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