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