Revert "libata-eh: Set 'information' field for autosense"
[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 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1634 * @dev: device to perform REQUEST_SENSE_SENSE_DATA_EXT 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 REQUEST_SENSE_DATA_EXT 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 * encoded sense data on success, 0 on failure or if sense data
1646 * is not available.
1647 */
1648 static u32 ata_eh_request_sense(struct ata_queued_cmd *qc,
1649 struct scsi_cmnd *cmd)
1650 {
1651 struct ata_device *dev = qc->dev;
1652 struct ata_taskfile tf;
1653 unsigned int err_mask;
1654
1655 if (!cmd)
1656 return 0;
1657
1658 DPRINTK("ATA request sense\n");
1659 ata_dev_warn(dev, "request sense\n");
1660 if (!ata_id_sense_reporting_enabled(dev->id)) {
1661 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1662 return 0;
1663 }
1664 ata_tf_init(dev, &tf);
1665
1666 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1667 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1668 tf.command = ATA_CMD_REQ_SENSE_DATA;
1669 tf.protocol = ATA_PROT_NODATA;
1670
1671 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1672 /*
1673 * ACS-4 states:
1674 * The device may set the SENSE DATA AVAILABLE bit to one in the
1675 * STATUS field and clear the ERROR bit to zero in the STATUS field
1676 * to indicate that the command returned completion without an error
1677 * and the sense data described in table 306 is available.
1678 *
1679 * IOW the 'ATA_SENSE' bit might not be set even though valid
1680 * sense data is available.
1681 * So check for both.
1682 */
1683 if ((tf.command & ATA_SENSE) ||
1684 tf.lbah != 0 || tf.lbam != 0 || tf.lbal != 0) {
1685 ata_scsi_set_sense(cmd, tf.lbah, tf.lbam, tf.lbal);
1686 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1687 ata_dev_warn(dev, "sense data %02x/%02x/%02x\n",
1688 tf.lbah, tf.lbam, tf.lbal);
1689 } else {
1690 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1691 tf.command, err_mask);
1692 }
1693 return err_mask;
1694 }
1695
1696 /**
1697 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1698 * @dev: device to perform REQUEST_SENSE to
1699 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1700 * @dfl_sense_key: default sense key to use
1701 *
1702 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1703 * SENSE. This function is EH helper.
1704 *
1705 * LOCKING:
1706 * Kernel thread context (may sleep).
1707 *
1708 * RETURNS:
1709 * 0 on success, AC_ERR_* mask on failure
1710 */
1711 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1712 u8 *sense_buf, u8 dfl_sense_key)
1713 {
1714 u8 cdb[ATAPI_CDB_LEN] =
1715 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1716 struct ata_port *ap = dev->link->ap;
1717 struct ata_taskfile tf;
1718
1719 DPRINTK("ATAPI request sense\n");
1720
1721 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1722
1723 /* initialize sense_buf with the error register,
1724 * for the case where they are -not- overwritten
1725 */
1726 sense_buf[0] = 0x70;
1727 sense_buf[2] = dfl_sense_key;
1728
1729 /* some devices time out if garbage left in tf */
1730 ata_tf_init(dev, &tf);
1731
1732 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1733 tf.command = ATA_CMD_PACKET;
1734
1735 /* is it pointless to prefer PIO for "safety reasons"? */
1736 if (ap->flags & ATA_FLAG_PIO_DMA) {
1737 tf.protocol = ATAPI_PROT_DMA;
1738 tf.feature |= ATAPI_PKT_DMA;
1739 } else {
1740 tf.protocol = ATAPI_PROT_PIO;
1741 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1742 tf.lbah = 0;
1743 }
1744
1745 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1746 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1747 }
1748
1749 /**
1750 * ata_eh_analyze_serror - analyze SError for a failed port
1751 * @link: ATA link to analyze SError for
1752 *
1753 * Analyze SError if available and further determine cause of
1754 * failure.
1755 *
1756 * LOCKING:
1757 * None.
1758 */
1759 static void ata_eh_analyze_serror(struct ata_link *link)
1760 {
1761 struct ata_eh_context *ehc = &link->eh_context;
1762 u32 serror = ehc->i.serror;
1763 unsigned int err_mask = 0, action = 0;
1764 u32 hotplug_mask;
1765
1766 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1767 err_mask |= AC_ERR_ATA_BUS;
1768 action |= ATA_EH_RESET;
1769 }
1770 if (serror & SERR_PROTOCOL) {
1771 err_mask |= AC_ERR_HSM;
1772 action |= ATA_EH_RESET;
1773 }
1774 if (serror & SERR_INTERNAL) {
1775 err_mask |= AC_ERR_SYSTEM;
1776 action |= ATA_EH_RESET;
1777 }
1778
1779 /* Determine whether a hotplug event has occurred. Both
1780 * SError.N/X are considered hotplug events for enabled or
1781 * host links. For disabled PMP links, only N bit is
1782 * considered as X bit is left at 1 for link plugging.
1783 */
1784 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1785 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1786 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1787 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1788 else
1789 hotplug_mask = SERR_PHYRDY_CHG;
1790
1791 if (serror & hotplug_mask)
1792 ata_ehi_hotplugged(&ehc->i);
1793
1794 ehc->i.err_mask |= err_mask;
1795 ehc->i.action |= action;
1796 }
1797
1798 /**
1799 * ata_eh_analyze_ncq_error - analyze NCQ error
1800 * @link: ATA link to analyze NCQ error for
1801 *
1802 * Read log page 10h, determine the offending qc and acquire
1803 * error status TF. For NCQ device errors, all LLDDs have to do
1804 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1805 * care of the rest.
1806 *
1807 * LOCKING:
1808 * Kernel thread context (may sleep).
1809 */
1810 void ata_eh_analyze_ncq_error(struct ata_link *link)
1811 {
1812 struct ata_port *ap = link->ap;
1813 struct ata_eh_context *ehc = &link->eh_context;
1814 struct ata_device *dev = link->device;
1815 struct ata_queued_cmd *qc;
1816 struct ata_taskfile tf;
1817 int tag, rc;
1818
1819 /* if frozen, we can't do much */
1820 if (ap->pflags & ATA_PFLAG_FROZEN)
1821 return;
1822
1823 /* is it NCQ device error? */
1824 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1825 return;
1826
1827 /* has LLDD analyzed already? */
1828 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1829 qc = __ata_qc_from_tag(ap, tag);
1830
1831 if (!(qc->flags & ATA_QCFLAG_FAILED))
1832 continue;
1833
1834 if (qc->err_mask)
1835 return;
1836 }
1837
1838 /* okay, this error is ours */
1839 memset(&tf, 0, sizeof(tf));
1840 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1841 if (rc) {
1842 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1843 rc);
1844 return;
1845 }
1846
1847 if (!(link->sactive & (1 << tag))) {
1848 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1849 tag);
1850 return;
1851 }
1852
1853 /* we've got the perpetrator, condemn it */
1854 qc = __ata_qc_from_tag(ap, tag);
1855 memcpy(&qc->result_tf, &tf, sizeof(tf));
1856 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1857 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1858 if (qc->result_tf.auxiliary) {
1859 char sense_key, asc, ascq;
1860
1861 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1862 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1863 ascq = qc->result_tf.auxiliary & 0xff;
1864 ata_dev_dbg(dev, "NCQ Autosense %02x/%02x/%02x\n",
1865 sense_key, asc, ascq);
1866 ata_scsi_set_sense(qc->scsicmd, sense_key, asc, ascq);
1867 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1868 }
1869
1870 ehc->i.err_mask &= ~AC_ERR_DEV;
1871 }
1872
1873 /**
1874 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1875 * @qc: qc to analyze
1876 * @tf: Taskfile registers to analyze
1877 *
1878 * Analyze taskfile of @qc and further determine cause of
1879 * failure. This function also requests ATAPI sense data if
1880 * available.
1881 *
1882 * LOCKING:
1883 * Kernel thread context (may sleep).
1884 *
1885 * RETURNS:
1886 * Determined recovery action
1887 */
1888 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1889 const struct ata_taskfile *tf)
1890 {
1891 unsigned int tmp, action = 0;
1892 u8 stat = tf->command, err = tf->feature;
1893
1894 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1895 qc->err_mask |= AC_ERR_HSM;
1896 return ATA_EH_RESET;
1897 }
1898
1899 /*
1900 * Sense data reporting does not work if the
1901 * device fault bit is set.
1902 */
1903 if ((stat & ATA_SENSE) && !(stat & ATA_DF) &&
1904 !(qc->flags & ATA_QCFLAG_SENSE_VALID)) {
1905 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1906 tmp = ata_eh_request_sense(qc, qc->scsicmd);
1907 if (tmp)
1908 qc->err_mask |= tmp;
1909 } else {
1910 ata_dev_warn(qc->dev, "sense data available but port frozen\n");
1911 }
1912 }
1913
1914 /* Set by NCQ autosense or request sense above */
1915 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1916 return 0;
1917
1918 if (stat & (ATA_ERR | ATA_DF))
1919 qc->err_mask |= AC_ERR_DEV;
1920 else
1921 return 0;
1922
1923 switch (qc->dev->class) {
1924 case ATA_DEV_ATA:
1925 case ATA_DEV_ZAC:
1926 if (err & ATA_ICRC)
1927 qc->err_mask |= AC_ERR_ATA_BUS;
1928 if (err & (ATA_UNC | ATA_AMNF))
1929 qc->err_mask |= AC_ERR_MEDIA;
1930 if (err & ATA_IDNF)
1931 qc->err_mask |= AC_ERR_INVALID;
1932 break;
1933
1934 case ATA_DEV_ATAPI:
1935 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1936 tmp = atapi_eh_request_sense(qc->dev,
1937 qc->scsicmd->sense_buffer,
1938 qc->result_tf.feature >> 4);
1939 if (!tmp) {
1940 /* ATA_QCFLAG_SENSE_VALID is used to
1941 * tell atapi_qc_complete() that sense
1942 * data is already valid.
1943 *
1944 * TODO: interpret sense data and set
1945 * appropriate err_mask.
1946 */
1947 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1948 } else
1949 qc->err_mask |= tmp;
1950 }
1951 }
1952
1953 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1954 action |= ATA_EH_RESET;
1955
1956 return action;
1957 }
1958
1959 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1960 int *xfer_ok)
1961 {
1962 int base = 0;
1963
1964 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1965 *xfer_ok = 1;
1966
1967 if (!*xfer_ok)
1968 base = ATA_ECAT_DUBIOUS_NONE;
1969
1970 if (err_mask & AC_ERR_ATA_BUS)
1971 return base + ATA_ECAT_ATA_BUS;
1972
1973 if (err_mask & AC_ERR_TIMEOUT)
1974 return base + ATA_ECAT_TOUT_HSM;
1975
1976 if (eflags & ATA_EFLAG_IS_IO) {
1977 if (err_mask & AC_ERR_HSM)
1978 return base + ATA_ECAT_TOUT_HSM;
1979 if ((err_mask &
1980 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1981 return base + ATA_ECAT_UNK_DEV;
1982 }
1983
1984 return 0;
1985 }
1986
1987 struct speed_down_verdict_arg {
1988 u64 since;
1989 int xfer_ok;
1990 int nr_errors[ATA_ECAT_NR];
1991 };
1992
1993 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1994 {
1995 struct speed_down_verdict_arg *arg = void_arg;
1996 int cat;
1997
1998 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1999 return -1;
2000
2001 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
2002 &arg->xfer_ok);
2003 arg->nr_errors[cat]++;
2004
2005 return 0;
2006 }
2007
2008 /**
2009 * ata_eh_speed_down_verdict - Determine speed down verdict
2010 * @dev: Device of interest
2011 *
2012 * This function examines error ring of @dev and determines
2013 * whether NCQ needs to be turned off, transfer speed should be
2014 * stepped down, or falling back to PIO is necessary.
2015 *
2016 * ECAT_ATA_BUS : ATA_BUS error for any command
2017 *
2018 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
2019 * IO commands
2020 *
2021 * ECAT_UNK_DEV : Unknown DEV error for IO commands
2022 *
2023 * ECAT_DUBIOUS_* : Identical to above three but occurred while
2024 * data transfer hasn't been verified.
2025 *
2026 * Verdicts are
2027 *
2028 * NCQ_OFF : Turn off NCQ.
2029 *
2030 * SPEED_DOWN : Speed down transfer speed but don't fall back
2031 * to PIO.
2032 *
2033 * FALLBACK_TO_PIO : Fall back to PIO.
2034 *
2035 * Even if multiple verdicts are returned, only one action is
2036 * taken per error. An action triggered by non-DUBIOUS errors
2037 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
2038 * This is to expedite speed down decisions right after device is
2039 * initially configured.
2040 *
2041 * The followings are speed down rules. #1 and #2 deal with
2042 * DUBIOUS errors.
2043 *
2044 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
2045 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
2046 *
2047 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
2048 * occurred during last 5 mins, NCQ_OFF.
2049 *
2050 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
2051 * occurred during last 5 mins, FALLBACK_TO_PIO
2052 *
2053 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
2054 * during last 10 mins, NCQ_OFF.
2055 *
2056 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
2057 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
2058 *
2059 * LOCKING:
2060 * Inherited from caller.
2061 *
2062 * RETURNS:
2063 * OR of ATA_EH_SPDN_* flags.
2064 */
2065 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
2066 {
2067 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
2068 u64 j64 = get_jiffies_64();
2069 struct speed_down_verdict_arg arg;
2070 unsigned int verdict = 0;
2071
2072 /* scan past 5 mins of error history */
2073 memset(&arg, 0, sizeof(arg));
2074 arg.since = j64 - min(j64, j5mins);
2075 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2076
2077 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
2078 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
2079 verdict |= ATA_EH_SPDN_SPEED_DOWN |
2080 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
2081
2082 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
2083 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
2084 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
2085
2086 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2087 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2088 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2089 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
2090
2091 /* scan past 10 mins of error history */
2092 memset(&arg, 0, sizeof(arg));
2093 arg.since = j64 - min(j64, j10mins);
2094 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2095
2096 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2097 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
2098 verdict |= ATA_EH_SPDN_NCQ_OFF;
2099
2100 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2101 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
2102 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2103 verdict |= ATA_EH_SPDN_SPEED_DOWN;
2104
2105 return verdict;
2106 }
2107
2108 /**
2109 * ata_eh_speed_down - record error and speed down if necessary
2110 * @dev: Failed device
2111 * @eflags: mask of ATA_EFLAG_* flags
2112 * @err_mask: err_mask of the error
2113 *
2114 * Record error and examine error history to determine whether
2115 * adjusting transmission speed is necessary. It also sets
2116 * transmission limits appropriately if such adjustment is
2117 * necessary.
2118 *
2119 * LOCKING:
2120 * Kernel thread context (may sleep).
2121 *
2122 * RETURNS:
2123 * Determined recovery action.
2124 */
2125 static unsigned int ata_eh_speed_down(struct ata_device *dev,
2126 unsigned int eflags, unsigned int err_mask)
2127 {
2128 struct ata_link *link = ata_dev_phys_link(dev);
2129 int xfer_ok = 0;
2130 unsigned int verdict;
2131 unsigned int action = 0;
2132
2133 /* don't bother if Cat-0 error */
2134 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2135 return 0;
2136
2137 /* record error and determine whether speed down is necessary */
2138 ata_ering_record(&dev->ering, eflags, err_mask);
2139 verdict = ata_eh_speed_down_verdict(dev);
2140
2141 /* turn off NCQ? */
2142 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2143 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2144 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2145 dev->flags |= ATA_DFLAG_NCQ_OFF;
2146 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2147 goto done;
2148 }
2149
2150 /* speed down? */
2151 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2152 /* speed down SATA link speed if possible */
2153 if (sata_down_spd_limit(link, 0) == 0) {
2154 action |= ATA_EH_RESET;
2155 goto done;
2156 }
2157
2158 /* lower transfer mode */
2159 if (dev->spdn_cnt < 2) {
2160 static const int dma_dnxfer_sel[] =
2161 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2162 static const int pio_dnxfer_sel[] =
2163 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2164 int sel;
2165
2166 if (dev->xfer_shift != ATA_SHIFT_PIO)
2167 sel = dma_dnxfer_sel[dev->spdn_cnt];
2168 else
2169 sel = pio_dnxfer_sel[dev->spdn_cnt];
2170
2171 dev->spdn_cnt++;
2172
2173 if (ata_down_xfermask_limit(dev, sel) == 0) {
2174 action |= ATA_EH_RESET;
2175 goto done;
2176 }
2177 }
2178 }
2179
2180 /* Fall back to PIO? Slowing down to PIO is meaningless for
2181 * SATA ATA devices. Consider it only for PATA and SATAPI.
2182 */
2183 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2184 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2185 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2186 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2187 dev->spdn_cnt = 0;
2188 action |= ATA_EH_RESET;
2189 goto done;
2190 }
2191 }
2192
2193 return 0;
2194 done:
2195 /* device has been slowed down, blow error history */
2196 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2197 ata_ering_clear(&dev->ering);
2198 return action;
2199 }
2200
2201 /**
2202 * ata_eh_worth_retry - analyze error and decide whether to retry
2203 * @qc: qc to possibly retry
2204 *
2205 * Look at the cause of the error and decide if a retry
2206 * might be useful or not. We don't want to retry media errors
2207 * because the drive itself has probably already taken 10-30 seconds
2208 * doing its own internal retries before reporting the failure.
2209 */
2210 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2211 {
2212 if (qc->err_mask & AC_ERR_MEDIA)
2213 return 0; /* don't retry media errors */
2214 if (qc->flags & ATA_QCFLAG_IO)
2215 return 1; /* otherwise retry anything from fs stack */
2216 if (qc->err_mask & AC_ERR_INVALID)
2217 return 0; /* don't retry these */
2218 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2219 }
2220
2221 /**
2222 * ata_eh_link_autopsy - analyze error and determine recovery action
2223 * @link: host link to perform autopsy on
2224 *
2225 * Analyze why @link failed and determine which recovery actions
2226 * are needed. This function also sets more detailed AC_ERR_*
2227 * values and fills sense data for ATAPI CHECK SENSE.
2228 *
2229 * LOCKING:
2230 * Kernel thread context (may sleep).
2231 */
2232 static void ata_eh_link_autopsy(struct ata_link *link)
2233 {
2234 struct ata_port *ap = link->ap;
2235 struct ata_eh_context *ehc = &link->eh_context;
2236 struct ata_device *dev;
2237 unsigned int all_err_mask = 0, eflags = 0;
2238 int tag;
2239 u32 serror;
2240 int rc;
2241
2242 DPRINTK("ENTER\n");
2243
2244 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2245 return;
2246
2247 /* obtain and analyze SError */
2248 rc = sata_scr_read(link, SCR_ERROR, &serror);
2249 if (rc == 0) {
2250 ehc->i.serror |= serror;
2251 ata_eh_analyze_serror(link);
2252 } else if (rc != -EOPNOTSUPP) {
2253 /* SError read failed, force reset and probing */
2254 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2255 ehc->i.action |= ATA_EH_RESET;
2256 ehc->i.err_mask |= AC_ERR_OTHER;
2257 }
2258
2259 /* analyze NCQ failure */
2260 ata_eh_analyze_ncq_error(link);
2261
2262 /* any real error trumps AC_ERR_OTHER */
2263 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2264 ehc->i.err_mask &= ~AC_ERR_OTHER;
2265
2266 all_err_mask |= ehc->i.err_mask;
2267
2268 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2269 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2270
2271 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2272 ata_dev_phys_link(qc->dev) != link)
2273 continue;
2274
2275 /* inherit upper level err_mask */
2276 qc->err_mask |= ehc->i.err_mask;
2277
2278 /* analyze TF */
2279 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2280
2281 /* DEV errors are probably spurious in case of ATA_BUS error */
2282 if (qc->err_mask & AC_ERR_ATA_BUS)
2283 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2284 AC_ERR_INVALID);
2285
2286 /* any real error trumps unknown error */
2287 if (qc->err_mask & ~AC_ERR_OTHER)
2288 qc->err_mask &= ~AC_ERR_OTHER;
2289
2290 /* SENSE_VALID trumps dev/unknown error and revalidation */
2291 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2292 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2293
2294 /* determine whether the command is worth retrying */
2295 if (ata_eh_worth_retry(qc))
2296 qc->flags |= ATA_QCFLAG_RETRY;
2297
2298 /* accumulate error info */
2299 ehc->i.dev = qc->dev;
2300 all_err_mask |= qc->err_mask;
2301 if (qc->flags & ATA_QCFLAG_IO)
2302 eflags |= ATA_EFLAG_IS_IO;
2303 trace_ata_eh_link_autopsy_qc(qc);
2304 }
2305
2306 /* enforce default EH actions */
2307 if (ap->pflags & ATA_PFLAG_FROZEN ||
2308 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2309 ehc->i.action |= ATA_EH_RESET;
2310 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2311 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2312 ehc->i.action |= ATA_EH_REVALIDATE;
2313
2314 /* If we have offending qcs and the associated failed device,
2315 * perform per-dev EH action only on the offending device.
2316 */
2317 if (ehc->i.dev) {
2318 ehc->i.dev_action[ehc->i.dev->devno] |=
2319 ehc->i.action & ATA_EH_PERDEV_MASK;
2320 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2321 }
2322
2323 /* propagate timeout to host link */
2324 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2325 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2326
2327 /* record error and consider speeding down */
2328 dev = ehc->i.dev;
2329 if (!dev && ((ata_link_max_devices(link) == 1 &&
2330 ata_dev_enabled(link->device))))
2331 dev = link->device;
2332
2333 if (dev) {
2334 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2335 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2336 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2337 }
2338 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2339 DPRINTK("EXIT\n");
2340 }
2341
2342 /**
2343 * ata_eh_autopsy - analyze error and determine recovery action
2344 * @ap: host port to perform autopsy on
2345 *
2346 * Analyze all links of @ap and determine why they failed and
2347 * which recovery actions are needed.
2348 *
2349 * LOCKING:
2350 * Kernel thread context (may sleep).
2351 */
2352 void ata_eh_autopsy(struct ata_port *ap)
2353 {
2354 struct ata_link *link;
2355
2356 ata_for_each_link(link, ap, EDGE)
2357 ata_eh_link_autopsy(link);
2358
2359 /* Handle the frigging slave link. Autopsy is done similarly
2360 * but actions and flags are transferred over to the master
2361 * link and handled from there.
2362 */
2363 if (ap->slave_link) {
2364 struct ata_eh_context *mehc = &ap->link.eh_context;
2365 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2366
2367 /* transfer control flags from master to slave */
2368 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2369
2370 /* perform autopsy on the slave link */
2371 ata_eh_link_autopsy(ap->slave_link);
2372
2373 /* transfer actions from slave to master and clear slave */
2374 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2375 mehc->i.action |= sehc->i.action;
2376 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2377 mehc->i.flags |= sehc->i.flags;
2378 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2379 }
2380
2381 /* Autopsy of fanout ports can affect host link autopsy.
2382 * Perform host link autopsy last.
2383 */
2384 if (sata_pmp_attached(ap))
2385 ata_eh_link_autopsy(&ap->link);
2386 }
2387
2388 /**
2389 * ata_get_cmd_descript - get description for ATA command
2390 * @command: ATA command code to get description for
2391 *
2392 * Return a textual description of the given command, or NULL if the
2393 * command is not known.
2394 *
2395 * LOCKING:
2396 * None
2397 */
2398 const char *ata_get_cmd_descript(u8 command)
2399 {
2400 #ifdef CONFIG_ATA_VERBOSE_ERROR
2401 static const struct
2402 {
2403 u8 command;
2404 const char *text;
2405 } cmd_descr[] = {
2406 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2407 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2408 { ATA_CMD_STANDBY, "STANDBY" },
2409 { ATA_CMD_IDLE, "IDLE" },
2410 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2411 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2412 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2413 { ATA_CMD_NOP, "NOP" },
2414 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2415 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2416 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2417 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2418 { ATA_CMD_SERVICE, "SERVICE" },
2419 { ATA_CMD_READ, "READ DMA" },
2420 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2421 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2422 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2423 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2424 { ATA_CMD_WRITE, "WRITE DMA" },
2425 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2426 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2427 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2428 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2429 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2430 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2431 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2432 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2433 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2434 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2435 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2436 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2437 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2438 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2439 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2440 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2441 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2442 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2443 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2444 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2445 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2446 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2447 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2448 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2449 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2450 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2451 { ATA_CMD_SLEEP, "SLEEP" },
2452 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2453 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2454 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2455 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2456 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2457 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2458 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2459 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2460 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2461 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2462 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2463 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2464 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2465 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2466 { ATA_CMD_PMP_READ, "READ BUFFER" },
2467 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2468 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2469 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2470 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2471 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2472 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2473 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2474 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2475 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2476 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2477 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2478 { ATA_CMD_SMART, "SMART" },
2479 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2480 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2481 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2482 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2483 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2484 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2485 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2486 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2487 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2488 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2489 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2490 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2491 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2492 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2493 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2494 { ATA_CMD_RESTORE, "RECALIBRATE" },
2495 { 0, NULL } /* terminate list */
2496 };
2497
2498 unsigned int i;
2499 for (i = 0; cmd_descr[i].text; i++)
2500 if (cmd_descr[i].command == command)
2501 return cmd_descr[i].text;
2502 #endif
2503
2504 return NULL;
2505 }
2506 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2507
2508 /**
2509 * ata_eh_link_report - report error handling to user
2510 * @link: ATA link EH is going on
2511 *
2512 * Report EH to user.
2513 *
2514 * LOCKING:
2515 * None.
2516 */
2517 static void ata_eh_link_report(struct ata_link *link)
2518 {
2519 struct ata_port *ap = link->ap;
2520 struct ata_eh_context *ehc = &link->eh_context;
2521 const char *frozen, *desc;
2522 char tries_buf[6] = "";
2523 int tag, nr_failed = 0;
2524
2525 if (ehc->i.flags & ATA_EHI_QUIET)
2526 return;
2527
2528 desc = NULL;
2529 if (ehc->i.desc[0] != '\0')
2530 desc = ehc->i.desc;
2531
2532 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2533 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2534
2535 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2536 ata_dev_phys_link(qc->dev) != link ||
2537 ((qc->flags & ATA_QCFLAG_QUIET) &&
2538 qc->err_mask == AC_ERR_DEV))
2539 continue;
2540 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2541 continue;
2542
2543 nr_failed++;
2544 }
2545
2546 if (!nr_failed && !ehc->i.err_mask)
2547 return;
2548
2549 frozen = "";
2550 if (ap->pflags & ATA_PFLAG_FROZEN)
2551 frozen = " frozen";
2552
2553 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2554 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2555 ap->eh_tries);
2556
2557 if (ehc->i.dev) {
2558 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2559 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2560 ehc->i.err_mask, link->sactive, ehc->i.serror,
2561 ehc->i.action, frozen, tries_buf);
2562 if (desc)
2563 ata_dev_err(ehc->i.dev, "%s\n", desc);
2564 } else {
2565 ata_link_err(link, "exception Emask 0x%x "
2566 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2567 ehc->i.err_mask, link->sactive, ehc->i.serror,
2568 ehc->i.action, frozen, tries_buf);
2569 if (desc)
2570 ata_link_err(link, "%s\n", desc);
2571 }
2572
2573 #ifdef CONFIG_ATA_VERBOSE_ERROR
2574 if (ehc->i.serror)
2575 ata_link_err(link,
2576 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2577 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2578 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2579 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2580 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2581 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2582 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2583 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2584 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2585 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2586 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2587 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2588 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2589 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2590 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2591 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2592 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2593 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2594 #endif
2595
2596 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2597 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2598 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2599 char data_buf[20] = "";
2600 char cdb_buf[70] = "";
2601
2602 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2603 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2604 continue;
2605
2606 if (qc->dma_dir != DMA_NONE) {
2607 static const char *dma_str[] = {
2608 [DMA_BIDIRECTIONAL] = "bidi",
2609 [DMA_TO_DEVICE] = "out",
2610 [DMA_FROM_DEVICE] = "in",
2611 };
2612 static const char *prot_str[] = {
2613 [ATA_PROT_PIO] = "pio",
2614 [ATA_PROT_DMA] = "dma",
2615 [ATA_PROT_NCQ] = "ncq",
2616 [ATAPI_PROT_PIO] = "pio",
2617 [ATAPI_PROT_DMA] = "dma",
2618 };
2619
2620 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2621 prot_str[qc->tf.protocol], qc->nbytes,
2622 dma_str[qc->dma_dir]);
2623 }
2624
2625 if (ata_is_atapi(qc->tf.protocol)) {
2626 const u8 *cdb = qc->cdb;
2627 size_t cdb_len = qc->dev->cdb_len;
2628
2629 if (qc->scsicmd) {
2630 cdb = qc->scsicmd->cmnd;
2631 cdb_len = qc->scsicmd->cmd_len;
2632 }
2633 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2634 cdb, cdb_len);
2635 } else {
2636 const char *descr = ata_get_cmd_descript(cmd->command);
2637 if (descr)
2638 ata_dev_err(qc->dev, "failed command: %s\n",
2639 descr);
2640 }
2641
2642 ata_dev_err(qc->dev,
2643 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2644 "tag %d%s\n %s"
2645 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2646 "Emask 0x%x (%s)%s\n",
2647 cmd->command, cmd->feature, cmd->nsect,
2648 cmd->lbal, cmd->lbam, cmd->lbah,
2649 cmd->hob_feature, cmd->hob_nsect,
2650 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2651 cmd->device, qc->tag, data_buf, cdb_buf,
2652 res->command, res->feature, res->nsect,
2653 res->lbal, res->lbam, res->lbah,
2654 res->hob_feature, res->hob_nsect,
2655 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2656 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2657 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2658
2659 #ifdef CONFIG_ATA_VERBOSE_ERROR
2660 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2661 ATA_SENSE | ATA_ERR)) {
2662 if (res->command & ATA_BUSY)
2663 ata_dev_err(qc->dev, "status: { Busy }\n");
2664 else
2665 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2666 res->command & ATA_DRDY ? "DRDY " : "",
2667 res->command & ATA_DF ? "DF " : "",
2668 res->command & ATA_DRQ ? "DRQ " : "",
2669 res->command & ATA_SENSE ? "SENSE " : "",
2670 res->command & ATA_ERR ? "ERR " : "");
2671 }
2672
2673 if (cmd->command != ATA_CMD_PACKET &&
2674 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2675 ATA_IDNF | ATA_ABORTED)))
2676 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2677 res->feature & ATA_ICRC ? "ICRC " : "",
2678 res->feature & ATA_UNC ? "UNC " : "",
2679 res->feature & ATA_AMNF ? "AMNF " : "",
2680 res->feature & ATA_IDNF ? "IDNF " : "",
2681 res->feature & ATA_ABORTED ? "ABRT " : "");
2682 #endif
2683 }
2684 }
2685
2686 /**
2687 * ata_eh_report - report error handling to user
2688 * @ap: ATA port to report EH about
2689 *
2690 * Report EH to user.
2691 *
2692 * LOCKING:
2693 * None.
2694 */
2695 void ata_eh_report(struct ata_port *ap)
2696 {
2697 struct ata_link *link;
2698
2699 ata_for_each_link(link, ap, HOST_FIRST)
2700 ata_eh_link_report(link);
2701 }
2702
2703 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2704 unsigned int *classes, unsigned long deadline,
2705 bool clear_classes)
2706 {
2707 struct ata_device *dev;
2708
2709 if (clear_classes)
2710 ata_for_each_dev(dev, link, ALL)
2711 classes[dev->devno] = ATA_DEV_UNKNOWN;
2712
2713 return reset(link, classes, deadline);
2714 }
2715
2716 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2717 {
2718 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2719 return 0;
2720 if (rc == -EAGAIN)
2721 return 1;
2722 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2723 return 1;
2724 return 0;
2725 }
2726
2727 int ata_eh_reset(struct ata_link *link, int classify,
2728 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2729 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2730 {
2731 struct ata_port *ap = link->ap;
2732 struct ata_link *slave = ap->slave_link;
2733 struct ata_eh_context *ehc = &link->eh_context;
2734 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2735 unsigned int *classes = ehc->classes;
2736 unsigned int lflags = link->flags;
2737 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2738 int max_tries = 0, try = 0;
2739 struct ata_link *failed_link;
2740 struct ata_device *dev;
2741 unsigned long deadline, now;
2742 ata_reset_fn_t reset;
2743 unsigned long flags;
2744 u32 sstatus;
2745 int nr_unknown, rc;
2746
2747 /*
2748 * Prepare to reset
2749 */
2750 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2751 max_tries++;
2752 if (link->flags & ATA_LFLAG_RST_ONCE)
2753 max_tries = 1;
2754 if (link->flags & ATA_LFLAG_NO_HRST)
2755 hardreset = NULL;
2756 if (link->flags & ATA_LFLAG_NO_SRST)
2757 softreset = NULL;
2758
2759 /* make sure each reset attempt is at least COOL_DOWN apart */
2760 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2761 now = jiffies;
2762 WARN_ON(time_after(ehc->last_reset, now));
2763 deadline = ata_deadline(ehc->last_reset,
2764 ATA_EH_RESET_COOL_DOWN);
2765 if (time_before(now, deadline))
2766 schedule_timeout_uninterruptible(deadline - now);
2767 }
2768
2769 spin_lock_irqsave(ap->lock, flags);
2770 ap->pflags |= ATA_PFLAG_RESETTING;
2771 spin_unlock_irqrestore(ap->lock, flags);
2772
2773 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2774
2775 ata_for_each_dev(dev, link, ALL) {
2776 /* If we issue an SRST then an ATA drive (not ATAPI)
2777 * may change configuration and be in PIO0 timing. If
2778 * we do a hard reset (or are coming from power on)
2779 * this is true for ATA or ATAPI. Until we've set a
2780 * suitable controller mode we should not touch the
2781 * bus as we may be talking too fast.
2782 */
2783 dev->pio_mode = XFER_PIO_0;
2784 dev->dma_mode = 0xff;
2785
2786 /* If the controller has a pio mode setup function
2787 * then use it to set the chipset to rights. Don't
2788 * touch the DMA setup as that will be dealt with when
2789 * configuring devices.
2790 */
2791 if (ap->ops->set_piomode)
2792 ap->ops->set_piomode(ap, dev);
2793 }
2794
2795 /* prefer hardreset */
2796 reset = NULL;
2797 ehc->i.action &= ~ATA_EH_RESET;
2798 if (hardreset) {
2799 reset = hardreset;
2800 ehc->i.action |= ATA_EH_HARDRESET;
2801 } else if (softreset) {
2802 reset = softreset;
2803 ehc->i.action |= ATA_EH_SOFTRESET;
2804 }
2805
2806 if (prereset) {
2807 unsigned long deadline = ata_deadline(jiffies,
2808 ATA_EH_PRERESET_TIMEOUT);
2809
2810 if (slave) {
2811 sehc->i.action &= ~ATA_EH_RESET;
2812 sehc->i.action |= ehc->i.action;
2813 }
2814
2815 rc = prereset(link, deadline);
2816
2817 /* If present, do prereset on slave link too. Reset
2818 * is skipped iff both master and slave links report
2819 * -ENOENT or clear ATA_EH_RESET.
2820 */
2821 if (slave && (rc == 0 || rc == -ENOENT)) {
2822 int tmp;
2823
2824 tmp = prereset(slave, deadline);
2825 if (tmp != -ENOENT)
2826 rc = tmp;
2827
2828 ehc->i.action |= sehc->i.action;
2829 }
2830
2831 if (rc) {
2832 if (rc == -ENOENT) {
2833 ata_link_dbg(link, "port disabled--ignoring\n");
2834 ehc->i.action &= ~ATA_EH_RESET;
2835
2836 ata_for_each_dev(dev, link, ALL)
2837 classes[dev->devno] = ATA_DEV_NONE;
2838
2839 rc = 0;
2840 } else
2841 ata_link_err(link,
2842 "prereset failed (errno=%d)\n",
2843 rc);
2844 goto out;
2845 }
2846
2847 /* prereset() might have cleared ATA_EH_RESET. If so,
2848 * bang classes, thaw and return.
2849 */
2850 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2851 ata_for_each_dev(dev, link, ALL)
2852 classes[dev->devno] = ATA_DEV_NONE;
2853 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2854 ata_is_host_link(link))
2855 ata_eh_thaw_port(ap);
2856 rc = 0;
2857 goto out;
2858 }
2859 }
2860
2861 retry:
2862 /*
2863 * Perform reset
2864 */
2865 if (ata_is_host_link(link))
2866 ata_eh_freeze_port(ap);
2867
2868 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2869
2870 if (reset) {
2871 if (verbose)
2872 ata_link_info(link, "%s resetting link\n",
2873 reset == softreset ? "soft" : "hard");
2874
2875 /* mark that this EH session started with reset */
2876 ehc->last_reset = jiffies;
2877 if (reset == hardreset)
2878 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2879 else
2880 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2881
2882 rc = ata_do_reset(link, reset, classes, deadline, true);
2883 if (rc && rc != -EAGAIN) {
2884 failed_link = link;
2885 goto fail;
2886 }
2887
2888 /* hardreset slave link if existent */
2889 if (slave && reset == hardreset) {
2890 int tmp;
2891
2892 if (verbose)
2893 ata_link_info(slave, "hard resetting link\n");
2894
2895 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2896 tmp = ata_do_reset(slave, reset, classes, deadline,
2897 false);
2898 switch (tmp) {
2899 case -EAGAIN:
2900 rc = -EAGAIN;
2901 case 0:
2902 break;
2903 default:
2904 failed_link = slave;
2905 rc = tmp;
2906 goto fail;
2907 }
2908 }
2909
2910 /* perform follow-up SRST if necessary */
2911 if (reset == hardreset &&
2912 ata_eh_followup_srst_needed(link, rc)) {
2913 reset = softreset;
2914
2915 if (!reset) {
2916 ata_link_err(link,
2917 "follow-up softreset required but no softreset available\n");
2918 failed_link = link;
2919 rc = -EINVAL;
2920 goto fail;
2921 }
2922
2923 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2924 rc = ata_do_reset(link, reset, classes, deadline, true);
2925 if (rc) {
2926 failed_link = link;
2927 goto fail;
2928 }
2929 }
2930 } else {
2931 if (verbose)
2932 ata_link_info(link,
2933 "no reset method available, skipping reset\n");
2934 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2935 lflags |= ATA_LFLAG_ASSUME_ATA;
2936 }
2937
2938 /*
2939 * Post-reset processing
2940 */
2941 ata_for_each_dev(dev, link, ALL) {
2942 /* After the reset, the device state is PIO 0 and the
2943 * controller state is undefined. Reset also wakes up
2944 * drives from sleeping mode.
2945 */
2946 dev->pio_mode = XFER_PIO_0;
2947 dev->flags &= ~ATA_DFLAG_SLEEPING;
2948
2949 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2950 continue;
2951
2952 /* apply class override */
2953 if (lflags & ATA_LFLAG_ASSUME_ATA)
2954 classes[dev->devno] = ATA_DEV_ATA;
2955 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2956 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2957 }
2958
2959 /* record current link speed */
2960 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2961 link->sata_spd = (sstatus >> 4) & 0xf;
2962 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2963 slave->sata_spd = (sstatus >> 4) & 0xf;
2964
2965 /* thaw the port */
2966 if (ata_is_host_link(link))
2967 ata_eh_thaw_port(ap);
2968
2969 /* postreset() should clear hardware SError. Although SError
2970 * is cleared during link resume, clearing SError here is
2971 * necessary as some PHYs raise hotplug events after SRST.
2972 * This introduces race condition where hotplug occurs between
2973 * reset and here. This race is mediated by cross checking
2974 * link onlineness and classification result later.
2975 */
2976 if (postreset) {
2977 postreset(link, classes);
2978 if (slave)
2979 postreset(slave, classes);
2980 }
2981
2982 /*
2983 * Some controllers can't be frozen very well and may set spurious
2984 * error conditions during reset. Clear accumulated error
2985 * information and re-thaw the port if frozen. As reset is the
2986 * final recovery action and we cross check link onlineness against
2987 * device classification later, no hotplug event is lost by this.
2988 */
2989 spin_lock_irqsave(link->ap->lock, flags);
2990 memset(&link->eh_info, 0, sizeof(link->eh_info));
2991 if (slave)
2992 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2993 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2994 spin_unlock_irqrestore(link->ap->lock, flags);
2995
2996 if (ap->pflags & ATA_PFLAG_FROZEN)
2997 ata_eh_thaw_port(ap);
2998
2999 /*
3000 * Make sure onlineness and classification result correspond.
3001 * Hotplug could have happened during reset and some
3002 * controllers fail to wait while a drive is spinning up after
3003 * being hotplugged causing misdetection. By cross checking
3004 * link on/offlineness and classification result, those
3005 * conditions can be reliably detected and retried.
3006 */
3007 nr_unknown = 0;
3008 ata_for_each_dev(dev, link, ALL) {
3009 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
3010 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
3011 ata_dev_dbg(dev, "link online but device misclassified\n");
3012 classes[dev->devno] = ATA_DEV_NONE;
3013 nr_unknown++;
3014 }
3015 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3016 if (ata_class_enabled(classes[dev->devno]))
3017 ata_dev_dbg(dev,
3018 "link offline, clearing class %d to NONE\n",
3019 classes[dev->devno]);
3020 classes[dev->devno] = ATA_DEV_NONE;
3021 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
3022 ata_dev_dbg(dev,
3023 "link status unknown, clearing UNKNOWN to NONE\n");
3024 classes[dev->devno] = ATA_DEV_NONE;
3025 }
3026 }
3027
3028 if (classify && nr_unknown) {
3029 if (try < max_tries) {
3030 ata_link_warn(link,
3031 "link online but %d devices misclassified, retrying\n",
3032 nr_unknown);
3033 failed_link = link;
3034 rc = -EAGAIN;
3035 goto fail;
3036 }
3037 ata_link_warn(link,
3038 "link online but %d devices misclassified, "
3039 "device detection might fail\n", nr_unknown);
3040 }
3041
3042 /* reset successful, schedule revalidation */
3043 ata_eh_done(link, NULL, ATA_EH_RESET);
3044 if (slave)
3045 ata_eh_done(slave, NULL, ATA_EH_RESET);
3046 ehc->last_reset = jiffies; /* update to completion time */
3047 ehc->i.action |= ATA_EH_REVALIDATE;
3048 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
3049
3050 rc = 0;
3051 out:
3052 /* clear hotplug flag */
3053 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3054 if (slave)
3055 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3056
3057 spin_lock_irqsave(ap->lock, flags);
3058 ap->pflags &= ~ATA_PFLAG_RESETTING;
3059 spin_unlock_irqrestore(ap->lock, flags);
3060
3061 return rc;
3062
3063 fail:
3064 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
3065 if (!ata_is_host_link(link) &&
3066 sata_scr_read(link, SCR_STATUS, &sstatus))
3067 rc = -ERESTART;
3068
3069 if (try >= max_tries) {
3070 /*
3071 * Thaw host port even if reset failed, so that the port
3072 * can be retried on the next phy event. This risks
3073 * repeated EH runs but seems to be a better tradeoff than
3074 * shutting down a port after a botched hotplug attempt.
3075 */
3076 if (ata_is_host_link(link))
3077 ata_eh_thaw_port(ap);
3078 goto out;
3079 }
3080
3081 now = jiffies;
3082 if (time_before(now, deadline)) {
3083 unsigned long delta = deadline - now;
3084
3085 ata_link_warn(failed_link,
3086 "reset failed (errno=%d), retrying in %u secs\n",
3087 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3088
3089 ata_eh_release(ap);
3090 while (delta)
3091 delta = schedule_timeout_uninterruptible(delta);
3092 ata_eh_acquire(ap);
3093 }
3094
3095 /*
3096 * While disks spinup behind PMP, some controllers fail sending SRST.
3097 * They need to be reset - as well as the PMP - before retrying.
3098 */
3099 if (rc == -ERESTART) {
3100 if (ata_is_host_link(link))
3101 ata_eh_thaw_port(ap);
3102 goto out;
3103 }
3104
3105 if (try == max_tries - 1) {
3106 sata_down_spd_limit(link, 0);
3107 if (slave)
3108 sata_down_spd_limit(slave, 0);
3109 } else if (rc == -EPIPE)
3110 sata_down_spd_limit(failed_link, 0);
3111
3112 if (hardreset)
3113 reset = hardreset;
3114 goto retry;
3115 }
3116
3117 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3118 {
3119 struct ata_link *link;
3120 struct ata_device *dev;
3121 unsigned long flags;
3122
3123 /*
3124 * This function can be thought of as an extended version of
3125 * ata_eh_about_to_do() specially crafted to accommodate the
3126 * requirements of ATA_EH_PARK handling. Since the EH thread
3127 * does not leave the do {} while () loop in ata_eh_recover as
3128 * long as the timeout for a park request to *one* device on
3129 * the port has not expired, and since we still want to pick
3130 * up park requests to other devices on the same port or
3131 * timeout updates for the same device, we have to pull
3132 * ATA_EH_PARK actions from eh_info into eh_context.i
3133 * ourselves at the beginning of each pass over the loop.
3134 *
3135 * Additionally, all write accesses to &ap->park_req_pending
3136 * through reinit_completion() (see below) or complete_all()
3137 * (see ata_scsi_park_store()) are protected by the host lock.
3138 * As a result we have that park_req_pending.done is zero on
3139 * exit from this function, i.e. when ATA_EH_PARK actions for
3140 * *all* devices on port ap have been pulled into the
3141 * respective eh_context structs. If, and only if,
3142 * park_req_pending.done is non-zero by the time we reach
3143 * wait_for_completion_timeout(), another ATA_EH_PARK action
3144 * has been scheduled for at least one of the devices on port
3145 * ap and we have to cycle over the do {} while () loop in
3146 * ata_eh_recover() again.
3147 */
3148
3149 spin_lock_irqsave(ap->lock, flags);
3150 reinit_completion(&ap->park_req_pending);
3151 ata_for_each_link(link, ap, EDGE) {
3152 ata_for_each_dev(dev, link, ALL) {
3153 struct ata_eh_info *ehi = &link->eh_info;
3154
3155 link->eh_context.i.dev_action[dev->devno] |=
3156 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3157 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3158 }
3159 }
3160 spin_unlock_irqrestore(ap->lock, flags);
3161 }
3162
3163 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3164 {
3165 struct ata_eh_context *ehc = &dev->link->eh_context;
3166 struct ata_taskfile tf;
3167 unsigned int err_mask;
3168
3169 ata_tf_init(dev, &tf);
3170 if (park) {
3171 ehc->unloaded_mask |= 1 << dev->devno;
3172 tf.command = ATA_CMD_IDLEIMMEDIATE;
3173 tf.feature = 0x44;
3174 tf.lbal = 0x4c;
3175 tf.lbam = 0x4e;
3176 tf.lbah = 0x55;
3177 } else {
3178 ehc->unloaded_mask &= ~(1 << dev->devno);
3179 tf.command = ATA_CMD_CHK_POWER;
3180 }
3181
3182 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3183 tf.protocol |= ATA_PROT_NODATA;
3184 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3185 if (park && (err_mask || tf.lbal != 0xc4)) {
3186 ata_dev_err(dev, "head unload failed!\n");
3187 ehc->unloaded_mask &= ~(1 << dev->devno);
3188 }
3189 }
3190
3191 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3192 struct ata_device **r_failed_dev)
3193 {
3194 struct ata_port *ap = link->ap;
3195 struct ata_eh_context *ehc = &link->eh_context;
3196 struct ata_device *dev;
3197 unsigned int new_mask = 0;
3198 unsigned long flags;
3199 int rc = 0;
3200
3201 DPRINTK("ENTER\n");
3202
3203 /* For PATA drive side cable detection to work, IDENTIFY must
3204 * be done backwards such that PDIAG- is released by the slave
3205 * device before the master device is identified.
3206 */
3207 ata_for_each_dev(dev, link, ALL_REVERSE) {
3208 unsigned int action = ata_eh_dev_action(dev);
3209 unsigned int readid_flags = 0;
3210
3211 if (ehc->i.flags & ATA_EHI_DID_RESET)
3212 readid_flags |= ATA_READID_POSTRESET;
3213
3214 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3215 WARN_ON(dev->class == ATA_DEV_PMP);
3216
3217 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3218 rc = -EIO;
3219 goto err;
3220 }
3221
3222 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3223 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3224 readid_flags);
3225 if (rc)
3226 goto err;
3227
3228 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3229
3230 /* Configuration may have changed, reconfigure
3231 * transfer mode.
3232 */
3233 ehc->i.flags |= ATA_EHI_SETMODE;
3234
3235 /* schedule the scsi_rescan_device() here */
3236 schedule_work(&(ap->scsi_rescan_task));
3237 } else if (dev->class == ATA_DEV_UNKNOWN &&
3238 ehc->tries[dev->devno] &&
3239 ata_class_enabled(ehc->classes[dev->devno])) {
3240 /* Temporarily set dev->class, it will be
3241 * permanently set once all configurations are
3242 * complete. This is necessary because new
3243 * device configuration is done in two
3244 * separate loops.
3245 */
3246 dev->class = ehc->classes[dev->devno];
3247
3248 if (dev->class == ATA_DEV_PMP)
3249 rc = sata_pmp_attach(dev);
3250 else
3251 rc = ata_dev_read_id(dev, &dev->class,
3252 readid_flags, dev->id);
3253
3254 /* read_id might have changed class, store and reset */
3255 ehc->classes[dev->devno] = dev->class;
3256 dev->class = ATA_DEV_UNKNOWN;
3257
3258 switch (rc) {
3259 case 0:
3260 /* clear error info accumulated during probe */
3261 ata_ering_clear(&dev->ering);
3262 new_mask |= 1 << dev->devno;
3263 break;
3264 case -ENOENT:
3265 /* IDENTIFY was issued to non-existent
3266 * device. No need to reset. Just
3267 * thaw and ignore the device.
3268 */
3269 ata_eh_thaw_port(ap);
3270 break;
3271 default:
3272 goto err;
3273 }
3274 }
3275 }
3276
3277 /* PDIAG- should have been released, ask cable type if post-reset */
3278 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3279 if (ap->ops->cable_detect)
3280 ap->cbl = ap->ops->cable_detect(ap);
3281 ata_force_cbl(ap);
3282 }
3283
3284 /* Configure new devices forward such that user doesn't see
3285 * device detection messages backwards.
3286 */
3287 ata_for_each_dev(dev, link, ALL) {
3288 if (!(new_mask & (1 << dev->devno)))
3289 continue;
3290
3291 dev->class = ehc->classes[dev->devno];
3292
3293 if (dev->class == ATA_DEV_PMP)
3294 continue;
3295
3296 ehc->i.flags |= ATA_EHI_PRINTINFO;
3297 rc = ata_dev_configure(dev);
3298 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3299 if (rc) {
3300 dev->class = ATA_DEV_UNKNOWN;
3301 goto err;
3302 }
3303
3304 spin_lock_irqsave(ap->lock, flags);
3305 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3306 spin_unlock_irqrestore(ap->lock, flags);
3307
3308 /* new device discovered, configure xfermode */
3309 ehc->i.flags |= ATA_EHI_SETMODE;
3310 }
3311
3312 return 0;
3313
3314 err:
3315 *r_failed_dev = dev;
3316 DPRINTK("EXIT rc=%d\n", rc);
3317 return rc;
3318 }
3319
3320 /**
3321 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3322 * @link: link on which timings will be programmed
3323 * @r_failed_dev: out parameter for failed device
3324 *
3325 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3326 * ata_set_mode() fails, pointer to the failing device is
3327 * returned in @r_failed_dev.
3328 *
3329 * LOCKING:
3330 * PCI/etc. bus probe sem.
3331 *
3332 * RETURNS:
3333 * 0 on success, negative errno otherwise
3334 */
3335 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3336 {
3337 struct ata_port *ap = link->ap;
3338 struct ata_device *dev;
3339 int rc;
3340
3341 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3342 ata_for_each_dev(dev, link, ENABLED) {
3343 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3344 struct ata_ering_entry *ent;
3345
3346 ent = ata_ering_top(&dev->ering);
3347 if (ent)
3348 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3349 }
3350 }
3351
3352 /* has private set_mode? */
3353 if (ap->ops->set_mode)
3354 rc = ap->ops->set_mode(link, r_failed_dev);
3355 else
3356 rc = ata_do_set_mode(link, r_failed_dev);
3357
3358 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3359 ata_for_each_dev(dev, link, ENABLED) {
3360 struct ata_eh_context *ehc = &link->eh_context;
3361 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3362 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3363
3364 if (dev->xfer_mode != saved_xfer_mode ||
3365 ata_ncq_enabled(dev) != saved_ncq)
3366 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3367 }
3368
3369 return rc;
3370 }
3371
3372 /**
3373 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3374 * @dev: ATAPI device to clear UA for
3375 *
3376 * Resets and other operations can make an ATAPI device raise
3377 * UNIT ATTENTION which causes the next operation to fail. This
3378 * function clears UA.
3379 *
3380 * LOCKING:
3381 * EH context (may sleep).
3382 *
3383 * RETURNS:
3384 * 0 on success, -errno on failure.
3385 */
3386 static int atapi_eh_clear_ua(struct ata_device *dev)
3387 {
3388 int i;
3389
3390 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3391 u8 *sense_buffer = dev->link->ap->sector_buf;
3392 u8 sense_key = 0;
3393 unsigned int err_mask;
3394
3395 err_mask = atapi_eh_tur(dev, &sense_key);
3396 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3397 ata_dev_warn(dev,
3398 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3399 err_mask);
3400 return -EIO;
3401 }
3402
3403 if (!err_mask || sense_key != UNIT_ATTENTION)
3404 return 0;
3405
3406 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3407 if (err_mask) {
3408 ata_dev_warn(dev, "failed to clear "
3409 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3410 return -EIO;
3411 }
3412 }
3413
3414 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3415 ATA_EH_UA_TRIES);
3416
3417 return 0;
3418 }
3419
3420 /**
3421 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3422 * @dev: ATA device which may need FLUSH retry
3423 *
3424 * If @dev failed FLUSH, it needs to be reported upper layer
3425 * immediately as it means that @dev failed to remap and already
3426 * lost at least a sector and further FLUSH retrials won't make
3427 * any difference to the lost sector. However, if FLUSH failed
3428 * for other reasons, for example transmission error, FLUSH needs
3429 * to be retried.
3430 *
3431 * This function determines whether FLUSH failure retry is
3432 * necessary and performs it if so.
3433 *
3434 * RETURNS:
3435 * 0 if EH can continue, -errno if EH needs to be repeated.
3436 */
3437 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3438 {
3439 struct ata_link *link = dev->link;
3440 struct ata_port *ap = link->ap;
3441 struct ata_queued_cmd *qc;
3442 struct ata_taskfile tf;
3443 unsigned int err_mask;
3444 int rc = 0;
3445
3446 /* did flush fail for this device? */
3447 if (!ata_tag_valid(link->active_tag))
3448 return 0;
3449
3450 qc = __ata_qc_from_tag(ap, link->active_tag);
3451 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3452 qc->tf.command != ATA_CMD_FLUSH))
3453 return 0;
3454
3455 /* if the device failed it, it should be reported to upper layers */
3456 if (qc->err_mask & AC_ERR_DEV)
3457 return 0;
3458
3459 /* flush failed for some other reason, give it another shot */
3460 ata_tf_init(dev, &tf);
3461
3462 tf.command = qc->tf.command;
3463 tf.flags |= ATA_TFLAG_DEVICE;
3464 tf.protocol = ATA_PROT_NODATA;
3465
3466 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3467 tf.command, qc->err_mask);
3468
3469 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3470 if (!err_mask) {
3471 /*
3472 * FLUSH is complete but there's no way to
3473 * successfully complete a failed command from EH.
3474 * Making sure retry is allowed at least once and
3475 * retrying it should do the trick - whatever was in
3476 * the cache is already on the platter and this won't
3477 * cause infinite loop.
3478 */
3479 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3480 } else {
3481 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3482 err_mask);
3483 rc = -EIO;
3484
3485 /* if device failed it, report it to upper layers */
3486 if (err_mask & AC_ERR_DEV) {
3487 qc->err_mask |= AC_ERR_DEV;
3488 qc->result_tf = tf;
3489 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3490 rc = 0;
3491 }
3492 }
3493 return rc;
3494 }
3495
3496 /**
3497 * ata_eh_set_lpm - configure SATA interface power management
3498 * @link: link to configure power management
3499 * @policy: the link power management policy
3500 * @r_failed_dev: out parameter for failed device
3501 *
3502 * Enable SATA Interface power management. This will enable
3503 * Device Interface Power Management (DIPM) for min_power
3504 * policy, and then call driver specific callbacks for
3505 * enabling Host Initiated Power management.
3506 *
3507 * LOCKING:
3508 * EH context.
3509 *
3510 * RETURNS:
3511 * 0 on success, -errno on failure.
3512 */
3513 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3514 struct ata_device **r_failed_dev)
3515 {
3516 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3517 struct ata_eh_context *ehc = &link->eh_context;
3518 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3519 enum ata_lpm_policy old_policy = link->lpm_policy;
3520 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3521 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3522 unsigned int err_mask;
3523 int rc;
3524
3525 /* if the link or host doesn't do LPM, noop */
3526 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3527 return 0;
3528
3529 /*
3530 * DIPM is enabled only for MIN_POWER as some devices
3531 * misbehave when the host NACKs transition to SLUMBER. Order
3532 * device and link configurations such that the host always
3533 * allows DIPM requests.
3534 */
3535 ata_for_each_dev(dev, link, ENABLED) {
3536 bool hipm = ata_id_has_hipm(dev->id);
3537 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3538
3539 /* find the first enabled and LPM enabled devices */
3540 if (!link_dev)
3541 link_dev = dev;
3542
3543 if (!lpm_dev && (hipm || dipm))
3544 lpm_dev = dev;
3545
3546 hints &= ~ATA_LPM_EMPTY;
3547 if (!hipm)
3548 hints &= ~ATA_LPM_HIPM;
3549
3550 /* disable DIPM before changing link config */
3551 if (policy != ATA_LPM_MIN_POWER && dipm) {
3552 err_mask = ata_dev_set_feature(dev,
3553 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3554 if (err_mask && err_mask != AC_ERR_DEV) {
3555 ata_dev_warn(dev,
3556 "failed to disable DIPM, Emask 0x%x\n",
3557 err_mask);
3558 rc = -EIO;
3559 goto fail;
3560 }
3561 }
3562 }
3563
3564 if (ap) {
3565 rc = ap->ops->set_lpm(link, policy, hints);
3566 if (!rc && ap->slave_link)
3567 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3568 } else
3569 rc = sata_pmp_set_lpm(link, policy, hints);
3570
3571 /*
3572 * Attribute link config failure to the first (LPM) enabled
3573 * device on the link.
3574 */
3575 if (rc) {
3576 if (rc == -EOPNOTSUPP) {
3577 link->flags |= ATA_LFLAG_NO_LPM;
3578 return 0;
3579 }
3580 dev = lpm_dev ? lpm_dev : link_dev;
3581 goto fail;
3582 }
3583
3584 /*
3585 * Low level driver acked the transition. Issue DIPM command
3586 * with the new policy set.
3587 */
3588 link->lpm_policy = policy;
3589 if (ap && ap->slave_link)
3590 ap->slave_link->lpm_policy = policy;
3591
3592 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3593 ata_for_each_dev(dev, link, ENABLED) {
3594 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3595 ata_id_has_dipm(dev->id)) {
3596 err_mask = ata_dev_set_feature(dev,
3597 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3598 if (err_mask && err_mask != AC_ERR_DEV) {
3599 ata_dev_warn(dev,
3600 "failed to enable DIPM, Emask 0x%x\n",
3601 err_mask);
3602 rc = -EIO;
3603 goto fail;
3604 }
3605 }
3606 }
3607
3608 link->last_lpm_change = jiffies;
3609 link->flags |= ATA_LFLAG_CHANGED;
3610
3611 return 0;
3612
3613 fail:
3614 /* restore the old policy */
3615 link->lpm_policy = old_policy;
3616 if (ap && ap->slave_link)
3617 ap->slave_link->lpm_policy = old_policy;
3618
3619 /* if no device or only one more chance is left, disable LPM */
3620 if (!dev || ehc->tries[dev->devno] <= 2) {
3621 ata_link_warn(link, "disabling LPM on the link\n");
3622 link->flags |= ATA_LFLAG_NO_LPM;
3623 }
3624 if (r_failed_dev)
3625 *r_failed_dev = dev;
3626 return rc;
3627 }
3628
3629 int ata_link_nr_enabled(struct ata_link *link)
3630 {
3631 struct ata_device *dev;
3632 int cnt = 0;
3633
3634 ata_for_each_dev(dev, link, ENABLED)
3635 cnt++;
3636 return cnt;
3637 }
3638
3639 static int ata_link_nr_vacant(struct ata_link *link)
3640 {
3641 struct ata_device *dev;
3642 int cnt = 0;
3643
3644 ata_for_each_dev(dev, link, ALL)
3645 if (dev->class == ATA_DEV_UNKNOWN)
3646 cnt++;
3647 return cnt;
3648 }
3649
3650 static int ata_eh_skip_recovery(struct ata_link *link)
3651 {
3652 struct ata_port *ap = link->ap;
3653 struct ata_eh_context *ehc = &link->eh_context;
3654 struct ata_device *dev;
3655
3656 /* skip disabled links */
3657 if (link->flags & ATA_LFLAG_DISABLED)
3658 return 1;
3659
3660 /* skip if explicitly requested */
3661 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3662 return 1;
3663
3664 /* thaw frozen port and recover failed devices */
3665 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3666 return 0;
3667
3668 /* reset at least once if reset is requested */
3669 if ((ehc->i.action & ATA_EH_RESET) &&
3670 !(ehc->i.flags & ATA_EHI_DID_RESET))
3671 return 0;
3672
3673 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3674 ata_for_each_dev(dev, link, ALL) {
3675 if (dev->class == ATA_DEV_UNKNOWN &&
3676 ehc->classes[dev->devno] != ATA_DEV_NONE)
3677 return 0;
3678 }
3679
3680 return 1;
3681 }
3682
3683 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3684 {
3685 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3686 u64 now = get_jiffies_64();
3687 int *trials = void_arg;
3688
3689 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3690 (ent->timestamp < now - min(now, interval)))
3691 return -1;
3692
3693 (*trials)++;
3694 return 0;
3695 }
3696
3697 static int ata_eh_schedule_probe(struct ata_device *dev)
3698 {
3699 struct ata_eh_context *ehc = &dev->link->eh_context;
3700 struct ata_link *link = ata_dev_phys_link(dev);
3701 int trials = 0;
3702
3703 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3704 (ehc->did_probe_mask & (1 << dev->devno)))
3705 return 0;
3706
3707 ata_eh_detach_dev(dev);
3708 ata_dev_init(dev);
3709 ehc->did_probe_mask |= (1 << dev->devno);
3710 ehc->i.action |= ATA_EH_RESET;
3711 ehc->saved_xfer_mode[dev->devno] = 0;
3712 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3713
3714 /* the link maybe in a deep sleep, wake it up */
3715 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3716 if (ata_is_host_link(link))
3717 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3718 ATA_LPM_EMPTY);
3719 else
3720 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3721 ATA_LPM_EMPTY);
3722 }
3723
3724 /* Record and count probe trials on the ering. The specific
3725 * error mask used is irrelevant. Because a successful device
3726 * detection clears the ering, this count accumulates only if
3727 * there are consecutive failed probes.
3728 *
3729 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3730 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3731 * forced to 1.5Gbps.
3732 *
3733 * This is to work around cases where failed link speed
3734 * negotiation results in device misdetection leading to
3735 * infinite DEVXCHG or PHRDY CHG events.
3736 */
3737 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3738 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3739
3740 if (trials > ATA_EH_PROBE_TRIALS)
3741 sata_down_spd_limit(link, 1);
3742
3743 return 1;
3744 }
3745
3746 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3747 {
3748 struct ata_eh_context *ehc = &dev->link->eh_context;
3749
3750 /* -EAGAIN from EH routine indicates retry without prejudice.
3751 * The requester is responsible for ensuring forward progress.
3752 */
3753 if (err != -EAGAIN)
3754 ehc->tries[dev->devno]--;
3755
3756 switch (err) {
3757 case -ENODEV:
3758 /* device missing or wrong IDENTIFY data, schedule probing */
3759 ehc->i.probe_mask |= (1 << dev->devno);
3760 case -EINVAL:
3761 /* give it just one more chance */
3762 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3763 case -EIO:
3764 if (ehc->tries[dev->devno] == 1) {
3765 /* This is the last chance, better to slow
3766 * down than lose it.
3767 */
3768 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3769 if (dev->pio_mode > XFER_PIO_0)
3770 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3771 }
3772 }
3773
3774 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3775 /* disable device if it has used up all its chances */
3776 ata_dev_disable(dev);
3777
3778 /* detach if offline */
3779 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3780 ata_eh_detach_dev(dev);
3781
3782 /* schedule probe if necessary */
3783 if (ata_eh_schedule_probe(dev)) {
3784 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3785 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3786 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3787 }
3788
3789 return 1;
3790 } else {
3791 ehc->i.action |= ATA_EH_RESET;
3792 return 0;
3793 }
3794 }
3795
3796 /**
3797 * ata_eh_recover - recover host port after error
3798 * @ap: host port to recover
3799 * @prereset: prereset method (can be NULL)
3800 * @softreset: softreset method (can be NULL)
3801 * @hardreset: hardreset method (can be NULL)
3802 * @postreset: postreset method (can be NULL)
3803 * @r_failed_link: out parameter for failed link
3804 *
3805 * This is the alpha and omega, eum and yang, heart and soul of
3806 * libata exception handling. On entry, actions required to
3807 * recover each link and hotplug requests are recorded in the
3808 * link's eh_context. This function executes all the operations
3809 * with appropriate retrials and fallbacks to resurrect failed
3810 * devices, detach goners and greet newcomers.
3811 *
3812 * LOCKING:
3813 * Kernel thread context (may sleep).
3814 *
3815 * RETURNS:
3816 * 0 on success, -errno on failure.
3817 */
3818 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3819 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3820 ata_postreset_fn_t postreset,
3821 struct ata_link **r_failed_link)
3822 {
3823 struct ata_link *link;
3824 struct ata_device *dev;
3825 int rc, nr_fails;
3826 unsigned long flags, deadline;
3827
3828 DPRINTK("ENTER\n");
3829
3830 /* prep for recovery */
3831 ata_for_each_link(link, ap, EDGE) {
3832 struct ata_eh_context *ehc = &link->eh_context;
3833
3834 /* re-enable link? */
3835 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3836 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3837 spin_lock_irqsave(ap->lock, flags);
3838 link->flags &= ~ATA_LFLAG_DISABLED;
3839 spin_unlock_irqrestore(ap->lock, flags);
3840 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3841 }
3842
3843 ata_for_each_dev(dev, link, ALL) {
3844 if (link->flags & ATA_LFLAG_NO_RETRY)
3845 ehc->tries[dev->devno] = 1;
3846 else
3847 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3848
3849 /* collect port action mask recorded in dev actions */
3850 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3851 ~ATA_EH_PERDEV_MASK;
3852 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3853
3854 /* process hotplug request */
3855 if (dev->flags & ATA_DFLAG_DETACH)
3856 ata_eh_detach_dev(dev);
3857
3858 /* schedule probe if necessary */
3859 if (!ata_dev_enabled(dev))
3860 ata_eh_schedule_probe(dev);
3861 }
3862 }
3863
3864 retry:
3865 rc = 0;
3866
3867 /* if UNLOADING, finish immediately */
3868 if (ap->pflags & ATA_PFLAG_UNLOADING)
3869 goto out;
3870
3871 /* prep for EH */
3872 ata_for_each_link(link, ap, EDGE) {
3873 struct ata_eh_context *ehc = &link->eh_context;
3874
3875 /* skip EH if possible. */
3876 if (ata_eh_skip_recovery(link))
3877 ehc->i.action = 0;
3878
3879 ata_for_each_dev(dev, link, ALL)
3880 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3881 }
3882
3883 /* reset */
3884 ata_for_each_link(link, ap, EDGE) {
3885 struct ata_eh_context *ehc = &link->eh_context;
3886
3887 if (!(ehc->i.action & ATA_EH_RESET))
3888 continue;
3889
3890 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3891 prereset, softreset, hardreset, postreset);
3892 if (rc) {
3893 ata_link_err(link, "reset failed, giving up\n");
3894 goto out;
3895 }
3896 }
3897
3898 do {
3899 unsigned long now;
3900
3901 /*
3902 * clears ATA_EH_PARK in eh_info and resets
3903 * ap->park_req_pending
3904 */
3905 ata_eh_pull_park_action(ap);
3906
3907 deadline = jiffies;
3908 ata_for_each_link(link, ap, EDGE) {
3909 ata_for_each_dev(dev, link, ALL) {
3910 struct ata_eh_context *ehc = &link->eh_context;
3911 unsigned long tmp;
3912
3913 if (dev->class != ATA_DEV_ATA &&
3914 dev->class != ATA_DEV_ZAC)
3915 continue;
3916 if (!(ehc->i.dev_action[dev->devno] &
3917 ATA_EH_PARK))
3918 continue;
3919 tmp = dev->unpark_deadline;
3920 if (time_before(deadline, tmp))
3921 deadline = tmp;
3922 else if (time_before_eq(tmp, jiffies))
3923 continue;
3924 if (ehc->unloaded_mask & (1 << dev->devno))
3925 continue;
3926
3927 ata_eh_park_issue_cmd(dev, 1);
3928 }
3929 }
3930
3931 now = jiffies;
3932 if (time_before_eq(deadline, now))
3933 break;
3934
3935 ata_eh_release(ap);
3936 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3937 deadline - now);
3938 ata_eh_acquire(ap);
3939 } while (deadline);
3940 ata_for_each_link(link, ap, EDGE) {
3941 ata_for_each_dev(dev, link, ALL) {
3942 if (!(link->eh_context.unloaded_mask &
3943 (1 << dev->devno)))
3944 continue;
3945
3946 ata_eh_park_issue_cmd(dev, 0);
3947 ata_eh_done(link, dev, ATA_EH_PARK);
3948 }
3949 }
3950
3951 /* the rest */
3952 nr_fails = 0;
3953 ata_for_each_link(link, ap, PMP_FIRST) {
3954 struct ata_eh_context *ehc = &link->eh_context;
3955
3956 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3957 goto config_lpm;
3958
3959 /* revalidate existing devices and attach new ones */
3960 rc = ata_eh_revalidate_and_attach(link, &dev);
3961 if (rc)
3962 goto rest_fail;
3963
3964 /* if PMP got attached, return, pmp EH will take care of it */
3965 if (link->device->class == ATA_DEV_PMP) {
3966 ehc->i.action = 0;
3967 return 0;
3968 }
3969
3970 /* configure transfer mode if necessary */
3971 if (ehc->i.flags & ATA_EHI_SETMODE) {
3972 rc = ata_set_mode(link, &dev);
3973 if (rc)
3974 goto rest_fail;
3975 ehc->i.flags &= ~ATA_EHI_SETMODE;
3976 }
3977
3978 /* If reset has been issued, clear UA to avoid
3979 * disrupting the current users of the device.
3980 */
3981 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3982 ata_for_each_dev(dev, link, ALL) {
3983 if (dev->class != ATA_DEV_ATAPI)
3984 continue;
3985 rc = atapi_eh_clear_ua(dev);
3986 if (rc)
3987 goto rest_fail;
3988 if (zpodd_dev_enabled(dev))
3989 zpodd_post_poweron(dev);
3990 }
3991 }
3992
3993 /* retry flush if necessary */
3994 ata_for_each_dev(dev, link, ALL) {
3995 if (dev->class != ATA_DEV_ATA &&
3996 dev->class != ATA_DEV_ZAC)
3997 continue;
3998 rc = ata_eh_maybe_retry_flush(dev);
3999 if (rc)
4000 goto rest_fail;
4001 }
4002
4003 config_lpm:
4004 /* configure link power saving */
4005 if (link->lpm_policy != ap->target_lpm_policy) {
4006 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
4007 if (rc)
4008 goto rest_fail;
4009 }
4010
4011 /* this link is okay now */
4012 ehc->i.flags = 0;
4013 continue;
4014
4015 rest_fail:
4016 nr_fails++;
4017 if (dev)
4018 ata_eh_handle_dev_fail(dev, rc);
4019
4020 if (ap->pflags & ATA_PFLAG_FROZEN) {
4021 /* PMP reset requires working host port.
4022 * Can't retry if it's frozen.
4023 */
4024 if (sata_pmp_attached(ap))
4025 goto out;
4026 break;
4027 }
4028 }
4029
4030 if (nr_fails)
4031 goto retry;
4032
4033 out:
4034 if (rc && r_failed_link)
4035 *r_failed_link = link;
4036
4037 DPRINTK("EXIT, rc=%d\n", rc);
4038 return rc;
4039 }
4040
4041 /**
4042 * ata_eh_finish - finish up EH
4043 * @ap: host port to finish EH for
4044 *
4045 * Recovery is complete. Clean up EH states and retry or finish
4046 * failed qcs.
4047 *
4048 * LOCKING:
4049 * None.
4050 */
4051 void ata_eh_finish(struct ata_port *ap)
4052 {
4053 int tag;
4054
4055 /* retry or finish qcs */
4056 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
4057 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
4058
4059 if (!(qc->flags & ATA_QCFLAG_FAILED))
4060 continue;
4061
4062 if (qc->err_mask) {
4063 /* FIXME: Once EH migration is complete,
4064 * generate sense data in this function,
4065 * considering both err_mask and tf.
4066 */
4067 if (qc->flags & ATA_QCFLAG_RETRY)
4068 ata_eh_qc_retry(qc);
4069 else
4070 ata_eh_qc_complete(qc);
4071 } else {
4072 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
4073 ata_eh_qc_complete(qc);
4074 } else {
4075 /* feed zero TF to sense generation */
4076 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
4077 ata_eh_qc_retry(qc);
4078 }
4079 }
4080 }
4081
4082 /* make sure nr_active_links is zero after EH */
4083 WARN_ON(ap->nr_active_links);
4084 ap->nr_active_links = 0;
4085 }
4086
4087 /**
4088 * ata_do_eh - do standard error handling
4089 * @ap: host port to handle error for
4090 *
4091 * @prereset: prereset method (can be NULL)
4092 * @softreset: softreset method (can be NULL)
4093 * @hardreset: hardreset method (can be NULL)
4094 * @postreset: postreset method (can be NULL)
4095 *
4096 * Perform standard error handling sequence.
4097 *
4098 * LOCKING:
4099 * Kernel thread context (may sleep).
4100 */
4101 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4102 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4103 ata_postreset_fn_t postreset)
4104 {
4105 struct ata_device *dev;
4106 int rc;
4107
4108 ata_eh_autopsy(ap);
4109 ata_eh_report(ap);
4110
4111 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4112 NULL);
4113 if (rc) {
4114 ata_for_each_dev(dev, &ap->link, ALL)
4115 ata_dev_disable(dev);
4116 }
4117
4118 ata_eh_finish(ap);
4119 }
4120
4121 /**
4122 * ata_std_error_handler - standard error handler
4123 * @ap: host port to handle error for
4124 *
4125 * Standard error handler
4126 *
4127 * LOCKING:
4128 * Kernel thread context (may sleep).
4129 */
4130 void ata_std_error_handler(struct ata_port *ap)
4131 {
4132 struct ata_port_operations *ops = ap->ops;
4133 ata_reset_fn_t hardreset = ops->hardreset;
4134
4135 /* ignore built-in hardreset if SCR access is not available */
4136 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4137 hardreset = NULL;
4138
4139 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4140 }
4141
4142 #ifdef CONFIG_PM
4143 /**
4144 * ata_eh_handle_port_suspend - perform port suspend operation
4145 * @ap: port to suspend
4146 *
4147 * Suspend @ap.
4148 *
4149 * LOCKING:
4150 * Kernel thread context (may sleep).
4151 */
4152 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4153 {
4154 unsigned long flags;
4155 int rc = 0;
4156 struct ata_device *dev;
4157
4158 /* are we suspending? */
4159 spin_lock_irqsave(ap->lock, flags);
4160 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4161 ap->pm_mesg.event & PM_EVENT_RESUME) {
4162 spin_unlock_irqrestore(ap->lock, flags);
4163 return;
4164 }
4165 spin_unlock_irqrestore(ap->lock, flags);
4166
4167 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4168
4169 /*
4170 * If we have a ZPODD attached, check its zero
4171 * power ready status before the port is frozen.
4172 * Only needed for runtime suspend.
4173 */
4174 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4175 ata_for_each_dev(dev, &ap->link, ENABLED) {
4176 if (zpodd_dev_enabled(dev))
4177 zpodd_on_suspend(dev);
4178 }
4179 }
4180
4181 /* tell ACPI we're suspending */
4182 rc = ata_acpi_on_suspend(ap);
4183 if (rc)
4184 goto out;
4185
4186 /* suspend */
4187 ata_eh_freeze_port(ap);
4188
4189 if (ap->ops->port_suspend)
4190 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4191
4192 ata_acpi_set_state(ap, ap->pm_mesg);
4193 out:
4194 /* update the flags */
4195 spin_lock_irqsave(ap->lock, flags);
4196
4197 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4198 if (rc == 0)
4199 ap->pflags |= ATA_PFLAG_SUSPENDED;
4200 else if (ap->pflags & ATA_PFLAG_FROZEN)
4201 ata_port_schedule_eh(ap);
4202
4203 spin_unlock_irqrestore(ap->lock, flags);
4204
4205 return;
4206 }
4207
4208 /**
4209 * ata_eh_handle_port_resume - perform port resume operation
4210 * @ap: port to resume
4211 *
4212 * Resume @ap.
4213 *
4214 * LOCKING:
4215 * Kernel thread context (may sleep).
4216 */
4217 static void ata_eh_handle_port_resume(struct ata_port *ap)
4218 {
4219 struct ata_link *link;
4220 struct ata_device *dev;
4221 unsigned long flags;
4222 int rc = 0;
4223
4224 /* are we resuming? */
4225 spin_lock_irqsave(ap->lock, flags);
4226 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4227 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4228 spin_unlock_irqrestore(ap->lock, flags);
4229 return;
4230 }
4231 spin_unlock_irqrestore(ap->lock, flags);
4232
4233 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4234
4235 /*
4236 * Error timestamps are in jiffies which doesn't run while
4237 * suspended and PHY events during resume isn't too uncommon.
4238 * When the two are combined, it can lead to unnecessary speed
4239 * downs if the machine is suspended and resumed repeatedly.
4240 * Clear error history.
4241 */
4242 ata_for_each_link(link, ap, HOST_FIRST)
4243 ata_for_each_dev(dev, link, ALL)
4244 ata_ering_clear(&dev->ering);
4245
4246 ata_acpi_set_state(ap, ap->pm_mesg);
4247
4248 if (ap->ops->port_resume)
4249 rc = ap->ops->port_resume(ap);
4250
4251 /* tell ACPI that we're resuming */
4252 ata_acpi_on_resume(ap);
4253
4254 /* update the flags */
4255 spin_lock_irqsave(ap->lock, flags);
4256 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4257 spin_unlock_irqrestore(ap->lock, flags);
4258 }
4259 #endif /* CONFIG_PM */
This page took 0.174646 seconds and 5 git commands to generate.