projects/linux·blob master · refs

4268 lines · 114.2 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * libata-eh.c - libata error handling
4 *
5 * Copyright 2006 Tejun Heo <[email protected]>
6 *
7 * libata documentation is available via 'make {ps|pdf}docs',
8 * as Documentation/driver-api/libata.rst
9 *
10 * Hardware documentation available from http://www.t13.org/ and
11 * http://www.sata-io.org/
12 */
13
14#include <linux/kernel.h>
15#include <linux/blkdev.h>
16#include <linux/export.h>
17#include <linux/pci.h>
18#include <scsi/scsi.h>
19#include <scsi/scsi_host.h>
20#include <scsi/scsi_eh.h>
21#include <scsi/scsi_device.h>
22#include <scsi/scsi_cmnd.h>
23#include <scsi/scsi_dbg.h>
24#include "../scsi/scsi_transport_api.h"
25
26#include <linux/libata.h>
27
28#include <trace/events/libata.h>
29#include "libata.h"
30
31enum {
32 /* speed down verdicts */
33 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
34 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
35 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
36 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
37
38 /* error flags */
39 ATA_EFLAG_IS_IO = (1 << 0),
40 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
41 ATA_EFLAG_OLD_ER = (1 << 31),
42
43 /* error categories */
44 ATA_ECAT_NONE = 0,
45 ATA_ECAT_ATA_BUS = 1,
46 ATA_ECAT_TOUT_HSM = 2,
47 ATA_ECAT_UNK_DEV = 3,
48 ATA_ECAT_DUBIOUS_NONE = 4,
49 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
50 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
51 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
52 ATA_ECAT_NR = 8,
53
54 ATA_EH_CMD_DFL_TIMEOUT = 5000,
55
56 /* always put at least this amount of time between resets */
57 ATA_EH_RESET_COOL_DOWN = 5000,
58
59 /* Waiting in ->prereset can never be reliable. It's
60 * sometimes nice to wait there but it can't be depended upon;
61 * otherwise, we wouldn't be resetting. Just give it enough
62 * time for most drives to spin up.
63 */
64 ATA_EH_PRERESET_TIMEOUT = 10000,
65 ATA_EH_FASTDRAIN_INTERVAL = 3000,
66
67 ATA_EH_UA_TRIES = 5,
68
69 /* probe speed down parameters, see ata_eh_schedule_probe() */
70 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
71 ATA_EH_PROBE_TRIALS = 2,
72};
73
74/* The following table determines how we sequence resets. Each entry
75 * represents timeout for that try. The first try can be soft or
76 * hardreset. All others are hardreset if available. In most cases
77 * the first reset w/ 10sec timeout should succeed. Following entries
78 * are mostly for error handling, hotplug and those outlier devices that
79 * take an exceptionally long time to recover from reset.
80 */
81static const unsigned int ata_eh_reset_timeouts[] = {
82 10000, /* most drives spin up by 10sec */
83 10000, /* > 99% working drives spin up before 20sec */
84 35000, /* give > 30 secs of idleness for outlier devices */
85 5000, /* and sweet one last chance */
86 UINT_MAX, /* > 1 min has elapsed, give up */
87};
88
89static const unsigned int ata_eh_identify_timeouts[] = {
90 5000, /* covers > 99% of successes and not too boring on failures */
91 10000, /* combined time till here is enough even for media access */
92 30000, /* for true idiots */
93 UINT_MAX,
94};
95
96static const unsigned int ata_eh_revalidate_timeouts[] = {
97 15000, /* Some drives are slow to read log pages when waking-up */
98 15000, /* combined time till here is enough even for media access */
99 UINT_MAX,
100};
101
102static const unsigned int ata_eh_flush_timeouts[] = {
103 15000, /* be generous with flush */
104 15000, /* ditto */
105 30000, /* and even more generous */
106 UINT_MAX,
107};
108
109static const unsigned int ata_eh_standby_timeouts[] = {
110 15000, /* Some drives may be slow to standby */
111 /* but don't hold up a suspend too long waiting for them */
112 UINT_MAX,
113};
114
115static const unsigned int ata_eh_other_timeouts[] = {
116 5000, /* same rationale as identify timeout */
117 10000, /* ditto */
118 /* but no merciful 30sec for other commands, it just isn't worth it */
119 UINT_MAX,
120};
121
122struct ata_eh_cmd_timeout_ent {
123 const u8 *commands;
124 const unsigned int *timeouts;
125};
126
127/* The following table determines timeouts to use for EH internal
128 * commands. Each table entry is a command class and matches the
129 * commands the entry applies to and the timeout table to use.
130 *
131 * On the retry after a command timed out, the next timeout value from
132 * the table is used. If the table doesn't contain further entries,
133 * the last value is used.
134 *
135 * ehc->cmd_timeout_idx keeps track of which timeout to use per
136 * command class, so if SET_FEATURES times out on the first try, the
137 * next try will use the second timeout value only for that class.
138 */
139#define CMDS(cmds...) (const u8 []){ cmds, 0 }
140static const struct ata_eh_cmd_timeout_ent
141ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
142 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
143 .timeouts = ata_eh_identify_timeouts, },
144 { .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT),
145 .timeouts = ata_eh_revalidate_timeouts, },
146 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
149 .timeouts = ata_eh_other_timeouts, },
150 { .commands = CMDS(ATA_CMD_SET_FEATURES),
151 .timeouts = ata_eh_other_timeouts, },
152 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
153 .timeouts = ata_eh_other_timeouts, },
154 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
155 .timeouts = ata_eh_flush_timeouts },
156 { .commands = CMDS(ATA_CMD_STANDBYNOW1),
157 .timeouts = ata_eh_standby_timeouts },
158 { .commands = CMDS(ATA_CMD_VERIFY),
159 .timeouts = ata_eh_reset_timeouts },
160};
161#undef CMDS
162
163static void __ata_port_freeze(struct ata_port *ap);
164#ifdef CONFIG_PM
165static void ata_eh_handle_port_suspend(struct ata_port *ap);
166static void ata_eh_handle_port_resume(struct ata_port *ap);
167#else /* CONFIG_PM */
168static void ata_eh_handle_port_suspend(struct ata_port *ap)
169{ }
170
171static void ata_eh_handle_port_resume(struct ata_port *ap)
172{ }
173#endif /* CONFIG_PM */
174
175static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
176 const char *fmt, va_list args)
177{
178 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
179 ATA_EH_DESC_LEN - ehi->desc_len,
180 fmt, args);
181}
182
183/**
184 * __ata_ehi_push_desc - push error description without adding separator
185 * @ehi: target EHI
186 * @fmt: printf format string
187 *
188 * Format string according to @fmt and append it to @ehi->desc.
189 *
190 * LOCKING:
191 * spin_lock_irqsave(host lock)
192 */
193void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
194{
195 va_list args;
196
197 va_start(args, fmt);
198 __ata_ehi_pushv_desc(ehi, fmt, args);
199 va_end(args);
200}
201EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
202
203/**
204 * ata_ehi_push_desc - push error description with separator
205 * @ehi: target EHI
206 * @fmt: printf format string
207 *
208 * Format string according to @fmt and append it to @ehi->desc.
209 * If @ehi->desc is not empty, ", " is added in-between.
210 *
211 * LOCKING:
212 * spin_lock_irqsave(host lock)
213 */
214void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
215{
216 va_list args;
217
218 if (ehi->desc_len)
219 __ata_ehi_push_desc(ehi, ", ");
220
221 va_start(args, fmt);
222 __ata_ehi_pushv_desc(ehi, fmt, args);
223 va_end(args);
224}
225EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
226
227/**
228 * ata_ehi_clear_desc - clean error description
229 * @ehi: target EHI
230 *
231 * Clear @ehi->desc.
232 *
233 * LOCKING:
234 * spin_lock_irqsave(host lock)
235 */
236void ata_ehi_clear_desc(struct ata_eh_info *ehi)
237{
238 ehi->desc[0] = '\0';
239 ehi->desc_len = 0;
240}
241EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
242
243/**
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
247 *
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
252 *
253 * LOCKING:
254 * None.
255 */
256void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
257{
258 va_list args;
259
260 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
261
262 if (ap->link.eh_info.desc_len)
263 __ata_ehi_push_desc(&ap->link.eh_info, " ");
264
265 va_start(args, fmt);
266 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
267 va_end(args);
268}
269EXPORT_SYMBOL_GPL(ata_port_desc);
270
271#ifdef CONFIG_PCI
272/**
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
278 *
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
283 *
284 * LOCKING:
285 * None.
286 */
287void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
288 const char *name)
289{
290 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
291 char *type = "";
292 unsigned long long start, len;
293
294 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
295 type = "m";
296 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
297 type = "i";
298
299 start = (unsigned long long)pci_resource_start(pdev, bar);
300 len = (unsigned long long)pci_resource_len(pdev, bar);
301
302 if (offset < 0)
303 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
304 else
305 ata_port_desc(ap, "%s 0x%llx", name,
306 start + (unsigned long long)offset);
307}
308EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
309#endif /* CONFIG_PCI */
310
311static int ata_lookup_timeout_table(u8 cmd)
312{
313 int i;
314
315 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
316 const u8 *cur;
317
318 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
319 if (*cur == cmd)
320 return i;
321 }
322
323 return -1;
324}
325
326/**
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
330 *
331 * Determine timeout for internal command @cmd for @dev.
332 *
333 * LOCKING:
334 * EH context.
335 *
336 * RETURNS:
337 * Determined timeout.
338 */
339unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
340{
341 struct ata_eh_context *ehc = &dev->link->eh_context;
342 int ent = ata_lookup_timeout_table(cmd);
343 int idx;
344
345 if (ent < 0)
346 return ATA_EH_CMD_DFL_TIMEOUT;
347
348 idx = ehc->cmd_timeout_idx[dev->devno][ent];
349 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
350}
351
352/**
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
356 *
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
360 *
361 * LOCKING:
362 * EH context.
363 */
364void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
365{
366 struct ata_eh_context *ehc = &dev->link->eh_context;
367 int ent = ata_lookup_timeout_table(cmd);
368 int idx;
369
370 if (ent < 0)
371 return;
372
373 idx = ehc->cmd_timeout_idx[dev->devno][ent];
374 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
375 ehc->cmd_timeout_idx[dev->devno][ent]++;
376}
377
378static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
379 unsigned int err_mask)
380{
381 struct ata_ering_entry *ent;
382
383 WARN_ON(!err_mask);
384
385 ering->cursor++;
386 ering->cursor %= ATA_ERING_SIZE;
387
388 ent = &ering->ring[ering->cursor];
389 ent->eflags = eflags;
390 ent->err_mask = err_mask;
391 ent->timestamp = get_jiffies_64();
392}
393
394static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
395{
396 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
397
398 if (ent->err_mask)
399 return ent;
400 return NULL;
401}
402
403int ata_ering_map(struct ata_ering *ering,
404 int (*map_fn)(struct ata_ering_entry *, void *),
405 void *arg)
406{
407 int idx, rc = 0;
408 struct ata_ering_entry *ent;
409
410 idx = ering->cursor;
411 do {
412 ent = &ering->ring[idx];
413 if (!ent->err_mask)
414 break;
415 rc = map_fn(ent, arg);
416 if (rc)
417 break;
418 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
419 } while (idx != ering->cursor);
420
421 return rc;
422}
423
424static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
425{
426 ent->eflags |= ATA_EFLAG_OLD_ER;
427 return 0;
428}
429
430static void ata_ering_clear(struct ata_ering *ering)
431{
432 ata_ering_map(ering, ata_ering_clear_cb, NULL);
433}
434
435static unsigned int ata_eh_dev_action(struct ata_device *dev)
436{
437 struct ata_eh_context *ehc = &dev->link->eh_context;
438
439 return ehc->i.action | ehc->i.dev_action[dev->devno];
440}
441
442static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
443 struct ata_eh_info *ehi, unsigned int action)
444{
445 struct ata_device *tdev;
446
447 if (!dev) {
448 ehi->action &= ~action;
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] &= ~action;
451 } else {
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
454
455 /* break ehi->action into ehi->dev_action */
456 if (ehi->action & action) {
457 ata_for_each_dev(tdev, link, ALL)
458 ehi->dev_action[tdev->devno] |=
459 ehi->action & action;
460 ehi->action &= ~action;
461 }
462
463 /* turn off the specified per-dev action */
464 ehi->dev_action[dev->devno] &= ~action;
465 }
466}
467
468/**
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
471 *
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
475 *
476 * LOCKING:
477 * EH context.
478 */
479void ata_eh_acquire(struct ata_port *ap)
480 __acquires(&ap->host->eh_mutex)
481{
482 mutex_lock(&ap->host->eh_mutex);
483 WARN_ON_ONCE(ap->host->eh_owner);
484 ap->host->eh_owner = current;
485}
486
487/**
488 * ata_eh_release - release EH ownership
489 * @ap: ATA port to release EH ownership for
490 *
491 * Release EH ownership for @ap if the caller. The caller must
492 * have acquired EH ownership using ata_eh_acquire() previously.
493 *
494 * LOCKING:
495 * EH context.
496 */
497void ata_eh_release(struct ata_port *ap)
498 __releases(&ap->host->eh_mutex)
499{
500 WARN_ON_ONCE(ap->host->eh_owner != current);
501 ap->host->eh_owner = NULL;
502 mutex_unlock(&ap->host->eh_mutex);
503}
504
505static void ata_eh_dev_disable(struct ata_device *dev)
506{
507 ata_acpi_on_disable(dev);
508 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
509 dev->class++;
510
511 /*
512 * From now till the next successful probe, ering is used to
513 * track probe failures. Clear accumulated device error info.
514 */
515 ata_ering_clear(&dev->ering);
516
517 ata_dev_free_resources(dev);
518}
519
520static void ata_eh_unload(struct ata_port *ap)
521{
522 struct ata_link *link;
523 struct ata_device *dev;
524 unsigned long flags;
525
526 /*
527 * Unless we are restarting, transition all enabled devices to
528 * standby power mode.
529 */
530 if (system_state != SYSTEM_RESTART) {
531 ata_for_each_link(link, ap, PMP_FIRST) {
532 ata_for_each_dev(dev, link, ENABLED)
533 ata_dev_power_set_standby(dev);
534 }
535 }
536
537 /*
538 * Restore SControl IPM and SPD for the next driver and
539 * disable attached devices.
540 */
541 ata_for_each_link(link, ap, PMP_FIRST) {
542 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
543 ata_for_each_dev(dev, link, ENABLED)
544 ata_eh_dev_disable(dev);
545 }
546
547 /* freeze and set UNLOADED */
548 spin_lock_irqsave(ap->lock, flags);
549
550 ata_port_freeze(ap); /* won't be thawed */
551 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
552 ap->pflags |= ATA_PFLAG_UNLOADED;
553
554 spin_unlock_irqrestore(ap->lock, flags);
555}
556
557/**
558 * ata_scsi_error - SCSI layer error handler callback
559 * @host: SCSI host on which error occurred
560 *
561 * Handles SCSI-layer-thrown error events.
562 *
563 * LOCKING:
564 * Inherited from SCSI layer (none, can sleep)
565 *
566 * RETURNS:
567 * Zero.
568 */
569void ata_scsi_error(struct Scsi_Host *host)
570{
571 struct ata_port *ap = ata_shost_to_port(host);
572 unsigned long flags;
573 int nr_timedout;
574 LIST_HEAD(eh_work_q);
575
576 spin_lock_irqsave(host->host_lock, flags);
577 list_splice_init(&host->eh_cmd_q, &eh_work_q);
578 spin_unlock_irqrestore(host->host_lock, flags);
579
580 /*
581 * First check what errors we got with ata_scsi_cmd_error_handler().
582 * If we had no command timeouts and EH is not scheduled for this port,
583 * meaning that we do not have any failed command, then there is no
584 * need to go through the full port error handling. We only need to
585 * flush the completed commands we have.
586 */
587 nr_timedout = ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
588 if (nr_timedout || ata_port_eh_scheduled(ap))
589 ata_scsi_port_error_handler(host, ap);
590 else
591 scsi_eh_flush_done_q(&ap->eh_done_q);
592
593 WARN_ON(!list_empty(&eh_work_q));
594}
595
596/**
597 * ata_scsi_cmd_error_handler - error callback for a list of commands
598 * @host: scsi host containing the port
599 * @ap: ATA port within the host
600 * @eh_work_q: list of commands to process
601 *
602 * process the given list of commands and return those finished to the
603 * ap->eh_done_q. This function is the first part of the libata error
604 * handler which processes a given list of failed commands.
605 *
606 * Return the number of commands that timed out.
607 */
608int ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
609 struct list_head *eh_work_q)
610{
611 int i;
612 unsigned long flags;
613 struct scsi_cmnd *scmd, *tmp;
614 int nr_timedout = 0;
615
616 /* make sure sff pio task is not running */
617 ata_sff_flush_pio_task(ap);
618
619 /* synchronize with host lock and sort out timeouts */
620
621 /*
622 * For EH, all qcs are finished in one of three ways -
623 * normal completion, error completion, and SCSI timeout.
624 * Both completions can race against SCSI timeout. When normal
625 * completion wins, the qc never reaches EH. When error
626 * completion wins, the qc has ATA_QCFLAG_EH set.
627 *
628 * When SCSI timeout wins, things are a bit more complex.
629 * Normal or error completion can occur after the timeout but
630 * before this point. In such cases, both types of
631 * completions are honored. A scmd is determined to have
632 * timed out iff its associated qc is active and not failed.
633 */
634 spin_lock_irqsave(ap->lock, flags);
635
636 /*
637 * This must occur under the ap->lock as we don't want
638 * a polled recovery to race the real interrupt handler
639 *
640 * The lost_interrupt handler checks for any completed but
641 * non-notified command and completes much like an IRQ handler.
642 *
643 * We then fall into the error recovery code which will treat
644 * this as if normal completion won the race
645 */
646 if (ap->ops->lost_interrupt)
647 ap->ops->lost_interrupt(ap);
648
649 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
650 struct ata_queued_cmd *qc;
651
652 /*
653 * If the scmd was added to EH, via ata_qc_schedule_eh() ->
654 * scsi_timeout() -> scsi_eh_scmd_add(), scsi_timeout() will
655 * have set DID_TIME_OUT (since libata does not have an abort
656 * handler). Thus, to clear DID_TIME_OUT, clear the host byte.
657 */
658 set_host_byte(scmd, DID_OK);
659
660 ata_qc_for_each_raw(ap, qc, i) {
661 if (qc->scsicmd == scmd &&
662 qc->flags & ATA_QCFLAG_ACTIVE)
663 break;
664 }
665
666 if (i < ATA_MAX_QUEUE) {
667 /* the scmd has an associated qc */
668 if (!(qc->flags & ATA_QCFLAG_EH)) {
669 /* which hasn't failed yet, timeout */
670 set_host_byte(scmd, DID_TIME_OUT);
671 qc->err_mask |= AC_ERR_TIMEOUT;
672 qc->flags |= ATA_QCFLAG_EH;
673 nr_timedout++;
674 }
675 } else {
676 /* Normal completion occurred after
677 * SCSI timeout but before this point.
678 * Successfully complete it.
679 */
680 scmd->retries = scmd->allowed;
681 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
682 }
683 }
684
685 /*
686 * If we have timed out qcs. They belong to EH from
687 * this point but the state of the controller is
688 * unknown. Freeze the port to make sure the IRQ
689 * handler doesn't diddle with those qcs. This must
690 * be done atomically w.r.t. setting ATA_QCFLAG_EH.
691 */
692 if (nr_timedout)
693 __ata_port_freeze(ap);
694
695 /* initialize eh_tries */
696 ap->eh_tries = ATA_EH_MAX_TRIES;
697
698 spin_unlock_irqrestore(ap->lock, flags);
699
700 return nr_timedout;
701}
702EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
703
704/**
705 * ata_scsi_port_error_handler - recover the port after the commands
706 * @host: SCSI host containing the port
707 * @ap: the ATA port
708 *
709 * Handle the recovery of the port @ap after all the commands
710 * have been recovered.
711 */
712void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
713{
714 unsigned long flags;
715 struct ata_link *link;
716
717 /* acquire EH ownership */
718 ata_eh_acquire(ap);
719 repeat:
720 /* kill fast drain timer */
721 timer_delete_sync(&ap->fastdrain_timer);
722
723 /* process port resume request */
724 ata_eh_handle_port_resume(ap);
725
726 /* fetch & clear EH info */
727 spin_lock_irqsave(ap->lock, flags);
728
729 ata_for_each_link(link, ap, HOST_FIRST) {
730 struct ata_eh_context *ehc = &link->eh_context;
731 struct ata_device *dev;
732
733 memset(&link->eh_context, 0, sizeof(link->eh_context));
734 link->eh_context.i = link->eh_info;
735 memset(&link->eh_info, 0, sizeof(link->eh_info));
736
737 ata_for_each_dev(dev, link, ENABLED) {
738 int devno = dev->devno;
739
740 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
741 if (ata_ncq_enabled(dev))
742 ehc->saved_ncq_enabled |= 1 << devno;
743
744 /* If we are resuming, wake up the device */
745 if (ap->pflags & ATA_PFLAG_RESUMING) {
746 dev->flags |= ATA_DFLAG_RESUMING;
747 ehc->i.dev_action[devno] |= ATA_EH_SET_ACTIVE;
748 }
749 }
750 }
751
752 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
753 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
754 ap->excl_link = NULL; /* don't maintain exclusion over EH */
755
756 spin_unlock_irqrestore(ap->lock, flags);
757
758 /* invoke EH, skip if unloading or suspended */
759 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)) &&
760 ata_adapter_is_online(ap))
761 ap->ops->error_handler(ap);
762 else {
763 /* if unloading, commence suicide */
764 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
765 !(ap->pflags & ATA_PFLAG_UNLOADED))
766 ata_eh_unload(ap);
767 ata_eh_finish(ap);
768 }
769
770 /* process port suspend request */
771 ata_eh_handle_port_suspend(ap);
772
773 /*
774 * Exception might have happened after ->error_handler recovered the
775 * port but before this point. Repeat EH in such case.
776 */
777 spin_lock_irqsave(ap->lock, flags);
778
779 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
780 if (--ap->eh_tries) {
781 spin_unlock_irqrestore(ap->lock, flags);
782 goto repeat;
783 }
784 ata_port_err(ap,
785 "EH pending after %d tries, giving up\n",
786 ATA_EH_MAX_TRIES);
787 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
788 }
789
790 /* this run is complete, make sure EH info is clear */
791 ata_for_each_link(link, ap, HOST_FIRST)
792 memset(&link->eh_info, 0, sizeof(link->eh_info));
793
794 /*
795 * end eh (clear host_eh_scheduled) while holding ap->lock such that if
796 * exception occurs after this point but before EH completion, SCSI
797 * midlayer will re-initiate EH.
798 */
799 ap->ops->end_eh(ap);
800
801 spin_unlock_irqrestore(ap->lock, flags);
802 ata_eh_release(ap);
803
804 scsi_eh_flush_done_q(&ap->eh_done_q);
805
806 /* clean up */
807 spin_lock_irqsave(ap->lock, flags);
808
809 ap->pflags &= ~ATA_PFLAG_RESUMING;
810
811 if (ap->pflags & ATA_PFLAG_LOADING)
812 ap->pflags &= ~ATA_PFLAG_LOADING;
813 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
814 !(ap->flags & ATA_FLAG_SAS_HOST))
815 queue_delayed_work(system_dfl_long_wq, &ap->hotplug_task, 0);
816
817 if (ap->pflags & ATA_PFLAG_RECOVERED)
818 ata_port_info(ap, "EH complete\n");
819
820 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
821
822 /* tell wait_eh that we're done */
823 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
824 wake_up_all(&ap->eh_wait_q);
825
826 spin_unlock_irqrestore(ap->lock, flags);
827}
828EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
829
830/**
831 * ata_port_wait_eh - Wait for the currently pending EH to complete
832 * @ap: Port to wait EH for
833 *
834 * Wait until the currently pending EH is complete.
835 *
836 * LOCKING:
837 * Kernel thread context (may sleep).
838 */
839void ata_port_wait_eh(struct ata_port *ap)
840{
841 unsigned long flags;
842 DEFINE_WAIT(wait);
843
844 retry:
845 spin_lock_irqsave(ap->lock, flags);
846
847 while (ata_port_eh_scheduled(ap)) {
848 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
849 spin_unlock_irqrestore(ap->lock, flags);
850 schedule();
851 spin_lock_irqsave(ap->lock, flags);
852 }
853 finish_wait(&ap->eh_wait_q, &wait);
854
855 spin_unlock_irqrestore(ap->lock, flags);
856
857 /* make sure SCSI EH is complete */
858 if (scsi_host_in_recovery(ap->scsi_host)) {
859 ata_msleep(ap, 10);
860 goto retry;
861 }
862}
863EXPORT_SYMBOL_GPL(ata_port_wait_eh);
864
865static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
866{
867 struct ata_queued_cmd *qc;
868 unsigned int tag;
869 unsigned int nr = 0;
870
871 /* count only non-internal commands */
872 ata_qc_for_each(ap, qc, tag) {
873 if (qc)
874 nr++;
875 }
876
877 return nr;
878}
879
880void ata_eh_fastdrain_timerfn(struct timer_list *t)
881{
882 struct ata_port *ap = timer_container_of(ap, t, fastdrain_timer);
883 unsigned long flags;
884 unsigned int cnt;
885
886 spin_lock_irqsave(ap->lock, flags);
887
888 cnt = ata_eh_nr_in_flight(ap);
889
890 /* are we done? */
891 if (!cnt)
892 goto out_unlock;
893
894 if (cnt == ap->fastdrain_cnt) {
895 struct ata_queued_cmd *qc;
896 unsigned int tag;
897
898 /* No progress during the last interval, tag all
899 * in-flight qcs as timed out and freeze the port.
900 */
901 ata_qc_for_each(ap, qc, tag) {
902 if (qc)
903 qc->err_mask |= AC_ERR_TIMEOUT;
904 }
905
906 ata_port_freeze(ap);
907 } else {
908 /* some qcs have finished, give it another chance */
909 ap->fastdrain_cnt = cnt;
910 ap->fastdrain_timer.expires =
911 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
912 add_timer(&ap->fastdrain_timer);
913 }
914
915 out_unlock:
916 spin_unlock_irqrestore(ap->lock, flags);
917}
918
919/**
920 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
921 * @ap: target ATA port
922 * @fastdrain: activate fast drain
923 *
924 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
925 * is non-zero and EH wasn't pending before. Fast drain ensures
926 * that EH kicks in in timely manner.
927 *
928 * LOCKING:
929 * spin_lock_irqsave(host lock)
930 */
931static void ata_eh_set_pending(struct ata_port *ap, bool fastdrain)
932{
933 unsigned int cnt;
934
935 /* already scheduled? */
936 if (ap->pflags & ATA_PFLAG_EH_PENDING)
937 return;
938
939 ap->pflags |= ATA_PFLAG_EH_PENDING;
940
941 /*
942 * If we have deferred QCs, requeue them so that the SCSI EH task can
943 * run.
944 */
945 ata_scsi_requeue_deferred_qc(ap, NULL);
946
947 if (!fastdrain)
948 return;
949
950 /* do we have in-flight qcs? */
951 cnt = ata_eh_nr_in_flight(ap);
952 if (!cnt)
953 return;
954
955 /* activate fast drain */
956 ap->fastdrain_cnt = cnt;
957 ap->fastdrain_timer.expires =
958 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
959 add_timer(&ap->fastdrain_timer);
960}
961
962/**
963 * ata_qc_schedule_eh - schedule qc for error handling
964 * @qc: command to schedule error handling for
965 *
966 * Schedule error handling for @qc. EH will kick in as soon as
967 * other commands are drained.
968 *
969 * LOCKING:
970 * spin_lock_irqsave(host lock)
971 */
972void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
973{
974 struct ata_port *ap = qc->ap;
975
976 qc->flags |= ATA_QCFLAG_EH;
977 ata_eh_set_pending(ap, true);
978
979 /* The following will fail if timeout has already expired.
980 * ata_scsi_error() takes care of such scmds on EH entry.
981 * Note that ATA_QCFLAG_EH is unconditionally set after
982 * this function completes.
983 */
984 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
985}
986
987/**
988 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
989 * @ap: ATA port to schedule EH for
990 *
991 * LOCKING: inherited from ata_port_schedule_eh
992 * spin_lock_irqsave(host lock)
993 */
994void ata_std_sched_eh(struct ata_port *ap)
995{
996 if (ap->pflags & ATA_PFLAG_INITIALIZING)
997 return;
998
999 ata_eh_set_pending(ap, true);
1000 scsi_schedule_eh(ap->scsi_host);
1001
1002 trace_ata_std_sched_eh(ap);
1003}
1004EXPORT_SYMBOL_GPL(ata_std_sched_eh);
1005
1006/**
1007 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1008 * @ap: ATA port to end EH for
1009 *
1010 * In the libata object model there is a 1:1 mapping of ata_port to
1011 * shost, so host fields can be directly manipulated under ap->lock, in
1012 * the libsas case we need to hold a lock at the ha->level to coordinate
1013 * these events.
1014 *
1015 * LOCKING:
1016 * spin_lock_irqsave(host lock)
1017 */
1018void ata_std_end_eh(struct ata_port *ap)
1019{
1020 struct Scsi_Host *host = ap->scsi_host;
1021
1022 host->host_eh_scheduled = 0;
1023}
1024EXPORT_SYMBOL(ata_std_end_eh);
1025
1026
1027/**
1028 * ata_port_schedule_eh - schedule error handling without a qc
1029 * @ap: ATA port to schedule EH for
1030 *
1031 * Schedule error handling for @ap. EH will kick in as soon as
1032 * all commands are drained.
1033 *
1034 * LOCKING:
1035 * spin_lock_irqsave(host lock)
1036 */
1037void ata_port_schedule_eh(struct ata_port *ap)
1038{
1039 /* see: ata_std_sched_eh, unless you know better */
1040 ap->ops->sched_eh(ap);
1041}
1042EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
1043
1044static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1045{
1046 struct ata_queued_cmd *qc;
1047 int tag, nr_aborted = 0;
1048
1049 /* we're gonna abort all commands, no need for fast drain */
1050 ata_eh_set_pending(ap, false);
1051
1052 /* include internal tag in iteration */
1053 ata_qc_for_each_with_internal(ap, qc, tag) {
1054 if (qc && (!link || qc->dev->link == link)) {
1055 qc->flags |= ATA_QCFLAG_EH;
1056 ata_qc_complete(qc);
1057 nr_aborted++;
1058 }
1059 }
1060
1061 if (!nr_aborted)
1062 ata_port_schedule_eh(ap);
1063
1064 return nr_aborted;
1065}
1066
1067/**
1068 * ata_link_abort - abort all qc's on the link
1069 * @link: ATA link to abort qc's for
1070 *
1071 * Abort all active qc's active on @link and schedule EH.
1072 *
1073 * LOCKING:
1074 * spin_lock_irqsave(host lock)
1075 *
1076 * RETURNS:
1077 * Number of aborted qc's.
1078 */
1079int ata_link_abort(struct ata_link *link)
1080{
1081 return ata_do_link_abort(link->ap, link);
1082}
1083EXPORT_SYMBOL_GPL(ata_link_abort);
1084
1085/**
1086 * ata_port_abort - abort all qc's on the port
1087 * @ap: ATA port to abort qc's for
1088 *
1089 * Abort all active qc's of @ap and schedule EH.
1090 *
1091 * LOCKING:
1092 * spin_lock_irqsave(host_set lock)
1093 *
1094 * RETURNS:
1095 * Number of aborted qc's.
1096 */
1097int ata_port_abort(struct ata_port *ap)
1098{
1099 return ata_do_link_abort(ap, NULL);
1100}
1101EXPORT_SYMBOL_GPL(ata_port_abort);
1102
1103/**
1104 * __ata_port_freeze - freeze port
1105 * @ap: ATA port to freeze
1106 *
1107 * This function is called when HSM violation or some other
1108 * condition disrupts normal operation of the port. Frozen port
1109 * is not allowed to perform any operation until the port is
1110 * thawed, which usually follows a successful reset.
1111 *
1112 * ap->ops->freeze() callback can be used for freezing the port
1113 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1114 * port cannot be frozen hardware-wise, the interrupt handler
1115 * must ack and clear interrupts unconditionally while the port
1116 * is frozen.
1117 *
1118 * LOCKING:
1119 * spin_lock_irqsave(host lock)
1120 */
1121static void __ata_port_freeze(struct ata_port *ap)
1122{
1123 if (ap->ops->freeze)
1124 ap->ops->freeze(ap);
1125
1126 ap->pflags |= ATA_PFLAG_FROZEN;
1127
1128 trace_ata_port_freeze(ap);
1129}
1130
1131/**
1132 * ata_port_freeze - abort & freeze port
1133 * @ap: ATA port to freeze
1134 *
1135 * Abort and freeze @ap. The freeze operation must be called
1136 * first, because some hardware requires special operations
1137 * before the taskfile registers are accessible.
1138 *
1139 * LOCKING:
1140 * spin_lock_irqsave(host lock)
1141 *
1142 * RETURNS:
1143 * Number of aborted commands.
1144 */
1145int ata_port_freeze(struct ata_port *ap)
1146{
1147 __ata_port_freeze(ap);
1148
1149 return ata_port_abort(ap);
1150}
1151EXPORT_SYMBOL_GPL(ata_port_freeze);
1152
1153/**
1154 * ata_eh_freeze_port - EH helper to freeze port
1155 * @ap: ATA port to freeze
1156 *
1157 * Freeze @ap.
1158 *
1159 * LOCKING:
1160 * None.
1161 */
1162void ata_eh_freeze_port(struct ata_port *ap)
1163{
1164 unsigned long flags;
1165
1166 spin_lock_irqsave(ap->lock, flags);
1167 __ata_port_freeze(ap);
1168 spin_unlock_irqrestore(ap->lock, flags);
1169}
1170EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1171
1172/**
1173 * ata_eh_thaw_port - EH helper to thaw port
1174 * @ap: ATA port to thaw
1175 *
1176 * Thaw frozen port @ap.
1177 *
1178 * LOCKING:
1179 * None.
1180 */
1181void ata_eh_thaw_port(struct ata_port *ap)
1182{
1183 unsigned long flags;
1184
1185 spin_lock_irqsave(ap->lock, flags);
1186
1187 ap->pflags &= ~ATA_PFLAG_FROZEN;
1188
1189 if (ap->ops->thaw)
1190 ap->ops->thaw(ap);
1191
1192 spin_unlock_irqrestore(ap->lock, flags);
1193
1194 trace_ata_port_thaw(ap);
1195}
1196
1197static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1198{
1199 /* nada */
1200}
1201
1202static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1203{
1204 struct ata_port *ap = qc->ap;
1205 struct scsi_cmnd *scmd = qc->scsicmd;
1206 unsigned long flags;
1207
1208 spin_lock_irqsave(ap->lock, flags);
1209 qc->scsidone = ata_eh_scsidone;
1210 __ata_qc_complete(qc);
1211 WARN_ON(ata_tag_valid(qc->tag));
1212 spin_unlock_irqrestore(ap->lock, flags);
1213
1214 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1215}
1216
1217/**
1218 * ata_eh_qc_complete - Complete an active ATA command from EH
1219 * @qc: Command to complete
1220 *
1221 * Indicate to the mid and upper layers that an ATA command has
1222 * completed. To be used from EH.
1223 */
1224static void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1225{
1226 struct scsi_cmnd *scmd = qc->scsicmd;
1227
1228 scmd->retries = scmd->allowed;
1229 __ata_eh_qc_complete(qc);
1230}
1231
1232/**
1233 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1234 * @qc: Command to retry
1235 *
1236 * Indicate to the mid and upper layers that an ATA command
1237 * should be retried. To be used from EH.
1238 *
1239 * SCSI midlayer limits the number of retries to scmd->allowed.
1240 * scmd->allowed is incremented for commands which get retried
1241 * due to unrelated failures (qc->err_mask is zero).
1242 */
1243static void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1244{
1245 struct scsi_cmnd *scmd = qc->scsicmd;
1246
1247 if (!qc->err_mask)
1248 scmd->allowed++;
1249 __ata_eh_qc_complete(qc);
1250}
1251
1252/**
1253 * ata_dev_disable - disable ATA device
1254 * @dev: ATA device to disable
1255 *
1256 * Disable @dev.
1257 *
1258 * Locking:
1259 * EH context.
1260 */
1261void ata_dev_disable(struct ata_device *dev)
1262{
1263 if (!ata_dev_enabled(dev))
1264 return;
1265
1266 ata_dev_warn(dev, "disable device\n");
1267
1268 ata_eh_dev_disable(dev);
1269}
1270EXPORT_SYMBOL_GPL(ata_dev_disable);
1271
1272/**
1273 * ata_eh_detach_dev - detach ATA device
1274 * @dev: ATA device to detach
1275 *
1276 * Detach @dev.
1277 *
1278 * LOCKING:
1279 * None.
1280 */
1281void ata_eh_detach_dev(struct ata_device *dev)
1282{
1283 struct ata_link *link = dev->link;
1284 struct ata_port *ap = link->ap;
1285 struct ata_eh_context *ehc = &link->eh_context;
1286 unsigned long flags;
1287
1288 /*
1289 * If the device is still enabled, transition it to standby power mode
1290 * (i.e. spin down HDDs) and disable it.
1291 */
1292 if (ata_dev_enabled(dev)) {
1293 ata_dev_power_set_standby(dev);
1294 ata_eh_dev_disable(dev);
1295 }
1296
1297 spin_lock_irqsave(ap->lock, flags);
1298
1299 dev->flags &= ~ATA_DFLAG_DETACH;
1300
1301 if (ata_scsi_offline_dev(dev)) {
1302 dev->flags |= ATA_DFLAG_DETACHED;
1303 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1304 }
1305
1306 /* clear per-dev EH info */
1307 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1308 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1309 ehc->saved_xfer_mode[dev->devno] = 0;
1310 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1311
1312 spin_unlock_irqrestore(ap->lock, flags);
1313}
1314
1315/**
1316 * ata_eh_about_to_do - about to perform eh_action
1317 * @link: target ATA link
1318 * @dev: target ATA dev for per-dev action (can be NULL)
1319 * @action: action about to be performed
1320 *
1321 * Called just before performing EH actions to clear related bits
1322 * in @link->eh_info such that eh actions are not unnecessarily
1323 * repeated.
1324 *
1325 * LOCKING:
1326 * None.
1327 */
1328void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1329 unsigned int action)
1330{
1331 struct ata_port *ap = link->ap;
1332 struct ata_eh_info *ehi = &link->eh_info;
1333 struct ata_eh_context *ehc = &link->eh_context;
1334 unsigned long flags;
1335
1336 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1337
1338 spin_lock_irqsave(ap->lock, flags);
1339
1340 ata_eh_clear_action(link, dev, ehi, action);
1341
1342 /* About to take EH action, set RECOVERED. Ignore actions on
1343 * slave links as master will do them again.
1344 */
1345 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1346 ap->pflags |= ATA_PFLAG_RECOVERED;
1347
1348 spin_unlock_irqrestore(ap->lock, flags);
1349}
1350
1351/**
1352 * ata_eh_done - EH action complete
1353 * @link: ATA link for which EH actions are complete
1354 * @dev: target ATA dev for per-dev action (can be NULL)
1355 * @action: action just completed
1356 *
1357 * Called right after performing EH actions to clear related bits
1358 * in @link->eh_context.
1359 *
1360 * LOCKING:
1361 * None.
1362 */
1363void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1364 unsigned int action)
1365{
1366 struct ata_eh_context *ehc = &link->eh_context;
1367
1368 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1369
1370 ata_eh_clear_action(link, dev, &ehc->i, action);
1371}
1372
1373/**
1374 * ata_err_string - convert err_mask to descriptive string
1375 * @err_mask: error mask to convert to string
1376 *
1377 * Convert @err_mask to descriptive string. Errors are
1378 * prioritized according to severity and only the most severe
1379 * error is reported.
1380 *
1381 * LOCKING:
1382 * None.
1383 *
1384 * RETURNS:
1385 * Descriptive string for @err_mask
1386 */
1387static const char *ata_err_string(unsigned int err_mask)
1388{
1389 if (err_mask & AC_ERR_HOST_BUS)
1390 return "host bus error";
1391 if (err_mask & AC_ERR_ATA_BUS)
1392 return "ATA bus error";
1393 if (err_mask & AC_ERR_TIMEOUT)
1394 return "timeout";
1395 if (err_mask & AC_ERR_HSM)
1396 return "HSM violation";
1397 if (err_mask & AC_ERR_SYSTEM)
1398 return "internal error";
1399 if (err_mask & AC_ERR_MEDIA)
1400 return "media error";
1401 if (err_mask & AC_ERR_INVALID)
1402 return "invalid argument";
1403 if (err_mask & AC_ERR_DEV)
1404 return "device error";
1405 if (err_mask & AC_ERR_NCQ)
1406 return "NCQ error";
1407 if (err_mask & AC_ERR_NODEV_HINT)
1408 return "Polling detection error";
1409 return "unknown error";
1410}
1411
1412/**
1413 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1414 * @dev: target ATAPI device
1415 * @r_sense_key: out parameter for sense_key
1416 *
1417 * Perform ATAPI TEST_UNIT_READY.
1418 *
1419 * LOCKING:
1420 * EH context (may sleep).
1421 *
1422 * RETURNS:
1423 * 0 on success, AC_ERR_* mask on failure.
1424 */
1425unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1426{
1427 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1428 struct ata_taskfile tf;
1429 unsigned int err_mask;
1430
1431 ata_tf_init(dev, &tf);
1432
1433 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1434 tf.command = ATA_CMD_PACKET;
1435 tf.protocol = ATAPI_PROT_NODATA;
1436
1437 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1438 if (err_mask == AC_ERR_DEV)
1439 *r_sense_key = tf.error >> 4;
1440 return err_mask;
1441}
1442
1443/**
1444 * ata_eh_decide_disposition - Disposition a qc based on sense data
1445 * @qc: qc to examine
1446 *
1447 * For a regular SCSI command, the SCSI completion callback (scsi_done())
1448 * will call scsi_complete(), which will call scsi_decide_disposition(),
1449 * which will call scsi_check_sense(). scsi_complete() finally calls
1450 * scsi_finish_command(). This is fine for SCSI, since any eventual sense
1451 * data is usually returned in the completion itself (without invoking SCSI
1452 * EH). However, for a QC, we always need to fetch the sense data
1453 * explicitly using SCSI EH.
1454 *
1455 * A command that is completed via SCSI EH will instead be completed using
1456 * scsi_eh_flush_done_q(), which will call scsi_finish_command() directly
1457 * (without ever calling scsi_check_sense()).
1458 *
1459 * For a command that went through SCSI EH, it is the responsibility of the
1460 * SCSI EH strategy handler to call scsi_decide_disposition(), see e.g. how
1461 * scsi_eh_get_sense() calls scsi_decide_disposition() for SCSI LLDDs that
1462 * do not get the sense data as part of the completion.
1463 *
1464 * Thus, for QC commands that went via SCSI EH, we need to call
1465 * scsi_check_sense() ourselves, similar to how scsi_eh_get_sense() calls
1466 * scsi_decide_disposition(), which calls scsi_check_sense(), in order to
1467 * set the correct SCSI ML byte (if any).
1468 *
1469 * LOCKING:
1470 * EH context.
1471 *
1472 * RETURNS:
1473 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
1474 */
1475enum scsi_disposition ata_eh_decide_disposition(struct ata_queued_cmd *qc)
1476{
1477 return scsi_check_sense(qc->scsicmd);
1478}
1479
1480/**
1481 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1482 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1483 *
1484 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1485 * SENSE. This function is an EH helper.
1486 *
1487 * LOCKING:
1488 * Kernel thread context (may sleep).
1489 *
1490 * RETURNS:
1491 * true if sense data could be fetched, false otherwise.
1492 */
1493static bool ata_eh_request_sense(struct ata_queued_cmd *qc)
1494{
1495 struct scsi_cmnd *cmd = qc->scsicmd;
1496 struct ata_device *dev = qc->dev;
1497 struct ata_taskfile tf;
1498 unsigned int err_mask;
1499
1500 if (ata_port_is_frozen(qc->ap)) {
1501 ata_dev_warn(dev, "sense data available but port frozen\n");
1502 return false;
1503 }
1504
1505 if (!ata_id_sense_reporting_enabled(dev->id)) {
1506 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1507 return false;
1508 }
1509
1510 ata_tf_init(dev, &tf);
1511 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1512 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1513 tf.command = ATA_CMD_REQ_SENSE_DATA;
1514 tf.protocol = ATA_PROT_NODATA;
1515
1516 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1517 /* Ignore err_mask; ATA_ERR might be set */
1518 if (tf.status & ATA_SENSE) {
1519 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1520 /* Set sense without also setting scsicmd->result */
1521 scsi_build_sense_buffer(dev->flags & ATA_DFLAG_D_SENSE,
1522 cmd->sense_buffer, tf.lbah,
1523 tf.lbam, tf.lbal);
1524 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1525 return true;
1526 }
1527 } else {
1528 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1529 tf.status, err_mask);
1530 }
1531
1532 return false;
1533}
1534
1535/**
1536 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1537 * @dev: device to perform REQUEST_SENSE to
1538 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1539 * @dfl_sense_key: default sense key to use
1540 *
1541 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1542 * SENSE. This function is EH helper.
1543 *
1544 * LOCKING:
1545 * Kernel thread context (may sleep).
1546 *
1547 * RETURNS:
1548 * 0 on success, AC_ERR_* mask on failure
1549 */
1550unsigned int atapi_eh_request_sense(struct ata_device *dev,
1551 u8 *sense_buf, u8 dfl_sense_key)
1552{
1553 u8 cdb[ATAPI_CDB_LEN] =
1554 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1555 struct ata_port *ap = dev->link->ap;
1556 struct ata_taskfile tf;
1557
1558 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1559
1560 /* initialize sense_buf with the error register,
1561 * for the case where they are -not- overwritten
1562 */
1563 sense_buf[0] = 0x70;
1564 sense_buf[2] = dfl_sense_key;
1565
1566 /* some devices time out if garbage left in tf */
1567 ata_tf_init(dev, &tf);
1568
1569 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1570 tf.command = ATA_CMD_PACKET;
1571
1572 /*
1573 * Do not use DMA if the connected device only supports PIO, even if the
1574 * port prefers PIO commands via DMA.
1575 *
1576 * Ideally, we should call atapi_check_dma() to check if it is safe for
1577 * the LLD to use DMA for REQUEST_SENSE, but we don't have a qc.
1578 * Since we can't check the command, perhaps we should only use pio?
1579 */
1580 if ((ap->flags & ATA_FLAG_PIO_DMA) && !(dev->flags & ATA_DFLAG_PIO)) {
1581 tf.protocol = ATAPI_PROT_DMA;
1582 tf.feature |= ATAPI_PKT_DMA;
1583 } else {
1584 tf.protocol = ATAPI_PROT_PIO;
1585 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1586 tf.lbah = 0;
1587 }
1588
1589 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1590 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1591}
1592
1593/**
1594 * ata_eh_analyze_serror - analyze SError for a failed port
1595 * @link: ATA link to analyze SError for
1596 *
1597 * Analyze SError if available and further determine cause of
1598 * failure.
1599 *
1600 * LOCKING:
1601 * None.
1602 */
1603static void ata_eh_analyze_serror(struct ata_link *link)
1604{
1605 struct ata_eh_context *ehc = &link->eh_context;
1606 u32 serror = ehc->i.serror;
1607 unsigned int err_mask = 0, action = 0;
1608 u32 hotplug_mask;
1609
1610 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1611 err_mask |= AC_ERR_ATA_BUS;
1612 action |= ATA_EH_RESET;
1613 }
1614 if (serror & SERR_PROTOCOL) {
1615 err_mask |= AC_ERR_HSM;
1616 action |= ATA_EH_RESET;
1617 }
1618 if (serror & SERR_INTERNAL) {
1619 err_mask |= AC_ERR_SYSTEM;
1620 action |= ATA_EH_RESET;
1621 }
1622
1623 /* Determine whether a hotplug event has occurred. Both
1624 * SError.N/X are considered hotplug events for enabled or
1625 * host links. For disabled PMP links, only N bit is
1626 * considered as X bit is left at 1 for link plugging.
1627 */
1628 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1629 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1630 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1631 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1632 else
1633 hotplug_mask = SERR_PHYRDY_CHG;
1634
1635 if (serror & hotplug_mask)
1636 ata_ehi_hotplugged(&ehc->i);
1637
1638 ehc->i.err_mask |= err_mask;
1639 ehc->i.action |= action;
1640}
1641
1642/**
1643 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1644 * @qc: qc to analyze
1645 *
1646 * Analyze taskfile of @qc and further determine cause of
1647 * failure. This function also requests ATAPI sense data if
1648 * available.
1649 *
1650 * LOCKING:
1651 * Kernel thread context (may sleep).
1652 *
1653 * RETURNS:
1654 * Determined recovery action
1655 */
1656static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1657{
1658 const struct ata_taskfile *tf = &qc->result_tf;
1659 unsigned int tmp, action = 0;
1660 u8 stat = tf->status, err = tf->error;
1661
1662 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1663 qc->err_mask |= AC_ERR_HSM;
1664 return ATA_EH_RESET;
1665 }
1666
1667 if (stat & (ATA_ERR | ATA_DF)) {
1668 qc->err_mask |= AC_ERR_DEV;
1669 /*
1670 * Sense data reporting does not work if the
1671 * device fault bit is set.
1672 */
1673 if (stat & ATA_DF)
1674 stat &= ~ATA_SENSE;
1675 } else {
1676 return 0;
1677 }
1678
1679 switch (qc->dev->class) {
1680 case ATA_DEV_ATA:
1681 case ATA_DEV_ZAC:
1682 /*
1683 * Fetch the sense data explicitly if:
1684 * -It was a non-NCQ command that failed, or
1685 * -It was a NCQ command that failed, but the sense data
1686 * was not included in the NCQ command error log
1687 * (i.e. NCQ autosense is not supported by the device).
1688 */
1689 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) &&
1690 (stat & ATA_SENSE) && ata_eh_request_sense(qc))
1691 set_status_byte(qc->scsicmd, SAM_STAT_CHECK_CONDITION);
1692 if (err & ATA_ICRC)
1693 qc->err_mask |= AC_ERR_ATA_BUS;
1694 if (err & (ATA_UNC | ATA_AMNF))
1695 qc->err_mask |= AC_ERR_MEDIA;
1696 if (err & ATA_IDNF)
1697 qc->err_mask |= AC_ERR_INVALID;
1698 break;
1699
1700 case ATA_DEV_ATAPI:
1701 if (!ata_port_is_frozen(qc->ap)) {
1702 tmp = atapi_eh_request_sense(qc->dev,
1703 qc->scsicmd->sense_buffer,
1704 qc->result_tf.error >> 4);
1705 if (!tmp)
1706 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1707 else
1708 qc->err_mask |= tmp;
1709 }
1710 }
1711
1712 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1713 enum scsi_disposition ret = ata_eh_decide_disposition(qc);
1714
1715 /*
1716 * SUCCESS here means that the sense code could be
1717 * evaluated and should be passed to the upper layers
1718 * for correct evaluation.
1719 * FAILED means the sense code could not be interpreted
1720 * and the device would need to be reset.
1721 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1722 * command would need to be retried.
1723 */
1724 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1725 qc->flags |= ATA_QCFLAG_RETRY;
1726 qc->err_mask |= AC_ERR_OTHER;
1727 } else if (ret != SUCCESS) {
1728 qc->err_mask |= AC_ERR_HSM;
1729 }
1730 }
1731 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1732 action |= ATA_EH_RESET;
1733
1734 return action;
1735}
1736
1737static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1738 int *xfer_ok)
1739{
1740 int base = 0;
1741
1742 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1743 *xfer_ok = 1;
1744
1745 if (!*xfer_ok)
1746 base = ATA_ECAT_DUBIOUS_NONE;
1747
1748 if (err_mask & AC_ERR_ATA_BUS)
1749 return base + ATA_ECAT_ATA_BUS;
1750
1751 if (err_mask & AC_ERR_TIMEOUT)
1752 return base + ATA_ECAT_TOUT_HSM;
1753
1754 if (eflags & ATA_EFLAG_IS_IO) {
1755 if (err_mask & AC_ERR_HSM)
1756 return base + ATA_ECAT_TOUT_HSM;
1757 if ((err_mask &
1758 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1759 return base + ATA_ECAT_UNK_DEV;
1760 }
1761
1762 return 0;
1763}
1764
1765struct speed_down_verdict_arg {
1766 u64 since;
1767 int xfer_ok;
1768 int nr_errors[ATA_ECAT_NR];
1769};
1770
1771static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1772{
1773 struct speed_down_verdict_arg *arg = void_arg;
1774 int cat;
1775
1776 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1777 return -1;
1778
1779 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1780 &arg->xfer_ok);
1781 arg->nr_errors[cat]++;
1782
1783 return 0;
1784}
1785
1786/**
1787 * ata_eh_speed_down_verdict - Determine speed down verdict
1788 * @dev: Device of interest
1789 *
1790 * This function examines error ring of @dev and determines
1791 * whether NCQ needs to be turned off, transfer speed should be
1792 * stepped down, or falling back to PIO is necessary.
1793 *
1794 * ECAT_ATA_BUS : ATA_BUS error for any command
1795 *
1796 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1797 * IO commands
1798 *
1799 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1800 *
1801 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1802 * data transfer hasn't been verified.
1803 *
1804 * Verdicts are
1805 *
1806 * NCQ_OFF : Turn off NCQ.
1807 *
1808 * SPEED_DOWN : Speed down transfer speed but don't fall back
1809 * to PIO.
1810 *
1811 * FALLBACK_TO_PIO : Fall back to PIO.
1812 *
1813 * Even if multiple verdicts are returned, only one action is
1814 * taken per error. An action triggered by non-DUBIOUS errors
1815 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1816 * This is to expedite speed down decisions right after device is
1817 * initially configured.
1818 *
1819 * The following are speed down rules. #1 and #2 deal with
1820 * DUBIOUS errors.
1821 *
1822 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1823 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1824 *
1825 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1826 * occurred during last 5 mins, NCQ_OFF.
1827 *
1828 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1829 * occurred during last 5 mins, FALLBACK_TO_PIO
1830 *
1831 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1832 * during last 10 mins, NCQ_OFF.
1833 *
1834 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1835 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1836 *
1837 * LOCKING:
1838 * Inherited from caller.
1839 *
1840 * RETURNS:
1841 * OR of ATA_EH_SPDN_* flags.
1842 */
1843static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1844{
1845 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1846 u64 j64 = get_jiffies_64();
1847 struct speed_down_verdict_arg arg;
1848 unsigned int verdict = 0;
1849
1850 /* scan past 5 mins of error history */
1851 memset(&arg, 0, sizeof(arg));
1852 arg.since = j64 - min(j64, j5mins);
1853 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1854
1855 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1856 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1857 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1858 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1859
1860 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1861 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1862 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1863
1864 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1865 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1866 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1867 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1868
1869 /* scan past 10 mins of error history */
1870 memset(&arg, 0, sizeof(arg));
1871 arg.since = j64 - min(j64, j10mins);
1872 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1873
1874 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1875 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1876 verdict |= ATA_EH_SPDN_NCQ_OFF;
1877
1878 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1879 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1880 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1881 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1882
1883 return verdict;
1884}
1885
1886/**
1887 * ata_eh_speed_down - record error and speed down if necessary
1888 * @dev: Failed device
1889 * @eflags: mask of ATA_EFLAG_* flags
1890 * @err_mask: err_mask of the error
1891 *
1892 * Record error and examine error history to determine whether
1893 * adjusting transmission speed is necessary. It also sets
1894 * transmission limits appropriately if such adjustment is
1895 * necessary.
1896 *
1897 * LOCKING:
1898 * Kernel thread context (may sleep).
1899 *
1900 * RETURNS:
1901 * Determined recovery action.
1902 */
1903static unsigned int ata_eh_speed_down(struct ata_device *dev,
1904 unsigned int eflags, unsigned int err_mask)
1905{
1906 struct ata_link *link = ata_dev_phys_link(dev);
1907 int xfer_ok = 0;
1908 unsigned int verdict;
1909 unsigned int action = 0;
1910
1911 /* don't bother if Cat-0 error */
1912 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1913 return 0;
1914
1915 /* record error and determine whether speed down is necessary */
1916 ata_ering_record(&dev->ering, eflags, err_mask);
1917 verdict = ata_eh_speed_down_verdict(dev);
1918
1919 /* turn off NCQ? */
1920 if ((verdict & ATA_EH_SPDN_NCQ_OFF) && ata_ncq_enabled(dev)) {
1921 dev->flags |= ATA_DFLAG_NCQ_OFF;
1922 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1923 goto done;
1924 }
1925
1926 /* speed down? */
1927 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1928 /* speed down SATA link speed if possible */
1929 if (sata_down_spd_limit(link, 0) == 0) {
1930 action |= ATA_EH_RESET;
1931 goto done;
1932 }
1933
1934 /* lower transfer mode */
1935 if (dev->spdn_cnt < 2) {
1936 static const int dma_dnxfer_sel[] =
1937 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1938 static const int pio_dnxfer_sel[] =
1939 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1940 int sel;
1941
1942 if (dev->xfer_shift != ATA_SHIFT_PIO)
1943 sel = dma_dnxfer_sel[dev->spdn_cnt];
1944 else
1945 sel = pio_dnxfer_sel[dev->spdn_cnt];
1946
1947 dev->spdn_cnt++;
1948
1949 if (ata_down_xfermask_limit(dev, sel) == 0) {
1950 action |= ATA_EH_RESET;
1951 goto done;
1952 }
1953 }
1954 }
1955
1956 /* Fall back to PIO? Slowing down to PIO is meaningless for
1957 * SATA ATA devices. Consider it only for PATA and SATAPI.
1958 */
1959 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1960 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1961 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1962 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1963 dev->spdn_cnt = 0;
1964 action |= ATA_EH_RESET;
1965 goto done;
1966 }
1967 }
1968
1969 return 0;
1970 done:
1971 /* device has been slowed down, blow error history */
1972 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1973 ata_ering_clear(&dev->ering);
1974 return action;
1975}
1976
1977/**
1978 * ata_eh_worth_retry - analyze error and decide whether to retry
1979 * @qc: qc to possibly retry
1980 *
1981 * Look at the cause of the error and decide if a retry
1982 * might be useful or not. We don't want to retry media errors
1983 * because the drive itself has probably already taken 10-30 seconds
1984 * doing its own internal retries before reporting the failure.
1985 */
1986static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1987{
1988 if (qc->err_mask & AC_ERR_MEDIA)
1989 return 0; /* don't retry media errors */
1990 if (qc->flags & ATA_QCFLAG_IO)
1991 return 1; /* otherwise retry anything from fs stack */
1992 if (qc->err_mask & AC_ERR_INVALID)
1993 return 0; /* don't retry these */
1994 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1995}
1996
1997/**
1998 * ata_eh_quiet - check if we need to be quiet about a command error
1999 * @qc: qc to check
2000 *
2001 * Look at the qc flags anbd its scsi command request flags to determine
2002 * if we need to be quiet about the command failure.
2003 */
2004static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
2005{
2006 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
2007 qc->flags |= ATA_QCFLAG_QUIET;
2008 return qc->flags & ATA_QCFLAG_QUIET;
2009}
2010
2011static int ata_eh_get_non_ncq_success_sense(struct ata_link *link)
2012{
2013 struct ata_port *ap = link->ap;
2014 struct ata_queued_cmd *qc;
2015
2016 qc = __ata_qc_from_tag(ap, link->active_tag);
2017 if (!qc)
2018 return -EIO;
2019
2020 if (!(qc->flags & ATA_QCFLAG_EH) ||
2021 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
2022 qc->err_mask)
2023 return -EIO;
2024
2025 if (!ata_eh_request_sense(qc))
2026 return -EIO;
2027
2028 /*
2029 * No point in checking the return value, since the command has already
2030 * completed successfully.
2031 */
2032 ata_eh_decide_disposition(qc);
2033
2034 return 0;
2035}
2036
2037static void ata_eh_get_success_sense(struct ata_link *link)
2038{
2039 struct ata_eh_context *ehc = &link->eh_context;
2040 struct ata_device *dev = link->device;
2041 struct ata_port *ap = link->ap;
2042 struct ata_queued_cmd *qc;
2043 int tag, ret = 0;
2044
2045 if (!(ehc->i.dev_action[dev->devno] & ATA_EH_GET_SUCCESS_SENSE))
2046 return;
2047
2048 /* if frozen, we can't do much */
2049 if (ata_port_is_frozen(ap)) {
2050 ata_dev_warn(dev,
2051 "successful sense data available but port frozen\n");
2052 goto out;
2053 }
2054
2055 /*
2056 * If the link has sactive set, then we have outstanding NCQ commands
2057 * and have to read the Successful NCQ Commands log to get the sense
2058 * data. Otherwise, we are dealing with a non-NCQ command and use
2059 * request sense ext command to retrieve the sense data.
2060 */
2061 if (link->sactive)
2062 ret = ata_eh_get_ncq_success_sense(link);
2063 else
2064 ret = ata_eh_get_non_ncq_success_sense(link);
2065 if (ret)
2066 goto out;
2067
2068 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
2069 return;
2070
2071out:
2072 /*
2073 * If we failed to get sense data for a successful command that ought to
2074 * have sense data, we cannot simply return BLK_STS_OK to user space.
2075 * This is because we can't know if the sense data that we couldn't get
2076 * was actually "DATA CURRENTLY UNAVAILABLE". Reporting such a command
2077 * as success to user space would result in a silent data corruption.
2078 * Thus, add a bogus ABORTED_COMMAND sense data to such commands, such
2079 * that SCSI will report these commands as BLK_STS_IOERR to user space.
2080 */
2081 ata_qc_for_each_raw(ap, qc, tag) {
2082 if (!(qc->flags & ATA_QCFLAG_EH) ||
2083 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
2084 qc->err_mask ||
2085 ata_dev_phys_link(qc->dev) != link)
2086 continue;
2087
2088 /* We managed to get sense for this success command, skip. */
2089 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2090 continue;
2091
2092 /* This success command did not have any sense data, skip. */
2093 if (!(qc->result_tf.status & ATA_SENSE))
2094 continue;
2095
2096 /* This success command had sense data, but we failed to get. */
2097 ata_scsi_set_sense(dev, qc->scsicmd, ABORTED_COMMAND, 0, 0);
2098 qc->flags |= ATA_QCFLAG_SENSE_VALID;
2099 }
2100 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
2101}
2102
2103/*
2104 * Check if a link is established. This is a relaxed version of
2105 * ata_phys_link_online() which accounts for the fact that this is potentially
2106 * called after changing the link power management policy, which may not be
2107 * reflected immediately in the SStatus register (e.g., we may still be seeing
2108 * the PHY in partial, slumber or devsleep Partial power management state.
2109 * So check that:
2110 * - A device is still present, that is, DET is 1h (Device presence detected
2111 * but Phy communication not established) or 3h (Device presence detected and
2112 * Phy communication established)
2113 * - Communication is established, that is, IPM is not 0h, indicating that PHY
2114 * is online or in a low power state.
2115 */
2116static bool ata_eh_link_established(struct ata_link *link)
2117{
2118 u32 sstatus;
2119 u8 det, ipm;
2120
2121 /*
2122 * For old IDE/PATA adapters that do not have a valid scr_read method,
2123 * or if reading the SStatus register fails, assume that the device is
2124 * present. Device probe will determine if that is really the case.
2125 */
2126 if (sata_scr_read(link, SCR_STATUS, &sstatus))
2127 return true;
2128
2129 det = sstatus & 0x0f;
2130 ipm = (sstatus >> 8) & 0x0f;
2131
2132 return (det & 0x01) && ipm;
2133}
2134
2135/**
2136 * ata_eh_link_set_lpm - configure SATA interface power management
2137 * @link: link to configure
2138 * @policy: the link power management policy
2139 * @r_failed_dev: out parameter for failed device
2140 *
2141 * Enable SATA Interface power management. This will enable
2142 * Device Interface Power Management (DIPM) for min_power and
2143 * medium_power_with_dipm policies, and then call driver specific
2144 * callbacks for enabling Host Initiated Power management.
2145 *
2146 * LOCKING:
2147 * EH context.
2148 *
2149 * RETURNS:
2150 * 0 on success, -errno on failure.
2151 */
2152static int ata_eh_link_set_lpm(struct ata_link *link,
2153 enum ata_lpm_policy policy,
2154 struct ata_device **r_failed_dev)
2155{
2156 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
2157 struct ata_eh_context *ehc = &link->eh_context;
2158 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
2159 enum ata_lpm_policy old_policy = link->lpm_policy;
2160 bool host_has_dipm = !(link->ap->flags & ATA_FLAG_NO_DIPM);
2161 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
2162 unsigned int err_mask;
2163 int rc;
2164
2165 /* if the link or host doesn't do LPM, noop */
2166 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
2167 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
2168 return 0;
2169
2170 /*
2171 * This function currently assumes that it will never be supplied policy
2172 * ATA_LPM_UNKNOWN.
2173 */
2174 if (WARN_ON_ONCE(policy == ATA_LPM_UNKNOWN))
2175 return 0;
2176
2177 ata_link_dbg(link, "Set LPM policy: %d -> %d\n", old_policy, policy);
2178
2179 /*
2180 * DIPM is enabled only for ATA_LPM_MIN_POWER,
2181 * ATA_LPM_MIN_POWER_WITH_PARTIAL, and ATA_LPM_MED_POWER_WITH_DIPM, as
2182 * some devices misbehave when the host NACKs transition to SLUMBER.
2183 */
2184 ata_for_each_dev(dev, link, ENABLED) {
2185 bool dev_has_hipm = ata_id_has_hipm(dev->id);
2186 bool dev_has_dipm = ata_id_has_dipm(dev->id);
2187
2188 /* find the first enabled and LPM enabled devices */
2189 if (!link_dev)
2190 link_dev = dev;
2191
2192 if (!lpm_dev &&
2193 (dev_has_hipm || (dev_has_dipm && host_has_dipm)))
2194 lpm_dev = dev;
2195
2196 hints &= ~ATA_LPM_EMPTY;
2197 if (!dev_has_hipm)
2198 hints &= ~ATA_LPM_HIPM;
2199
2200 /* disable DIPM before changing link config */
2201 if (dev_has_dipm) {
2202 err_mask = ata_dev_set_feature(dev,
2203 SETFEATURES_SATA_DISABLE, SATA_DIPM);
2204 if (err_mask && err_mask != AC_ERR_DEV) {
2205 ata_dev_warn(dev,
2206 "failed to disable DIPM, Emask 0x%x\n",
2207 err_mask);
2208 rc = -EIO;
2209 goto fail;
2210 }
2211 }
2212 }
2213
2214 if (ap) {
2215 rc = ap->ops->set_lpm(link, policy, hints);
2216 if (!rc && ap->slave_link)
2217 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
2218 } else
2219 rc = sata_pmp_set_lpm(link, policy, hints);
2220
2221 /*
2222 * Attribute link config failure to the first (LPM) enabled
2223 * device on the link.
2224 */
2225 if (rc) {
2226 if (rc == -EOPNOTSUPP) {
2227 link->flags |= ATA_LFLAG_NO_LPM;
2228 return 0;
2229 }
2230 dev = lpm_dev ? lpm_dev : link_dev;
2231 goto fail;
2232 }
2233
2234 /*
2235 * Low level driver acked the transition. Issue DIPM command
2236 * with the new policy set.
2237 */
2238 link->lpm_policy = policy;
2239 if (ap && ap->slave_link)
2240 ap->slave_link->lpm_policy = policy;
2241
2242 /*
2243 * Host config updated, enable DIPM if transitioning to
2244 * ATA_LPM_MIN_POWER, ATA_LPM_MIN_POWER_WITH_PARTIAL, or
2245 * ATA_LPM_MED_POWER_WITH_DIPM.
2246 */
2247 ata_for_each_dev(dev, link, ENABLED) {
2248 bool dev_has_dipm = ata_id_has_dipm(dev->id);
2249
2250 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && host_has_dipm &&
2251 dev_has_dipm) {
2252 err_mask = ata_dev_set_feature(dev,
2253 SETFEATURES_SATA_ENABLE, SATA_DIPM);
2254 if (err_mask && err_mask != AC_ERR_DEV) {
2255 ata_dev_warn(dev,
2256 "failed to enable DIPM, Emask 0x%x\n",
2257 err_mask);
2258 rc = -EIO;
2259 goto fail;
2260 }
2261 }
2262 }
2263
2264 link->last_lpm_change = jiffies;
2265 link->flags |= ATA_LFLAG_CHANGED;
2266
2267 return 0;
2268
2269fail:
2270 /* restore the old policy */
2271 link->lpm_policy = old_policy;
2272 if (ap && ap->slave_link)
2273 ap->slave_link->lpm_policy = old_policy;
2274
2275 /* if no device or only one more chance is left, disable LPM */
2276 if (!dev || ehc->tries[dev->devno] <= 2) {
2277 ata_link_warn(link, "disabling LPM on the link\n");
2278 link->flags |= ATA_LFLAG_NO_LPM;
2279 }
2280 if (r_failed_dev)
2281 *r_failed_dev = dev;
2282 return rc;
2283}
2284
2285/**
2286 * ata_eh_link_autopsy - analyze error and determine recovery action
2287 * @link: host link to perform autopsy on
2288 *
2289 * Analyze why @link failed and determine which recovery actions
2290 * are needed. This function also sets more detailed AC_ERR_*
2291 * values and fills sense data for ATAPI CHECK SENSE.
2292 *
2293 * LOCKING:
2294 * Kernel thread context (may sleep).
2295 */
2296static void ata_eh_link_autopsy(struct ata_link *link)
2297{
2298 struct ata_port *ap = link->ap;
2299 struct ata_eh_context *ehc = &link->eh_context;
2300 struct ata_queued_cmd *qc;
2301 struct ata_device *dev;
2302 unsigned int all_err_mask = 0, eflags = 0;
2303 int tag, nr_failed = 0, nr_quiet = 0;
2304 u32 serror;
2305 int rc;
2306
2307 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2308 return;
2309
2310 /* obtain and analyze SError */
2311 rc = sata_scr_read(link, SCR_ERROR, &serror);
2312 if (rc == 0) {
2313 ehc->i.serror |= serror;
2314 ata_eh_analyze_serror(link);
2315 } else if (rc != -EOPNOTSUPP) {
2316 /* SError read failed, force reset and probing */
2317 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2318 ehc->i.action |= ATA_EH_RESET;
2319 ehc->i.err_mask |= AC_ERR_OTHER;
2320 }
2321
2322 /* analyze NCQ failure */
2323 ata_eh_analyze_ncq_error(link);
2324
2325 /*
2326 * Check if this was a successful command that simply needs sense data.
2327 * Since the sense data is not part of the completion, we need to fetch
2328 * it using an additional command. Since this can't be done from irq
2329 * context, the sense data for successful commands are fetched by EH.
2330 */
2331 ata_eh_get_success_sense(link);
2332
2333 /* any real error trumps AC_ERR_OTHER */
2334 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2335 ehc->i.err_mask &= ~AC_ERR_OTHER;
2336
2337 all_err_mask |= ehc->i.err_mask;
2338
2339 ata_qc_for_each_raw(ap, qc, tag) {
2340 if (!(qc->flags & ATA_QCFLAG_EH) ||
2341 qc->flags & ATA_QCFLAG_RETRY ||
2342 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD ||
2343 ata_dev_phys_link(qc->dev) != link)
2344 continue;
2345
2346 /* inherit upper level err_mask */
2347 qc->err_mask |= ehc->i.err_mask;
2348
2349 /* analyze TF */
2350 ehc->i.action |= ata_eh_analyze_tf(qc);
2351
2352 /* DEV errors are probably spurious in case of ATA_BUS error */
2353 if (qc->err_mask & AC_ERR_ATA_BUS)
2354 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2355 AC_ERR_INVALID);
2356
2357 /* any real error trumps unknown error */
2358 if (qc->err_mask & ~AC_ERR_OTHER)
2359 qc->err_mask &= ~AC_ERR_OTHER;
2360
2361 /*
2362 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2363 * layers will determine whether the command is worth retrying
2364 * based on the sense data and device class/type. Otherwise,
2365 * determine directly if the command is worth retrying using its
2366 * error mask and flags.
2367 */
2368 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2369 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2370 else if (ata_eh_worth_retry(qc))
2371 qc->flags |= ATA_QCFLAG_RETRY;
2372
2373 /* accumulate error info */
2374 ehc->i.dev = qc->dev;
2375 all_err_mask |= qc->err_mask;
2376 if (qc->flags & ATA_QCFLAG_IO)
2377 eflags |= ATA_EFLAG_IS_IO;
2378 trace_ata_eh_link_autopsy_qc(qc);
2379
2380 /* Count quiet errors */
2381 if (ata_eh_quiet(qc))
2382 nr_quiet++;
2383 nr_failed++;
2384 }
2385
2386 /* If all failed commands requested silence, then be quiet */
2387 if (nr_quiet == nr_failed)
2388 ehc->i.flags |= ATA_EHI_QUIET;
2389
2390 /* enforce default EH actions */
2391 if (ata_port_is_frozen(ap) ||
2392 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2393 ehc->i.action |= ATA_EH_RESET;
2394 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2395 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2396 ehc->i.action |= ATA_EH_REVALIDATE;
2397
2398 /* If we have offending qcs and the associated failed device,
2399 * perform per-dev EH action only on the offending device.
2400 */
2401 if (ehc->i.dev) {
2402 ehc->i.dev_action[ehc->i.dev->devno] |=
2403 ehc->i.action & ATA_EH_PERDEV_MASK;
2404 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2405 }
2406
2407 /* propagate timeout to host link */
2408 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2409 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2410
2411 /* record error and consider speeding down */
2412 dev = ehc->i.dev;
2413 if (!dev && ((ata_link_max_devices(link) == 1 &&
2414 ata_dev_enabled(link->device))))
2415 dev = link->device;
2416
2417 if (dev) {
2418 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2419 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2420 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2421 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2422 }
2423}
2424
2425/**
2426 * ata_eh_autopsy - analyze error and determine recovery action
2427 * @ap: host port to perform autopsy on
2428 *
2429 * Analyze all links of @ap and determine why they failed and
2430 * which recovery actions are needed.
2431 *
2432 * LOCKING:
2433 * Kernel thread context (may sleep).
2434 */
2435void ata_eh_autopsy(struct ata_port *ap)
2436{
2437 struct ata_link *link;
2438
2439 ata_for_each_link(link, ap, EDGE)
2440 ata_eh_link_autopsy(link);
2441
2442 /* Handle the frigging slave link. Autopsy is done similarly
2443 * but actions and flags are transferred over to the master
2444 * link and handled from there.
2445 */
2446 if (ap->slave_link) {
2447 struct ata_eh_context *mehc = &ap->link.eh_context;
2448 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2449
2450 /* transfer control flags from master to slave */
2451 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2452
2453 /* perform autopsy on the slave link */
2454 ata_eh_link_autopsy(ap->slave_link);
2455
2456 /* transfer actions from slave to master and clear slave */
2457 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2458 mehc->i.action |= sehc->i.action;
2459 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2460 mehc->i.flags |= sehc->i.flags;
2461 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2462 }
2463
2464 /* Autopsy of fanout ports can affect host link autopsy.
2465 * Perform host link autopsy last.
2466 */
2467 if (sata_pmp_attached(ap))
2468 ata_eh_link_autopsy(&ap->link);
2469}
2470
2471/**
2472 * ata_get_cmd_name - get name for ATA command
2473 * @command: ATA command code to get name for
2474 *
2475 * Return a textual name of the given command or "unknown"
2476 *
2477 * LOCKING:
2478 * None
2479 */
2480const char *ata_get_cmd_name(u8 command)
2481{
2482#ifdef CONFIG_ATA_VERBOSE_ERROR
2483 static const struct
2484 {
2485 u8 command;
2486 const char *text;
2487 } cmd_descr[] = {
2488 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2489 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2490 { ATA_CMD_STANDBY, "STANDBY" },
2491 { ATA_CMD_IDLE, "IDLE" },
2492 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2493 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2494 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2495 { ATA_CMD_NOP, "NOP" },
2496 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2497 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2498 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2499 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2500 { ATA_CMD_SERVICE, "SERVICE" },
2501 { ATA_CMD_READ, "READ DMA" },
2502 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2503 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2504 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2505 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2506 { ATA_CMD_WRITE, "WRITE DMA" },
2507 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2508 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2509 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2510 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2511 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2512 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2513 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2514 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2515 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" },
2516 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2517 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2518 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2519 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2520 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2521 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2522 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2523 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2524 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2525 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2526 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2527 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2528 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2529 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2530 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2531 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2532 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2533 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2534 { ATA_CMD_SLEEP, "SLEEP" },
2535 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2536 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2537 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2538 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2539 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2540 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2541 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2542 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2543 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2544 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2545 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2546 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2547 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2548 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2549 { ATA_CMD_PMP_READ, "READ BUFFER" },
2550 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2551 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2552 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2553 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2554 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2555 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2556 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2557 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2558 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2559 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2560 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2561 { ATA_CMD_SMART, "SMART" },
2562 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2563 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2564 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2565 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2566 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2567 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2568 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2569 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2570 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2571 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2572 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2573 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2574 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2575 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2576 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2577 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2578 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2579 { ATA_CMD_RESTORE, "RECALIBRATE" },
2580 { 0, NULL } /* terminate list */
2581 };
2582
2583 unsigned int i;
2584 for (i = 0; cmd_descr[i].text; i++)
2585 if (cmd_descr[i].command == command)
2586 return cmd_descr[i].text;
2587#endif
2588
2589 return "unknown";
2590}
2591EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2592
2593/**
2594 * ata_eh_link_report - report error handling to user
2595 * @link: ATA link EH is going on
2596 *
2597 * Report EH to user.
2598 *
2599 * LOCKING:
2600 * None.
2601 */
2602static void ata_eh_link_report(struct ata_link *link)
2603{
2604 struct ata_port *ap = link->ap;
2605 struct ata_eh_context *ehc = &link->eh_context;
2606 struct ata_queued_cmd *qc;
2607 const char *frozen, *desc;
2608 char tries_buf[16] = "";
2609 int tag, nr_failed = 0;
2610
2611 if (ehc->i.flags & ATA_EHI_QUIET)
2612 return;
2613
2614 desc = NULL;
2615 if (ehc->i.desc[0] != '\0')
2616 desc = ehc->i.desc;
2617
2618 ata_qc_for_each_raw(ap, qc, tag) {
2619 if (!(qc->flags & ATA_QCFLAG_EH) ||
2620 ata_dev_phys_link(qc->dev) != link ||
2621 ((qc->flags & ATA_QCFLAG_QUIET) &&
2622 qc->err_mask == AC_ERR_DEV))
2623 continue;
2624 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2625 continue;
2626
2627 nr_failed++;
2628 }
2629
2630 if (!nr_failed && !ehc->i.err_mask)
2631 return;
2632
2633 frozen = "";
2634 if (ata_port_is_frozen(ap))
2635 frozen = " frozen";
2636
2637 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2638 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2639 ap->eh_tries);
2640
2641 if (ehc->i.dev) {
2642 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2643 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2644 ehc->i.err_mask, link->sactive, ehc->i.serror,
2645 ehc->i.action, frozen, tries_buf);
2646 if (desc)
2647 ata_dev_err(ehc->i.dev, "%s\n", desc);
2648 } else {
2649 ata_link_err(link, "exception Emask 0x%x "
2650 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2651 ehc->i.err_mask, link->sactive, ehc->i.serror,
2652 ehc->i.action, frozen, tries_buf);
2653 if (desc)
2654 ata_link_err(link, "%s\n", desc);
2655 }
2656
2657#ifdef CONFIG_ATA_VERBOSE_ERROR
2658 if (ehc->i.serror)
2659 ata_link_err(link,
2660 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2661 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2662 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2663 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2664 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2665 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2666 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2667 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2668 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2669 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2670 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2671 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2672 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2673 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2674 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2675 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2676 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2677 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2678#endif
2679
2680 ata_qc_for_each_raw(ap, qc, tag) {
2681 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2682 char data_buf[20] = "";
2683 char cdb_buf[70] = "";
2684
2685 if (!(qc->flags & ATA_QCFLAG_EH) ||
2686 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2687 continue;
2688
2689 if (qc->dma_dir != DMA_NONE) {
2690 static const char *dma_str[] = {
2691 [DMA_BIDIRECTIONAL] = "bidi",
2692 [DMA_TO_DEVICE] = "out",
2693 [DMA_FROM_DEVICE] = "in",
2694 };
2695 const char *prot_str = NULL;
2696
2697 switch (qc->tf.protocol) {
2698 case ATA_PROT_UNKNOWN:
2699 prot_str = "unknown";
2700 break;
2701 case ATA_PROT_NODATA:
2702 prot_str = "nodata";
2703 break;
2704 case ATA_PROT_PIO:
2705 prot_str = "pio";
2706 break;
2707 case ATA_PROT_DMA:
2708 prot_str = "dma";
2709 break;
2710 case ATA_PROT_NCQ:
2711 prot_str = "ncq dma";
2712 break;
2713 case ATA_PROT_NCQ_NODATA:
2714 prot_str = "ncq nodata";
2715 break;
2716 case ATAPI_PROT_NODATA:
2717 prot_str = "nodata";
2718 break;
2719 case ATAPI_PROT_PIO:
2720 prot_str = "pio";
2721 break;
2722 case ATAPI_PROT_DMA:
2723 prot_str = "dma";
2724 break;
2725 }
2726 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2727 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2728 }
2729
2730 if (ata_is_atapi(qc->tf.protocol)) {
2731 const u8 *cdb = qc->cdb;
2732 size_t cdb_len = qc->dev->cdb_len;
2733
2734 if (qc->scsicmd) {
2735 cdb = qc->scsicmd->cmnd;
2736 cdb_len = qc->scsicmd->cmd_len;
2737 }
2738 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2739 cdb, cdb_len);
2740 } else
2741 ata_dev_err(qc->dev, "failed command: %s\n",
2742 ata_get_cmd_name(cmd->command));
2743
2744 ata_dev_err(qc->dev,
2745 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2746 "tag %d%s\n %s"
2747 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2748 "Emask 0x%x (%s)%s\n",
2749 cmd->command, cmd->feature, cmd->nsect,
2750 cmd->lbal, cmd->lbam, cmd->lbah,
2751 cmd->hob_feature, cmd->hob_nsect,
2752 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2753 cmd->device, qc->tag, data_buf, cdb_buf,
2754 res->status, res->error, res->nsect,
2755 res->lbal, res->lbam, res->lbah,
2756 res->hob_feature, res->hob_nsect,
2757 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2758 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2759 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2760
2761#ifdef CONFIG_ATA_VERBOSE_ERROR
2762 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2763 ATA_SENSE | ATA_ERR)) {
2764 if (res->status & ATA_BUSY)
2765 ata_dev_err(qc->dev, "status: { Busy }\n");
2766 else
2767 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2768 res->status & ATA_DRDY ? "DRDY " : "",
2769 res->status & ATA_DF ? "DF " : "",
2770 res->status & ATA_DRQ ? "DRQ " : "",
2771 res->status & ATA_SENSE ? "SENSE " : "",
2772 res->status & ATA_ERR ? "ERR " : "");
2773 }
2774
2775 if (cmd->command != ATA_CMD_PACKET &&
2776 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2777 ATA_ABORTED)))
2778 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2779 res->error & ATA_ICRC ? "ICRC " : "",
2780 res->error & ATA_UNC ? "UNC " : "",
2781 res->error & ATA_AMNF ? "AMNF " : "",
2782 res->error & ATA_IDNF ? "IDNF " : "",
2783 res->error & ATA_ABORTED ? "ABRT " : "");
2784#endif
2785 }
2786}
2787
2788/**
2789 * ata_eh_report - report error handling to user
2790 * @ap: ATA port to report EH about
2791 *
2792 * Report EH to user.
2793 *
2794 * LOCKING:
2795 * None.
2796 */
2797void ata_eh_report(struct ata_port *ap)
2798{
2799 struct ata_link *link;
2800
2801 ata_for_each_link(link, ap, HOST_FIRST)
2802 ata_eh_link_report(link);
2803}
2804
2805static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2806 unsigned int *classes, unsigned long deadline,
2807 bool clear_classes)
2808{
2809 struct ata_device *dev;
2810
2811 if (clear_classes)
2812 ata_for_each_dev(dev, link, ALL)
2813 classes[dev->devno] = ATA_DEV_UNKNOWN;
2814
2815 return reset(link, classes, deadline);
2816}
2817
2818static bool ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2819{
2820 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2821 return false;
2822 if (rc == -EAGAIN)
2823 return true;
2824 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2825 return true;
2826 return false;
2827}
2828
2829int ata_eh_reset(struct ata_port *ap, struct ata_link *link, int classify,
2830 struct ata_reset_operations *reset_ops)
2831 __must_hold(&ap->host->eh_mutex)
2832{
2833 struct ata_link *slave = ap->slave_link;
2834 struct ata_eh_context *ehc = &link->eh_context;
2835 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2836 ata_reset_fn_t hardreset = reset_ops->hardreset;
2837 ata_reset_fn_t softreset = reset_ops->softreset;
2838 ata_prereset_fn_t prereset = reset_ops->prereset;
2839 ata_postreset_fn_t postreset = reset_ops->postreset;
2840 unsigned int *classes = ehc->classes;
2841 unsigned int lflags = link->flags;
2842 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2843 int max_tries = 0, try = 0;
2844 struct ata_link *failed_link;
2845 struct ata_device *dev;
2846 unsigned long deadline, now;
2847 ata_reset_fn_t reset;
2848 unsigned long flags;
2849 u32 sstatus;
2850 int nr_unknown, rc;
2851
2852 /*
2853 * Prepare to reset
2854 */
2855 while (ata_eh_reset_timeouts[max_tries] != UINT_MAX)
2856 max_tries++;
2857 if (link->flags & ATA_LFLAG_RST_ONCE)
2858 max_tries = 1;
2859 if (link->flags & ATA_LFLAG_NO_HRST)
2860 hardreset = NULL;
2861 if (link->flags & ATA_LFLAG_NO_SRST)
2862 softreset = NULL;
2863
2864 /* make sure each reset attempt is at least COOL_DOWN apart */
2865 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2866 now = jiffies;
2867 WARN_ON(time_after(ehc->last_reset, now));
2868 deadline = ata_deadline(ehc->last_reset,
2869 ATA_EH_RESET_COOL_DOWN);
2870 if (time_before(now, deadline))
2871 schedule_timeout_uninterruptible(deadline - now);
2872 }
2873
2874 spin_lock_irqsave(ap->lock, flags);
2875 ap->pflags |= ATA_PFLAG_RESETTING;
2876 spin_unlock_irqrestore(ap->lock, flags);
2877
2878 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2879
2880 ata_for_each_dev(dev, link, ALL) {
2881 /* If we issue an SRST then an ATA drive (not ATAPI)
2882 * may change configuration and be in PIO0 timing. If
2883 * we do a hard reset (or are coming from power on)
2884 * this is true for ATA or ATAPI. Until we've set a
2885 * suitable controller mode we should not touch the
2886 * bus as we may be talking too fast.
2887 */
2888 dev->pio_mode = XFER_PIO_0;
2889 dev->dma_mode = 0xff;
2890
2891 /* If the controller has a pio mode setup function
2892 * then use it to set the chipset to rights. Don't
2893 * touch the DMA setup as that will be dealt with when
2894 * configuring devices.
2895 */
2896 if (ap->ops->set_piomode)
2897 ap->ops->set_piomode(ap, dev);
2898 }
2899
2900 /* prefer hardreset */
2901 reset = NULL;
2902 ehc->i.action &= ~ATA_EH_RESET;
2903 if (hardreset) {
2904 reset = hardreset;
2905 ehc->i.action |= ATA_EH_HARDRESET;
2906 } else if (softreset) {
2907 reset = softreset;
2908 ehc->i.action |= ATA_EH_SOFTRESET;
2909 }
2910
2911 if (prereset) {
2912 unsigned long deadline = ata_deadline(jiffies,
2913 ATA_EH_PRERESET_TIMEOUT);
2914
2915 if (slave) {
2916 sehc->i.action &= ~ATA_EH_RESET;
2917 sehc->i.action |= ehc->i.action;
2918 }
2919
2920 rc = prereset(link, deadline);
2921
2922 /* If present, do prereset on slave link too. Reset
2923 * is skipped iff both master and slave links report
2924 * -ENOENT or clear ATA_EH_RESET.
2925 */
2926 if (slave && (rc == 0 || rc == -ENOENT)) {
2927 int tmp;
2928
2929 tmp = prereset(slave, deadline);
2930 if (tmp != -ENOENT)
2931 rc = tmp;
2932
2933 ehc->i.action |= sehc->i.action;
2934 }
2935
2936 if (rc) {
2937 if (rc == -ENOENT) {
2938 ata_link_dbg(link, "port disabled--ignoring\n");
2939 ehc->i.action &= ~ATA_EH_RESET;
2940
2941 ata_for_each_dev(dev, link, ALL)
2942 classes[dev->devno] = ATA_DEV_NONE;
2943
2944 rc = 0;
2945 } else
2946 ata_link_err(link,
2947 "prereset failed (errno=%d)\n",
2948 rc);
2949 goto out;
2950 }
2951
2952 /* prereset() might have cleared ATA_EH_RESET. If so,
2953 * bang classes, thaw and return.
2954 */
2955 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2956 ata_for_each_dev(dev, link, ALL)
2957 classes[dev->devno] = ATA_DEV_NONE;
2958 if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2959 ata_eh_thaw_port(ap);
2960 rc = 0;
2961 goto out;
2962 }
2963 }
2964
2965 retry:
2966 /*
2967 * Perform reset
2968 */
2969 if (ata_is_host_link(link))
2970 ata_eh_freeze_port(ap);
2971
2972 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2973
2974 if (reset) {
2975 if (verbose)
2976 ata_link_info(link, "%s resetting link\n",
2977 reset == softreset ? "soft" : "hard");
2978
2979 /* mark that this EH session started with reset */
2980 ehc->last_reset = jiffies;
2981 if (reset == hardreset) {
2982 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2983 trace_ata_link_hardreset_begin(link, classes, deadline);
2984 } else {
2985 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2986 trace_ata_link_softreset_begin(link, classes, deadline);
2987 }
2988
2989 rc = ata_do_reset(link, reset, classes, deadline, true);
2990 if (reset == hardreset)
2991 trace_ata_link_hardreset_end(link, classes, rc);
2992 else
2993 trace_ata_link_softreset_end(link, classes, rc);
2994 if (rc && rc != -EAGAIN) {
2995 failed_link = link;
2996 goto fail;
2997 }
2998
2999 /* hardreset slave link if existent */
3000 if (slave && reset == hardreset) {
3001 int tmp;
3002
3003 if (verbose)
3004 ata_link_info(slave, "hard resetting link\n");
3005
3006 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
3007 trace_ata_slave_hardreset_begin(slave, classes,
3008 deadline);
3009 tmp = ata_do_reset(slave, reset, classes, deadline,
3010 false);
3011 trace_ata_slave_hardreset_end(slave, classes, tmp);
3012 switch (tmp) {
3013 case -EAGAIN:
3014 rc = -EAGAIN;
3015 break;
3016 case 0:
3017 break;
3018 default:
3019 failed_link = slave;
3020 rc = tmp;
3021 goto fail;
3022 }
3023 }
3024
3025 /* perform follow-up SRST if necessary */
3026 if (reset == hardreset &&
3027 ata_eh_followup_srst_needed(link, rc)) {
3028 reset = softreset;
3029
3030 if (!reset) {
3031 ata_link_err(link,
3032 "follow-up softreset required but no softreset available\n");
3033 failed_link = link;
3034 rc = -EINVAL;
3035 goto fail;
3036 }
3037
3038 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
3039 trace_ata_link_softreset_begin(link, classes, deadline);
3040 rc = ata_do_reset(link, reset, classes, deadline, true);
3041 trace_ata_link_softreset_end(link, classes, rc);
3042 if (rc) {
3043 failed_link = link;
3044 goto fail;
3045 }
3046 }
3047 } else {
3048 if (verbose)
3049 ata_link_info(link,
3050 "no reset method available, skipping reset\n");
3051 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
3052 lflags |= ATA_LFLAG_ASSUME_ATA;
3053 }
3054
3055 /*
3056 * Post-reset processing
3057 */
3058 ata_for_each_dev(dev, link, ALL) {
3059 /* After the reset, the device state is PIO 0 and the
3060 * controller state is undefined. Reset also wakes up
3061 * drives from sleeping mode.
3062 */
3063 dev->pio_mode = XFER_PIO_0;
3064 dev->flags &= ~ATA_DFLAG_SLEEPING;
3065
3066 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3067 continue;
3068
3069 /* apply class override */
3070 if (lflags & ATA_LFLAG_ASSUME_ATA)
3071 classes[dev->devno] = ATA_DEV_ATA;
3072 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
3073 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
3074 }
3075
3076 /* record current link speed */
3077 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
3078 link->sata_spd = (sstatus >> 4) & 0xf;
3079 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
3080 slave->sata_spd = (sstatus >> 4) & 0xf;
3081
3082 /* thaw the port */
3083 if (ata_is_host_link(link))
3084 ata_eh_thaw_port(ap);
3085
3086 /* postreset() should clear hardware SError. Although SError
3087 * is cleared during link resume, clearing SError here is
3088 * necessary as some PHYs raise hotplug events after SRST.
3089 * This introduces race condition where hotplug occurs between
3090 * reset and here. This race is mediated by cross checking
3091 * link onlineness and classification result later.
3092 */
3093 if (postreset) {
3094 postreset(link, classes);
3095 trace_ata_link_postreset(link, classes, rc);
3096 if (slave) {
3097 postreset(slave, classes);
3098 trace_ata_slave_postreset(slave, classes, rc);
3099 }
3100 }
3101
3102 /* clear cached SError */
3103 spin_lock_irqsave(link->ap->lock, flags);
3104 link->eh_info.serror = 0;
3105 if (slave)
3106 slave->eh_info.serror = 0;
3107 spin_unlock_irqrestore(link->ap->lock, flags);
3108
3109 /*
3110 * Make sure onlineness and classification result correspond.
3111 * Hotplug could have happened during reset and some
3112 * controllers fail to wait while a drive is spinning up after
3113 * being hotplugged causing misdetection. By cross checking
3114 * link on/offlineness and classification result, those
3115 * conditions can be reliably detected and retried.
3116 */
3117 nr_unknown = 0;
3118 ata_for_each_dev(dev, link, ALL) {
3119 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
3120 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
3121 ata_dev_dbg(dev, "link online but device misclassified\n");
3122 classes[dev->devno] = ATA_DEV_NONE;
3123 nr_unknown++;
3124 }
3125 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3126 if (ata_class_enabled(classes[dev->devno]))
3127 ata_dev_dbg(dev,
3128 "link offline, clearing class %d to NONE\n",
3129 classes[dev->devno]);
3130 classes[dev->devno] = ATA_DEV_NONE;
3131 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
3132 ata_dev_dbg(dev,
3133 "link status unknown, clearing UNKNOWN to NONE\n");
3134 classes[dev->devno] = ATA_DEV_NONE;
3135 }
3136 }
3137
3138 if (classify && nr_unknown) {
3139 if (try < max_tries) {
3140 ata_link_warn(link,
3141 "link online but %d devices misclassified, retrying\n",
3142 nr_unknown);
3143 failed_link = link;
3144 rc = -EAGAIN;
3145 goto fail;
3146 }
3147 ata_link_warn(link,
3148 "link online but %d devices misclassified, "
3149 "device detection might fail\n", nr_unknown);
3150 }
3151
3152 /* reset successful, schedule revalidation */
3153 ata_eh_done(link, NULL, ATA_EH_RESET);
3154 if (slave)
3155 ata_eh_done(slave, NULL, ATA_EH_RESET);
3156 ehc->last_reset = jiffies; /* update to completion time */
3157 ehc->i.action |= ATA_EH_REVALIDATE;
3158 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
3159
3160 rc = 0;
3161 out:
3162 /* clear hotplug flag */
3163 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3164 if (slave)
3165 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3166
3167 spin_lock_irqsave(ap->lock, flags);
3168 ap->pflags &= ~ATA_PFLAG_RESETTING;
3169 spin_unlock_irqrestore(ap->lock, flags);
3170
3171 return rc;
3172
3173 fail:
3174 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
3175 if (!ata_is_host_link(link) &&
3176 sata_scr_read(link, SCR_STATUS, &sstatus))
3177 rc = -ERESTART;
3178
3179 if (try >= max_tries || rc == -ENODEV) {
3180 /*
3181 * Thaw host port even if reset failed, so that the port
3182 * can be retried on the next phy event. This risks
3183 * repeated EH runs but seems to be a better tradeoff than
3184 * shutting down a port after a botched hotplug attempt.
3185 */
3186 if (ata_is_host_link(link))
3187 ata_eh_thaw_port(ap);
3188 ata_link_warn(link, "%s failed\n",
3189 reset == hardreset ? "hardreset" : "softreset");
3190 goto out;
3191 }
3192
3193 now = jiffies;
3194 if (time_before(now, deadline)) {
3195 unsigned long delta = deadline - now;
3196
3197 ata_link_warn(failed_link,
3198 "reset failed (errno=%d), retrying in %u secs\n",
3199 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3200
3201 ata_eh_release(ap);
3202 while (delta)
3203 delta = schedule_timeout_uninterruptible(delta);
3204 ata_eh_acquire(ap);
3205 }
3206
3207 /*
3208 * While disks spinup behind PMP, some controllers fail sending SRST.
3209 * They need to be reset - as well as the PMP - before retrying.
3210 */
3211 if (rc == -ERESTART) {
3212 if (ata_is_host_link(link))
3213 ata_eh_thaw_port(ap);
3214 goto out;
3215 }
3216
3217 if (try == max_tries - 1) {
3218 sata_down_spd_limit(link, 0);
3219 if (slave)
3220 sata_down_spd_limit(slave, 0);
3221 } else if (rc == -EPIPE)
3222 sata_down_spd_limit(failed_link, 0);
3223
3224 if (hardreset)
3225 reset = hardreset;
3226 goto retry;
3227}
3228
3229static inline void ata_eh_pull_park_action(struct ata_port *ap)
3230{
3231 struct ata_link *link;
3232 struct ata_device *dev;
3233 unsigned long flags;
3234
3235 /*
3236 * This function can be thought of as an extended version of
3237 * ata_eh_about_to_do() specially crafted to accommodate the
3238 * requirements of ATA_EH_PARK handling. Since the EH thread
3239 * does not leave the do {} while () loop in ata_eh_recover as
3240 * long as the timeout for a park request to *one* device on
3241 * the port has not expired, and since we still want to pick
3242 * up park requests to other devices on the same port or
3243 * timeout updates for the same device, we have to pull
3244 * ATA_EH_PARK actions from eh_info into eh_context.i
3245 * ourselves at the beginning of each pass over the loop.
3246 *
3247 * Additionally, all write accesses to &ap->park_req_pending
3248 * through reinit_completion() (see below) or complete_all()
3249 * (see ata_scsi_park_store()) are protected by the host lock.
3250 * As a result we have that park_req_pending.done is zero on
3251 * exit from this function, i.e. when ATA_EH_PARK actions for
3252 * *all* devices on port ap have been pulled into the
3253 * respective eh_context structs. If, and only if,
3254 * park_req_pending.done is non-zero by the time we reach
3255 * wait_for_completion_timeout(), another ATA_EH_PARK action
3256 * has been scheduled for at least one of the devices on port
3257 * ap and we have to cycle over the do {} while () loop in
3258 * ata_eh_recover() again.
3259 */
3260
3261 spin_lock_irqsave(ap->lock, flags);
3262 reinit_completion(&ap->park_req_pending);
3263 ata_for_each_link(link, ap, EDGE) {
3264 ata_for_each_dev(dev, link, ALL) {
3265 struct ata_eh_info *ehi = &link->eh_info;
3266
3267 link->eh_context.i.dev_action[dev->devno] |=
3268 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3269 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3270 }
3271 }
3272 spin_unlock_irqrestore(ap->lock, flags);
3273}
3274
3275static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3276{
3277 struct ata_eh_context *ehc = &dev->link->eh_context;
3278 struct ata_taskfile tf;
3279 unsigned int err_mask;
3280
3281 ata_tf_init(dev, &tf);
3282 if (park) {
3283 ehc->unloaded_mask |= 1 << dev->devno;
3284 tf.command = ATA_CMD_IDLEIMMEDIATE;
3285 tf.feature = 0x44;
3286 tf.lbal = 0x4c;
3287 tf.lbam = 0x4e;
3288 tf.lbah = 0x55;
3289 } else {
3290 ehc->unloaded_mask &= ~(1 << dev->devno);
3291 tf.command = ATA_CMD_CHK_POWER;
3292 }
3293
3294 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3295 tf.protocol = ATA_PROT_NODATA;
3296 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3297 if (park && (err_mask || tf.lbal != 0xc4)) {
3298 ata_dev_err(dev, "head unload failed!\n");
3299 ehc->unloaded_mask &= ~(1 << dev->devno);
3300 }
3301}
3302
3303static int ata_eh_revalidate_and_attach(struct ata_link *link,
3304 struct ata_device **r_failed_dev)
3305{
3306 struct ata_port *ap = link->ap;
3307 struct ata_eh_context *ehc = &link->eh_context;
3308 struct ata_device *dev;
3309 unsigned int new_mask = 0;
3310 unsigned long flags;
3311 int rc = 0;
3312
3313 /* For PATA drive side cable detection to work, IDENTIFY must
3314 * be done backwards such that PDIAG- is released by the slave
3315 * device before the master device is identified.
3316 */
3317 ata_for_each_dev(dev, link, ALL_REVERSE) {
3318 unsigned int action = ata_eh_dev_action(dev);
3319 unsigned int readid_flags = 0;
3320
3321 if (ehc->i.flags & ATA_EHI_DID_RESET)
3322 readid_flags |= ATA_READID_POSTRESET;
3323
3324 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3325 WARN_ON(dev->class == ATA_DEV_PMP);
3326
3327 /*
3328 * The link may be in a deep sleep, wake it up.
3329 *
3330 * If the link is in deep sleep, ata_phys_link_offline()
3331 * will return true, causing the revalidation to fail,
3332 * which leads to a (potentially) needless hard reset.
3333 *
3334 * ata_eh_recover() will later restore the link policy
3335 * to ap->target_lpm_policy after revalidation is done.
3336 */
3337 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3338 rc = ata_eh_link_set_lpm(link, ATA_LPM_MAX_POWER,
3339 r_failed_dev);
3340 if (rc)
3341 goto err;
3342 }
3343
3344 if (!ata_eh_link_established(ata_dev_phys_link(dev))) {
3345 rc = -EIO;
3346 goto err;
3347 }
3348
3349 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3350 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3351 readid_flags);
3352 if (rc)
3353 goto err;
3354
3355 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3356
3357 /* Configuration may have changed, reconfigure
3358 * transfer mode.
3359 */
3360 ehc->i.flags |= ATA_EHI_SETMODE;
3361
3362 /* schedule the scsi_rescan_device() here */
3363 schedule_delayed_work(&ap->scsi_rescan_task, 0);
3364 } else if (dev->class == ATA_DEV_UNKNOWN &&
3365 ehc->tries[dev->devno] &&
3366 ata_class_enabled(ehc->classes[dev->devno])) {
3367 /* Temporarily set dev->class, it will be
3368 * permanently set once all configurations are
3369 * complete. This is necessary because new
3370 * device configuration is done in two
3371 * separate loops.
3372 */
3373 dev->class = ehc->classes[dev->devno];
3374
3375 if (dev->class == ATA_DEV_PMP)
3376 rc = sata_pmp_attach(dev);
3377 else
3378 rc = ata_dev_read_id(dev, &dev->class,
3379 readid_flags, dev->id);
3380
3381 /* read_id might have changed class, store and reset */
3382 ehc->classes[dev->devno] = dev->class;
3383 dev->class = ATA_DEV_UNKNOWN;
3384
3385 switch (rc) {
3386 case 0:
3387 /* clear error info accumulated during probe */
3388 ata_ering_clear(&dev->ering);
3389 new_mask |= 1 << dev->devno;
3390 break;
3391 case -ENOENT:
3392 /* IDENTIFY was issued to non-existent
3393 * device. No need to reset. Just
3394 * thaw and ignore the device.
3395 */
3396 ata_eh_thaw_port(ap);
3397 break;
3398 default:
3399 goto err;
3400 }
3401 }
3402 }
3403
3404 /* PDIAG- should have been released, ask cable type if post-reset */
3405 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3406 if (ap->ops->cable_detect)
3407 ap->cbl = ap->ops->cable_detect(ap);
3408 ata_force_cbl(ap);
3409 }
3410
3411 /* Configure new devices forward such that user doesn't see
3412 * device detection messages backwards.
3413 */
3414 ata_for_each_dev(dev, link, ALL) {
3415 if (!(new_mask & (1 << dev->devno)))
3416 continue;
3417
3418 dev->class = ehc->classes[dev->devno];
3419
3420 if (dev->class == ATA_DEV_PMP)
3421 continue;
3422
3423 ehc->i.flags |= ATA_EHI_PRINTINFO;
3424 rc = ata_dev_configure(dev);
3425 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3426 if (rc) {
3427 dev->class = ATA_DEV_UNKNOWN;
3428 goto err;
3429 }
3430
3431 spin_lock_irqsave(ap->lock, flags);
3432 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3433 spin_unlock_irqrestore(ap->lock, flags);
3434
3435 /* new device discovered, configure xfermode */
3436 ehc->i.flags |= ATA_EHI_SETMODE;
3437 }
3438
3439 return 0;
3440
3441 err:
3442 dev->flags &= ~ATA_DFLAG_RESUMING;
3443 *r_failed_dev = dev;
3444 return rc;
3445}
3446
3447/**
3448 * ata_eh_set_mode - Program timings and issue SET FEATURES - XFER
3449 * @link: link on which timings will be programmed
3450 * @r_failed_dev: out parameter for failed device
3451 *
3452 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3453 * ata_eh_set_mode() fails, pointer to the failing device is
3454 * returned in @r_failed_dev.
3455 *
3456 * LOCKING:
3457 * PCI/etc. bus probe sem.
3458 *
3459 * RETURNS:
3460 * 0 on success, negative errno otherwise
3461 */
3462static int ata_eh_set_mode(struct ata_link *link,
3463 struct ata_device **r_failed_dev)
3464{
3465 struct ata_port *ap = link->ap;
3466 struct ata_device *dev;
3467 int rc;
3468
3469 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3470 ata_for_each_dev(dev, link, ENABLED) {
3471 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3472 struct ata_ering_entry *ent;
3473
3474 ent = ata_ering_top(&dev->ering);
3475 if (ent)
3476 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3477 }
3478 }
3479
3480 /* has private set_mode? */
3481 if (ap->ops->set_mode)
3482 rc = ap->ops->set_mode(link, r_failed_dev);
3483 else
3484 rc = ata_set_mode(link, r_failed_dev);
3485
3486 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3487 ata_for_each_dev(dev, link, ENABLED) {
3488 struct ata_eh_context *ehc = &link->eh_context;
3489 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3490 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3491
3492 if (dev->xfer_mode != saved_xfer_mode ||
3493 ata_ncq_enabled(dev) != saved_ncq)
3494 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3495 }
3496
3497 return rc;
3498}
3499
3500/**
3501 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3502 * @dev: ATAPI device to clear UA for
3503 *
3504 * Resets and other operations can make an ATAPI device raise
3505 * UNIT ATTENTION which causes the next operation to fail. This
3506 * function clears UA.
3507 *
3508 * LOCKING:
3509 * EH context (may sleep).
3510 *
3511 * RETURNS:
3512 * 0 on success, -errno on failure.
3513 */
3514static int atapi_eh_clear_ua(struct ata_device *dev)
3515{
3516 int i;
3517
3518 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3519 u8 *sense_buffer = dev->sector_buf;
3520 u8 sense_key = 0;
3521 unsigned int err_mask;
3522
3523 err_mask = atapi_eh_tur(dev, &sense_key);
3524 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3525 ata_dev_warn(dev,
3526 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3527 err_mask);
3528 return -EIO;
3529 }
3530
3531 if (!err_mask || sense_key != UNIT_ATTENTION)
3532 return 0;
3533
3534 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3535 if (err_mask) {
3536 ata_dev_warn(dev, "failed to clear "
3537 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3538 return -EIO;
3539 }
3540 }
3541
3542 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3543 ATA_EH_UA_TRIES);
3544
3545 return 0;
3546}
3547
3548/**
3549 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3550 * @dev: ATA device which may need FLUSH retry
3551 *
3552 * If @dev failed FLUSH, it needs to be reported upper layer
3553 * immediately as it means that @dev failed to remap and already
3554 * lost at least a sector and further FLUSH retrials won't make
3555 * any difference to the lost sector. However, if FLUSH failed
3556 * for other reasons, for example transmission error, FLUSH needs
3557 * to be retried.
3558 *
3559 * This function determines whether FLUSH failure retry is
3560 * necessary and performs it if so.
3561 *
3562 * RETURNS:
3563 * 0 if EH can continue, -errno if EH needs to be repeated.
3564 */
3565static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3566{
3567 struct ata_link *link = dev->link;
3568 struct ata_port *ap = link->ap;
3569 struct ata_queued_cmd *qc;
3570 struct ata_taskfile tf;
3571 unsigned int err_mask;
3572 int rc = 0;
3573
3574 /* did flush fail for this device? */
3575 if (!ata_tag_valid(link->active_tag))
3576 return 0;
3577
3578 qc = __ata_qc_from_tag(ap, link->active_tag);
3579 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3580 qc->tf.command != ATA_CMD_FLUSH))
3581 return 0;
3582
3583 /* if the device failed it, it should be reported to upper layers */
3584 if (qc->err_mask & AC_ERR_DEV)
3585 return 0;
3586
3587 /* flush failed for some other reason, give it another shot */
3588 ata_tf_init(dev, &tf);
3589
3590 tf.command = qc->tf.command;
3591 tf.flags |= ATA_TFLAG_DEVICE;
3592 tf.protocol = ATA_PROT_NODATA;
3593
3594 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3595 tf.command, qc->err_mask);
3596
3597 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3598 if (!err_mask) {
3599 /*
3600 * FLUSH is complete but there's no way to
3601 * successfully complete a failed command from EH.
3602 * Making sure retry is allowed at least once and
3603 * retrying it should do the trick - whatever was in
3604 * the cache is already on the platter and this won't
3605 * cause infinite loop.
3606 */
3607 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3608 } else {
3609 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3610 err_mask);
3611 rc = -EIO;
3612
3613 /* if device failed it, report it to upper layers */
3614 if (err_mask & AC_ERR_DEV) {
3615 qc->err_mask |= AC_ERR_DEV;
3616 qc->result_tf = tf;
3617 if (!ata_port_is_frozen(ap))
3618 rc = 0;
3619 }
3620 }
3621 return rc;
3622}
3623
3624int ata_link_nr_enabled(struct ata_link *link)
3625{
3626 struct ata_device *dev;
3627 int cnt = 0;
3628
3629 ata_for_each_dev(dev, link, ENABLED)
3630 cnt++;
3631 return cnt;
3632}
3633
3634static int ata_link_nr_vacant(struct ata_link *link)
3635{
3636 struct ata_device *dev;
3637 int cnt = 0;
3638
3639 ata_for_each_dev(dev, link, ALL)
3640 if (dev->class == ATA_DEV_UNKNOWN)
3641 cnt++;
3642 return cnt;
3643}
3644
3645static int ata_eh_skip_recovery(struct ata_link *link)
3646{
3647 struct ata_port *ap = link->ap;
3648 struct ata_eh_context *ehc = &link->eh_context;
3649 struct ata_device *dev;
3650
3651 /* skip disabled links */
3652 if (link->flags & ATA_LFLAG_DISABLED)
3653 return 1;
3654
3655 /* skip if explicitly requested */
3656 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3657 return 1;
3658
3659 /* thaw frozen port and recover failed devices */
3660 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3661 return 0;
3662
3663 /* reset at least once if reset is requested */
3664 if ((ehc->i.action & ATA_EH_RESET) &&
3665 !(ehc->i.flags & ATA_EHI_DID_RESET))
3666 return 0;
3667
3668 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3669 ata_for_each_dev(dev, link, ALL) {
3670 if (dev->class == ATA_DEV_UNKNOWN &&
3671 ehc->classes[dev->devno] != ATA_DEV_NONE)
3672 return 0;
3673 }
3674
3675 return 1;
3676}
3677
3678static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3679{
3680 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3681 u64 now = get_jiffies_64();
3682 int *trials = void_arg;
3683
3684 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3685 (ent->timestamp < now - min(now, interval)))
3686 return -1;
3687
3688 (*trials)++;
3689 return 0;
3690}
3691
3692static int ata_eh_schedule_probe(struct ata_device *dev)
3693{
3694 struct ata_eh_context *ehc = &dev->link->eh_context;
3695 struct ata_link *link = ata_dev_phys_link(dev);
3696 int trials = 0;
3697
3698 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3699 (ehc->did_probe_mask & (1 << dev->devno)))
3700 return 0;
3701
3702 ata_eh_detach_dev(dev);
3703 ata_dev_init(dev);
3704 ehc->did_probe_mask |= (1 << dev->devno);
3705 ehc->i.action |= ATA_EH_RESET;
3706 ehc->saved_xfer_mode[dev->devno] = 0;
3707 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3708
3709 /* the link maybe in a deep sleep, wake it up */
3710 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3711 if (ata_is_host_link(link))
3712 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3713 ATA_LPM_EMPTY);
3714 else
3715 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3716 ATA_LPM_EMPTY);
3717 }
3718
3719 /* Record and count probe trials on the ering. The specific
3720 * error mask used is irrelevant. Because a successful device
3721 * detection clears the ering, this count accumulates only if
3722 * there are consecutive failed probes.
3723 *
3724 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3725 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3726 * forced to 1.5Gbps.
3727 *
3728 * This is to work around cases where failed link speed
3729 * negotiation results in device misdetection leading to
3730 * infinite DEVXCHG or PHRDY CHG events.
3731 */
3732 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3733 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3734
3735 if (trials > ATA_EH_PROBE_TRIALS)
3736 sata_down_spd_limit(link, 1);
3737
3738 return 1;
3739}
3740
3741static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3742{
3743 struct ata_eh_context *ehc = &dev->link->eh_context;
3744
3745 /* -EAGAIN from EH routine indicates retry without prejudice.
3746 * The requester is responsible for ensuring forward progress.
3747 */
3748 if (err != -EAGAIN)
3749 ehc->tries[dev->devno]--;
3750
3751 switch (err) {
3752 case -ENODEV:
3753 /* device missing or wrong IDENTIFY data, schedule probing */
3754 ehc->i.probe_mask |= (1 << dev->devno);
3755 fallthrough;
3756 case -EINVAL:
3757 /* give it just one more chance */
3758 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3759 fallthrough;
3760 case -EIO:
3761 if (ehc->tries[dev->devno] == 1) {
3762 /* This is the last chance, better to slow
3763 * down than lose it.
3764 */
3765 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3766 if (dev->pio_mode > XFER_PIO_0)
3767 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3768 }
3769 }
3770
3771 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3772 /* disable device if it has used up all its chances */
3773 ata_dev_disable(dev);
3774
3775 /* detach if offline */
3776 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3777 ata_eh_detach_dev(dev);
3778
3779 /* schedule probe if necessary */
3780 if (ata_eh_schedule_probe(dev)) {
3781 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3782 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3783 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3784 }
3785
3786 return 1;
3787 } else {
3788 ehc->i.action |= ATA_EH_RESET;
3789 return 0;
3790 }
3791}
3792
3793/**
3794 * ata_eh_recover - recover host port after error
3795 * @ap: host port to recover
3796 * @reset_ops: The set of reset operations to use
3797 * @r_failed_link: out parameter for failed link
3798 *
3799 * This is the alpha and omega, eum and yang, heart and soul of
3800 * libata exception handling. On entry, actions required to
3801 * recover each link and hotplug requests are recorded in the
3802 * link's eh_context. This function executes all the operations
3803 * with appropriate retrials and fallbacks to resurrect failed
3804 * devices, detach goners and greet newcomers.
3805 *
3806 * LOCKING:
3807 * Kernel thread context (may sleep).
3808 *
3809 * RETURNS:
3810 * 0 on success, -errno on failure.
3811 */
3812int ata_eh_recover(struct ata_port *ap, struct ata_reset_operations *reset_ops,
3813 struct ata_link **r_failed_link)
3814 __must_hold(&ap->host->eh_mutex)
3815{
3816 struct ata_link *link;
3817 struct ata_device *dev;
3818 int rc, nr_fails;
3819 unsigned long flags, deadline;
3820
3821 /* prep for recovery */
3822 ata_for_each_link(link, ap, EDGE) {
3823 struct ata_eh_context *ehc = &link->eh_context;
3824
3825 /* re-enable link? */
3826 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3827 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3828 spin_lock_irqsave(ap->lock, flags);
3829 link->flags &= ~ATA_LFLAG_DISABLED;
3830 spin_unlock_irqrestore(ap->lock, flags);
3831 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3832 }
3833
3834 ata_for_each_dev(dev, link, ALL) {
3835 if (link->flags & ATA_LFLAG_NO_RETRY)
3836 ehc->tries[dev->devno] = 1;
3837 else
3838 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3839
3840 /* collect port action mask recorded in dev actions */
3841 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3842 ~ATA_EH_PERDEV_MASK;
3843 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3844
3845 /* process hotplug request */
3846 if (dev->flags & ATA_DFLAG_DETACH)
3847 ata_eh_detach_dev(dev);
3848
3849 /* schedule probe if necessary */
3850 if (!ata_dev_enabled(dev))
3851 ata_eh_schedule_probe(dev);
3852 }
3853 }
3854
3855 retry:
3856 rc = 0;
3857
3858 /* if UNLOADING, finish immediately */
3859 if (ap->pflags & ATA_PFLAG_UNLOADING)
3860 goto out;
3861
3862 /* prep for EH */
3863 ata_for_each_link(link, ap, EDGE) {
3864 struct ata_eh_context *ehc = &link->eh_context;
3865
3866 /* skip EH if possible. */
3867 if (ata_eh_skip_recovery(link))
3868 ehc->i.action = 0;
3869
3870 ata_for_each_dev(dev, link, ALL)
3871 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3872 }
3873
3874 /* reset */
3875 ata_for_each_link(link, ap, EDGE) {
3876 struct ata_eh_context *ehc = &link->eh_context;
3877
3878 if (!(ehc->i.action & ATA_EH_RESET))
3879 continue;
3880
3881 rc = ata_eh_reset(ap, link, ata_link_nr_vacant(link),
3882 reset_ops);
3883 if (rc) {
3884 ata_link_err(link, "reset failed, giving up\n");
3885 goto out;
3886 }
3887 }
3888
3889 do {
3890 unsigned long now;
3891
3892 /*
3893 * clears ATA_EH_PARK in eh_info and resets
3894 * ap->park_req_pending
3895 */
3896 ata_eh_pull_park_action(ap);
3897
3898 deadline = jiffies;
3899 ata_for_each_link(link, ap, EDGE) {
3900 ata_for_each_dev(dev, link, ALL) {
3901 struct ata_eh_context *ehc = &link->eh_context;
3902 unsigned long tmp;
3903
3904 if (dev->class != ATA_DEV_ATA &&
3905 dev->class != ATA_DEV_ZAC)
3906 continue;
3907 if (!(ehc->i.dev_action[dev->devno] &
3908 ATA_EH_PARK))
3909 continue;
3910 tmp = dev->unpark_deadline;
3911 if (time_before(deadline, tmp))
3912 deadline = tmp;
3913 else if (time_before_eq(tmp, jiffies))
3914 continue;
3915 if (ehc->unloaded_mask & (1 << dev->devno))
3916 continue;
3917
3918 ata_eh_park_issue_cmd(dev, 1);
3919 }
3920 }
3921
3922 now = jiffies;
3923 if (time_before_eq(deadline, now))
3924 break;
3925
3926 ata_eh_release(ap);
3927 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3928 deadline - now);
3929 ata_eh_acquire(ap);
3930 } while (deadline);
3931 ata_for_each_link(link, ap, EDGE) {
3932 ata_for_each_dev(dev, link, ALL) {
3933 if (!(link->eh_context.unloaded_mask &
3934 (1 << dev->devno)))
3935 continue;
3936
3937 ata_eh_park_issue_cmd(dev, 0);
3938 ata_eh_done(link, dev, ATA_EH_PARK);
3939 }
3940 }
3941
3942 /* the rest */
3943 nr_fails = 0;
3944 ata_for_each_link(link, ap, PMP_FIRST) {
3945 struct ata_eh_context *ehc = &link->eh_context;
3946
3947 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3948 goto config_lpm;
3949
3950 /* revalidate existing devices and attach new ones */
3951 rc = ata_eh_revalidate_and_attach(link, &dev);
3952 if (rc)
3953 goto rest_fail;
3954
3955 /* if PMP got attached, return, pmp EH will take care of it */
3956 if (link->device->class == ATA_DEV_PMP) {
3957 ehc->i.action = 0;
3958 return 0;
3959 }
3960
3961 /* configure transfer mode if necessary */
3962 if (ehc->i.flags & ATA_EHI_SETMODE) {
3963 rc = ata_eh_set_mode(link, &dev);
3964 if (rc)
3965 goto rest_fail;
3966 ehc->i.flags &= ~ATA_EHI_SETMODE;
3967 }
3968
3969 /* If reset has been issued, clear UA to avoid
3970 * disrupting the current users of the device.
3971 */
3972 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3973 ata_for_each_dev(dev, link, ALL) {
3974 if (dev->class != ATA_DEV_ATAPI)
3975 continue;
3976 rc = atapi_eh_clear_ua(dev);
3977 if (rc)
3978 goto rest_fail;
3979 if (zpodd_dev_enabled(dev))
3980 zpodd_post_poweron(dev);
3981 }
3982 }
3983
3984 /*
3985 * Make sure to transition devices to the active power mode
3986 * if needed (e.g. if we were scheduled on system resume).
3987 */
3988 ata_for_each_dev(dev, link, ENABLED) {
3989 if (ehc->i.dev_action[dev->devno] & ATA_EH_SET_ACTIVE) {
3990 ata_dev_power_set_active(dev);
3991 ata_eh_done(link, dev, ATA_EH_SET_ACTIVE);
3992 }
3993 }
3994
3995 /* retry flush if necessary */
3996 ata_for_each_dev(dev, link, ALL) {
3997 if (dev->class != ATA_DEV_ATA &&
3998 dev->class != ATA_DEV_ZAC)
3999 continue;
4000 rc = ata_eh_maybe_retry_flush(dev);
4001 if (rc)
4002 goto rest_fail;
4003 }
4004
4005 config_lpm:
4006 /* configure link power saving */
4007 if (link->lpm_policy != ap->target_lpm_policy) {
4008 rc = ata_eh_link_set_lpm(link, ap->target_lpm_policy,
4009 &dev);
4010 if (rc)
4011 goto rest_fail;
4012 }
4013
4014 /* this link is okay now */
4015 ehc->i.flags = 0;
4016 continue;
4017
4018 rest_fail:
4019 nr_fails++;
4020 if (dev)
4021 ata_eh_handle_dev_fail(dev, rc);
4022
4023 if (ata_port_is_frozen(ap)) {
4024 /* PMP reset requires working host port.
4025 * Can't retry if it's frozen.
4026 */
4027 if (sata_pmp_attached(ap))
4028 goto out;
4029 break;
4030 }
4031 }
4032
4033 if (nr_fails)
4034 goto retry;
4035
4036 out:
4037 if (rc && r_failed_link)
4038 *r_failed_link = link;
4039
4040 return rc;
4041}
4042
4043/**
4044 * ata_eh_finish - finish up EH
4045 * @ap: host port to finish EH for
4046 *
4047 * Recovery is complete. Clean up EH states and retry or finish
4048 * failed qcs.
4049 *
4050 * LOCKING:
4051 * None.
4052 */
4053void ata_eh_finish(struct ata_port *ap)
4054{
4055 struct ata_queued_cmd *qc;
4056 int tag;
4057
4058 /* retry or finish qcs */
4059 ata_qc_for_each_raw(ap, qc, tag) {
4060 if (!(qc->flags & ATA_QCFLAG_EH))
4061 continue;
4062
4063 if (qc->err_mask) {
4064 /* FIXME: Once EH migration is complete,
4065 * generate sense data in this function,
4066 * considering both err_mask and tf.
4067 */
4068 if (qc->flags & ATA_QCFLAG_RETRY) {
4069 /*
4070 * Since qc->err_mask is set, ata_eh_qc_retry()
4071 * will not increment scmd->allowed, so upper
4072 * layer will only retry the command if it has
4073 * not already been retried too many times.
4074 */
4075 ata_eh_qc_retry(qc);
4076 } else {
4077 ata_eh_qc_complete(qc);
4078 }
4079 } else {
4080 if (qc->flags & ATA_QCFLAG_SENSE_VALID ||
4081 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) {
4082 ata_eh_qc_complete(qc);
4083 } else {
4084 /* feed zero TF to sense generation */
4085 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
4086 /*
4087 * Since qc->err_mask is not set,
4088 * ata_eh_qc_retry() will increment
4089 * scmd->allowed, so upper layer is guaranteed
4090 * to retry the command.
4091 */
4092 ata_eh_qc_retry(qc);
4093 }
4094 }
4095 }
4096
4097 /* make sure nr_active_links is zero after EH */
4098 WARN_ON(ap->nr_active_links);
4099 ap->nr_active_links = 0;
4100}
4101
4102/**
4103 * ata_std_error_handler - standard error handler
4104 * @ap: host port to handle error for
4105 *
4106 * Perform standard error handling sequence.
4107 *
4108 * LOCKING:
4109 * Kernel thread context (may sleep).
4110 */
4111void ata_std_error_handler(struct ata_port *ap)
4112 __must_hold(&ap->host->eh_mutex)
4113{
4114 struct ata_reset_operations *reset_ops = &ap->ops->reset;
4115 struct ata_link *link = &ap->link;
4116 int rc;
4117
4118 /* Ignore built-in hardresets if SCR access is not available */
4119 if ((reset_ops->hardreset == sata_std_hardreset ||
4120 reset_ops->hardreset == sata_sff_hardreset) &&
4121 !sata_scr_valid(link))
4122 link->flags |= ATA_LFLAG_NO_HRST;
4123
4124 ata_eh_autopsy(ap);
4125 ata_eh_report(ap);
4126
4127 rc = ata_eh_recover(ap, reset_ops, NULL);
4128 if (rc) {
4129 struct ata_device *dev;
4130
4131 ata_for_each_dev(dev, link, ALL)
4132 ata_dev_disable(dev);
4133 }
4134
4135 ata_eh_finish(ap);
4136}
4137EXPORT_SYMBOL_GPL(ata_std_error_handler);
4138
4139#ifdef CONFIG_PM
4140/**
4141 * ata_eh_handle_port_suspend - perform port suspend operation
4142 * @ap: port to suspend
4143 *
4144 * Suspend @ap.
4145 *
4146 * LOCKING:
4147 * Kernel thread context (may sleep).
4148 */
4149static void ata_eh_handle_port_suspend(struct ata_port *ap)
4150{
4151 unsigned long flags;
4152 int rc = 0;
4153 struct ata_device *dev;
4154 struct ata_link *link;
4155
4156 /* are we suspending? */
4157 spin_lock_irqsave(ap->lock, flags);
4158 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4159 ap->pm_mesg.event & PM_EVENT_RESUME) {
4160 spin_unlock_irqrestore(ap->lock, flags);
4161 return;
4162 }
4163 spin_unlock_irqrestore(ap->lock, flags);
4164
4165 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4166
4167 /*
4168 * We will reach this point for all of the PM events:
4169 * PM_EVENT_SUSPEND (if runtime pm, PM_EVENT_AUTO will also be set)
4170 * PM_EVENT_FREEZE, and PM_EVENT_HIBERNATE.
4171 *
4172 * We do not want to perform disk spin down for PM_EVENT_FREEZE.
4173 * (Spin down will be performed by the subsequent PM_EVENT_HIBERNATE.)
4174 */
4175 if (!(ap->pm_mesg.event & PM_EVENT_FREEZE)) {
4176 /* Set all devices attached to the port in standby mode */
4177 ata_for_each_link(link, ap, HOST_FIRST) {
4178 ata_for_each_dev(dev, link, ENABLED)
4179 ata_dev_power_set_standby(dev);
4180 }
4181 }
4182
4183 /*
4184 * If we have a ZPODD attached, check its zero
4185 * power ready status before the port is frozen.
4186 * Only needed for runtime suspend.
4187 */
4188 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4189 ata_for_each_dev(dev, &ap->link, ENABLED) {
4190 if (zpodd_dev_enabled(dev))
4191 zpodd_on_suspend(dev);
4192 }
4193 }
4194
4195 /* suspend */
4196 ata_eh_freeze_port(ap);
4197
4198 if (ap->ops->port_suspend)
4199 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4200
4201 ata_acpi_set_state(ap, ap->pm_mesg);
4202
4203 /* update the flags */
4204 spin_lock_irqsave(ap->lock, flags);
4205
4206 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4207 if (rc == 0)
4208 ap->pflags |= ATA_PFLAG_SUSPENDED;
4209 else if (ata_port_is_frozen(ap))
4210 ata_port_schedule_eh(ap);
4211
4212 spin_unlock_irqrestore(ap->lock, flags);
4213
4214 return;
4215}
4216
4217/**
4218 * ata_eh_handle_port_resume - perform port resume operation
4219 * @ap: port to resume
4220 *
4221 * Resume @ap.
4222 *
4223 * LOCKING:
4224 * Kernel thread context (may sleep).
4225 */
4226static void ata_eh_handle_port_resume(struct ata_port *ap)
4227{
4228 struct ata_link *link;
4229 struct ata_device *dev;
4230 unsigned long flags;
4231
4232 /* are we resuming? */
4233 spin_lock_irqsave(ap->lock, flags);
4234 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4235 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4236 spin_unlock_irqrestore(ap->lock, flags);
4237 return;
4238 }
4239 spin_unlock_irqrestore(ap->lock, flags);
4240
4241 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4242
4243 /*
4244 * Error timestamps are in jiffies which doesn't run while
4245 * suspended and PHY events during resume isn't too uncommon.
4246 * When the two are combined, it can lead to unnecessary speed
4247 * downs if the machine is suspended and resumed repeatedly.
4248 * Clear error history.
4249 */
4250 ata_for_each_link(link, ap, HOST_FIRST)
4251 ata_for_each_dev(dev, link, ALL)
4252 ata_ering_clear(&dev->ering);
4253
4254 ata_acpi_set_state(ap, ap->pm_mesg);
4255
4256 if (ap->ops->port_resume)
4257 ap->ops->port_resume(ap);
4258
4259 /* tell ACPI that we're resuming */
4260 ata_acpi_on_resume(ap);
4261
4262 /* update the flags */
4263 spin_lock_irqsave(ap->lock, flags);
4264 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4265 ap->pflags |= ATA_PFLAG_RESUMING;
4266 spin_unlock_irqrestore(ap->lock, flags);
4267}
4268#endif /* CONFIG_PM */