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