Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * libata-eh.c - libata error handling
4 *
5 * Copyright 2006 Tejun Heo <htejun@gmail.com>
6 *
7 * libata documentation is available via 'make {ps|pdf}docs',
8 * as Documentation/driver-api/libata.rst
9 *
10 * Hardware documentation available from http://www.t13.org/ and
11 * http://www.sata-io.org/
12 */
13
14#include <linux/kernel.h>
15#include <linux/blkdev.h>
16#include <linux/export.h>
17#include <linux/pci.h>
18#include <scsi/scsi.h>
19#include <scsi/scsi_host.h>
20#include <scsi/scsi_eh.h>
21#include <scsi/scsi_device.h>
22#include <scsi/scsi_cmnd.h>
23#include <scsi/scsi_dbg.h>
24#include "../scsi/scsi_transport_api.h"
25
26#include <linux/libata.h>
27
28#include <trace/events/libata.h>
29#include "libata.h"
30
31enum {
32 /* speed down verdicts */
33 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
34 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
35 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
36 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
37
38 /* error flags */
39 ATA_EFLAG_IS_IO = (1 << 0),
40 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
41 ATA_EFLAG_OLD_ER = (1 << 31),
42
43 /* error categories */
44 ATA_ECAT_NONE = 0,
45 ATA_ECAT_ATA_BUS = 1,
46 ATA_ECAT_TOUT_HSM = 2,
47 ATA_ECAT_UNK_DEV = 3,
48 ATA_ECAT_DUBIOUS_NONE = 4,
49 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
50 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
51 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
52 ATA_ECAT_NR = 8,
53
54 ATA_EH_CMD_DFL_TIMEOUT = 5000,
55
56 /* always put at least this amount of time between resets */
57 ATA_EH_RESET_COOL_DOWN = 5000,
58
59 /* Waiting in ->prereset can never be reliable. It's
60 * sometimes nice to wait there but it can't be depended upon;
61 * otherwise, we wouldn't be resetting. Just give it enough
62 * time for most drives to spin up.
63 */
64 ATA_EH_PRERESET_TIMEOUT = 10000,
65 ATA_EH_FASTDRAIN_INTERVAL = 3000,
66
67 ATA_EH_UA_TRIES = 5,
68
69 /* probe speed down parameters, see ata_eh_schedule_probe() */
70 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
71 ATA_EH_PROBE_TRIALS = 2,
72};
73
74/* The following table determines how we sequence resets. Each entry
75 * represents timeout for that try. The first try can be soft or
76 * hardreset. All others are hardreset if available. In most cases
77 * the first reset w/ 10sec timeout should succeed. Following entries
78 * are mostly for error handling, hotplug and those outlier devices that
79 * take an exceptionally long time to recover from reset.
80 */
81static const unsigned long ata_eh_reset_timeouts[] = {
82 10000, /* most drives spin up by 10sec */
83 10000, /* > 99% working drives spin up before 20sec */
84 35000, /* give > 30 secs of idleness for outlier devices */
85 5000, /* and sweet one last chance */
86 ULONG_MAX, /* > 1 min has elapsed, give up */
87};
88
89static const unsigned int ata_eh_identify_timeouts[] = {
90 5000, /* covers > 99% of successes and not too boring on failures */
91 10000, /* combined time till here is enough even for media access */
92 30000, /* for true idiots */
93 UINT_MAX,
94};
95
96static const unsigned int ata_eh_revalidate_timeouts[] = {
97 15000, /* Some drives are slow to read log pages when waking-up */
98 15000, /* combined time till here is enough even for media access */
99 UINT_MAX,
100};
101
102static const unsigned int ata_eh_flush_timeouts[] = {
103 15000, /* be generous with flush */
104 15000, /* ditto */
105 30000, /* and even more generous */
106 UINT_MAX,
107};
108
109static const unsigned int ata_eh_other_timeouts[] = {
110 5000, /* same rationale as identify timeout */
111 10000, /* ditto */
112 /* but no merciful 30sec for other commands, it just isn't worth it */
113 UINT_MAX,
114};
115
116struct ata_eh_cmd_timeout_ent {
117 const u8 *commands;
118 const unsigned int *timeouts;
119};
120
121/* The following table determines timeouts to use for EH internal
122 * commands. Each table entry is a command class and matches the
123 * commands the entry applies to and the timeout table to use.
124 *
125 * On the retry after a command timed out, the next timeout value from
126 * the table is used. If the table doesn't contain further entries,
127 * the last value is used.
128 *
129 * ehc->cmd_timeout_idx keeps track of which timeout to use per
130 * command class, so if SET_FEATURES times out on the first try, the
131 * next try will use the second timeout value only for that class.
132 */
133#define CMDS(cmds...) (const u8 []){ cmds, 0 }
134static const struct ata_eh_cmd_timeout_ent
135ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
136 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
137 .timeouts = ata_eh_identify_timeouts, },
138 { .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT),
139 .timeouts = ata_eh_revalidate_timeouts, },
140 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
141 .timeouts = ata_eh_other_timeouts, },
142 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_SET_FEATURES),
145 .timeouts = ata_eh_other_timeouts, },
146 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
149 .timeouts = ata_eh_flush_timeouts },
150};
151#undef CMDS
152
153static void __ata_port_freeze(struct ata_port *ap);
154static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
155 struct ata_device **r_failed_dev);
156#ifdef CONFIG_PM
157static void ata_eh_handle_port_suspend(struct ata_port *ap);
158static void ata_eh_handle_port_resume(struct ata_port *ap);
159#else /* CONFIG_PM */
160static void ata_eh_handle_port_suspend(struct ata_port *ap)
161{ }
162
163static void ata_eh_handle_port_resume(struct ata_port *ap)
164{ }
165#endif /* CONFIG_PM */
166
167static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
168 const char *fmt, va_list args)
169{
170 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
171 ATA_EH_DESC_LEN - ehi->desc_len,
172 fmt, args);
173}
174
175/**
176 * __ata_ehi_push_desc - push error description without adding separator
177 * @ehi: target EHI
178 * @fmt: printf format string
179 *
180 * Format string according to @fmt and append it to @ehi->desc.
181 *
182 * LOCKING:
183 * spin_lock_irqsave(host lock)
184 */
185void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
186{
187 va_list args;
188
189 va_start(args, fmt);
190 __ata_ehi_pushv_desc(ehi, fmt, args);
191 va_end(args);
192}
193EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
194
195/**
196 * ata_ehi_push_desc - push error description with separator
197 * @ehi: target EHI
198 * @fmt: printf format string
199 *
200 * Format string according to @fmt and append it to @ehi->desc.
201 * If @ehi->desc is not empty, ", " is added in-between.
202 *
203 * LOCKING:
204 * spin_lock_irqsave(host lock)
205 */
206void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
207{
208 va_list args;
209
210 if (ehi->desc_len)
211 __ata_ehi_push_desc(ehi, ", ");
212
213 va_start(args, fmt);
214 __ata_ehi_pushv_desc(ehi, fmt, args);
215 va_end(args);
216}
217EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
218
219/**
220 * ata_ehi_clear_desc - clean error description
221 * @ehi: target EHI
222 *
223 * Clear @ehi->desc.
224 *
225 * LOCKING:
226 * spin_lock_irqsave(host lock)
227 */
228void ata_ehi_clear_desc(struct ata_eh_info *ehi)
229{
230 ehi->desc[0] = '\0';
231 ehi->desc_len = 0;
232}
233EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
234
235/**
236 * ata_port_desc - append port description
237 * @ap: target ATA port
238 * @fmt: printf format string
239 *
240 * Format string according to @fmt and append it to port
241 * description. If port description is not empty, " " is added
242 * in-between. This function is to be used while initializing
243 * ata_host. The description is printed on host registration.
244 *
245 * LOCKING:
246 * None.
247 */
248void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
249{
250 va_list args;
251
252 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
253
254 if (ap->link.eh_info.desc_len)
255 __ata_ehi_push_desc(&ap->link.eh_info, " ");
256
257 va_start(args, fmt);
258 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
259 va_end(args);
260}
261EXPORT_SYMBOL_GPL(ata_port_desc);
262
263#ifdef CONFIG_PCI
264/**
265 * ata_port_pbar_desc - append PCI BAR description
266 * @ap: target ATA port
267 * @bar: target PCI BAR
268 * @offset: offset into PCI BAR
269 * @name: name of the area
270 *
271 * If @offset is negative, this function formats a string which
272 * contains the name, address, size and type of the BAR and
273 * appends it to the port description. If @offset is zero or
274 * positive, only name and offsetted address is appended.
275 *
276 * LOCKING:
277 * None.
278 */
279void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
280 const char *name)
281{
282 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
283 char *type = "";
284 unsigned long long start, len;
285
286 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
287 type = "m";
288 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
289 type = "i";
290
291 start = (unsigned long long)pci_resource_start(pdev, bar);
292 len = (unsigned long long)pci_resource_len(pdev, bar);
293
294 if (offset < 0)
295 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
296 else
297 ata_port_desc(ap, "%s 0x%llx", name,
298 start + (unsigned long long)offset);
299}
300EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
301#endif /* CONFIG_PCI */
302
303static int ata_lookup_timeout_table(u8 cmd)
304{
305 int i;
306
307 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
308 const u8 *cur;
309
310 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
311 if (*cur == cmd)
312 return i;
313 }
314
315 return -1;
316}
317
318/**
319 * ata_internal_cmd_timeout - determine timeout for an internal command
320 * @dev: target device
321 * @cmd: internal command to be issued
322 *
323 * Determine timeout for internal command @cmd for @dev.
324 *
325 * LOCKING:
326 * EH context.
327 *
328 * RETURNS:
329 * Determined timeout.
330 */
331unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
332{
333 struct ata_eh_context *ehc = &dev->link->eh_context;
334 int ent = ata_lookup_timeout_table(cmd);
335 int idx;
336
337 if (ent < 0)
338 return ATA_EH_CMD_DFL_TIMEOUT;
339
340 idx = ehc->cmd_timeout_idx[dev->devno][ent];
341 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
342}
343
344/**
345 * ata_internal_cmd_timed_out - notification for internal command timeout
346 * @dev: target device
347 * @cmd: internal command which timed out
348 *
349 * Notify EH that internal command @cmd for @dev timed out. This
350 * function should be called only for commands whose timeouts are
351 * determined using ata_internal_cmd_timeout().
352 *
353 * LOCKING:
354 * EH context.
355 */
356void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
357{
358 struct ata_eh_context *ehc = &dev->link->eh_context;
359 int ent = ata_lookup_timeout_table(cmd);
360 int idx;
361
362 if (ent < 0)
363 return;
364
365 idx = ehc->cmd_timeout_idx[dev->devno][ent];
366 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
367 ehc->cmd_timeout_idx[dev->devno][ent]++;
368}
369
370static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
371 unsigned int err_mask)
372{
373 struct ata_ering_entry *ent;
374
375 WARN_ON(!err_mask);
376
377 ering->cursor++;
378 ering->cursor %= ATA_ERING_SIZE;
379
380 ent = &ering->ring[ering->cursor];
381 ent->eflags = eflags;
382 ent->err_mask = err_mask;
383 ent->timestamp = get_jiffies_64();
384}
385
386static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
387{
388 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
389
390 if (ent->err_mask)
391 return ent;
392 return NULL;
393}
394
395int 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
416static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
417{
418 ent->eflags |= ATA_EFLAG_OLD_ER;
419 return 0;
420}
421
422static void ata_ering_clear(struct ata_ering *ering)
423{
424 ata_ering_map(ering, ata_ering_clear_cb, NULL);
425}
426
427static unsigned int ata_eh_dev_action(struct ata_device *dev)
428{
429 struct ata_eh_context *ehc = &dev->link->eh_context;
430
431 return ehc->i.action | ehc->i.dev_action[dev->devno];
432}
433
434static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
435 struct ata_eh_info *ehi, unsigned int action)
436{
437 struct ata_device *tdev;
438
439 if (!dev) {
440 ehi->action &= ~action;
441 ata_for_each_dev(tdev, link, ALL)
442 ehi->dev_action[tdev->devno] &= ~action;
443 } else {
444 /* doesn't make sense for port-wide EH actions */
445 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
446
447 /* break ehi->action into ehi->dev_action */
448 if (ehi->action & action) {
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] |=
451 ehi->action & action;
452 ehi->action &= ~action;
453 }
454
455 /* turn off the specified per-dev action */
456 ehi->dev_action[dev->devno] &= ~action;
457 }
458}
459
460/**
461 * ata_eh_acquire - acquire EH ownership
462 * @ap: ATA port to acquire EH ownership for
463 *
464 * Acquire EH ownership for @ap. This is the basic exclusion
465 * mechanism for ports sharing a host. Only one port hanging off
466 * the same host can claim the ownership of EH.
467 *
468 * LOCKING:
469 * EH context.
470 */
471void ata_eh_acquire(struct ata_port *ap)
472{
473 mutex_lock(&ap->host->eh_mutex);
474 WARN_ON_ONCE(ap->host->eh_owner);
475 ap->host->eh_owner = current;
476}
477
478/**
479 * ata_eh_release - release EH ownership
480 * @ap: ATA port to release EH ownership for
481 *
482 * Release EH ownership for @ap if the caller. The caller must
483 * have acquired EH ownership using ata_eh_acquire() previously.
484 *
485 * LOCKING:
486 * EH context.
487 */
488void ata_eh_release(struct ata_port *ap)
489{
490 WARN_ON_ONCE(ap->host->eh_owner != current);
491 ap->host->eh_owner = NULL;
492 mutex_unlock(&ap->host->eh_mutex);
493}
494
495static void ata_eh_unload(struct ata_port *ap)
496{
497 struct ata_link *link;
498 struct ata_device *dev;
499 unsigned long flags;
500
501 /* Restore SControl IPM and SPD for the next driver and
502 * disable attached devices.
503 */
504 ata_for_each_link(link, ap, PMP_FIRST) {
505 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
506 ata_for_each_dev(dev, link, ALL)
507 ata_dev_disable(dev);
508 }
509
510 /* freeze and set UNLOADED */
511 spin_lock_irqsave(ap->lock, flags);
512
513 ata_port_freeze(ap); /* won't be thawed */
514 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
515 ap->pflags |= ATA_PFLAG_UNLOADED;
516
517 spin_unlock_irqrestore(ap->lock, flags);
518}
519
520/**
521 * ata_scsi_error - SCSI layer error handler callback
522 * @host: SCSI host on which error occurred
523 *
524 * Handles SCSI-layer-thrown error events.
525 *
526 * LOCKING:
527 * Inherited from SCSI layer (none, can sleep)
528 *
529 * RETURNS:
530 * Zero.
531 */
532void ata_scsi_error(struct Scsi_Host *host)
533{
534 struct ata_port *ap = ata_shost_to_port(host);
535 unsigned long flags;
536 LIST_HEAD(eh_work_q);
537
538 spin_lock_irqsave(host->host_lock, flags);
539 list_splice_init(&host->eh_cmd_q, &eh_work_q);
540 spin_unlock_irqrestore(host->host_lock, flags);
541
542 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
543
544 /* If we timed raced normal completion and there is nothing to
545 recover nr_timedout == 0 why exactly are we doing error recovery ? */
546 ata_scsi_port_error_handler(host, ap);
547
548 /* finish or retry handled scmd's and clean up */
549 WARN_ON(!list_empty(&eh_work_q));
550
551}
552
553/**
554 * ata_scsi_cmd_error_handler - error callback for a list of commands
555 * @host: scsi host containing the port
556 * @ap: ATA port within the host
557 * @eh_work_q: list of commands to process
558 *
559 * process the given list of commands and return those finished to the
560 * ap->eh_done_q. This function is the first part of the libata error
561 * handler which processes a given list of failed commands.
562 */
563void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
564 struct list_head *eh_work_q)
565{
566 int i;
567 unsigned long flags;
568
569 /* make sure sff pio task is not running */
570 ata_sff_flush_pio_task(ap);
571
572 /* synchronize with host lock and sort out timeouts */
573
574 /* For new EH, all qcs are finished in one of three ways -
575 * normal completion, error completion, and SCSI timeout.
576 * Both completions can race against SCSI timeout. When normal
577 * completion wins, the qc never reaches EH. When error
578 * completion wins, the qc has ATA_QCFLAG_FAILED set.
579 *
580 * When SCSI timeout wins, things are a bit more complex.
581 * Normal or error completion can occur after the timeout but
582 * before this point. In such cases, both types of
583 * completions are honored. A scmd is determined to have
584 * timed out iff its associated qc is active and not failed.
585 */
586 spin_lock_irqsave(ap->lock, flags);
587 if (ap->ops->error_handler) {
588 struct scsi_cmnd *scmd, *tmp;
589 int nr_timedout = 0;
590
591 /* This must occur under the ap->lock as we don't want
592 a polled recovery to race the real interrupt handler
593
594 The lost_interrupt handler checks for any completed but
595 non-notified command and completes much like an IRQ handler.
596
597 We then fall into the error recovery code which will treat
598 this as if normal completion won the race */
599
600 if (ap->ops->lost_interrupt)
601 ap->ops->lost_interrupt(ap);
602
603 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
604 struct ata_queued_cmd *qc;
605
606 ata_qc_for_each_raw(ap, qc, i) {
607 if (qc->flags & ATA_QCFLAG_ACTIVE &&
608 qc->scsicmd == scmd)
609 break;
610 }
611
612 if (i < ATA_MAX_QUEUE) {
613 /* the scmd has an associated qc */
614 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
615 /* which hasn't failed yet, timeout */
616 qc->err_mask |= AC_ERR_TIMEOUT;
617 qc->flags |= ATA_QCFLAG_FAILED;
618 nr_timedout++;
619 }
620 } else {
621 /* Normal completion occurred after
622 * SCSI timeout but before this point.
623 * Successfully complete it.
624 */
625 scmd->retries = scmd->allowed;
626 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
627 }
628 }
629
630 /* If we have timed out qcs. They belong to EH from
631 * this point but the state of the controller is
632 * unknown. Freeze the port to make sure the IRQ
633 * handler doesn't diddle with those qcs. This must
634 * be done atomically w.r.t. setting QCFLAG_FAILED.
635 */
636 if (nr_timedout)
637 __ata_port_freeze(ap);
638
639
640 /* initialize eh_tries */
641 ap->eh_tries = ATA_EH_MAX_TRIES;
642 }
643 spin_unlock_irqrestore(ap->lock, flags);
644
645}
646EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
647
648/**
649 * ata_scsi_port_error_handler - recover the port after the commands
650 * @host: SCSI host containing the port
651 * @ap: the ATA port
652 *
653 * Handle the recovery of the port @ap after all the commands
654 * have been recovered.
655 */
656void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
657{
658 unsigned long flags;
659
660 /* invoke error handler */
661 if (ap->ops->error_handler) {
662 struct ata_link *link;
663
664 /* acquire EH ownership */
665 ata_eh_acquire(ap);
666 repeat:
667 /* kill fast drain timer */
668 del_timer_sync(&ap->fastdrain_timer);
669
670 /* process port resume request */
671 ata_eh_handle_port_resume(ap);
672
673 /* fetch & clear EH info */
674 spin_lock_irqsave(ap->lock, flags);
675
676 ata_for_each_link(link, ap, HOST_FIRST) {
677 struct ata_eh_context *ehc = &link->eh_context;
678 struct ata_device *dev;
679
680 memset(&link->eh_context, 0, sizeof(link->eh_context));
681 link->eh_context.i = link->eh_info;
682 memset(&link->eh_info, 0, sizeof(link->eh_info));
683
684 ata_for_each_dev(dev, link, ENABLED) {
685 int devno = dev->devno;
686
687 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
688 if (ata_ncq_enabled(dev))
689 ehc->saved_ncq_enabled |= 1 << devno;
690 }
691 }
692
693 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
694 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
695 ap->excl_link = NULL; /* don't maintain exclusion over EH */
696
697 spin_unlock_irqrestore(ap->lock, flags);
698
699 /* invoke EH, skip if unloading or suspended */
700 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
701 ap->ops->error_handler(ap);
702 else {
703 /* if unloading, commence suicide */
704 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
705 !(ap->pflags & ATA_PFLAG_UNLOADED))
706 ata_eh_unload(ap);
707 ata_eh_finish(ap);
708 }
709
710 /* process port suspend request */
711 ata_eh_handle_port_suspend(ap);
712
713 /* Exception might have happened after ->error_handler
714 * recovered the port but before this point. Repeat
715 * EH in such case.
716 */
717 spin_lock_irqsave(ap->lock, flags);
718
719 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
720 if (--ap->eh_tries) {
721 spin_unlock_irqrestore(ap->lock, flags);
722 goto repeat;
723 }
724 ata_port_err(ap,
725 "EH pending after %d tries, giving up\n",
726 ATA_EH_MAX_TRIES);
727 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
728 }
729
730 /* this run is complete, make sure EH info is clear */
731 ata_for_each_link(link, ap, HOST_FIRST)
732 memset(&link->eh_info, 0, sizeof(link->eh_info));
733
734 /* end eh (clear host_eh_scheduled) while holding
735 * ap->lock such that if exception occurs after this
736 * point but before EH completion, SCSI midlayer will
737 * re-initiate EH.
738 */
739 ap->ops->end_eh(ap);
740
741 spin_unlock_irqrestore(ap->lock, flags);
742 ata_eh_release(ap);
743 } else {
744 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
745 ap->ops->eng_timeout(ap);
746 }
747
748 scsi_eh_flush_done_q(&ap->eh_done_q);
749
750 /* clean up */
751 spin_lock_irqsave(ap->lock, flags);
752
753 if (ap->pflags & ATA_PFLAG_LOADING)
754 ap->pflags &= ~ATA_PFLAG_LOADING;
755 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
756 !(ap->flags & ATA_FLAG_SAS_HOST))
757 schedule_delayed_work(&ap->hotplug_task, 0);
758
759 if (ap->pflags & ATA_PFLAG_RECOVERED)
760 ata_port_info(ap, "EH complete\n");
761
762 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
763
764 /* tell wait_eh that we're done */
765 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
766 wake_up_all(&ap->eh_wait_q);
767
768 spin_unlock_irqrestore(ap->lock, flags);
769}
770EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
771
772/**
773 * ata_port_wait_eh - Wait for the currently pending EH to complete
774 * @ap: Port to wait EH for
775 *
776 * Wait until the currently pending EH is complete.
777 *
778 * LOCKING:
779 * Kernel thread context (may sleep).
780 */
781void ata_port_wait_eh(struct ata_port *ap)
782{
783 unsigned long flags;
784 DEFINE_WAIT(wait);
785
786 retry:
787 spin_lock_irqsave(ap->lock, flags);
788
789 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
790 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
791 spin_unlock_irqrestore(ap->lock, flags);
792 schedule();
793 spin_lock_irqsave(ap->lock, flags);
794 }
795 finish_wait(&ap->eh_wait_q, &wait);
796
797 spin_unlock_irqrestore(ap->lock, flags);
798
799 /* make sure SCSI EH is complete */
800 if (scsi_host_in_recovery(ap->scsi_host)) {
801 ata_msleep(ap, 10);
802 goto retry;
803 }
804}
805EXPORT_SYMBOL_GPL(ata_port_wait_eh);
806
807static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
808{
809 struct ata_queued_cmd *qc;
810 unsigned int tag;
811 unsigned int nr = 0;
812
813 /* count only non-internal commands */
814 ata_qc_for_each(ap, qc, tag) {
815 if (qc)
816 nr++;
817 }
818
819 return nr;
820}
821
822void ata_eh_fastdrain_timerfn(struct timer_list *t)
823{
824 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
825 unsigned long flags;
826 unsigned int cnt;
827
828 spin_lock_irqsave(ap->lock, flags);
829
830 cnt = ata_eh_nr_in_flight(ap);
831
832 /* are we done? */
833 if (!cnt)
834 goto out_unlock;
835
836 if (cnt == ap->fastdrain_cnt) {
837 struct ata_queued_cmd *qc;
838 unsigned int tag;
839
840 /* No progress during the last interval, tag all
841 * in-flight qcs as timed out and freeze the port.
842 */
843 ata_qc_for_each(ap, qc, tag) {
844 if (qc)
845 qc->err_mask |= AC_ERR_TIMEOUT;
846 }
847
848 ata_port_freeze(ap);
849 } else {
850 /* some qcs have finished, give it another chance */
851 ap->fastdrain_cnt = cnt;
852 ap->fastdrain_timer.expires =
853 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
854 add_timer(&ap->fastdrain_timer);
855 }
856
857 out_unlock:
858 spin_unlock_irqrestore(ap->lock, flags);
859}
860
861/**
862 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
863 * @ap: target ATA port
864 * @fastdrain: activate fast drain
865 *
866 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
867 * is non-zero and EH wasn't pending before. Fast drain ensures
868 * that EH kicks in in timely manner.
869 *
870 * LOCKING:
871 * spin_lock_irqsave(host lock)
872 */
873static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
874{
875 unsigned int cnt;
876
877 /* already scheduled? */
878 if (ap->pflags & ATA_PFLAG_EH_PENDING)
879 return;
880
881 ap->pflags |= ATA_PFLAG_EH_PENDING;
882
883 if (!fastdrain)
884 return;
885
886 /* do we have in-flight qcs? */
887 cnt = ata_eh_nr_in_flight(ap);
888 if (!cnt)
889 return;
890
891 /* activate fast drain */
892 ap->fastdrain_cnt = cnt;
893 ap->fastdrain_timer.expires =
894 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
895 add_timer(&ap->fastdrain_timer);
896}
897
898/**
899 * ata_qc_schedule_eh - schedule qc for error handling
900 * @qc: command to schedule error handling for
901 *
902 * Schedule error handling for @qc. EH will kick in as soon as
903 * other commands are drained.
904 *
905 * LOCKING:
906 * spin_lock_irqsave(host lock)
907 */
908void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
909{
910 struct ata_port *ap = qc->ap;
911
912 WARN_ON(!ap->ops->error_handler);
913
914 qc->flags |= ATA_QCFLAG_FAILED;
915 ata_eh_set_pending(ap, 1);
916
917 /* The following will fail if timeout has already expired.
918 * ata_scsi_error() takes care of such scmds on EH entry.
919 * Note that ATA_QCFLAG_FAILED is unconditionally set after
920 * this function completes.
921 */
922 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
923}
924
925/**
926 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
927 * @ap: ATA port to schedule EH for
928 *
929 * LOCKING: inherited from ata_port_schedule_eh
930 * spin_lock_irqsave(host lock)
931 */
932void ata_std_sched_eh(struct ata_port *ap)
933{
934 WARN_ON(!ap->ops->error_handler);
935
936 if (ap->pflags & ATA_PFLAG_INITIALIZING)
937 return;
938
939 ata_eh_set_pending(ap, 1);
940 scsi_schedule_eh(ap->scsi_host);
941
942 trace_ata_std_sched_eh(ap);
943}
944EXPORT_SYMBOL_GPL(ata_std_sched_eh);
945
946/**
947 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
948 * @ap: ATA port to end EH for
949 *
950 * In the libata object model there is a 1:1 mapping of ata_port to
951 * shost, so host fields can be directly manipulated under ap->lock, in
952 * the libsas case we need to hold a lock at the ha->level to coordinate
953 * these events.
954 *
955 * LOCKING:
956 * spin_lock_irqsave(host lock)
957 */
958void ata_std_end_eh(struct ata_port *ap)
959{
960 struct Scsi_Host *host = ap->scsi_host;
961
962 host->host_eh_scheduled = 0;
963}
964EXPORT_SYMBOL(ata_std_end_eh);
965
966
967/**
968 * ata_port_schedule_eh - schedule error handling without a qc
969 * @ap: ATA port to schedule EH for
970 *
971 * Schedule error handling for @ap. EH will kick in as soon as
972 * all commands are drained.
973 *
974 * LOCKING:
975 * spin_lock_irqsave(host lock)
976 */
977void ata_port_schedule_eh(struct ata_port *ap)
978{
979 /* see: ata_std_sched_eh, unless you know better */
980 ap->ops->sched_eh(ap);
981}
982EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
983
984static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
985{
986 struct ata_queued_cmd *qc;
987 int tag, nr_aborted = 0;
988
989 WARN_ON(!ap->ops->error_handler);
990
991 /* we're gonna abort all commands, no need for fast drain */
992 ata_eh_set_pending(ap, 0);
993
994 /* include internal tag in iteration */
995 ata_qc_for_each_with_internal(ap, qc, tag) {
996 if (qc && (!link || qc->dev->link == link)) {
997 qc->flags |= ATA_QCFLAG_FAILED;
998 ata_qc_complete(qc);
999 nr_aborted++;
1000 }
1001 }
1002
1003 if (!nr_aborted)
1004 ata_port_schedule_eh(ap);
1005
1006 return nr_aborted;
1007}
1008
1009/**
1010 * ata_link_abort - abort all qc's on the link
1011 * @link: ATA link to abort qc's for
1012 *
1013 * Abort all active qc's active on @link and schedule EH.
1014 *
1015 * LOCKING:
1016 * spin_lock_irqsave(host lock)
1017 *
1018 * RETURNS:
1019 * Number of aborted qc's.
1020 */
1021int ata_link_abort(struct ata_link *link)
1022{
1023 return ata_do_link_abort(link->ap, link);
1024}
1025EXPORT_SYMBOL_GPL(ata_link_abort);
1026
1027/**
1028 * ata_port_abort - abort all qc's on the port
1029 * @ap: ATA port to abort qc's for
1030 *
1031 * Abort all active qc's of @ap and schedule EH.
1032 *
1033 * LOCKING:
1034 * spin_lock_irqsave(host_set lock)
1035 *
1036 * RETURNS:
1037 * Number of aborted qc's.
1038 */
1039int ata_port_abort(struct ata_port *ap)
1040{
1041 return ata_do_link_abort(ap, NULL);
1042}
1043EXPORT_SYMBOL_GPL(ata_port_abort);
1044
1045/**
1046 * __ata_port_freeze - freeze port
1047 * @ap: ATA port to freeze
1048 *
1049 * This function is called when HSM violation or some other
1050 * condition disrupts normal operation of the port. Frozen port
1051 * is not allowed to perform any operation until the port is
1052 * thawed, which usually follows a successful reset.
1053 *
1054 * ap->ops->freeze() callback can be used for freezing the port
1055 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1056 * port cannot be frozen hardware-wise, the interrupt handler
1057 * must ack and clear interrupts unconditionally while the port
1058 * is frozen.
1059 *
1060 * LOCKING:
1061 * spin_lock_irqsave(host lock)
1062 */
1063static void __ata_port_freeze(struct ata_port *ap)
1064{
1065 WARN_ON(!ap->ops->error_handler);
1066
1067 if (ap->ops->freeze)
1068 ap->ops->freeze(ap);
1069
1070 ap->pflags |= ATA_PFLAG_FROZEN;
1071
1072 trace_ata_port_freeze(ap);
1073}
1074
1075/**
1076 * ata_port_freeze - abort & freeze port
1077 * @ap: ATA port to freeze
1078 *
1079 * Abort and freeze @ap. The freeze operation must be called
1080 * first, because some hardware requires special operations
1081 * before the taskfile registers are accessible.
1082 *
1083 * LOCKING:
1084 * spin_lock_irqsave(host lock)
1085 *
1086 * RETURNS:
1087 * Number of aborted commands.
1088 */
1089int ata_port_freeze(struct ata_port *ap)
1090{
1091 WARN_ON(!ap->ops->error_handler);
1092
1093 __ata_port_freeze(ap);
1094
1095 return ata_port_abort(ap);
1096}
1097EXPORT_SYMBOL_GPL(ata_port_freeze);
1098
1099/**
1100 * ata_eh_freeze_port - EH helper to freeze port
1101 * @ap: ATA port to freeze
1102 *
1103 * Freeze @ap.
1104 *
1105 * LOCKING:
1106 * None.
1107 */
1108void ata_eh_freeze_port(struct ata_port *ap)
1109{
1110 unsigned long flags;
1111
1112 if (!ap->ops->error_handler)
1113 return;
1114
1115 spin_lock_irqsave(ap->lock, flags);
1116 __ata_port_freeze(ap);
1117 spin_unlock_irqrestore(ap->lock, flags);
1118}
1119EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1120
1121/**
1122 * ata_eh_thaw_port - EH helper to thaw port
1123 * @ap: ATA port to thaw
1124 *
1125 * Thaw frozen port @ap.
1126 *
1127 * LOCKING:
1128 * None.
1129 */
1130void ata_eh_thaw_port(struct ata_port *ap)
1131{
1132 unsigned long flags;
1133
1134 if (!ap->ops->error_handler)
1135 return;
1136
1137 spin_lock_irqsave(ap->lock, flags);
1138
1139 ap->pflags &= ~ATA_PFLAG_FROZEN;
1140
1141 if (ap->ops->thaw)
1142 ap->ops->thaw(ap);
1143
1144 spin_unlock_irqrestore(ap->lock, flags);
1145
1146 trace_ata_port_thaw(ap);
1147}
1148
1149static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1150{
1151 /* nada */
1152}
1153
1154static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1155{
1156 struct ata_port *ap = qc->ap;
1157 struct scsi_cmnd *scmd = qc->scsicmd;
1158 unsigned long flags;
1159
1160 spin_lock_irqsave(ap->lock, flags);
1161 qc->scsidone = ata_eh_scsidone;
1162 __ata_qc_complete(qc);
1163 WARN_ON(ata_tag_valid(qc->tag));
1164 spin_unlock_irqrestore(ap->lock, flags);
1165
1166 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1167}
1168
1169/**
1170 * ata_eh_qc_complete - Complete an active ATA command from EH
1171 * @qc: Command to complete
1172 *
1173 * Indicate to the mid and upper layers that an ATA command has
1174 * completed. To be used from EH.
1175 */
1176void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1177{
1178 struct scsi_cmnd *scmd = qc->scsicmd;
1179 scmd->retries = scmd->allowed;
1180 __ata_eh_qc_complete(qc);
1181}
1182
1183/**
1184 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1185 * @qc: Command to retry
1186 *
1187 * Indicate to the mid and upper layers that an ATA command
1188 * should be retried. To be used from EH.
1189 *
1190 * SCSI midlayer limits the number of retries to scmd->allowed.
1191 * scmd->allowed is incremented for commands which get retried
1192 * due to unrelated failures (qc->err_mask is zero).
1193 */
1194void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1195{
1196 struct scsi_cmnd *scmd = qc->scsicmd;
1197 if (!qc->err_mask)
1198 scmd->allowed++;
1199 __ata_eh_qc_complete(qc);
1200}
1201
1202/**
1203 * ata_dev_disable - disable ATA device
1204 * @dev: ATA device to disable
1205 *
1206 * Disable @dev.
1207 *
1208 * Locking:
1209 * EH context.
1210 */
1211void ata_dev_disable(struct ata_device *dev)
1212{
1213 if (!ata_dev_enabled(dev))
1214 return;
1215
1216 ata_dev_warn(dev, "disable device\n");
1217 ata_acpi_on_disable(dev);
1218 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1219 dev->class++;
1220
1221 /* From now till the next successful probe, ering is used to
1222 * track probe failures. Clear accumulated device error info.
1223 */
1224 ata_ering_clear(&dev->ering);
1225}
1226EXPORT_SYMBOL_GPL(ata_dev_disable);
1227
1228/**
1229 * ata_eh_detach_dev - detach ATA device
1230 * @dev: ATA device to detach
1231 *
1232 * Detach @dev.
1233 *
1234 * LOCKING:
1235 * None.
1236 */
1237void ata_eh_detach_dev(struct ata_device *dev)
1238{
1239 struct ata_link *link = dev->link;
1240 struct ata_port *ap = link->ap;
1241 struct ata_eh_context *ehc = &link->eh_context;
1242 unsigned long flags;
1243
1244 ata_dev_disable(dev);
1245
1246 spin_lock_irqsave(ap->lock, flags);
1247
1248 dev->flags &= ~ATA_DFLAG_DETACH;
1249
1250 if (ata_scsi_offline_dev(dev)) {
1251 dev->flags |= ATA_DFLAG_DETACHED;
1252 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1253 }
1254
1255 /* clear per-dev EH info */
1256 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1257 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1258 ehc->saved_xfer_mode[dev->devno] = 0;
1259 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1260
1261 spin_unlock_irqrestore(ap->lock, flags);
1262}
1263
1264/**
1265 * ata_eh_about_to_do - about to perform eh_action
1266 * @link: target ATA link
1267 * @dev: target ATA dev for per-dev action (can be NULL)
1268 * @action: action about to be performed
1269 *
1270 * Called just before performing EH actions to clear related bits
1271 * in @link->eh_info such that eh actions are not unnecessarily
1272 * repeated.
1273 *
1274 * LOCKING:
1275 * None.
1276 */
1277void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1278 unsigned int action)
1279{
1280 struct ata_port *ap = link->ap;
1281 struct ata_eh_info *ehi = &link->eh_info;
1282 struct ata_eh_context *ehc = &link->eh_context;
1283 unsigned long flags;
1284
1285 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1286
1287 spin_lock_irqsave(ap->lock, flags);
1288
1289 ata_eh_clear_action(link, dev, ehi, action);
1290
1291 /* About to take EH action, set RECOVERED. Ignore actions on
1292 * slave links as master will do them again.
1293 */
1294 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1295 ap->pflags |= ATA_PFLAG_RECOVERED;
1296
1297 spin_unlock_irqrestore(ap->lock, flags);
1298}
1299
1300/**
1301 * ata_eh_done - EH action complete
1302 * @link: ATA link for which EH actions are complete
1303 * @dev: target ATA dev for per-dev action (can be NULL)
1304 * @action: action just completed
1305 *
1306 * Called right after performing EH actions to clear related bits
1307 * in @link->eh_context.
1308 *
1309 * LOCKING:
1310 * None.
1311 */
1312void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1313 unsigned int action)
1314{
1315 struct ata_eh_context *ehc = &link->eh_context;
1316
1317 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1318
1319 ata_eh_clear_action(link, dev, &ehc->i, action);
1320}
1321
1322/**
1323 * ata_err_string - convert err_mask to descriptive string
1324 * @err_mask: error mask to convert to string
1325 *
1326 * Convert @err_mask to descriptive string. Errors are
1327 * prioritized according to severity and only the most severe
1328 * error is reported.
1329 *
1330 * LOCKING:
1331 * None.
1332 *
1333 * RETURNS:
1334 * Descriptive string for @err_mask
1335 */
1336static const char *ata_err_string(unsigned int err_mask)
1337{
1338 if (err_mask & AC_ERR_HOST_BUS)
1339 return "host bus error";
1340 if (err_mask & AC_ERR_ATA_BUS)
1341 return "ATA bus error";
1342 if (err_mask & AC_ERR_TIMEOUT)
1343 return "timeout";
1344 if (err_mask & AC_ERR_HSM)
1345 return "HSM violation";
1346 if (err_mask & AC_ERR_SYSTEM)
1347 return "internal error";
1348 if (err_mask & AC_ERR_MEDIA)
1349 return "media error";
1350 if (err_mask & AC_ERR_INVALID)
1351 return "invalid argument";
1352 if (err_mask & AC_ERR_DEV)
1353 return "device error";
1354 if (err_mask & AC_ERR_NCQ)
1355 return "NCQ error";
1356 if (err_mask & AC_ERR_NODEV_HINT)
1357 return "Polling detection error";
1358 return "unknown error";
1359}
1360
1361/**
1362 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1363 * @dev: target ATAPI device
1364 * @r_sense_key: out parameter for sense_key
1365 *
1366 * Perform ATAPI TEST_UNIT_READY.
1367 *
1368 * LOCKING:
1369 * EH context (may sleep).
1370 *
1371 * RETURNS:
1372 * 0 on success, AC_ERR_* mask on failure.
1373 */
1374unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1375{
1376 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1377 struct ata_taskfile tf;
1378 unsigned int err_mask;
1379
1380 ata_tf_init(dev, &tf);
1381
1382 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1383 tf.command = ATA_CMD_PACKET;
1384 tf.protocol = ATAPI_PROT_NODATA;
1385
1386 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1387 if (err_mask == AC_ERR_DEV)
1388 *r_sense_key = tf.error >> 4;
1389 return err_mask;
1390}
1391
1392/**
1393 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1394 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1395 *
1396 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1397 * SENSE. This function is an EH helper.
1398 *
1399 * LOCKING:
1400 * Kernel thread context (may sleep).
1401 */
1402static void ata_eh_request_sense(struct ata_queued_cmd *qc)
1403{
1404 struct scsi_cmnd *cmd = qc->scsicmd;
1405 struct ata_device *dev = qc->dev;
1406 struct ata_taskfile tf;
1407 unsigned int err_mask;
1408
1409 if (ata_port_is_frozen(qc->ap)) {
1410 ata_dev_warn(dev, "sense data available but port frozen\n");
1411 return;
1412 }
1413
1414 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1415 return;
1416
1417 if (!ata_id_sense_reporting_enabled(dev->id)) {
1418 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1419 return;
1420 }
1421
1422 ata_tf_init(dev, &tf);
1423 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1424 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1425 tf.command = ATA_CMD_REQ_SENSE_DATA;
1426 tf.protocol = ATA_PROT_NODATA;
1427
1428 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1429 /* Ignore err_mask; ATA_ERR might be set */
1430 if (tf.status & ATA_SENSE) {
1431 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1432 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1433 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1434 }
1435 } else {
1436 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1437 tf.status, err_mask);
1438 }
1439}
1440
1441/**
1442 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1443 * @dev: device to perform REQUEST_SENSE to
1444 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1445 * @dfl_sense_key: default sense key to use
1446 *
1447 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1448 * SENSE. This function is EH helper.
1449 *
1450 * LOCKING:
1451 * Kernel thread context (may sleep).
1452 *
1453 * RETURNS:
1454 * 0 on success, AC_ERR_* mask on failure
1455 */
1456unsigned int atapi_eh_request_sense(struct ata_device *dev,
1457 u8 *sense_buf, u8 dfl_sense_key)
1458{
1459 u8 cdb[ATAPI_CDB_LEN] =
1460 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1461 struct ata_port *ap = dev->link->ap;
1462 struct ata_taskfile tf;
1463
1464 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1465
1466 /* initialize sense_buf with the error register,
1467 * for the case where they are -not- overwritten
1468 */
1469 sense_buf[0] = 0x70;
1470 sense_buf[2] = dfl_sense_key;
1471
1472 /* some devices time out if garbage left in tf */
1473 ata_tf_init(dev, &tf);
1474
1475 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1476 tf.command = ATA_CMD_PACKET;
1477
1478 /* is it pointless to prefer PIO for "safety reasons"? */
1479 if (ap->flags & ATA_FLAG_PIO_DMA) {
1480 tf.protocol = ATAPI_PROT_DMA;
1481 tf.feature |= ATAPI_PKT_DMA;
1482 } else {
1483 tf.protocol = ATAPI_PROT_PIO;
1484 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1485 tf.lbah = 0;
1486 }
1487
1488 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1489 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1490}
1491
1492/**
1493 * ata_eh_analyze_serror - analyze SError for a failed port
1494 * @link: ATA link to analyze SError for
1495 *
1496 * Analyze SError if available and further determine cause of
1497 * failure.
1498 *
1499 * LOCKING:
1500 * None.
1501 */
1502static void ata_eh_analyze_serror(struct ata_link *link)
1503{
1504 struct ata_eh_context *ehc = &link->eh_context;
1505 u32 serror = ehc->i.serror;
1506 unsigned int err_mask = 0, action = 0;
1507 u32 hotplug_mask;
1508
1509 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1510 err_mask |= AC_ERR_ATA_BUS;
1511 action |= ATA_EH_RESET;
1512 }
1513 if (serror & SERR_PROTOCOL) {
1514 err_mask |= AC_ERR_HSM;
1515 action |= ATA_EH_RESET;
1516 }
1517 if (serror & SERR_INTERNAL) {
1518 err_mask |= AC_ERR_SYSTEM;
1519 action |= ATA_EH_RESET;
1520 }
1521
1522 /* Determine whether a hotplug event has occurred. Both
1523 * SError.N/X are considered hotplug events for enabled or
1524 * host links. For disabled PMP links, only N bit is
1525 * considered as X bit is left at 1 for link plugging.
1526 */
1527 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1528 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1529 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1530 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1531 else
1532 hotplug_mask = SERR_PHYRDY_CHG;
1533
1534 if (serror & hotplug_mask)
1535 ata_ehi_hotplugged(&ehc->i);
1536
1537 ehc->i.err_mask |= err_mask;
1538 ehc->i.action |= action;
1539}
1540
1541/**
1542 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1543 * @qc: qc to analyze
1544 *
1545 * Analyze taskfile of @qc and further determine cause of
1546 * failure. This function also requests ATAPI sense data if
1547 * available.
1548 *
1549 * LOCKING:
1550 * Kernel thread context (may sleep).
1551 *
1552 * RETURNS:
1553 * Determined recovery action
1554 */
1555static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1556{
1557 const struct ata_taskfile *tf = &qc->result_tf;
1558 unsigned int tmp, action = 0;
1559 u8 stat = tf->status, err = tf->error;
1560
1561 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1562 qc->err_mask |= AC_ERR_HSM;
1563 return ATA_EH_RESET;
1564 }
1565
1566 if (stat & (ATA_ERR | ATA_DF)) {
1567 qc->err_mask |= AC_ERR_DEV;
1568 /*
1569 * Sense data reporting does not work if the
1570 * device fault bit is set.
1571 */
1572 if (stat & ATA_DF)
1573 stat &= ~ATA_SENSE;
1574 } else {
1575 return 0;
1576 }
1577
1578 switch (qc->dev->class) {
1579 case ATA_DEV_ATA:
1580 case ATA_DEV_ZAC:
1581 /*
1582 * Fetch the sense data explicitly if:
1583 * -It was a non-NCQ command that failed, or
1584 * -It was a NCQ command that failed, but the sense data
1585 * was not included in the NCQ command error log
1586 * (i.e. NCQ autosense is not supported by the device).
1587 */
1588 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) && (stat & ATA_SENSE))
1589 ata_eh_request_sense(qc);
1590 if (err & ATA_ICRC)
1591 qc->err_mask |= AC_ERR_ATA_BUS;
1592 if (err & (ATA_UNC | ATA_AMNF))
1593 qc->err_mask |= AC_ERR_MEDIA;
1594 if (err & ATA_IDNF)
1595 qc->err_mask |= AC_ERR_INVALID;
1596 break;
1597
1598 case ATA_DEV_ATAPI:
1599 if (!ata_port_is_frozen(qc->ap)) {
1600 tmp = atapi_eh_request_sense(qc->dev,
1601 qc->scsicmd->sense_buffer,
1602 qc->result_tf.error >> 4);
1603 if (!tmp)
1604 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1605 else
1606 qc->err_mask |= tmp;
1607 }
1608 }
1609
1610 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1611 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
1612 /*
1613 * SUCCESS here means that the sense code could be
1614 * evaluated and should be passed to the upper layers
1615 * for correct evaluation.
1616 * FAILED means the sense code could not be interpreted
1617 * and the device would need to be reset.
1618 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1619 * command would need to be retried.
1620 */
1621 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1622 qc->flags |= ATA_QCFLAG_RETRY;
1623 qc->err_mask |= AC_ERR_OTHER;
1624 } else if (ret != SUCCESS) {
1625 qc->err_mask |= AC_ERR_HSM;
1626 }
1627 }
1628 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1629 action |= ATA_EH_RESET;
1630
1631 return action;
1632}
1633
1634static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1635 int *xfer_ok)
1636{
1637 int base = 0;
1638
1639 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1640 *xfer_ok = 1;
1641
1642 if (!*xfer_ok)
1643 base = ATA_ECAT_DUBIOUS_NONE;
1644
1645 if (err_mask & AC_ERR_ATA_BUS)
1646 return base + ATA_ECAT_ATA_BUS;
1647
1648 if (err_mask & AC_ERR_TIMEOUT)
1649 return base + ATA_ECAT_TOUT_HSM;
1650
1651 if (eflags & ATA_EFLAG_IS_IO) {
1652 if (err_mask & AC_ERR_HSM)
1653 return base + ATA_ECAT_TOUT_HSM;
1654 if ((err_mask &
1655 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1656 return base + ATA_ECAT_UNK_DEV;
1657 }
1658
1659 return 0;
1660}
1661
1662struct speed_down_verdict_arg {
1663 u64 since;
1664 int xfer_ok;
1665 int nr_errors[ATA_ECAT_NR];
1666};
1667
1668static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1669{
1670 struct speed_down_verdict_arg *arg = void_arg;
1671 int cat;
1672
1673 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1674 return -1;
1675
1676 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1677 &arg->xfer_ok);
1678 arg->nr_errors[cat]++;
1679
1680 return 0;
1681}
1682
1683/**
1684 * ata_eh_speed_down_verdict - Determine speed down verdict
1685 * @dev: Device of interest
1686 *
1687 * This function examines error ring of @dev and determines
1688 * whether NCQ needs to be turned off, transfer speed should be
1689 * stepped down, or falling back to PIO is necessary.
1690 *
1691 * ECAT_ATA_BUS : ATA_BUS error for any command
1692 *
1693 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1694 * IO commands
1695 *
1696 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1697 *
1698 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1699 * data transfer hasn't been verified.
1700 *
1701 * Verdicts are
1702 *
1703 * NCQ_OFF : Turn off NCQ.
1704 *
1705 * SPEED_DOWN : Speed down transfer speed but don't fall back
1706 * to PIO.
1707 *
1708 * FALLBACK_TO_PIO : Fall back to PIO.
1709 *
1710 * Even if multiple verdicts are returned, only one action is
1711 * taken per error. An action triggered by non-DUBIOUS errors
1712 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1713 * This is to expedite speed down decisions right after device is
1714 * initially configured.
1715 *
1716 * The following are speed down rules. #1 and #2 deal with
1717 * DUBIOUS errors.
1718 *
1719 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1720 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1721 *
1722 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1723 * occurred during last 5 mins, NCQ_OFF.
1724 *
1725 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1726 * occurred during last 5 mins, FALLBACK_TO_PIO
1727 *
1728 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1729 * during last 10 mins, NCQ_OFF.
1730 *
1731 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1732 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1733 *
1734 * LOCKING:
1735 * Inherited from caller.
1736 *
1737 * RETURNS:
1738 * OR of ATA_EH_SPDN_* flags.
1739 */
1740static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1741{
1742 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1743 u64 j64 = get_jiffies_64();
1744 struct speed_down_verdict_arg arg;
1745 unsigned int verdict = 0;
1746
1747 /* scan past 5 mins of error history */
1748 memset(&arg, 0, sizeof(arg));
1749 arg.since = j64 - min(j64, j5mins);
1750 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1751
1752 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1753 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1754 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1755 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1756
1757 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1758 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1759 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1760
1761 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1762 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1763 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1764 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1765
1766 /* scan past 10 mins of error history */
1767 memset(&arg, 0, sizeof(arg));
1768 arg.since = j64 - min(j64, j10mins);
1769 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1770
1771 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1772 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1773 verdict |= ATA_EH_SPDN_NCQ_OFF;
1774
1775 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1776 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1777 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1778 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1779
1780 return verdict;
1781}
1782
1783/**
1784 * ata_eh_speed_down - record error and speed down if necessary
1785 * @dev: Failed device
1786 * @eflags: mask of ATA_EFLAG_* flags
1787 * @err_mask: err_mask of the error
1788 *
1789 * Record error and examine error history to determine whether
1790 * adjusting transmission speed is necessary. It also sets
1791 * transmission limits appropriately if such adjustment is
1792 * necessary.
1793 *
1794 * LOCKING:
1795 * Kernel thread context (may sleep).
1796 *
1797 * RETURNS:
1798 * Determined recovery action.
1799 */
1800static unsigned int ata_eh_speed_down(struct ata_device *dev,
1801 unsigned int eflags, unsigned int err_mask)
1802{
1803 struct ata_link *link = ata_dev_phys_link(dev);
1804 int xfer_ok = 0;
1805 unsigned int verdict;
1806 unsigned int action = 0;
1807
1808 /* don't bother if Cat-0 error */
1809 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1810 return 0;
1811
1812 /* record error and determine whether speed down is necessary */
1813 ata_ering_record(&dev->ering, eflags, err_mask);
1814 verdict = ata_eh_speed_down_verdict(dev);
1815
1816 /* turn off NCQ? */
1817 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1818 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1819 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1820 dev->flags |= ATA_DFLAG_NCQ_OFF;
1821 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1822 goto done;
1823 }
1824
1825 /* speed down? */
1826 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1827 /* speed down SATA link speed if possible */
1828 if (sata_down_spd_limit(link, 0) == 0) {
1829 action |= ATA_EH_RESET;
1830 goto done;
1831 }
1832
1833 /* lower transfer mode */
1834 if (dev->spdn_cnt < 2) {
1835 static const int dma_dnxfer_sel[] =
1836 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1837 static const int pio_dnxfer_sel[] =
1838 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1839 int sel;
1840
1841 if (dev->xfer_shift != ATA_SHIFT_PIO)
1842 sel = dma_dnxfer_sel[dev->spdn_cnt];
1843 else
1844 sel = pio_dnxfer_sel[dev->spdn_cnt];
1845
1846 dev->spdn_cnt++;
1847
1848 if (ata_down_xfermask_limit(dev, sel) == 0) {
1849 action |= ATA_EH_RESET;
1850 goto done;
1851 }
1852 }
1853 }
1854
1855 /* Fall back to PIO? Slowing down to PIO is meaningless for
1856 * SATA ATA devices. Consider it only for PATA and SATAPI.
1857 */
1858 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1859 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1860 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1861 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1862 dev->spdn_cnt = 0;
1863 action |= ATA_EH_RESET;
1864 goto done;
1865 }
1866 }
1867
1868 return 0;
1869 done:
1870 /* device has been slowed down, blow error history */
1871 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1872 ata_ering_clear(&dev->ering);
1873 return action;
1874}
1875
1876/**
1877 * ata_eh_worth_retry - analyze error and decide whether to retry
1878 * @qc: qc to possibly retry
1879 *
1880 * Look at the cause of the error and decide if a retry
1881 * might be useful or not. We don't want to retry media errors
1882 * because the drive itself has probably already taken 10-30 seconds
1883 * doing its own internal retries before reporting the failure.
1884 */
1885static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1886{
1887 if (qc->err_mask & AC_ERR_MEDIA)
1888 return 0; /* don't retry media errors */
1889 if (qc->flags & ATA_QCFLAG_IO)
1890 return 1; /* otherwise retry anything from fs stack */
1891 if (qc->err_mask & AC_ERR_INVALID)
1892 return 0; /* don't retry these */
1893 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1894}
1895
1896/**
1897 * ata_eh_quiet - check if we need to be quiet about a command error
1898 * @qc: qc to check
1899 *
1900 * Look at the qc flags anbd its scsi command request flags to determine
1901 * if we need to be quiet about the command failure.
1902 */
1903static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1904{
1905 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1906 qc->flags |= ATA_QCFLAG_QUIET;
1907 return qc->flags & ATA_QCFLAG_QUIET;
1908}
1909
1910/**
1911 * ata_eh_link_autopsy - analyze error and determine recovery action
1912 * @link: host link to perform autopsy on
1913 *
1914 * Analyze why @link failed and determine which recovery actions
1915 * are needed. This function also sets more detailed AC_ERR_*
1916 * values and fills sense data for ATAPI CHECK SENSE.
1917 *
1918 * LOCKING:
1919 * Kernel thread context (may sleep).
1920 */
1921static void ata_eh_link_autopsy(struct ata_link *link)
1922{
1923 struct ata_port *ap = link->ap;
1924 struct ata_eh_context *ehc = &link->eh_context;
1925 struct ata_queued_cmd *qc;
1926 struct ata_device *dev;
1927 unsigned int all_err_mask = 0, eflags = 0;
1928 int tag, nr_failed = 0, nr_quiet = 0;
1929 u32 serror;
1930 int rc;
1931
1932 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1933 return;
1934
1935 /* obtain and analyze SError */
1936 rc = sata_scr_read(link, SCR_ERROR, &serror);
1937 if (rc == 0) {
1938 ehc->i.serror |= serror;
1939 ata_eh_analyze_serror(link);
1940 } else if (rc != -EOPNOTSUPP) {
1941 /* SError read failed, force reset and probing */
1942 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1943 ehc->i.action |= ATA_EH_RESET;
1944 ehc->i.err_mask |= AC_ERR_OTHER;
1945 }
1946
1947 /* analyze NCQ failure */
1948 ata_eh_analyze_ncq_error(link);
1949
1950 /* any real error trumps AC_ERR_OTHER */
1951 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1952 ehc->i.err_mask &= ~AC_ERR_OTHER;
1953
1954 all_err_mask |= ehc->i.err_mask;
1955
1956 ata_qc_for_each_raw(ap, qc, tag) {
1957 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1958 qc->flags & ATA_QCFLAG_RETRY ||
1959 ata_dev_phys_link(qc->dev) != link)
1960 continue;
1961
1962 /* inherit upper level err_mask */
1963 qc->err_mask |= ehc->i.err_mask;
1964
1965 /* analyze TF */
1966 ehc->i.action |= ata_eh_analyze_tf(qc);
1967
1968 /* DEV errors are probably spurious in case of ATA_BUS error */
1969 if (qc->err_mask & AC_ERR_ATA_BUS)
1970 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1971 AC_ERR_INVALID);
1972
1973 /* any real error trumps unknown error */
1974 if (qc->err_mask & ~AC_ERR_OTHER)
1975 qc->err_mask &= ~AC_ERR_OTHER;
1976
1977 /*
1978 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
1979 * layers will determine whether the command is worth retrying
1980 * based on the sense data and device class/type. Otherwise,
1981 * determine directly if the command is worth retrying using its
1982 * error mask and flags.
1983 */
1984 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1985 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1986 else if (ata_eh_worth_retry(qc))
1987 qc->flags |= ATA_QCFLAG_RETRY;
1988
1989 /* accumulate error info */
1990 ehc->i.dev = qc->dev;
1991 all_err_mask |= qc->err_mask;
1992 if (qc->flags & ATA_QCFLAG_IO)
1993 eflags |= ATA_EFLAG_IS_IO;
1994 trace_ata_eh_link_autopsy_qc(qc);
1995
1996 /* Count quiet errors */
1997 if (ata_eh_quiet(qc))
1998 nr_quiet++;
1999 nr_failed++;
2000 }
2001
2002 /* If all failed commands requested silence, then be quiet */
2003 if (nr_quiet == nr_failed)
2004 ehc->i.flags |= ATA_EHI_QUIET;
2005
2006 /* enforce default EH actions */
2007 if (ata_port_is_frozen(ap) ||
2008 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2009 ehc->i.action |= ATA_EH_RESET;
2010 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2011 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2012 ehc->i.action |= ATA_EH_REVALIDATE;
2013
2014 /* If we have offending qcs and the associated failed device,
2015 * perform per-dev EH action only on the offending device.
2016 */
2017 if (ehc->i.dev) {
2018 ehc->i.dev_action[ehc->i.dev->devno] |=
2019 ehc->i.action & ATA_EH_PERDEV_MASK;
2020 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2021 }
2022
2023 /* propagate timeout to host link */
2024 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2025 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2026
2027 /* record error and consider speeding down */
2028 dev = ehc->i.dev;
2029 if (!dev && ((ata_link_max_devices(link) == 1 &&
2030 ata_dev_enabled(link->device))))
2031 dev = link->device;
2032
2033 if (dev) {
2034 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2035 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2036 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2037 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2038 }
2039}
2040
2041/**
2042 * ata_eh_autopsy - analyze error and determine recovery action
2043 * @ap: host port to perform autopsy on
2044 *
2045 * Analyze all links of @ap and determine why they failed and
2046 * which recovery actions are needed.
2047 *
2048 * LOCKING:
2049 * Kernel thread context (may sleep).
2050 */
2051void ata_eh_autopsy(struct ata_port *ap)
2052{
2053 struct ata_link *link;
2054
2055 ata_for_each_link(link, ap, EDGE)
2056 ata_eh_link_autopsy(link);
2057
2058 /* Handle the frigging slave link. Autopsy is done similarly
2059 * but actions and flags are transferred over to the master
2060 * link and handled from there.
2061 */
2062 if (ap->slave_link) {
2063 struct ata_eh_context *mehc = &ap->link.eh_context;
2064 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2065
2066 /* transfer control flags from master to slave */
2067 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2068
2069 /* perform autopsy on the slave link */
2070 ata_eh_link_autopsy(ap->slave_link);
2071
2072 /* transfer actions from slave to master and clear slave */
2073 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2074 mehc->i.action |= sehc->i.action;
2075 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2076 mehc->i.flags |= sehc->i.flags;
2077 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2078 }
2079
2080 /* Autopsy of fanout ports can affect host link autopsy.
2081 * Perform host link autopsy last.
2082 */
2083 if (sata_pmp_attached(ap))
2084 ata_eh_link_autopsy(&ap->link);
2085}
2086
2087/**
2088 * ata_get_cmd_name - get name for ATA command
2089 * @command: ATA command code to get name for
2090 *
2091 * Return a textual name of the given command or "unknown"
2092 *
2093 * LOCKING:
2094 * None
2095 */
2096const char *ata_get_cmd_name(u8 command)
2097{
2098#ifdef CONFIG_ATA_VERBOSE_ERROR
2099 static const struct
2100 {
2101 u8 command;
2102 const char *text;
2103 } cmd_descr[] = {
2104 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2105 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2106 { ATA_CMD_STANDBY, "STANDBY" },
2107 { ATA_CMD_IDLE, "IDLE" },
2108 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2109 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2110 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2111 { ATA_CMD_NOP, "NOP" },
2112 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2113 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2114 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2115 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2116 { ATA_CMD_SERVICE, "SERVICE" },
2117 { ATA_CMD_READ, "READ DMA" },
2118 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2119 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2120 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2121 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2122 { ATA_CMD_WRITE, "WRITE DMA" },
2123 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2124 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2125 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2126 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2127 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2128 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2129 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2130 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2131 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" },
2132 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2133 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2134 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2135 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2136 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2137 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2138 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2139 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2140 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2141 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2142 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2143 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2144 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2145 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2146 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2147 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2148 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2149 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2150 { ATA_CMD_SLEEP, "SLEEP" },
2151 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2152 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2153 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2154 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2155 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2156 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2157 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2158 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2159 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2160 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2161 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2162 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2163 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2164 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2165 { ATA_CMD_PMP_READ, "READ BUFFER" },
2166 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2167 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2168 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2169 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2170 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2171 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2172 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2173 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2174 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2175 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2176 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2177 { ATA_CMD_SMART, "SMART" },
2178 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2179 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2180 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2181 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2182 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2183 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2184 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2185 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2186 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2187 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2188 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2189 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2190 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2191 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2192 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2193 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2194 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2195 { ATA_CMD_RESTORE, "RECALIBRATE" },
2196 { 0, NULL } /* terminate list */
2197 };
2198
2199 unsigned int i;
2200 for (i = 0; cmd_descr[i].text; i++)
2201 if (cmd_descr[i].command == command)
2202 return cmd_descr[i].text;
2203#endif
2204
2205 return "unknown";
2206}
2207EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2208
2209/**
2210 * ata_eh_link_report - report error handling to user
2211 * @link: ATA link EH is going on
2212 *
2213 * Report EH to user.
2214 *
2215 * LOCKING:
2216 * None.
2217 */
2218static void ata_eh_link_report(struct ata_link *link)
2219{
2220 struct ata_port *ap = link->ap;
2221 struct ata_eh_context *ehc = &link->eh_context;
2222 struct ata_queued_cmd *qc;
2223 const char *frozen, *desc;
2224 char tries_buf[6] = "";
2225 int tag, nr_failed = 0;
2226
2227 if (ehc->i.flags & ATA_EHI_QUIET)
2228 return;
2229
2230 desc = NULL;
2231 if (ehc->i.desc[0] != '\0')
2232 desc = ehc->i.desc;
2233
2234 ata_qc_for_each_raw(ap, qc, tag) {
2235 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2236 ata_dev_phys_link(qc->dev) != link ||
2237 ((qc->flags & ATA_QCFLAG_QUIET) &&
2238 qc->err_mask == AC_ERR_DEV))
2239 continue;
2240 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2241 continue;
2242
2243 nr_failed++;
2244 }
2245
2246 if (!nr_failed && !ehc->i.err_mask)
2247 return;
2248
2249 frozen = "";
2250 if (ata_port_is_frozen(ap))
2251 frozen = " frozen";
2252
2253 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2254 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2255 ap->eh_tries);
2256
2257 if (ehc->i.dev) {
2258 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2259 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2260 ehc->i.err_mask, link->sactive, ehc->i.serror,
2261 ehc->i.action, frozen, tries_buf);
2262 if (desc)
2263 ata_dev_err(ehc->i.dev, "%s\n", desc);
2264 } else {
2265 ata_link_err(link, "exception Emask 0x%x "
2266 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2267 ehc->i.err_mask, link->sactive, ehc->i.serror,
2268 ehc->i.action, frozen, tries_buf);
2269 if (desc)
2270 ata_link_err(link, "%s\n", desc);
2271 }
2272
2273#ifdef CONFIG_ATA_VERBOSE_ERROR
2274 if (ehc->i.serror)
2275 ata_link_err(link,
2276 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2277 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2278 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2279 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2280 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2281 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2282 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2283 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2284 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2285 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2286 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2287 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2288 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2289 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2290 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2291 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2292 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2293 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2294#endif
2295
2296 ata_qc_for_each_raw(ap, qc, tag) {
2297 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2298 char data_buf[20] = "";
2299 char cdb_buf[70] = "";
2300
2301 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2302 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2303 continue;
2304
2305 if (qc->dma_dir != DMA_NONE) {
2306 static const char *dma_str[] = {
2307 [DMA_BIDIRECTIONAL] = "bidi",
2308 [DMA_TO_DEVICE] = "out",
2309 [DMA_FROM_DEVICE] = "in",
2310 };
2311 const char *prot_str = NULL;
2312
2313 switch (qc->tf.protocol) {
2314 case ATA_PROT_UNKNOWN:
2315 prot_str = "unknown";
2316 break;
2317 case ATA_PROT_NODATA:
2318 prot_str = "nodata";
2319 break;
2320 case ATA_PROT_PIO:
2321 prot_str = "pio";
2322 break;
2323 case ATA_PROT_DMA:
2324 prot_str = "dma";
2325 break;
2326 case ATA_PROT_NCQ:
2327 prot_str = "ncq dma";
2328 break;
2329 case ATA_PROT_NCQ_NODATA:
2330 prot_str = "ncq nodata";
2331 break;
2332 case ATAPI_PROT_NODATA:
2333 prot_str = "nodata";
2334 break;
2335 case ATAPI_PROT_PIO:
2336 prot_str = "pio";
2337 break;
2338 case ATAPI_PROT_DMA:
2339 prot_str = "dma";
2340 break;
2341 }
2342 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2343 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2344 }
2345
2346 if (ata_is_atapi(qc->tf.protocol)) {
2347 const u8 *cdb = qc->cdb;
2348 size_t cdb_len = qc->dev->cdb_len;
2349
2350 if (qc->scsicmd) {
2351 cdb = qc->scsicmd->cmnd;
2352 cdb_len = qc->scsicmd->cmd_len;
2353 }
2354 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2355 cdb, cdb_len);
2356 } else
2357 ata_dev_err(qc->dev, "failed command: %s\n",
2358 ata_get_cmd_name(cmd->command));
2359
2360 ata_dev_err(qc->dev,
2361 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2362 "tag %d%s\n %s"
2363 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2364 "Emask 0x%x (%s)%s\n",
2365 cmd->command, cmd->feature, cmd->nsect,
2366 cmd->lbal, cmd->lbam, cmd->lbah,
2367 cmd->hob_feature, cmd->hob_nsect,
2368 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2369 cmd->device, qc->tag, data_buf, cdb_buf,
2370 res->status, res->error, res->nsect,
2371 res->lbal, res->lbam, res->lbah,
2372 res->hob_feature, res->hob_nsect,
2373 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2374 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2375 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2376
2377#ifdef CONFIG_ATA_VERBOSE_ERROR
2378 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2379 ATA_SENSE | ATA_ERR)) {
2380 if (res->status & ATA_BUSY)
2381 ata_dev_err(qc->dev, "status: { Busy }\n");
2382 else
2383 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2384 res->status & ATA_DRDY ? "DRDY " : "",
2385 res->status & ATA_DF ? "DF " : "",
2386 res->status & ATA_DRQ ? "DRQ " : "",
2387 res->status & ATA_SENSE ? "SENSE " : "",
2388 res->status & ATA_ERR ? "ERR " : "");
2389 }
2390
2391 if (cmd->command != ATA_CMD_PACKET &&
2392 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2393 ATA_ABORTED)))
2394 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2395 res->error & ATA_ICRC ? "ICRC " : "",
2396 res->error & ATA_UNC ? "UNC " : "",
2397 res->error & ATA_AMNF ? "AMNF " : "",
2398 res->error & ATA_IDNF ? "IDNF " : "",
2399 res->error & ATA_ABORTED ? "ABRT " : "");
2400#endif
2401 }
2402}
2403
2404/**
2405 * ata_eh_report - report error handling to user
2406 * @ap: ATA port to report EH about
2407 *
2408 * Report EH to user.
2409 *
2410 * LOCKING:
2411 * None.
2412 */
2413void ata_eh_report(struct ata_port *ap)
2414{
2415 struct ata_link *link;
2416
2417 ata_for_each_link(link, ap, HOST_FIRST)
2418 ata_eh_link_report(link);
2419}
2420
2421static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2422 unsigned int *classes, unsigned long deadline,
2423 bool clear_classes)
2424{
2425 struct ata_device *dev;
2426
2427 if (clear_classes)
2428 ata_for_each_dev(dev, link, ALL)
2429 classes[dev->devno] = ATA_DEV_UNKNOWN;
2430
2431 return reset(link, classes, deadline);
2432}
2433
2434static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2435{
2436 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2437 return 0;
2438 if (rc == -EAGAIN)
2439 return 1;
2440 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2441 return 1;
2442 return 0;
2443}
2444
2445int ata_eh_reset(struct ata_link *link, int classify,
2446 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2447 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2448{
2449 struct ata_port *ap = link->ap;
2450 struct ata_link *slave = ap->slave_link;
2451 struct ata_eh_context *ehc = &link->eh_context;
2452 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2453 unsigned int *classes = ehc->classes;
2454 unsigned int lflags = link->flags;
2455 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2456 int max_tries = 0, try = 0;
2457 struct ata_link *failed_link;
2458 struct ata_device *dev;
2459 unsigned long deadline, now;
2460 ata_reset_fn_t reset;
2461 unsigned long flags;
2462 u32 sstatus;
2463 int nr_unknown, rc;
2464
2465 /*
2466 * Prepare to reset
2467 */
2468 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2469 max_tries++;
2470 if (link->flags & ATA_LFLAG_RST_ONCE)
2471 max_tries = 1;
2472 if (link->flags & ATA_LFLAG_NO_HRST)
2473 hardreset = NULL;
2474 if (link->flags & ATA_LFLAG_NO_SRST)
2475 softreset = NULL;
2476
2477 /* make sure each reset attempt is at least COOL_DOWN apart */
2478 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2479 now = jiffies;
2480 WARN_ON(time_after(ehc->last_reset, now));
2481 deadline = ata_deadline(ehc->last_reset,
2482 ATA_EH_RESET_COOL_DOWN);
2483 if (time_before(now, deadline))
2484 schedule_timeout_uninterruptible(deadline - now);
2485 }
2486
2487 spin_lock_irqsave(ap->lock, flags);
2488 ap->pflags |= ATA_PFLAG_RESETTING;
2489 spin_unlock_irqrestore(ap->lock, flags);
2490
2491 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2492
2493 ata_for_each_dev(dev, link, ALL) {
2494 /* If we issue an SRST then an ATA drive (not ATAPI)
2495 * may change configuration and be in PIO0 timing. If
2496 * we do a hard reset (or are coming from power on)
2497 * this is true for ATA or ATAPI. Until we've set a
2498 * suitable controller mode we should not touch the
2499 * bus as we may be talking too fast.
2500 */
2501 dev->pio_mode = XFER_PIO_0;
2502 dev->dma_mode = 0xff;
2503
2504 /* If the controller has a pio mode setup function
2505 * then use it to set the chipset to rights. Don't
2506 * touch the DMA setup as that will be dealt with when
2507 * configuring devices.
2508 */
2509 if (ap->ops->set_piomode)
2510 ap->ops->set_piomode(ap, dev);
2511 }
2512
2513 /* prefer hardreset */
2514 reset = NULL;
2515 ehc->i.action &= ~ATA_EH_RESET;
2516 if (hardreset) {
2517 reset = hardreset;
2518 ehc->i.action |= ATA_EH_HARDRESET;
2519 } else if (softreset) {
2520 reset = softreset;
2521 ehc->i.action |= ATA_EH_SOFTRESET;
2522 }
2523
2524 if (prereset) {
2525 unsigned long deadline = ata_deadline(jiffies,
2526 ATA_EH_PRERESET_TIMEOUT);
2527
2528 if (slave) {
2529 sehc->i.action &= ~ATA_EH_RESET;
2530 sehc->i.action |= ehc->i.action;
2531 }
2532
2533 rc = prereset(link, deadline);
2534
2535 /* If present, do prereset on slave link too. Reset
2536 * is skipped iff both master and slave links report
2537 * -ENOENT or clear ATA_EH_RESET.
2538 */
2539 if (slave && (rc == 0 || rc == -ENOENT)) {
2540 int tmp;
2541
2542 tmp = prereset(slave, deadline);
2543 if (tmp != -ENOENT)
2544 rc = tmp;
2545
2546 ehc->i.action |= sehc->i.action;
2547 }
2548
2549 if (rc) {
2550 if (rc == -ENOENT) {
2551 ata_link_dbg(link, "port disabled--ignoring\n");
2552 ehc->i.action &= ~ATA_EH_RESET;
2553
2554 ata_for_each_dev(dev, link, ALL)
2555 classes[dev->devno] = ATA_DEV_NONE;
2556
2557 rc = 0;
2558 } else
2559 ata_link_err(link,
2560 "prereset failed (errno=%d)\n",
2561 rc);
2562 goto out;
2563 }
2564
2565 /* prereset() might have cleared ATA_EH_RESET. If so,
2566 * bang classes, thaw and return.
2567 */
2568 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2569 ata_for_each_dev(dev, link, ALL)
2570 classes[dev->devno] = ATA_DEV_NONE;
2571 if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2572 ata_eh_thaw_port(ap);
2573 rc = 0;
2574 goto out;
2575 }
2576 }
2577
2578 retry:
2579 /*
2580 * Perform reset
2581 */
2582 if (ata_is_host_link(link))
2583 ata_eh_freeze_port(ap);
2584
2585 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2586
2587 if (reset) {
2588 if (verbose)
2589 ata_link_info(link, "%s resetting link\n",
2590 reset == softreset ? "soft" : "hard");
2591
2592 /* mark that this EH session started with reset */
2593 ehc->last_reset = jiffies;
2594 if (reset == hardreset) {
2595 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2596 trace_ata_link_hardreset_begin(link, classes, deadline);
2597 } else {
2598 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2599 trace_ata_link_softreset_begin(link, classes, deadline);
2600 }
2601
2602 rc = ata_do_reset(link, reset, classes, deadline, true);
2603 if (reset == hardreset)
2604 trace_ata_link_hardreset_end(link, classes, rc);
2605 else
2606 trace_ata_link_softreset_end(link, classes, rc);
2607 if (rc && rc != -EAGAIN) {
2608 failed_link = link;
2609 goto fail;
2610 }
2611
2612 /* hardreset slave link if existent */
2613 if (slave && reset == hardreset) {
2614 int tmp;
2615
2616 if (verbose)
2617 ata_link_info(slave, "hard resetting link\n");
2618
2619 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2620 trace_ata_slave_hardreset_begin(slave, classes,
2621 deadline);
2622 tmp = ata_do_reset(slave, reset, classes, deadline,
2623 false);
2624 trace_ata_slave_hardreset_end(slave, classes, tmp);
2625 switch (tmp) {
2626 case -EAGAIN:
2627 rc = -EAGAIN;
2628 break;
2629 case 0:
2630 break;
2631 default:
2632 failed_link = slave;
2633 rc = tmp;
2634 goto fail;
2635 }
2636 }
2637
2638 /* perform follow-up SRST if necessary */
2639 if (reset == hardreset &&
2640 ata_eh_followup_srst_needed(link, rc)) {
2641 reset = softreset;
2642
2643 if (!reset) {
2644 ata_link_err(link,
2645 "follow-up softreset required but no softreset available\n");
2646 failed_link = link;
2647 rc = -EINVAL;
2648 goto fail;
2649 }
2650
2651 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2652 trace_ata_link_softreset_begin(link, classes, deadline);
2653 rc = ata_do_reset(link, reset, classes, deadline, true);
2654 trace_ata_link_softreset_end(link, classes, rc);
2655 if (rc) {
2656 failed_link = link;
2657 goto fail;
2658 }
2659 }
2660 } else {
2661 if (verbose)
2662 ata_link_info(link,
2663 "no reset method available, skipping reset\n");
2664 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2665 lflags |= ATA_LFLAG_ASSUME_ATA;
2666 }
2667
2668 /*
2669 * Post-reset processing
2670 */
2671 ata_for_each_dev(dev, link, ALL) {
2672 /* After the reset, the device state is PIO 0 and the
2673 * controller state is undefined. Reset also wakes up
2674 * drives from sleeping mode.
2675 */
2676 dev->pio_mode = XFER_PIO_0;
2677 dev->flags &= ~ATA_DFLAG_SLEEPING;
2678
2679 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2680 continue;
2681
2682 /* apply class override */
2683 if (lflags & ATA_LFLAG_ASSUME_ATA)
2684 classes[dev->devno] = ATA_DEV_ATA;
2685 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2686 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2687 }
2688
2689 /* record current link speed */
2690 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2691 link->sata_spd = (sstatus >> 4) & 0xf;
2692 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2693 slave->sata_spd = (sstatus >> 4) & 0xf;
2694
2695 /* thaw the port */
2696 if (ata_is_host_link(link))
2697 ata_eh_thaw_port(ap);
2698
2699 /* postreset() should clear hardware SError. Although SError
2700 * is cleared during link resume, clearing SError here is
2701 * necessary as some PHYs raise hotplug events after SRST.
2702 * This introduces race condition where hotplug occurs between
2703 * reset and here. This race is mediated by cross checking
2704 * link onlineness and classification result later.
2705 */
2706 if (postreset) {
2707 postreset(link, classes);
2708 trace_ata_link_postreset(link, classes, rc);
2709 if (slave) {
2710 postreset(slave, classes);
2711 trace_ata_slave_postreset(slave, classes, rc);
2712 }
2713 }
2714
2715 /*
2716 * Some controllers can't be frozen very well and may set spurious
2717 * error conditions during reset. Clear accumulated error
2718 * information and re-thaw the port if frozen. As reset is the
2719 * final recovery action and we cross check link onlineness against
2720 * device classification later, no hotplug event is lost by this.
2721 */
2722 spin_lock_irqsave(link->ap->lock, flags);
2723 memset(&link->eh_info, 0, sizeof(link->eh_info));
2724 if (slave)
2725 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2726 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2727 spin_unlock_irqrestore(link->ap->lock, flags);
2728
2729 if (ata_port_is_frozen(ap))
2730 ata_eh_thaw_port(ap);
2731
2732 /*
2733 * Make sure onlineness and classification result correspond.
2734 * Hotplug could have happened during reset and some
2735 * controllers fail to wait while a drive is spinning up after
2736 * being hotplugged causing misdetection. By cross checking
2737 * link on/offlineness and classification result, those
2738 * conditions can be reliably detected and retried.
2739 */
2740 nr_unknown = 0;
2741 ata_for_each_dev(dev, link, ALL) {
2742 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2743 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2744 ata_dev_dbg(dev, "link online but device misclassified\n");
2745 classes[dev->devno] = ATA_DEV_NONE;
2746 nr_unknown++;
2747 }
2748 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2749 if (ata_class_enabled(classes[dev->devno]))
2750 ata_dev_dbg(dev,
2751 "link offline, clearing class %d to NONE\n",
2752 classes[dev->devno]);
2753 classes[dev->devno] = ATA_DEV_NONE;
2754 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2755 ata_dev_dbg(dev,
2756 "link status unknown, clearing UNKNOWN to NONE\n");
2757 classes[dev->devno] = ATA_DEV_NONE;
2758 }
2759 }
2760
2761 if (classify && nr_unknown) {
2762 if (try < max_tries) {
2763 ata_link_warn(link,
2764 "link online but %d devices misclassified, retrying\n",
2765 nr_unknown);
2766 failed_link = link;
2767 rc = -EAGAIN;
2768 goto fail;
2769 }
2770 ata_link_warn(link,
2771 "link online but %d devices misclassified, "
2772 "device detection might fail\n", nr_unknown);
2773 }
2774
2775 /* reset successful, schedule revalidation */
2776 ata_eh_done(link, NULL, ATA_EH_RESET);
2777 if (slave)
2778 ata_eh_done(slave, NULL, ATA_EH_RESET);
2779 ehc->last_reset = jiffies; /* update to completion time */
2780 ehc->i.action |= ATA_EH_REVALIDATE;
2781 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2782
2783 rc = 0;
2784 out:
2785 /* clear hotplug flag */
2786 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2787 if (slave)
2788 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2789
2790 spin_lock_irqsave(ap->lock, flags);
2791 ap->pflags &= ~ATA_PFLAG_RESETTING;
2792 spin_unlock_irqrestore(ap->lock, flags);
2793
2794 return rc;
2795
2796 fail:
2797 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2798 if (!ata_is_host_link(link) &&
2799 sata_scr_read(link, SCR_STATUS, &sstatus))
2800 rc = -ERESTART;
2801
2802 if (try >= max_tries) {
2803 /*
2804 * Thaw host port even if reset failed, so that the port
2805 * can be retried on the next phy event. This risks
2806 * repeated EH runs but seems to be a better tradeoff than
2807 * shutting down a port after a botched hotplug attempt.
2808 */
2809 if (ata_is_host_link(link))
2810 ata_eh_thaw_port(ap);
2811 goto out;
2812 }
2813
2814 now = jiffies;
2815 if (time_before(now, deadline)) {
2816 unsigned long delta = deadline - now;
2817
2818 ata_link_warn(failed_link,
2819 "reset failed (errno=%d), retrying in %u secs\n",
2820 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2821
2822 ata_eh_release(ap);
2823 while (delta)
2824 delta = schedule_timeout_uninterruptible(delta);
2825 ata_eh_acquire(ap);
2826 }
2827
2828 /*
2829 * While disks spinup behind PMP, some controllers fail sending SRST.
2830 * They need to be reset - as well as the PMP - before retrying.
2831 */
2832 if (rc == -ERESTART) {
2833 if (ata_is_host_link(link))
2834 ata_eh_thaw_port(ap);
2835 goto out;
2836 }
2837
2838 if (try == max_tries - 1) {
2839 sata_down_spd_limit(link, 0);
2840 if (slave)
2841 sata_down_spd_limit(slave, 0);
2842 } else if (rc == -EPIPE)
2843 sata_down_spd_limit(failed_link, 0);
2844
2845 if (hardreset)
2846 reset = hardreset;
2847 goto retry;
2848}
2849
2850static inline void ata_eh_pull_park_action(struct ata_port *ap)
2851{
2852 struct ata_link *link;
2853 struct ata_device *dev;
2854 unsigned long flags;
2855
2856 /*
2857 * This function can be thought of as an extended version of
2858 * ata_eh_about_to_do() specially crafted to accommodate the
2859 * requirements of ATA_EH_PARK handling. Since the EH thread
2860 * does not leave the do {} while () loop in ata_eh_recover as
2861 * long as the timeout for a park request to *one* device on
2862 * the port has not expired, and since we still want to pick
2863 * up park requests to other devices on the same port or
2864 * timeout updates for the same device, we have to pull
2865 * ATA_EH_PARK actions from eh_info into eh_context.i
2866 * ourselves at the beginning of each pass over the loop.
2867 *
2868 * Additionally, all write accesses to &ap->park_req_pending
2869 * through reinit_completion() (see below) or complete_all()
2870 * (see ata_scsi_park_store()) are protected by the host lock.
2871 * As a result we have that park_req_pending.done is zero on
2872 * exit from this function, i.e. when ATA_EH_PARK actions for
2873 * *all* devices on port ap have been pulled into the
2874 * respective eh_context structs. If, and only if,
2875 * park_req_pending.done is non-zero by the time we reach
2876 * wait_for_completion_timeout(), another ATA_EH_PARK action
2877 * has been scheduled for at least one of the devices on port
2878 * ap and we have to cycle over the do {} while () loop in
2879 * ata_eh_recover() again.
2880 */
2881
2882 spin_lock_irqsave(ap->lock, flags);
2883 reinit_completion(&ap->park_req_pending);
2884 ata_for_each_link(link, ap, EDGE) {
2885 ata_for_each_dev(dev, link, ALL) {
2886 struct ata_eh_info *ehi = &link->eh_info;
2887
2888 link->eh_context.i.dev_action[dev->devno] |=
2889 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2890 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2891 }
2892 }
2893 spin_unlock_irqrestore(ap->lock, flags);
2894}
2895
2896static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2897{
2898 struct ata_eh_context *ehc = &dev->link->eh_context;
2899 struct ata_taskfile tf;
2900 unsigned int err_mask;
2901
2902 ata_tf_init(dev, &tf);
2903 if (park) {
2904 ehc->unloaded_mask |= 1 << dev->devno;
2905 tf.command = ATA_CMD_IDLEIMMEDIATE;
2906 tf.feature = 0x44;
2907 tf.lbal = 0x4c;
2908 tf.lbam = 0x4e;
2909 tf.lbah = 0x55;
2910 } else {
2911 ehc->unloaded_mask &= ~(1 << dev->devno);
2912 tf.command = ATA_CMD_CHK_POWER;
2913 }
2914
2915 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2916 tf.protocol = ATA_PROT_NODATA;
2917 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2918 if (park && (err_mask || tf.lbal != 0xc4)) {
2919 ata_dev_err(dev, "head unload failed!\n");
2920 ehc->unloaded_mask &= ~(1 << dev->devno);
2921 }
2922}
2923
2924static int ata_eh_revalidate_and_attach(struct ata_link *link,
2925 struct ata_device **r_failed_dev)
2926{
2927 struct ata_port *ap = link->ap;
2928 struct ata_eh_context *ehc = &link->eh_context;
2929 struct ata_device *dev;
2930 unsigned int new_mask = 0;
2931 unsigned long flags;
2932 int rc = 0;
2933
2934 /* For PATA drive side cable detection to work, IDENTIFY must
2935 * be done backwards such that PDIAG- is released by the slave
2936 * device before the master device is identified.
2937 */
2938 ata_for_each_dev(dev, link, ALL_REVERSE) {
2939 unsigned int action = ata_eh_dev_action(dev);
2940 unsigned int readid_flags = 0;
2941
2942 if (ehc->i.flags & ATA_EHI_DID_RESET)
2943 readid_flags |= ATA_READID_POSTRESET;
2944
2945 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2946 WARN_ON(dev->class == ATA_DEV_PMP);
2947
2948 /*
2949 * The link may be in a deep sleep, wake it up.
2950 *
2951 * If the link is in deep sleep, ata_phys_link_offline()
2952 * will return true, causing the revalidation to fail,
2953 * which leads to a (potentially) needless hard reset.
2954 *
2955 * ata_eh_recover() will later restore the link policy
2956 * to ap->target_lpm_policy after revalidation is done.
2957 */
2958 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
2959 rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER,
2960 r_failed_dev);
2961 if (rc)
2962 goto err;
2963 }
2964
2965 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2966 rc = -EIO;
2967 goto err;
2968 }
2969
2970 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2971 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2972 readid_flags);
2973 if (rc)
2974 goto err;
2975
2976 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2977
2978 /* Configuration may have changed, reconfigure
2979 * transfer mode.
2980 */
2981 ehc->i.flags |= ATA_EHI_SETMODE;
2982
2983 /* schedule the scsi_rescan_device() here */
2984 schedule_work(&(ap->scsi_rescan_task));
2985 } else if (dev->class == ATA_DEV_UNKNOWN &&
2986 ehc->tries[dev->devno] &&
2987 ata_class_enabled(ehc->classes[dev->devno])) {
2988 /* Temporarily set dev->class, it will be
2989 * permanently set once all configurations are
2990 * complete. This is necessary because new
2991 * device configuration is done in two
2992 * separate loops.
2993 */
2994 dev->class = ehc->classes[dev->devno];
2995
2996 if (dev->class == ATA_DEV_PMP)
2997 rc = sata_pmp_attach(dev);
2998 else
2999 rc = ata_dev_read_id(dev, &dev->class,
3000 readid_flags, dev->id);
3001
3002 /* read_id might have changed class, store and reset */
3003 ehc->classes[dev->devno] = dev->class;
3004 dev->class = ATA_DEV_UNKNOWN;
3005
3006 switch (rc) {
3007 case 0:
3008 /* clear error info accumulated during probe */
3009 ata_ering_clear(&dev->ering);
3010 new_mask |= 1 << dev->devno;
3011 break;
3012 case -ENOENT:
3013 /* IDENTIFY was issued to non-existent
3014 * device. No need to reset. Just
3015 * thaw and ignore the device.
3016 */
3017 ata_eh_thaw_port(ap);
3018 break;
3019 default:
3020 goto err;
3021 }
3022 }
3023 }
3024
3025 /* PDIAG- should have been released, ask cable type if post-reset */
3026 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3027 if (ap->ops->cable_detect)
3028 ap->cbl = ap->ops->cable_detect(ap);
3029 ata_force_cbl(ap);
3030 }
3031
3032 /* Configure new devices forward such that user doesn't see
3033 * device detection messages backwards.
3034 */
3035 ata_for_each_dev(dev, link, ALL) {
3036 if (!(new_mask & (1 << dev->devno)))
3037 continue;
3038
3039 dev->class = ehc->classes[dev->devno];
3040
3041 if (dev->class == ATA_DEV_PMP)
3042 continue;
3043
3044 ehc->i.flags |= ATA_EHI_PRINTINFO;
3045 rc = ata_dev_configure(dev);
3046 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3047 if (rc) {
3048 dev->class = ATA_DEV_UNKNOWN;
3049 goto err;
3050 }
3051
3052 spin_lock_irqsave(ap->lock, flags);
3053 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3054 spin_unlock_irqrestore(ap->lock, flags);
3055
3056 /* new device discovered, configure xfermode */
3057 ehc->i.flags |= ATA_EHI_SETMODE;
3058 }
3059
3060 return 0;
3061
3062 err:
3063 *r_failed_dev = dev;
3064 return rc;
3065}
3066
3067/**
3068 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3069 * @link: link on which timings will be programmed
3070 * @r_failed_dev: out parameter for failed device
3071 *
3072 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3073 * ata_set_mode() fails, pointer to the failing device is
3074 * returned in @r_failed_dev.
3075 *
3076 * LOCKING:
3077 * PCI/etc. bus probe sem.
3078 *
3079 * RETURNS:
3080 * 0 on success, negative errno otherwise
3081 */
3082int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3083{
3084 struct ata_port *ap = link->ap;
3085 struct ata_device *dev;
3086 int rc;
3087
3088 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3089 ata_for_each_dev(dev, link, ENABLED) {
3090 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3091 struct ata_ering_entry *ent;
3092
3093 ent = ata_ering_top(&dev->ering);
3094 if (ent)
3095 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3096 }
3097 }
3098
3099 /* has private set_mode? */
3100 if (ap->ops->set_mode)
3101 rc = ap->ops->set_mode(link, r_failed_dev);
3102 else
3103 rc = ata_do_set_mode(link, r_failed_dev);
3104
3105 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3106 ata_for_each_dev(dev, link, ENABLED) {
3107 struct ata_eh_context *ehc = &link->eh_context;
3108 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3109 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3110
3111 if (dev->xfer_mode != saved_xfer_mode ||
3112 ata_ncq_enabled(dev) != saved_ncq)
3113 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3114 }
3115
3116 return rc;
3117}
3118
3119/**
3120 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3121 * @dev: ATAPI device to clear UA for
3122 *
3123 * Resets and other operations can make an ATAPI device raise
3124 * UNIT ATTENTION which causes the next operation to fail. This
3125 * function clears UA.
3126 *
3127 * LOCKING:
3128 * EH context (may sleep).
3129 *
3130 * RETURNS:
3131 * 0 on success, -errno on failure.
3132 */
3133static int atapi_eh_clear_ua(struct ata_device *dev)
3134{
3135 int i;
3136
3137 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3138 u8 *sense_buffer = dev->link->ap->sector_buf;
3139 u8 sense_key = 0;
3140 unsigned int err_mask;
3141
3142 err_mask = atapi_eh_tur(dev, &sense_key);
3143 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3144 ata_dev_warn(dev,
3145 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3146 err_mask);
3147 return -EIO;
3148 }
3149
3150 if (!err_mask || sense_key != UNIT_ATTENTION)
3151 return 0;
3152
3153 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3154 if (err_mask) {
3155 ata_dev_warn(dev, "failed to clear "
3156 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3157 return -EIO;
3158 }
3159 }
3160
3161 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3162 ATA_EH_UA_TRIES);
3163
3164 return 0;
3165}
3166
3167/**
3168 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3169 * @dev: ATA device which may need FLUSH retry
3170 *
3171 * If @dev failed FLUSH, it needs to be reported upper layer
3172 * immediately as it means that @dev failed to remap and already
3173 * lost at least a sector and further FLUSH retrials won't make
3174 * any difference to the lost sector. However, if FLUSH failed
3175 * for other reasons, for example transmission error, FLUSH needs
3176 * to be retried.
3177 *
3178 * This function determines whether FLUSH failure retry is
3179 * necessary and performs it if so.
3180 *
3181 * RETURNS:
3182 * 0 if EH can continue, -errno if EH needs to be repeated.
3183 */
3184static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3185{
3186 struct ata_link *link = dev->link;
3187 struct ata_port *ap = link->ap;
3188 struct ata_queued_cmd *qc;
3189 struct ata_taskfile tf;
3190 unsigned int err_mask;
3191 int rc = 0;
3192
3193 /* did flush fail for this device? */
3194 if (!ata_tag_valid(link->active_tag))
3195 return 0;
3196
3197 qc = __ata_qc_from_tag(ap, link->active_tag);
3198 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3199 qc->tf.command != ATA_CMD_FLUSH))
3200 return 0;
3201
3202 /* if the device failed it, it should be reported to upper layers */
3203 if (qc->err_mask & AC_ERR_DEV)
3204 return 0;
3205
3206 /* flush failed for some other reason, give it another shot */
3207 ata_tf_init(dev, &tf);
3208
3209 tf.command = qc->tf.command;
3210 tf.flags |= ATA_TFLAG_DEVICE;
3211 tf.protocol = ATA_PROT_NODATA;
3212
3213 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3214 tf.command, qc->err_mask);
3215
3216 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3217 if (!err_mask) {
3218 /*
3219 * FLUSH is complete but there's no way to
3220 * successfully complete a failed command from EH.
3221 * Making sure retry is allowed at least once and
3222 * retrying it should do the trick - whatever was in
3223 * the cache is already on the platter and this won't
3224 * cause infinite loop.
3225 */
3226 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3227 } else {
3228 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3229 err_mask);
3230 rc = -EIO;
3231
3232 /* if device failed it, report it to upper layers */
3233 if (err_mask & AC_ERR_DEV) {
3234 qc->err_mask |= AC_ERR_DEV;
3235 qc->result_tf = tf;
3236 if (!ata_port_is_frozen(ap))
3237 rc = 0;
3238 }
3239 }
3240 return rc;
3241}
3242
3243/**
3244 * ata_eh_set_lpm - configure SATA interface power management
3245 * @link: link to configure power management
3246 * @policy: the link power management policy
3247 * @r_failed_dev: out parameter for failed device
3248 *
3249 * Enable SATA Interface power management. This will enable
3250 * Device Interface Power Management (DIPM) for min_power and
3251 * medium_power_with_dipm policies, and then call driver specific
3252 * callbacks for enabling Host Initiated Power management.
3253 *
3254 * LOCKING:
3255 * EH context.
3256 *
3257 * RETURNS:
3258 * 0 on success, -errno on failure.
3259 */
3260static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3261 struct ata_device **r_failed_dev)
3262{
3263 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3264 struct ata_eh_context *ehc = &link->eh_context;
3265 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3266 enum ata_lpm_policy old_policy = link->lpm_policy;
3267 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3268 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3269 unsigned int err_mask;
3270 int rc;
3271
3272 /* if the link or host doesn't do LPM, noop */
3273 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3274 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3275 return 0;
3276
3277 /*
3278 * DIPM is enabled only for MIN_POWER as some devices
3279 * misbehave when the host NACKs transition to SLUMBER. Order
3280 * device and link configurations such that the host always
3281 * allows DIPM requests.
3282 */
3283 ata_for_each_dev(dev, link, ENABLED) {
3284 bool hipm = ata_id_has_hipm(dev->id);
3285 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3286
3287 /* find the first enabled and LPM enabled devices */
3288 if (!link_dev)
3289 link_dev = dev;
3290
3291 if (!lpm_dev && (hipm || dipm))
3292 lpm_dev = dev;
3293
3294 hints &= ~ATA_LPM_EMPTY;
3295 if (!hipm)
3296 hints &= ~ATA_LPM_HIPM;
3297
3298 /* disable DIPM before changing link config */
3299 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3300 err_mask = ata_dev_set_feature(dev,
3301 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3302 if (err_mask && err_mask != AC_ERR_DEV) {
3303 ata_dev_warn(dev,
3304 "failed to disable DIPM, Emask 0x%x\n",
3305 err_mask);
3306 rc = -EIO;
3307 goto fail;
3308 }
3309 }
3310 }
3311
3312 if (ap) {
3313 rc = ap->ops->set_lpm(link, policy, hints);
3314 if (!rc && ap->slave_link)
3315 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3316 } else
3317 rc = sata_pmp_set_lpm(link, policy, hints);
3318
3319 /*
3320 * Attribute link config failure to the first (LPM) enabled
3321 * device on the link.
3322 */
3323 if (rc) {
3324 if (rc == -EOPNOTSUPP) {
3325 link->flags |= ATA_LFLAG_NO_LPM;
3326 return 0;
3327 }
3328 dev = lpm_dev ? lpm_dev : link_dev;
3329 goto fail;
3330 }
3331
3332 /*
3333 * Low level driver acked the transition. Issue DIPM command
3334 * with the new policy set.
3335 */
3336 link->lpm_policy = policy;
3337 if (ap && ap->slave_link)
3338 ap->slave_link->lpm_policy = policy;
3339
3340 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3341 ata_for_each_dev(dev, link, ENABLED) {
3342 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3343 ata_id_has_dipm(dev->id)) {
3344 err_mask = ata_dev_set_feature(dev,
3345 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3346 if (err_mask && err_mask != AC_ERR_DEV) {
3347 ata_dev_warn(dev,
3348 "failed to enable DIPM, Emask 0x%x\n",
3349 err_mask);
3350 rc = -EIO;
3351 goto fail;
3352 }
3353 }
3354 }
3355
3356 link->last_lpm_change = jiffies;
3357 link->flags |= ATA_LFLAG_CHANGED;
3358
3359 return 0;
3360
3361fail:
3362 /* restore the old policy */
3363 link->lpm_policy = old_policy;
3364 if (ap && ap->slave_link)
3365 ap->slave_link->lpm_policy = old_policy;
3366
3367 /* if no device or only one more chance is left, disable LPM */
3368 if (!dev || ehc->tries[dev->devno] <= 2) {
3369 ata_link_warn(link, "disabling LPM on the link\n");
3370 link->flags |= ATA_LFLAG_NO_LPM;
3371 }
3372 if (r_failed_dev)
3373 *r_failed_dev = dev;
3374 return rc;
3375}
3376
3377int ata_link_nr_enabled(struct ata_link *link)
3378{
3379 struct ata_device *dev;
3380 int cnt = 0;
3381
3382 ata_for_each_dev(dev, link, ENABLED)
3383 cnt++;
3384 return cnt;
3385}
3386
3387static int ata_link_nr_vacant(struct ata_link *link)
3388{
3389 struct ata_device *dev;
3390 int cnt = 0;
3391
3392 ata_for_each_dev(dev, link, ALL)
3393 if (dev->class == ATA_DEV_UNKNOWN)
3394 cnt++;
3395 return cnt;
3396}
3397
3398static int ata_eh_skip_recovery(struct ata_link *link)
3399{
3400 struct ata_port *ap = link->ap;
3401 struct ata_eh_context *ehc = &link->eh_context;
3402 struct ata_device *dev;
3403
3404 /* skip disabled links */
3405 if (link->flags & ATA_LFLAG_DISABLED)
3406 return 1;
3407
3408 /* skip if explicitly requested */
3409 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3410 return 1;
3411
3412 /* thaw frozen port and recover failed devices */
3413 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3414 return 0;
3415
3416 /* reset at least once if reset is requested */
3417 if ((ehc->i.action & ATA_EH_RESET) &&
3418 !(ehc->i.flags & ATA_EHI_DID_RESET))
3419 return 0;
3420
3421 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3422 ata_for_each_dev(dev, link, ALL) {
3423 if (dev->class == ATA_DEV_UNKNOWN &&
3424 ehc->classes[dev->devno] != ATA_DEV_NONE)
3425 return 0;
3426 }
3427
3428 return 1;
3429}
3430
3431static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3432{
3433 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3434 u64 now = get_jiffies_64();
3435 int *trials = void_arg;
3436
3437 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3438 (ent->timestamp < now - min(now, interval)))
3439 return -1;
3440
3441 (*trials)++;
3442 return 0;
3443}
3444
3445static int ata_eh_schedule_probe(struct ata_device *dev)
3446{
3447 struct ata_eh_context *ehc = &dev->link->eh_context;
3448 struct ata_link *link = ata_dev_phys_link(dev);
3449 int trials = 0;
3450
3451 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3452 (ehc->did_probe_mask & (1 << dev->devno)))
3453 return 0;
3454
3455 ata_eh_detach_dev(dev);
3456 ata_dev_init(dev);
3457 ehc->did_probe_mask |= (1 << dev->devno);
3458 ehc->i.action |= ATA_EH_RESET;
3459 ehc->saved_xfer_mode[dev->devno] = 0;
3460 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3461
3462 /* the link maybe in a deep sleep, wake it up */
3463 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3464 if (ata_is_host_link(link))
3465 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3466 ATA_LPM_EMPTY);
3467 else
3468 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3469 ATA_LPM_EMPTY);
3470 }
3471
3472 /* Record and count probe trials on the ering. The specific
3473 * error mask used is irrelevant. Because a successful device
3474 * detection clears the ering, this count accumulates only if
3475 * there are consecutive failed probes.
3476 *
3477 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3478 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3479 * forced to 1.5Gbps.
3480 *
3481 * This is to work around cases where failed link speed
3482 * negotiation results in device misdetection leading to
3483 * infinite DEVXCHG or PHRDY CHG events.
3484 */
3485 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3486 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3487
3488 if (trials > ATA_EH_PROBE_TRIALS)
3489 sata_down_spd_limit(link, 1);
3490
3491 return 1;
3492}
3493
3494static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3495{
3496 struct ata_eh_context *ehc = &dev->link->eh_context;
3497
3498 /* -EAGAIN from EH routine indicates retry without prejudice.
3499 * The requester is responsible for ensuring forward progress.
3500 */
3501 if (err != -EAGAIN)
3502 ehc->tries[dev->devno]--;
3503
3504 switch (err) {
3505 case -ENODEV:
3506 /* device missing or wrong IDENTIFY data, schedule probing */
3507 ehc->i.probe_mask |= (1 << dev->devno);
3508 fallthrough;
3509 case -EINVAL:
3510 /* give it just one more chance */
3511 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3512 fallthrough;
3513 case -EIO:
3514 if (ehc->tries[dev->devno] == 1) {
3515 /* This is the last chance, better to slow
3516 * down than lose it.
3517 */
3518 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3519 if (dev->pio_mode > XFER_PIO_0)
3520 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3521 }
3522 }
3523
3524 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3525 /* disable device if it has used up all its chances */
3526 ata_dev_disable(dev);
3527
3528 /* detach if offline */
3529 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3530 ata_eh_detach_dev(dev);
3531
3532 /* schedule probe if necessary */
3533 if (ata_eh_schedule_probe(dev)) {
3534 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3535 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3536 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3537 }
3538
3539 return 1;
3540 } else {
3541 ehc->i.action |= ATA_EH_RESET;
3542 return 0;
3543 }
3544}
3545
3546/**
3547 * ata_eh_recover - recover host port after error
3548 * @ap: host port to recover
3549 * @prereset: prereset method (can be NULL)
3550 * @softreset: softreset method (can be NULL)
3551 * @hardreset: hardreset method (can be NULL)
3552 * @postreset: postreset method (can be NULL)
3553 * @r_failed_link: out parameter for failed link
3554 *
3555 * This is the alpha and omega, eum and yang, heart and soul of
3556 * libata exception handling. On entry, actions required to
3557 * recover each link and hotplug requests are recorded in the
3558 * link's eh_context. This function executes all the operations
3559 * with appropriate retrials and fallbacks to resurrect failed
3560 * devices, detach goners and greet newcomers.
3561 *
3562 * LOCKING:
3563 * Kernel thread context (may sleep).
3564 *
3565 * RETURNS:
3566 * 0 on success, -errno on failure.
3567 */
3568int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3569 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3570 ata_postreset_fn_t postreset,
3571 struct ata_link **r_failed_link)
3572{
3573 struct ata_link *link;
3574 struct ata_device *dev;
3575 int rc, nr_fails;
3576 unsigned long flags, deadline;
3577
3578 /* prep for recovery */
3579 ata_for_each_link(link, ap, EDGE) {
3580 struct ata_eh_context *ehc = &link->eh_context;
3581
3582 /* re-enable link? */
3583 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3584 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3585 spin_lock_irqsave(ap->lock, flags);
3586 link->flags &= ~ATA_LFLAG_DISABLED;
3587 spin_unlock_irqrestore(ap->lock, flags);
3588 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3589 }
3590
3591 ata_for_each_dev(dev, link, ALL) {
3592 if (link->flags & ATA_LFLAG_NO_RETRY)
3593 ehc->tries[dev->devno] = 1;
3594 else
3595 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3596
3597 /* collect port action mask recorded in dev actions */
3598 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3599 ~ATA_EH_PERDEV_MASK;
3600 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3601
3602 /* process hotplug request */
3603 if (dev->flags & ATA_DFLAG_DETACH)
3604 ata_eh_detach_dev(dev);
3605
3606 /* schedule probe if necessary */
3607 if (!ata_dev_enabled(dev))
3608 ata_eh_schedule_probe(dev);
3609 }
3610 }
3611
3612 retry:
3613 rc = 0;
3614
3615 /* if UNLOADING, finish immediately */
3616 if (ap->pflags & ATA_PFLAG_UNLOADING)
3617 goto out;
3618
3619 /* prep for EH */
3620 ata_for_each_link(link, ap, EDGE) {
3621 struct ata_eh_context *ehc = &link->eh_context;
3622
3623 /* skip EH if possible. */
3624 if (ata_eh_skip_recovery(link))
3625 ehc->i.action = 0;
3626
3627 ata_for_each_dev(dev, link, ALL)
3628 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3629 }
3630
3631 /* reset */
3632 ata_for_each_link(link, ap, EDGE) {
3633 struct ata_eh_context *ehc = &link->eh_context;
3634
3635 if (!(ehc->i.action & ATA_EH_RESET))
3636 continue;
3637
3638 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3639 prereset, softreset, hardreset, postreset);
3640 if (rc) {
3641 ata_link_err(link, "reset failed, giving up\n");
3642 goto out;
3643 }
3644 }
3645
3646 do {
3647 unsigned long now;
3648
3649 /*
3650 * clears ATA_EH_PARK in eh_info and resets
3651 * ap->park_req_pending
3652 */
3653 ata_eh_pull_park_action(ap);
3654
3655 deadline = jiffies;
3656 ata_for_each_link(link, ap, EDGE) {
3657 ata_for_each_dev(dev, link, ALL) {
3658 struct ata_eh_context *ehc = &link->eh_context;
3659 unsigned long tmp;
3660
3661 if (dev->class != ATA_DEV_ATA &&
3662 dev->class != ATA_DEV_ZAC)
3663 continue;
3664 if (!(ehc->i.dev_action[dev->devno] &
3665 ATA_EH_PARK))
3666 continue;
3667 tmp = dev->unpark_deadline;
3668 if (time_before(deadline, tmp))
3669 deadline = tmp;
3670 else if (time_before_eq(tmp, jiffies))
3671 continue;
3672 if (ehc->unloaded_mask & (1 << dev->devno))
3673 continue;
3674
3675 ata_eh_park_issue_cmd(dev, 1);
3676 }
3677 }
3678
3679 now = jiffies;
3680 if (time_before_eq(deadline, now))
3681 break;
3682
3683 ata_eh_release(ap);
3684 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3685 deadline - now);
3686 ata_eh_acquire(ap);
3687 } while (deadline);
3688 ata_for_each_link(link, ap, EDGE) {
3689 ata_for_each_dev(dev, link, ALL) {
3690 if (!(link->eh_context.unloaded_mask &
3691 (1 << dev->devno)))
3692 continue;
3693
3694 ata_eh_park_issue_cmd(dev, 0);
3695 ata_eh_done(link, dev, ATA_EH_PARK);
3696 }
3697 }
3698
3699 /* the rest */
3700 nr_fails = 0;
3701 ata_for_each_link(link, ap, PMP_FIRST) {
3702 struct ata_eh_context *ehc = &link->eh_context;
3703
3704 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3705 goto config_lpm;
3706
3707 /* revalidate existing devices and attach new ones */
3708 rc = ata_eh_revalidate_and_attach(link, &dev);
3709 if (rc)
3710 goto rest_fail;
3711
3712 /* if PMP got attached, return, pmp EH will take care of it */
3713 if (link->device->class == ATA_DEV_PMP) {
3714 ehc->i.action = 0;
3715 return 0;
3716 }
3717
3718 /* configure transfer mode if necessary */
3719 if (ehc->i.flags & ATA_EHI_SETMODE) {
3720 rc = ata_set_mode(link, &dev);
3721 if (rc)
3722 goto rest_fail;
3723 ehc->i.flags &= ~ATA_EHI_SETMODE;
3724 }
3725
3726 /* If reset has been issued, clear UA to avoid
3727 * disrupting the current users of the device.
3728 */
3729 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3730 ata_for_each_dev(dev, link, ALL) {
3731 if (dev->class != ATA_DEV_ATAPI)
3732 continue;
3733 rc = atapi_eh_clear_ua(dev);
3734 if (rc)
3735 goto rest_fail;
3736 if (zpodd_dev_enabled(dev))
3737 zpodd_post_poweron(dev);
3738 }
3739 }
3740
3741 /* retry flush if necessary */
3742 ata_for_each_dev(dev, link, ALL) {
3743 if (dev->class != ATA_DEV_ATA &&
3744 dev->class != ATA_DEV_ZAC)
3745 continue;
3746 rc = ata_eh_maybe_retry_flush(dev);
3747 if (rc)
3748 goto rest_fail;
3749 }
3750
3751 config_lpm:
3752 /* configure link power saving */
3753 if (link->lpm_policy != ap->target_lpm_policy) {
3754 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3755 if (rc)
3756 goto rest_fail;
3757 }
3758
3759 /* this link is okay now */
3760 ehc->i.flags = 0;
3761 continue;
3762
3763 rest_fail:
3764 nr_fails++;
3765 if (dev)
3766 ata_eh_handle_dev_fail(dev, rc);
3767
3768 if (ata_port_is_frozen(ap)) {
3769 /* PMP reset requires working host port.
3770 * Can't retry if it's frozen.
3771 */
3772 if (sata_pmp_attached(ap))
3773 goto out;
3774 break;
3775 }
3776 }
3777
3778 if (nr_fails)
3779 goto retry;
3780
3781 out:
3782 if (rc && r_failed_link)
3783 *r_failed_link = link;
3784
3785 return rc;
3786}
3787
3788/**
3789 * ata_eh_finish - finish up EH
3790 * @ap: host port to finish EH for
3791 *
3792 * Recovery is complete. Clean up EH states and retry or finish
3793 * failed qcs.
3794 *
3795 * LOCKING:
3796 * None.
3797 */
3798void ata_eh_finish(struct ata_port *ap)
3799{
3800 struct ata_queued_cmd *qc;
3801 int tag;
3802
3803 /* retry or finish qcs */
3804 ata_qc_for_each_raw(ap, qc, tag) {
3805 if (!(qc->flags & ATA_QCFLAG_FAILED))
3806 continue;
3807
3808 if (qc->err_mask) {
3809 /* FIXME: Once EH migration is complete,
3810 * generate sense data in this function,
3811 * considering both err_mask and tf.
3812 */
3813 if (qc->flags & ATA_QCFLAG_RETRY)
3814 ata_eh_qc_retry(qc);
3815 else
3816 ata_eh_qc_complete(qc);
3817 } else {
3818 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3819 ata_eh_qc_complete(qc);
3820 } else {
3821 /* feed zero TF to sense generation */
3822 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3823 ata_eh_qc_retry(qc);
3824 }
3825 }
3826 }
3827
3828 /* make sure nr_active_links is zero after EH */
3829 WARN_ON(ap->nr_active_links);
3830 ap->nr_active_links = 0;
3831}
3832
3833/**
3834 * ata_do_eh - do standard error handling
3835 * @ap: host port to handle error for
3836 *
3837 * @prereset: prereset method (can be NULL)
3838 * @softreset: softreset method (can be NULL)
3839 * @hardreset: hardreset method (can be NULL)
3840 * @postreset: postreset method (can be NULL)
3841 *
3842 * Perform standard error handling sequence.
3843 *
3844 * LOCKING:
3845 * Kernel thread context (may sleep).
3846 */
3847void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3848 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3849 ata_postreset_fn_t postreset)
3850{
3851 struct ata_device *dev;
3852 int rc;
3853
3854 ata_eh_autopsy(ap);
3855 ata_eh_report(ap);
3856
3857 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3858 NULL);
3859 if (rc) {
3860 ata_for_each_dev(dev, &ap->link, ALL)
3861 ata_dev_disable(dev);
3862 }
3863
3864 ata_eh_finish(ap);
3865}
3866
3867/**
3868 * ata_std_error_handler - standard error handler
3869 * @ap: host port to handle error for
3870 *
3871 * Standard error handler
3872 *
3873 * LOCKING:
3874 * Kernel thread context (may sleep).
3875 */
3876void ata_std_error_handler(struct ata_port *ap)
3877{
3878 struct ata_port_operations *ops = ap->ops;
3879 ata_reset_fn_t hardreset = ops->hardreset;
3880
3881 /* ignore built-in hardreset if SCR access is not available */
3882 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3883 hardreset = NULL;
3884
3885 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3886}
3887EXPORT_SYMBOL_GPL(ata_std_error_handler);
3888
3889#ifdef CONFIG_PM
3890/**
3891 * ata_eh_handle_port_suspend - perform port suspend operation
3892 * @ap: port to suspend
3893 *
3894 * Suspend @ap.
3895 *
3896 * LOCKING:
3897 * Kernel thread context (may sleep).
3898 */
3899static void ata_eh_handle_port_suspend(struct ata_port *ap)
3900{
3901 unsigned long flags;
3902 int rc = 0;
3903 struct ata_device *dev;
3904
3905 /* are we suspending? */
3906 spin_lock_irqsave(ap->lock, flags);
3907 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3908 ap->pm_mesg.event & PM_EVENT_RESUME) {
3909 spin_unlock_irqrestore(ap->lock, flags);
3910 return;
3911 }
3912 spin_unlock_irqrestore(ap->lock, flags);
3913
3914 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3915
3916 /*
3917 * If we have a ZPODD attached, check its zero
3918 * power ready status before the port is frozen.
3919 * Only needed for runtime suspend.
3920 */
3921 if (PMSG_IS_AUTO(ap->pm_mesg)) {
3922 ata_for_each_dev(dev, &ap->link, ENABLED) {
3923 if (zpodd_dev_enabled(dev))
3924 zpodd_on_suspend(dev);
3925 }
3926 }
3927
3928 /* suspend */
3929 ata_eh_freeze_port(ap);
3930
3931 if (ap->ops->port_suspend)
3932 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3933
3934 ata_acpi_set_state(ap, ap->pm_mesg);
3935
3936 /* update the flags */
3937 spin_lock_irqsave(ap->lock, flags);
3938
3939 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3940 if (rc == 0)
3941 ap->pflags |= ATA_PFLAG_SUSPENDED;
3942 else if (ata_port_is_frozen(ap))
3943 ata_port_schedule_eh(ap);
3944
3945 spin_unlock_irqrestore(ap->lock, flags);
3946
3947 return;
3948}
3949
3950/**
3951 * ata_eh_handle_port_resume - perform port resume operation
3952 * @ap: port to resume
3953 *
3954 * Resume @ap.
3955 *
3956 * LOCKING:
3957 * Kernel thread context (may sleep).
3958 */
3959static void ata_eh_handle_port_resume(struct ata_port *ap)
3960{
3961 struct ata_link *link;
3962 struct ata_device *dev;
3963 unsigned long flags;
3964
3965 /* are we resuming? */
3966 spin_lock_irqsave(ap->lock, flags);
3967 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3968 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
3969 spin_unlock_irqrestore(ap->lock, flags);
3970 return;
3971 }
3972 spin_unlock_irqrestore(ap->lock, flags);
3973
3974 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3975
3976 /*
3977 * Error timestamps are in jiffies which doesn't run while
3978 * suspended and PHY events during resume isn't too uncommon.
3979 * When the two are combined, it can lead to unnecessary speed
3980 * downs if the machine is suspended and resumed repeatedly.
3981 * Clear error history.
3982 */
3983 ata_for_each_link(link, ap, HOST_FIRST)
3984 ata_for_each_dev(dev, link, ALL)
3985 ata_ering_clear(&dev->ering);
3986
3987 ata_acpi_set_state(ap, ap->pm_mesg);
3988
3989 if (ap->ops->port_resume)
3990 ap->ops->port_resume(ap);
3991
3992 /* tell ACPI that we're resuming */
3993 ata_acpi_on_resume(ap);
3994
3995 /* update the flags */
3996 spin_lock_irqsave(ap->lock, flags);
3997 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3998 spin_unlock_irqrestore(ap->lock, flags);
3999}
4000#endif /* CONFIG_PM */
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * libata-eh.c - libata error handling
4 *
5 * Copyright 2006 Tejun Heo <htejun@gmail.com>
6 *
7 * libata documentation is available via 'make {ps|pdf}docs',
8 * as Documentation/driver-api/libata.rst
9 *
10 * Hardware documentation available from http://www.t13.org/ and
11 * http://www.sata-io.org/
12 */
13
14#include <linux/kernel.h>
15#include <linux/blkdev.h>
16#include <linux/export.h>
17#include <linux/pci.h>
18#include <scsi/scsi.h>
19#include <scsi/scsi_host.h>
20#include <scsi/scsi_eh.h>
21#include <scsi/scsi_device.h>
22#include <scsi/scsi_cmnd.h>
23#include <scsi/scsi_dbg.h>
24#include "../scsi/scsi_transport_api.h"
25
26#include <linux/libata.h>
27
28#include <trace/events/libata.h>
29#include "libata.h"
30
31enum {
32 /* speed down verdicts */
33 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
34 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
35 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
36 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
37
38 /* error flags */
39 ATA_EFLAG_IS_IO = (1 << 0),
40 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
41 ATA_EFLAG_OLD_ER = (1 << 31),
42
43 /* error categories */
44 ATA_ECAT_NONE = 0,
45 ATA_ECAT_ATA_BUS = 1,
46 ATA_ECAT_TOUT_HSM = 2,
47 ATA_ECAT_UNK_DEV = 3,
48 ATA_ECAT_DUBIOUS_NONE = 4,
49 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
50 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
51 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
52 ATA_ECAT_NR = 8,
53
54 ATA_EH_CMD_DFL_TIMEOUT = 5000,
55
56 /* always put at least this amount of time between resets */
57 ATA_EH_RESET_COOL_DOWN = 5000,
58
59 /* Waiting in ->prereset can never be reliable. It's
60 * sometimes nice to wait there but it can't be depended upon;
61 * otherwise, we wouldn't be resetting. Just give it enough
62 * time for most drives to spin up.
63 */
64 ATA_EH_PRERESET_TIMEOUT = 10000,
65 ATA_EH_FASTDRAIN_INTERVAL = 3000,
66
67 ATA_EH_UA_TRIES = 5,
68
69 /* probe speed down parameters, see ata_eh_schedule_probe() */
70 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
71 ATA_EH_PROBE_TRIALS = 2,
72};
73
74/* The following table determines how we sequence resets. Each entry
75 * represents timeout for that try. The first try can be soft or
76 * hardreset. All others are hardreset if available. In most cases
77 * the first reset w/ 10sec timeout should succeed. Following entries
78 * are mostly for error handling, hotplug and those outlier devices that
79 * take an exceptionally long time to recover from reset.
80 */
81static const unsigned int ata_eh_reset_timeouts[] = {
82 10000, /* most drives spin up by 10sec */
83 10000, /* > 99% working drives spin up before 20sec */
84 35000, /* give > 30 secs of idleness for outlier devices */
85 5000, /* and sweet one last chance */
86 UINT_MAX, /* > 1 min has elapsed, give up */
87};
88
89static const unsigned int ata_eh_identify_timeouts[] = {
90 5000, /* covers > 99% of successes and not too boring on failures */
91 10000, /* combined time till here is enough even for media access */
92 30000, /* for true idiots */
93 UINT_MAX,
94};
95
96static const unsigned int ata_eh_revalidate_timeouts[] = {
97 15000, /* Some drives are slow to read log pages when waking-up */
98 15000, /* combined time till here is enough even for media access */
99 UINT_MAX,
100};
101
102static const unsigned int ata_eh_flush_timeouts[] = {
103 15000, /* be generous with flush */
104 15000, /* ditto */
105 30000, /* and even more generous */
106 UINT_MAX,
107};
108
109static const unsigned int ata_eh_other_timeouts[] = {
110 5000, /* same rationale as identify timeout */
111 10000, /* ditto */
112 /* but no merciful 30sec for other commands, it just isn't worth it */
113 UINT_MAX,
114};
115
116struct ata_eh_cmd_timeout_ent {
117 const u8 *commands;
118 const unsigned int *timeouts;
119};
120
121/* The following table determines timeouts to use for EH internal
122 * commands. Each table entry is a command class and matches the
123 * commands the entry applies to and the timeout table to use.
124 *
125 * On the retry after a command timed out, the next timeout value from
126 * the table is used. If the table doesn't contain further entries,
127 * the last value is used.
128 *
129 * ehc->cmd_timeout_idx keeps track of which timeout to use per
130 * command class, so if SET_FEATURES times out on the first try, the
131 * next try will use the second timeout value only for that class.
132 */
133#define CMDS(cmds...) (const u8 []){ cmds, 0 }
134static const struct ata_eh_cmd_timeout_ent
135ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
136 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
137 .timeouts = ata_eh_identify_timeouts, },
138 { .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT),
139 .timeouts = ata_eh_revalidate_timeouts, },
140 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
141 .timeouts = ata_eh_other_timeouts, },
142 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_SET_FEATURES),
145 .timeouts = ata_eh_other_timeouts, },
146 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
149 .timeouts = ata_eh_flush_timeouts },
150 { .commands = CMDS(ATA_CMD_VERIFY),
151 .timeouts = ata_eh_reset_timeouts },
152};
153#undef CMDS
154
155static void __ata_port_freeze(struct ata_port *ap);
156static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
157 struct ata_device **r_failed_dev);
158#ifdef CONFIG_PM
159static void ata_eh_handle_port_suspend(struct ata_port *ap);
160static void ata_eh_handle_port_resume(struct ata_port *ap);
161#else /* CONFIG_PM */
162static void ata_eh_handle_port_suspend(struct ata_port *ap)
163{ }
164
165static void ata_eh_handle_port_resume(struct ata_port *ap)
166{ }
167#endif /* CONFIG_PM */
168
169static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
170 const char *fmt, va_list args)
171{
172 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
173 ATA_EH_DESC_LEN - ehi->desc_len,
174 fmt, args);
175}
176
177/**
178 * __ata_ehi_push_desc - push error description without adding separator
179 * @ehi: target EHI
180 * @fmt: printf format string
181 *
182 * Format string according to @fmt and append it to @ehi->desc.
183 *
184 * LOCKING:
185 * spin_lock_irqsave(host lock)
186 */
187void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
188{
189 va_list args;
190
191 va_start(args, fmt);
192 __ata_ehi_pushv_desc(ehi, fmt, args);
193 va_end(args);
194}
195EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
196
197/**
198 * ata_ehi_push_desc - push error description with separator
199 * @ehi: target EHI
200 * @fmt: printf format string
201 *
202 * Format string according to @fmt and append it to @ehi->desc.
203 * If @ehi->desc is not empty, ", " is added in-between.
204 *
205 * LOCKING:
206 * spin_lock_irqsave(host lock)
207 */
208void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
209{
210 va_list args;
211
212 if (ehi->desc_len)
213 __ata_ehi_push_desc(ehi, ", ");
214
215 va_start(args, fmt);
216 __ata_ehi_pushv_desc(ehi, fmt, args);
217 va_end(args);
218}
219EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
220
221/**
222 * ata_ehi_clear_desc - clean error description
223 * @ehi: target EHI
224 *
225 * Clear @ehi->desc.
226 *
227 * LOCKING:
228 * spin_lock_irqsave(host lock)
229 */
230void ata_ehi_clear_desc(struct ata_eh_info *ehi)
231{
232 ehi->desc[0] = '\0';
233 ehi->desc_len = 0;
234}
235EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
236
237/**
238 * ata_port_desc - append port description
239 * @ap: target ATA port
240 * @fmt: printf format string
241 *
242 * Format string according to @fmt and append it to port
243 * description. If port description is not empty, " " is added
244 * in-between. This function is to be used while initializing
245 * ata_host. The description is printed on host registration.
246 *
247 * LOCKING:
248 * None.
249 */
250void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
251{
252 va_list args;
253
254 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
255
256 if (ap->link.eh_info.desc_len)
257 __ata_ehi_push_desc(&ap->link.eh_info, " ");
258
259 va_start(args, fmt);
260 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
261 va_end(args);
262}
263EXPORT_SYMBOL_GPL(ata_port_desc);
264
265#ifdef CONFIG_PCI
266/**
267 * ata_port_pbar_desc - append PCI BAR description
268 * @ap: target ATA port
269 * @bar: target PCI BAR
270 * @offset: offset into PCI BAR
271 * @name: name of the area
272 *
273 * If @offset is negative, this function formats a string which
274 * contains the name, address, size and type of the BAR and
275 * appends it to the port description. If @offset is zero or
276 * positive, only name and offsetted address is appended.
277 *
278 * LOCKING:
279 * None.
280 */
281void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
282 const char *name)
283{
284 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
285 char *type = "";
286 unsigned long long start, len;
287
288 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
289 type = "m";
290 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
291 type = "i";
292
293 start = (unsigned long long)pci_resource_start(pdev, bar);
294 len = (unsigned long long)pci_resource_len(pdev, bar);
295
296 if (offset < 0)
297 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
298 else
299 ata_port_desc(ap, "%s 0x%llx", name,
300 start + (unsigned long long)offset);
301}
302EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
303#endif /* CONFIG_PCI */
304
305static int ata_lookup_timeout_table(u8 cmd)
306{
307 int i;
308
309 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
310 const u8 *cur;
311
312 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
313 if (*cur == cmd)
314 return i;
315 }
316
317 return -1;
318}
319
320/**
321 * ata_internal_cmd_timeout - determine timeout for an internal command
322 * @dev: target device
323 * @cmd: internal command to be issued
324 *
325 * Determine timeout for internal command @cmd for @dev.
326 *
327 * LOCKING:
328 * EH context.
329 *
330 * RETURNS:
331 * Determined timeout.
332 */
333unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
334{
335 struct ata_eh_context *ehc = &dev->link->eh_context;
336 int ent = ata_lookup_timeout_table(cmd);
337 int idx;
338
339 if (ent < 0)
340 return ATA_EH_CMD_DFL_TIMEOUT;
341
342 idx = ehc->cmd_timeout_idx[dev->devno][ent];
343 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
344}
345
346/**
347 * ata_internal_cmd_timed_out - notification for internal command timeout
348 * @dev: target device
349 * @cmd: internal command which timed out
350 *
351 * Notify EH that internal command @cmd for @dev timed out. This
352 * function should be called only for commands whose timeouts are
353 * determined using ata_internal_cmd_timeout().
354 *
355 * LOCKING:
356 * EH context.
357 */
358void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
359{
360 struct ata_eh_context *ehc = &dev->link->eh_context;
361 int ent = ata_lookup_timeout_table(cmd);
362 int idx;
363
364 if (ent < 0)
365 return;
366
367 idx = ehc->cmd_timeout_idx[dev->devno][ent];
368 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
369 ehc->cmd_timeout_idx[dev->devno][ent]++;
370}
371
372static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
373 unsigned int err_mask)
374{
375 struct ata_ering_entry *ent;
376
377 WARN_ON(!err_mask);
378
379 ering->cursor++;
380 ering->cursor %= ATA_ERING_SIZE;
381
382 ent = &ering->ring[ering->cursor];
383 ent->eflags = eflags;
384 ent->err_mask = err_mask;
385 ent->timestamp = get_jiffies_64();
386}
387
388static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
389{
390 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
391
392 if (ent->err_mask)
393 return ent;
394 return NULL;
395}
396
397int ata_ering_map(struct ata_ering *ering,
398 int (*map_fn)(struct ata_ering_entry *, void *),
399 void *arg)
400{
401 int idx, rc = 0;
402 struct ata_ering_entry *ent;
403
404 idx = ering->cursor;
405 do {
406 ent = &ering->ring[idx];
407 if (!ent->err_mask)
408 break;
409 rc = map_fn(ent, arg);
410 if (rc)
411 break;
412 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
413 } while (idx != ering->cursor);
414
415 return rc;
416}
417
418static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
419{
420 ent->eflags |= ATA_EFLAG_OLD_ER;
421 return 0;
422}
423
424static void ata_ering_clear(struct ata_ering *ering)
425{
426 ata_ering_map(ering, ata_ering_clear_cb, NULL);
427}
428
429static unsigned int ata_eh_dev_action(struct ata_device *dev)
430{
431 struct ata_eh_context *ehc = &dev->link->eh_context;
432
433 return ehc->i.action | ehc->i.dev_action[dev->devno];
434}
435
436static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
437 struct ata_eh_info *ehi, unsigned int action)
438{
439 struct ata_device *tdev;
440
441 if (!dev) {
442 ehi->action &= ~action;
443 ata_for_each_dev(tdev, link, ALL)
444 ehi->dev_action[tdev->devno] &= ~action;
445 } else {
446 /* doesn't make sense for port-wide EH actions */
447 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
448
449 /* break ehi->action into ehi->dev_action */
450 if (ehi->action & action) {
451 ata_for_each_dev(tdev, link, ALL)
452 ehi->dev_action[tdev->devno] |=
453 ehi->action & action;
454 ehi->action &= ~action;
455 }
456
457 /* turn off the specified per-dev action */
458 ehi->dev_action[dev->devno] &= ~action;
459 }
460}
461
462/**
463 * ata_eh_acquire - acquire EH ownership
464 * @ap: ATA port to acquire EH ownership for
465 *
466 * Acquire EH ownership for @ap. This is the basic exclusion
467 * mechanism for ports sharing a host. Only one port hanging off
468 * the same host can claim the ownership of EH.
469 *
470 * LOCKING:
471 * EH context.
472 */
473void ata_eh_acquire(struct ata_port *ap)
474{
475 mutex_lock(&ap->host->eh_mutex);
476 WARN_ON_ONCE(ap->host->eh_owner);
477 ap->host->eh_owner = current;
478}
479
480/**
481 * ata_eh_release - release EH ownership
482 * @ap: ATA port to release EH ownership for
483 *
484 * Release EH ownership for @ap if the caller. The caller must
485 * have acquired EH ownership using ata_eh_acquire() previously.
486 *
487 * LOCKING:
488 * EH context.
489 */
490void ata_eh_release(struct ata_port *ap)
491{
492 WARN_ON_ONCE(ap->host->eh_owner != current);
493 ap->host->eh_owner = NULL;
494 mutex_unlock(&ap->host->eh_mutex);
495}
496
497static void ata_eh_dev_disable(struct ata_device *dev)
498{
499 ata_acpi_on_disable(dev);
500 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
501 dev->class++;
502
503 /* From now till the next successful probe, ering is used to
504 * track probe failures. Clear accumulated device error info.
505 */
506 ata_ering_clear(&dev->ering);
507}
508
509static void ata_eh_unload(struct ata_port *ap)
510{
511 struct ata_link *link;
512 struct ata_device *dev;
513 unsigned long flags;
514
515 /*
516 * Unless we are restarting, transition all enabled devices to
517 * standby power mode.
518 */
519 if (system_state != SYSTEM_RESTART) {
520 ata_for_each_link(link, ap, PMP_FIRST) {
521 ata_for_each_dev(dev, link, ENABLED)
522 ata_dev_power_set_standby(dev);
523 }
524 }
525
526 /*
527 * Restore SControl IPM and SPD for the next driver and
528 * disable attached devices.
529 */
530 ata_for_each_link(link, ap, PMP_FIRST) {
531 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
532 ata_for_each_dev(dev, link, ENABLED)
533 ata_eh_dev_disable(dev);
534 }
535
536 /* freeze and set UNLOADED */
537 spin_lock_irqsave(ap->lock, flags);
538
539 ata_port_freeze(ap); /* won't be thawed */
540 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
541 ap->pflags |= ATA_PFLAG_UNLOADED;
542
543 spin_unlock_irqrestore(ap->lock, flags);
544}
545
546/**
547 * ata_scsi_error - SCSI layer error handler callback
548 * @host: SCSI host on which error occurred
549 *
550 * Handles SCSI-layer-thrown error events.
551 *
552 * LOCKING:
553 * Inherited from SCSI layer (none, can sleep)
554 *
555 * RETURNS:
556 * Zero.
557 */
558void ata_scsi_error(struct Scsi_Host *host)
559{
560 struct ata_port *ap = ata_shost_to_port(host);
561 unsigned long flags;
562 LIST_HEAD(eh_work_q);
563
564 spin_lock_irqsave(host->host_lock, flags);
565 list_splice_init(&host->eh_cmd_q, &eh_work_q);
566 spin_unlock_irqrestore(host->host_lock, flags);
567
568 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
569
570 /* If we timed raced normal completion and there is nothing to
571 recover nr_timedout == 0 why exactly are we doing error recovery ? */
572 ata_scsi_port_error_handler(host, ap);
573
574 /* finish or retry handled scmd's and clean up */
575 WARN_ON(!list_empty(&eh_work_q));
576
577}
578
579/**
580 * ata_scsi_cmd_error_handler - error callback for a list of commands
581 * @host: scsi host containing the port
582 * @ap: ATA port within the host
583 * @eh_work_q: list of commands to process
584 *
585 * process the given list of commands and return those finished to the
586 * ap->eh_done_q. This function is the first part of the libata error
587 * handler which processes a given list of failed commands.
588 */
589void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
590 struct list_head *eh_work_q)
591{
592 int i;
593 unsigned long flags;
594 struct scsi_cmnd *scmd, *tmp;
595 int nr_timedout = 0;
596
597 /* make sure sff pio task is not running */
598 ata_sff_flush_pio_task(ap);
599
600 /* synchronize with host lock and sort out timeouts */
601
602 /*
603 * For EH, all qcs are finished in one of three ways -
604 * normal completion, error completion, and SCSI timeout.
605 * Both completions can race against SCSI timeout. When normal
606 * completion wins, the qc never reaches EH. When error
607 * completion wins, the qc has ATA_QCFLAG_EH set.
608 *
609 * When SCSI timeout wins, things are a bit more complex.
610 * Normal or error completion can occur after the timeout but
611 * before this point. In such cases, both types of
612 * completions are honored. A scmd is determined to have
613 * timed out iff its associated qc is active and not failed.
614 */
615 spin_lock_irqsave(ap->lock, flags);
616
617 /*
618 * This must occur under the ap->lock as we don't want
619 * a polled recovery to race the real interrupt handler
620 *
621 * The lost_interrupt handler checks for any completed but
622 * non-notified command and completes much like an IRQ handler.
623 *
624 * We then fall into the error recovery code which will treat
625 * this as if normal completion won the race
626 */
627 if (ap->ops->lost_interrupt)
628 ap->ops->lost_interrupt(ap);
629
630 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
631 struct ata_queued_cmd *qc;
632
633 ata_qc_for_each_raw(ap, qc, i) {
634 if (qc->flags & ATA_QCFLAG_ACTIVE &&
635 qc->scsicmd == scmd)
636 break;
637 }
638
639 if (i < ATA_MAX_QUEUE) {
640 /* the scmd has an associated qc */
641 if (!(qc->flags & ATA_QCFLAG_EH)) {
642 /* which hasn't failed yet, timeout */
643 qc->err_mask |= AC_ERR_TIMEOUT;
644 qc->flags |= ATA_QCFLAG_EH;
645 nr_timedout++;
646 }
647 } else {
648 /* Normal completion occurred after
649 * SCSI timeout but before this point.
650 * Successfully complete it.
651 */
652 scmd->retries = scmd->allowed;
653 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
654 }
655 }
656
657 /*
658 * If we have timed out qcs. They belong to EH from
659 * this point but the state of the controller is
660 * unknown. Freeze the port to make sure the IRQ
661 * handler doesn't diddle with those qcs. This must
662 * be done atomically w.r.t. setting ATA_QCFLAG_EH.
663 */
664 if (nr_timedout)
665 __ata_port_freeze(ap);
666
667 /* initialize eh_tries */
668 ap->eh_tries = ATA_EH_MAX_TRIES;
669
670 spin_unlock_irqrestore(ap->lock, flags);
671}
672EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
673
674/**
675 * ata_scsi_port_error_handler - recover the port after the commands
676 * @host: SCSI host containing the port
677 * @ap: the ATA port
678 *
679 * Handle the recovery of the port @ap after all the commands
680 * have been recovered.
681 */
682void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
683{
684 unsigned long flags;
685 struct ata_link *link;
686
687 /* acquire EH ownership */
688 ata_eh_acquire(ap);
689 repeat:
690 /* kill fast drain timer */
691 del_timer_sync(&ap->fastdrain_timer);
692
693 /* process port resume request */
694 ata_eh_handle_port_resume(ap);
695
696 /* fetch & clear EH info */
697 spin_lock_irqsave(ap->lock, flags);
698
699 ata_for_each_link(link, ap, HOST_FIRST) {
700 struct ata_eh_context *ehc = &link->eh_context;
701 struct ata_device *dev;
702
703 memset(&link->eh_context, 0, sizeof(link->eh_context));
704 link->eh_context.i = link->eh_info;
705 memset(&link->eh_info, 0, sizeof(link->eh_info));
706
707 ata_for_each_dev(dev, link, ENABLED) {
708 int devno = dev->devno;
709
710 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
711 if (ata_ncq_enabled(dev))
712 ehc->saved_ncq_enabled |= 1 << devno;
713
714 /* If we are resuming, wake up the device */
715 if (ap->pflags & ATA_PFLAG_RESUMING)
716 ehc->i.dev_action[devno] |= ATA_EH_SET_ACTIVE;
717 }
718 }
719
720 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
721 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
722 ap->excl_link = NULL; /* don't maintain exclusion over EH */
723
724 spin_unlock_irqrestore(ap->lock, flags);
725
726 /* invoke EH, skip if unloading or suspended */
727 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
728 ap->ops->error_handler(ap);
729 else {
730 /* if unloading, commence suicide */
731 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
732 !(ap->pflags & ATA_PFLAG_UNLOADED))
733 ata_eh_unload(ap);
734 ata_eh_finish(ap);
735 }
736
737 /* process port suspend request */
738 ata_eh_handle_port_suspend(ap);
739
740 /*
741 * Exception might have happened after ->error_handler recovered the
742 * port but before this point. Repeat EH in such case.
743 */
744 spin_lock_irqsave(ap->lock, flags);
745
746 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
747 if (--ap->eh_tries) {
748 spin_unlock_irqrestore(ap->lock, flags);
749 goto repeat;
750 }
751 ata_port_err(ap,
752 "EH pending after %d tries, giving up\n",
753 ATA_EH_MAX_TRIES);
754 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
755 }
756
757 /* this run is complete, make sure EH info is clear */
758 ata_for_each_link(link, ap, HOST_FIRST)
759 memset(&link->eh_info, 0, sizeof(link->eh_info));
760
761 /*
762 * end eh (clear host_eh_scheduled) while holding ap->lock such that if
763 * exception occurs after this point but before EH completion, SCSI
764 * midlayer will re-initiate EH.
765 */
766 ap->ops->end_eh(ap);
767
768 spin_unlock_irqrestore(ap->lock, flags);
769 ata_eh_release(ap);
770
771 scsi_eh_flush_done_q(&ap->eh_done_q);
772
773 /* clean up */
774 spin_lock_irqsave(ap->lock, flags);
775
776 ap->pflags &= ~ATA_PFLAG_RESUMING;
777
778 if (ap->pflags & ATA_PFLAG_LOADING)
779 ap->pflags &= ~ATA_PFLAG_LOADING;
780 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
781 !(ap->flags & ATA_FLAG_SAS_HOST))
782 schedule_delayed_work(&ap->hotplug_task, 0);
783
784 if (ap->pflags & ATA_PFLAG_RECOVERED)
785 ata_port_info(ap, "EH complete\n");
786
787 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
788
789 /* tell wait_eh that we're done */
790 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
791 wake_up_all(&ap->eh_wait_q);
792
793 spin_unlock_irqrestore(ap->lock, flags);
794}
795EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
796
797/**
798 * ata_port_wait_eh - Wait for the currently pending EH to complete
799 * @ap: Port to wait EH for
800 *
801 * Wait until the currently pending EH is complete.
802 *
803 * LOCKING:
804 * Kernel thread context (may sleep).
805 */
806void ata_port_wait_eh(struct ata_port *ap)
807{
808 unsigned long flags;
809 DEFINE_WAIT(wait);
810
811 retry:
812 spin_lock_irqsave(ap->lock, flags);
813
814 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
815 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
816 spin_unlock_irqrestore(ap->lock, flags);
817 schedule();
818 spin_lock_irqsave(ap->lock, flags);
819 }
820 finish_wait(&ap->eh_wait_q, &wait);
821
822 spin_unlock_irqrestore(ap->lock, flags);
823
824 /* make sure SCSI EH is complete */
825 if (scsi_host_in_recovery(ap->scsi_host)) {
826 ata_msleep(ap, 10);
827 goto retry;
828 }
829}
830EXPORT_SYMBOL_GPL(ata_port_wait_eh);
831
832static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
833{
834 struct ata_queued_cmd *qc;
835 unsigned int tag;
836 unsigned int nr = 0;
837
838 /* count only non-internal commands */
839 ata_qc_for_each(ap, qc, tag) {
840 if (qc)
841 nr++;
842 }
843
844 return nr;
845}
846
847void ata_eh_fastdrain_timerfn(struct timer_list *t)
848{
849 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
850 unsigned long flags;
851 unsigned int cnt;
852
853 spin_lock_irqsave(ap->lock, flags);
854
855 cnt = ata_eh_nr_in_flight(ap);
856
857 /* are we done? */
858 if (!cnt)
859 goto out_unlock;
860
861 if (cnt == ap->fastdrain_cnt) {
862 struct ata_queued_cmd *qc;
863 unsigned int tag;
864
865 /* No progress during the last interval, tag all
866 * in-flight qcs as timed out and freeze the port.
867 */
868 ata_qc_for_each(ap, qc, tag) {
869 if (qc)
870 qc->err_mask |= AC_ERR_TIMEOUT;
871 }
872
873 ata_port_freeze(ap);
874 } else {
875 /* some qcs have finished, give it another chance */
876 ap->fastdrain_cnt = cnt;
877 ap->fastdrain_timer.expires =
878 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
879 add_timer(&ap->fastdrain_timer);
880 }
881
882 out_unlock:
883 spin_unlock_irqrestore(ap->lock, flags);
884}
885
886/**
887 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
888 * @ap: target ATA port
889 * @fastdrain: activate fast drain
890 *
891 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
892 * is non-zero and EH wasn't pending before. Fast drain ensures
893 * that EH kicks in in timely manner.
894 *
895 * LOCKING:
896 * spin_lock_irqsave(host lock)
897 */
898static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
899{
900 unsigned int cnt;
901
902 /* already scheduled? */
903 if (ap->pflags & ATA_PFLAG_EH_PENDING)
904 return;
905
906 ap->pflags |= ATA_PFLAG_EH_PENDING;
907
908 if (!fastdrain)
909 return;
910
911 /* do we have in-flight qcs? */
912 cnt = ata_eh_nr_in_flight(ap);
913 if (!cnt)
914 return;
915
916 /* activate fast drain */
917 ap->fastdrain_cnt = cnt;
918 ap->fastdrain_timer.expires =
919 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
920 add_timer(&ap->fastdrain_timer);
921}
922
923/**
924 * ata_qc_schedule_eh - schedule qc for error handling
925 * @qc: command to schedule error handling for
926 *
927 * Schedule error handling for @qc. EH will kick in as soon as
928 * other commands are drained.
929 *
930 * LOCKING:
931 * spin_lock_irqsave(host lock)
932 */
933void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
934{
935 struct ata_port *ap = qc->ap;
936
937 qc->flags |= ATA_QCFLAG_EH;
938 ata_eh_set_pending(ap, 1);
939
940 /* The following will fail if timeout has already expired.
941 * ata_scsi_error() takes care of such scmds on EH entry.
942 * Note that ATA_QCFLAG_EH is unconditionally set after
943 * this function completes.
944 */
945 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
946}
947
948/**
949 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
950 * @ap: ATA port to schedule EH for
951 *
952 * LOCKING: inherited from ata_port_schedule_eh
953 * spin_lock_irqsave(host lock)
954 */
955void ata_std_sched_eh(struct ata_port *ap)
956{
957 if (ap->pflags & ATA_PFLAG_INITIALIZING)
958 return;
959
960 ata_eh_set_pending(ap, 1);
961 scsi_schedule_eh(ap->scsi_host);
962
963 trace_ata_std_sched_eh(ap);
964}
965EXPORT_SYMBOL_GPL(ata_std_sched_eh);
966
967/**
968 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
969 * @ap: ATA port to end EH for
970 *
971 * In the libata object model there is a 1:1 mapping of ata_port to
972 * shost, so host fields can be directly manipulated under ap->lock, in
973 * the libsas case we need to hold a lock at the ha->level to coordinate
974 * these events.
975 *
976 * LOCKING:
977 * spin_lock_irqsave(host lock)
978 */
979void ata_std_end_eh(struct ata_port *ap)
980{
981 struct Scsi_Host *host = ap->scsi_host;
982
983 host->host_eh_scheduled = 0;
984}
985EXPORT_SYMBOL(ata_std_end_eh);
986
987
988/**
989 * ata_port_schedule_eh - schedule error handling without a qc
990 * @ap: ATA port to schedule EH for
991 *
992 * Schedule error handling for @ap. EH will kick in as soon as
993 * all commands are drained.
994 *
995 * LOCKING:
996 * spin_lock_irqsave(host lock)
997 */
998void ata_port_schedule_eh(struct ata_port *ap)
999{
1000 /* see: ata_std_sched_eh, unless you know better */
1001 ap->ops->sched_eh(ap);
1002}
1003EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
1004
1005static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1006{
1007 struct ata_queued_cmd *qc;
1008 int tag, nr_aborted = 0;
1009
1010 /* we're gonna abort all commands, no need for fast drain */
1011 ata_eh_set_pending(ap, 0);
1012
1013 /* include internal tag in iteration */
1014 ata_qc_for_each_with_internal(ap, qc, tag) {
1015 if (qc && (!link || qc->dev->link == link)) {
1016 qc->flags |= ATA_QCFLAG_EH;
1017 ata_qc_complete(qc);
1018 nr_aborted++;
1019 }
1020 }
1021
1022 if (!nr_aborted)
1023 ata_port_schedule_eh(ap);
1024
1025 return nr_aborted;
1026}
1027
1028/**
1029 * ata_link_abort - abort all qc's on the link
1030 * @link: ATA link to abort qc's for
1031 *
1032 * Abort all active qc's active on @link and schedule EH.
1033 *
1034 * LOCKING:
1035 * spin_lock_irqsave(host lock)
1036 *
1037 * RETURNS:
1038 * Number of aborted qc's.
1039 */
1040int ata_link_abort(struct ata_link *link)
1041{
1042 return ata_do_link_abort(link->ap, link);
1043}
1044EXPORT_SYMBOL_GPL(ata_link_abort);
1045
1046/**
1047 * ata_port_abort - abort all qc's on the port
1048 * @ap: ATA port to abort qc's for
1049 *
1050 * Abort all active qc's of @ap and schedule EH.
1051 *
1052 * LOCKING:
1053 * spin_lock_irqsave(host_set lock)
1054 *
1055 * RETURNS:
1056 * Number of aborted qc's.
1057 */
1058int ata_port_abort(struct ata_port *ap)
1059{
1060 return ata_do_link_abort(ap, NULL);
1061}
1062EXPORT_SYMBOL_GPL(ata_port_abort);
1063
1064/**
1065 * __ata_port_freeze - freeze port
1066 * @ap: ATA port to freeze
1067 *
1068 * This function is called when HSM violation or some other
1069 * condition disrupts normal operation of the port. Frozen port
1070 * is not allowed to perform any operation until the port is
1071 * thawed, which usually follows a successful reset.
1072 *
1073 * ap->ops->freeze() callback can be used for freezing the port
1074 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1075 * port cannot be frozen hardware-wise, the interrupt handler
1076 * must ack and clear interrupts unconditionally while the port
1077 * is frozen.
1078 *
1079 * LOCKING:
1080 * spin_lock_irqsave(host lock)
1081 */
1082static void __ata_port_freeze(struct ata_port *ap)
1083{
1084 if (ap->ops->freeze)
1085 ap->ops->freeze(ap);
1086
1087 ap->pflags |= ATA_PFLAG_FROZEN;
1088
1089 trace_ata_port_freeze(ap);
1090}
1091
1092/**
1093 * ata_port_freeze - abort & freeze port
1094 * @ap: ATA port to freeze
1095 *
1096 * Abort and freeze @ap. The freeze operation must be called
1097 * first, because some hardware requires special operations
1098 * before the taskfile registers are accessible.
1099 *
1100 * LOCKING:
1101 * spin_lock_irqsave(host lock)
1102 *
1103 * RETURNS:
1104 * Number of aborted commands.
1105 */
1106int ata_port_freeze(struct ata_port *ap)
1107{
1108 __ata_port_freeze(ap);
1109
1110 return ata_port_abort(ap);
1111}
1112EXPORT_SYMBOL_GPL(ata_port_freeze);
1113
1114/**
1115 * ata_eh_freeze_port - EH helper to freeze port
1116 * @ap: ATA port to freeze
1117 *
1118 * Freeze @ap.
1119 *
1120 * LOCKING:
1121 * None.
1122 */
1123void ata_eh_freeze_port(struct ata_port *ap)
1124{
1125 unsigned long flags;
1126
1127 spin_lock_irqsave(ap->lock, flags);
1128 __ata_port_freeze(ap);
1129 spin_unlock_irqrestore(ap->lock, flags);
1130}
1131EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1132
1133/**
1134 * ata_eh_thaw_port - EH helper to thaw port
1135 * @ap: ATA port to thaw
1136 *
1137 * Thaw frozen port @ap.
1138 *
1139 * LOCKING:
1140 * None.
1141 */
1142void ata_eh_thaw_port(struct ata_port *ap)
1143{
1144 unsigned long flags;
1145
1146 spin_lock_irqsave(ap->lock, flags);
1147
1148 ap->pflags &= ~ATA_PFLAG_FROZEN;
1149
1150 if (ap->ops->thaw)
1151 ap->ops->thaw(ap);
1152
1153 spin_unlock_irqrestore(ap->lock, flags);
1154
1155 trace_ata_port_thaw(ap);
1156}
1157
1158static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1159{
1160 /* nada */
1161}
1162
1163static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1164{
1165 struct ata_port *ap = qc->ap;
1166 struct scsi_cmnd *scmd = qc->scsicmd;
1167 unsigned long flags;
1168
1169 spin_lock_irqsave(ap->lock, flags);
1170 qc->scsidone = ata_eh_scsidone;
1171 __ata_qc_complete(qc);
1172 WARN_ON(ata_tag_valid(qc->tag));
1173 spin_unlock_irqrestore(ap->lock, flags);
1174
1175 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1176}
1177
1178/**
1179 * ata_eh_qc_complete - Complete an active ATA command from EH
1180 * @qc: Command to complete
1181 *
1182 * Indicate to the mid and upper layers that an ATA command has
1183 * completed. To be used from EH.
1184 */
1185void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1186{
1187 struct scsi_cmnd *scmd = qc->scsicmd;
1188 scmd->retries = scmd->allowed;
1189 __ata_eh_qc_complete(qc);
1190}
1191
1192/**
1193 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1194 * @qc: Command to retry
1195 *
1196 * Indicate to the mid and upper layers that an ATA command
1197 * should be retried. To be used from EH.
1198 *
1199 * SCSI midlayer limits the number of retries to scmd->allowed.
1200 * scmd->allowed is incremented for commands which get retried
1201 * due to unrelated failures (qc->err_mask is zero).
1202 */
1203void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1204{
1205 struct scsi_cmnd *scmd = qc->scsicmd;
1206 if (!qc->err_mask)
1207 scmd->allowed++;
1208 __ata_eh_qc_complete(qc);
1209}
1210
1211/**
1212 * ata_dev_disable - disable ATA device
1213 * @dev: ATA device to disable
1214 *
1215 * Disable @dev.
1216 *
1217 * Locking:
1218 * EH context.
1219 */
1220void ata_dev_disable(struct ata_device *dev)
1221{
1222 if (!ata_dev_enabled(dev))
1223 return;
1224
1225 ata_dev_warn(dev, "disable device\n");
1226
1227 ata_eh_dev_disable(dev);
1228}
1229EXPORT_SYMBOL_GPL(ata_dev_disable);
1230
1231/**
1232 * ata_eh_detach_dev - detach ATA device
1233 * @dev: ATA device to detach
1234 *
1235 * Detach @dev.
1236 *
1237 * LOCKING:
1238 * None.
1239 */
1240void ata_eh_detach_dev(struct ata_device *dev)
1241{
1242 struct ata_link *link = dev->link;
1243 struct ata_port *ap = link->ap;
1244 struct ata_eh_context *ehc = &link->eh_context;
1245 unsigned long flags;
1246
1247 /*
1248 * If the device is still enabled, transition it to standby power mode
1249 * (i.e. spin down HDDs) and disable it.
1250 */
1251 if (ata_dev_enabled(dev)) {
1252 ata_dev_power_set_standby(dev);
1253 ata_eh_dev_disable(dev);
1254 }
1255
1256 spin_lock_irqsave(ap->lock, flags);
1257
1258 dev->flags &= ~ATA_DFLAG_DETACH;
1259
1260 if (ata_scsi_offline_dev(dev)) {
1261 dev->flags |= ATA_DFLAG_DETACHED;
1262 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1263 }
1264
1265 /* clear per-dev EH info */
1266 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1267 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1268 ehc->saved_xfer_mode[dev->devno] = 0;
1269 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1270
1271 spin_unlock_irqrestore(ap->lock, flags);
1272}
1273
1274/**
1275 * ata_eh_about_to_do - about to perform eh_action
1276 * @link: target ATA link
1277 * @dev: target ATA dev for per-dev action (can be NULL)
1278 * @action: action about to be performed
1279 *
1280 * Called just before performing EH actions to clear related bits
1281 * in @link->eh_info such that eh actions are not unnecessarily
1282 * repeated.
1283 *
1284 * LOCKING:
1285 * None.
1286 */
1287void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1288 unsigned int action)
1289{
1290 struct ata_port *ap = link->ap;
1291 struct ata_eh_info *ehi = &link->eh_info;
1292 struct ata_eh_context *ehc = &link->eh_context;
1293 unsigned long flags;
1294
1295 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1296
1297 spin_lock_irqsave(ap->lock, flags);
1298
1299 ata_eh_clear_action(link, dev, ehi, action);
1300
1301 /* About to take EH action, set RECOVERED. Ignore actions on
1302 * slave links as master will do them again.
1303 */
1304 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1305 ap->pflags |= ATA_PFLAG_RECOVERED;
1306
1307 spin_unlock_irqrestore(ap->lock, flags);
1308}
1309
1310/**
1311 * ata_eh_done - EH action complete
1312 * @link: ATA link for which EH actions are complete
1313 * @dev: target ATA dev for per-dev action (can be NULL)
1314 * @action: action just completed
1315 *
1316 * Called right after performing EH actions to clear related bits
1317 * in @link->eh_context.
1318 *
1319 * LOCKING:
1320 * None.
1321 */
1322void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1323 unsigned int action)
1324{
1325 struct ata_eh_context *ehc = &link->eh_context;
1326
1327 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1328
1329 ata_eh_clear_action(link, dev, &ehc->i, action);
1330}
1331
1332/**
1333 * ata_err_string - convert err_mask to descriptive string
1334 * @err_mask: error mask to convert to string
1335 *
1336 * Convert @err_mask to descriptive string. Errors are
1337 * prioritized according to severity and only the most severe
1338 * error is reported.
1339 *
1340 * LOCKING:
1341 * None.
1342 *
1343 * RETURNS:
1344 * Descriptive string for @err_mask
1345 */
1346static const char *ata_err_string(unsigned int err_mask)
1347{
1348 if (err_mask & AC_ERR_HOST_BUS)
1349 return "host bus error";
1350 if (err_mask & AC_ERR_ATA_BUS)
1351 return "ATA bus error";
1352 if (err_mask & AC_ERR_TIMEOUT)
1353 return "timeout";
1354 if (err_mask & AC_ERR_HSM)
1355 return "HSM violation";
1356 if (err_mask & AC_ERR_SYSTEM)
1357 return "internal error";
1358 if (err_mask & AC_ERR_MEDIA)
1359 return "media error";
1360 if (err_mask & AC_ERR_INVALID)
1361 return "invalid argument";
1362 if (err_mask & AC_ERR_DEV)
1363 return "device error";
1364 if (err_mask & AC_ERR_NCQ)
1365 return "NCQ error";
1366 if (err_mask & AC_ERR_NODEV_HINT)
1367 return "Polling detection error";
1368 return "unknown error";
1369}
1370
1371/**
1372 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1373 * @dev: target ATAPI device
1374 * @r_sense_key: out parameter for sense_key
1375 *
1376 * Perform ATAPI TEST_UNIT_READY.
1377 *
1378 * LOCKING:
1379 * EH context (may sleep).
1380 *
1381 * RETURNS:
1382 * 0 on success, AC_ERR_* mask on failure.
1383 */
1384unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1385{
1386 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1387 struct ata_taskfile tf;
1388 unsigned int err_mask;
1389
1390 ata_tf_init(dev, &tf);
1391
1392 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1393 tf.command = ATA_CMD_PACKET;
1394 tf.protocol = ATAPI_PROT_NODATA;
1395
1396 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1397 if (err_mask == AC_ERR_DEV)
1398 *r_sense_key = tf.error >> 4;
1399 return err_mask;
1400}
1401
1402/**
1403 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1404 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1405 *
1406 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1407 * SENSE. This function is an EH helper.
1408 *
1409 * LOCKING:
1410 * Kernel thread context (may sleep).
1411 *
1412 * RETURNS:
1413 * true if sense data could be fetched, false otherwise.
1414 */
1415static bool ata_eh_request_sense(struct ata_queued_cmd *qc)
1416{
1417 struct scsi_cmnd *cmd = qc->scsicmd;
1418 struct ata_device *dev = qc->dev;
1419 struct ata_taskfile tf;
1420 unsigned int err_mask;
1421
1422 if (ata_port_is_frozen(qc->ap)) {
1423 ata_dev_warn(dev, "sense data available but port frozen\n");
1424 return false;
1425 }
1426
1427 if (!ata_id_sense_reporting_enabled(dev->id)) {
1428 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1429 return false;
1430 }
1431
1432 ata_tf_init(dev, &tf);
1433 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1434 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1435 tf.command = ATA_CMD_REQ_SENSE_DATA;
1436 tf.protocol = ATA_PROT_NODATA;
1437
1438 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1439 /* Ignore err_mask; ATA_ERR might be set */
1440 if (tf.status & ATA_SENSE) {
1441 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1442 /* Set sense without also setting scsicmd->result */
1443 scsi_build_sense_buffer(dev->flags & ATA_DFLAG_D_SENSE,
1444 cmd->sense_buffer, tf.lbah,
1445 tf.lbam, tf.lbal);
1446 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1447 return true;
1448 }
1449 } else {
1450 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1451 tf.status, err_mask);
1452 }
1453
1454 return false;
1455}
1456
1457/**
1458 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1459 * @dev: device to perform REQUEST_SENSE to
1460 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1461 * @dfl_sense_key: default sense key to use
1462 *
1463 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1464 * SENSE. This function is EH helper.
1465 *
1466 * LOCKING:
1467 * Kernel thread context (may sleep).
1468 *
1469 * RETURNS:
1470 * 0 on success, AC_ERR_* mask on failure
1471 */
1472unsigned int atapi_eh_request_sense(struct ata_device *dev,
1473 u8 *sense_buf, u8 dfl_sense_key)
1474{
1475 u8 cdb[ATAPI_CDB_LEN] =
1476 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1477 struct ata_port *ap = dev->link->ap;
1478 struct ata_taskfile tf;
1479
1480 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1481
1482 /* initialize sense_buf with the error register,
1483 * for the case where they are -not- overwritten
1484 */
1485 sense_buf[0] = 0x70;
1486 sense_buf[2] = dfl_sense_key;
1487
1488 /* some devices time out if garbage left in tf */
1489 ata_tf_init(dev, &tf);
1490
1491 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1492 tf.command = ATA_CMD_PACKET;
1493
1494 /* is it pointless to prefer PIO for "safety reasons"? */
1495 if (ap->flags & ATA_FLAG_PIO_DMA) {
1496 tf.protocol = ATAPI_PROT_DMA;
1497 tf.feature |= ATAPI_PKT_DMA;
1498 } else {
1499 tf.protocol = ATAPI_PROT_PIO;
1500 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1501 tf.lbah = 0;
1502 }
1503
1504 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1505 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1506}
1507
1508/**
1509 * ata_eh_analyze_serror - analyze SError for a failed port
1510 * @link: ATA link to analyze SError for
1511 *
1512 * Analyze SError if available and further determine cause of
1513 * failure.
1514 *
1515 * LOCKING:
1516 * None.
1517 */
1518static void ata_eh_analyze_serror(struct ata_link *link)
1519{
1520 struct ata_eh_context *ehc = &link->eh_context;
1521 u32 serror = ehc->i.serror;
1522 unsigned int err_mask = 0, action = 0;
1523 u32 hotplug_mask;
1524
1525 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1526 err_mask |= AC_ERR_ATA_BUS;
1527 action |= ATA_EH_RESET;
1528 }
1529 if (serror & SERR_PROTOCOL) {
1530 err_mask |= AC_ERR_HSM;
1531 action |= ATA_EH_RESET;
1532 }
1533 if (serror & SERR_INTERNAL) {
1534 err_mask |= AC_ERR_SYSTEM;
1535 action |= ATA_EH_RESET;
1536 }
1537
1538 /* Determine whether a hotplug event has occurred. Both
1539 * SError.N/X are considered hotplug events for enabled or
1540 * host links. For disabled PMP links, only N bit is
1541 * considered as X bit is left at 1 for link plugging.
1542 */
1543 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1544 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1545 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1546 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1547 else
1548 hotplug_mask = SERR_PHYRDY_CHG;
1549
1550 if (serror & hotplug_mask)
1551 ata_ehi_hotplugged(&ehc->i);
1552
1553 ehc->i.err_mask |= err_mask;
1554 ehc->i.action |= action;
1555}
1556
1557/**
1558 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1559 * @qc: qc to analyze
1560 *
1561 * Analyze taskfile of @qc and further determine cause of
1562 * failure. This function also requests ATAPI sense data if
1563 * available.
1564 *
1565 * LOCKING:
1566 * Kernel thread context (may sleep).
1567 *
1568 * RETURNS:
1569 * Determined recovery action
1570 */
1571static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1572{
1573 const struct ata_taskfile *tf = &qc->result_tf;
1574 unsigned int tmp, action = 0;
1575 u8 stat = tf->status, err = tf->error;
1576
1577 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1578 qc->err_mask |= AC_ERR_HSM;
1579 return ATA_EH_RESET;
1580 }
1581
1582 if (stat & (ATA_ERR | ATA_DF)) {
1583 qc->err_mask |= AC_ERR_DEV;
1584 /*
1585 * Sense data reporting does not work if the
1586 * device fault bit is set.
1587 */
1588 if (stat & ATA_DF)
1589 stat &= ~ATA_SENSE;
1590 } else {
1591 return 0;
1592 }
1593
1594 switch (qc->dev->class) {
1595 case ATA_DEV_ATA:
1596 case ATA_DEV_ZAC:
1597 /*
1598 * Fetch the sense data explicitly if:
1599 * -It was a non-NCQ command that failed, or
1600 * -It was a NCQ command that failed, but the sense data
1601 * was not included in the NCQ command error log
1602 * (i.e. NCQ autosense is not supported by the device).
1603 */
1604 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) &&
1605 (stat & ATA_SENSE) && ata_eh_request_sense(qc))
1606 set_status_byte(qc->scsicmd, SAM_STAT_CHECK_CONDITION);
1607 if (err & ATA_ICRC)
1608 qc->err_mask |= AC_ERR_ATA_BUS;
1609 if (err & (ATA_UNC | ATA_AMNF))
1610 qc->err_mask |= AC_ERR_MEDIA;
1611 if (err & ATA_IDNF)
1612 qc->err_mask |= AC_ERR_INVALID;
1613 break;
1614
1615 case ATA_DEV_ATAPI:
1616 if (!ata_port_is_frozen(qc->ap)) {
1617 tmp = atapi_eh_request_sense(qc->dev,
1618 qc->scsicmd->sense_buffer,
1619 qc->result_tf.error >> 4);
1620 if (!tmp)
1621 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1622 else
1623 qc->err_mask |= tmp;
1624 }
1625 }
1626
1627 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1628 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
1629 /*
1630 * SUCCESS here means that the sense code could be
1631 * evaluated and should be passed to the upper layers
1632 * for correct evaluation.
1633 * FAILED means the sense code could not be interpreted
1634 * and the device would need to be reset.
1635 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1636 * command would need to be retried.
1637 */
1638 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1639 qc->flags |= ATA_QCFLAG_RETRY;
1640 qc->err_mask |= AC_ERR_OTHER;
1641 } else if (ret != SUCCESS) {
1642 qc->err_mask |= AC_ERR_HSM;
1643 }
1644 }
1645 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1646 action |= ATA_EH_RESET;
1647
1648 return action;
1649}
1650
1651static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1652 int *xfer_ok)
1653{
1654 int base = 0;
1655
1656 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1657 *xfer_ok = 1;
1658
1659 if (!*xfer_ok)
1660 base = ATA_ECAT_DUBIOUS_NONE;
1661
1662 if (err_mask & AC_ERR_ATA_BUS)
1663 return base + ATA_ECAT_ATA_BUS;
1664
1665 if (err_mask & AC_ERR_TIMEOUT)
1666 return base + ATA_ECAT_TOUT_HSM;
1667
1668 if (eflags & ATA_EFLAG_IS_IO) {
1669 if (err_mask & AC_ERR_HSM)
1670 return base + ATA_ECAT_TOUT_HSM;
1671 if ((err_mask &
1672 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1673 return base + ATA_ECAT_UNK_DEV;
1674 }
1675
1676 return 0;
1677}
1678
1679struct speed_down_verdict_arg {
1680 u64 since;
1681 int xfer_ok;
1682 int nr_errors[ATA_ECAT_NR];
1683};
1684
1685static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1686{
1687 struct speed_down_verdict_arg *arg = void_arg;
1688 int cat;
1689
1690 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1691 return -1;
1692
1693 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1694 &arg->xfer_ok);
1695 arg->nr_errors[cat]++;
1696
1697 return 0;
1698}
1699
1700/**
1701 * ata_eh_speed_down_verdict - Determine speed down verdict
1702 * @dev: Device of interest
1703 *
1704 * This function examines error ring of @dev and determines
1705 * whether NCQ needs to be turned off, transfer speed should be
1706 * stepped down, or falling back to PIO is necessary.
1707 *
1708 * ECAT_ATA_BUS : ATA_BUS error for any command
1709 *
1710 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1711 * IO commands
1712 *
1713 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1714 *
1715 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1716 * data transfer hasn't been verified.
1717 *
1718 * Verdicts are
1719 *
1720 * NCQ_OFF : Turn off NCQ.
1721 *
1722 * SPEED_DOWN : Speed down transfer speed but don't fall back
1723 * to PIO.
1724 *
1725 * FALLBACK_TO_PIO : Fall back to PIO.
1726 *
1727 * Even if multiple verdicts are returned, only one action is
1728 * taken per error. An action triggered by non-DUBIOUS errors
1729 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1730 * This is to expedite speed down decisions right after device is
1731 * initially configured.
1732 *
1733 * The following are speed down rules. #1 and #2 deal with
1734 * DUBIOUS errors.
1735 *
1736 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1737 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1738 *
1739 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1740 * occurred during last 5 mins, NCQ_OFF.
1741 *
1742 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1743 * occurred during last 5 mins, FALLBACK_TO_PIO
1744 *
1745 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1746 * during last 10 mins, NCQ_OFF.
1747 *
1748 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1749 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1750 *
1751 * LOCKING:
1752 * Inherited from caller.
1753 *
1754 * RETURNS:
1755 * OR of ATA_EH_SPDN_* flags.
1756 */
1757static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1758{
1759 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1760 u64 j64 = get_jiffies_64();
1761 struct speed_down_verdict_arg arg;
1762 unsigned int verdict = 0;
1763
1764 /* scan past 5 mins of error history */
1765 memset(&arg, 0, sizeof(arg));
1766 arg.since = j64 - min(j64, j5mins);
1767 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1768
1769 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1770 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1771 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1772 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1773
1774 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1775 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1776 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1777
1778 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1779 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1780 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1781 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1782
1783 /* scan past 10 mins of error history */
1784 memset(&arg, 0, sizeof(arg));
1785 arg.since = j64 - min(j64, j10mins);
1786 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1787
1788 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1789 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1790 verdict |= ATA_EH_SPDN_NCQ_OFF;
1791
1792 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1793 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1794 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1795 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1796
1797 return verdict;
1798}
1799
1800/**
1801 * ata_eh_speed_down - record error and speed down if necessary
1802 * @dev: Failed device
1803 * @eflags: mask of ATA_EFLAG_* flags
1804 * @err_mask: err_mask of the error
1805 *
1806 * Record error and examine error history to determine whether
1807 * adjusting transmission speed is necessary. It also sets
1808 * transmission limits appropriately if such adjustment is
1809 * necessary.
1810 *
1811 * LOCKING:
1812 * Kernel thread context (may sleep).
1813 *
1814 * RETURNS:
1815 * Determined recovery action.
1816 */
1817static unsigned int ata_eh_speed_down(struct ata_device *dev,
1818 unsigned int eflags, unsigned int err_mask)
1819{
1820 struct ata_link *link = ata_dev_phys_link(dev);
1821 int xfer_ok = 0;
1822 unsigned int verdict;
1823 unsigned int action = 0;
1824
1825 /* don't bother if Cat-0 error */
1826 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1827 return 0;
1828
1829 /* record error and determine whether speed down is necessary */
1830 ata_ering_record(&dev->ering, eflags, err_mask);
1831 verdict = ata_eh_speed_down_verdict(dev);
1832
1833 /* turn off NCQ? */
1834 if ((verdict & ATA_EH_SPDN_NCQ_OFF) && ata_ncq_enabled(dev)) {
1835 dev->flags |= ATA_DFLAG_NCQ_OFF;
1836 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1837 goto done;
1838 }
1839
1840 /* speed down? */
1841 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1842 /* speed down SATA link speed if possible */
1843 if (sata_down_spd_limit(link, 0) == 0) {
1844 action |= ATA_EH_RESET;
1845 goto done;
1846 }
1847
1848 /* lower transfer mode */
1849 if (dev->spdn_cnt < 2) {
1850 static const int dma_dnxfer_sel[] =
1851 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1852 static const int pio_dnxfer_sel[] =
1853 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1854 int sel;
1855
1856 if (dev->xfer_shift != ATA_SHIFT_PIO)
1857 sel = dma_dnxfer_sel[dev->spdn_cnt];
1858 else
1859 sel = pio_dnxfer_sel[dev->spdn_cnt];
1860
1861 dev->spdn_cnt++;
1862
1863 if (ata_down_xfermask_limit(dev, sel) == 0) {
1864 action |= ATA_EH_RESET;
1865 goto done;
1866 }
1867 }
1868 }
1869
1870 /* Fall back to PIO? Slowing down to PIO is meaningless for
1871 * SATA ATA devices. Consider it only for PATA and SATAPI.
1872 */
1873 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1874 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1875 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1876 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1877 dev->spdn_cnt = 0;
1878 action |= ATA_EH_RESET;
1879 goto done;
1880 }
1881 }
1882
1883 return 0;
1884 done:
1885 /* device has been slowed down, blow error history */
1886 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1887 ata_ering_clear(&dev->ering);
1888 return action;
1889}
1890
1891/**
1892 * ata_eh_worth_retry - analyze error and decide whether to retry
1893 * @qc: qc to possibly retry
1894 *
1895 * Look at the cause of the error and decide if a retry
1896 * might be useful or not. We don't want to retry media errors
1897 * because the drive itself has probably already taken 10-30 seconds
1898 * doing its own internal retries before reporting the failure.
1899 */
1900static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1901{
1902 if (qc->err_mask & AC_ERR_MEDIA)
1903 return 0; /* don't retry media errors */
1904 if (qc->flags & ATA_QCFLAG_IO)
1905 return 1; /* otherwise retry anything from fs stack */
1906 if (qc->err_mask & AC_ERR_INVALID)
1907 return 0; /* don't retry these */
1908 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1909}
1910
1911/**
1912 * ata_eh_quiet - check if we need to be quiet about a command error
1913 * @qc: qc to check
1914 *
1915 * Look at the qc flags anbd its scsi command request flags to determine
1916 * if we need to be quiet about the command failure.
1917 */
1918static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1919{
1920 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1921 qc->flags |= ATA_QCFLAG_QUIET;
1922 return qc->flags & ATA_QCFLAG_QUIET;
1923}
1924
1925static int ata_eh_read_sense_success_non_ncq(struct ata_link *link)
1926{
1927 struct ata_port *ap = link->ap;
1928 struct ata_queued_cmd *qc;
1929
1930 qc = __ata_qc_from_tag(ap, link->active_tag);
1931 if (!qc)
1932 return -EIO;
1933
1934 if (!(qc->flags & ATA_QCFLAG_EH) ||
1935 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
1936 qc->err_mask)
1937 return -EIO;
1938
1939 if (!ata_eh_request_sense(qc))
1940 return -EIO;
1941
1942 /*
1943 * If we have sense data, call scsi_check_sense() in order to set the
1944 * correct SCSI ML byte (if any). No point in checking the return value,
1945 * since the command has already completed successfully.
1946 */
1947 scsi_check_sense(qc->scsicmd);
1948
1949 return 0;
1950}
1951
1952static void ata_eh_get_success_sense(struct ata_link *link)
1953{
1954 struct ata_eh_context *ehc = &link->eh_context;
1955 struct ata_device *dev = link->device;
1956 struct ata_port *ap = link->ap;
1957 struct ata_queued_cmd *qc;
1958 int tag, ret = 0;
1959
1960 if (!(ehc->i.dev_action[dev->devno] & ATA_EH_GET_SUCCESS_SENSE))
1961 return;
1962
1963 /* if frozen, we can't do much */
1964 if (ata_port_is_frozen(ap)) {
1965 ata_dev_warn(dev,
1966 "successful sense data available but port frozen\n");
1967 goto out;
1968 }
1969
1970 /*
1971 * If the link has sactive set, then we have outstanding NCQ commands
1972 * and have to read the Successful NCQ Commands log to get the sense
1973 * data. Otherwise, we are dealing with a non-NCQ command and use
1974 * request sense ext command to retrieve the sense data.
1975 */
1976 if (link->sactive)
1977 ret = ata_eh_read_sense_success_ncq_log(link);
1978 else
1979 ret = ata_eh_read_sense_success_non_ncq(link);
1980 if (ret)
1981 goto out;
1982
1983 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
1984 return;
1985
1986out:
1987 /*
1988 * If we failed to get sense data for a successful command that ought to
1989 * have sense data, we cannot simply return BLK_STS_OK to user space.
1990 * This is because we can't know if the sense data that we couldn't get
1991 * was actually "DATA CURRENTLY UNAVAILABLE". Reporting such a command
1992 * as success to user space would result in a silent data corruption.
1993 * Thus, add a bogus ABORTED_COMMAND sense data to such commands, such
1994 * that SCSI will report these commands as BLK_STS_IOERR to user space.
1995 */
1996 ata_qc_for_each_raw(ap, qc, tag) {
1997 if (!(qc->flags & ATA_QCFLAG_EH) ||
1998 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
1999 qc->err_mask ||
2000 ata_dev_phys_link(qc->dev) != link)
2001 continue;
2002
2003 /* We managed to get sense for this success command, skip. */
2004 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2005 continue;
2006
2007 /* This success command did not have any sense data, skip. */
2008 if (!(qc->result_tf.status & ATA_SENSE))
2009 continue;
2010
2011 /* This success command had sense data, but we failed to get. */
2012 ata_scsi_set_sense(dev, qc->scsicmd, ABORTED_COMMAND, 0, 0);
2013 qc->flags |= ATA_QCFLAG_SENSE_VALID;
2014 }
2015 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
2016}
2017
2018/**
2019 * ata_eh_link_autopsy - analyze error and determine recovery action
2020 * @link: host link to perform autopsy on
2021 *
2022 * Analyze why @link failed and determine which recovery actions
2023 * are needed. This function also sets more detailed AC_ERR_*
2024 * values and fills sense data for ATAPI CHECK SENSE.
2025 *
2026 * LOCKING:
2027 * Kernel thread context (may sleep).
2028 */
2029static void ata_eh_link_autopsy(struct ata_link *link)
2030{
2031 struct ata_port *ap = link->ap;
2032 struct ata_eh_context *ehc = &link->eh_context;
2033 struct ata_queued_cmd *qc;
2034 struct ata_device *dev;
2035 unsigned int all_err_mask = 0, eflags = 0;
2036 int tag, nr_failed = 0, nr_quiet = 0;
2037 u32 serror;
2038 int rc;
2039
2040 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2041 return;
2042
2043 /* obtain and analyze SError */
2044 rc = sata_scr_read(link, SCR_ERROR, &serror);
2045 if (rc == 0) {
2046 ehc->i.serror |= serror;
2047 ata_eh_analyze_serror(link);
2048 } else if (rc != -EOPNOTSUPP) {
2049 /* SError read failed, force reset and probing */
2050 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2051 ehc->i.action |= ATA_EH_RESET;
2052 ehc->i.err_mask |= AC_ERR_OTHER;
2053 }
2054
2055 /* analyze NCQ failure */
2056 ata_eh_analyze_ncq_error(link);
2057
2058 /*
2059 * Check if this was a successful command that simply needs sense data.
2060 * Since the sense data is not part of the completion, we need to fetch
2061 * it using an additional command. Since this can't be done from irq
2062 * context, the sense data for successful commands are fetched by EH.
2063 */
2064 ata_eh_get_success_sense(link);
2065
2066 /* any real error trumps AC_ERR_OTHER */
2067 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2068 ehc->i.err_mask &= ~AC_ERR_OTHER;
2069
2070 all_err_mask |= ehc->i.err_mask;
2071
2072 ata_qc_for_each_raw(ap, qc, tag) {
2073 if (!(qc->flags & ATA_QCFLAG_EH) ||
2074 qc->flags & ATA_QCFLAG_RETRY ||
2075 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD ||
2076 ata_dev_phys_link(qc->dev) != link)
2077 continue;
2078
2079 /* inherit upper level err_mask */
2080 qc->err_mask |= ehc->i.err_mask;
2081
2082 /* analyze TF */
2083 ehc->i.action |= ata_eh_analyze_tf(qc);
2084
2085 /* DEV errors are probably spurious in case of ATA_BUS error */
2086 if (qc->err_mask & AC_ERR_ATA_BUS)
2087 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2088 AC_ERR_INVALID);
2089
2090 /* any real error trumps unknown error */
2091 if (qc->err_mask & ~AC_ERR_OTHER)
2092 qc->err_mask &= ~AC_ERR_OTHER;
2093
2094 /*
2095 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2096 * layers will determine whether the command is worth retrying
2097 * based on the sense data and device class/type. Otherwise,
2098 * determine directly if the command is worth retrying using its
2099 * error mask and flags.
2100 */
2101 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2102 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2103 else if (ata_eh_worth_retry(qc))
2104 qc->flags |= ATA_QCFLAG_RETRY;
2105
2106 /* accumulate error info */
2107 ehc->i.dev = qc->dev;
2108 all_err_mask |= qc->err_mask;
2109 if (qc->flags & ATA_QCFLAG_IO)
2110 eflags |= ATA_EFLAG_IS_IO;
2111 trace_ata_eh_link_autopsy_qc(qc);
2112
2113 /* Count quiet errors */
2114 if (ata_eh_quiet(qc))
2115 nr_quiet++;
2116 nr_failed++;
2117 }
2118
2119 /* If all failed commands requested silence, then be quiet */
2120 if (nr_quiet == nr_failed)
2121 ehc->i.flags |= ATA_EHI_QUIET;
2122
2123 /* enforce default EH actions */
2124 if (ata_port_is_frozen(ap) ||
2125 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2126 ehc->i.action |= ATA_EH_RESET;
2127 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2128 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2129 ehc->i.action |= ATA_EH_REVALIDATE;
2130
2131 /* If we have offending qcs and the associated failed device,
2132 * perform per-dev EH action only on the offending device.
2133 */
2134 if (ehc->i.dev) {
2135 ehc->i.dev_action[ehc->i.dev->devno] |=
2136 ehc->i.action & ATA_EH_PERDEV_MASK;
2137 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2138 }
2139
2140 /* propagate timeout to host link */
2141 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2142 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2143
2144 /* record error and consider speeding down */
2145 dev = ehc->i.dev;
2146 if (!dev && ((ata_link_max_devices(link) == 1 &&
2147 ata_dev_enabled(link->device))))
2148 dev = link->device;
2149
2150 if (dev) {
2151 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2152 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2153 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2154 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2155 }
2156}
2157
2158/**
2159 * ata_eh_autopsy - analyze error and determine recovery action
2160 * @ap: host port to perform autopsy on
2161 *
2162 * Analyze all links of @ap and determine why they failed and
2163 * which recovery actions are needed.
2164 *
2165 * LOCKING:
2166 * Kernel thread context (may sleep).
2167 */
2168void ata_eh_autopsy(struct ata_port *ap)
2169{
2170 struct ata_link *link;
2171
2172 ata_for_each_link(link, ap, EDGE)
2173 ata_eh_link_autopsy(link);
2174
2175 /* Handle the frigging slave link. Autopsy is done similarly
2176 * but actions and flags are transferred over to the master
2177 * link and handled from there.
2178 */
2179 if (ap->slave_link) {
2180 struct ata_eh_context *mehc = &ap->link.eh_context;
2181 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2182
2183 /* transfer control flags from master to slave */
2184 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2185
2186 /* perform autopsy on the slave link */
2187 ata_eh_link_autopsy(ap->slave_link);
2188
2189 /* transfer actions from slave to master and clear slave */
2190 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2191 mehc->i.action |= sehc->i.action;
2192 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2193 mehc->i.flags |= sehc->i.flags;
2194 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2195 }
2196
2197 /* Autopsy of fanout ports can affect host link autopsy.
2198 * Perform host link autopsy last.
2199 */
2200 if (sata_pmp_attached(ap))
2201 ata_eh_link_autopsy(&ap->link);
2202}
2203
2204/**
2205 * ata_get_cmd_name - get name for ATA command
2206 * @command: ATA command code to get name for
2207 *
2208 * Return a textual name of the given command or "unknown"
2209 *
2210 * LOCKING:
2211 * None
2212 */
2213const char *ata_get_cmd_name(u8 command)
2214{
2215#ifdef CONFIG_ATA_VERBOSE_ERROR
2216 static const struct
2217 {
2218 u8 command;
2219 const char *text;
2220 } cmd_descr[] = {
2221 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2222 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2223 { ATA_CMD_STANDBY, "STANDBY" },
2224 { ATA_CMD_IDLE, "IDLE" },
2225 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2226 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2227 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2228 { ATA_CMD_NOP, "NOP" },
2229 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2230 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2231 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2232 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2233 { ATA_CMD_SERVICE, "SERVICE" },
2234 { ATA_CMD_READ, "READ DMA" },
2235 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2236 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2237 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2238 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2239 { ATA_CMD_WRITE, "WRITE DMA" },
2240 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2241 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2242 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2243 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2244 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2245 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2246 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2247 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2248 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" },
2249 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2250 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2251 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2252 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2253 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2254 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2255 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2256 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2257 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2258 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2259 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2260 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2261 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2262 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2263 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2264 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2265 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2266 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2267 { ATA_CMD_SLEEP, "SLEEP" },
2268 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2269 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2270 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2271 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2272 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2273 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2274 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2275 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2276 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2277 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2278 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2279 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2280 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2281 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2282 { ATA_CMD_PMP_READ, "READ BUFFER" },
2283 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2284 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2285 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2286 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2287 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2288 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2289 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2290 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2291 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2292 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2293 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2294 { ATA_CMD_SMART, "SMART" },
2295 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2296 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2297 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2298 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2299 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2300 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2301 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2302 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2303 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2304 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2305 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2306 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2307 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2308 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2309 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2310 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2311 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2312 { ATA_CMD_RESTORE, "RECALIBRATE" },
2313 { 0, NULL } /* terminate list */
2314 };
2315
2316 unsigned int i;
2317 for (i = 0; cmd_descr[i].text; i++)
2318 if (cmd_descr[i].command == command)
2319 return cmd_descr[i].text;
2320#endif
2321
2322 return "unknown";
2323}
2324EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2325
2326/**
2327 * ata_eh_link_report - report error handling to user
2328 * @link: ATA link EH is going on
2329 *
2330 * Report EH to user.
2331 *
2332 * LOCKING:
2333 * None.
2334 */
2335static void ata_eh_link_report(struct ata_link *link)
2336{
2337 struct ata_port *ap = link->ap;
2338 struct ata_eh_context *ehc = &link->eh_context;
2339 struct ata_queued_cmd *qc;
2340 const char *frozen, *desc;
2341 char tries_buf[16] = "";
2342 int tag, nr_failed = 0;
2343
2344 if (ehc->i.flags & ATA_EHI_QUIET)
2345 return;
2346
2347 desc = NULL;
2348 if (ehc->i.desc[0] != '\0')
2349 desc = ehc->i.desc;
2350
2351 ata_qc_for_each_raw(ap, qc, tag) {
2352 if (!(qc->flags & ATA_QCFLAG_EH) ||
2353 ata_dev_phys_link(qc->dev) != link ||
2354 ((qc->flags & ATA_QCFLAG_QUIET) &&
2355 qc->err_mask == AC_ERR_DEV))
2356 continue;
2357 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2358 continue;
2359
2360 nr_failed++;
2361 }
2362
2363 if (!nr_failed && !ehc->i.err_mask)
2364 return;
2365
2366 frozen = "";
2367 if (ata_port_is_frozen(ap))
2368 frozen = " frozen";
2369
2370 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2371 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2372 ap->eh_tries);
2373
2374 if (ehc->i.dev) {
2375 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2376 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2377 ehc->i.err_mask, link->sactive, ehc->i.serror,
2378 ehc->i.action, frozen, tries_buf);
2379 if (desc)
2380 ata_dev_err(ehc->i.dev, "%s\n", desc);
2381 } else {
2382 ata_link_err(link, "exception Emask 0x%x "
2383 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2384 ehc->i.err_mask, link->sactive, ehc->i.serror,
2385 ehc->i.action, frozen, tries_buf);
2386 if (desc)
2387 ata_link_err(link, "%s\n", desc);
2388 }
2389
2390#ifdef CONFIG_ATA_VERBOSE_ERROR
2391 if (ehc->i.serror)
2392 ata_link_err(link,
2393 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2394 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2395 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2396 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2397 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2398 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2399 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2400 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2401 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2402 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2403 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2404 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2405 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2406 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2407 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2408 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2409 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2410 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2411#endif
2412
2413 ata_qc_for_each_raw(ap, qc, tag) {
2414 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2415 char data_buf[20] = "";
2416 char cdb_buf[70] = "";
2417
2418 if (!(qc->flags & ATA_QCFLAG_EH) ||
2419 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2420 continue;
2421
2422 if (qc->dma_dir != DMA_NONE) {
2423 static const char *dma_str[] = {
2424 [DMA_BIDIRECTIONAL] = "bidi",
2425 [DMA_TO_DEVICE] = "out",
2426 [DMA_FROM_DEVICE] = "in",
2427 };
2428 const char *prot_str = NULL;
2429
2430 switch (qc->tf.protocol) {
2431 case ATA_PROT_UNKNOWN:
2432 prot_str = "unknown";
2433 break;
2434 case ATA_PROT_NODATA:
2435 prot_str = "nodata";
2436 break;
2437 case ATA_PROT_PIO:
2438 prot_str = "pio";
2439 break;
2440 case ATA_PROT_DMA:
2441 prot_str = "dma";
2442 break;
2443 case ATA_PROT_NCQ:
2444 prot_str = "ncq dma";
2445 break;
2446 case ATA_PROT_NCQ_NODATA:
2447 prot_str = "ncq nodata";
2448 break;
2449 case ATAPI_PROT_NODATA:
2450 prot_str = "nodata";
2451 break;
2452 case ATAPI_PROT_PIO:
2453 prot_str = "pio";
2454 break;
2455 case ATAPI_PROT_DMA:
2456 prot_str = "dma";
2457 break;
2458 }
2459 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2460 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2461 }
2462
2463 if (ata_is_atapi(qc->tf.protocol)) {
2464 const u8 *cdb = qc->cdb;
2465 size_t cdb_len = qc->dev->cdb_len;
2466
2467 if (qc->scsicmd) {
2468 cdb = qc->scsicmd->cmnd;
2469 cdb_len = qc->scsicmd->cmd_len;
2470 }
2471 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2472 cdb, cdb_len);
2473 } else
2474 ata_dev_err(qc->dev, "failed command: %s\n",
2475 ata_get_cmd_name(cmd->command));
2476
2477 ata_dev_err(qc->dev,
2478 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2479 "tag %d%s\n %s"
2480 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2481 "Emask 0x%x (%s)%s\n",
2482 cmd->command, cmd->feature, cmd->nsect,
2483 cmd->lbal, cmd->lbam, cmd->lbah,
2484 cmd->hob_feature, cmd->hob_nsect,
2485 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2486 cmd->device, qc->tag, data_buf, cdb_buf,
2487 res->status, res->error, res->nsect,
2488 res->lbal, res->lbam, res->lbah,
2489 res->hob_feature, res->hob_nsect,
2490 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2491 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2492 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2493
2494#ifdef CONFIG_ATA_VERBOSE_ERROR
2495 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2496 ATA_SENSE | ATA_ERR)) {
2497 if (res->status & ATA_BUSY)
2498 ata_dev_err(qc->dev, "status: { Busy }\n");
2499 else
2500 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2501 res->status & ATA_DRDY ? "DRDY " : "",
2502 res->status & ATA_DF ? "DF " : "",
2503 res->status & ATA_DRQ ? "DRQ " : "",
2504 res->status & ATA_SENSE ? "SENSE " : "",
2505 res->status & ATA_ERR ? "ERR " : "");
2506 }
2507
2508 if (cmd->command != ATA_CMD_PACKET &&
2509 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2510 ATA_ABORTED)))
2511 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2512 res->error & ATA_ICRC ? "ICRC " : "",
2513 res->error & ATA_UNC ? "UNC " : "",
2514 res->error & ATA_AMNF ? "AMNF " : "",
2515 res->error & ATA_IDNF ? "IDNF " : "",
2516 res->error & ATA_ABORTED ? "ABRT " : "");
2517#endif
2518 }
2519}
2520
2521/**
2522 * ata_eh_report - report error handling to user
2523 * @ap: ATA port to report EH about
2524 *
2525 * Report EH to user.
2526 *
2527 * LOCKING:
2528 * None.
2529 */
2530void ata_eh_report(struct ata_port *ap)
2531{
2532 struct ata_link *link;
2533
2534 ata_for_each_link(link, ap, HOST_FIRST)
2535 ata_eh_link_report(link);
2536}
2537
2538static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2539 unsigned int *classes, unsigned long deadline,
2540 bool clear_classes)
2541{
2542 struct ata_device *dev;
2543
2544 if (clear_classes)
2545 ata_for_each_dev(dev, link, ALL)
2546 classes[dev->devno] = ATA_DEV_UNKNOWN;
2547
2548 return reset(link, classes, deadline);
2549}
2550
2551static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2552{
2553 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2554 return 0;
2555 if (rc == -EAGAIN)
2556 return 1;
2557 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2558 return 1;
2559 return 0;
2560}
2561
2562int ata_eh_reset(struct ata_link *link, int classify,
2563 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2564 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2565{
2566 struct ata_port *ap = link->ap;
2567 struct ata_link *slave = ap->slave_link;
2568 struct ata_eh_context *ehc = &link->eh_context;
2569 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2570 unsigned int *classes = ehc->classes;
2571 unsigned int lflags = link->flags;
2572 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2573 int max_tries = 0, try = 0;
2574 struct ata_link *failed_link;
2575 struct ata_device *dev;
2576 unsigned long deadline, now;
2577 ata_reset_fn_t reset;
2578 unsigned long flags;
2579 u32 sstatus;
2580 int nr_unknown, rc;
2581
2582 /*
2583 * Prepare to reset
2584 */
2585 while (ata_eh_reset_timeouts[max_tries] != UINT_MAX)
2586 max_tries++;
2587 if (link->flags & ATA_LFLAG_RST_ONCE)
2588 max_tries = 1;
2589 if (link->flags & ATA_LFLAG_NO_HRST)
2590 hardreset = NULL;
2591 if (link->flags & ATA_LFLAG_NO_SRST)
2592 softreset = NULL;
2593
2594 /* make sure each reset attempt is at least COOL_DOWN apart */
2595 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2596 now = jiffies;
2597 WARN_ON(time_after(ehc->last_reset, now));
2598 deadline = ata_deadline(ehc->last_reset,
2599 ATA_EH_RESET_COOL_DOWN);
2600 if (time_before(now, deadline))
2601 schedule_timeout_uninterruptible(deadline - now);
2602 }
2603
2604 spin_lock_irqsave(ap->lock, flags);
2605 ap->pflags |= ATA_PFLAG_RESETTING;
2606 spin_unlock_irqrestore(ap->lock, flags);
2607
2608 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2609
2610 ata_for_each_dev(dev, link, ALL) {
2611 /* If we issue an SRST then an ATA drive (not ATAPI)
2612 * may change configuration and be in PIO0 timing. If
2613 * we do a hard reset (or are coming from power on)
2614 * this is true for ATA or ATAPI. Until we've set a
2615 * suitable controller mode we should not touch the
2616 * bus as we may be talking too fast.
2617 */
2618 dev->pio_mode = XFER_PIO_0;
2619 dev->dma_mode = 0xff;
2620
2621 /* If the controller has a pio mode setup function
2622 * then use it to set the chipset to rights. Don't
2623 * touch the DMA setup as that will be dealt with when
2624 * configuring devices.
2625 */
2626 if (ap->ops->set_piomode)
2627 ap->ops->set_piomode(ap, dev);
2628 }
2629
2630 /* prefer hardreset */
2631 reset = NULL;
2632 ehc->i.action &= ~ATA_EH_RESET;
2633 if (hardreset) {
2634 reset = hardreset;
2635 ehc->i.action |= ATA_EH_HARDRESET;
2636 } else if (softreset) {
2637 reset = softreset;
2638 ehc->i.action |= ATA_EH_SOFTRESET;
2639 }
2640
2641 if (prereset) {
2642 unsigned long deadline = ata_deadline(jiffies,
2643 ATA_EH_PRERESET_TIMEOUT);
2644
2645 if (slave) {
2646 sehc->i.action &= ~ATA_EH_RESET;
2647 sehc->i.action |= ehc->i.action;
2648 }
2649
2650 rc = prereset(link, deadline);
2651
2652 /* If present, do prereset on slave link too. Reset
2653 * is skipped iff both master and slave links report
2654 * -ENOENT or clear ATA_EH_RESET.
2655 */
2656 if (slave && (rc == 0 || rc == -ENOENT)) {
2657 int tmp;
2658
2659 tmp = prereset(slave, deadline);
2660 if (tmp != -ENOENT)
2661 rc = tmp;
2662
2663 ehc->i.action |= sehc->i.action;
2664 }
2665
2666 if (rc) {
2667 if (rc == -ENOENT) {
2668 ata_link_dbg(link, "port disabled--ignoring\n");
2669 ehc->i.action &= ~ATA_EH_RESET;
2670
2671 ata_for_each_dev(dev, link, ALL)
2672 classes[dev->devno] = ATA_DEV_NONE;
2673
2674 rc = 0;
2675 } else
2676 ata_link_err(link,
2677 "prereset failed (errno=%d)\n",
2678 rc);
2679 goto out;
2680 }
2681
2682 /* prereset() might have cleared ATA_EH_RESET. If so,
2683 * bang classes, thaw and return.
2684 */
2685 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2686 ata_for_each_dev(dev, link, ALL)
2687 classes[dev->devno] = ATA_DEV_NONE;
2688 if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2689 ata_eh_thaw_port(ap);
2690 rc = 0;
2691 goto out;
2692 }
2693 }
2694
2695 retry:
2696 /*
2697 * Perform reset
2698 */
2699 if (ata_is_host_link(link))
2700 ata_eh_freeze_port(ap);
2701
2702 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2703
2704 if (reset) {
2705 if (verbose)
2706 ata_link_info(link, "%s resetting link\n",
2707 reset == softreset ? "soft" : "hard");
2708
2709 /* mark that this EH session started with reset */
2710 ehc->last_reset = jiffies;
2711 if (reset == hardreset) {
2712 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2713 trace_ata_link_hardreset_begin(link, classes, deadline);
2714 } else {
2715 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2716 trace_ata_link_softreset_begin(link, classes, deadline);
2717 }
2718
2719 rc = ata_do_reset(link, reset, classes, deadline, true);
2720 if (reset == hardreset)
2721 trace_ata_link_hardreset_end(link, classes, rc);
2722 else
2723 trace_ata_link_softreset_end(link, classes, rc);
2724 if (rc && rc != -EAGAIN) {
2725 failed_link = link;
2726 goto fail;
2727 }
2728
2729 /* hardreset slave link if existent */
2730 if (slave && reset == hardreset) {
2731 int tmp;
2732
2733 if (verbose)
2734 ata_link_info(slave, "hard resetting link\n");
2735
2736 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2737 trace_ata_slave_hardreset_begin(slave, classes,
2738 deadline);
2739 tmp = ata_do_reset(slave, reset, classes, deadline,
2740 false);
2741 trace_ata_slave_hardreset_end(slave, classes, tmp);
2742 switch (tmp) {
2743 case -EAGAIN:
2744 rc = -EAGAIN;
2745 break;
2746 case 0:
2747 break;
2748 default:
2749 failed_link = slave;
2750 rc = tmp;
2751 goto fail;
2752 }
2753 }
2754
2755 /* perform follow-up SRST if necessary */
2756 if (reset == hardreset &&
2757 ata_eh_followup_srst_needed(link, rc)) {
2758 reset = softreset;
2759
2760 if (!reset) {
2761 ata_link_err(link,
2762 "follow-up softreset required but no softreset available\n");
2763 failed_link = link;
2764 rc = -EINVAL;
2765 goto fail;
2766 }
2767
2768 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2769 trace_ata_link_softreset_begin(link, classes, deadline);
2770 rc = ata_do_reset(link, reset, classes, deadline, true);
2771 trace_ata_link_softreset_end(link, classes, rc);
2772 if (rc) {
2773 failed_link = link;
2774 goto fail;
2775 }
2776 }
2777 } else {
2778 if (verbose)
2779 ata_link_info(link,
2780 "no reset method available, skipping reset\n");
2781 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2782 lflags |= ATA_LFLAG_ASSUME_ATA;
2783 }
2784
2785 /*
2786 * Post-reset processing
2787 */
2788 ata_for_each_dev(dev, link, ALL) {
2789 /* After the reset, the device state is PIO 0 and the
2790 * controller state is undefined. Reset also wakes up
2791 * drives from sleeping mode.
2792 */
2793 dev->pio_mode = XFER_PIO_0;
2794 dev->flags &= ~ATA_DFLAG_SLEEPING;
2795
2796 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2797 continue;
2798
2799 /* apply class override */
2800 if (lflags & ATA_LFLAG_ASSUME_ATA)
2801 classes[dev->devno] = ATA_DEV_ATA;
2802 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2803 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2804 }
2805
2806 /* record current link speed */
2807 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2808 link->sata_spd = (sstatus >> 4) & 0xf;
2809 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2810 slave->sata_spd = (sstatus >> 4) & 0xf;
2811
2812 /* thaw the port */
2813 if (ata_is_host_link(link))
2814 ata_eh_thaw_port(ap);
2815
2816 /* postreset() should clear hardware SError. Although SError
2817 * is cleared during link resume, clearing SError here is
2818 * necessary as some PHYs raise hotplug events after SRST.
2819 * This introduces race condition where hotplug occurs between
2820 * reset and here. This race is mediated by cross checking
2821 * link onlineness and classification result later.
2822 */
2823 if (postreset) {
2824 postreset(link, classes);
2825 trace_ata_link_postreset(link, classes, rc);
2826 if (slave) {
2827 postreset(slave, classes);
2828 trace_ata_slave_postreset(slave, classes, rc);
2829 }
2830 }
2831
2832 /* clear cached SError */
2833 spin_lock_irqsave(link->ap->lock, flags);
2834 link->eh_info.serror = 0;
2835 if (slave)
2836 slave->eh_info.serror = 0;
2837 spin_unlock_irqrestore(link->ap->lock, flags);
2838
2839 /*
2840 * Make sure onlineness and classification result correspond.
2841 * Hotplug could have happened during reset and some
2842 * controllers fail to wait while a drive is spinning up after
2843 * being hotplugged causing misdetection. By cross checking
2844 * link on/offlineness and classification result, those
2845 * conditions can be reliably detected and retried.
2846 */
2847 nr_unknown = 0;
2848 ata_for_each_dev(dev, link, ALL) {
2849 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2850 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2851 ata_dev_dbg(dev, "link online but device misclassified\n");
2852 classes[dev->devno] = ATA_DEV_NONE;
2853 nr_unknown++;
2854 }
2855 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2856 if (ata_class_enabled(classes[dev->devno]))
2857 ata_dev_dbg(dev,
2858 "link offline, clearing class %d to NONE\n",
2859 classes[dev->devno]);
2860 classes[dev->devno] = ATA_DEV_NONE;
2861 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2862 ata_dev_dbg(dev,
2863 "link status unknown, clearing UNKNOWN to NONE\n");
2864 classes[dev->devno] = ATA_DEV_NONE;
2865 }
2866 }
2867
2868 if (classify && nr_unknown) {
2869 if (try < max_tries) {
2870 ata_link_warn(link,
2871 "link online but %d devices misclassified, retrying\n",
2872 nr_unknown);
2873 failed_link = link;
2874 rc = -EAGAIN;
2875 goto fail;
2876 }
2877 ata_link_warn(link,
2878 "link online but %d devices misclassified, "
2879 "device detection might fail\n", nr_unknown);
2880 }
2881
2882 /* reset successful, schedule revalidation */
2883 ata_eh_done(link, NULL, ATA_EH_RESET);
2884 if (slave)
2885 ata_eh_done(slave, NULL, ATA_EH_RESET);
2886 ehc->last_reset = jiffies; /* update to completion time */
2887 ehc->i.action |= ATA_EH_REVALIDATE;
2888 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2889
2890 rc = 0;
2891 out:
2892 /* clear hotplug flag */
2893 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2894 if (slave)
2895 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2896
2897 spin_lock_irqsave(ap->lock, flags);
2898 ap->pflags &= ~ATA_PFLAG_RESETTING;
2899 spin_unlock_irqrestore(ap->lock, flags);
2900
2901 return rc;
2902
2903 fail:
2904 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2905 if (!ata_is_host_link(link) &&
2906 sata_scr_read(link, SCR_STATUS, &sstatus))
2907 rc = -ERESTART;
2908
2909 if (try >= max_tries) {
2910 /*
2911 * Thaw host port even if reset failed, so that the port
2912 * can be retried on the next phy event. This risks
2913 * repeated EH runs but seems to be a better tradeoff than
2914 * shutting down a port after a botched hotplug attempt.
2915 */
2916 if (ata_is_host_link(link))
2917 ata_eh_thaw_port(ap);
2918 ata_link_warn(link, "%s failed\n",
2919 reset == hardreset ? "hardreset" : "softreset");
2920 goto out;
2921 }
2922
2923 now = jiffies;
2924 if (time_before(now, deadline)) {
2925 unsigned long delta = deadline - now;
2926
2927 ata_link_warn(failed_link,
2928 "reset failed (errno=%d), retrying in %u secs\n",
2929 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2930
2931 ata_eh_release(ap);
2932 while (delta)
2933 delta = schedule_timeout_uninterruptible(delta);
2934 ata_eh_acquire(ap);
2935 }
2936
2937 /*
2938 * While disks spinup behind PMP, some controllers fail sending SRST.
2939 * They need to be reset - as well as the PMP - before retrying.
2940 */
2941 if (rc == -ERESTART) {
2942 if (ata_is_host_link(link))
2943 ata_eh_thaw_port(ap);
2944 goto out;
2945 }
2946
2947 if (try == max_tries - 1) {
2948 sata_down_spd_limit(link, 0);
2949 if (slave)
2950 sata_down_spd_limit(slave, 0);
2951 } else if (rc == -EPIPE)
2952 sata_down_spd_limit(failed_link, 0);
2953
2954 if (hardreset)
2955 reset = hardreset;
2956 goto retry;
2957}
2958
2959static inline void ata_eh_pull_park_action(struct ata_port *ap)
2960{
2961 struct ata_link *link;
2962 struct ata_device *dev;
2963 unsigned long flags;
2964
2965 /*
2966 * This function can be thought of as an extended version of
2967 * ata_eh_about_to_do() specially crafted to accommodate the
2968 * requirements of ATA_EH_PARK handling. Since the EH thread
2969 * does not leave the do {} while () loop in ata_eh_recover as
2970 * long as the timeout for a park request to *one* device on
2971 * the port has not expired, and since we still want to pick
2972 * up park requests to other devices on the same port or
2973 * timeout updates for the same device, we have to pull
2974 * ATA_EH_PARK actions from eh_info into eh_context.i
2975 * ourselves at the beginning of each pass over the loop.
2976 *
2977 * Additionally, all write accesses to &ap->park_req_pending
2978 * through reinit_completion() (see below) or complete_all()
2979 * (see ata_scsi_park_store()) are protected by the host lock.
2980 * As a result we have that park_req_pending.done is zero on
2981 * exit from this function, i.e. when ATA_EH_PARK actions for
2982 * *all* devices on port ap have been pulled into the
2983 * respective eh_context structs. If, and only if,
2984 * park_req_pending.done is non-zero by the time we reach
2985 * wait_for_completion_timeout(), another ATA_EH_PARK action
2986 * has been scheduled for at least one of the devices on port
2987 * ap and we have to cycle over the do {} while () loop in
2988 * ata_eh_recover() again.
2989 */
2990
2991 spin_lock_irqsave(ap->lock, flags);
2992 reinit_completion(&ap->park_req_pending);
2993 ata_for_each_link(link, ap, EDGE) {
2994 ata_for_each_dev(dev, link, ALL) {
2995 struct ata_eh_info *ehi = &link->eh_info;
2996
2997 link->eh_context.i.dev_action[dev->devno] |=
2998 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2999 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3000 }
3001 }
3002 spin_unlock_irqrestore(ap->lock, flags);
3003}
3004
3005static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3006{
3007 struct ata_eh_context *ehc = &dev->link->eh_context;
3008 struct ata_taskfile tf;
3009 unsigned int err_mask;
3010
3011 ata_tf_init(dev, &tf);
3012 if (park) {
3013 ehc->unloaded_mask |= 1 << dev->devno;
3014 tf.command = ATA_CMD_IDLEIMMEDIATE;
3015 tf.feature = 0x44;
3016 tf.lbal = 0x4c;
3017 tf.lbam = 0x4e;
3018 tf.lbah = 0x55;
3019 } else {
3020 ehc->unloaded_mask &= ~(1 << dev->devno);
3021 tf.command = ATA_CMD_CHK_POWER;
3022 }
3023
3024 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3025 tf.protocol = ATA_PROT_NODATA;
3026 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3027 if (park && (err_mask || tf.lbal != 0xc4)) {
3028 ata_dev_err(dev, "head unload failed!\n");
3029 ehc->unloaded_mask &= ~(1 << dev->devno);
3030 }
3031}
3032
3033static int ata_eh_revalidate_and_attach(struct ata_link *link,
3034 struct ata_device **r_failed_dev)
3035{
3036 struct ata_port *ap = link->ap;
3037 struct ata_eh_context *ehc = &link->eh_context;
3038 struct ata_device *dev;
3039 unsigned int new_mask = 0;
3040 unsigned long flags;
3041 int rc = 0;
3042
3043 /* For PATA drive side cable detection to work, IDENTIFY must
3044 * be done backwards such that PDIAG- is released by the slave
3045 * device before the master device is identified.
3046 */
3047 ata_for_each_dev(dev, link, ALL_REVERSE) {
3048 unsigned int action = ata_eh_dev_action(dev);
3049 unsigned int readid_flags = 0;
3050
3051 if (ehc->i.flags & ATA_EHI_DID_RESET)
3052 readid_flags |= ATA_READID_POSTRESET;
3053
3054 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3055 WARN_ON(dev->class == ATA_DEV_PMP);
3056
3057 /*
3058 * The link may be in a deep sleep, wake it up.
3059 *
3060 * If the link is in deep sleep, ata_phys_link_offline()
3061 * will return true, causing the revalidation to fail,
3062 * which leads to a (potentially) needless hard reset.
3063 *
3064 * ata_eh_recover() will later restore the link policy
3065 * to ap->target_lpm_policy after revalidation is done.
3066 */
3067 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3068 rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER,
3069 r_failed_dev);
3070 if (rc)
3071 goto err;
3072 }
3073
3074 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3075 rc = -EIO;
3076 goto err;
3077 }
3078
3079 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3080 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3081 readid_flags);
3082 if (rc)
3083 goto err;
3084
3085 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3086
3087 /* Configuration may have changed, reconfigure
3088 * transfer mode.
3089 */
3090 ehc->i.flags |= ATA_EHI_SETMODE;
3091
3092 /* schedule the scsi_rescan_device() here */
3093 schedule_delayed_work(&ap->scsi_rescan_task, 0);
3094 } else if (dev->class == ATA_DEV_UNKNOWN &&
3095 ehc->tries[dev->devno] &&
3096 ata_class_enabled(ehc->classes[dev->devno])) {
3097 /* Temporarily set dev->class, it will be
3098 * permanently set once all configurations are
3099 * complete. This is necessary because new
3100 * device configuration is done in two
3101 * separate loops.
3102 */
3103 dev->class = ehc->classes[dev->devno];
3104
3105 if (dev->class == ATA_DEV_PMP)
3106 rc = sata_pmp_attach(dev);
3107 else
3108 rc = ata_dev_read_id(dev, &dev->class,
3109 readid_flags, dev->id);
3110
3111 /* read_id might have changed class, store and reset */
3112 ehc->classes[dev->devno] = dev->class;
3113 dev->class = ATA_DEV_UNKNOWN;
3114
3115 switch (rc) {
3116 case 0:
3117 /* clear error info accumulated during probe */
3118 ata_ering_clear(&dev->ering);
3119 new_mask |= 1 << dev->devno;
3120 break;
3121 case -ENOENT:
3122 /* IDENTIFY was issued to non-existent
3123 * device. No need to reset. Just
3124 * thaw and ignore the device.
3125 */
3126 ata_eh_thaw_port(ap);
3127 break;
3128 default:
3129 goto err;
3130 }
3131 }
3132 }
3133
3134 /* PDIAG- should have been released, ask cable type if post-reset */
3135 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3136 if (ap->ops->cable_detect)
3137 ap->cbl = ap->ops->cable_detect(ap);
3138 ata_force_cbl(ap);
3139 }
3140
3141 /* Configure new devices forward such that user doesn't see
3142 * device detection messages backwards.
3143 */
3144 ata_for_each_dev(dev, link, ALL) {
3145 if (!(new_mask & (1 << dev->devno)))
3146 continue;
3147
3148 dev->class = ehc->classes[dev->devno];
3149
3150 if (dev->class == ATA_DEV_PMP)
3151 continue;
3152
3153 ehc->i.flags |= ATA_EHI_PRINTINFO;
3154 rc = ata_dev_configure(dev);
3155 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3156 if (rc) {
3157 dev->class = ATA_DEV_UNKNOWN;
3158 goto err;
3159 }
3160
3161 spin_lock_irqsave(ap->lock, flags);
3162 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3163 spin_unlock_irqrestore(ap->lock, flags);
3164
3165 /* new device discovered, configure xfermode */
3166 ehc->i.flags |= ATA_EHI_SETMODE;
3167 }
3168
3169 return 0;
3170
3171 err:
3172 *r_failed_dev = dev;
3173 return rc;
3174}
3175
3176/**
3177 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3178 * @link: link on which timings will be programmed
3179 * @r_failed_dev: out parameter for failed device
3180 *
3181 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3182 * ata_set_mode() fails, pointer to the failing device is
3183 * returned in @r_failed_dev.
3184 *
3185 * LOCKING:
3186 * PCI/etc. bus probe sem.
3187 *
3188 * RETURNS:
3189 * 0 on success, negative errno otherwise
3190 */
3191int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3192{
3193 struct ata_port *ap = link->ap;
3194 struct ata_device *dev;
3195 int rc;
3196
3197 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3198 ata_for_each_dev(dev, link, ENABLED) {
3199 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3200 struct ata_ering_entry *ent;
3201
3202 ent = ata_ering_top(&dev->ering);
3203 if (ent)
3204 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3205 }
3206 }
3207
3208 /* has private set_mode? */
3209 if (ap->ops->set_mode)
3210 rc = ap->ops->set_mode(link, r_failed_dev);
3211 else
3212 rc = ata_do_set_mode(link, r_failed_dev);
3213
3214 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3215 ata_for_each_dev(dev, link, ENABLED) {
3216 struct ata_eh_context *ehc = &link->eh_context;
3217 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3218 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3219
3220 if (dev->xfer_mode != saved_xfer_mode ||
3221 ata_ncq_enabled(dev) != saved_ncq)
3222 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3223 }
3224
3225 return rc;
3226}
3227
3228/**
3229 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3230 * @dev: ATAPI device to clear UA for
3231 *
3232 * Resets and other operations can make an ATAPI device raise
3233 * UNIT ATTENTION which causes the next operation to fail. This
3234 * function clears UA.
3235 *
3236 * LOCKING:
3237 * EH context (may sleep).
3238 *
3239 * RETURNS:
3240 * 0 on success, -errno on failure.
3241 */
3242static int atapi_eh_clear_ua(struct ata_device *dev)
3243{
3244 int i;
3245
3246 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3247 u8 *sense_buffer = dev->link->ap->sector_buf;
3248 u8 sense_key = 0;
3249 unsigned int err_mask;
3250
3251 err_mask = atapi_eh_tur(dev, &sense_key);
3252 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3253 ata_dev_warn(dev,
3254 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3255 err_mask);
3256 return -EIO;
3257 }
3258
3259 if (!err_mask || sense_key != UNIT_ATTENTION)
3260 return 0;
3261
3262 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3263 if (err_mask) {
3264 ata_dev_warn(dev, "failed to clear "
3265 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3266 return -EIO;
3267 }
3268 }
3269
3270 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3271 ATA_EH_UA_TRIES);
3272
3273 return 0;
3274}
3275
3276/**
3277 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3278 * @dev: ATA device which may need FLUSH retry
3279 *
3280 * If @dev failed FLUSH, it needs to be reported upper layer
3281 * immediately as it means that @dev failed to remap and already
3282 * lost at least a sector and further FLUSH retrials won't make
3283 * any difference to the lost sector. However, if FLUSH failed
3284 * for other reasons, for example transmission error, FLUSH needs
3285 * to be retried.
3286 *
3287 * This function determines whether FLUSH failure retry is
3288 * necessary and performs it if so.
3289 *
3290 * RETURNS:
3291 * 0 if EH can continue, -errno if EH needs to be repeated.
3292 */
3293static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3294{
3295 struct ata_link *link = dev->link;
3296 struct ata_port *ap = link->ap;
3297 struct ata_queued_cmd *qc;
3298 struct ata_taskfile tf;
3299 unsigned int err_mask;
3300 int rc = 0;
3301
3302 /* did flush fail for this device? */
3303 if (!ata_tag_valid(link->active_tag))
3304 return 0;
3305
3306 qc = __ata_qc_from_tag(ap, link->active_tag);
3307 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3308 qc->tf.command != ATA_CMD_FLUSH))
3309 return 0;
3310
3311 /* if the device failed it, it should be reported to upper layers */
3312 if (qc->err_mask & AC_ERR_DEV)
3313 return 0;
3314
3315 /* flush failed for some other reason, give it another shot */
3316 ata_tf_init(dev, &tf);
3317
3318 tf.command = qc->tf.command;
3319 tf.flags |= ATA_TFLAG_DEVICE;
3320 tf.protocol = ATA_PROT_NODATA;
3321
3322 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3323 tf.command, qc->err_mask);
3324
3325 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3326 if (!err_mask) {
3327 /*
3328 * FLUSH is complete but there's no way to
3329 * successfully complete a failed command from EH.
3330 * Making sure retry is allowed at least once and
3331 * retrying it should do the trick - whatever was in
3332 * the cache is already on the platter and this won't
3333 * cause infinite loop.
3334 */
3335 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3336 } else {
3337 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3338 err_mask);
3339 rc = -EIO;
3340
3341 /* if device failed it, report it to upper layers */
3342 if (err_mask & AC_ERR_DEV) {
3343 qc->err_mask |= AC_ERR_DEV;
3344 qc->result_tf = tf;
3345 if (!ata_port_is_frozen(ap))
3346 rc = 0;
3347 }
3348 }
3349 return rc;
3350}
3351
3352/**
3353 * ata_eh_set_lpm - configure SATA interface power management
3354 * @link: link to configure power management
3355 * @policy: the link power management policy
3356 * @r_failed_dev: out parameter for failed device
3357 *
3358 * Enable SATA Interface power management. This will enable
3359 * Device Interface Power Management (DIPM) for min_power and
3360 * medium_power_with_dipm policies, and then call driver specific
3361 * callbacks for enabling Host Initiated Power management.
3362 *
3363 * LOCKING:
3364 * EH context.
3365 *
3366 * RETURNS:
3367 * 0 on success, -errno on failure.
3368 */
3369static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3370 struct ata_device **r_failed_dev)
3371{
3372 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3373 struct ata_eh_context *ehc = &link->eh_context;
3374 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3375 enum ata_lpm_policy old_policy = link->lpm_policy;
3376 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3377 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3378 unsigned int err_mask;
3379 int rc;
3380
3381 /* if the link or host doesn't do LPM, noop */
3382 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3383 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3384 return 0;
3385
3386 /*
3387 * DIPM is enabled only for MIN_POWER as some devices
3388 * misbehave when the host NACKs transition to SLUMBER. Order
3389 * device and link configurations such that the host always
3390 * allows DIPM requests.
3391 */
3392 ata_for_each_dev(dev, link, ENABLED) {
3393 bool hipm = ata_id_has_hipm(dev->id);
3394 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3395
3396 /* find the first enabled and LPM enabled devices */
3397 if (!link_dev)
3398 link_dev = dev;
3399
3400 if (!lpm_dev && (hipm || dipm))
3401 lpm_dev = dev;
3402
3403 hints &= ~ATA_LPM_EMPTY;
3404 if (!hipm)
3405 hints &= ~ATA_LPM_HIPM;
3406
3407 /* disable DIPM before changing link config */
3408 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3409 err_mask = ata_dev_set_feature(dev,
3410 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3411 if (err_mask && err_mask != AC_ERR_DEV) {
3412 ata_dev_warn(dev,
3413 "failed to disable DIPM, Emask 0x%x\n",
3414 err_mask);
3415 rc = -EIO;
3416 goto fail;
3417 }
3418 }
3419 }
3420
3421 if (ap) {
3422 rc = ap->ops->set_lpm(link, policy, hints);
3423 if (!rc && ap->slave_link)
3424 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3425 } else
3426 rc = sata_pmp_set_lpm(link, policy, hints);
3427
3428 /*
3429 * Attribute link config failure to the first (LPM) enabled
3430 * device on the link.
3431 */
3432 if (rc) {
3433 if (rc == -EOPNOTSUPP) {
3434 link->flags |= ATA_LFLAG_NO_LPM;
3435 return 0;
3436 }
3437 dev = lpm_dev ? lpm_dev : link_dev;
3438 goto fail;
3439 }
3440
3441 /*
3442 * Low level driver acked the transition. Issue DIPM command
3443 * with the new policy set.
3444 */
3445 link->lpm_policy = policy;
3446 if (ap && ap->slave_link)
3447 ap->slave_link->lpm_policy = policy;
3448
3449 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3450 ata_for_each_dev(dev, link, ENABLED) {
3451 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3452 ata_id_has_dipm(dev->id)) {
3453 err_mask = ata_dev_set_feature(dev,
3454 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3455 if (err_mask && err_mask != AC_ERR_DEV) {
3456 ata_dev_warn(dev,
3457 "failed to enable DIPM, Emask 0x%x\n",
3458 err_mask);
3459 rc = -EIO;
3460 goto fail;
3461 }
3462 }
3463 }
3464
3465 link->last_lpm_change = jiffies;
3466 link->flags |= ATA_LFLAG_CHANGED;
3467
3468 return 0;
3469
3470fail:
3471 /* restore the old policy */
3472 link->lpm_policy = old_policy;
3473 if (ap && ap->slave_link)
3474 ap->slave_link->lpm_policy = old_policy;
3475
3476 /* if no device or only one more chance is left, disable LPM */
3477 if (!dev || ehc->tries[dev->devno] <= 2) {
3478 ata_link_warn(link, "disabling LPM on the link\n");
3479 link->flags |= ATA_LFLAG_NO_LPM;
3480 }
3481 if (r_failed_dev)
3482 *r_failed_dev = dev;
3483 return rc;
3484}
3485
3486int ata_link_nr_enabled(struct ata_link *link)
3487{
3488 struct ata_device *dev;
3489 int cnt = 0;
3490
3491 ata_for_each_dev(dev, link, ENABLED)
3492 cnt++;
3493 return cnt;
3494}
3495
3496static int ata_link_nr_vacant(struct ata_link *link)
3497{
3498 struct ata_device *dev;
3499 int cnt = 0;
3500
3501 ata_for_each_dev(dev, link, ALL)
3502 if (dev->class == ATA_DEV_UNKNOWN)
3503 cnt++;
3504 return cnt;
3505}
3506
3507static int ata_eh_skip_recovery(struct ata_link *link)
3508{
3509 struct ata_port *ap = link->ap;
3510 struct ata_eh_context *ehc = &link->eh_context;
3511 struct ata_device *dev;
3512
3513 /* skip disabled links */
3514 if (link->flags & ATA_LFLAG_DISABLED)
3515 return 1;
3516
3517 /* skip if explicitly requested */
3518 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3519 return 1;
3520
3521 /* thaw frozen port and recover failed devices */
3522 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3523 return 0;
3524
3525 /* reset at least once if reset is requested */
3526 if ((ehc->i.action & ATA_EH_RESET) &&
3527 !(ehc->i.flags & ATA_EHI_DID_RESET))
3528 return 0;
3529
3530 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3531 ata_for_each_dev(dev, link, ALL) {
3532 if (dev->class == ATA_DEV_UNKNOWN &&
3533 ehc->classes[dev->devno] != ATA_DEV_NONE)
3534 return 0;
3535 }
3536
3537 return 1;
3538}
3539
3540static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3541{
3542 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3543 u64 now = get_jiffies_64();
3544 int *trials = void_arg;
3545
3546 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3547 (ent->timestamp < now - min(now, interval)))
3548 return -1;
3549
3550 (*trials)++;
3551 return 0;
3552}
3553
3554static int ata_eh_schedule_probe(struct ata_device *dev)
3555{
3556 struct ata_eh_context *ehc = &dev->link->eh_context;
3557 struct ata_link *link = ata_dev_phys_link(dev);
3558 int trials = 0;
3559
3560 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3561 (ehc->did_probe_mask & (1 << dev->devno)))
3562 return 0;
3563
3564 ata_eh_detach_dev(dev);
3565 ata_dev_init(dev);
3566 ehc->did_probe_mask |= (1 << dev->devno);
3567 ehc->i.action |= ATA_EH_RESET;
3568 ehc->saved_xfer_mode[dev->devno] = 0;
3569 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3570
3571 /* the link maybe in a deep sleep, wake it up */
3572 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3573 if (ata_is_host_link(link))
3574 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3575 ATA_LPM_EMPTY);
3576 else
3577 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3578 ATA_LPM_EMPTY);
3579 }
3580
3581 /* Record and count probe trials on the ering. The specific
3582 * error mask used is irrelevant. Because a successful device
3583 * detection clears the ering, this count accumulates only if
3584 * there are consecutive failed probes.
3585 *
3586 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3587 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3588 * forced to 1.5Gbps.
3589 *
3590 * This is to work around cases where failed link speed
3591 * negotiation results in device misdetection leading to
3592 * infinite DEVXCHG or PHRDY CHG events.
3593 */
3594 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3595 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3596
3597 if (trials > ATA_EH_PROBE_TRIALS)
3598 sata_down_spd_limit(link, 1);
3599
3600 return 1;
3601}
3602
3603static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3604{
3605 struct ata_eh_context *ehc = &dev->link->eh_context;
3606
3607 /* -EAGAIN from EH routine indicates retry without prejudice.
3608 * The requester is responsible for ensuring forward progress.
3609 */
3610 if (err != -EAGAIN)
3611 ehc->tries[dev->devno]--;
3612
3613 switch (err) {
3614 case -ENODEV:
3615 /* device missing or wrong IDENTIFY data, schedule probing */
3616 ehc->i.probe_mask |= (1 << dev->devno);
3617 fallthrough;
3618 case -EINVAL:
3619 /* give it just one more chance */
3620 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3621 fallthrough;
3622 case -EIO:
3623 if (ehc->tries[dev->devno] == 1) {
3624 /* This is the last chance, better to slow
3625 * down than lose it.
3626 */
3627 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3628 if (dev->pio_mode > XFER_PIO_0)
3629 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3630 }
3631 }
3632
3633 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3634 /* disable device if it has used up all its chances */
3635 ata_dev_disable(dev);
3636
3637 /* detach if offline */
3638 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3639 ata_eh_detach_dev(dev);
3640
3641 /* schedule probe if necessary */
3642 if (ata_eh_schedule_probe(dev)) {
3643 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3644 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3645 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3646 }
3647
3648 return 1;
3649 } else {
3650 ehc->i.action |= ATA_EH_RESET;
3651 return 0;
3652 }
3653}
3654
3655/**
3656 * ata_eh_recover - recover host port after error
3657 * @ap: host port to recover
3658 * @prereset: prereset method (can be NULL)
3659 * @softreset: softreset method (can be NULL)
3660 * @hardreset: hardreset method (can be NULL)
3661 * @postreset: postreset method (can be NULL)
3662 * @r_failed_link: out parameter for failed link
3663 *
3664 * This is the alpha and omega, eum and yang, heart and soul of
3665 * libata exception handling. On entry, actions required to
3666 * recover each link and hotplug requests are recorded in the
3667 * link's eh_context. This function executes all the operations
3668 * with appropriate retrials and fallbacks to resurrect failed
3669 * devices, detach goners and greet newcomers.
3670 *
3671 * LOCKING:
3672 * Kernel thread context (may sleep).
3673 *
3674 * RETURNS:
3675 * 0 on success, -errno on failure.
3676 */
3677int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3678 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3679 ata_postreset_fn_t postreset,
3680 struct ata_link **r_failed_link)
3681{
3682 struct ata_link *link;
3683 struct ata_device *dev;
3684 int rc, nr_fails;
3685 unsigned long flags, deadline;
3686
3687 /* prep for recovery */
3688 ata_for_each_link(link, ap, EDGE) {
3689 struct ata_eh_context *ehc = &link->eh_context;
3690
3691 /* re-enable link? */
3692 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3693 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3694 spin_lock_irqsave(ap->lock, flags);
3695 link->flags &= ~ATA_LFLAG_DISABLED;
3696 spin_unlock_irqrestore(ap->lock, flags);
3697 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3698 }
3699
3700 ata_for_each_dev(dev, link, ALL) {
3701 if (link->flags & ATA_LFLAG_NO_RETRY)
3702 ehc->tries[dev->devno] = 1;
3703 else
3704 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3705
3706 /* collect port action mask recorded in dev actions */
3707 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3708 ~ATA_EH_PERDEV_MASK;
3709 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3710
3711 /* process hotplug request */
3712 if (dev->flags & ATA_DFLAG_DETACH)
3713 ata_eh_detach_dev(dev);
3714
3715 /* schedule probe if necessary */
3716 if (!ata_dev_enabled(dev))
3717 ata_eh_schedule_probe(dev);
3718 }
3719 }
3720
3721 retry:
3722 rc = 0;
3723
3724 /* if UNLOADING, finish immediately */
3725 if (ap->pflags & ATA_PFLAG_UNLOADING)
3726 goto out;
3727
3728 /* prep for EH */
3729 ata_for_each_link(link, ap, EDGE) {
3730 struct ata_eh_context *ehc = &link->eh_context;
3731
3732 /* skip EH if possible. */
3733 if (ata_eh_skip_recovery(link))
3734 ehc->i.action = 0;
3735
3736 ata_for_each_dev(dev, link, ALL)
3737 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3738 }
3739
3740 /* reset */
3741 ata_for_each_link(link, ap, EDGE) {
3742 struct ata_eh_context *ehc = &link->eh_context;
3743
3744 if (!(ehc->i.action & ATA_EH_RESET))
3745 continue;
3746
3747 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3748 prereset, softreset, hardreset, postreset);
3749 if (rc) {
3750 ata_link_err(link, "reset failed, giving up\n");
3751 goto out;
3752 }
3753 }
3754
3755 do {
3756 unsigned long now;
3757
3758 /*
3759 * clears ATA_EH_PARK in eh_info and resets
3760 * ap->park_req_pending
3761 */
3762 ata_eh_pull_park_action(ap);
3763
3764 deadline = jiffies;
3765 ata_for_each_link(link, ap, EDGE) {
3766 ata_for_each_dev(dev, link, ALL) {
3767 struct ata_eh_context *ehc = &link->eh_context;
3768 unsigned long tmp;
3769
3770 if (dev->class != ATA_DEV_ATA &&
3771 dev->class != ATA_DEV_ZAC)
3772 continue;
3773 if (!(ehc->i.dev_action[dev->devno] &
3774 ATA_EH_PARK))
3775 continue;
3776 tmp = dev->unpark_deadline;
3777 if (time_before(deadline, tmp))
3778 deadline = tmp;
3779 else if (time_before_eq(tmp, jiffies))
3780 continue;
3781 if (ehc->unloaded_mask & (1 << dev->devno))
3782 continue;
3783
3784 ata_eh_park_issue_cmd(dev, 1);
3785 }
3786 }
3787
3788 now = jiffies;
3789 if (time_before_eq(deadline, now))
3790 break;
3791
3792 ata_eh_release(ap);
3793 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3794 deadline - now);
3795 ata_eh_acquire(ap);
3796 } while (deadline);
3797 ata_for_each_link(link, ap, EDGE) {
3798 ata_for_each_dev(dev, link, ALL) {
3799 if (!(link->eh_context.unloaded_mask &
3800 (1 << dev->devno)))
3801 continue;
3802
3803 ata_eh_park_issue_cmd(dev, 0);
3804 ata_eh_done(link, dev, ATA_EH_PARK);
3805 }
3806 }
3807
3808 /* the rest */
3809 nr_fails = 0;
3810 ata_for_each_link(link, ap, PMP_FIRST) {
3811 struct ata_eh_context *ehc = &link->eh_context;
3812
3813 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3814 goto config_lpm;
3815
3816 /* revalidate existing devices and attach new ones */
3817 rc = ata_eh_revalidate_and_attach(link, &dev);
3818 if (rc)
3819 goto rest_fail;
3820
3821 /* if PMP got attached, return, pmp EH will take care of it */
3822 if (link->device->class == ATA_DEV_PMP) {
3823 ehc->i.action = 0;
3824 return 0;
3825 }
3826
3827 /* configure transfer mode if necessary */
3828 if (ehc->i.flags & ATA_EHI_SETMODE) {
3829 rc = ata_set_mode(link, &dev);
3830 if (rc)
3831 goto rest_fail;
3832 ehc->i.flags &= ~ATA_EHI_SETMODE;
3833 }
3834
3835 /* If reset has been issued, clear UA to avoid
3836 * disrupting the current users of the device.
3837 */
3838 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3839 ata_for_each_dev(dev, link, ALL) {
3840 if (dev->class != ATA_DEV_ATAPI)
3841 continue;
3842 rc = atapi_eh_clear_ua(dev);
3843 if (rc)
3844 goto rest_fail;
3845 if (zpodd_dev_enabled(dev))
3846 zpodd_post_poweron(dev);
3847 }
3848 }
3849
3850 /*
3851 * Make sure to transition devices to the active power mode
3852 * if needed (e.g. if we were scheduled on system resume).
3853 */
3854 ata_for_each_dev(dev, link, ENABLED) {
3855 if (ehc->i.dev_action[dev->devno] & ATA_EH_SET_ACTIVE) {
3856 ata_dev_power_set_active(dev);
3857 ata_eh_done(link, dev, ATA_EH_SET_ACTIVE);
3858 }
3859 }
3860
3861 /* retry flush if necessary */
3862 ata_for_each_dev(dev, link, ALL) {
3863 if (dev->class != ATA_DEV_ATA &&
3864 dev->class != ATA_DEV_ZAC)
3865 continue;
3866 rc = ata_eh_maybe_retry_flush(dev);
3867 if (rc)
3868 goto rest_fail;
3869 }
3870
3871 config_lpm:
3872 /* configure link power saving */
3873 if (link->lpm_policy != ap->target_lpm_policy) {
3874 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3875 if (rc)
3876 goto rest_fail;
3877 }
3878
3879 /* this link is okay now */
3880 ehc->i.flags = 0;
3881 continue;
3882
3883 rest_fail:
3884 nr_fails++;
3885 if (dev)
3886 ata_eh_handle_dev_fail(dev, rc);
3887
3888 if (ata_port_is_frozen(ap)) {
3889 /* PMP reset requires working host port.
3890 * Can't retry if it's frozen.
3891 */
3892 if (sata_pmp_attached(ap))
3893 goto out;
3894 break;
3895 }
3896 }
3897
3898 if (nr_fails)
3899 goto retry;
3900
3901 out:
3902 if (rc && r_failed_link)
3903 *r_failed_link = link;
3904
3905 return rc;
3906}
3907
3908/**
3909 * ata_eh_finish - finish up EH
3910 * @ap: host port to finish EH for
3911 *
3912 * Recovery is complete. Clean up EH states and retry or finish
3913 * failed qcs.
3914 *
3915 * LOCKING:
3916 * None.
3917 */
3918void ata_eh_finish(struct ata_port *ap)
3919{
3920 struct ata_queued_cmd *qc;
3921 int tag;
3922
3923 /* retry or finish qcs */
3924 ata_qc_for_each_raw(ap, qc, tag) {
3925 if (!(qc->flags & ATA_QCFLAG_EH))
3926 continue;
3927
3928 if (qc->err_mask) {
3929 /* FIXME: Once EH migration is complete,
3930 * generate sense data in this function,
3931 * considering both err_mask and tf.
3932 */
3933 if (qc->flags & ATA_QCFLAG_RETRY) {
3934 /*
3935 * Since qc->err_mask is set, ata_eh_qc_retry()
3936 * will not increment scmd->allowed, so upper
3937 * layer will only retry the command if it has
3938 * not already been retried too many times.
3939 */
3940 ata_eh_qc_retry(qc);
3941 } else {
3942 ata_eh_qc_complete(qc);
3943 }
3944 } else {
3945 if (qc->flags & ATA_QCFLAG_SENSE_VALID ||
3946 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) {
3947 ata_eh_qc_complete(qc);
3948 } else {
3949 /* feed zero TF to sense generation */
3950 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3951 /*
3952 * Since qc->err_mask is not set,
3953 * ata_eh_qc_retry() will increment
3954 * scmd->allowed, so upper layer is guaranteed
3955 * to retry the command.
3956 */
3957 ata_eh_qc_retry(qc);
3958 }
3959 }
3960 }
3961
3962 /* make sure nr_active_links is zero after EH */
3963 WARN_ON(ap->nr_active_links);
3964 ap->nr_active_links = 0;
3965}
3966
3967/**
3968 * ata_do_eh - do standard error handling
3969 * @ap: host port to handle error for
3970 *
3971 * @prereset: prereset method (can be NULL)
3972 * @softreset: softreset method (can be NULL)
3973 * @hardreset: hardreset method (can be NULL)
3974 * @postreset: postreset method (can be NULL)
3975 *
3976 * Perform standard error handling sequence.
3977 *
3978 * LOCKING:
3979 * Kernel thread context (may sleep).
3980 */
3981void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3982 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3983 ata_postreset_fn_t postreset)
3984{
3985 struct ata_device *dev;
3986 int rc;
3987
3988 ata_eh_autopsy(ap);
3989 ata_eh_report(ap);
3990
3991 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3992 NULL);
3993 if (rc) {
3994 ata_for_each_dev(dev, &ap->link, ALL)
3995 ata_dev_disable(dev);
3996 }
3997
3998 ata_eh_finish(ap);
3999}
4000
4001/**
4002 * ata_std_error_handler - standard error handler
4003 * @ap: host port to handle error for
4004 *
4005 * Standard error handler
4006 *
4007 * LOCKING:
4008 * Kernel thread context (may sleep).
4009 */
4010void ata_std_error_handler(struct ata_port *ap)
4011{
4012 struct ata_port_operations *ops = ap->ops;
4013 ata_reset_fn_t hardreset = ops->hardreset;
4014
4015 /* ignore built-in hardreset if SCR access is not available */
4016 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4017 hardreset = NULL;
4018
4019 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4020}
4021EXPORT_SYMBOL_GPL(ata_std_error_handler);
4022
4023#ifdef CONFIG_PM
4024/**
4025 * ata_eh_handle_port_suspend - perform port suspend operation
4026 * @ap: port to suspend
4027 *
4028 * Suspend @ap.
4029 *
4030 * LOCKING:
4031 * Kernel thread context (may sleep).
4032 */
4033static void ata_eh_handle_port_suspend(struct ata_port *ap)
4034{
4035 unsigned long flags;
4036 int rc = 0;
4037 struct ata_device *dev;
4038 struct ata_link *link;
4039
4040 /* are we suspending? */
4041 spin_lock_irqsave(ap->lock, flags);
4042 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4043 ap->pm_mesg.event & PM_EVENT_RESUME) {
4044 spin_unlock_irqrestore(ap->lock, flags);
4045 return;
4046 }
4047 spin_unlock_irqrestore(ap->lock, flags);
4048
4049 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4050
4051 /* Set all devices attached to the port in standby mode */
4052 ata_for_each_link(link, ap, HOST_FIRST) {
4053 ata_for_each_dev(dev, link, ENABLED)
4054 ata_dev_power_set_standby(dev);
4055 }
4056
4057 /*
4058 * If we have a ZPODD attached, check its zero
4059 * power ready status before the port is frozen.
4060 * Only needed for runtime suspend.
4061 */
4062 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4063 ata_for_each_dev(dev, &ap->link, ENABLED) {
4064 if (zpodd_dev_enabled(dev))
4065 zpodd_on_suspend(dev);
4066 }
4067 }
4068
4069 /* suspend */
4070 ata_eh_freeze_port(ap);
4071
4072 if (ap->ops->port_suspend)
4073 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4074
4075 ata_acpi_set_state(ap, ap->pm_mesg);
4076
4077 /* update the flags */
4078 spin_lock_irqsave(ap->lock, flags);
4079
4080 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4081 if (rc == 0)
4082 ap->pflags |= ATA_PFLAG_SUSPENDED;
4083 else if (ata_port_is_frozen(ap))
4084 ata_port_schedule_eh(ap);
4085
4086 spin_unlock_irqrestore(ap->lock, flags);
4087
4088 return;
4089}
4090
4091/**
4092 * ata_eh_handle_port_resume - perform port resume operation
4093 * @ap: port to resume
4094 *
4095 * Resume @ap.
4096 *
4097 * LOCKING:
4098 * Kernel thread context (may sleep).
4099 */
4100static void ata_eh_handle_port_resume(struct ata_port *ap)
4101{
4102 struct ata_link *link;
4103 struct ata_device *dev;
4104 unsigned long flags;
4105
4106 /* are we resuming? */
4107 spin_lock_irqsave(ap->lock, flags);
4108 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4109 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4110 spin_unlock_irqrestore(ap->lock, flags);
4111 return;
4112 }
4113 spin_unlock_irqrestore(ap->lock, flags);
4114
4115 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4116
4117 /*
4118 * Error timestamps are in jiffies which doesn't run while
4119 * suspended and PHY events during resume isn't too uncommon.
4120 * When the two are combined, it can lead to unnecessary speed
4121 * downs if the machine is suspended and resumed repeatedly.
4122 * Clear error history.
4123 */
4124 ata_for_each_link(link, ap, HOST_FIRST)
4125 ata_for_each_dev(dev, link, ALL)
4126 ata_ering_clear(&dev->ering);
4127
4128 ata_acpi_set_state(ap, ap->pm_mesg);
4129
4130 if (ap->ops->port_resume)
4131 ap->ops->port_resume(ap);
4132
4133 /* tell ACPI that we're resuming */
4134 ata_acpi_on_resume(ap);
4135
4136 /* update the flags */
4137 spin_lock_irqsave(ap->lock, flags);
4138 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4139 ap->pflags |= ATA_PFLAG_RESUMING;
4140 spin_unlock_irqrestore(ap->lock, flags);
4141}
4142#endif /* CONFIG_PM */