libata: add command name parsing for error output
[deliverable/linux.git] / drivers / ata / libata-eh.c
1 /*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
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/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include <scsi/scsi_dbg.h>
44 #include "../scsi/scsi_transport_api.h"
45
46 #include <linux/libata.h>
47
48 #include "libata.h"
49
50 enum {
51 /* speed down verdicts */
52 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
55 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
56
57 /* error flags */
58 ATA_EFLAG_IS_IO = (1 << 0),
59 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
60
61 /* error categories */
62 ATA_ECAT_NONE = 0,
63 ATA_ECAT_ATA_BUS = 1,
64 ATA_ECAT_TOUT_HSM = 2,
65 ATA_ECAT_UNK_DEV = 3,
66 ATA_ECAT_DUBIOUS_NONE = 4,
67 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
68 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
69 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
70 ATA_ECAT_NR = 8,
71
72 ATA_EH_CMD_DFL_TIMEOUT = 5000,
73
74 /* always put at least this amount of time between resets */
75 ATA_EH_RESET_COOL_DOWN = 5000,
76
77 /* Waiting in ->prereset can never be reliable. It's
78 * sometimes nice to wait there but it can't be depended upon;
79 * otherwise, we wouldn't be resetting. Just give it enough
80 * time for most drives to spin up.
81 */
82 ATA_EH_PRERESET_TIMEOUT = 10000,
83 ATA_EH_FASTDRAIN_INTERVAL = 3000,
84
85 ATA_EH_UA_TRIES = 5,
86
87 /* probe speed down parameters, see ata_eh_schedule_probe() */
88 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
89 ATA_EH_PROBE_TRIALS = 2,
90 };
91
92 /* The following table determines how we sequence resets. Each entry
93 * represents timeout for that try. The first try can be soft or
94 * hardreset. All others are hardreset if available. In most cases
95 * the first reset w/ 10sec timeout should succeed. Following entries
96 * are mostly for error handling, hotplug and retarded devices.
97 */
98 static const unsigned long ata_eh_reset_timeouts[] = {
99 10000, /* most drives spin up by 10sec */
100 10000, /* > 99% working drives spin up before 20sec */
101 35000, /* give > 30 secs of idleness for retarded devices */
102 5000, /* and sweet one last chance */
103 ULONG_MAX, /* > 1 min has elapsed, give up */
104 };
105
106 static const unsigned long ata_eh_identify_timeouts[] = {
107 5000, /* covers > 99% of successes and not too boring on failures */
108 10000, /* combined time till here is enough even for media access */
109 30000, /* for true idiots */
110 ULONG_MAX,
111 };
112
113 static const unsigned long ata_eh_other_timeouts[] = {
114 5000, /* same rationale as identify timeout */
115 10000, /* ditto */
116 /* but no merciful 30sec for other commands, it just isn't worth it */
117 ULONG_MAX,
118 };
119
120 struct ata_eh_cmd_timeout_ent {
121 const u8 *commands;
122 const unsigned long *timeouts;
123 };
124
125 /* The following table determines timeouts to use for EH internal
126 * commands. Each table entry is a command class and matches the
127 * commands the entry applies to and the timeout table to use.
128 *
129 * On the retry after a command timed out, the next timeout value from
130 * the table is used. If the table doesn't contain further entries,
131 * the last value is used.
132 *
133 * ehc->cmd_timeout_idx keeps track of which timeout to use per
134 * command class, so if SET_FEATURES times out on the first try, the
135 * next try will use the second timeout value only for that class.
136 */
137 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
138 static const struct ata_eh_cmd_timeout_ent
139 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
140 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
141 .timeouts = ata_eh_identify_timeouts, },
142 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
145 .timeouts = ata_eh_other_timeouts, },
146 { .commands = CMDS(ATA_CMD_SET_FEATURES),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
149 .timeouts = ata_eh_other_timeouts, },
150 };
151 #undef CMDS
152
153 static void __ata_port_freeze(struct ata_port *ap);
154 #ifdef CONFIG_PM
155 static void ata_eh_handle_port_suspend(struct ata_port *ap);
156 static void ata_eh_handle_port_resume(struct ata_port *ap);
157 #else /* CONFIG_PM */
158 static void ata_eh_handle_port_suspend(struct ata_port *ap)
159 { }
160
161 static void ata_eh_handle_port_resume(struct ata_port *ap)
162 { }
163 #endif /* CONFIG_PM */
164
165 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
166 va_list args)
167 {
168 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
169 ATA_EH_DESC_LEN - ehi->desc_len,
170 fmt, args);
171 }
172
173 /**
174 * __ata_ehi_push_desc - push error description without adding separator
175 * @ehi: target EHI
176 * @fmt: printf format string
177 *
178 * Format string according to @fmt and append it to @ehi->desc.
179 *
180 * LOCKING:
181 * spin_lock_irqsave(host lock)
182 */
183 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
184 {
185 va_list args;
186
187 va_start(args, fmt);
188 __ata_ehi_pushv_desc(ehi, fmt, args);
189 va_end(args);
190 }
191
192 /**
193 * ata_ehi_push_desc - push error description with separator
194 * @ehi: target EHI
195 * @fmt: printf format string
196 *
197 * Format string according to @fmt and append it to @ehi->desc.
198 * If @ehi->desc is not empty, ", " is added in-between.
199 *
200 * LOCKING:
201 * spin_lock_irqsave(host lock)
202 */
203 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
204 {
205 va_list args;
206
207 if (ehi->desc_len)
208 __ata_ehi_push_desc(ehi, ", ");
209
210 va_start(args, fmt);
211 __ata_ehi_pushv_desc(ehi, fmt, args);
212 va_end(args);
213 }
214
215 /**
216 * ata_ehi_clear_desc - clean error description
217 * @ehi: target EHI
218 *
219 * Clear @ehi->desc.
220 *
221 * LOCKING:
222 * spin_lock_irqsave(host lock)
223 */
224 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
225 {
226 ehi->desc[0] = '\0';
227 ehi->desc_len = 0;
228 }
229
230 /**
231 * ata_port_desc - append port description
232 * @ap: target ATA port
233 * @fmt: printf format string
234 *
235 * Format string according to @fmt and append it to port
236 * description. If port description is not empty, " " is added
237 * in-between. This function is to be used while initializing
238 * ata_host. The description is printed on host registration.
239 *
240 * LOCKING:
241 * None.
242 */
243 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
244 {
245 va_list args;
246
247 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
248
249 if (ap->link.eh_info.desc_len)
250 __ata_ehi_push_desc(&ap->link.eh_info, " ");
251
252 va_start(args, fmt);
253 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
254 va_end(args);
255 }
256
257 #ifdef CONFIG_PCI
258
259 /**
260 * ata_port_pbar_desc - append PCI BAR description
261 * @ap: target ATA port
262 * @bar: target PCI BAR
263 * @offset: offset into PCI BAR
264 * @name: name of the area
265 *
266 * If @offset is negative, this function formats a string which
267 * contains the name, address, size and type of the BAR and
268 * appends it to the port description. If @offset is zero or
269 * positive, only name and offsetted address is appended.
270 *
271 * LOCKING:
272 * None.
273 */
274 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
275 const char *name)
276 {
277 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
278 char *type = "";
279 unsigned long long start, len;
280
281 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
282 type = "m";
283 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
284 type = "i";
285
286 start = (unsigned long long)pci_resource_start(pdev, bar);
287 len = (unsigned long long)pci_resource_len(pdev, bar);
288
289 if (offset < 0)
290 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
291 else
292 ata_port_desc(ap, "%s 0x%llx", name,
293 start + (unsigned long long)offset);
294 }
295
296 #endif /* CONFIG_PCI */
297
298 static int ata_lookup_timeout_table(u8 cmd)
299 {
300 int i;
301
302 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
303 const u8 *cur;
304
305 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
306 if (*cur == cmd)
307 return i;
308 }
309
310 return -1;
311 }
312
313 /**
314 * ata_internal_cmd_timeout - determine timeout for an internal command
315 * @dev: target device
316 * @cmd: internal command to be issued
317 *
318 * Determine timeout for internal command @cmd for @dev.
319 *
320 * LOCKING:
321 * EH context.
322 *
323 * RETURNS:
324 * Determined timeout.
325 */
326 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
327 {
328 struct ata_eh_context *ehc = &dev->link->eh_context;
329 int ent = ata_lookup_timeout_table(cmd);
330 int idx;
331
332 if (ent < 0)
333 return ATA_EH_CMD_DFL_TIMEOUT;
334
335 idx = ehc->cmd_timeout_idx[dev->devno][ent];
336 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
337 }
338
339 /**
340 * ata_internal_cmd_timed_out - notification for internal command timeout
341 * @dev: target device
342 * @cmd: internal command which timed out
343 *
344 * Notify EH that internal command @cmd for @dev timed out. This
345 * function should be called only for commands whose timeouts are
346 * determined using ata_internal_cmd_timeout().
347 *
348 * LOCKING:
349 * EH context.
350 */
351 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
352 {
353 struct ata_eh_context *ehc = &dev->link->eh_context;
354 int ent = ata_lookup_timeout_table(cmd);
355 int idx;
356
357 if (ent < 0)
358 return;
359
360 idx = ehc->cmd_timeout_idx[dev->devno][ent];
361 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
362 ehc->cmd_timeout_idx[dev->devno][ent]++;
363 }
364
365 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
366 unsigned int err_mask)
367 {
368 struct ata_ering_entry *ent;
369
370 WARN_ON(!err_mask);
371
372 ering->cursor++;
373 ering->cursor %= ATA_ERING_SIZE;
374
375 ent = &ering->ring[ering->cursor];
376 ent->eflags = eflags;
377 ent->err_mask = err_mask;
378 ent->timestamp = get_jiffies_64();
379 }
380
381 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
382 {
383 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
384
385 if (ent->err_mask)
386 return ent;
387 return NULL;
388 }
389
390 static void ata_ering_clear(struct ata_ering *ering)
391 {
392 memset(ering, 0, sizeof(*ering));
393 }
394
395 static int ata_ering_map(struct ata_ering *ering,
396 int (*map_fn)(struct ata_ering_entry *, void *),
397 void *arg)
398 {
399 int idx, rc = 0;
400 struct ata_ering_entry *ent;
401
402 idx = ering->cursor;
403 do {
404 ent = &ering->ring[idx];
405 if (!ent->err_mask)
406 break;
407 rc = map_fn(ent, arg);
408 if (rc)
409 break;
410 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
411 } while (idx != ering->cursor);
412
413 return rc;
414 }
415
416 static unsigned int ata_eh_dev_action(struct ata_device *dev)
417 {
418 struct ata_eh_context *ehc = &dev->link->eh_context;
419
420 return ehc->i.action | ehc->i.dev_action[dev->devno];
421 }
422
423 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
424 struct ata_eh_info *ehi, unsigned int action)
425 {
426 struct ata_device *tdev;
427
428 if (!dev) {
429 ehi->action &= ~action;
430 ata_for_each_dev(tdev, link, ALL)
431 ehi->dev_action[tdev->devno] &= ~action;
432 } else {
433 /* doesn't make sense for port-wide EH actions */
434 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
435
436 /* break ehi->action into ehi->dev_action */
437 if (ehi->action & action) {
438 ata_for_each_dev(tdev, link, ALL)
439 ehi->dev_action[tdev->devno] |=
440 ehi->action & action;
441 ehi->action &= ~action;
442 }
443
444 /* turn off the specified per-dev action */
445 ehi->dev_action[dev->devno] &= ~action;
446 }
447 }
448
449 /**
450 * ata_scsi_timed_out - SCSI layer time out callback
451 * @cmd: timed out SCSI command
452 *
453 * Handles SCSI layer timeout. We race with normal completion of
454 * the qc for @cmd. If the qc is already gone, we lose and let
455 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
456 * timed out and EH should be invoked. Prevent ata_qc_complete()
457 * from finishing it by setting EH_SCHEDULED and return
458 * EH_NOT_HANDLED.
459 *
460 * TODO: kill this function once old EH is gone.
461 *
462 * LOCKING:
463 * Called from timer context
464 *
465 * RETURNS:
466 * EH_HANDLED or EH_NOT_HANDLED
467 */
468 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
469 {
470 struct Scsi_Host *host = cmd->device->host;
471 struct ata_port *ap = ata_shost_to_port(host);
472 unsigned long flags;
473 struct ata_queued_cmd *qc;
474 enum blk_eh_timer_return ret;
475
476 DPRINTK("ENTER\n");
477
478 if (ap->ops->error_handler) {
479 ret = BLK_EH_NOT_HANDLED;
480 goto out;
481 }
482
483 ret = BLK_EH_HANDLED;
484 spin_lock_irqsave(ap->lock, flags);
485 qc = ata_qc_from_tag(ap, ap->link.active_tag);
486 if (qc) {
487 WARN_ON(qc->scsicmd != cmd);
488 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
489 qc->err_mask |= AC_ERR_TIMEOUT;
490 ret = BLK_EH_NOT_HANDLED;
491 }
492 spin_unlock_irqrestore(ap->lock, flags);
493
494 out:
495 DPRINTK("EXIT, ret=%d\n", ret);
496 return ret;
497 }
498
499 static void ata_eh_unload(struct ata_port *ap)
500 {
501 struct ata_link *link;
502 struct ata_device *dev;
503 unsigned long flags;
504
505 /* Restore SControl IPM and SPD for the next driver and
506 * disable attached devices.
507 */
508 ata_for_each_link(link, ap, PMP_FIRST) {
509 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
510 ata_for_each_dev(dev, link, ALL)
511 ata_dev_disable(dev);
512 }
513
514 /* freeze and set UNLOADED */
515 spin_lock_irqsave(ap->lock, flags);
516
517 ata_port_freeze(ap); /* won't be thawed */
518 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
519 ap->pflags |= ATA_PFLAG_UNLOADED;
520
521 spin_unlock_irqrestore(ap->lock, flags);
522 }
523
524 /**
525 * ata_scsi_error - SCSI layer error handler callback
526 * @host: SCSI host on which error occurred
527 *
528 * Handles SCSI-layer-thrown error events.
529 *
530 * LOCKING:
531 * Inherited from SCSI layer (none, can sleep)
532 *
533 * RETURNS:
534 * Zero.
535 */
536 void ata_scsi_error(struct Scsi_Host *host)
537 {
538 struct ata_port *ap = ata_shost_to_port(host);
539 int i;
540 unsigned long flags;
541
542 DPRINTK("ENTER\n");
543
544 /* synchronize with port task */
545 ata_port_flush_task(ap);
546
547 /* synchronize with host lock and sort out timeouts */
548
549 /* For new EH, all qcs are finished in one of three ways -
550 * normal completion, error completion, and SCSI timeout.
551 * Both completions can race against SCSI timeout. When normal
552 * completion wins, the qc never reaches EH. When error
553 * completion wins, the qc has ATA_QCFLAG_FAILED set.
554 *
555 * When SCSI timeout wins, things are a bit more complex.
556 * Normal or error completion can occur after the timeout but
557 * before this point. In such cases, both types of
558 * completions are honored. A scmd is determined to have
559 * timed out iff its associated qc is active and not failed.
560 */
561 if (ap->ops->error_handler) {
562 struct scsi_cmnd *scmd, *tmp;
563 int nr_timedout = 0;
564
565 spin_lock_irqsave(ap->lock, flags);
566
567 /* This must occur under the ap->lock as we don't want
568 a polled recovery to race the real interrupt handler
569
570 The lost_interrupt handler checks for any completed but
571 non-notified command and completes much like an IRQ handler.
572
573 We then fall into the error recovery code which will treat
574 this as if normal completion won the race */
575
576 if (ap->ops->lost_interrupt)
577 ap->ops->lost_interrupt(ap);
578
579 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
580 struct ata_queued_cmd *qc;
581
582 for (i = 0; i < ATA_MAX_QUEUE; i++) {
583 qc = __ata_qc_from_tag(ap, i);
584 if (qc->flags & ATA_QCFLAG_ACTIVE &&
585 qc->scsicmd == scmd)
586 break;
587 }
588
589 if (i < ATA_MAX_QUEUE) {
590 /* the scmd has an associated qc */
591 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
592 /* which hasn't failed yet, timeout */
593 qc->err_mask |= AC_ERR_TIMEOUT;
594 qc->flags |= ATA_QCFLAG_FAILED;
595 nr_timedout++;
596 }
597 } else {
598 /* Normal completion occurred after
599 * SCSI timeout but before this point.
600 * Successfully complete it.
601 */
602 scmd->retries = scmd->allowed;
603 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
604 }
605 }
606
607 /* If we have timed out qcs. They belong to EH from
608 * this point but the state of the controller is
609 * unknown. Freeze the port to make sure the IRQ
610 * handler doesn't diddle with those qcs. This must
611 * be done atomically w.r.t. setting QCFLAG_FAILED.
612 */
613 if (nr_timedout)
614 __ata_port_freeze(ap);
615
616 spin_unlock_irqrestore(ap->lock, flags);
617
618 /* initialize eh_tries */
619 ap->eh_tries = ATA_EH_MAX_TRIES;
620 } else
621 spin_unlock_wait(ap->lock);
622
623 /* If we timed raced normal completion and there is nothing to
624 recover nr_timedout == 0 why exactly are we doing error recovery ? */
625
626 repeat:
627 /* invoke error handler */
628 if (ap->ops->error_handler) {
629 struct ata_link *link;
630
631 /* kill fast drain timer */
632 del_timer_sync(&ap->fastdrain_timer);
633
634 /* process port resume request */
635 ata_eh_handle_port_resume(ap);
636
637 /* fetch & clear EH info */
638 spin_lock_irqsave(ap->lock, flags);
639
640 ata_for_each_link(link, ap, HOST_FIRST) {
641 struct ata_eh_context *ehc = &link->eh_context;
642 struct ata_device *dev;
643
644 memset(&link->eh_context, 0, sizeof(link->eh_context));
645 link->eh_context.i = link->eh_info;
646 memset(&link->eh_info, 0, sizeof(link->eh_info));
647
648 ata_for_each_dev(dev, link, ENABLED) {
649 int devno = dev->devno;
650
651 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
652 if (ata_ncq_enabled(dev))
653 ehc->saved_ncq_enabled |= 1 << devno;
654 }
655 }
656
657 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
658 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
659 ap->excl_link = NULL; /* don't maintain exclusion over EH */
660
661 spin_unlock_irqrestore(ap->lock, flags);
662
663 /* invoke EH, skip if unloading or suspended */
664 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
665 ap->ops->error_handler(ap);
666 else {
667 /* if unloading, commence suicide */
668 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
669 !(ap->pflags & ATA_PFLAG_UNLOADED))
670 ata_eh_unload(ap);
671 ata_eh_finish(ap);
672 }
673
674 /* process port suspend request */
675 ata_eh_handle_port_suspend(ap);
676
677 /* Exception might have happend after ->error_handler
678 * recovered the port but before this point. Repeat
679 * EH in such case.
680 */
681 spin_lock_irqsave(ap->lock, flags);
682
683 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
684 if (--ap->eh_tries) {
685 spin_unlock_irqrestore(ap->lock, flags);
686 goto repeat;
687 }
688 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
689 "tries, giving up\n", ATA_EH_MAX_TRIES);
690 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
691 }
692
693 /* this run is complete, make sure EH info is clear */
694 ata_for_each_link(link, ap, HOST_FIRST)
695 memset(&link->eh_info, 0, sizeof(link->eh_info));
696
697 /* Clear host_eh_scheduled while holding ap->lock such
698 * that if exception occurs after this point but
699 * before EH completion, SCSI midlayer will
700 * re-initiate EH.
701 */
702 host->host_eh_scheduled = 0;
703
704 spin_unlock_irqrestore(ap->lock, flags);
705 } else {
706 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
707 ap->ops->eng_timeout(ap);
708 }
709
710 /* finish or retry handled scmd's and clean up */
711 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
712
713 scsi_eh_flush_done_q(&ap->eh_done_q);
714
715 /* clean up */
716 spin_lock_irqsave(ap->lock, flags);
717
718 if (ap->pflags & ATA_PFLAG_LOADING)
719 ap->pflags &= ~ATA_PFLAG_LOADING;
720 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
721 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
722
723 if (ap->pflags & ATA_PFLAG_RECOVERED)
724 ata_port_printk(ap, KERN_INFO, "EH complete\n");
725
726 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
727
728 /* tell wait_eh that we're done */
729 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
730 wake_up_all(&ap->eh_wait_q);
731
732 spin_unlock_irqrestore(ap->lock, flags);
733
734 DPRINTK("EXIT\n");
735 }
736
737 /**
738 * ata_port_wait_eh - Wait for the currently pending EH to complete
739 * @ap: Port to wait EH for
740 *
741 * Wait until the currently pending EH is complete.
742 *
743 * LOCKING:
744 * Kernel thread context (may sleep).
745 */
746 void ata_port_wait_eh(struct ata_port *ap)
747 {
748 unsigned long flags;
749 DEFINE_WAIT(wait);
750
751 retry:
752 spin_lock_irqsave(ap->lock, flags);
753
754 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
755 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
756 spin_unlock_irqrestore(ap->lock, flags);
757 schedule();
758 spin_lock_irqsave(ap->lock, flags);
759 }
760 finish_wait(&ap->eh_wait_q, &wait);
761
762 spin_unlock_irqrestore(ap->lock, flags);
763
764 /* make sure SCSI EH is complete */
765 if (scsi_host_in_recovery(ap->scsi_host)) {
766 msleep(10);
767 goto retry;
768 }
769 }
770
771 static int ata_eh_nr_in_flight(struct ata_port *ap)
772 {
773 unsigned int tag;
774 int nr = 0;
775
776 /* count only non-internal commands */
777 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
778 if (ata_qc_from_tag(ap, tag))
779 nr++;
780
781 return nr;
782 }
783
784 void ata_eh_fastdrain_timerfn(unsigned long arg)
785 {
786 struct ata_port *ap = (void *)arg;
787 unsigned long flags;
788 int cnt;
789
790 spin_lock_irqsave(ap->lock, flags);
791
792 cnt = ata_eh_nr_in_flight(ap);
793
794 /* are we done? */
795 if (!cnt)
796 goto out_unlock;
797
798 if (cnt == ap->fastdrain_cnt) {
799 unsigned int tag;
800
801 /* No progress during the last interval, tag all
802 * in-flight qcs as timed out and freeze the port.
803 */
804 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
805 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
806 if (qc)
807 qc->err_mask |= AC_ERR_TIMEOUT;
808 }
809
810 ata_port_freeze(ap);
811 } else {
812 /* some qcs have finished, give it another chance */
813 ap->fastdrain_cnt = cnt;
814 ap->fastdrain_timer.expires =
815 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
816 add_timer(&ap->fastdrain_timer);
817 }
818
819 out_unlock:
820 spin_unlock_irqrestore(ap->lock, flags);
821 }
822
823 /**
824 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
825 * @ap: target ATA port
826 * @fastdrain: activate fast drain
827 *
828 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
829 * is non-zero and EH wasn't pending before. Fast drain ensures
830 * that EH kicks in in timely manner.
831 *
832 * LOCKING:
833 * spin_lock_irqsave(host lock)
834 */
835 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
836 {
837 int cnt;
838
839 /* already scheduled? */
840 if (ap->pflags & ATA_PFLAG_EH_PENDING)
841 return;
842
843 ap->pflags |= ATA_PFLAG_EH_PENDING;
844
845 if (!fastdrain)
846 return;
847
848 /* do we have in-flight qcs? */
849 cnt = ata_eh_nr_in_flight(ap);
850 if (!cnt)
851 return;
852
853 /* activate fast drain */
854 ap->fastdrain_cnt = cnt;
855 ap->fastdrain_timer.expires =
856 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
857 add_timer(&ap->fastdrain_timer);
858 }
859
860 /**
861 * ata_qc_schedule_eh - schedule qc for error handling
862 * @qc: command to schedule error handling for
863 *
864 * Schedule error handling for @qc. EH will kick in as soon as
865 * other commands are drained.
866 *
867 * LOCKING:
868 * spin_lock_irqsave(host lock)
869 */
870 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
871 {
872 struct ata_port *ap = qc->ap;
873
874 WARN_ON(!ap->ops->error_handler);
875
876 qc->flags |= ATA_QCFLAG_FAILED;
877 ata_eh_set_pending(ap, 1);
878
879 /* The following will fail if timeout has already expired.
880 * ata_scsi_error() takes care of such scmds on EH entry.
881 * Note that ATA_QCFLAG_FAILED is unconditionally set after
882 * this function completes.
883 */
884 blk_abort_request(qc->scsicmd->request);
885 }
886
887 /**
888 * ata_port_schedule_eh - schedule error handling without a qc
889 * @ap: ATA port to schedule EH for
890 *
891 * Schedule error handling for @ap. EH will kick in as soon as
892 * all commands are drained.
893 *
894 * LOCKING:
895 * spin_lock_irqsave(host lock)
896 */
897 void ata_port_schedule_eh(struct ata_port *ap)
898 {
899 WARN_ON(!ap->ops->error_handler);
900
901 if (ap->pflags & ATA_PFLAG_INITIALIZING)
902 return;
903
904 ata_eh_set_pending(ap, 1);
905 scsi_schedule_eh(ap->scsi_host);
906
907 DPRINTK("port EH scheduled\n");
908 }
909
910 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
911 {
912 int tag, nr_aborted = 0;
913
914 WARN_ON(!ap->ops->error_handler);
915
916 /* we're gonna abort all commands, no need for fast drain */
917 ata_eh_set_pending(ap, 0);
918
919 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
920 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
921
922 if (qc && (!link || qc->dev->link == link)) {
923 qc->flags |= ATA_QCFLAG_FAILED;
924 ata_qc_complete(qc);
925 nr_aborted++;
926 }
927 }
928
929 if (!nr_aborted)
930 ata_port_schedule_eh(ap);
931
932 return nr_aborted;
933 }
934
935 /**
936 * ata_link_abort - abort all qc's on the link
937 * @link: ATA link to abort qc's for
938 *
939 * Abort all active qc's active on @link and schedule EH.
940 *
941 * LOCKING:
942 * spin_lock_irqsave(host lock)
943 *
944 * RETURNS:
945 * Number of aborted qc's.
946 */
947 int ata_link_abort(struct ata_link *link)
948 {
949 return ata_do_link_abort(link->ap, link);
950 }
951
952 /**
953 * ata_port_abort - abort all qc's on the port
954 * @ap: ATA port to abort qc's for
955 *
956 * Abort all active qc's of @ap and schedule EH.
957 *
958 * LOCKING:
959 * spin_lock_irqsave(host_set lock)
960 *
961 * RETURNS:
962 * Number of aborted qc's.
963 */
964 int ata_port_abort(struct ata_port *ap)
965 {
966 return ata_do_link_abort(ap, NULL);
967 }
968
969 /**
970 * __ata_port_freeze - freeze port
971 * @ap: ATA port to freeze
972 *
973 * This function is called when HSM violation or some other
974 * condition disrupts normal operation of the port. Frozen port
975 * is not allowed to perform any operation until the port is
976 * thawed, which usually follows a successful reset.
977 *
978 * ap->ops->freeze() callback can be used for freezing the port
979 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
980 * port cannot be frozen hardware-wise, the interrupt handler
981 * must ack and clear interrupts unconditionally while the port
982 * is frozen.
983 *
984 * LOCKING:
985 * spin_lock_irqsave(host lock)
986 */
987 static void __ata_port_freeze(struct ata_port *ap)
988 {
989 WARN_ON(!ap->ops->error_handler);
990
991 if (ap->ops->freeze)
992 ap->ops->freeze(ap);
993
994 ap->pflags |= ATA_PFLAG_FROZEN;
995
996 DPRINTK("ata%u port frozen\n", ap->print_id);
997 }
998
999 /**
1000 * ata_port_freeze - abort & freeze port
1001 * @ap: ATA port to freeze
1002 *
1003 * Abort and freeze @ap. The freeze operation must be called
1004 * first, because some hardware requires special operations
1005 * before the taskfile registers are accessible.
1006 *
1007 * LOCKING:
1008 * spin_lock_irqsave(host lock)
1009 *
1010 * RETURNS:
1011 * Number of aborted commands.
1012 */
1013 int ata_port_freeze(struct ata_port *ap)
1014 {
1015 int nr_aborted;
1016
1017 WARN_ON(!ap->ops->error_handler);
1018
1019 __ata_port_freeze(ap);
1020 nr_aborted = ata_port_abort(ap);
1021
1022 return nr_aborted;
1023 }
1024
1025 /**
1026 * sata_async_notification - SATA async notification handler
1027 * @ap: ATA port where async notification is received
1028 *
1029 * Handler to be called when async notification via SDB FIS is
1030 * received. This function schedules EH if necessary.
1031 *
1032 * LOCKING:
1033 * spin_lock_irqsave(host lock)
1034 *
1035 * RETURNS:
1036 * 1 if EH is scheduled, 0 otherwise.
1037 */
1038 int sata_async_notification(struct ata_port *ap)
1039 {
1040 u32 sntf;
1041 int rc;
1042
1043 if (!(ap->flags & ATA_FLAG_AN))
1044 return 0;
1045
1046 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1047 if (rc == 0)
1048 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1049
1050 if (!sata_pmp_attached(ap) || rc) {
1051 /* PMP is not attached or SNTF is not available */
1052 if (!sata_pmp_attached(ap)) {
1053 /* PMP is not attached. Check whether ATAPI
1054 * AN is configured. If so, notify media
1055 * change.
1056 */
1057 struct ata_device *dev = ap->link.device;
1058
1059 if ((dev->class == ATA_DEV_ATAPI) &&
1060 (dev->flags & ATA_DFLAG_AN))
1061 ata_scsi_media_change_notify(dev);
1062 return 0;
1063 } else {
1064 /* PMP is attached but SNTF is not available.
1065 * ATAPI async media change notification is
1066 * not used. The PMP must be reporting PHY
1067 * status change, schedule EH.
1068 */
1069 ata_port_schedule_eh(ap);
1070 return 1;
1071 }
1072 } else {
1073 /* PMP is attached and SNTF is available */
1074 struct ata_link *link;
1075
1076 /* check and notify ATAPI AN */
1077 ata_for_each_link(link, ap, EDGE) {
1078 if (!(sntf & (1 << link->pmp)))
1079 continue;
1080
1081 if ((link->device->class == ATA_DEV_ATAPI) &&
1082 (link->device->flags & ATA_DFLAG_AN))
1083 ata_scsi_media_change_notify(link->device);
1084 }
1085
1086 /* If PMP is reporting that PHY status of some
1087 * downstream ports has changed, schedule EH.
1088 */
1089 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1090 ata_port_schedule_eh(ap);
1091 return 1;
1092 }
1093
1094 return 0;
1095 }
1096 }
1097
1098 /**
1099 * ata_eh_freeze_port - EH helper to freeze port
1100 * @ap: ATA port to freeze
1101 *
1102 * Freeze @ap.
1103 *
1104 * LOCKING:
1105 * None.
1106 */
1107 void ata_eh_freeze_port(struct ata_port *ap)
1108 {
1109 unsigned long flags;
1110
1111 if (!ap->ops->error_handler)
1112 return;
1113
1114 spin_lock_irqsave(ap->lock, flags);
1115 __ata_port_freeze(ap);
1116 spin_unlock_irqrestore(ap->lock, flags);
1117 }
1118
1119 /**
1120 * ata_port_thaw_port - EH helper to thaw port
1121 * @ap: ATA port to thaw
1122 *
1123 * Thaw frozen port @ap.
1124 *
1125 * LOCKING:
1126 * None.
1127 */
1128 void ata_eh_thaw_port(struct ata_port *ap)
1129 {
1130 unsigned long flags;
1131
1132 if (!ap->ops->error_handler)
1133 return;
1134
1135 spin_lock_irqsave(ap->lock, flags);
1136
1137 ap->pflags &= ~ATA_PFLAG_FROZEN;
1138
1139 if (ap->ops->thaw)
1140 ap->ops->thaw(ap);
1141
1142 spin_unlock_irqrestore(ap->lock, flags);
1143
1144 DPRINTK("ata%u port thawed\n", ap->print_id);
1145 }
1146
1147 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1148 {
1149 /* nada */
1150 }
1151
1152 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1153 {
1154 struct ata_port *ap = qc->ap;
1155 struct scsi_cmnd *scmd = qc->scsicmd;
1156 unsigned long flags;
1157
1158 spin_lock_irqsave(ap->lock, flags);
1159 qc->scsidone = ata_eh_scsidone;
1160 __ata_qc_complete(qc);
1161 WARN_ON(ata_tag_valid(qc->tag));
1162 spin_unlock_irqrestore(ap->lock, flags);
1163
1164 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1165 }
1166
1167 /**
1168 * ata_eh_qc_complete - Complete an active ATA command from EH
1169 * @qc: Command to complete
1170 *
1171 * Indicate to the mid and upper layers that an ATA command has
1172 * completed. To be used from EH.
1173 */
1174 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1175 {
1176 struct scsi_cmnd *scmd = qc->scsicmd;
1177 scmd->retries = scmd->allowed;
1178 __ata_eh_qc_complete(qc);
1179 }
1180
1181 /**
1182 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1183 * @qc: Command to retry
1184 *
1185 * Indicate to the mid and upper layers that an ATA command
1186 * should be retried. To be used from EH.
1187 *
1188 * SCSI midlayer limits the number of retries to scmd->allowed.
1189 * scmd->retries is decremented for commands which get retried
1190 * due to unrelated failures (qc->err_mask is zero).
1191 */
1192 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1193 {
1194 struct scsi_cmnd *scmd = qc->scsicmd;
1195 if (!qc->err_mask && scmd->retries)
1196 scmd->retries--;
1197 __ata_eh_qc_complete(qc);
1198 }
1199
1200 /**
1201 * ata_dev_disable - disable ATA device
1202 * @dev: ATA device to disable
1203 *
1204 * Disable @dev.
1205 *
1206 * Locking:
1207 * EH context.
1208 */
1209 void ata_dev_disable(struct ata_device *dev)
1210 {
1211 if (!ata_dev_enabled(dev))
1212 return;
1213
1214 if (ata_msg_drv(dev->link->ap))
1215 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1216 ata_acpi_on_disable(dev);
1217 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1218 dev->class++;
1219
1220 /* From now till the next successful probe, ering is used to
1221 * track probe failures. Clear accumulated device error info.
1222 */
1223 ata_ering_clear(&dev->ering);
1224 }
1225
1226 /**
1227 * ata_eh_detach_dev - detach ATA device
1228 * @dev: ATA device to detach
1229 *
1230 * Detach @dev.
1231 *
1232 * LOCKING:
1233 * None.
1234 */
1235 void ata_eh_detach_dev(struct ata_device *dev)
1236 {
1237 struct ata_link *link = dev->link;
1238 struct ata_port *ap = link->ap;
1239 struct ata_eh_context *ehc = &link->eh_context;
1240 unsigned long flags;
1241
1242 ata_dev_disable(dev);
1243
1244 spin_lock_irqsave(ap->lock, flags);
1245
1246 dev->flags &= ~ATA_DFLAG_DETACH;
1247
1248 if (ata_scsi_offline_dev(dev)) {
1249 dev->flags |= ATA_DFLAG_DETACHED;
1250 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1251 }
1252
1253 /* clear per-dev EH info */
1254 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1255 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1256 ehc->saved_xfer_mode[dev->devno] = 0;
1257 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1258
1259 spin_unlock_irqrestore(ap->lock, flags);
1260 }
1261
1262 /**
1263 * ata_eh_about_to_do - about to perform eh_action
1264 * @link: target ATA link
1265 * @dev: target ATA dev for per-dev action (can be NULL)
1266 * @action: action about to be performed
1267 *
1268 * Called just before performing EH actions to clear related bits
1269 * in @link->eh_info such that eh actions are not unnecessarily
1270 * repeated.
1271 *
1272 * LOCKING:
1273 * None.
1274 */
1275 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1276 unsigned int action)
1277 {
1278 struct ata_port *ap = link->ap;
1279 struct ata_eh_info *ehi = &link->eh_info;
1280 struct ata_eh_context *ehc = &link->eh_context;
1281 unsigned long flags;
1282
1283 spin_lock_irqsave(ap->lock, flags);
1284
1285 ata_eh_clear_action(link, dev, ehi, action);
1286
1287 /* About to take EH action, set RECOVERED. Ignore actions on
1288 * slave links as master will do them again.
1289 */
1290 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1291 ap->pflags |= ATA_PFLAG_RECOVERED;
1292
1293 spin_unlock_irqrestore(ap->lock, flags);
1294 }
1295
1296 /**
1297 * ata_eh_done - EH action complete
1298 * @ap: target ATA port
1299 * @dev: target ATA dev for per-dev action (can be NULL)
1300 * @action: action just completed
1301 *
1302 * Called right after performing EH actions to clear related bits
1303 * in @link->eh_context.
1304 *
1305 * LOCKING:
1306 * None.
1307 */
1308 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1309 unsigned int action)
1310 {
1311 struct ata_eh_context *ehc = &link->eh_context;
1312
1313 ata_eh_clear_action(link, dev, &ehc->i, action);
1314 }
1315
1316 /**
1317 * ata_err_string - convert err_mask to descriptive string
1318 * @err_mask: error mask to convert to string
1319 *
1320 * Convert @err_mask to descriptive string. Errors are
1321 * prioritized according to severity and only the most severe
1322 * error is reported.
1323 *
1324 * LOCKING:
1325 * None.
1326 *
1327 * RETURNS:
1328 * Descriptive string for @err_mask
1329 */
1330 static const char *ata_err_string(unsigned int err_mask)
1331 {
1332 if (err_mask & AC_ERR_HOST_BUS)
1333 return "host bus error";
1334 if (err_mask & AC_ERR_ATA_BUS)
1335 return "ATA bus error";
1336 if (err_mask & AC_ERR_TIMEOUT)
1337 return "timeout";
1338 if (err_mask & AC_ERR_HSM)
1339 return "HSM violation";
1340 if (err_mask & AC_ERR_SYSTEM)
1341 return "internal error";
1342 if (err_mask & AC_ERR_MEDIA)
1343 return "media error";
1344 if (err_mask & AC_ERR_INVALID)
1345 return "invalid argument";
1346 if (err_mask & AC_ERR_DEV)
1347 return "device error";
1348 return "unknown error";
1349 }
1350
1351 /**
1352 * ata_read_log_page - read a specific log page
1353 * @dev: target device
1354 * @page: page to read
1355 * @buf: buffer to store read page
1356 * @sectors: number of sectors to read
1357 *
1358 * Read log page using READ_LOG_EXT command.
1359 *
1360 * LOCKING:
1361 * Kernel thread context (may sleep).
1362 *
1363 * RETURNS:
1364 * 0 on success, AC_ERR_* mask otherwise.
1365 */
1366 static unsigned int ata_read_log_page(struct ata_device *dev,
1367 u8 page, void *buf, unsigned int sectors)
1368 {
1369 struct ata_taskfile tf;
1370 unsigned int err_mask;
1371
1372 DPRINTK("read log page - page %d\n", page);
1373
1374 ata_tf_init(dev, &tf);
1375 tf.command = ATA_CMD_READ_LOG_EXT;
1376 tf.lbal = page;
1377 tf.nsect = sectors;
1378 tf.hob_nsect = sectors >> 8;
1379 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1380 tf.protocol = ATA_PROT_PIO;
1381
1382 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1383 buf, sectors * ATA_SECT_SIZE, 0);
1384
1385 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1386 return err_mask;
1387 }
1388
1389 /**
1390 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1391 * @dev: Device to read log page 10h from
1392 * @tag: Resulting tag of the failed command
1393 * @tf: Resulting taskfile registers of the failed command
1394 *
1395 * Read log page 10h to obtain NCQ error details and clear error
1396 * condition.
1397 *
1398 * LOCKING:
1399 * Kernel thread context (may sleep).
1400 *
1401 * RETURNS:
1402 * 0 on success, -errno otherwise.
1403 */
1404 static int ata_eh_read_log_10h(struct ata_device *dev,
1405 int *tag, struct ata_taskfile *tf)
1406 {
1407 u8 *buf = dev->link->ap->sector_buf;
1408 unsigned int err_mask;
1409 u8 csum;
1410 int i;
1411
1412 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1413 if (err_mask)
1414 return -EIO;
1415
1416 csum = 0;
1417 for (i = 0; i < ATA_SECT_SIZE; i++)
1418 csum += buf[i];
1419 if (csum)
1420 ata_dev_printk(dev, KERN_WARNING,
1421 "invalid checksum 0x%x on log page 10h\n", csum);
1422
1423 if (buf[0] & 0x80)
1424 return -ENOENT;
1425
1426 *tag = buf[0] & 0x1f;
1427
1428 tf->command = buf[2];
1429 tf->feature = buf[3];
1430 tf->lbal = buf[4];
1431 tf->lbam = buf[5];
1432 tf->lbah = buf[6];
1433 tf->device = buf[7];
1434 tf->hob_lbal = buf[8];
1435 tf->hob_lbam = buf[9];
1436 tf->hob_lbah = buf[10];
1437 tf->nsect = buf[12];
1438 tf->hob_nsect = buf[13];
1439
1440 return 0;
1441 }
1442
1443 /**
1444 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1445 * @dev: target ATAPI device
1446 * @r_sense_key: out parameter for sense_key
1447 *
1448 * Perform ATAPI TEST_UNIT_READY.
1449 *
1450 * LOCKING:
1451 * EH context (may sleep).
1452 *
1453 * RETURNS:
1454 * 0 on success, AC_ERR_* mask on failure.
1455 */
1456 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1457 {
1458 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1459 struct ata_taskfile tf;
1460 unsigned int err_mask;
1461
1462 ata_tf_init(dev, &tf);
1463
1464 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1465 tf.command = ATA_CMD_PACKET;
1466 tf.protocol = ATAPI_PROT_NODATA;
1467
1468 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1469 if (err_mask == AC_ERR_DEV)
1470 *r_sense_key = tf.feature >> 4;
1471 return err_mask;
1472 }
1473
1474 /**
1475 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1476 * @dev: device to perform REQUEST_SENSE to
1477 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1478 * @dfl_sense_key: default sense key to use
1479 *
1480 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1481 * SENSE. This function is EH helper.
1482 *
1483 * LOCKING:
1484 * Kernel thread context (may sleep).
1485 *
1486 * RETURNS:
1487 * 0 on success, AC_ERR_* mask on failure
1488 */
1489 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1490 u8 *sense_buf, u8 dfl_sense_key)
1491 {
1492 u8 cdb[ATAPI_CDB_LEN] =
1493 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1494 struct ata_port *ap = dev->link->ap;
1495 struct ata_taskfile tf;
1496
1497 DPRINTK("ATAPI request sense\n");
1498
1499 /* FIXME: is this needed? */
1500 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1501
1502 /* initialize sense_buf with the error register,
1503 * for the case where they are -not- overwritten
1504 */
1505 sense_buf[0] = 0x70;
1506 sense_buf[2] = dfl_sense_key;
1507
1508 /* some devices time out if garbage left in tf */
1509 ata_tf_init(dev, &tf);
1510
1511 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1512 tf.command = ATA_CMD_PACKET;
1513
1514 /* is it pointless to prefer PIO for "safety reasons"? */
1515 if (ap->flags & ATA_FLAG_PIO_DMA) {
1516 tf.protocol = ATAPI_PROT_DMA;
1517 tf.feature |= ATAPI_PKT_DMA;
1518 } else {
1519 tf.protocol = ATAPI_PROT_PIO;
1520 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1521 tf.lbah = 0;
1522 }
1523
1524 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1525 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1526 }
1527
1528 /**
1529 * ata_eh_analyze_serror - analyze SError for a failed port
1530 * @link: ATA link to analyze SError for
1531 *
1532 * Analyze SError if available and further determine cause of
1533 * failure.
1534 *
1535 * LOCKING:
1536 * None.
1537 */
1538 static void ata_eh_analyze_serror(struct ata_link *link)
1539 {
1540 struct ata_eh_context *ehc = &link->eh_context;
1541 u32 serror = ehc->i.serror;
1542 unsigned int err_mask = 0, action = 0;
1543 u32 hotplug_mask;
1544
1545 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1546 err_mask |= AC_ERR_ATA_BUS;
1547 action |= ATA_EH_RESET;
1548 }
1549 if (serror & SERR_PROTOCOL) {
1550 err_mask |= AC_ERR_HSM;
1551 action |= ATA_EH_RESET;
1552 }
1553 if (serror & SERR_INTERNAL) {
1554 err_mask |= AC_ERR_SYSTEM;
1555 action |= ATA_EH_RESET;
1556 }
1557
1558 /* Determine whether a hotplug event has occurred. Both
1559 * SError.N/X are considered hotplug events for enabled or
1560 * host links. For disabled PMP links, only N bit is
1561 * considered as X bit is left at 1 for link plugging.
1562 */
1563 hotplug_mask = 0;
1564
1565 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1566 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1567 else
1568 hotplug_mask = SERR_PHYRDY_CHG;
1569
1570 if (serror & hotplug_mask)
1571 ata_ehi_hotplugged(&ehc->i);
1572
1573 ehc->i.err_mask |= err_mask;
1574 ehc->i.action |= action;
1575 }
1576
1577 /**
1578 * ata_eh_analyze_ncq_error - analyze NCQ error
1579 * @link: ATA link to analyze NCQ error for
1580 *
1581 * Read log page 10h, determine the offending qc and acquire
1582 * error status TF. For NCQ device errors, all LLDDs have to do
1583 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1584 * care of the rest.
1585 *
1586 * LOCKING:
1587 * Kernel thread context (may sleep).
1588 */
1589 void ata_eh_analyze_ncq_error(struct ata_link *link)
1590 {
1591 struct ata_port *ap = link->ap;
1592 struct ata_eh_context *ehc = &link->eh_context;
1593 struct ata_device *dev = link->device;
1594 struct ata_queued_cmd *qc;
1595 struct ata_taskfile tf;
1596 int tag, rc;
1597
1598 /* if frozen, we can't do much */
1599 if (ap->pflags & ATA_PFLAG_FROZEN)
1600 return;
1601
1602 /* is it NCQ device error? */
1603 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1604 return;
1605
1606 /* has LLDD analyzed already? */
1607 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1608 qc = __ata_qc_from_tag(ap, tag);
1609
1610 if (!(qc->flags & ATA_QCFLAG_FAILED))
1611 continue;
1612
1613 if (qc->err_mask)
1614 return;
1615 }
1616
1617 /* okay, this error is ours */
1618 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1619 if (rc) {
1620 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1621 "(errno=%d)\n", rc);
1622 return;
1623 }
1624
1625 if (!(link->sactive & (1 << tag))) {
1626 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1627 "inactive tag %d\n", tag);
1628 return;
1629 }
1630
1631 /* we've got the perpetrator, condemn it */
1632 qc = __ata_qc_from_tag(ap, tag);
1633 memcpy(&qc->result_tf, &tf, sizeof(tf));
1634 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1635 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1636 ehc->i.err_mask &= ~AC_ERR_DEV;
1637 }
1638
1639 /**
1640 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1641 * @qc: qc to analyze
1642 * @tf: Taskfile registers to analyze
1643 *
1644 * Analyze taskfile of @qc and further determine cause of
1645 * failure. This function also requests ATAPI sense data if
1646 * avaliable.
1647 *
1648 * LOCKING:
1649 * Kernel thread context (may sleep).
1650 *
1651 * RETURNS:
1652 * Determined recovery action
1653 */
1654 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1655 const struct ata_taskfile *tf)
1656 {
1657 unsigned int tmp, action = 0;
1658 u8 stat = tf->command, err = tf->feature;
1659
1660 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1661 qc->err_mask |= AC_ERR_HSM;
1662 return ATA_EH_RESET;
1663 }
1664
1665 if (stat & (ATA_ERR | ATA_DF))
1666 qc->err_mask |= AC_ERR_DEV;
1667 else
1668 return 0;
1669
1670 switch (qc->dev->class) {
1671 case ATA_DEV_ATA:
1672 if (err & ATA_ICRC)
1673 qc->err_mask |= AC_ERR_ATA_BUS;
1674 if (err & ATA_UNC)
1675 qc->err_mask |= AC_ERR_MEDIA;
1676 if (err & ATA_IDNF)
1677 qc->err_mask |= AC_ERR_INVALID;
1678 break;
1679
1680 case ATA_DEV_ATAPI:
1681 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1682 tmp = atapi_eh_request_sense(qc->dev,
1683 qc->scsicmd->sense_buffer,
1684 qc->result_tf.feature >> 4);
1685 if (!tmp) {
1686 /* ATA_QCFLAG_SENSE_VALID is used to
1687 * tell atapi_qc_complete() that sense
1688 * data is already valid.
1689 *
1690 * TODO: interpret sense data and set
1691 * appropriate err_mask.
1692 */
1693 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1694 } else
1695 qc->err_mask |= tmp;
1696 }
1697 }
1698
1699 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1700 action |= ATA_EH_RESET;
1701
1702 return action;
1703 }
1704
1705 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1706 int *xfer_ok)
1707 {
1708 int base = 0;
1709
1710 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1711 *xfer_ok = 1;
1712
1713 if (!*xfer_ok)
1714 base = ATA_ECAT_DUBIOUS_NONE;
1715
1716 if (err_mask & AC_ERR_ATA_BUS)
1717 return base + ATA_ECAT_ATA_BUS;
1718
1719 if (err_mask & AC_ERR_TIMEOUT)
1720 return base + ATA_ECAT_TOUT_HSM;
1721
1722 if (eflags & ATA_EFLAG_IS_IO) {
1723 if (err_mask & AC_ERR_HSM)
1724 return base + ATA_ECAT_TOUT_HSM;
1725 if ((err_mask &
1726 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1727 return base + ATA_ECAT_UNK_DEV;
1728 }
1729
1730 return 0;
1731 }
1732
1733 struct speed_down_verdict_arg {
1734 u64 since;
1735 int xfer_ok;
1736 int nr_errors[ATA_ECAT_NR];
1737 };
1738
1739 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1740 {
1741 struct speed_down_verdict_arg *arg = void_arg;
1742 int cat;
1743
1744 if (ent->timestamp < arg->since)
1745 return -1;
1746
1747 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1748 &arg->xfer_ok);
1749 arg->nr_errors[cat]++;
1750
1751 return 0;
1752 }
1753
1754 /**
1755 * ata_eh_speed_down_verdict - Determine speed down verdict
1756 * @dev: Device of interest
1757 *
1758 * This function examines error ring of @dev and determines
1759 * whether NCQ needs to be turned off, transfer speed should be
1760 * stepped down, or falling back to PIO is necessary.
1761 *
1762 * ECAT_ATA_BUS : ATA_BUS error for any command
1763 *
1764 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1765 * IO commands
1766 *
1767 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1768 *
1769 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1770 * data transfer hasn't been verified.
1771 *
1772 * Verdicts are
1773 *
1774 * NCQ_OFF : Turn off NCQ.
1775 *
1776 * SPEED_DOWN : Speed down transfer speed but don't fall back
1777 * to PIO.
1778 *
1779 * FALLBACK_TO_PIO : Fall back to PIO.
1780 *
1781 * Even if multiple verdicts are returned, only one action is
1782 * taken per error. An action triggered by non-DUBIOUS errors
1783 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1784 * This is to expedite speed down decisions right after device is
1785 * initially configured.
1786 *
1787 * The followings are speed down rules. #1 and #2 deal with
1788 * DUBIOUS errors.
1789 *
1790 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1791 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1792 *
1793 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1794 * occurred during last 5 mins, NCQ_OFF.
1795 *
1796 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1797 * ocurred during last 5 mins, FALLBACK_TO_PIO
1798 *
1799 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1800 * during last 10 mins, NCQ_OFF.
1801 *
1802 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1803 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1804 *
1805 * LOCKING:
1806 * Inherited from caller.
1807 *
1808 * RETURNS:
1809 * OR of ATA_EH_SPDN_* flags.
1810 */
1811 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1812 {
1813 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1814 u64 j64 = get_jiffies_64();
1815 struct speed_down_verdict_arg arg;
1816 unsigned int verdict = 0;
1817
1818 /* scan past 5 mins of error history */
1819 memset(&arg, 0, sizeof(arg));
1820 arg.since = j64 - min(j64, j5mins);
1821 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1822
1823 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1824 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1825 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1826 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1827
1828 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1829 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1830 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1831
1832 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1833 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1834 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1835 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1836
1837 /* scan past 10 mins of error history */
1838 memset(&arg, 0, sizeof(arg));
1839 arg.since = j64 - min(j64, j10mins);
1840 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1841
1842 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1843 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1844 verdict |= ATA_EH_SPDN_NCQ_OFF;
1845
1846 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1847 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1848 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1849 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1850
1851 return verdict;
1852 }
1853
1854 /**
1855 * ata_eh_speed_down - record error and speed down if necessary
1856 * @dev: Failed device
1857 * @eflags: mask of ATA_EFLAG_* flags
1858 * @err_mask: err_mask of the error
1859 *
1860 * Record error and examine error history to determine whether
1861 * adjusting transmission speed is necessary. It also sets
1862 * transmission limits appropriately if such adjustment is
1863 * necessary.
1864 *
1865 * LOCKING:
1866 * Kernel thread context (may sleep).
1867 *
1868 * RETURNS:
1869 * Determined recovery action.
1870 */
1871 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1872 unsigned int eflags, unsigned int err_mask)
1873 {
1874 struct ata_link *link = ata_dev_phys_link(dev);
1875 int xfer_ok = 0;
1876 unsigned int verdict;
1877 unsigned int action = 0;
1878
1879 /* don't bother if Cat-0 error */
1880 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1881 return 0;
1882
1883 /* record error and determine whether speed down is necessary */
1884 ata_ering_record(&dev->ering, eflags, err_mask);
1885 verdict = ata_eh_speed_down_verdict(dev);
1886
1887 /* turn off NCQ? */
1888 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1889 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1890 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1891 dev->flags |= ATA_DFLAG_NCQ_OFF;
1892 ata_dev_printk(dev, KERN_WARNING,
1893 "NCQ disabled due to excessive errors\n");
1894 goto done;
1895 }
1896
1897 /* speed down? */
1898 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1899 /* speed down SATA link speed if possible */
1900 if (sata_down_spd_limit(link, 0) == 0) {
1901 action |= ATA_EH_RESET;
1902 goto done;
1903 }
1904
1905 /* lower transfer mode */
1906 if (dev->spdn_cnt < 2) {
1907 static const int dma_dnxfer_sel[] =
1908 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1909 static const int pio_dnxfer_sel[] =
1910 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1911 int sel;
1912
1913 if (dev->xfer_shift != ATA_SHIFT_PIO)
1914 sel = dma_dnxfer_sel[dev->spdn_cnt];
1915 else
1916 sel = pio_dnxfer_sel[dev->spdn_cnt];
1917
1918 dev->spdn_cnt++;
1919
1920 if (ata_down_xfermask_limit(dev, sel) == 0) {
1921 action |= ATA_EH_RESET;
1922 goto done;
1923 }
1924 }
1925 }
1926
1927 /* Fall back to PIO? Slowing down to PIO is meaningless for
1928 * SATA ATA devices. Consider it only for PATA and SATAPI.
1929 */
1930 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1931 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1932 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1933 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1934 dev->spdn_cnt = 0;
1935 action |= ATA_EH_RESET;
1936 goto done;
1937 }
1938 }
1939
1940 return 0;
1941 done:
1942 /* device has been slowed down, blow error history */
1943 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1944 ata_ering_clear(&dev->ering);
1945 return action;
1946 }
1947
1948 /**
1949 * ata_eh_link_autopsy - analyze error and determine recovery action
1950 * @link: host link to perform autopsy on
1951 *
1952 * Analyze why @link failed and determine which recovery actions
1953 * are needed. This function also sets more detailed AC_ERR_*
1954 * values and fills sense data for ATAPI CHECK SENSE.
1955 *
1956 * LOCKING:
1957 * Kernel thread context (may sleep).
1958 */
1959 static void ata_eh_link_autopsy(struct ata_link *link)
1960 {
1961 struct ata_port *ap = link->ap;
1962 struct ata_eh_context *ehc = &link->eh_context;
1963 struct ata_device *dev;
1964 unsigned int all_err_mask = 0, eflags = 0;
1965 int tag;
1966 u32 serror;
1967 int rc;
1968
1969 DPRINTK("ENTER\n");
1970
1971 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1972 return;
1973
1974 /* obtain and analyze SError */
1975 rc = sata_scr_read(link, SCR_ERROR, &serror);
1976 if (rc == 0) {
1977 ehc->i.serror |= serror;
1978 ata_eh_analyze_serror(link);
1979 } else if (rc != -EOPNOTSUPP) {
1980 /* SError read failed, force reset and probing */
1981 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1982 ehc->i.action |= ATA_EH_RESET;
1983 ehc->i.err_mask |= AC_ERR_OTHER;
1984 }
1985
1986 /* analyze NCQ failure */
1987 ata_eh_analyze_ncq_error(link);
1988
1989 /* any real error trumps AC_ERR_OTHER */
1990 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1991 ehc->i.err_mask &= ~AC_ERR_OTHER;
1992
1993 all_err_mask |= ehc->i.err_mask;
1994
1995 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1996 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1997
1998 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1999 ata_dev_phys_link(qc->dev) != link)
2000 continue;
2001
2002 /* inherit upper level err_mask */
2003 qc->err_mask |= ehc->i.err_mask;
2004
2005 /* analyze TF */
2006 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2007
2008 /* DEV errors are probably spurious in case of ATA_BUS error */
2009 if (qc->err_mask & AC_ERR_ATA_BUS)
2010 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2011 AC_ERR_INVALID);
2012
2013 /* any real error trumps unknown error */
2014 if (qc->err_mask & ~AC_ERR_OTHER)
2015 qc->err_mask &= ~AC_ERR_OTHER;
2016
2017 /* SENSE_VALID trumps dev/unknown error and revalidation */
2018 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2019 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2020
2021 /* determine whether the command is worth retrying */
2022 if (!(qc->err_mask & AC_ERR_INVALID) &&
2023 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
2024 qc->flags |= ATA_QCFLAG_RETRY;
2025
2026 /* accumulate error info */
2027 ehc->i.dev = qc->dev;
2028 all_err_mask |= qc->err_mask;
2029 if (qc->flags & ATA_QCFLAG_IO)
2030 eflags |= ATA_EFLAG_IS_IO;
2031 }
2032
2033 /* enforce default EH actions */
2034 if (ap->pflags & ATA_PFLAG_FROZEN ||
2035 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2036 ehc->i.action |= ATA_EH_RESET;
2037 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2038 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2039 ehc->i.action |= ATA_EH_REVALIDATE;
2040
2041 /* If we have offending qcs and the associated failed device,
2042 * perform per-dev EH action only on the offending device.
2043 */
2044 if (ehc->i.dev) {
2045 ehc->i.dev_action[ehc->i.dev->devno] |=
2046 ehc->i.action & ATA_EH_PERDEV_MASK;
2047 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2048 }
2049
2050 /* propagate timeout to host link */
2051 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2052 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2053
2054 /* record error and consider speeding down */
2055 dev = ehc->i.dev;
2056 if (!dev && ((ata_link_max_devices(link) == 1 &&
2057 ata_dev_enabled(link->device))))
2058 dev = link->device;
2059
2060 if (dev) {
2061 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2062 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2063 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2064 }
2065
2066 DPRINTK("EXIT\n");
2067 }
2068
2069 /**
2070 * ata_eh_autopsy - analyze error and determine recovery action
2071 * @ap: host port to perform autopsy on
2072 *
2073 * Analyze all links of @ap and determine why they failed and
2074 * which recovery actions are needed.
2075 *
2076 * LOCKING:
2077 * Kernel thread context (may sleep).
2078 */
2079 void ata_eh_autopsy(struct ata_port *ap)
2080 {
2081 struct ata_link *link;
2082
2083 ata_for_each_link(link, ap, EDGE)
2084 ata_eh_link_autopsy(link);
2085
2086 /* Handle the frigging slave link. Autopsy is done similarly
2087 * but actions and flags are transferred over to the master
2088 * link and handled from there.
2089 */
2090 if (ap->slave_link) {
2091 struct ata_eh_context *mehc = &ap->link.eh_context;
2092 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2093
2094 /* transfer control flags from master to slave */
2095 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2096
2097 /* perform autopsy on the slave link */
2098 ata_eh_link_autopsy(ap->slave_link);
2099
2100 /* transfer actions from slave to master and clear slave */
2101 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2102 mehc->i.action |= sehc->i.action;
2103 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2104 mehc->i.flags |= sehc->i.flags;
2105 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2106 }
2107
2108 /* Autopsy of fanout ports can affect host link autopsy.
2109 * Perform host link autopsy last.
2110 */
2111 if (sata_pmp_attached(ap))
2112 ata_eh_link_autopsy(&ap->link);
2113 }
2114
2115 /**
2116 * ata_get_cmd_descript - get description for ATA command
2117 * @command: ATA command code to get description for
2118 *
2119 * Return a textual description of the given command, or NULL if the
2120 * command is not known.
2121 *
2122 * LOCKING:
2123 * None
2124 */
2125 const char *ata_get_cmd_descript(u8 command)
2126 {
2127 #ifdef CONFIG_ATA_VERBOSE_ERROR
2128 static const struct
2129 {
2130 u8 command;
2131 const char *text;
2132 } cmd_descr[] = {
2133 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2134 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2135 { ATA_CMD_STANDBY, "STANDBY" },
2136 { ATA_CMD_IDLE, "IDLE" },
2137 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2138 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2139 { ATA_CMD_NOP, "NOP" },
2140 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2141 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2142 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2143 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2144 { ATA_CMD_SERVICE, "SERVICE" },
2145 { ATA_CMD_READ, "READ DMA" },
2146 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2147 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2148 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2149 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2150 { ATA_CMD_WRITE, "WRITE DMA" },
2151 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2152 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2153 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2154 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2155 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2156 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2157 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2158 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2159 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2160 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2161 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2162 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2163 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2164 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2165 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2166 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2167 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2168 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2169 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2170 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2171 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2172 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2173 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2174 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2175 { ATA_CMD_SLEEP, "SLEEP" },
2176 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2177 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2178 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2179 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2180 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2181 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2182 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2183 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2184 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2185 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2186 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2187 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2188 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2189 { ATA_CMD_PMP_READ, "READ BUFFER" },
2190 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2191 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2192 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2193 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2194 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2195 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2196 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2197 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2198 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2199 { ATA_CMD_SMART, "SMART" },
2200 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2201 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2202 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2203 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2204 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2205 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2206 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2207 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2208 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2209 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2210 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2211 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2212 { ATA_CMD_RESTORE, "RECALIBRATE" },
2213 { 0, NULL } /* terminate list */
2214 };
2215
2216 unsigned int i;
2217 for (i = 0; cmd_descr[i].text; i++)
2218 if (cmd_descr[i].command == command)
2219 return cmd_descr[i].text;
2220 #endif
2221
2222 return NULL;
2223 }
2224
2225 /**
2226 * ata_eh_link_report - report error handling to user
2227 * @link: ATA link EH is going on
2228 *
2229 * Report EH to user.
2230 *
2231 * LOCKING:
2232 * None.
2233 */
2234 static void ata_eh_link_report(struct ata_link *link)
2235 {
2236 struct ata_port *ap = link->ap;
2237 struct ata_eh_context *ehc = &link->eh_context;
2238 const char *frozen, *desc;
2239 char tries_buf[6];
2240 int tag, nr_failed = 0;
2241
2242 if (ehc->i.flags & ATA_EHI_QUIET)
2243 return;
2244
2245 desc = NULL;
2246 if (ehc->i.desc[0] != '\0')
2247 desc = ehc->i.desc;
2248
2249 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2250 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2251
2252 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2253 ata_dev_phys_link(qc->dev) != link ||
2254 ((qc->flags & ATA_QCFLAG_QUIET) &&
2255 qc->err_mask == AC_ERR_DEV))
2256 continue;
2257 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2258 continue;
2259
2260 nr_failed++;
2261 }
2262
2263 if (!nr_failed && !ehc->i.err_mask)
2264 return;
2265
2266 frozen = "";
2267 if (ap->pflags & ATA_PFLAG_FROZEN)
2268 frozen = " frozen";
2269
2270 memset(tries_buf, 0, sizeof(tries_buf));
2271 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2272 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2273 ap->eh_tries);
2274
2275 if (ehc->i.dev) {
2276 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2277 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2278 ehc->i.err_mask, link->sactive, ehc->i.serror,
2279 ehc->i.action, frozen, tries_buf);
2280 if (desc)
2281 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2282 } else {
2283 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2284 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2285 ehc->i.err_mask, link->sactive, ehc->i.serror,
2286 ehc->i.action, frozen, tries_buf);
2287 if (desc)
2288 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2289 }
2290
2291 #ifdef CONFIG_ATA_VERBOSE_ERROR
2292 if (ehc->i.serror)
2293 ata_link_printk(link, KERN_ERR,
2294 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2295 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2296 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2297 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2298 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2299 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2300 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2301 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2302 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2303 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2304 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2305 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2306 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2307 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2308 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2309 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2310 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2311 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2312 #endif
2313
2314 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2315 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2316 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2317 const u8 *cdb = qc->cdb;
2318 char data_buf[20] = "";
2319 char cdb_buf[70] = "";
2320
2321 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2322 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2323 continue;
2324
2325 if (qc->dma_dir != DMA_NONE) {
2326 static const char *dma_str[] = {
2327 [DMA_BIDIRECTIONAL] = "bidi",
2328 [DMA_TO_DEVICE] = "out",
2329 [DMA_FROM_DEVICE] = "in",
2330 };
2331 static const char *prot_str[] = {
2332 [ATA_PROT_PIO] = "pio",
2333 [ATA_PROT_DMA] = "dma",
2334 [ATA_PROT_NCQ] = "ncq",
2335 [ATAPI_PROT_PIO] = "pio",
2336 [ATAPI_PROT_DMA] = "dma",
2337 };
2338
2339 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2340 prot_str[qc->tf.protocol], qc->nbytes,
2341 dma_str[qc->dma_dir]);
2342 }
2343
2344 if (ata_is_atapi(qc->tf.protocol)) {
2345 if (qc->scsicmd)
2346 scsi_print_command(qc->scsicmd);
2347 else
2348 snprintf(cdb_buf, sizeof(cdb_buf),
2349 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2350 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2351 cdb[0], cdb[1], cdb[2], cdb[3],
2352 cdb[4], cdb[5], cdb[6], cdb[7],
2353 cdb[8], cdb[9], cdb[10], cdb[11],
2354 cdb[12], cdb[13], cdb[14], cdb[15]);
2355 } else {
2356 const char *descr = ata_get_cmd_descript(cmd->command);
2357 if (descr)
2358 ata_dev_printk(qc->dev, KERN_ERR,
2359 "failed command: %s\n", descr);
2360 }
2361
2362 ata_dev_printk(qc->dev, KERN_ERR,
2363 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2364 "tag %d%s\n %s"
2365 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2366 "Emask 0x%x (%s)%s\n",
2367 cmd->command, cmd->feature, cmd->nsect,
2368 cmd->lbal, cmd->lbam, cmd->lbah,
2369 cmd->hob_feature, cmd->hob_nsect,
2370 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2371 cmd->device, qc->tag, data_buf, cdb_buf,
2372 res->command, res->feature, res->nsect,
2373 res->lbal, res->lbam, res->lbah,
2374 res->hob_feature, res->hob_nsect,
2375 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2376 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2377 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2378
2379 #ifdef CONFIG_ATA_VERBOSE_ERROR
2380 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2381 ATA_ERR)) {
2382 if (res->command & ATA_BUSY)
2383 ata_dev_printk(qc->dev, KERN_ERR,
2384 "status: { Busy }\n");
2385 else
2386 ata_dev_printk(qc->dev, KERN_ERR,
2387 "status: { %s%s%s%s}\n",
2388 res->command & ATA_DRDY ? "DRDY " : "",
2389 res->command & ATA_DF ? "DF " : "",
2390 res->command & ATA_DRQ ? "DRQ " : "",
2391 res->command & ATA_ERR ? "ERR " : "");
2392 }
2393
2394 if (cmd->command != ATA_CMD_PACKET &&
2395 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2396 ATA_ABORTED)))
2397 ata_dev_printk(qc->dev, KERN_ERR,
2398 "error: { %s%s%s%s}\n",
2399 res->feature & ATA_ICRC ? "ICRC " : "",
2400 res->feature & ATA_UNC ? "UNC " : "",
2401 res->feature & ATA_IDNF ? "IDNF " : "",
2402 res->feature & ATA_ABORTED ? "ABRT " : "");
2403 #endif
2404 }
2405 }
2406
2407 /**
2408 * ata_eh_report - report error handling to user
2409 * @ap: ATA port to report EH about
2410 *
2411 * Report EH to user.
2412 *
2413 * LOCKING:
2414 * None.
2415 */
2416 void ata_eh_report(struct ata_port *ap)
2417 {
2418 struct ata_link *link;
2419
2420 ata_for_each_link(link, ap, HOST_FIRST)
2421 ata_eh_link_report(link);
2422 }
2423
2424 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2425 unsigned int *classes, unsigned long deadline,
2426 bool clear_classes)
2427 {
2428 struct ata_device *dev;
2429
2430 if (clear_classes)
2431 ata_for_each_dev(dev, link, ALL)
2432 classes[dev->devno] = ATA_DEV_UNKNOWN;
2433
2434 return reset(link, classes, deadline);
2435 }
2436
2437 static int ata_eh_followup_srst_needed(struct ata_link *link,
2438 int rc, const unsigned int *classes)
2439 {
2440 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2441 return 0;
2442 if (rc == -EAGAIN)
2443 return 1;
2444 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2445 return 1;
2446 return 0;
2447 }
2448
2449 int ata_eh_reset(struct ata_link *link, int classify,
2450 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2451 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2452 {
2453 struct ata_port *ap = link->ap;
2454 struct ata_link *slave = ap->slave_link;
2455 struct ata_eh_context *ehc = &link->eh_context;
2456 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2457 unsigned int *classes = ehc->classes;
2458 unsigned int lflags = link->flags;
2459 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2460 int max_tries = 0, try = 0;
2461 struct ata_link *failed_link;
2462 struct ata_device *dev;
2463 unsigned long deadline, now;
2464 ata_reset_fn_t reset;
2465 unsigned long flags;
2466 u32 sstatus;
2467 int nr_unknown, rc;
2468
2469 /*
2470 * Prepare to reset
2471 */
2472 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2473 max_tries++;
2474 if (link->flags & ATA_LFLAG_NO_HRST)
2475 hardreset = NULL;
2476 if (link->flags & ATA_LFLAG_NO_SRST)
2477 softreset = NULL;
2478
2479 /* make sure each reset attemp is at least COOL_DOWN apart */
2480 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2481 now = jiffies;
2482 WARN_ON(time_after(ehc->last_reset, now));
2483 deadline = ata_deadline(ehc->last_reset,
2484 ATA_EH_RESET_COOL_DOWN);
2485 if (time_before(now, deadline))
2486 schedule_timeout_uninterruptible(deadline - now);
2487 }
2488
2489 spin_lock_irqsave(ap->lock, flags);
2490 ap->pflags |= ATA_PFLAG_RESETTING;
2491 spin_unlock_irqrestore(ap->lock, flags);
2492
2493 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2494
2495 ata_for_each_dev(dev, link, ALL) {
2496 /* If we issue an SRST then an ATA drive (not ATAPI)
2497 * may change configuration and be in PIO0 timing. If
2498 * we do a hard reset (or are coming from power on)
2499 * this is true for ATA or ATAPI. Until we've set a
2500 * suitable controller mode we should not touch the
2501 * bus as we may be talking too fast.
2502 */
2503 dev->pio_mode = XFER_PIO_0;
2504
2505 /* If the controller has a pio mode setup function
2506 * then use it to set the chipset to rights. Don't
2507 * touch the DMA setup as that will be dealt with when
2508 * configuring devices.
2509 */
2510 if (ap->ops->set_piomode)
2511 ap->ops->set_piomode(ap, dev);
2512 }
2513
2514 /* prefer hardreset */
2515 reset = NULL;
2516 ehc->i.action &= ~ATA_EH_RESET;
2517 if (hardreset) {
2518 reset = hardreset;
2519 ehc->i.action |= ATA_EH_HARDRESET;
2520 } else if (softreset) {
2521 reset = softreset;
2522 ehc->i.action |= ATA_EH_SOFTRESET;
2523 }
2524
2525 if (prereset) {
2526 unsigned long deadline = ata_deadline(jiffies,
2527 ATA_EH_PRERESET_TIMEOUT);
2528
2529 if (slave) {
2530 sehc->i.action &= ~ATA_EH_RESET;
2531 sehc->i.action |= ehc->i.action;
2532 }
2533
2534 rc = prereset(link, deadline);
2535
2536 /* If present, do prereset on slave link too. Reset
2537 * is skipped iff both master and slave links report
2538 * -ENOENT or clear ATA_EH_RESET.
2539 */
2540 if (slave && (rc == 0 || rc == -ENOENT)) {
2541 int tmp;
2542
2543 tmp = prereset(slave, deadline);
2544 if (tmp != -ENOENT)
2545 rc = tmp;
2546
2547 ehc->i.action |= sehc->i.action;
2548 }
2549
2550 if (rc) {
2551 if (rc == -ENOENT) {
2552 ata_link_printk(link, KERN_DEBUG,
2553 "port disabled. ignoring.\n");
2554 ehc->i.action &= ~ATA_EH_RESET;
2555
2556 ata_for_each_dev(dev, link, ALL)
2557 classes[dev->devno] = ATA_DEV_NONE;
2558
2559 rc = 0;
2560 } else
2561 ata_link_printk(link, KERN_ERR,
2562 "prereset failed (errno=%d)\n", rc);
2563 goto out;
2564 }
2565
2566 /* prereset() might have cleared ATA_EH_RESET. If so,
2567 * bang classes, thaw and return.
2568 */
2569 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2570 ata_for_each_dev(dev, link, ALL)
2571 classes[dev->devno] = ATA_DEV_NONE;
2572 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2573 ata_is_host_link(link))
2574 ata_eh_thaw_port(ap);
2575 rc = 0;
2576 goto out;
2577 }
2578 }
2579
2580 retry:
2581 /*
2582 * Perform reset
2583 */
2584 if (ata_is_host_link(link))
2585 ata_eh_freeze_port(ap);
2586
2587 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2588
2589 if (reset) {
2590 if (verbose)
2591 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2592 reset == softreset ? "soft" : "hard");
2593
2594 /* mark that this EH session started with reset */
2595 ehc->last_reset = jiffies;
2596 if (reset == hardreset)
2597 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2598 else
2599 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2600
2601 rc = ata_do_reset(link, reset, classes, deadline, true);
2602 if (rc && rc != -EAGAIN) {
2603 failed_link = link;
2604 goto fail;
2605 }
2606
2607 /* hardreset slave link if existent */
2608 if (slave && reset == hardreset) {
2609 int tmp;
2610
2611 if (verbose)
2612 ata_link_printk(slave, KERN_INFO,
2613 "hard resetting link\n");
2614
2615 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2616 tmp = ata_do_reset(slave, reset, classes, deadline,
2617 false);
2618 switch (tmp) {
2619 case -EAGAIN:
2620 rc = -EAGAIN;
2621 case 0:
2622 break;
2623 default:
2624 failed_link = slave;
2625 rc = tmp;
2626 goto fail;
2627 }
2628 }
2629
2630 /* perform follow-up SRST if necessary */
2631 if (reset == hardreset &&
2632 ata_eh_followup_srst_needed(link, rc, classes)) {
2633 reset = softreset;
2634
2635 if (!reset) {
2636 ata_link_printk(link, KERN_ERR,
2637 "follow-up softreset required "
2638 "but no softreset avaliable\n");
2639 failed_link = link;
2640 rc = -EINVAL;
2641 goto fail;
2642 }
2643
2644 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2645 rc = ata_do_reset(link, reset, classes, deadline, true);
2646 if (rc) {
2647 failed_link = link;
2648 goto fail;
2649 }
2650 }
2651 } else {
2652 if (verbose)
2653 ata_link_printk(link, KERN_INFO, "no reset method "
2654 "available, skipping reset\n");
2655 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2656 lflags |= ATA_LFLAG_ASSUME_ATA;
2657 }
2658
2659 /*
2660 * Post-reset processing
2661 */
2662 ata_for_each_dev(dev, link, ALL) {
2663 /* After the reset, the device state is PIO 0 and the
2664 * controller state is undefined. Reset also wakes up
2665 * drives from sleeping mode.
2666 */
2667 dev->pio_mode = XFER_PIO_0;
2668 dev->flags &= ~ATA_DFLAG_SLEEPING;
2669
2670 if (!ata_phys_link_offline(ata_dev_phys_link(dev))) {
2671 /* apply class override */
2672 if (lflags & ATA_LFLAG_ASSUME_ATA)
2673 classes[dev->devno] = ATA_DEV_ATA;
2674 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2675 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2676 } else
2677 classes[dev->devno] = ATA_DEV_NONE;
2678 }
2679
2680 /* record current link speed */
2681 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2682 link->sata_spd = (sstatus >> 4) & 0xf;
2683 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2684 slave->sata_spd = (sstatus >> 4) & 0xf;
2685
2686 /* thaw the port */
2687 if (ata_is_host_link(link))
2688 ata_eh_thaw_port(ap);
2689
2690 /* postreset() should clear hardware SError. Although SError
2691 * is cleared during link resume, clearing SError here is
2692 * necessary as some PHYs raise hotplug events after SRST.
2693 * This introduces race condition where hotplug occurs between
2694 * reset and here. This race is mediated by cross checking
2695 * link onlineness and classification result later.
2696 */
2697 if (postreset) {
2698 postreset(link, classes);
2699 if (slave)
2700 postreset(slave, classes);
2701 }
2702
2703 /*
2704 * Some controllers can't be frozen very well and may set
2705 * spuruious error conditions during reset. Clear accumulated
2706 * error information. As reset is the final recovery action,
2707 * nothing is lost by doing this.
2708 */
2709 spin_lock_irqsave(link->ap->lock, flags);
2710 memset(&link->eh_info, 0, sizeof(link->eh_info));
2711 if (slave)
2712 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2713 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2714 spin_unlock_irqrestore(link->ap->lock, flags);
2715
2716 /* Make sure onlineness and classification result correspond.
2717 * Hotplug could have happened during reset and some
2718 * controllers fail to wait while a drive is spinning up after
2719 * being hotplugged causing misdetection. By cross checking
2720 * link onlineness and classification result, those conditions
2721 * can be reliably detected and retried.
2722 */
2723 nr_unknown = 0;
2724 ata_for_each_dev(dev, link, ALL) {
2725 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2726 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2727 classes[dev->devno] = ATA_DEV_NONE;
2728 if (ata_phys_link_online(ata_dev_phys_link(dev)))
2729 nr_unknown++;
2730 }
2731 }
2732
2733 if (classify && nr_unknown) {
2734 if (try < max_tries) {
2735 ata_link_printk(link, KERN_WARNING, "link online but "
2736 "device misclassified, retrying\n");
2737 failed_link = link;
2738 rc = -EAGAIN;
2739 goto fail;
2740 }
2741 ata_link_printk(link, KERN_WARNING,
2742 "link online but device misclassified, "
2743 "device detection might fail\n");
2744 }
2745
2746 /* reset successful, schedule revalidation */
2747 ata_eh_done(link, NULL, ATA_EH_RESET);
2748 if (slave)
2749 ata_eh_done(slave, NULL, ATA_EH_RESET);
2750 ehc->last_reset = jiffies; /* update to completion time */
2751 ehc->i.action |= ATA_EH_REVALIDATE;
2752
2753 rc = 0;
2754 out:
2755 /* clear hotplug flag */
2756 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2757 if (slave)
2758 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2759
2760 spin_lock_irqsave(ap->lock, flags);
2761 ap->pflags &= ~ATA_PFLAG_RESETTING;
2762 spin_unlock_irqrestore(ap->lock, flags);
2763
2764 return rc;
2765
2766 fail:
2767 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2768 if (!ata_is_host_link(link) &&
2769 sata_scr_read(link, SCR_STATUS, &sstatus))
2770 rc = -ERESTART;
2771
2772 if (rc == -ERESTART || try >= max_tries)
2773 goto out;
2774
2775 now = jiffies;
2776 if (time_before(now, deadline)) {
2777 unsigned long delta = deadline - now;
2778
2779 ata_link_printk(failed_link, KERN_WARNING,
2780 "reset failed (errno=%d), retrying in %u secs\n",
2781 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2782
2783 while (delta)
2784 delta = schedule_timeout_uninterruptible(delta);
2785 }
2786
2787 if (try == max_tries - 1) {
2788 sata_down_spd_limit(link, 0);
2789 if (slave)
2790 sata_down_spd_limit(slave, 0);
2791 } else if (rc == -EPIPE)
2792 sata_down_spd_limit(failed_link, 0);
2793
2794 if (hardreset)
2795 reset = hardreset;
2796 goto retry;
2797 }
2798
2799 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2800 {
2801 struct ata_link *link;
2802 struct ata_device *dev;
2803 unsigned long flags;
2804
2805 /*
2806 * This function can be thought of as an extended version of
2807 * ata_eh_about_to_do() specially crafted to accommodate the
2808 * requirements of ATA_EH_PARK handling. Since the EH thread
2809 * does not leave the do {} while () loop in ata_eh_recover as
2810 * long as the timeout for a park request to *one* device on
2811 * the port has not expired, and since we still want to pick
2812 * up park requests to other devices on the same port or
2813 * timeout updates for the same device, we have to pull
2814 * ATA_EH_PARK actions from eh_info into eh_context.i
2815 * ourselves at the beginning of each pass over the loop.
2816 *
2817 * Additionally, all write accesses to &ap->park_req_pending
2818 * through INIT_COMPLETION() (see below) or complete_all()
2819 * (see ata_scsi_park_store()) are protected by the host lock.
2820 * As a result we have that park_req_pending.done is zero on
2821 * exit from this function, i.e. when ATA_EH_PARK actions for
2822 * *all* devices on port ap have been pulled into the
2823 * respective eh_context structs. If, and only if,
2824 * park_req_pending.done is non-zero by the time we reach
2825 * wait_for_completion_timeout(), another ATA_EH_PARK action
2826 * has been scheduled for at least one of the devices on port
2827 * ap and we have to cycle over the do {} while () loop in
2828 * ata_eh_recover() again.
2829 */
2830
2831 spin_lock_irqsave(ap->lock, flags);
2832 INIT_COMPLETION(ap->park_req_pending);
2833 ata_for_each_link(link, ap, EDGE) {
2834 ata_for_each_dev(dev, link, ALL) {
2835 struct ata_eh_info *ehi = &link->eh_info;
2836
2837 link->eh_context.i.dev_action[dev->devno] |=
2838 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2839 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2840 }
2841 }
2842 spin_unlock_irqrestore(ap->lock, flags);
2843 }
2844
2845 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2846 {
2847 struct ata_eh_context *ehc = &dev->link->eh_context;
2848 struct ata_taskfile tf;
2849 unsigned int err_mask;
2850
2851 ata_tf_init(dev, &tf);
2852 if (park) {
2853 ehc->unloaded_mask |= 1 << dev->devno;
2854 tf.command = ATA_CMD_IDLEIMMEDIATE;
2855 tf.feature = 0x44;
2856 tf.lbal = 0x4c;
2857 tf.lbam = 0x4e;
2858 tf.lbah = 0x55;
2859 } else {
2860 ehc->unloaded_mask &= ~(1 << dev->devno);
2861 tf.command = ATA_CMD_CHK_POWER;
2862 }
2863
2864 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2865 tf.protocol |= ATA_PROT_NODATA;
2866 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2867 if (park && (err_mask || tf.lbal != 0xc4)) {
2868 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2869 ehc->unloaded_mask &= ~(1 << dev->devno);
2870 }
2871 }
2872
2873 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2874 struct ata_device **r_failed_dev)
2875 {
2876 struct ata_port *ap = link->ap;
2877 struct ata_eh_context *ehc = &link->eh_context;
2878 struct ata_device *dev;
2879 unsigned int new_mask = 0;
2880 unsigned long flags;
2881 int rc = 0;
2882
2883 DPRINTK("ENTER\n");
2884
2885 /* For PATA drive side cable detection to work, IDENTIFY must
2886 * be done backwards such that PDIAG- is released by the slave
2887 * device before the master device is identified.
2888 */
2889 ata_for_each_dev(dev, link, ALL_REVERSE) {
2890 unsigned int action = ata_eh_dev_action(dev);
2891 unsigned int readid_flags = 0;
2892
2893 if (ehc->i.flags & ATA_EHI_DID_RESET)
2894 readid_flags |= ATA_READID_POSTRESET;
2895
2896 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2897 WARN_ON(dev->class == ATA_DEV_PMP);
2898
2899 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2900 rc = -EIO;
2901 goto err;
2902 }
2903
2904 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2905 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2906 readid_flags);
2907 if (rc)
2908 goto err;
2909
2910 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2911
2912 /* Configuration may have changed, reconfigure
2913 * transfer mode.
2914 */
2915 ehc->i.flags |= ATA_EHI_SETMODE;
2916
2917 /* schedule the scsi_rescan_device() here */
2918 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2919 } else if (dev->class == ATA_DEV_UNKNOWN &&
2920 ehc->tries[dev->devno] &&
2921 ata_class_enabled(ehc->classes[dev->devno])) {
2922 /* Temporarily set dev->class, it will be
2923 * permanently set once all configurations are
2924 * complete. This is necessary because new
2925 * device configuration is done in two
2926 * separate loops.
2927 */
2928 dev->class = ehc->classes[dev->devno];
2929
2930 if (dev->class == ATA_DEV_PMP)
2931 rc = sata_pmp_attach(dev);
2932 else
2933 rc = ata_dev_read_id(dev, &dev->class,
2934 readid_flags, dev->id);
2935
2936 /* read_id might have changed class, store and reset */
2937 ehc->classes[dev->devno] = dev->class;
2938 dev->class = ATA_DEV_UNKNOWN;
2939
2940 switch (rc) {
2941 case 0:
2942 /* clear error info accumulated during probe */
2943 ata_ering_clear(&dev->ering);
2944 new_mask |= 1 << dev->devno;
2945 break;
2946 case -ENOENT:
2947 /* IDENTIFY was issued to non-existent
2948 * device. No need to reset. Just
2949 * thaw and ignore the device.
2950 */
2951 ata_eh_thaw_port(ap);
2952 break;
2953 default:
2954 goto err;
2955 }
2956 }
2957 }
2958
2959 /* PDIAG- should have been released, ask cable type if post-reset */
2960 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2961 if (ap->ops->cable_detect)
2962 ap->cbl = ap->ops->cable_detect(ap);
2963 ata_force_cbl(ap);
2964 }
2965
2966 /* Configure new devices forward such that user doesn't see
2967 * device detection messages backwards.
2968 */
2969 ata_for_each_dev(dev, link, ALL) {
2970 if (!(new_mask & (1 << dev->devno)) ||
2971 dev->class == ATA_DEV_PMP)
2972 continue;
2973
2974 dev->class = ehc->classes[dev->devno];
2975
2976 ehc->i.flags |= ATA_EHI_PRINTINFO;
2977 rc = ata_dev_configure(dev);
2978 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2979 if (rc) {
2980 dev->class = ATA_DEV_UNKNOWN;
2981 goto err;
2982 }
2983
2984 spin_lock_irqsave(ap->lock, flags);
2985 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2986 spin_unlock_irqrestore(ap->lock, flags);
2987
2988 /* new device discovered, configure xfermode */
2989 ehc->i.flags |= ATA_EHI_SETMODE;
2990 }
2991
2992 return 0;
2993
2994 err:
2995 *r_failed_dev = dev;
2996 DPRINTK("EXIT rc=%d\n", rc);
2997 return rc;
2998 }
2999
3000 /**
3001 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3002 * @link: link on which timings will be programmed
3003 * @r_failed_dev: out parameter for failed device
3004 *
3005 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3006 * ata_set_mode() fails, pointer to the failing device is
3007 * returned in @r_failed_dev.
3008 *
3009 * LOCKING:
3010 * PCI/etc. bus probe sem.
3011 *
3012 * RETURNS:
3013 * 0 on success, negative errno otherwise
3014 */
3015 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3016 {
3017 struct ata_port *ap = link->ap;
3018 struct ata_device *dev;
3019 int rc;
3020
3021 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3022 ata_for_each_dev(dev, link, ENABLED) {
3023 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3024 struct ata_ering_entry *ent;
3025
3026 ent = ata_ering_top(&dev->ering);
3027 if (ent)
3028 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3029 }
3030 }
3031
3032 /* has private set_mode? */
3033 if (ap->ops->set_mode)
3034 rc = ap->ops->set_mode(link, r_failed_dev);
3035 else
3036 rc = ata_do_set_mode(link, r_failed_dev);
3037
3038 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3039 ata_for_each_dev(dev, link, ENABLED) {
3040 struct ata_eh_context *ehc = &link->eh_context;
3041 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3042 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3043
3044 if (dev->xfer_mode != saved_xfer_mode ||
3045 ata_ncq_enabled(dev) != saved_ncq)
3046 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3047 }
3048
3049 return rc;
3050 }
3051
3052 /**
3053 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3054 * @dev: ATAPI device to clear UA for
3055 *
3056 * Resets and other operations can make an ATAPI device raise
3057 * UNIT ATTENTION which causes the next operation to fail. This
3058 * function clears UA.
3059 *
3060 * LOCKING:
3061 * EH context (may sleep).
3062 *
3063 * RETURNS:
3064 * 0 on success, -errno on failure.
3065 */
3066 static int atapi_eh_clear_ua(struct ata_device *dev)
3067 {
3068 int i;
3069
3070 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3071 u8 *sense_buffer = dev->link->ap->sector_buf;
3072 u8 sense_key = 0;
3073 unsigned int err_mask;
3074
3075 err_mask = atapi_eh_tur(dev, &sense_key);
3076 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3077 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3078 "failed (err_mask=0x%x)\n", err_mask);
3079 return -EIO;
3080 }
3081
3082 if (!err_mask || sense_key != UNIT_ATTENTION)
3083 return 0;
3084
3085 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3086 if (err_mask) {
3087 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3088 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3089 return -EIO;
3090 }
3091 }
3092
3093 ata_dev_printk(dev, KERN_WARNING,
3094 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3095
3096 return 0;
3097 }
3098
3099 static int ata_link_nr_enabled(struct ata_link *link)
3100 {
3101 struct ata_device *dev;
3102 int cnt = 0;
3103
3104 ata_for_each_dev(dev, link, ENABLED)
3105 cnt++;
3106 return cnt;
3107 }
3108
3109 static int ata_link_nr_vacant(struct ata_link *link)
3110 {
3111 struct ata_device *dev;
3112 int cnt = 0;
3113
3114 ata_for_each_dev(dev, link, ALL)
3115 if (dev->class == ATA_DEV_UNKNOWN)
3116 cnt++;
3117 return cnt;
3118 }
3119
3120 static int ata_eh_skip_recovery(struct ata_link *link)
3121 {
3122 struct ata_port *ap = link->ap;
3123 struct ata_eh_context *ehc = &link->eh_context;
3124 struct ata_device *dev;
3125
3126 /* skip disabled links */
3127 if (link->flags & ATA_LFLAG_DISABLED)
3128 return 1;
3129
3130 /* thaw frozen port and recover failed devices */
3131 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3132 return 0;
3133
3134 /* reset at least once if reset is requested */
3135 if ((ehc->i.action & ATA_EH_RESET) &&
3136 !(ehc->i.flags & ATA_EHI_DID_RESET))
3137 return 0;
3138
3139 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3140 ata_for_each_dev(dev, link, ALL) {
3141 if (dev->class == ATA_DEV_UNKNOWN &&
3142 ehc->classes[dev->devno] != ATA_DEV_NONE)
3143 return 0;
3144 }
3145
3146 return 1;
3147 }
3148
3149 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3150 {
3151 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3152 u64 now = get_jiffies_64();
3153 int *trials = void_arg;
3154
3155 if (ent->timestamp < now - min(now, interval))
3156 return -1;
3157
3158 (*trials)++;
3159 return 0;
3160 }
3161
3162 static int ata_eh_schedule_probe(struct ata_device *dev)
3163 {
3164 struct ata_eh_context *ehc = &dev->link->eh_context;
3165 struct ata_link *link = ata_dev_phys_link(dev);
3166 int trials = 0;
3167
3168 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3169 (ehc->did_probe_mask & (1 << dev->devno)))
3170 return 0;
3171
3172 ata_eh_detach_dev(dev);
3173 ata_dev_init(dev);
3174 ehc->did_probe_mask |= (1 << dev->devno);
3175 ehc->i.action |= ATA_EH_RESET;
3176 ehc->saved_xfer_mode[dev->devno] = 0;
3177 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3178
3179 /* Record and count probe trials on the ering. The specific
3180 * error mask used is irrelevant. Because a successful device
3181 * detection clears the ering, this count accumulates only if
3182 * there are consecutive failed probes.
3183 *
3184 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3185 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3186 * forced to 1.5Gbps.
3187 *
3188 * This is to work around cases where failed link speed
3189 * negotiation results in device misdetection leading to
3190 * infinite DEVXCHG or PHRDY CHG events.
3191 */
3192 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3193 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3194
3195 if (trials > ATA_EH_PROBE_TRIALS)
3196 sata_down_spd_limit(link, 1);
3197
3198 return 1;
3199 }
3200
3201 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3202 {
3203 struct ata_eh_context *ehc = &dev->link->eh_context;
3204
3205 /* -EAGAIN from EH routine indicates retry without prejudice.
3206 * The requester is responsible for ensuring forward progress.
3207 */
3208 if (err != -EAGAIN)
3209 ehc->tries[dev->devno]--;
3210
3211 switch (err) {
3212 case -ENODEV:
3213 /* device missing or wrong IDENTIFY data, schedule probing */
3214 ehc->i.probe_mask |= (1 << dev->devno);
3215 case -EINVAL:
3216 /* give it just one more chance */
3217 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3218 case -EIO:
3219 if (ehc->tries[dev->devno] == 1) {
3220 /* This is the last chance, better to slow
3221 * down than lose it.
3222 */
3223 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3224 if (dev->pio_mode > XFER_PIO_0)
3225 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3226 }
3227 }
3228
3229 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3230 /* disable device if it has used up all its chances */
3231 ata_dev_disable(dev);
3232
3233 /* detach if offline */
3234 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3235 ata_eh_detach_dev(dev);
3236
3237 /* schedule probe if necessary */
3238 if (ata_eh_schedule_probe(dev)) {
3239 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3240 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3241 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3242 }
3243
3244 return 1;
3245 } else {
3246 ehc->i.action |= ATA_EH_RESET;
3247 return 0;
3248 }
3249 }
3250
3251 /**
3252 * ata_eh_recover - recover host port after error
3253 * @ap: host port to recover
3254 * @prereset: prereset method (can be NULL)
3255 * @softreset: softreset method (can be NULL)
3256 * @hardreset: hardreset method (can be NULL)
3257 * @postreset: postreset method (can be NULL)
3258 * @r_failed_link: out parameter for failed link
3259 *
3260 * This is the alpha and omega, eum and yang, heart and soul of
3261 * libata exception handling. On entry, actions required to
3262 * recover each link and hotplug requests are recorded in the
3263 * link's eh_context. This function executes all the operations
3264 * with appropriate retrials and fallbacks to resurrect failed
3265 * devices, detach goners and greet newcomers.
3266 *
3267 * LOCKING:
3268 * Kernel thread context (may sleep).
3269 *
3270 * RETURNS:
3271 * 0 on success, -errno on failure.
3272 */
3273 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3274 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3275 ata_postreset_fn_t postreset,
3276 struct ata_link **r_failed_link)
3277 {
3278 struct ata_link *link;
3279 struct ata_device *dev;
3280 int nr_failed_devs;
3281 int rc;
3282 unsigned long flags, deadline;
3283
3284 DPRINTK("ENTER\n");
3285
3286 /* prep for recovery */
3287 ata_for_each_link(link, ap, EDGE) {
3288 struct ata_eh_context *ehc = &link->eh_context;
3289
3290 /* re-enable link? */
3291 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3292 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3293 spin_lock_irqsave(ap->lock, flags);
3294 link->flags &= ~ATA_LFLAG_DISABLED;
3295 spin_unlock_irqrestore(ap->lock, flags);
3296 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3297 }
3298
3299 ata_for_each_dev(dev, link, ALL) {
3300 if (link->flags & ATA_LFLAG_NO_RETRY)
3301 ehc->tries[dev->devno] = 1;
3302 else
3303 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3304
3305 /* collect port action mask recorded in dev actions */
3306 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3307 ~ATA_EH_PERDEV_MASK;
3308 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3309
3310 /* process hotplug request */
3311 if (dev->flags & ATA_DFLAG_DETACH)
3312 ata_eh_detach_dev(dev);
3313
3314 /* schedule probe if necessary */
3315 if (!ata_dev_enabled(dev))
3316 ata_eh_schedule_probe(dev);
3317 }
3318 }
3319
3320 retry:
3321 rc = 0;
3322 nr_failed_devs = 0;
3323
3324 /* if UNLOADING, finish immediately */
3325 if (ap->pflags & ATA_PFLAG_UNLOADING)
3326 goto out;
3327
3328 /* prep for EH */
3329 ata_for_each_link(link, ap, EDGE) {
3330 struct ata_eh_context *ehc = &link->eh_context;
3331
3332 /* skip EH if possible. */
3333 if (ata_eh_skip_recovery(link))
3334 ehc->i.action = 0;
3335
3336 ata_for_each_dev(dev, link, ALL)
3337 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3338 }
3339
3340 /* reset */
3341 ata_for_each_link(link, ap, EDGE) {
3342 struct ata_eh_context *ehc = &link->eh_context;
3343
3344 if (!(ehc->i.action & ATA_EH_RESET))
3345 continue;
3346
3347 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3348 prereset, softreset, hardreset, postreset);
3349 if (rc) {
3350 ata_link_printk(link, KERN_ERR,
3351 "reset failed, giving up\n");
3352 goto out;
3353 }
3354 }
3355
3356 do {
3357 unsigned long now;
3358
3359 /*
3360 * clears ATA_EH_PARK in eh_info and resets
3361 * ap->park_req_pending
3362 */
3363 ata_eh_pull_park_action(ap);
3364
3365 deadline = jiffies;
3366 ata_for_each_link(link, ap, EDGE) {
3367 ata_for_each_dev(dev, link, ALL) {
3368 struct ata_eh_context *ehc = &link->eh_context;
3369 unsigned long tmp;
3370
3371 if (dev->class != ATA_DEV_ATA)
3372 continue;
3373 if (!(ehc->i.dev_action[dev->devno] &
3374 ATA_EH_PARK))
3375 continue;
3376 tmp = dev->unpark_deadline;
3377 if (time_before(deadline, tmp))
3378 deadline = tmp;
3379 else if (time_before_eq(tmp, jiffies))
3380 continue;
3381 if (ehc->unloaded_mask & (1 << dev->devno))
3382 continue;
3383
3384 ata_eh_park_issue_cmd(dev, 1);
3385 }
3386 }
3387
3388 now = jiffies;
3389 if (time_before_eq(deadline, now))
3390 break;
3391
3392 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3393 deadline - now);
3394 } while (deadline);
3395 ata_for_each_link(link, ap, EDGE) {
3396 ata_for_each_dev(dev, link, ALL) {
3397 if (!(link->eh_context.unloaded_mask &
3398 (1 << dev->devno)))
3399 continue;
3400
3401 ata_eh_park_issue_cmd(dev, 0);
3402 ata_eh_done(link, dev, ATA_EH_PARK);
3403 }
3404 }
3405
3406 /* the rest */
3407 ata_for_each_link(link, ap, EDGE) {
3408 struct ata_eh_context *ehc = &link->eh_context;
3409
3410 /* revalidate existing devices and attach new ones */
3411 rc = ata_eh_revalidate_and_attach(link, &dev);
3412 if (rc)
3413 goto dev_fail;
3414
3415 /* if PMP got attached, return, pmp EH will take care of it */
3416 if (link->device->class == ATA_DEV_PMP) {
3417 ehc->i.action = 0;
3418 return 0;
3419 }
3420
3421 /* configure transfer mode if necessary */
3422 if (ehc->i.flags & ATA_EHI_SETMODE) {
3423 rc = ata_set_mode(link, &dev);
3424 if (rc)
3425 goto dev_fail;
3426 ehc->i.flags &= ~ATA_EHI_SETMODE;
3427 }
3428
3429 /* If reset has been issued, clear UA to avoid
3430 * disrupting the current users of the device.
3431 */
3432 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3433 ata_for_each_dev(dev, link, ALL) {
3434 if (dev->class != ATA_DEV_ATAPI)
3435 continue;
3436 rc = atapi_eh_clear_ua(dev);
3437 if (rc)
3438 goto dev_fail;
3439 }
3440 }
3441
3442 /* configure link power saving */
3443 if (ehc->i.action & ATA_EH_LPM)
3444 ata_for_each_dev(dev, link, ALL)
3445 ata_dev_enable_pm(dev, ap->pm_policy);
3446
3447 /* this link is okay now */
3448 ehc->i.flags = 0;
3449 continue;
3450
3451 dev_fail:
3452 nr_failed_devs++;
3453 ata_eh_handle_dev_fail(dev, rc);
3454
3455 if (ap->pflags & ATA_PFLAG_FROZEN) {
3456 /* PMP reset requires working host port.
3457 * Can't retry if it's frozen.
3458 */
3459 if (sata_pmp_attached(ap))
3460 goto out;
3461 break;
3462 }
3463 }
3464
3465 if (nr_failed_devs)
3466 goto retry;
3467
3468 out:
3469 if (rc && r_failed_link)
3470 *r_failed_link = link;
3471
3472 DPRINTK("EXIT, rc=%d\n", rc);
3473 return rc;
3474 }
3475
3476 /**
3477 * ata_eh_finish - finish up EH
3478 * @ap: host port to finish EH for
3479 *
3480 * Recovery is complete. Clean up EH states and retry or finish
3481 * failed qcs.
3482 *
3483 * LOCKING:
3484 * None.
3485 */
3486 void ata_eh_finish(struct ata_port *ap)
3487 {
3488 int tag;
3489
3490 /* retry or finish qcs */
3491 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3492 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3493
3494 if (!(qc->flags & ATA_QCFLAG_FAILED))
3495 continue;
3496
3497 if (qc->err_mask) {
3498 /* FIXME: Once EH migration is complete,
3499 * generate sense data in this function,
3500 * considering both err_mask and tf.
3501 */
3502 if (qc->flags & ATA_QCFLAG_RETRY)
3503 ata_eh_qc_retry(qc);
3504 else
3505 ata_eh_qc_complete(qc);
3506 } else {
3507 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3508 ata_eh_qc_complete(qc);
3509 } else {
3510 /* feed zero TF to sense generation */
3511 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3512 ata_eh_qc_retry(qc);
3513 }
3514 }
3515 }
3516
3517 /* make sure nr_active_links is zero after EH */
3518 WARN_ON(ap->nr_active_links);
3519 ap->nr_active_links = 0;
3520 }
3521
3522 /**
3523 * ata_do_eh - do standard error handling
3524 * @ap: host port to handle error for
3525 *
3526 * @prereset: prereset method (can be NULL)
3527 * @softreset: softreset method (can be NULL)
3528 * @hardreset: hardreset method (can be NULL)
3529 * @postreset: postreset method (can be NULL)
3530 *
3531 * Perform standard error handling sequence.
3532 *
3533 * LOCKING:
3534 * Kernel thread context (may sleep).
3535 */
3536 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3537 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3538 ata_postreset_fn_t postreset)
3539 {
3540 struct ata_device *dev;
3541 int rc;
3542
3543 ata_eh_autopsy(ap);
3544 ata_eh_report(ap);
3545
3546 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3547 NULL);
3548 if (rc) {
3549 ata_for_each_dev(dev, &ap->link, ALL)
3550 ata_dev_disable(dev);
3551 }
3552
3553 ata_eh_finish(ap);
3554 }
3555
3556 /**
3557 * ata_std_error_handler - standard error handler
3558 * @ap: host port to handle error for
3559 *
3560 * Standard error handler
3561 *
3562 * LOCKING:
3563 * Kernel thread context (may sleep).
3564 */
3565 void ata_std_error_handler(struct ata_port *ap)
3566 {
3567 struct ata_port_operations *ops = ap->ops;
3568 ata_reset_fn_t hardreset = ops->hardreset;
3569
3570 /* ignore built-in hardreset if SCR access is not available */
3571 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3572 hardreset = NULL;
3573
3574 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3575 }
3576
3577 #ifdef CONFIG_PM
3578 /**
3579 * ata_eh_handle_port_suspend - perform port suspend operation
3580 * @ap: port to suspend
3581 *
3582 * Suspend @ap.
3583 *
3584 * LOCKING:
3585 * Kernel thread context (may sleep).
3586 */
3587 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3588 {
3589 unsigned long flags;
3590 int rc = 0;
3591
3592 /* are we suspending? */
3593 spin_lock_irqsave(ap->lock, flags);
3594 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3595 ap->pm_mesg.event == PM_EVENT_ON) {
3596 spin_unlock_irqrestore(ap->lock, flags);
3597 return;
3598 }
3599 spin_unlock_irqrestore(ap->lock, flags);
3600
3601 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3602
3603 /* tell ACPI we're suspending */
3604 rc = ata_acpi_on_suspend(ap);
3605 if (rc)
3606 goto out;
3607
3608 /* suspend */
3609 ata_eh_freeze_port(ap);
3610
3611 if (ap->ops->port_suspend)
3612 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3613
3614 ata_acpi_set_state(ap, PMSG_SUSPEND);
3615 out:
3616 /* report result */
3617 spin_lock_irqsave(ap->lock, flags);
3618
3619 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3620 if (rc == 0)
3621 ap->pflags |= ATA_PFLAG_SUSPENDED;
3622 else if (ap->pflags & ATA_PFLAG_FROZEN)
3623 ata_port_schedule_eh(ap);
3624
3625 if (ap->pm_result) {
3626 *ap->pm_result = rc;
3627 ap->pm_result = NULL;
3628 }
3629
3630 spin_unlock_irqrestore(ap->lock, flags);
3631
3632 return;
3633 }
3634
3635 /**
3636 * ata_eh_handle_port_resume - perform port resume operation
3637 * @ap: port to resume
3638 *
3639 * Resume @ap.
3640 *
3641 * LOCKING:
3642 * Kernel thread context (may sleep).
3643 */
3644 static void ata_eh_handle_port_resume(struct ata_port *ap)
3645 {
3646 struct ata_link *link;
3647 struct ata_device *dev;
3648 unsigned long flags;
3649 int rc = 0;
3650
3651 /* are we resuming? */
3652 spin_lock_irqsave(ap->lock, flags);
3653 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3654 ap->pm_mesg.event != PM_EVENT_ON) {
3655 spin_unlock_irqrestore(ap->lock, flags);
3656 return;
3657 }
3658 spin_unlock_irqrestore(ap->lock, flags);
3659
3660 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3661
3662 /*
3663 * Error timestamps are in jiffies which doesn't run while
3664 * suspended and PHY events during resume isn't too uncommon.
3665 * When the two are combined, it can lead to unnecessary speed
3666 * downs if the machine is suspended and resumed repeatedly.
3667 * Clear error history.
3668 */
3669 ata_for_each_link(link, ap, HOST_FIRST)
3670 ata_for_each_dev(dev, link, ALL)
3671 ata_ering_clear(&dev->ering);
3672
3673 ata_acpi_set_state(ap, PMSG_ON);
3674
3675 if (ap->ops->port_resume)
3676 rc = ap->ops->port_resume(ap);
3677
3678 /* tell ACPI that we're resuming */
3679 ata_acpi_on_resume(ap);
3680
3681 /* report result */
3682 spin_lock_irqsave(ap->lock, flags);
3683 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3684 if (ap->pm_result) {
3685 *ap->pm_result = rc;
3686 ap->pm_result = NULL;
3687 }
3688 spin_unlock_irqrestore(ap->lock, flags);
3689 }
3690 #endif /* CONFIG_PM */
This page took 0.113687 seconds and 5 git commands to generate.