ap: ATA port to freeze
1239 *
1240 * Freeze @ap.
1241 *
1242 * LOCKING:
1243 * None.
1244 */
1245 void ata_eh_freeze_port(struct ata_port *ap)
1246 {
1247 unsigned long flags;
1248
1249 if (!ap->ops->error_handler)
1250 return;
1251
1252 spin_lock_irqsave(ap->lock, flags);
1253 __ata_port_freeze(ap);
1254 spin_unlock_irqrestore(ap->lock, flags);
1255 }
1256
1257 /**
1258 * ata_port_thaw_port - EH helper to thaw port
1259 * @ap: ATA port to thaw
1260 *
1261 * Thaw frozen port @ap.
1262 *
1263 * LOCKING:
1264 * None.
1265 */
1266 void ata_eh_thaw_port(struct ata_port *ap)
1267 {
1268 unsigned long flags;
1269
1270 if (!ap->ops->error_handler)
1271 return;
1272
1273 spin_lock_irqsave(ap->lock, flags);
1274
1275 ap->pflags &= ~ATA_PFLAG_FROZEN;
1276
1277 if (ap->ops->thaw)
1278 ap->ops->thaw(ap);
1279
1280 spin_unlock_irqrestore(ap->lock, flags);
1281
1282 DPRINTK("ata%u port thawed\n", ap->print_id);
1283 }
1284
1285 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1286 {
1287 /* nada */
1288 }
1289
1290 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1291 {
1292 struct ata_port *ap = qc->ap;
1293 struct scsi_cmnd *scmd = qc->scsicmd;
1294 unsigned long flags;
1295
1296 spin_lock_irqsave(ap->lock, flags);
1297 qc->scsidone = ata_eh_scsidone;
1298 __ata_qc_complete(qc);
1299 WARN_ON(ata_tag_valid(qc->tag));
1300 spin_unlock_irqrestore(ap->lock, flags);
1301
1302 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1303 }
1304
1305 /**
1306 * ata_eh_qc_complete - Complete an active ATA command from EH
1307 * @qc: Command to complete
1308 *
1309 * Indicate to the mid and upper layers that an ATA command has
1310 * completed. To be used from EH.
1311 */
1312 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1313 {
1314 struct scsi_cmnd *scmd = qc->scsicmd;
1315 scmd->retries = scmd->allowed;
1316 __ata_eh_qc_complete(qc);
1317 }
1318
1319 /**
1320 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1321 * @qc: Command to retry
1322 *
1323 * Indicate to the mid and upper layers that an ATA command
1324 * should be retried. To be used from EH.
1325 *
1326 * SCSI midlayer limits the number of retries to scmd->allowed.
1327 * scmd->allowed is incremented for commands which get retried
1328 * due to unrelated failures (qc->err_mask is zero).
1329 */
1330 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1331 {
1332 struct scsi_cmnd *scmd = qc->scsicmd;
1333 if (!qc->err_mask)
1334 scmd->allowed++;
1335 __ata_eh_qc_complete(qc);
1336 }
1337
1338 /**
1339 * ata_dev_disable - disable ATA device
1340 * @dev: ATA device to disable
1341 *
1342 * Disable @dev.
1343 *
1344 * Locking:
1345 * EH context.
1346 */
1347 void ata_dev_disable(struct ata_device *dev)
1348 {
1349 if (!ata_dev_enabled(dev))
1350 return;
1351
1352 if (ata_msg_drv(dev->link->ap))
1353 ata_dev_warn(dev, "disabled\n");
1354 ata_acpi_on_disable(dev);
1355 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1356 dev->class++;
1357
1358 /* From now till the next successful probe, ering is used to
1359 * track probe failures. Clear accumulated device error info.
1360 */
1361 ata_ering_clear(&dev->ering);
1362 }
1363
1364 /**
1365 * ata_eh_detach_dev - detach ATA device
1366 * @dev: ATA device to detach
1367 *
1368 * Detach @dev.
1369 *
1370 * LOCKING:
1371 * None.
1372 */
1373 void ata_eh_detach_dev(struct ata_device *dev)
1374 {
1375 struct ata_link *link = dev->link;
1376 struct ata_port *ap = link->ap;
1377 struct ata_eh_context *ehc = &link->eh_context;
1378 unsigned long flags;
1379
1380 ata_dev_disable(dev);
1381
1382 spin_lock_irqsave(ap->lock, flags);
1383
1384 dev->flags &= ~ATA_DFLAG_DETACH;
1385
1386 if (ata_scsi_offline_dev(dev)) {
1387 dev->flags |= ATA_DFLAG_DETACHED;
1388 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1389 }
1390
1391 /* clear per-dev EH info */
1392 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1393 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1394 ehc->saved_xfer_mode[dev->devno] = 0;
1395 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1396
1397 spin_unlock_irqrestore(ap->lock, flags);
1398 }
1399
1400 /**
1401 * ata_eh_about_to_do - about to perform eh_action
1402 * @link: target ATA link
1403 * @dev: target ATA dev for per-dev action (can be NULL)
1404 * @action: action about to be performed
1405 *
1406 * Called just before performing EH actions to clear related bits
1407 * in @link->eh_info such that eh actions are not unnecessarily
1408 * repeated.
1409 *
1410 * LOCKING:
1411 * None.
1412 */
1413 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1414 unsigned int action)
1415 {
1416 struct ata_port *ap = link->ap;
1417 struct ata_eh_info *ehi = &link->eh_info;
1418 struct ata_eh_context *ehc = &link->eh_context;
1419 unsigned long flags;
1420
1421 spin_lock_irqsave(ap->lock, flags);
1422
1423 ata_eh_clear_action(link, dev, ehi, action);
1424
1425 /* About to take EH action, set RECOVERED. Ignore actions on
1426 * slave links as master will do them again.
1427 */
1428 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1429 ap->pflags |= ATA_PFLAG_RECOVERED;
1430
1431 spin_unlock_irqrestore(ap->lock, flags);
1432 }
1433
1434 /**
1435 * ata_eh_done - EH action complete
1436 * @ap: target ATA port
1437 * @dev: target ATA dev for per-dev action (can be NULL)
1438 * @action: action just completed
1439 *
1440 * Called right after performing EH actions to clear related bits
1441 * in @link->eh_context.
1442 *
1443 * LOCKING:
1444 * None.
1445 */
1446 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1447 unsigned int action)
1448 {
1449 struct ata_eh_context *ehc = &link->eh_context;
1450
1451 ata_eh_clear_action(link, dev, &ehc->i, action);
1452 }
1453
1454 /**
1455 * ata_err_string - convert err_mask to descriptive string
1456 * @err_mask: error mask to convert to string
1457 *
1458 * Convert @err_mask to descriptive string. Errors are
1459 * prioritized according to severity and only the most severe
1460 * error is reported.
1461 *
1462 * LOCKING:
1463 * None.
1464 *
1465 * RETURNS:
1466 * Descriptive string for @err_mask
1467 */
1468 static const char *ata_err_string(unsigned int err_mask)
1469 {
1470 if (err_mask & AC_ERR_HOST_BUS)
1471 return "host bus error";
1472 if (err_mask & AC_ERR_ATA_BUS)
1473 return "ATA bus error";
1474 if (err_mask & AC_ERR_TIMEOUT)
1475 return "timeout";
1476 if (err_mask & AC_ERR_HSM)
1477 return "HSM violation";
1478 if (err_mask & AC_ERR_SYSTEM)
1479 return "internal error";
1480 if (err_mask & AC_ERR_MEDIA)
1481 return "media error";
1482 if (err_mask & AC_ERR_INVALID)
1483 return "invalid argument";
1484 if (err_mask & AC_ERR_DEV)
1485 return "device error";
1486 return "unknown error";
1487 }
1488
1489 /**
1490 * ata_read_log_page - read a specific log page
1491 * @dev: target device
1492 * @log: log to read
1493 * @page: page to read
1494 * @buf: buffer to store read page
1495 * @sectors: number of sectors to read
1496 *
1497 * Read log page using READ_LOG_EXT command.
1498 *
1499 * LOCKING:
1500 * Kernel thread context (may sleep).
1501 *
1502 * RETURNS:
1503 * 0 on success, AC_ERR_* mask otherwise.
1504 */
1505 unsigned int ata_read_log_page(struct ata_device *dev, u8 log,
1506 u8 page, void *buf, unsigned int sectors)
1507 {
1508 struct ata_taskfile tf;
1509 unsigned int err_mask;
1510 bool dma = false;
1511
1512 DPRINTK("read log page - log 0x%x, page 0x%x\n", log, page);
1513
1514 retry:
1515 ata_tf_init(dev, &tf);
1516 if (dev->dma_mode && ata_id_has_read_log_dma_ext(dev->id) &&
1517 !(dev->horkage & ATA_HORKAGE_NO_NCQ_LOG)) {
1518 tf.command = ATA_CMD_READ_LOG_DMA_EXT;
1519 tf.protocol = ATA_PROT_DMA;
1520 dma = true;
1521 } else {
1522 tf.command = ATA_CMD_READ_LOG_EXT;
1523 tf.protocol = ATA_PROT_PIO;
1524 dma = false;
1525 }
1526 tf.lbal = log;
1527 tf.lbam = page;
1528 tf.nsect = sectors;
1529 tf.hob_nsect = sectors >> 8;
1530 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1531
1532 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1533 buf, sectors * ATA_SECT_SIZE, 0);
1534
1535 if (err_mask && dma) {
1536 dev->horkage |= ATA_HORKAGE_NO_NCQ_LOG;
1537 ata_dev_warn(dev, "READ LOG DMA EXT failed, trying unqueued\n");
1538 goto retry;
1539 }
1540
1541 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1542 return err_mask;
1543 }
1544
1545 /**
1546 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1547 * @dev: Device to read log page 10h from
1548 * @tag: Resulting tag of the failed command
1549 * @tf: Resulting taskfile registers of the failed command
1550 *
1551 * Read log page 10h to obtain NCQ error details and clear error
1552 * condition.
1553 *
1554 * LOCKING:
1555 * Kernel thread context (may sleep).
1556 *
1557 * RETURNS:
1558 * 0 on success, -errno otherwise.
1559 */
1560 static int ata_eh_read_log_10h(struct ata_device *dev,
1561 int *tag, struct ata_taskfile *tf)
1562 {
1563 u8 *buf = dev->link->ap->sector_buf;
1564 unsigned int err_mask;
1565 u8 csum;
1566 int i;
1567
1568 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1569 if (err_mask)
1570 return -EIO;
1571
1572 csum = 0;
1573 for (i = 0; i < ATA_SECT_SIZE; i++)
1574 csum += buf[i];
1575 if (csum)
1576 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1577 csum);
1578
1579 if (buf[0] & 0x80)
1580 return -ENOENT;
1581
1582 *tag = buf[0] & 0x1f;
1583
1584 tf->command = buf[2];
1585 tf->feature = buf[3];
1586 tf->lbal = buf[4];
1587 tf->lbam = buf[5];
1588 tf->lbah = buf[6];
1589 tf->device = buf[7];
1590 tf->hob_lbal = buf[8];
1591 tf->hob_lbam = buf[9];
1592 tf->hob_lbah = buf[10];
1593 tf->nsect = buf[12];
1594 tf->hob_nsect = buf[13];
1595 if (ata_id_has_ncq_autosense(dev->id))
1596 tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16];
1597
1598 return 0;
1599 }
1600
1601 /**
1602 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1603 * @dev: target ATAPI device
1604 * @r_sense_key: out parameter for sense_key
1605 *
1606 * Perform ATAPI TEST_UNIT_READY.
1607 *
1608 * LOCKING:
1609 * EH context (may sleep).
1610 *
1611 * RETURNS:
1612 * 0 on success, AC_ERR_* mask on failure.
1613 */
1614 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1615 {
1616 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1617 struct ata_taskfile tf;
1618 unsigned int err_mask;
1619
1620 ata_tf_init(dev, &tf);
1621
1622 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1623 tf.command = ATA_CMD_PACKET;
1624 tf.protocol = ATAPI_PROT_NODATA;
1625
1626 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1627 if (err_mask == AC_ERR_DEV)
1628 *r_sense_key = tf.feature >> 4;
1629 return err_mask;
1630 }
1631
1632 /**
1633 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1634 * @dev: device to perform REQUEST_SENSE to
1635 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1636 * @dfl_sense_key: default sense key to use
1637 *
1638 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1639 * SENSE. This function is EH helper.
1640 *
1641 * LOCKING:
1642 * Kernel thread context (may sleep).
1643 *
1644 * RETURNS:
1645 * 0 on success, AC_ERR_* mask on failure
1646 */
1647 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1648 u8 *sense_buf, u8 dfl_sense_key)
1649 {
1650 u8 cdb[ATAPI_CDB_LEN] =
1651 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1652 struct ata_port *ap = dev->link->ap;
1653 struct ata_taskfile tf;
1654
1655 DPRINTK("ATAPI request sense\n");
1656
1657 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1658
1659 /* initialize sense_buf with the error register,
1660 * for the case where they are -not- overwritten
1661 */
1662 sense_buf[0] = 0x70;
1663 sense_buf[2] = dfl_sense_key;
1664
1665 /* some devices time out if garbage left in tf */
1666 ata_tf_init(dev, &tf);
1667
1668 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1669 tf.command = ATA_CMD_PACKET;
1670
1671 /* is it pointless to prefer PIO for "safety reasons"? */
1672 if (ap->flags & ATA_FLAG_PIO_DMA) {
1673 tf.protocol = ATAPI_PROT_DMA;
1674 tf.feature |= ATAPI_PKT_DMA;
1675 } else {
1676 tf.protocol = ATAPI_PROT_PIO;
1677 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1678 tf.lbah = 0;
1679 }
1680
1681 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1682 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1683 }
1684
1685 /**
1686 * ata_eh_analyze_serror - analyze SError for a failed port
1687 * @link: ATA link to analyze SError for
1688 *
1689 * Analyze SError if available and further determine cause of
1690 * failure.
1691 *
1692 * LOCKING:
1693 * None.
1694 */
1695 static void ata_eh_analyze_serror(struct ata_link *link)
1696 {
1697 struct ata_eh_context *ehc = &link->eh_context;
1698 u32 serror = ehc->i.serror;
1699 unsigned int err_mask = 0, action = 0;
1700 u32 hotplug_mask;
1701
1702 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1703 err_mask |= AC_ERR_ATA_BUS;
1704 action |= ATA_EH_RESET;
1705 }
1706 if (serror & SERR_PROTOCOL) {
1707 err_mask |= AC_ERR_HSM;
1708 action |= ATA_EH_RESET;
1709 }
1710 if (serror & SERR_INTERNAL) {
1711 err_mask |= AC_ERR_SYSTEM;
1712 action |= ATA_EH_RESET;
1713 }
1714
1715 /* Determine whether a hotplug event has occurred. Both
1716 * SError.N/X are considered hotplug events for enabled or
1717 * host links. For disabled PMP links, only N bit is
1718 * considered as X bit is left at 1 for link plugging.
1719 */
1720 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1721 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1722 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1723 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1724 else
1725 hotplug_mask = SERR_PHYRDY_CHG;
1726
1727 if (serror & hotplug_mask)
1728 ata_ehi_hotplugged(&ehc->i);
1729
1730 ehc->i.err_mask |= err_mask;
1731 ehc->i.action |= action;
1732 }
1733
1734 /**
1735 * ata_eh_analyze_ncq_error - analyze NCQ error
1736 * @link: ATA link to analyze NCQ error for
1737 *
1738 * Read log page 10h, determine the offending qc and acquire
1739 * error status TF. For NCQ device errors, all LLDDs have to do
1740 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1741 * care of the rest.
1742 *
1743 * LOCKING:
1744 * Kernel thread context (may sleep).
1745 */
1746 void ata_eh_analyze_ncq_error(struct ata_link *link)
1747 {
1748 struct ata_port *ap = link->ap;
1749 struct ata_eh_context *ehc = &link->eh_context;
1750 struct ata_device *dev = link->device;
1751 struct ata_queued_cmd *qc;
1752 struct ata_taskfile tf;
1753 int tag, rc;
1754
1755 /* if frozen, we can't do much */
1756 if (ap->pflags & ATA_PFLAG_FROZEN)
1757 return;
1758
1759 /* is it NCQ device error? */
1760 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1761 return;
1762
1763 /* has LLDD analyzed already? */
1764 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1765 qc = __ata_qc_from_tag(ap, tag);
1766
1767 if (!(qc->flags & ATA_QCFLAG_FAILED))
1768 continue;
1769
1770 if (qc->err_mask)
1771 return;
1772 }
1773
1774 /* okay, this error is ours */
1775 memset(&tf, 0, sizeof(tf));
1776 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1777 if (rc) {
1778 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1779 rc);
1780 return;
1781 }
1782
1783 if (!(link->sactive & (1 << tag))) {
1784 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1785 tag);
1786 return;
1787 }
1788
1789 /* we've got the perpetrator, condemn it */
1790 qc = __ata_qc_from_tag(ap, tag);
1791 memcpy(&qc->result_tf, &tf, sizeof(tf));
1792 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1793 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1794 if (qc->result_tf.auxiliary) {
1795 char sense_key, asc, ascq;
1796
1797 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1798 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1799 ascq = qc->result_tf.auxiliary & 0xff;
1800 ata_dev_dbg(dev, "NCQ Autosense %02x/%02x/%02x\n",
1801 sense_key, asc, ascq);
1802 ata_scsi_set_sense(qc->scsicmd, sense_key, asc, ascq);
1803 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1804 }
1805
1806 ehc->i.err_mask &= ~AC_ERR_DEV;
1807 }
1808
1809 /**
1810 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1811 * @qc: qc to analyze
1812 * @tf: Taskfile registers to analyze
1813 *
1814 * Analyze taskfile of @qc and further determine cause of
1815 * failure. This function also requests ATAPI sense data if
1816 * available.
1817 *
1818 * LOCKING:
1819 * Kernel thread context (may sleep).
1820 *
1821 * RETURNS:
1822 * Determined recovery action
1823 */
1824 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1825 const struct ata_taskfile *tf)
1826 {
1827 unsigned int tmp, action = 0;
1828 u8 stat = tf->command, err = tf->feature;
1829
1830 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1831 qc->err_mask |= AC_ERR_HSM;
1832 return ATA_EH_RESET;
1833 }
1834
1835 /* Set by NCQ autosense */
1836 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1837 return 0;
1838
1839 if (stat & (ATA_ERR | ATA_DF))
1840 qc->err_mask |= AC_ERR_DEV;
1841 else
1842 return 0;
1843
1844 switch (qc->dev->class) {
1845 case ATA_DEV_ATA:
1846 case ATA_DEV_ZAC:
1847 if (err & ATA_ICRC)
1848 qc->err_mask |= AC_ERR_ATA_BUS;
1849 if (err & (ATA_UNC | ATA_AMNF))
1850 qc->err_mask |= AC_ERR_MEDIA;
1851 if (err & ATA_IDNF)
1852 qc->err_mask |= AC_ERR_INVALID;
1853 break;
1854
1855 case ATA_DEV_ATAPI:
1856 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1857 tmp = atapi_eh_request_sense(qc->dev,
1858 qc->scsicmd->sense_buffer,
1859 qc->result_tf.feature >> 4);
1860 if (!tmp) {
1861 /* ATA_QCFLAG_SENSE_VALID is used to
1862 * tell atapi_qc_complete() that sense
1863 * data is already valid.
1864 *
1865 * TODO: interpret sense data and set
1866 * appropriate err_mask.
1867 */
1868 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1869 } else
1870 qc->err_mask |= tmp;
1871 }
1872 }
1873
1874 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1875 action |= ATA_EH_RESET;
1876
1877 return action;
1878 }
1879
1880 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1881 int *xfer_ok)
1882 {
1883 int base = 0;
1884
1885 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1886 *xfer_ok = 1;
1887
1888 if (!*xfer_ok)
1889 base = ATA_ECAT_DUBIOUS_NONE;
1890
1891 if (err_mask & AC_ERR_ATA_BUS)
1892 return base + ATA_ECAT_ATA_BUS;
1893
1894 if (err_mask & AC_ERR_TIMEOUT)
1895 return base + ATA_ECAT_TOUT_HSM;
1896
1897 if (eflags & ATA_EFLAG_IS_IO) {
1898 if (err_mask & AC_ERR_HSM)
1899 return base + ATA_ECAT_TOUT_HSM;
1900 if ((err_mask &
1901 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1902 return base + ATA_ECAT_UNK_DEV;
1903 }
1904
1905 return 0;
1906 }
1907
1908 struct speed_down_verdict_arg {
1909 u64 since;
1910 int xfer_ok;
1911 int nr_errors[ATA_ECAT_NR];
1912 };
1913
1914 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1915 {
1916 struct speed_down_verdict_arg *arg = void_arg;
1917 int cat;
1918
1919 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1920 return -1;
1921
1922 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1923 &arg->xfer_ok);
1924 arg->nr_errors[cat]++;
1925
1926 return 0;
1927 }
1928
1929 /**
1930 * ata_eh_speed_down_verdict - Determine speed down verdict
1931 * @dev: Device of interest
1932 *
1933 * This function examines error ring of @dev and determines
1934 * whether NCQ needs to be turned off, transfer speed should be
1935 * stepped down, or falling back to PIO is necessary.
1936 *
1937 * ECAT_ATA_BUS : ATA_BUS error for any command
1938 *
1939 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1940 * IO commands
1941 *
1942 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1943 *
1944 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1945 * data transfer hasn't been verified.
1946 *
1947 * Verdicts are
1948 *
1949 * NCQ_OFF : Turn off NCQ.
1950 *
1951 * SPEED_DOWN : Speed down transfer speed but don't fall back
1952 * to PIO.
1953 *
1954 * FALLBACK_TO_PIO : Fall back to PIO.
1955 *
1956 * Even if multiple verdicts are returned, only one action is
1957 * taken per error. An action triggered by non-DUBIOUS errors
1958 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1959 * This is to expedite speed down decisions right after device is
1960 * initially configured.
1961 *
1962 * The followings are speed down rules. #1 and #2 deal with
1963 * DUBIOUS errors.
1964 *
1965 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1966 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1967 *
1968 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1969 * occurred during last 5 mins, NCQ_OFF.
1970 *
1971 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1972 * occurred during last 5 mins, FALLBACK_TO_PIO
1973 *
1974 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1975 * during last 10 mins, NCQ_OFF.
1976 *
1977 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1978 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1979 *
1980 * LOCKING:
1981 * Inherited from caller.
1982 *
1983 * RETURNS:
1984 * OR of ATA_EH_SPDN_* flags.
1985 */
1986 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1987 {
1988 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1989 u64 j64 = get_jiffies_64();
1990 struct speed_down_verdict_arg arg;
1991 unsigned int verdict = 0;
1992
1993 /* scan past 5 mins of error history */
1994 memset(&arg, 0, sizeof(arg));
1995 arg.since = j64 - min(j64, j5mins);
1996 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1997
1998 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1999 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
2000 verdict |= ATA_EH_SPDN_SPEED_DOWN |
2001 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
2002
2003 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
2004 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
2005 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
2006
2007 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2008 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2009 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2010 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
2011
2012 /* scan past 10 mins of error history */
2013 memset(&arg, 0, sizeof(arg));
2014 arg.since = j64 - min(j64, j10mins);
2015 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2016
2017 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2018 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
2019 verdict |= ATA_EH_SPDN_NCQ_OFF;
2020
2021 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2022 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
2023 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2024 verdict |= ATA_EH_SPDN_SPEED_DOWN;
2025
2026 return verdict;
2027 }
2028
2029 /**
2030 * ata_eh_speed_down - record error and speed down if necessary
2031 * @dev: Failed device
2032 * @eflags: mask of ATA_EFLAG_* flags
2033 * @err_mask: err_mask of the error
2034 *
2035 * Record error and examine error history to determine whether
2036 * adjusting transmission speed is necessary. It also sets
2037 * transmission limits appropriately if such adjustment is
2038 * necessary.
2039 *
2040 * LOCKING:
2041 * Kernel thread context (may sleep).
2042 *
2043 * RETURNS:
2044 * Determined recovery action.
2045 */
2046 static unsigned int ata_eh_speed_down(struct ata_device *dev,
2047 unsigned int eflags, unsigned int err_mask)
2048 {
2049 struct ata_link *link = ata_dev_phys_link(dev);
2050 int xfer_ok = 0;
2051 unsigned int verdict;
2052 unsigned int action = 0;
2053
2054 /* don't bother if Cat-0 error */
2055 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2056 return 0;
2057
2058 /* record error and determine whether speed down is necessary */
2059 ata_ering_record(&dev->ering, eflags, err_mask);
2060 verdict = ata_eh_speed_down_verdict(dev);
2061
2062 /* turn off NCQ? */
2063 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2064 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2065 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2066 dev->flags |= ATA_DFLAG_NCQ_OFF;
2067 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2068 goto done;
2069 }
2070
2071 /* speed down? */
2072 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2073 /* speed down SATA link speed if possible */
2074 if (sata_down_spd_limit(link, 0) == 0) {
2075 action |= ATA_EH_RESET;
2076 goto done;
2077 }
2078
2079 /* lower transfer mode */
2080 if (dev->spdn_cnt < 2) {
2081 static const int dma_dnxfer_sel[] =
2082 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2083 static const int pio_dnxfer_sel[] =
2084 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2085 int sel;
2086
2087 if (dev->xfer_shift != ATA_SHIFT_PIO)
2088 sel = dma_dnxfer_sel[dev->spdn_cnt];
2089 else
2090 sel = pio_dnxfer_sel[dev->spdn_cnt];
2091
2092 dev->spdn_cnt++;
2093
2094 if (ata_down_xfermask_limit(dev, sel) == 0) {
2095 action |= ATA_EH_RESET;
2096 goto done;
2097 }
2098 }
2099 }
2100
2101 /* Fall back to PIO? Slowing down to PIO is meaningless for
2102 * SATA ATA devices. Consider it only for PATA and SATAPI.
2103 */
2104 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2105 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2106 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2107 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2108 dev->spdn_cnt = 0;
2109 action |= ATA_EH_RESET;
2110 goto done;
2111 }
2112 }
2113
2114 return 0;
2115 done:
2116 /* device has been slowed down, blow error history */
2117 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2118 ata_ering_clear(&dev->ering);
2119 return action;
2120 }
2121
2122 /**
2123 * ata_eh_worth_retry - analyze error and decide whether to retry
2124 * @qc: qc to possibly retry
2125 *
2126 * Look at the cause of the error and decide if a retry
2127 * might be useful or not. We don't want to retry media errors
2128 * because the drive itself has probably already taken 10-30 seconds
2129 * doing its own internal retries before reporting the failure.
2130 */
2131 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2132 {
2133 if (qc->err_mask & AC_ERR_MEDIA)
2134 return 0; /* don't retry media errors */
2135 if (qc->flags & ATA_QCFLAG_IO)
2136 return 1; /* otherwise retry anything from fs stack */
2137 if (qc->err_mask & AC_ERR_INVALID)
2138 return 0; /* don't retry these */
2139 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2140 }
2141
2142 /**
2143 * ata_eh_link_autopsy - analyze error and determine recovery action
2144 * @link: host link to perform autopsy on
2145 *
2146 * Analyze why @link failed and determine which recovery actions
2147 * are needed. This function also sets more detailed AC_ERR_*
2148 * values and fills sense data for ATAPI CHECK SENSE.
2149 *
2150 * LOCKING:
2151 * Kernel thread context (may sleep).
2152 */
2153 static void ata_eh_link_autopsy(struct ata_link *link)
2154 {
2155 struct ata_port *ap = link->ap;
2156 struct ata_eh_context *ehc = &link->eh_context;
2157 struct ata_device *dev;
2158 unsigned int all_err_mask = 0, eflags = 0;
2159 int tag;
2160 u32 serror;
2161 int rc;
2162
2163 DPRINTK("ENTER\n");
2164
2165 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2166 return;
2167
2168 /* obtain and analyze SError */
2169 rc = sata_scr_read(link, SCR_ERROR, &serror);
2170 if (rc == 0) {
2171 ehc->i.serror |= serror;
2172 ata_eh_analyze_serror(link);
2173 } else if (rc != -EOPNOTSUPP) {
2174 /* SError read failed, force reset and probing */
2175 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2176 ehc->i.action |= ATA_EH_RESET;
2177 ehc->i.err_mask |= AC_ERR_OTHER;
2178 }
2179
2180 /* analyze NCQ failure */
2181 ata_eh_analyze_ncq_error(link);
2182
2183 /* any real error trumps AC_ERR_OTHER */
2184 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2185 ehc->i.err_mask &= ~AC_ERR_OTHER;
2186
2187 all_err_mask |= ehc->i.err_mask;
2188
2189 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2190 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2191
2192 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2193 ata_dev_phys_link(qc->dev) != link)
2194 continue;
2195
2196 /* inherit upper level err_mask */
2197 qc->err_mask |= ehc->i.err_mask;
2198
2199 /* analyze TF */
2200 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2201
2202 /* DEV errors are probably spurious in case of ATA_BUS error */
2203 if (qc->err_mask & AC_ERR_ATA_BUS)
2204 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2205 AC_ERR_INVALID);
2206
2207 /* any real error trumps unknown error */
2208 if (qc->err_mask & ~AC_ERR_OTHER)
2209 qc->err_mask &= ~AC_ERR_OTHER;
2210
2211 /* SENSE_VALID trumps dev/unknown error and revalidation */
2212 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2213 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2214
2215 /* determine whether the command is worth retrying */
2216 if (ata_eh_worth_retry(qc))
2217 qc->flags |= ATA_QCFLAG_RETRY;
2218
2219 /* accumulate error info */
2220 ehc->i.dev = qc->dev;
2221 all_err_mask |= qc->err_mask;
2222 if (qc->flags & ATA_QCFLAG_IO)
2223 eflags |= ATA_EFLAG_IS_IO;
2224 trace_ata_eh_link_autopsy_qc(qc);
2225 }
2226
2227 /* enforce default EH actions */
2228 if (ap->pflags & ATA_PFLAG_FROZEN ||
2229 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2230 ehc->i.action |= ATA_EH_RESET;
2231 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2232 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2233 ehc->i.action |= ATA_EH_REVALIDATE;
2234
2235 /* If we have offending qcs and the associated failed device,
2236 * perform per-dev EH action only on the offending device.
2237 */
2238 if (ehc->i.dev) {
2239 ehc->i.dev_action[ehc->i.dev->devno] |=
2240 ehc->i.action & ATA_EH_PERDEV_MASK;
2241 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2242 }
2243
2244 /* propagate timeout to host link */
2245 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2246 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2247
2248 /* record error and consider speeding down */
2249 dev = ehc->i.dev;
2250 if (!dev && ((ata_link_max_devices(link) == 1 &&
2251 ata_dev_enabled(link->device))))
2252 dev = link->device;
2253
2254 if (dev) {
2255 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2256 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2257 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2258 }
2259 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2260 DPRINTK("EXIT\n");
2261 }
2262
2263 /**
2264 * ata_eh_autopsy - analyze error and determine recovery action
2265 * @ap: host port to perform autopsy on
2266 *
2267 * Analyze all links of @ap and determine why they failed and
2268 * which recovery actions are needed.
2269 *
2270 * LOCKING:
2271 * Kernel thread context (may sleep).
2272 */
2273 void ata_eh_autopsy(struct ata_port *ap)
2274 {
2275 struct ata_link *link;
2276
2277 ata_for_each_link(link, ap, EDGE)
2278 ata_eh_link_autopsy(link);
2279
2280 /* Handle the frigging slave link. Autopsy is done similarly
2281 * but actions and flags are transferred over to the master
2282 * link and handled from there.
2283 */
2284 if (ap->slave_link) {
2285 struct ata_eh_context *mehc = &ap->link.eh_context;
2286 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2287
2288 /* transfer control flags from master to slave */
2289 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2290
2291 /* perform autopsy on the slave link */
2292 ata_eh_link_autopsy(ap->slave_link);
2293
2294 /* transfer actions from slave to master and clear slave */
2295 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2296 mehc->i.action |= sehc->i.action;
2297 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2298 mehc->i.flags |= sehc->i.flags;
2299 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2300 }
2301
2302 /* Autopsy of fanout ports can affect host link autopsy.
2303 * Perform host link autopsy last.
2304 */
2305 if (sata_pmp_attached(ap))
2306 ata_eh_link_autopsy(&ap->link);
2307 }
2308
2309 /**
2310 * ata_get_cmd_descript - get description for ATA command
2311 * @command: ATA command code to get description for
2312 *
2313 * Return a textual description of the given command, or NULL if the
2314 * command is not known.
2315 *
2316 * LOCKING:
2317 * None
2318 */
2319 const char *ata_get_cmd_descript(u8 command)
2320 {
2321 #ifdef CONFIG_ATA_VERBOSE_ERROR
2322 static const struct
2323 {
2324 u8 command;
2325 const char *text;
2326 } cmd_descr[] = {
2327 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2328 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2329 { ATA_CMD_STANDBY, "STANDBY" },
2330 { ATA_CMD_IDLE, "IDLE" },
2331 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2332 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2333 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2334 { ATA_CMD_NOP, "NOP" },
2335 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2336 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2337 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2338 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2339 { ATA_CMD_SERVICE, "SERVICE" },
2340 { ATA_CMD_READ, "READ DMA" },
2341 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2342 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2343 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2344 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2345 { ATA_CMD_WRITE, "WRITE DMA" },
2346 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2347 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2348 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2349 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2350 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2351 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2352 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2353 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2354 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2355 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2356 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2357 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2358 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2359 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2360 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2361 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2362 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2363 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2364 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2365 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2366 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2367 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2368 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2369 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2370 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2371 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2372 { ATA_CMD_SLEEP, "SLEEP" },
2373 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2374 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2375 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2376 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2377 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2378 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2379 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2380 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2381 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2382 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2383 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2384 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2385 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2386 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2387 { ATA_CMD_PMP_READ, "READ BUFFER" },
2388 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2389 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2390 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2391 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2392 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2393 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2394 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2395 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2396 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2397 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2398 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2399 { ATA_CMD_SMART, "SMART" },
2400 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2401 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2402 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2403 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2404 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2405 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2406 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2407 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2408 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2409 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2410 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2411 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2412 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2413 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2414 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2415 { ATA_CMD_RESTORE, "RECALIBRATE" },
2416 { 0, NULL } /* terminate list */
2417 };
2418
2419 unsigned int i;
2420 for (i = 0; cmd_descr[i].text; i++)
2421 if (cmd_descr[i].command == command)
2422 return cmd_descr[i].text;
2423 #endif
2424
2425 return NULL;
2426 }
2427 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2428
2429 /**
2430 * ata_eh_link_report - report error handling to user
2431 * @link: ATA link EH is going on
2432 *
2433 * Report EH to user.
2434 *
2435 * LOCKING:
2436 * None.
2437 */
2438 static void ata_eh_link_report(struct ata_link *link)
2439 {
2440 struct ata_port *ap = link->ap;
2441 struct ata_eh_context *ehc = &link->eh_context;
2442 const char *frozen, *desc;
2443 char tries_buf[6] = "";
2444 int tag, nr_failed = 0;
2445
2446 if (ehc->i.flags & ATA_EHI_QUIET)
2447 return;
2448
2449 desc = NULL;
2450 if (ehc->i.desc[0] != '\0')
2451 desc = ehc->i.desc;
2452
2453 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2454 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2455
2456 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2457 ata_dev_phys_link(qc->dev) != link ||
2458 ((qc->flags & ATA_QCFLAG_QUIET) &&
2459 qc->err_mask == AC_ERR_DEV))
2460 continue;
2461 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2462 continue;
2463
2464 nr_failed++;
2465 }
2466
2467 if (!nr_failed && !ehc->i.err_mask)
2468 return;
2469
2470 frozen = "";
2471 if (ap->pflags & ATA_PFLAG_FROZEN)
2472 frozen = " frozen";
2473
2474 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2475 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2476 ap->eh_tries);
2477
2478 if (ehc->i.dev) {
2479 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2480 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2481 ehc->i.err_mask, link->sactive, ehc->i.serror,
2482 ehc->i.action, frozen, tries_buf);
2483 if (desc)
2484 ata_dev_err(ehc->i.dev, "%s\n", desc);
2485 } else {
2486 ata_link_err(link, "exception Emask 0x%x "
2487 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2488 ehc->i.err_mask, link->sactive, ehc->i.serror,
2489 ehc->i.action, frozen, tries_buf);
2490 if (desc)
2491 ata_link_err(link, "%s\n", desc);
2492 }
2493
2494 #ifdef CONFIG_ATA_VERBOSE_ERROR
2495 if (ehc->i.serror)
2496 ata_link_err(link,
2497 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2498 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2499 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2500 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2501 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2502 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2503 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2504 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2505 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2506 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2507 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2508 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2509 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2510 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2511 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2512 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2513 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2514 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2515 #endif
2516
2517 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2518 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2519 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2520 char data_buf[20] = "";
2521 char cdb_buf[70] = "";
2522
2523 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2524 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2525 continue;
2526
2527 if (qc->dma_dir != DMA_NONE) {
2528 static const char *dma_str[] = {
2529 [DMA_BIDIRECTIONAL] = "bidi",
2530 [DMA_TO_DEVICE] = "out",
2531 [DMA_FROM_DEVICE] = "in",
2532 };
2533 static const char *prot_str[] = {
2534 [ATA_PROT_PIO] = "pio",
2535 [ATA_PROT_DMA] = "dma",
2536 [ATA_PROT_NCQ] = "ncq",
2537 [ATAPI_PROT_PIO] = "pio",
2538 [ATAPI_PROT_DMA] = "dma",
2539 };
2540
2541 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2542 prot_str[qc->tf.protocol], qc->nbytes,
2543 dma_str[qc->dma_dir]);
2544 }
2545
2546 if (ata_is_atapi(qc->tf.protocol)) {
2547 const u8 *cdb = qc->cdb;
2548 size_t cdb_len = qc->dev->cdb_len;
2549
2550 if (qc->scsicmd) {
2551 cdb = qc->scsicmd->cmnd;
2552 cdb_len = qc->scsicmd->cmd_len;
2553 }
2554 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2555 cdb, cdb_len);
2556 } else {
2557 const char *descr = ata_get_cmd_descript(cmd->command);
2558 if (descr)
2559 ata_dev_err(qc->dev, "failed command: %s\n",
2560 descr);
2561 }
2562
2563 ata_dev_err(qc->dev,
2564 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2565 "tag %d%s\n %s"
2566 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2567 "Emask 0x%x (%s)%s\n",
2568 cmd->command, cmd->feature, cmd->nsect,
2569 cmd->lbal, cmd->lbam, cmd->lbah,
2570 cmd->hob_feature, cmd->hob_nsect,
2571 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2572 cmd->device, qc->tag, data_buf, cdb_buf,
2573 res->command, res->feature, res->nsect,
2574 res->lbal, res->lbam, res->lbah,
2575 res->hob_feature, res->hob_nsect,
2576 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2577 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2578 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2579
2580 #ifdef CONFIG_ATA_VERBOSE_ERROR
2581 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2582 ATA_ERR)) {
2583 if (res->command & ATA_BUSY)
2584 ata_dev_err(qc->dev, "status: { Busy }\n");
2585 else
2586 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
2587 res->command & ATA_DRDY ? "DRDY " : "",
2588 res->command & ATA_DF ? "DF " : "",
2589 res->command & ATA_DRQ ? "DRQ " : "",
2590 res->command & ATA_ERR ? "ERR " : "");
2591 }
2592
2593 if (cmd->command != ATA_CMD_PACKET &&
2594 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2595 ATA_IDNF | ATA_ABORTED)))
2596 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2597 res->feature & ATA_ICRC ? "ICRC " : "",
2598 res->feature & ATA_UNC ? "UNC " : "",
2599 res->feature & ATA_AMNF ? "AMNF " : "",
2600 res->feature & ATA_IDNF ? "IDNF " : "",
2601 res->feature & ATA_ABORTED ? "ABRT " : "");
2602 #endif
2603 }
2604 }
2605
2606 /**
2607 * ata_eh_report - report error handling to user
2608 * @ap: ATA port to report EH about
2609 *
2610 * Report EH to user.
2611 *
2612 * LOCKING:
2613 * None.
2614 */
2615 void ata_eh_report(struct ata_port *ap)
2616 {
2617 struct ata_link *link;
2618
2619 ata_for_each_link(link, ap, HOST_FIRST)
2620 ata_eh_link_report(link);
2621 }
2622
2623 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2624 unsigned int *classes, unsigned long deadline,
2625 bool clear_classes)
2626 {
2627 struct ata_device *dev;
2628
2629 if (clear_classes)
2630 ata_for_each_dev(dev, link, ALL)
2631 classes[dev->devno] = ATA_DEV_UNKNOWN;
2632
2633 return reset(link, classes, deadline);
2634 }
2635
2636 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2637 {
2638 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2639 return 0;
2640 if (rc == -EAGAIN)
2641 return 1;
2642 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2643 return 1;
2644 return 0;
2645 }
2646
2647 int ata_eh_reset(struct ata_link *link, int classify,
2648 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2649 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2650 {
2651 struct ata_port *ap = link->ap;
2652 struct ata_link *slave = ap->slave_link;
2653 struct ata_eh_context *ehc = &link->eh_context;
2654 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2655 unsigned int *classes = ehc->classes;
2656 unsigned int lflags = link->flags;
2657 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2658 int max_tries = 0, try = 0;
2659 struct ata_link *failed_link;
2660 struct ata_device *dev;
2661 unsigned long deadline, now;
2662 ata_reset_fn_t reset;
2663 unsigned long flags;
2664 u32 sstatus;
2665 int nr_unknown, rc;
2666
2667 /*
2668 * Prepare to reset
2669 */
2670 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2671 max_tries++;
2672 if (link->flags & ATA_LFLAG_RST_ONCE)
2673 max_tries = 1;
2674 if (link->flags & ATA_LFLAG_NO_HRST)
2675 hardreset = NULL;
2676 if (link->flags & ATA_LFLAG_NO_SRST)
2677 softreset = NULL;
2678
2679 /* make sure each reset attempt is at least COOL_DOWN apart */
2680 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2681 now = jiffies;
2682 WARN_ON(time_after(ehc->last_reset, now));
2683 deadline = ata_deadline(ehc->last_reset,
2684 ATA_EH_RESET_COOL_DOWN);
2685 if (time_before(now, deadline))
2686 schedule_timeout_uninterruptible(deadline - now);
2687 }
2688
2689 spin_lock_irqsave(ap->lock, flags);
2690 ap->pflags |= ATA_PFLAG_RESETTING;
2691 spin_unlock_irqrestore(ap->lock, flags);
2692
2693 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2694
2695 ata_for_each_dev(dev, link, ALL) {
2696 /* If we issue an SRST then an ATA drive (not ATAPI)
2697 * may change configuration and be in PIO0 timing. If
2698 * we do a hard reset (or are coming from power on)
2699 * this is true for ATA or ATAPI. Until we've set a
2700 * suitable controller mode we should not touch the
2701 * bus as we may be talking too fast.
2702 */
2703 dev->pio_mode = XFER_PIO_0;
2704 dev->dma_mode = 0xff;
2705
2706 /* If the controller has a pio mode setup function
2707 * then use it to set the chipset to rights. Don't
2708 * touch the DMA setup as that will be dealt with when
2709 * configuring devices.
2710 */
2711 if (ap->ops->set_piomode)
2712 ap->ops->set_piomode(ap, dev);
2713 }
2714
2715 /* prefer hardreset */
2716 reset = NULL;
2717 ehc->i.action &= ~ATA_EH_RESET;
2718 if (hardreset) {
2719 reset = hardreset;
2720 ehc->i.action |= ATA_EH_HARDRESET;
2721 } else if (softreset) {
2722 reset = softreset;
2723 ehc->i.action |= ATA_EH_SOFTRESET;
2724 }
2725
2726 if (prereset) {
2727 unsigned long deadline = ata_deadline(jiffies,
2728 ATA_EH_PRERESET_TIMEOUT);
2729
2730 if (slave) {
2731 sehc->i.action &= ~ATA_EH_RESET;
2732 sehc->i.action |= ehc->i.action;
2733 }
2734
2735 rc = prereset(link, deadline);
2736
2737 /* If present, do prereset on slave link too. Reset
2738 * is skipped iff both master and slave links report
2739 * -ENOENT or clear ATA_EH_RESET.
2740 */
2741 if (slave && (rc == 0 || rc == -ENOENT)) {
2742 int tmp;
2743
2744 tmp = prereset(slave, deadline);
2745 if (tmp != -ENOENT)
2746 rc = tmp;
2747
2748 ehc->i.action |= sehc->i.action;
2749 }
2750
2751 if (rc) {
2752 if (rc == -ENOENT) {
2753 ata_link_dbg(link, "port disabled--ignoring\n");
2754 ehc->i.action &= ~ATA_EH_RESET;
2755
2756 ata_for_each_dev(dev, link, ALL)
2757 classes[dev->devno] = ATA_DEV_NONE;
2758
2759 rc = 0;
2760 } else
2761 ata_link_err(link,
2762 "prereset failed (errno=%d)\n",
2763 rc);
2764 goto out;
2765 }
2766
2767 /* prereset() might have cleared ATA_EH_RESET. If so,
2768 * bang classes, thaw and return.
2769 */
2770 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2771 ata_for_each_dev(dev, link, ALL)
2772 classes[dev->devno] = ATA_DEV_NONE;
2773 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2774 ata_is_host_link(link))
2775 ata_eh_thaw_port(ap);
2776 rc = 0;
2777 goto out;
2778 }
2779 }
2780
2781 retry:
2782 /*
2783 * Perform reset
2784 */
2785 if (ata_is_host_link(link))
2786 ata_eh_freeze_port(ap);
2787
2788 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2789
2790 if (reset) {
2791 if (verbose)
2792 ata_link_info(link, "%s resetting link\n",
2793 reset == softreset ? "soft" : "hard");
2794
2795 /* mark that this EH session started with reset */
2796 ehc->last_reset = jiffies;
2797 if (reset == hardreset)
2798 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2799 else
2800 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2801
2802 rc = ata_do_reset(link, reset, classes, deadline, true);
2803 if (rc && rc != -EAGAIN) {
2804 failed_link = link;
2805 goto fail;
2806 }
2807
2808 /* hardreset slave link if existent */
2809 if (slave && reset == hardreset) {
2810 int tmp;
2811
2812 if (verbose)
2813 ata_link_info(slave, "hard resetting link\n");
2814
2815 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2816 tmp = ata_do_reset(slave, reset, classes, deadline,
2817 false);
2818 switch (tmp) {
2819 case -EAGAIN:
2820 rc = -EAGAIN;
2821 case 0:
2822 break;
2823 default:
2824 failed_link = slave;
2825 rc = tmp;
2826 goto fail;
2827 }
2828 }
2829
2830 /* perform follow-up SRST if necessary */
2831 if (reset == hardreset &&
2832 ata_eh_followup_srst_needed(link, rc)) {
2833 reset = softreset;
2834
2835 if (!reset) {
2836 ata_link_err(link,
2837 "follow-up softreset required but no softreset available\n");
2838 failed_link = link;
2839 rc = -EINVAL;
2840 goto fail;
2841 }
2842
2843 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2844 rc = ata_do_reset(link, reset, classes, deadline, true);
2845 if (rc) {
2846 failed_link = link;
2847 goto fail;
2848 }
2849 }
2850 } else {
2851 if (verbose)
2852 ata_link_info(link,
2853 "no reset method available, skipping reset\n");
2854 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2855 lflags |= ATA_LFLAG_ASSUME_ATA;
2856 }
2857
2858 /*
2859 * Post-reset processing
2860 */
2861 ata_for_each_dev(dev, link, ALL) {
2862 /* After the reset, the device state is PIO 0 and the
2863 * controller state is undefined. Reset also wakes up
2864 * drives from sleeping mode.
2865 */
2866 dev->pio_mode = XFER_PIO_0;
2867 dev->flags &= ~ATA_DFLAG_SLEEPING;
2868
2869 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2870 continue;
2871
2872 /* apply class override */
2873 if (lflags & ATA_LFLAG_ASSUME_ATA)
2874 classes[dev->devno] = ATA_DEV_ATA;
2875 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2876 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2877 }
2878
2879 /* record current link speed */
2880 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2881 link->sata_spd = (sstatus >> 4) & 0xf;
2882 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2883 slave->sata_spd = (sstatus >> 4) & 0xf;
2884
2885 /* thaw the port */
2886 if (ata_is_host_link(link))
2887 ata_eh_thaw_port(ap);
2888
2889 /* postreset() should clear hardware SError. Although SError
2890 * is cleared during link resume, clearing SError here is
2891 * necessary as some PHYs raise hotplug events after SRST.
2892 * This introduces race condition where hotplug occurs between
2893 * reset and here. This race is mediated by cross checking
2894 * link onlineness and classification result later.
2895 */
2896 if (postreset) {
2897 postreset(link, classes);
2898 if (slave)
2899 postreset(slave, classes);
2900 }
2901
2902 /*
2903 * Some controllers can't be frozen very well and may set spurious
2904 * error conditions during reset. Clear accumulated error
2905 * information and re-thaw the port if frozen. As reset is the
2906 * final recovery action and we cross check link onlineness against
2907 * device classification later, no hotplug event is lost by this.
2908 */
2909 spin_lock_irqsave(link->ap->lock, flags);
2910 memset(&link->eh_info, 0, sizeof(link->eh_info));
2911 if (slave)
2912 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2913 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2914 spin_unlock_irqrestore(link->ap->lock, flags);
2915
2916 if (ap->pflags & ATA_PFLAG_FROZEN)
2917 ata_eh_thaw_port(ap);
2918
2919 /*
2920 * Make sure onlineness and classification result correspond.
2921 * Hotplug could have happened during reset and some
2922 * controllers fail to wait while a drive is spinning up after
2923 * being hotplugged causing misdetection. By cross checking
2924 * link on/offlineness and classification result, those
2925 * conditions can be reliably detected and retried.
2926 */
2927 nr_unknown = 0;
2928 ata_for_each_dev(dev, link, ALL) {
2929 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2930 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2931 ata_dev_dbg(dev, "link online but device misclassified\n");
2932 classes[dev->devno] = ATA_DEV_NONE;
2933 nr_unknown++;
2934 }
2935 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2936 if (ata_class_enabled(classes[dev->devno]))
2937 ata_dev_dbg(dev,
2938 "link offline, clearing class %d to NONE\n",
2939 classes[dev->devno]);
2940 classes[dev->devno] = ATA_DEV_NONE;
2941 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2942 ata_dev_dbg(dev,
2943 "link status unknown, clearing UNKNOWN to NONE\n");
2944 classes[dev->devno] = ATA_DEV_NONE;
2945 }
2946 }
2947
2948 if (classify && nr_unknown) {
2949 if (try < max_tries) {
2950 ata_link_warn(link,
2951 "link online but %d devices misclassified, retrying\n",
2952 nr_unknown);
2953 failed_link = link;
2954 rc = -EAGAIN;
2955 goto fail;
2956 }
2957 ata_link_warn(link,
2958 "link online but %d devices misclassified, "
2959 "device detection might fail\n", nr_unknown);
2960 }
2961
2962 /* reset successful, schedule revalidation */
2963 ata_eh_done(link, NULL, ATA_EH_RESET);
2964 if (slave)
2965 ata_eh_done(slave, NULL, ATA_EH_RESET);
2966 ehc->last_reset = jiffies; /* update to completion time */
2967 ehc->i.action |= ATA_EH_REVALIDATE;
2968 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2969
2970 rc = 0;
2971 out:
2972 /* clear hotplug flag */
2973 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2974 if (slave)
2975 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2976
2977 spin_lock_irqsave(ap->lock, flags);
2978 ap->pflags &= ~ATA_PFLAG_RESETTING;
2979 spin_unlock_irqrestore(ap->lock, flags);
2980
2981 return rc;
2982
2983 fail:
2984 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2985 if (!ata_is_host_link(link) &&
2986 sata_scr_read(link, SCR_STATUS, &sstatus))
2987 rc = -ERESTART;
2988
2989 if (try >= max_tries) {
2990 /*
2991 * Thaw host port even if reset failed, so that the port
2992 * can be retried on the next phy event. This risks
2993 * repeated EH runs but seems to be a better tradeoff than
2994 * shutting down a port after a botched hotplug attempt.
2995 */
2996 if (ata_is_host_link(link))
2997 ata_eh_thaw_port(ap);
2998 goto out;
2999 }
3000
3001 now = jiffies;
3002 if (time_before(now, deadline)) {
3003 unsigned long delta = deadline - now;
3004
3005 ata_link_warn(failed_link,
3006 "reset failed (errno=%d), retrying in %u secs\n",
3007 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3008
3009 ata_eh_release(ap);
3010 while (delta)
3011 delta = schedule_timeout_uninterruptible(delta);
3012 ata_eh_acquire(ap);
3013 }
3014
3015 /*
3016 * While disks spinup behind PMP, some controllers fail sending SRST.
3017 * They need to be reset - as well as the PMP - before retrying.
3018 */
3019 if (rc == -ERESTART) {
3020 if (ata_is_host_link(link))
3021 ata_eh_thaw_port(ap);
3022 goto out;
3023 }
3024
3025 if (try == max_tries - 1) {
3026 sata_down_spd_limit(link, 0);
3027 if (slave)
3028 sata_down_spd_limit(slave, 0);
3029 } else if (rc == -EPIPE)
3030 sata_down_spd_limit(failed_link, 0);
3031
3032 if (hardreset)
3033 reset = hardreset;
3034 goto retry;
3035 }
3036
3037 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3038 {
3039 struct ata_link *link;
3040 struct ata_device *dev;
3041 unsigned long flags;
3042
3043 /*
3044 * This function can be thought of as an extended version of
3045 * ata_eh_about_to_do() specially crafted to accommodate the
3046 * requirements of ATA_EH_PARK handling. Since the EH thread
3047 * does not leave the do {} while () loop in ata_eh_recover as
3048 * long as the timeout for a park request to *one* device on
3049 * the port has not expired, and since we still want to pick
3050 * up park requests to other devices on the same port or
3051 * timeout updates for the same device, we have to pull
3052 * ATA_EH_PARK actions from eh_info into eh_context.i
3053 * ourselves at the beginning of each pass over the loop.
3054 *
3055 * Additionally, all write accesses to &ap->park_req_pending
3056 * through reinit_completion() (see below) or complete_all()
3057 * (see ata_scsi_park_store()) are protected by the host lock.
3058 * As a result we have that park_req_pending.done is zero on
3059 * exit from this function, i.e. when ATA_EH_PARK actions for
3060 * *all* devices on port ap have been pulled into the
3061 * respective eh_context structs. If, and only if,
3062 * park_req_pending.done is non-zero by the time we reach
3063 * wait_for_completion_timeout(), another ATA_EH_PARK action
3064 * has been scheduled for at least one of the devices on port
3065 * ap and we have to cycle over the do {} while () loop in
3066 * ata_eh_recover() again.
3067 */
3068
3069 spin_lock_irqsave(ap->lock, flags);
3070 reinit_completion(&ap->park_req_pending);
3071 ata_for_each_link(link, ap, EDGE) {
3072 ata_for_each_dev(dev, link, ALL) {
3073 struct ata_eh_info *ehi = &link->eh_info;
3074
3075 link->eh_context.i.dev_action[dev->devno] |=
3076 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3077 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3078 }
3079 }
3080 spin_unlock_irqrestore(ap->lock, flags);
3081 }
3082
3083 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3084 {
3085 struct ata_eh_context *ehc = &dev->link->eh_context;
3086 struct ata_taskfile tf;
3087 unsigned int err_mask;
3088
3089 ata_tf_init(dev, &tf);
3090 if (park) {
3091 ehc->unloaded_mask |= 1 << dev->devno;
3092 tf.command = ATA_CMD_IDLEIMMEDIATE;
3093 tf.feature = 0x44;
3094 tf.lbal = 0x4c;
3095 tf.lbam = 0x4e;
3096 tf.lbah = 0x55;
3097 } else {
3098 ehc->unloaded_mask &= ~(1 << dev->devno);
3099 tf.command = ATA_CMD_CHK_POWER;
3100 }
3101
3102 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3103 tf.protocol |= ATA_PROT_NODATA;
3104 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3105 if (park && (err_mask || tf.lbal != 0xc4)) {
3106 ata_dev_err(dev, "head unload failed!\n");
3107 ehc->unloaded_mask &= ~(1 << dev->devno);
3108 }
3109 }
3110
3111 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3112 struct ata_device **r_failed_dev)
3113 {
3114 struct ata_port *ap = link->ap;
3115 struct ata_eh_context *ehc = &link->eh_context;
3116 struct ata_device *dev;
3117 unsigned int new_mask = 0;
3118 unsigned long flags;
3119 int rc = 0;
3120
3121 DPRINTK("ENTER\n");
3122
3123 /* For PATA drive side cable detection to work, IDENTIFY must
3124 * be done backwards such that PDIAG- is released by the slave
3125 * device before the master device is identified.
3126 */
3127 ata_for_each_dev(dev, link, ALL_REVERSE) {
3128 unsigned int action = ata_eh_dev_action(dev);
3129 unsigned int readid_flags = 0;
3130
3131 if (ehc->i.flags & ATA_EHI_DID_RESET)
3132 readid_flags |= ATA_READID_POSTRESET;
3133
3134 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3135 WARN_ON(dev->class == ATA_DEV_PMP);
3136
3137 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3138 rc = -EIO;
3139 goto err;
3140 }
3141
3142 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3143 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3144 readid_flags);
3145 if (rc)
3146 goto err;
3147
3148 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3149
3150 /* Configuration may have changed, reconfigure
3151 * transfer mode.
3152 */
3153 ehc->i.flags |= ATA_EHI_SETMODE;
3154
3155 /* schedule the scsi_rescan_device() here */
3156 schedule_work(&(ap->scsi_rescan_task));
3157 } else if (dev->class == ATA_DEV_UNKNOWN &&
3158 ehc->tries[dev->devno] &&
3159 ata_class_enabled(ehc->classes[dev->devno])) {
3160 /* Temporarily set dev->class, it will be
3161 * permanently set once all configurations are
3162 * complete. This is necessary because new
3163 * device configuration is done in two
3164 * separate loops.
3165 */
3166 dev->class = ehc->classes[dev->devno];
3167
3168 if (dev->class == ATA_DEV_PMP)
3169 rc = sata_pmp_attach(dev);
3170 else
3171 rc = ata_dev_read_id(dev, &dev->class,
3172 readid_flags, dev->id);
3173
3174 /* read_id might have changed class, store and reset */
3175 ehc->classes[dev->devno] = dev->class;
3176 dev->class = ATA_DEV_UNKNOWN;
3177
3178 switch (rc) {
3179 case 0:
3180 /* clear error info accumulated during probe */
3181 ata_ering_clear(&dev->ering);
3182 new_mask |= 1 << dev->devno;
3183 break;
3184 case -ENOENT:
3185 /* IDENTIFY was issued to non-existent
3186 * device. No need to reset. Just
3187 * thaw and ignore the device.
3188 */
3189 ata_eh_thaw_port(ap);
3190 break;
3191 default:
3192 goto err;
3193 }
3194 }
3195 }
3196
3197 /* PDIAG- should have been released, ask cable type if post-reset */
3198 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3199 if (ap->ops->cable_detect)
3200 ap->cbl = ap->ops->cable_detect(ap);
3201 ata_force_cbl(ap);
3202 }
3203
3204 /* Configure new devices forward such that user doesn't see
3205 * device detection messages backwards.
3206 */
3207 ata_for_each_dev(dev, link, ALL) {
3208 if (!(new_mask & (1 << dev->devno)))
3209 continue;
3210
3211 dev->class = ehc->classes[dev->devno];
3212
3213 if (dev->class == ATA_DEV_PMP)
3214 continue;
3215
3216 ehc->i.flags |= ATA_EHI_PRINTINFO;
3217 rc = ata_dev_configure(dev);
3218 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3219 if (rc) {
3220 dev->class = ATA_DEV_UNKNOWN;
3221 goto err;
3222 }
3223
3224 spin_lock_irqsave(ap->lock, flags);
3225 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3226 spin_unlock_irqrestore(ap->lock, flags);
3227
3228 /* new device discovered, configure xfermode */
3229 ehc->i.flags |= ATA_EHI_SETMODE;
3230 }
3231
3232 return 0;
3233
3234 err:
3235 *r_failed_dev = dev;
3236 DPRINTK("EXIT rc=%d\n", rc);
3237 return rc;
3238 }
3239
3240 /**
3241 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3242 * @link: link on which timings will be programmed
3243 * @r_failed_dev: out parameter for failed device
3244 *
3245 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3246 * ata_set_mode() fails, pointer to the failing device is
3247 * returned in @r_failed_dev.
3248 *
3249 * LOCKING:
3250 * PCI/etc. bus probe sem.
3251 *
3252 * RETURNS:
3253 * 0 on success, negative errno otherwise
3254 */
3255 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3256 {
3257 struct ata_port *ap = link->ap;
3258 struct ata_device *dev;
3259 int rc;
3260
3261 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3262 ata_for_each_dev(dev, link, ENABLED) {
3263 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3264 struct ata_ering_entry *ent;
3265
3266 ent = ata_ering_top(&dev->ering);
3267 if (ent)
3268 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3269 }
3270 }
3271
3272 /* has private set_mode? */
3273 if (ap->ops->set_mode)
3274 rc = ap->ops->set_mode(link, r_failed_dev);
3275 else
3276 rc = ata_do_set_mode(link, r_failed_dev);
3277
3278 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3279 ata_for_each_dev(dev, link, ENABLED) {
3280 struct ata_eh_context *ehc = &link->eh_context;
3281 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3282 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3283
3284 if (dev->xfer_mode != saved_xfer_mode ||
3285 ata_ncq_enabled(dev) != saved_ncq)
3286 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3287 }
3288
3289 return rc;
3290 }
3291
3292 /**
3293 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3294 * @dev: ATAPI device to clear UA for
3295 *
3296 * Resets and other operations can make an ATAPI device raise
3297 * UNIT ATTENTION which causes the next operation to fail. This
3298 * function clears UA.
3299 *
3300 * LOCKING:
3301 * EH context (may sleep).
3302 *
3303 * RETURNS:
3304 * 0 on success, -errno on failure.
3305 */
3306 static int atapi_eh_clear_ua(struct ata_device *dev)
3307 {
3308 int i;
3309
3310 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3311 u8 *sense_buffer = dev->link->ap->sector_buf;
3312 u8 sense_key = 0;
3313 unsigned int err_mask;
3314
3315 err_mask = atapi_eh_tur(dev, &sense_key);
3316 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3317 ata_dev_warn(dev,
3318 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3319 err_mask);
3320 return -EIO;
3321 }
3322
3323 if (!err_mask || sense_key != UNIT_ATTENTION)
3324 return 0;
3325
3326 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3327 if (err_mask) {
3328 ata_dev_warn(dev, "failed to clear "
3329 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3330 return -EIO;
3331 }
3332 }
3333
3334 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3335 ATA_EH_UA_TRIES);
3336
3337 return 0;
3338 }
3339
3340 /**
3341 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3342 * @dev: ATA device which may need FLUSH retry
3343 *
3344 * If @dev failed FLUSH, it needs to be reported upper layer
3345 * immediately as it means that @dev failed to remap and already
3346 * lost at least a sector and further FLUSH retrials won't make
3347 * any difference to the lost sector. However, if FLUSH failed
3348 * for other reasons, for example transmission error, FLUSH needs
3349 * to be retried.
3350 *
3351 * This function determines whether FLUSH failure retry is
3352 * necessary and performs it if so.
3353 *
3354 * RETURNS:
3355 * 0 if EH can continue, -errno if EH needs to be repeated.
3356 */
3357 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3358 {
3359 struct ata_link *link = dev->link;
3360 struct ata_port *ap = link->ap;
3361 struct ata_queued_cmd *qc;
3362 struct ata_taskfile tf;
3363 unsigned int err_mask;
3364 int rc = 0;
3365
3366 /* did flush fail for this device? */
3367 if (!ata_tag_valid(link->active_tag))
3368 return 0;
3369
3370 qc = __ata_qc_from_tag(ap, link->active_tag);
3371 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3372 qc->tf.command != ATA_CMD_FLUSH))
3373 return 0;
3374
3375 /* if the device failed it, it should be reported to upper layers */
3376 if (qc->err_mask & AC_ERR_DEV)
3377 return 0;
3378
3379 /* flush failed for some other reason, give it another shot */
3380 ata_tf_init(dev, &tf);
3381
3382 tf.command = qc->tf.command;
3383 tf.flags |= ATA_TFLAG_DEVICE;
3384 tf.protocol = ATA_PROT_NODATA;
3385
3386 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3387 tf.command, qc->err_mask);
3388
3389 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3390 if (!err_mask) {
3391 /*
3392 * FLUSH is complete but there's no way to
3393 * successfully complete a failed command from EH.
3394 * Making sure retry is allowed at least once and
3395 * retrying it should do the trick - whatever was in
3396 * the cache is already on the platter and this won't
3397 * cause infinite loop.
3398 */
3399 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3400 } else {
3401 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3402 err_mask);
3403 rc = -EIO;
3404
3405 /* if device failed it, report it to upper layers */
3406 if (err_mask & AC_ERR_DEV) {
3407 qc->err_mask |= AC_ERR_DEV;
3408 qc->result_tf = tf;
3409 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3410 rc = 0;
3411 }
3412 }
3413 return rc;
3414 }
3415
3416 /**
3417 * ata_eh_set_lpm - configure SATA interface power management
3418 * @link: link to configure power management
3419 * @policy: the link power management policy
3420 * @r_failed_dev: out parameter for failed device
3421 *
3422 * Enable SATA Interface power management. This will enable
3423 * Device Interface Power Management (DIPM) for min_power
3424 * policy, and then call driver specific callbacks for
3425 * enabling Host Initiated Power management.
3426 *
3427 * LOCKING:
3428 * EH context.
3429 *
3430 * RETURNS:
3431 * 0 on success, -errno on failure.
3432 */
3433 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3434 struct ata_device **r_failed_dev)
3435 {
3436 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3437 struct ata_eh_context *ehc = &link->eh_context;
3438 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3439 enum ata_lpm_policy old_policy = link->lpm_policy;
3440 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3441 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3442 unsigned int err_mask;
3443 int rc;
3444
3445 /* if the link or host doesn't do LPM, noop */
3446 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3447 return 0;
3448
3449 /*
3450 * DIPM is enabled only for MIN_POWER as some devices
3451 * misbehave when the host NACKs transition to SLUMBER. Order
3452 * device and link configurations such that the host always
3453 * allows DIPM requests.
3454 */
3455 ata_for_each_dev(dev, link, ENABLED) {
3456 bool hipm = ata_id_has_hipm(dev->id);
3457 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3458
3459 /* find the first enabled and LPM enabled devices */
3460 if (!link_dev)
3461 link_dev = dev;
3462
3463 if (!lpm_dev && (hipm || dipm))
3464 lpm_dev = dev;
3465
3466 hints &= ~ATA_LPM_EMPTY;
3467 if (!hipm)
3468 hints &= ~ATA_LPM_HIPM;
3469
3470 /* disable DIPM before changing link config */
3471 if (policy != ATA_LPM_MIN_POWER && dipm) {
3472 err_mask = ata_dev_set_feature(dev,
3473 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3474 if (err_mask && err_mask != AC_ERR_DEV) {
3475 ata_dev_warn(dev,
3476 "failed to disable DIPM, Emask 0x%x\n",
3477 err_mask);
3478 rc = -EIO;
3479 goto fail;
3480 }
3481 }
3482 }
3483
3484 if (ap) {
3485 rc = ap->ops->set_lpm(link, policy, hints);
3486 if (!rc && ap->slave_link)
3487 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3488 } else
3489 rc = sata_pmp_set_lpm(link, policy, hints);
3490
3491 /*
3492 * Attribute link config failure to the first (LPM) enabled
3493 * device on the link.
3494 */
3495 if (rc) {
3496 if (rc == -EOPNOTSUPP) {
3497 link->flags |= ATA_LFLAG_NO_LPM;
3498 return 0;
3499 }
3500 dev = lpm_dev ? lpm_dev : link_dev;
3501 goto fail;
3502 }
3503
3504 /*
3505 * Low level driver acked the transition. Issue DIPM command
3506 * with the new policy set.
3507 */
3508 link->lpm_policy = policy;
3509 if (ap && ap->slave_link)
3510 ap->slave_link->lpm_policy = policy;
3511
3512 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3513 ata_for_each_dev(dev, link, ENABLED) {
3514 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3515 ata_id_has_dipm(dev->id)) {
3516 err_mask = ata_dev_set_feature(dev,
3517 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3518 if (err_mask && err_mask != AC_ERR_DEV) {
3519 ata_dev_warn(dev,
3520 "failed to enable DIPM, Emask 0x%x\n",
3521 err_mask);
3522 rc = -EIO;
3523 goto fail;
3524 }
3525 }
3526 }
3527
3528 link->last_lpm_change = jiffies;
3529 link->flags |= ATA_LFLAG_CHANGED;
3530
3531 return 0;
3532
3533 fail:
3534 /* restore the old policy */
3535 link->lpm_policy = old_policy;
3536 if (ap && ap->slave_link)
3537 ap->slave_link->lpm_policy = old_policy;
3538
3539 /* if no device or only one more chance is left, disable LPM */
3540 if (!dev || ehc->tries[dev->devno] <= 2) {
3541 ata_link_warn(link, "disabling LPM on the link\n");
3542 link->flags |= ATA_LFLAG_NO_LPM;
3543 }
3544 if (r_failed_dev)
3545 *r_failed_dev = dev;
3546 return rc;
3547 }
3548
3549 int ata_link_nr_enabled(struct ata_link *link)
3550 {
3551 struct ata_device *dev;
3552 int cnt = 0;
3553
3554 ata_for_each_dev(dev, link, ENABLED)
3555 cnt++;
3556 return cnt;
3557 }
3558
3559 static int ata_link_nr_vacant(struct ata_link *link)
3560 {
3561 struct ata_device *dev;
3562 int cnt = 0;
3563
3564 ata_for_each_dev(dev, link, ALL)
3565 if (dev->class == ATA_DEV_UNKNOWN)
3566 cnt++;
3567 return cnt;
3568 }
3569
3570 static int ata_eh_skip_recovery(struct ata_link *link)
3571 {
3572 struct ata_port *ap = link->ap;
3573 struct ata_eh_context *ehc = &link->eh_context;
3574 struct ata_device *dev;
3575
3576 /* skip disabled links */
3577 if (link->flags & ATA_LFLAG_DISABLED)
3578 return 1;
3579
3580 /* skip if explicitly requested */
3581 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3582 return 1;
3583
3584 /* thaw frozen port and recover failed devices */
3585 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3586 return 0;
3587
3588 /* reset at least once if reset is requested */
3589 if ((ehc->i.action & ATA_EH_RESET) &&
3590 !(ehc->i.flags & ATA_EHI_DID_RESET))
3591 return 0;
3592
3593 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3594 ata_for_each_dev(dev, link, ALL) {
3595 if (dev->class == ATA_DEV_UNKNOWN &&
3596 ehc->classes[dev->devno] != ATA_DEV_NONE)
3597 return 0;
3598 }
3599
3600 return 1;
3601 }
3602
3603 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3604 {
3605 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3606 u64 now = get_jiffies_64();
3607 int *trials = void_arg;
3608
3609 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3610 (ent->timestamp < now - min(now, interval)))
3611 return -1;
3612
3613 (*trials)++;
3614 return 0;
3615 }
3616
3617 static int ata_eh_schedule_probe(struct ata_device *dev)
3618 {
3619 struct ata_eh_context *ehc = &dev->link->eh_context;
3620 struct ata_link *link = ata_dev_phys_link(dev);
3621 int trials = 0;
3622
3623 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3624 (ehc->did_probe_mask & (1 << dev->devno)))
3625 return 0;
3626
3627 ata_eh_detach_dev(dev);
3628 ata_dev_init(dev);
3629 ehc->did_probe_mask |= (1 << dev->devno);
3630 ehc->i.action |= ATA_EH_RESET;
3631 ehc->saved_xfer_mode[dev->devno] = 0;
3632 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3633
3634 /* the link maybe in a deep sleep, wake it up */
3635 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3636 if (ata_is_host_link(link))
3637 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3638 ATA_LPM_EMPTY);
3639 else
3640 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3641 ATA_LPM_EMPTY);
3642 }
3643
3644 /* Record and count probe trials on the ering. The specific
3645 * error mask used is irrelevant. Because a successful device
3646 * detection clears the ering, this count accumulates only if
3647 * there are consecutive failed probes.
3648 *
3649 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3650 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3651 * forced to 1.5Gbps.
3652 *
3653 * This is to work around cases where failed link speed
3654 * negotiation results in device misdetection leading to
3655 * infinite DEVXCHG or PHRDY CHG events.
3656 */
3657 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3658 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3659
3660 if (trials > ATA_EH_PROBE_TRIALS)
3661 sata_down_spd_limit(link, 1);
3662
3663 return 1;
3664 }
3665
3666 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3667 {
3668 struct ata_eh_context *ehc = &dev->link->eh_context;
3669
3670 /* -EAGAIN from EH routine indicates retry without prejudice.
3671 * The requester is responsible for ensuring forward progress.
3672 */
3673 if (err != -EAGAIN)
3674 ehc->tries[dev->devno]--;
3675
3676 switch (err) {
3677 case -ENODEV:
3678 /* device missing or wrong IDENTIFY data, schedule probing */
3679 ehc->i.probe_mask |= (1 << dev->devno);
3680 case -EINVAL:
3681 /* give it just one more chance */
3682 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3683 case -EIO:
3684 if (ehc->tries[dev->devno] == 1) {
3685 /* This is the last chance, better to slow
3686 * down than lose it.
3687 */
3688 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3689 if (dev->pio_mode > XFER_PIO_0)
3690 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3691 }
3692 }
3693
3694 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3695 /* disable device if it has used up all its chances */
3696 ata_dev_disable(dev);
3697
3698 /* detach if offline */
3699 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3700 ata_eh_detach_dev(dev);
3701
3702 /* schedule probe if necessary */
3703 if (ata_eh_schedule_probe(dev)) {
3704 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3705 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3706 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3707 }
3708
3709 return 1;
3710 } else {
3711 ehc->i.action |= ATA_EH_RESET;
3712 return 0;
3713 }
3714 }
3715
3716 /**
3717 * ata_eh_recover - recover host port after error
3718 * @ap: host port to recover
3719 * @prereset: prereset method (can be NULL)
3720 * @softreset: softreset method (can be NULL)
3721 * @hardreset: hardreset method (can be NULL)
3722 * @postreset: postreset method (can be NULL)
3723 * @r_failed_link: out parameter for failed link
3724 *
3725 * This is the alpha and omega, eum and yang, heart and soul of
3726 * libata exception handling. On entry, actions required to
3727 * recover each link and hotplug requests are recorded in the
3728 * link's eh_context. This function executes all the operations
3729 * with appropriate retrials and fallbacks to resurrect failed
3730 * devices, detach goners and greet newcomers.
3731 *
3732 * LOCKING:
3733 * Kernel thread context (may sleep).
3734 *
3735 * RETURNS:
3736 * 0 on success, -errno on failure.
3737 */
3738 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3739 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3740 ata_postreset_fn_t postreset,
3741 struct ata_link **r_failed_link)
3742 {
3743 struct ata_link *link;
3744 struct ata_device *dev;
3745 int rc, nr_fails;
3746 unsigned long flags, deadline;
3747
3748 DPRINTK("ENTER\n");
3749
3750 /* prep for recovery */
3751 ata_for_each_link(link, ap, EDGE) {
3752 struct ata_eh_context *ehc = &link->eh_context;
3753
3754 /* re-enable link? */
3755 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3756 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3757 spin_lock_irqsave(ap->lock, flags);
3758 link->flags &= ~ATA_LFLAG_DISABLED;
3759 spin_unlock_irqrestore(ap->lock, flags);
3760 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3761 }
3762
3763 ata_for_each_dev(dev, link, ALL) {
3764 if (link->flags & ATA_LFLAG_NO_RETRY)
3765 ehc->tries[dev->devno] = 1;
3766 else
3767 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3768
3769 /* collect port action mask recorded in dev actions */
3770 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3771 ~ATA_EH_PERDEV_MASK;
3772 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3773
3774 /* process hotplug request */
3775 if (dev->flags & ATA_DFLAG_DETACH)
3776 ata_eh_detach_dev(dev);
3777
3778 /* schedule probe if necessary */
3779 if (!ata_dev_enabled(dev))
3780 ata_eh_schedule_probe(dev);
3781 }
3782 }
3783
3784 retry:
3785 rc = 0;
3786
3787 /* if UNLOADING, finish immediately */
3788 if (ap->pflags & ATA_PFLAG_UNLOADING)
3789 goto out;
3790
3791 /* prep for EH */
3792 ata_for_each_link(link, ap, EDGE) {
3793 struct ata_eh_context *ehc = &link->eh_context;
3794
3795 /* skip EH if possible. */
3796 if (ata_eh_skip_recovery(link))
3797 ehc->i.action = 0;
3798
3799 ata_for_each_dev(dev, link, ALL)
3800 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3801 }
3802
3803 /* reset */
3804 ata_for_each_link(link, ap, EDGE) {
3805 struct ata_eh_context *ehc = &link->eh_context;
3806
3807 if (!(ehc->i.action & ATA_EH_RESET))
3808 continue;
3809
3810 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3811 prereset, softreset, hardreset, postreset);
3812 if (rc) {
3813 ata_link_err(link, "reset failed, giving up\n");
3814 goto out;
3815 }
3816 }
3817
3818 do {
3819 unsigned long now;
3820
3821 /*
3822 * clears ATA_EH_PARK in eh_info and resets
3823 * ap->park_req_pending
3824 */
3825 ata_eh_pull_park_action(ap);
3826
3827 deadline = jiffies;
3828 ata_for_each_link(link, ap, EDGE) {
3829 ata_for_each_dev(dev, link, ALL) {
3830 struct ata_eh_context *ehc = &link->eh_context;
3831 unsigned long tmp;
3832
3833 if (dev->class != ATA_DEV_ATA &&
3834 dev->class != ATA_DEV_ZAC)
3835 continue;
3836 if (!(ehc->i.dev_action[dev->devno] &
3837 ATA_EH_PARK))
3838 continue;
3839 tmp = dev->unpark_deadline;
3840 if (time_before(deadline, tmp))
3841 deadline = tmp;
3842 else if (time_before_eq(tmp, jiffies))
3843 continue;
3844 if (ehc->unloaded_mask & (1 << dev->devno))
3845 continue;
3846
3847 ata_eh_park_issue_cmd(dev, 1);
3848 }
3849 }
3850
3851 now = jiffies;
3852 if (time_before_eq(deadline, now))
3853 break;
3854
3855 ata_eh_release(ap);
3856 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3857 deadline - now);
3858 ata_eh_acquire(ap);
3859 } while (deadline);
3860 ata_for_each_link(link, ap, EDGE) {
3861 ata_for_each_dev(dev, link, ALL) {
3862 if (!(link->eh_context.unloaded_mask &
3863 (1 << dev->devno)))
3864 continue;
3865
3866 ata_eh_park_issue_cmd(dev, 0);
3867 ata_eh_done(link, dev, ATA_EH_PARK);
3868 }
3869 }
3870
3871 /* the rest */
3872 nr_fails = 0;
3873 ata_for_each_link(link, ap, PMP_FIRST) {
3874 struct ata_eh_context *ehc = &link->eh_context;
3875
3876 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3877 goto config_lpm;
3878
3879 /* revalidate existing devices and attach new ones */
3880 rc = ata_eh_revalidate_and_attach(link, &dev);
3881 if (rc)
3882 goto rest_fail;
3883
3884 /* if PMP got attached, return, pmp EH will take care of it */
3885 if (link->device->class == ATA_DEV_PMP) {
3886 ehc->i.action = 0;
3887 return 0;
3888 }
3889
3890 /* configure transfer mode if necessary */
3891 if (ehc->i.flags & ATA_EHI_SETMODE) {
3892 rc = ata_set_mode(link, &dev);
3893 if (rc)
3894 goto rest_fail;
3895 ehc->i.flags &= ~ATA_EHI_SETMODE;
3896 }
3897
3898 /* If reset has been issued, clear UA to avoid
3899 * disrupting the current users of the device.
3900 */
3901 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3902 ata_for_each_dev(dev, link, ALL) {
3903 if (dev->class != ATA_DEV_ATAPI)
3904 continue;
3905 rc = atapi_eh_clear_ua(dev);
3906 if (rc)
3907 goto rest_fail;
3908 if (zpodd_dev_enabled(dev))
3909 zpodd_post_poweron(dev);
3910 }
3911 }
3912
3913 /* retry flush if necessary */
3914 ata_for_each_dev(dev, link, ALL) {
3915 if (dev->class != ATA_DEV_ATA &&
3916 dev->class != ATA_DEV_ZAC)
3917 continue;
3918 rc = ata_eh_maybe_retry_flush(dev);
3919 if (rc)
3920 goto rest_fail;
3921 }
3922
3923 config_lpm:
3924 /* configure link power saving */
3925 if (link->lpm_policy != ap->target_lpm_policy) {
3926 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3927 if (rc)
3928 goto rest_fail;
3929 }
3930
3931 /* this link is okay now */
3932 ehc->i.flags = 0;
3933 continue;
3934
3935 rest_fail:
3936 nr_fails++;
3937 if (dev)
3938 ata_eh_handle_dev_fail(dev, rc);
3939
3940 if (ap->pflags & ATA_PFLAG_FROZEN) {
3941 /* PMP reset requires working host port.
3942 * Can't retry if it's frozen.
3943 */
3944 if (sata_pmp_attached(ap))
3945 goto out;
3946 break;
3947 }
3948 }
3949
3950 if (nr_fails)
3951 goto retry;
3952
3953 out:
3954 if (rc && r_failed_link)
3955 *r_failed_link = link;
3956
3957 DPRINTK("EXIT, rc=%d\n", rc);
3958 return rc;
3959 }
3960
3961 /**
3962 * ata_eh_finish - finish up EH
3963 * @ap: host port to finish EH for
3964 *
3965 * Recovery is complete. Clean up EH states and retry or finish
3966 * failed qcs.
3967 *
3968 * LOCKING:
3969 * None.
3970 */
3971 void ata_eh_finish(struct ata_port *ap)
3972 {
3973 int tag;
3974
3975 /* retry or finish qcs */
3976 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3977 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3978
3979 if (!(qc->flags & ATA_QCFLAG_FAILED))
3980 continue;
3981
3982 if (qc->err_mask) {
3983 /* FIXME: Once EH migration is complete,
3984 * generate sense data in this function,
3985 * considering both err_mask and tf.
3986 */
3987 if (qc->flags & ATA_QCFLAG_RETRY)
3988 ata_eh_qc_retry(qc);
3989 else
3990 ata_eh_qc_complete(qc);
3991 } else {
3992 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3993 ata_eh_qc_complete(qc);
3994 } else {
3995 /* feed zero TF to sense generation */
3996 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3997 ata_eh_qc_retry(qc);
3998 }
3999 }
4000 }
4001
4002 /* make sure nr_active_links is zero after EH */
4003 WARN_ON(ap->nr_active_links);
4004 ap->nr_active_links = 0;
4005 }
4006
4007 /**
4008 * ata_do_eh - do standard error handling
4009 * @ap: host port to handle error for
4010 *
4011 * @prereset: prereset method (can be NULL)
4012 * @softreset: softreset method (can be NULL)
4013 * @hardreset: hardreset method (can be NULL)
4014 * @postreset: postreset method (can be NULL)
4015 *
4016 * Perform standard error handling sequence.
4017 *
4018 * LOCKING:
4019 * Kernel thread context (may sleep).
4020 */
4021 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4022 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4023 ata_postreset_fn_t postreset)
4024 {
4025 struct ata_device *dev;
4026 int rc;
4027
4028 ata_eh_autopsy(ap);
4029 ata_eh_report(ap);
4030
4031 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4032 NULL);
4033 if (rc) {
4034 ata_for_each_dev(dev, &ap->link, ALL)
4035 ata_dev_disable(dev);
4036 }
4037
4038 ata_eh_finish(ap);
4039 }
4040
4041 /**
4042 * ata_std_error_handler - standard error handler
4043 * @ap: host port to handle error for
4044 *
4045 * Standard error handler
4046 *
4047 * LOCKING:
4048 * Kernel thread context (may sleep).
4049 */
4050 void ata_std_error_handler(struct ata_port *ap)
4051 {
4052 struct ata_port_operations *ops = ap->ops;
4053 ata_reset_fn_t hardreset = ops->hardreset;
4054
4055 /* ignore built-in hardreset if SCR access is not available */
4056 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4057 hardreset = NULL;
4058
4059 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4060 }
4061
4062 #ifdef CONFIG_PM
4063 /**
4064 * ata_eh_handle_port_suspend - perform port suspend operation
4065 * @ap: port to suspend
4066 *
4067 * Suspend @ap.
4068 *
4069 * LOCKING:
4070 * Kernel thread context (may sleep).
4071 */
4072 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4073 {
4074 unsigned long flags;
4075 int rc = 0;
4076 struct ata_device *dev;
4077
4078 /* are we suspending? */
4079 spin_lock_irqsave(ap->lock, flags);
4080 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4081 ap->pm_mesg.event & PM_EVENT_RESUME) {
4082 spin_unlock_irqrestore(ap->lock, flags);
4083 return;
4084 }
4085 spin_unlock_irqrestore(ap->lock, flags);
4086
4087 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4088
4089 /*
4090 * If we have a ZPODD attached, check its zero
4091 * power ready status before the port is frozen.
4092 * Only needed for runtime suspend.
4093 */
4094 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4095 ata_for_each_dev(dev, &ap->link, ENABLED) {
4096 if (zpodd_dev_enabled(dev))
4097 zpodd_on_suspend(dev);
4098 }
4099 }
4100
4101 /* tell ACPI we're suspending */
4102 rc = ata_acpi_on_suspend(ap);
4103 if (rc)
4104 goto out;
4105
4106 /* suspend */
4107 ata_eh_freeze_port(ap);
4108
4109 if (ap->ops->port_suspend)
4110 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4111
4112 ata_acpi_set_state(ap, ap->pm_mesg);
4113 out:
4114 /* update the flags */
4115 spin_lock_irqsave(ap->lock, flags);
4116
4117 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4118 if (rc == 0)
4119 ap->pflags |= ATA_PFLAG_SUSPENDED;
4120 else if (ap->pflags & ATA_PFLAG_FROZEN)
4121 ata_port_schedule_eh(ap);
4122
4123 spin_unlock_irqrestore(ap->lock, flags);
4124
4125 return;
4126 }
4127
4128 /**
4129 * ata_eh_handle_port_resume - perform port resume operation
4130 * @ap: port to resume
4131 *
4132 * Resume @ap.
4133 *
4134 * LOCKING:
4135 * Kernel thread context (may sleep).
4136 */
4137 static void ata_eh_handle_port_resume(struct ata_port *ap)
4138 {
4139 struct ata_link *link;
4140 struct ata_device *dev;
4141 unsigned long flags;
4142 int rc = 0;
4143
4144 /* are we resuming? */
4145 spin_lock_irqsave(ap->lock, flags);
4146 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4147 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4148 spin_unlock_irqrestore(ap->lock, flags);
4149 return;
4150 }
4151 spin_unlock_irqrestore(ap->lock, flags);
4152
4153 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4154
4155 /*
4156 * Error timestamps are in jiffies which doesn't run while
4157 * suspended and PHY events during resume isn't too uncommon.
4158 * When the two are combined, it can lead to unnecessary speed
4159 * downs if the machine is suspended and resumed repeatedly.
4160 * Clear error history.
4161 */
4162 ata_for_each_link(link, ap, HOST_FIRST)
4163 ata_for_each_dev(dev, link, ALL)
4164 ata_ering_clear(&dev->ering);
4165
4166 ata_acpi_set_state(ap, ap->pm_mesg);
4167
4168 if (ap->ops->port_resume)
4169 rc = ap->ops->port_resume(ap);
4170
4171 /* tell ACPI that we're resuming */
4172 ata_acpi_on_resume(ap);
4173
4174 /* update the flags */
4175 spin_lock_irqsave(ap->lock, flags);
4176 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4177 spin_unlock_irqrestore(ap->lock, flags);
4178 }
4179 #endif /* CONFIG_PM */
This page took 0.118989 seconds and 5 git commands to generate.