1 /*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include <scsi/scsi_dbg.h>
44 #include "../scsi/scsi_transport_api.h"
45
46 #include <linux/libata.h>
47
48 #include "libata.h"
49
50 enum {
51 /* speed down verdicts */
52 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
55 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
56
57 /* error flags */
58 ATA_EFLAG_IS_IO = (1 << 0),
59 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
60 ATA_EFLAG_OLD_ER = (1 << 31),
61
62 /* error categories */
63 ATA_ECAT_NONE = 0,
64 ATA_ECAT_ATA_BUS = 1,
65 ATA_ECAT_TOUT_HSM = 2,
66 ATA_ECAT_UNK_DEV = 3,
67 ATA_ECAT_DUBIOUS_NONE = 4,
68 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
69 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
70 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_ECAT_NR = 8,
72
73 ATA_EH_CMD_DFL_TIMEOUT = 5000,
74
75 /* always put at least this amount of time between resets */
76 ATA_EH_RESET_COOL_DOWN = 5000,
77
78 /* Waiting in ->prereset can never be reliable. It's
79 * sometimes nice to wait there but it can't be depended upon;
80 * otherwise, we wouldn't be resetting. Just give it enough
81 * time for most drives to spin up.
82 */
83 ATA_EH_PRERESET_TIMEOUT = 10000,
84 ATA_EH_FASTDRAIN_INTERVAL = 3000,
85
86 ATA_EH_UA_TRIES = 5,
87
88 /* probe speed down parameters, see ata_eh_schedule_probe() */
89 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
90 ATA_EH_PROBE_TRIALS = 2,
91 };
92
93 /* The following table determines how we sequence resets. Each entry
94 * represents timeout for that try. The first try can be soft or
95 * hardreset. All others are hardreset if available. In most cases
96 * the first reset w/ 10sec timeout should succeed. Following entries
97 * are mostly for error handling, hotplug and retarded devices.
98 */
99 static const unsigned long ata_eh_reset_timeouts[] = {
100 10000, /* most drives spin up by 10sec */
101 10000, /* > 99% working drives spin up before 20sec */
102 35000, /* give > 30 secs of idleness for retarded devices */
103 5000, /* and sweet one last chance */
104 ULONG_MAX, /* > 1 min has elapsed, give up */
105 };
106
107 static const unsigned long ata_eh_identify_timeouts[] = {
108 5000, /* covers > 99% of successes and not too boring on failures */
109 10000, /* combined time till here is enough even for media access */
110 30000, /* for true idiots */
111 ULONG_MAX,
112 };
113
114 static const unsigned long ata_eh_flush_timeouts[] = {
115 15000, /* be generous with flush */
116 15000, /* ditto */
117 30000, /* and even more generous */
118 ULONG_MAX,
119 };
120
121 static const unsigned long ata_eh_other_timeouts[] = {
122 5000, /* same rationale as identify timeout */
123 10000, /* ditto */
124 /* but no merciful 30sec for other commands, it just isn't worth it */
125 ULONG_MAX,
126 };
127
128 struct ata_eh_cmd_timeout_ent {
129 const u8 *commands;
130 const unsigned long *timeouts;
131 };
132
133 /* The following table determines timeouts to use for EH internal
134 * commands. Each table entry is a command class and matches the
135 * commands the entry applies to and the timeout table to use.
136 *
137 * On the retry after a command timed out, the next timeout value from
138 * the table is used. If the table doesn't contain further entries,
139 * the last value is used.
140 *
141 * ehc->cmd_timeout_idx keeps track of which timeout to use per
142 * command class, so if SET_FEATURES times out on the first try, the
143 * next try will use the second timeout value only for that class.
144 */
145 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
146 static const struct ata_eh_cmd_timeout_ent
147 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
148 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
149 .timeouts = ata_eh_identify_timeouts, },
150 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
151 .timeouts = ata_eh_other_timeouts, },
152 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
153 .timeouts = ata_eh_other_timeouts, },
154 { .commands = CMDS(ATA_CMD_SET_FEATURES),
155 .timeouts = ata_eh_other_timeouts, },
156 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
157 .timeouts = ata_eh_other_timeouts, },
158 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
159 .timeouts = ata_eh_flush_timeouts },
160 };
161 #undef CMDS
162
163 static void __ata_port_freeze(struct ata_port *ap);
164 #ifdef CONFIG_PM
165 static void ata_eh_handle_port_suspend(struct ata_port *ap);
166 static void ata_eh_handle_port_resume(struct ata_port *ap);
167 #else /* CONFIG_PM */
ata_eh_handle_port_suspend(struct ata_port * ap)168 static void ata_eh_handle_port_suspend(struct ata_port *ap)
169 { }
170
ata_eh_handle_port_resume(struct ata_port * ap)171 static void ata_eh_handle_port_resume(struct ata_port *ap)
172 { }
173 #endif /* CONFIG_PM */
174
__ata_ehi_pushv_desc(struct ata_eh_info * ehi,const char * fmt,va_list args)175 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
176 va_list args)
177 {
178 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
179 ATA_EH_DESC_LEN - ehi->desc_len,
180 fmt, args);
181 }
182
183 /**
184 * __ata_ehi_push_desc - push error description without adding separator
185 * @ehi: target EHI
186 * @fmt: printf format string
187 *
188 * Format string according to @fmt and append it to @ehi->desc.
189 *
190 * LOCKING:
191 * spin_lock_irqsave(host lock)
192 */
__ata_ehi_push_desc(struct ata_eh_info * ehi,const char * fmt,...)193 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
194 {
195 va_list args;
196
197 va_start(args, fmt);
198 __ata_ehi_pushv_desc(ehi, fmt, args);
199 va_end(args);
200 }
201
202 /**
203 * ata_ehi_push_desc - push error description with separator
204 * @ehi: target EHI
205 * @fmt: printf format string
206 *
207 * Format string according to @fmt and append it to @ehi->desc.
208 * If @ehi->desc is not empty, ", " is added in-between.
209 *
210 * LOCKING:
211 * spin_lock_irqsave(host lock)
212 */
ata_ehi_push_desc(struct ata_eh_info * ehi,const char * fmt,...)213 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
214 {
215 va_list args;
216
217 if (ehi->desc_len)
218 __ata_ehi_push_desc(ehi, ", ");
219
220 va_start(args, fmt);
221 __ata_ehi_pushv_desc(ehi, fmt, args);
222 va_end(args);
223 }
224
225 /**
226 * ata_ehi_clear_desc - clean error description
227 * @ehi: target EHI
228 *
229 * Clear @ehi->desc.
230 *
231 * LOCKING:
232 * spin_lock_irqsave(host lock)
233 */
ata_ehi_clear_desc(struct ata_eh_info * ehi)234 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
235 {
236 ehi->desc[0] = '\0';
237 ehi->desc_len = 0;
238 }
239
240 /**
241 * ata_port_desc - append port description
242 * @ap: target ATA port
243 * @fmt: printf format string
244 *
245 * Format string according to @fmt and append it to port
246 * description. If port description is not empty, " " is added
247 * in-between. This function is to be used while initializing
248 * ata_host. The description is printed on host registration.
249 *
250 * LOCKING:
251 * None.
252 */
ata_port_desc(struct ata_port * ap,const char * fmt,...)253 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
254 {
255 va_list args;
256
257 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
258
259 if (ap->link.eh_info.desc_len)
260 __ata_ehi_push_desc(&ap->link.eh_info, " ");
261
262 va_start(args, fmt);
263 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
264 va_end(args);
265 }
266
267 #ifdef CONFIG_PCI
268
269 /**
270 * ata_port_pbar_desc - append PCI BAR description
271 * @ap: target ATA port
272 * @bar: target PCI BAR
273 * @offset: offset into PCI BAR
274 * @name: name of the area
275 *
276 * If @offset is negative, this function formats a string which
277 * contains the name, address, size and type of the BAR and
278 * appends it to the port description. If @offset is zero or
279 * positive, only name and offsetted address is appended.
280 *
281 * LOCKING:
282 * None.
283 */
ata_port_pbar_desc(struct ata_port * ap,int bar,ssize_t offset,const char * name)284 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
285 const char *name)
286 {
287 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
288 char *type = "";
289 unsigned long long start, len;
290
291 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
292 type = "m";
293 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
294 type = "i";
295
296 start = (unsigned long long)pci_resource_start(pdev, bar);
297 len = (unsigned long long)pci_resource_len(pdev, bar);
298
299 if (offset < 0)
300 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
301 else
302 ata_port_desc(ap, "%s 0x%llx", name,
303 start + (unsigned long long)offset);
304 }
305
306 #endif /* CONFIG_PCI */
307
ata_lookup_timeout_table(u8 cmd)308 static int ata_lookup_timeout_table(u8 cmd)
309 {
310 int i;
311
312 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
313 const u8 *cur;
314
315 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
316 if (*cur == cmd)
317 return i;
318 }
319
320 return -1;
321 }
322
323 /**
324 * ata_internal_cmd_timeout - determine timeout for an internal command
325 * @dev: target device
326 * @cmd: internal command to be issued
327 *
328 * Determine timeout for internal command @cmd for @dev.
329 *
330 * LOCKING:
331 * EH context.
332 *
333 * RETURNS:
334 * Determined timeout.
335 */
ata_internal_cmd_timeout(struct ata_device * dev,u8 cmd)336 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
337 {
338 struct ata_eh_context *ehc = &dev->link->eh_context;
339 int ent = ata_lookup_timeout_table(cmd);
340 int idx;
341
342 if (ent < 0)
343 return ATA_EH_CMD_DFL_TIMEOUT;
344
345 idx = ehc->cmd_timeout_idx[dev->devno][ent];
346 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
347 }
348
349 /**
350 * ata_internal_cmd_timed_out - notification for internal command timeout
351 * @dev: target device
352 * @cmd: internal command which timed out
353 *
354 * Notify EH that internal command @cmd for @dev timed out. This
355 * function should be called only for commands whose timeouts are
356 * determined using ata_internal_cmd_timeout().
357 *
358 * LOCKING:
359 * EH context.
360 */
ata_internal_cmd_timed_out(struct ata_device * dev,u8 cmd)361 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
362 {
363 struct ata_eh_context *ehc = &dev->link->eh_context;
364 int ent = ata_lookup_timeout_table(cmd);
365 int idx;
366
367 if (ent < 0)
368 return;
369
370 idx = ehc->cmd_timeout_idx[dev->devno][ent];
371 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
372 ehc->cmd_timeout_idx[dev->devno][ent]++;
373 }
374
ata_ering_record(struct ata_ering * ering,unsigned int eflags,unsigned int err_mask)375 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
376 unsigned int err_mask)
377 {
378 struct ata_ering_entry *ent;
379
380 WARN_ON(!err_mask);
381
382 ering->cursor++;
383 ering->cursor %= ATA_ERING_SIZE;
384
385 ent = &ering->ring[ering->cursor];
386 ent->eflags = eflags;
387 ent->err_mask = err_mask;
388 ent->timestamp = get_jiffies_64();
389 }
390
ata_ering_top(struct ata_ering * ering)391 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
392 {
393 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
394
395 if (ent->err_mask)
396 return ent;
397 return NULL;
398 }
399
ata_ering_map(struct ata_ering * ering,int (* map_fn)(struct ata_ering_entry *,void *),void * arg)400 int ata_ering_map(struct ata_ering *ering,
401 int (*map_fn)(struct ata_ering_entry *, void *),
402 void *arg)
403 {
404 int idx, rc = 0;
405 struct ata_ering_entry *ent;
406
407 idx = ering->cursor;
408 do {
409 ent = &ering->ring[idx];
410 if (!ent->err_mask)
411 break;
412 rc = map_fn(ent, arg);
413 if (rc)
414 break;
415 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
416 } while (idx != ering->cursor);
417
418 return rc;
419 }
420
ata_ering_clear_cb(struct ata_ering_entry * ent,void * void_arg)421 int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
422 {
423 ent->eflags |= ATA_EFLAG_OLD_ER;
424 return 0;
425 }
426
ata_ering_clear(struct ata_ering * ering)427 static void ata_ering_clear(struct ata_ering *ering)
428 {
429 ata_ering_map(ering, ata_ering_clear_cb, NULL);
430 }
431
ata_eh_dev_action(struct ata_device * dev)432 static unsigned int ata_eh_dev_action(struct ata_device *dev)
433 {
434 struct ata_eh_context *ehc = &dev->link->eh_context;
435
436 return ehc->i.action | ehc->i.dev_action[dev->devno];
437 }
438
ata_eh_clear_action(struct ata_link * link,struct ata_device * dev,struct ata_eh_info * ehi,unsigned int action)439 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
440 struct ata_eh_info *ehi, unsigned int action)
441 {
442 struct ata_device *tdev;
443
444 if (!dev) {
445 ehi->action &= ~action;
446 ata_for_each_dev(tdev, link, ALL)
447 ehi->dev_action[tdev->devno] &= ~action;
448 } else {
449 /* doesn't make sense for port-wide EH actions */
450 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
451
452 /* break ehi->action into ehi->dev_action */
453 if (ehi->action & action) {
454 ata_for_each_dev(tdev, link, ALL)
455 ehi->dev_action[tdev->devno] |=
456 ehi->action & action;
457 ehi->action &= ~action;
458 }
459
460 /* turn off the specified per-dev action */
461 ehi->dev_action[dev->devno] &= ~action;
462 }
463 }
464
465 /**
466 * ata_eh_acquire - acquire EH ownership
467 * @ap: ATA port to acquire EH ownership for
468 *
469 * Acquire EH ownership for @ap. This is the basic exclusion
470 * mechanism for ports sharing a host. Only one port hanging off
471 * the same host can claim the ownership of EH.
472 *
473 * LOCKING:
474 * EH context.
475 */
ata_eh_acquire(struct ata_port * ap)476 void ata_eh_acquire(struct ata_port *ap)
477 {
478 mutex_lock(&ap->host->eh_mutex);
479 WARN_ON_ONCE(ap->host->eh_owner);
480 ap->host->eh_owner = current;
481 }
482
483 /**
484 * ata_eh_release - release EH ownership
485 * @ap: ATA port to release EH ownership for
486 *
487 * Release EH ownership for @ap if the caller. The caller must
488 * have acquired EH ownership using ata_eh_acquire() previously.
489 *
490 * LOCKING:
491 * EH context.
492 */
ata_eh_release(struct ata_port * ap)493 void ata_eh_release(struct ata_port *ap)
494 {
495 WARN_ON_ONCE(ap->host->eh_owner != current);
496 ap->host->eh_owner = NULL;
497 mutex_unlock(&ap->host->eh_mutex);
498 }
499
500 /**
501 * ata_scsi_timed_out - SCSI layer time out callback
502 * @cmd: timed out SCSI command
503 *
504 * Handles SCSI layer timeout. We race with normal completion of
505 * the qc for @cmd. If the qc is already gone, we lose and let
506 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
507 * timed out and EH should be invoked. Prevent ata_qc_complete()
508 * from finishing it by setting EH_SCHEDULED and return
509 * EH_NOT_HANDLED.
510 *
511 * TODO: kill this function once old EH is gone.
512 *
513 * LOCKING:
514 * Called from timer context
515 *
516 * RETURNS:
517 * EH_HANDLED or EH_NOT_HANDLED
518 */
ata_scsi_timed_out(struct scsi_cmnd * cmd)519 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
520 {
521 struct Scsi_Host *host = cmd->device->host;
522 struct ata_port *ap = ata_shost_to_port(host);
523 unsigned long flags;
524 struct ata_queued_cmd *qc;
525 enum blk_eh_timer_return ret;
526
527 DPRINTK("ENTER\n");
528
529 if (ap->ops->error_handler) {
530 ret = BLK_EH_NOT_HANDLED;
531 goto out;
532 }
533
534 ret = BLK_EH_HANDLED;
535 spin_lock_irqsave(ap->lock, flags);
536 qc = ata_qc_from_tag(ap, ap->link.active_tag);
537 if (qc) {
538 WARN_ON(qc->scsicmd != cmd);
539 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
540 qc->err_mask |= AC_ERR_TIMEOUT;
541 ret = BLK_EH_NOT_HANDLED;
542 }
543 spin_unlock_irqrestore(ap->lock, flags);
544
545 out:
546 DPRINTK("EXIT, ret=%d\n", ret);
547 return ret;
548 }
549
ata_eh_unload(struct ata_port * ap)550 static void ata_eh_unload(struct ata_port *ap)
551 {
552 struct ata_link *link;
553 struct ata_device *dev;
554 unsigned long flags;
555
556 /* Restore SControl IPM and SPD for the next driver and
557 * disable attached devices.
558 */
559 ata_for_each_link(link, ap, PMP_FIRST) {
560 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
561 ata_for_each_dev(dev, link, ALL)
562 ata_dev_disable(dev);
563 }
564
565 /* freeze and set UNLOADED */
566 spin_lock_irqsave(ap->lock, flags);
567
568 ata_port_freeze(ap); /* won't be thawed */
569 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
570 ap->pflags |= ATA_PFLAG_UNLOADED;
571
572 spin_unlock_irqrestore(ap->lock, flags);
573 }
574
575 /**
576 * ata_scsi_error - SCSI layer error handler callback
577 * @host: SCSI host on which error occurred
578 *
579 * Handles SCSI-layer-thrown error events.
580 *
581 * LOCKING:
582 * Inherited from SCSI layer (none, can sleep)
583 *
584 * RETURNS:
585 * Zero.
586 */
ata_scsi_error(struct Scsi_Host * host)587 void ata_scsi_error(struct Scsi_Host *host)
588 {
589 struct ata_port *ap = ata_shost_to_port(host);
590 unsigned long flags;
591 LIST_HEAD(eh_work_q);
592
593 DPRINTK("ENTER\n");
594
595 spin_lock_irqsave(host->host_lock, flags);
596 list_splice_init(&host->eh_cmd_q, &eh_work_q);
597 spin_unlock_irqrestore(host->host_lock, flags);
598
599 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
600
601 /* If we timed raced normal completion and there is nothing to
602 recover nr_timedout == 0 why exactly are we doing error recovery ? */
603 ata_scsi_port_error_handler(host, ap);
604
605 /* finish or retry handled scmd's and clean up */
606 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
607
608 DPRINTK("EXIT\n");
609 }
610
611 /**
612 * ata_scsi_cmd_error_handler - error callback for a list of commands
613 * @host: scsi host containing the port
614 * @ap: ATA port within the host
615 * @eh_work_q: list of commands to process
616 *
617 * process the given list of commands and return those finished to the
618 * ap->eh_done_q. This function is the first part of the libata error
619 * handler which processes a given list of failed commands.
620 */
ata_scsi_cmd_error_handler(struct Scsi_Host * host,struct ata_port * ap,struct list_head * eh_work_q)621 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
622 struct list_head *eh_work_q)
623 {
624 int i;
625 unsigned long flags;
626
627 /* make sure sff pio task is not running */
628 ata_sff_flush_pio_task(ap);
629
630 /* synchronize with host lock and sort out timeouts */
631
632 /* For new EH, all qcs are finished in one of three ways -
633 * normal completion, error completion, and SCSI timeout.
634 * Both completions can race against SCSI timeout. When normal
635 * completion wins, the qc never reaches EH. When error
636 * completion wins, the qc has ATA_QCFLAG_FAILED set.
637 *
638 * When SCSI timeout wins, things are a bit more complex.
639 * Normal or error completion can occur after the timeout but
640 * before this point. In such cases, both types of
641 * completions are honored. A scmd is determined to have
642 * timed out iff its associated qc is active and not failed.
643 */
644 if (ap->ops->error_handler) {
645 struct scsi_cmnd *scmd, *tmp;
646 int nr_timedout = 0;
647
648 spin_lock_irqsave(ap->lock, flags);
649
650 /* This must occur under the ap->lock as we don't want
651 a polled recovery to race the real interrupt handler
652
653 The lost_interrupt handler checks for any completed but
654 non-notified command and completes much like an IRQ handler.
655
656 We then fall into the error recovery code which will treat
657 this as if normal completion won the race */
658
659 if (ap->ops->lost_interrupt)
660 ap->ops->lost_interrupt(ap);
661
662 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
663 struct ata_queued_cmd *qc;
664
665 for (i = 0; i < ATA_MAX_QUEUE; i++) {
666 qc = __ata_qc_from_tag(ap, i);
667 if (qc->flags & ATA_QCFLAG_ACTIVE &&
668 qc->scsicmd == scmd)
669 break;
670 }
671
672 if (i < ATA_MAX_QUEUE) {
673 /* the scmd has an associated qc */
674 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
675 /* which hasn't failed yet, timeout */
676 qc->err_mask |= AC_ERR_TIMEOUT;
677 qc->flags |= ATA_QCFLAG_FAILED;
678 nr_timedout++;
679 }
680 } else {
681 /* Normal completion occurred after
682 * SCSI timeout but before this point.
683 * Successfully complete it.
684 */
685 scmd->retries = scmd->allowed;
686 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
687 }
688 }
689
690 /* If we have timed out qcs. They belong to EH from
691 * this point but the state of the controller is
692 * unknown. Freeze the port to make sure the IRQ
693 * handler doesn't diddle with those qcs. This must
694 * be done atomically w.r.t. setting QCFLAG_FAILED.
695 */
696 if (nr_timedout)
697 __ata_port_freeze(ap);
698
699 spin_unlock_irqrestore(ap->lock, flags);
700
701 /* initialize eh_tries */
702 ap->eh_tries = ATA_EH_MAX_TRIES;
703 } else
704 spin_unlock_wait(ap->lock);
705
706 }
707 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
708
709 /**
710 * ata_scsi_port_error_handler - recover the port after the commands
711 * @host: SCSI host containing the port
712 * @ap: the ATA port
713 *
714 * Handle the recovery of the port @ap after all the commands
715 * have been recovered.
716 */
ata_scsi_port_error_handler(struct Scsi_Host * host,struct ata_port * ap)717 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
718 {
719 unsigned long flags;
720
721 /* invoke error handler */
722 if (ap->ops->error_handler) {
723 struct ata_link *link;
724
725 /* acquire EH ownership */
726 ata_eh_acquire(ap);
727 repeat:
728 /* kill fast drain timer */
729 del_timer_sync(&ap->fastdrain_timer);
730
731 /* process port resume request */
732 ata_eh_handle_port_resume(ap);
733
734 /* fetch & clear EH info */
735 spin_lock_irqsave(ap->lock, flags);
736
737 ata_for_each_link(link, ap, HOST_FIRST) {
738 struct ata_eh_context *ehc = &link->eh_context;
739 struct ata_device *dev;
740
741 memset(&link->eh_context, 0, sizeof(link->eh_context));
742 link->eh_context.i = link->eh_info;
743 memset(&link->eh_info, 0, sizeof(link->eh_info));
744
745 ata_for_each_dev(dev, link, ENABLED) {
746 int devno = dev->devno;
747
748 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
749 if (ata_ncq_enabled(dev))
750 ehc->saved_ncq_enabled |= 1 << devno;
751 }
752 }
753
754 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
755 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
756 ap->excl_link = NULL; /* don't maintain exclusion over EH */
757
758 spin_unlock_irqrestore(ap->lock, flags);
759
760 /* invoke EH, skip if unloading or suspended */
761 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
762 ap->ops->error_handler(ap);
763 else {
764 /* if unloading, commence suicide */
765 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
766 !(ap->pflags & ATA_PFLAG_UNLOADED))
767 ata_eh_unload(ap);
768 ata_eh_finish(ap);
769 }
770
771 /* process port suspend request */
772 ata_eh_handle_port_suspend(ap);
773
774 /* Exception might have happened after ->error_handler
775 * recovered the port but before this point. Repeat
776 * EH in such case.
777 */
778 spin_lock_irqsave(ap->lock, flags);
779
780 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
781 if (--ap->eh_tries) {
782 spin_unlock_irqrestore(ap->lock, flags);
783 goto repeat;
784 }
785 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
786 "tries, giving up\n", ATA_EH_MAX_TRIES);
787 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
788 }
789
790 /* this run is complete, make sure EH info is clear */
791 ata_for_each_link(link, ap, HOST_FIRST)
792 memset(&link->eh_info, 0, sizeof(link->eh_info));
793
794 /* Clear host_eh_scheduled while holding ap->lock such
795 * that if exception occurs after this point but
796 * before EH completion, SCSI midlayer will
797 * re-initiate EH.
798 */
799 host->host_eh_scheduled = 0;
800
801 spin_unlock_irqrestore(ap->lock, flags);
802 ata_eh_release(ap);
803 } else {
804 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
805 ap->ops->eng_timeout(ap);
806 }
807
808 scsi_eh_flush_done_q(&ap->eh_done_q);
809
810 /* clean up */
811 spin_lock_irqsave(ap->lock, flags);
812
813 if (ap->pflags & ATA_PFLAG_LOADING)
814 ap->pflags &= ~ATA_PFLAG_LOADING;
815 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
816 schedule_delayed_work(&ap->hotplug_task, 0);
817
818 if (ap->pflags & ATA_PFLAG_RECOVERED)
819 ata_port_printk(ap, KERN_INFO, "EH complete\n");
820
821 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
822
823 /* tell wait_eh that we're done */
824 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
825 wake_up_all(&ap->eh_wait_q);
826
827 spin_unlock_irqrestore(ap->lock, flags);
828 }
829 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
830
831 /**
832 * ata_port_wait_eh - Wait for the currently pending EH to complete
833 * @ap: Port to wait EH for
834 *
835 * Wait until the currently pending EH is complete.
836 *
837 * LOCKING:
838 * Kernel thread context (may sleep).
839 */
ata_port_wait_eh(struct ata_port * ap)840 void ata_port_wait_eh(struct ata_port *ap)
841 {
842 unsigned long flags;
843 DEFINE_WAIT(wait);
844
845 retry:
846 spin_lock_irqsave(ap->lock, flags);
847
848 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
849 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
850 spin_unlock_irqrestore(ap->lock, flags);
851 schedule();
852 spin_lock_irqsave(ap->lock, flags);
853 }
854 finish_wait(&ap->eh_wait_q, &wait);
855
856 spin_unlock_irqrestore(ap->lock, flags);
857
858 /* make sure SCSI EH is complete */
859 if (scsi_host_in_recovery(ap->scsi_host)) {
860 ata_msleep(ap, 10);
861 goto retry;
862 }
863 }
864
ata_eh_nr_in_flight(struct ata_port * ap)865 static int ata_eh_nr_in_flight(struct ata_port *ap)
866 {
867 unsigned int tag;
868 int nr = 0;
869
870 /* count only non-internal commands */
871 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
872 if (ata_qc_from_tag(ap, tag))
873 nr++;
874
875 return nr;
876 }
877
ata_eh_fastdrain_timerfn(unsigned long arg)878 void ata_eh_fastdrain_timerfn(unsigned long arg)
879 {
880 struct ata_port *ap = (void *)arg;
881 unsigned long flags;
882 int cnt;
883
884 spin_lock_irqsave(ap->lock, flags);
885
886 cnt = ata_eh_nr_in_flight(ap);
887
888 /* are we done? */
889 if (!cnt)
890 goto out_unlock;
891
892 if (cnt == ap->fastdrain_cnt) {
893 unsigned int tag;
894
895 /* No progress during the last interval, tag all
896 * in-flight qcs as timed out and freeze the port.
897 */
898 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
899 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
900 if (qc)
901 qc->err_mask |= AC_ERR_TIMEOUT;
902 }
903
904 ata_port_freeze(ap);
905 } else {
906 /* some qcs have finished, give it another chance */
907 ap->fastdrain_cnt = cnt;
908 ap->fastdrain_timer.expires =
909 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
910 add_timer(&ap->fastdrain_timer);
911 }
912
913 out_unlock:
914 spin_unlock_irqrestore(ap->lock, flags);
915 }
916
917 /**
918 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
919 * @ap: target ATA port
920 * @fastdrain: activate fast drain
921 *
922 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
923 * is non-zero and EH wasn't pending before. Fast drain ensures
924 * that EH kicks in in timely manner.
925 *
926 * LOCKING:
927 * spin_lock_irqsave(host lock)
928 */
ata_eh_set_pending(struct ata_port * ap,int fastdrain)929 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
930 {
931 int cnt;
932
933 /* already scheduled? */
934 if (ap->pflags & ATA_PFLAG_EH_PENDING)
935 return;
936
937 ap->pflags |= ATA_PFLAG_EH_PENDING;
938
939 if (!fastdrain)
940 return;
941
942 /* do we have in-flight qcs? */
943 cnt = ata_eh_nr_in_flight(ap);
944 if (!cnt)
945 return;
946
947 /* activate fast drain */
948 ap->fastdrain_cnt = cnt;
949 ap->fastdrain_timer.expires =
950 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
951 add_timer(&ap->fastdrain_timer);
952 }
953
954 /**
955 * ata_qc_schedule_eh - schedule qc for error handling
956 * @qc: command to schedule error handling for
957 *
958 * Schedule error handling for @qc. EH will kick in as soon as
959 * other commands are drained.
960 *
961 * LOCKING:
962 * spin_lock_irqsave(host lock)
963 */
ata_qc_schedule_eh(struct ata_queued_cmd * qc)964 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
965 {
966 struct ata_port *ap = qc->ap;
967 struct request_queue *q = qc->scsicmd->device->request_queue;
968 unsigned long flags;
969
970 WARN_ON(!ap->ops->error_handler);
971
972 qc->flags |= ATA_QCFLAG_FAILED;
973 ata_eh_set_pending(ap, 1);
974
975 /* The following will fail if timeout has already expired.
976 * ata_scsi_error() takes care of such scmds on EH entry.
977 * Note that ATA_QCFLAG_FAILED is unconditionally set after
978 * this function completes.
979 */
980 spin_lock_irqsave(q->queue_lock, flags);
981 blk_abort_request(qc->scsicmd->request);
982 spin_unlock_irqrestore(q->queue_lock, flags);
983 }
984
985 /**
986 * ata_port_schedule_eh - schedule error handling without a qc
987 * @ap: ATA port to schedule EH for
988 *
989 * Schedule error handling for @ap. EH will kick in as soon as
990 * all commands are drained.
991 *
992 * LOCKING:
993 * spin_lock_irqsave(host lock)
994 */
ata_port_schedule_eh(struct ata_port * ap)995 void ata_port_schedule_eh(struct ata_port *ap)
996 {
997 WARN_ON(!ap->ops->error_handler);
998
999 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1000 return;
1001
1002 ata_eh_set_pending(ap, 1);
1003 scsi_schedule_eh(ap->scsi_host);
1004
1005 DPRINTK("port EH scheduled\n");
1006 }
1007
ata_do_link_abort(struct ata_port * ap,struct ata_link * link)1008 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1009 {
1010 int tag, nr_aborted = 0;
1011
1012 WARN_ON(!ap->ops->error_handler);
1013
1014 /* we're gonna abort all commands, no need for fast drain */
1015 ata_eh_set_pending(ap, 0);
1016
1017 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1018 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1019
1020 if (qc && (!link || qc->dev->link == link)) {
1021 qc->flags |= ATA_QCFLAG_FAILED;
1022 ata_qc_complete(qc);
1023 nr_aborted++;
1024 }
1025 }
1026
1027 if (!nr_aborted)
1028 ata_port_schedule_eh(ap);
1029
1030 return nr_aborted;
1031 }
1032
1033 /**
1034 * ata_link_abort - abort all qc's on the link
1035 * @link: ATA link to abort qc's for
1036 *
1037 * Abort all active qc's active on @link and schedule EH.
1038 *
1039 * LOCKING:
1040 * spin_lock_irqsave(host lock)
1041 *
1042 * RETURNS:
1043 * Number of aborted qc's.
1044 */
ata_link_abort(struct ata_link * link)1045 int ata_link_abort(struct ata_link *link)
1046 {
1047 return ata_do_link_abort(link->ap, link);
1048 }
1049
1050 /**
1051 * ata_port_abort - abort all qc's on the port
1052 * @ap: ATA port to abort qc's for
1053 *
1054 * Abort all active qc's of @ap and schedule EH.
1055 *
1056 * LOCKING:
1057 * spin_lock_irqsave(host_set lock)
1058 *
1059 * RETURNS:
1060 * Number of aborted qc's.
1061 */
ata_port_abort(struct ata_port * ap)1062 int ata_port_abort(struct ata_port *ap)
1063 {
1064 return ata_do_link_abort(ap, NULL);
1065 }
1066
1067 /**
1068 * __ata_port_freeze - freeze port
1069 * @ap: ATA port to freeze
1070 *
1071 * This function is called when HSM violation or some other
1072 * condition disrupts normal operation of the port. Frozen port
1073 * is not allowed to perform any operation until the port is
1074 * thawed, which usually follows a successful reset.
1075 *
1076 * ap->ops->freeze() callback can be used for freezing the port
1077 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1078 * port cannot be frozen hardware-wise, the interrupt handler
1079 * must ack and clear interrupts unconditionally while the port
1080 * is frozen.
1081 *
1082 * LOCKING:
1083 * spin_lock_irqsave(host lock)
1084 */
__ata_port_freeze(struct ata_port * ap)1085 static void __ata_port_freeze(struct ata_port *ap)
1086 {
1087 WARN_ON(!ap->ops->error_handler);
1088
1089 if (ap->ops->freeze)
1090 ap->ops->freeze(ap);
1091
1092 ap->pflags |= ATA_PFLAG_FROZEN;
1093
1094 DPRINTK("ata%u port frozen\n", ap->print_id);
1095 }
1096
1097 /**
1098 * ata_port_freeze - abort & freeze port
1099 * @ap: ATA port to freeze
1100 *
1101 * Abort and freeze @ap. The freeze operation must be called
1102 * first, because some hardware requires special operations
1103 * before the taskfile registers are accessible.
1104 *
1105 * LOCKING:
1106 * spin_lock_irqsave(host lock)
1107 *
1108 * RETURNS:
1109 * Number of aborted commands.
1110 */
ata_port_freeze(struct ata_port * ap)1111 int ata_port_freeze(struct ata_port *ap)
1112 {
1113 int nr_aborted;
1114
1115 WARN_ON(!ap->ops->error_handler);
1116
1117 __ata_port_freeze(ap);
1118 nr_aborted = ata_port_abort(ap);
1119
1120 return nr_aborted;
1121 }
1122
1123 /**
1124 * sata_async_notification - SATA async notification handler
1125 * @ap: ATA port where async notification is received
1126 *
1127 * Handler to be called when async notification via SDB FIS is
1128 * received. This function schedules EH if necessary.
1129 *
1130 * LOCKING:
1131 * spin_lock_irqsave(host lock)
1132 *
1133 * RETURNS:
1134 * 1 if EH is scheduled, 0 otherwise.
1135 */
sata_async_notification(struct ata_port * ap)1136 int sata_async_notification(struct ata_port *ap)
1137 {
1138 u32 sntf;
1139 int rc;
1140
1141 if (!(ap->flags & ATA_FLAG_AN))
1142 return 0;
1143
1144 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1145 if (rc == 0)
1146 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1147
1148 if (!sata_pmp_attached(ap) || rc) {
1149 /* PMP is not attached or SNTF is not available */
1150 if (!sata_pmp_attached(ap)) {
1151 /* PMP is not attached. Check whether ATAPI
1152 * AN is configured. If so, notify media
1153 * change.
1154 */
1155 struct ata_device *dev = ap->link.device;
1156
1157 if ((dev->class == ATA_DEV_ATAPI) &&
1158 (dev->flags & ATA_DFLAG_AN))
1159 ata_scsi_media_change_notify(dev);
1160 return 0;
1161 } else {
1162 /* PMP is attached but SNTF is not available.
1163 * ATAPI async media change notification is
1164 * not used. The PMP must be reporting PHY
1165 * status change, schedule EH.
1166 */
1167 ata_port_schedule_eh(ap);
1168 return 1;
1169 }
1170 } else {
1171 /* PMP is attached and SNTF is available */
1172 struct ata_link *link;
1173
1174 /* check and notify ATAPI AN */
1175 ata_for_each_link(link, ap, EDGE) {
1176 if (!(sntf & (1 << link->pmp)))
1177 continue;
1178
1179 if ((link->device->class == ATA_DEV_ATAPI) &&
1180 (link->device->flags & ATA_DFLAG_AN))
1181 ata_scsi_media_change_notify(link->device);
1182 }
1183
1184 /* If PMP is reporting that PHY status of some
1185 * downstream ports has changed, schedule EH.
1186 */
1187 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1188 ata_port_schedule_eh(ap);
1189 return 1;
1190 }
1191
1192 return 0;
1193 }
1194 }
1195
1196 /**
1197 * ata_eh_freeze_port - EH helper to freeze port
1198 * @ap: ATA port to freeze
1199 *
1200 * Freeze @ap.
1201 *
1202 * LOCKING:
1203 * None.
1204 */
ata_eh_freeze_port(struct ata_port * ap)1205 void ata_eh_freeze_port(struct ata_port *ap)
1206 {
1207 unsigned long flags;
1208
1209 if (!ap->ops->error_handler)
1210 return;
1211
1212 spin_lock_irqsave(ap->lock, flags);
1213 __ata_port_freeze(ap);
1214 spin_unlock_irqrestore(ap->lock, flags);
1215 }
1216
1217 /**
1218 * ata_port_thaw_port - EH helper to thaw port
1219 * @ap: ATA port to thaw
1220 *
1221 * Thaw frozen port @ap.
1222 *
1223 * LOCKING:
1224 * None.
1225 */
ata_eh_thaw_port(struct ata_port * ap)1226 void ata_eh_thaw_port(struct ata_port *ap)
1227 {
1228 unsigned long flags;
1229
1230 if (!ap->ops->error_handler)
1231 return;
1232
1233 spin_lock_irqsave(ap->lock, flags);
1234
1235 ap->pflags &= ~ATA_PFLAG_FROZEN;
1236
1237 if (ap->ops->thaw)
1238 ap->ops->thaw(ap);
1239
1240 spin_unlock_irqrestore(ap->lock, flags);
1241
1242 DPRINTK("ata%u port thawed\n", ap->print_id);
1243 }
1244
ata_eh_scsidone(struct scsi_cmnd * scmd)1245 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1246 {
1247 /* nada */
1248 }
1249
__ata_eh_qc_complete(struct ata_queued_cmd * qc)1250 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1251 {
1252 struct ata_port *ap = qc->ap;
1253 struct scsi_cmnd *scmd = qc->scsicmd;
1254 unsigned long flags;
1255
1256 spin_lock_irqsave(ap->lock, flags);
1257 qc->scsidone = ata_eh_scsidone;
1258 __ata_qc_complete(qc);
1259 WARN_ON(ata_tag_valid(qc->tag));
1260 spin_unlock_irqrestore(ap->lock, flags);
1261
1262 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1263 }
1264
1265 /**
1266 * ata_eh_qc_complete - Complete an active ATA command from EH
1267 * @qc: Command to complete
1268 *
1269 * Indicate to the mid and upper layers that an ATA command has
1270 * completed. To be used from EH.
1271 */
ata_eh_qc_complete(struct ata_queued_cmd * qc)1272 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1273 {
1274 struct scsi_cmnd *scmd = qc->scsicmd;
1275 scmd->retries = scmd->allowed;
1276 __ata_eh_qc_complete(qc);
1277 }
1278
1279 /**
1280 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1281 * @qc: Command to retry
1282 *
1283 * Indicate to the mid and upper layers that an ATA command
1284 * should be retried. To be used from EH.
1285 *
1286 * SCSI midlayer limits the number of retries to scmd->allowed.
1287 * scmd->retries is decremented for commands which get retried
1288 * due to unrelated failures (qc->err_mask is zero).
1289 */
ata_eh_qc_retry(struct ata_queued_cmd * qc)1290 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1291 {
1292 struct scsi_cmnd *scmd = qc->scsicmd;
1293 if (!qc->err_mask && scmd->retries)
1294 scmd->retries--;
1295 __ata_eh_qc_complete(qc);
1296 }
1297
1298 /**
1299 * ata_dev_disable - disable ATA device
1300 * @dev: ATA device to disable
1301 *
1302 * Disable @dev.
1303 *
1304 * Locking:
1305 * EH context.
1306 */
ata_dev_disable(struct ata_device * dev)1307 void ata_dev_disable(struct ata_device *dev)
1308 {
1309 if (!ata_dev_enabled(dev))
1310 return;
1311
1312 if (ata_msg_drv(dev->link->ap))
1313 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1314 ata_acpi_on_disable(dev);
1315 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1316 dev->class++;
1317
1318 /* From now till the next successful probe, ering is used to
1319 * track probe failures. Clear accumulated device error info.
1320 */
1321 ata_ering_clear(&dev->ering);
1322 }
1323
1324 /**
1325 * ata_eh_detach_dev - detach ATA device
1326 * @dev: ATA device to detach
1327 *
1328 * Detach @dev.
1329 *
1330 * LOCKING:
1331 * None.
1332 */
ata_eh_detach_dev(struct ata_device * dev)1333 void ata_eh_detach_dev(struct ata_device *dev)
1334 {
1335 struct ata_link *link = dev->link;
1336 struct ata_port *ap = link->ap;
1337 struct ata_eh_context *ehc = &link->eh_context;
1338 unsigned long flags;
1339
1340 ata_dev_disable(dev);
1341
1342 spin_lock_irqsave(ap->lock, flags);
1343
1344 dev->flags &= ~ATA_DFLAG_DETACH;
1345
1346 if (ata_scsi_offline_dev(dev)) {
1347 dev->flags |= ATA_DFLAG_DETACHED;
1348 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1349 }
1350
1351 /* clear per-dev EH info */
1352 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1353 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1354 ehc->saved_xfer_mode[dev->devno] = 0;
1355 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1356
1357 spin_unlock_irqrestore(ap->lock, flags);
1358 }
1359
1360 /**
1361 * ata_eh_about_to_do - about to perform eh_action
1362 * @link: target ATA link
1363 * @dev: target ATA dev for per-dev action (can be NULL)
1364 * @action: action about to be performed
1365 *
1366 * Called just before performing EH actions to clear related bits
1367 * in @link->eh_info such that eh actions are not unnecessarily
1368 * repeated.
1369 *
1370 * LOCKING:
1371 * None.
1372 */
ata_eh_about_to_do(struct ata_link * link,struct ata_device * dev,unsigned int action)1373 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1374 unsigned int action)
1375 {
1376 struct ata_port *ap = link->ap;
1377 struct ata_eh_info *ehi = &link->eh_info;
1378 struct ata_eh_context *ehc = &link->eh_context;
1379 unsigned long flags;
1380
1381 spin_lock_irqsave(ap->lock, flags);
1382
1383 ata_eh_clear_action(link, dev, ehi, action);
1384
1385 /* About to take EH action, set RECOVERED. Ignore actions on
1386 * slave links as master will do them again.
1387 */
1388 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1389 ap->pflags |= ATA_PFLAG_RECOVERED;
1390
1391 spin_unlock_irqrestore(ap->lock, flags);
1392 }
1393
1394 /**
1395 * ata_eh_done - EH action complete
1396 * @ap: target ATA port
1397 * @dev: target ATA dev for per-dev action (can be NULL)
1398 * @action: action just completed
1399 *
1400 * Called right after performing EH actions to clear related bits
1401 * in @link->eh_context.
1402 *
1403 * LOCKING:
1404 * None.
1405 */
ata_eh_done(struct ata_link * link,struct ata_device * dev,unsigned int action)1406 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1407 unsigned int action)
1408 {
1409 struct ata_eh_context *ehc = &link->eh_context;
1410
1411 ata_eh_clear_action(link, dev, &ehc->i, action);
1412 }
1413
1414 /**
1415 * ata_err_string - convert err_mask to descriptive string
1416 * @err_mask: error mask to convert to string
1417 *
1418 * Convert @err_mask to descriptive string. Errors are
1419 * prioritized according to severity and only the most severe
1420 * error is reported.
1421 *
1422 * LOCKING:
1423 * None.
1424 *
1425 * RETURNS:
1426 * Descriptive string for @err_mask
1427 */
ata_err_string(unsigned int err_mask)1428 static const char *ata_err_string(unsigned int err_mask)
1429 {
1430 if (err_mask & AC_ERR_HOST_BUS)
1431 return "host bus error";
1432 if (err_mask & AC_ERR_ATA_BUS)
1433 return "ATA bus error";
1434 if (err_mask & AC_ERR_TIMEOUT)
1435 return "timeout";
1436 if (err_mask & AC_ERR_HSM)
1437 return "HSM violation";
1438 if (err_mask & AC_ERR_SYSTEM)
1439 return "internal error";
1440 if (err_mask & AC_ERR_MEDIA)
1441 return "media error";
1442 if (err_mask & AC_ERR_INVALID)
1443 return "invalid argument";
1444 if (err_mask & AC_ERR_DEV)
1445 return "device error";
1446 return "unknown error";
1447 }
1448
1449 /**
1450 * ata_read_log_page - read a specific log page
1451 * @dev: target device
1452 * @page: page to read
1453 * @buf: buffer to store read page
1454 * @sectors: number of sectors to read
1455 *
1456 * Read log page using READ_LOG_EXT command.
1457 *
1458 * LOCKING:
1459 * Kernel thread context (may sleep).
1460 *
1461 * RETURNS:
1462 * 0 on success, AC_ERR_* mask otherwise.
1463 */
ata_read_log_page(struct ata_device * dev,u8 page,void * buf,unsigned int sectors)1464 static unsigned int ata_read_log_page(struct ata_device *dev,
1465 u8 page, void *buf, unsigned int sectors)
1466 {
1467 struct ata_taskfile tf;
1468 unsigned int err_mask;
1469
1470 DPRINTK("read log page - page %d\n", page);
1471
1472 ata_tf_init(dev, &tf);
1473 tf.command = ATA_CMD_READ_LOG_EXT;
1474 tf.lbal = page;
1475 tf.nsect = sectors;
1476 tf.hob_nsect = sectors >> 8;
1477 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1478 tf.protocol = ATA_PROT_PIO;
1479
1480 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1481 buf, sectors * ATA_SECT_SIZE, 0);
1482
1483 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1484 return err_mask;
1485 }
1486
1487 /**
1488 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1489 * @dev: Device to read log page 10h from
1490 * @tag: Resulting tag of the failed command
1491 * @tf: Resulting taskfile registers of the failed command
1492 *
1493 * Read log page 10h to obtain NCQ error details and clear error
1494 * condition.
1495 *
1496 * LOCKING:
1497 * Kernel thread context (may sleep).
1498 *
1499 * RETURNS:
1500 * 0 on success, -errno otherwise.
1501 */
ata_eh_read_log_10h(struct ata_device * dev,int * tag,struct ata_taskfile * tf)1502 static int ata_eh_read_log_10h(struct ata_device *dev,
1503 int *tag, struct ata_taskfile *tf)
1504 {
1505 u8 *buf = dev->link->ap->sector_buf;
1506 unsigned int err_mask;
1507 u8 csum;
1508 int i;
1509
1510 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1511 if (err_mask)
1512 return -EIO;
1513
1514 csum = 0;
1515 for (i = 0; i < ATA_SECT_SIZE; i++)
1516 csum += buf[i];
1517 if (csum)
1518 ata_dev_printk(dev, KERN_WARNING,
1519 "invalid checksum 0x%x on log page 10h\n", csum);
1520
1521 if (buf[0] & 0x80)
1522 return -ENOENT;
1523
1524 *tag = buf[0] & 0x1f;
1525
1526 tf->command = buf[2];
1527 tf->feature = buf[3];
1528 tf->lbal = buf[4];
1529 tf->lbam = buf[5];
1530 tf->lbah = buf[6];
1531 tf->device = buf[7];
1532 tf->hob_lbal = buf[8];
1533 tf->hob_lbam = buf[9];
1534 tf->hob_lbah = buf[10];
1535 tf->nsect = buf[12];
1536 tf->hob_nsect = buf[13];
1537
1538 return 0;
1539 }
1540
1541 /**
1542 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1543 * @dev: target ATAPI device
1544 * @r_sense_key: out parameter for sense_key
1545 *
1546 * Perform ATAPI TEST_UNIT_READY.
1547 *
1548 * LOCKING:
1549 * EH context (may sleep).
1550 *
1551 * RETURNS:
1552 * 0 on success, AC_ERR_* mask on failure.
1553 */
atapi_eh_tur(struct ata_device * dev,u8 * r_sense_key)1554 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1555 {
1556 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1557 struct ata_taskfile tf;
1558 unsigned int err_mask;
1559
1560 ata_tf_init(dev, &tf);
1561
1562 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1563 tf.command = ATA_CMD_PACKET;
1564 tf.protocol = ATAPI_PROT_NODATA;
1565
1566 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1567 if (err_mask == AC_ERR_DEV)
1568 *r_sense_key = tf.feature >> 4;
1569 return err_mask;
1570 }
1571
1572 /**
1573 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1574 * @dev: device to perform REQUEST_SENSE to
1575 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1576 * @dfl_sense_key: default sense key to use
1577 *
1578 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1579 * SENSE. This function is EH helper.
1580 *
1581 * LOCKING:
1582 * Kernel thread context (may sleep).
1583 *
1584 * RETURNS:
1585 * 0 on success, AC_ERR_* mask on failure
1586 */
atapi_eh_request_sense(struct ata_device * dev,u8 * sense_buf,u8 dfl_sense_key)1587 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1588 u8 *sense_buf, u8 dfl_sense_key)
1589 {
1590 u8 cdb[ATAPI_CDB_LEN] =
1591 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1592 struct ata_port *ap = dev->link->ap;
1593 struct ata_taskfile tf;
1594
1595 DPRINTK("ATAPI request sense\n");
1596
1597 /* FIXME: is this needed? */
1598 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1599
1600 /* initialize sense_buf with the error register,
1601 * for the case where they are -not- overwritten
1602 */
1603 sense_buf[0] = 0x70;
1604 sense_buf[2] = dfl_sense_key;
1605
1606 /* some devices time out if garbage left in tf */
1607 ata_tf_init(dev, &tf);
1608
1609 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1610 tf.command = ATA_CMD_PACKET;
1611
1612 /* is it pointless to prefer PIO for "safety reasons"? */
1613 if (ap->flags & ATA_FLAG_PIO_DMA) {
1614 tf.protocol = ATAPI_PROT_DMA;
1615 tf.feature |= ATAPI_PKT_DMA;
1616 } else {
1617 tf.protocol = ATAPI_PROT_PIO;
1618 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1619 tf.lbah = 0;
1620 }
1621
1622 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1623 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1624 }
1625
1626 /**
1627 * ata_eh_analyze_serror - analyze SError for a failed port
1628 * @link: ATA link to analyze SError for
1629 *
1630 * Analyze SError if available and further determine cause of
1631 * failure.
1632 *
1633 * LOCKING:
1634 * None.
1635 */
ata_eh_analyze_serror(struct ata_link * link)1636 static void ata_eh_analyze_serror(struct ata_link *link)
1637 {
1638 struct ata_eh_context *ehc = &link->eh_context;
1639 u32 serror = ehc->i.serror;
1640 unsigned int err_mask = 0, action = 0;
1641 u32 hotplug_mask;
1642
1643 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1644 err_mask |= AC_ERR_ATA_BUS;
1645 action |= ATA_EH_RESET;
1646 }
1647 if (serror & SERR_PROTOCOL) {
1648 err_mask |= AC_ERR_HSM;
1649 action |= ATA_EH_RESET;
1650 }
1651 if (serror & SERR_INTERNAL) {
1652 err_mask |= AC_ERR_SYSTEM;
1653 action |= ATA_EH_RESET;
1654 }
1655
1656 /* Determine whether a hotplug event has occurred. Both
1657 * SError.N/X are considered hotplug events for enabled or
1658 * host links. For disabled PMP links, only N bit is
1659 * considered as X bit is left at 1 for link plugging.
1660 */
1661 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1662 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1663 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1664 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1665 else
1666 hotplug_mask = SERR_PHYRDY_CHG;
1667
1668 if (serror & hotplug_mask)
1669 ata_ehi_hotplugged(&ehc->i);
1670
1671 ehc->i.err_mask |= err_mask;
1672 ehc->i.action |= action;
1673 }
1674
1675 /**
1676 * ata_eh_analyze_ncq_error - analyze NCQ error
1677 * @link: ATA link to analyze NCQ error for
1678 *
1679 * Read log page 10h, determine the offending qc and acquire
1680 * error status TF. For NCQ device errors, all LLDDs have to do
1681 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1682 * care of the rest.
1683 *
1684 * LOCKING:
1685 * Kernel thread context (may sleep).
1686 */
ata_eh_analyze_ncq_error(struct ata_link * link)1687 void ata_eh_analyze_ncq_error(struct ata_link *link)
1688 {
1689 struct ata_port *ap = link->ap;
1690 struct ata_eh_context *ehc = &link->eh_context;
1691 struct ata_device *dev = link->device;
1692 struct ata_queued_cmd *qc;
1693 struct ata_taskfile tf;
1694 int tag, rc;
1695
1696 /* if frozen, we can't do much */
1697 if (ap->pflags & ATA_PFLAG_FROZEN)
1698 return;
1699
1700 /* is it NCQ device error? */
1701 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1702 return;
1703
1704 /* has LLDD analyzed already? */
1705 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1706 qc = __ata_qc_from_tag(ap, tag);
1707
1708 if (!(qc->flags & ATA_QCFLAG_FAILED))
1709 continue;
1710
1711 if (qc->err_mask)
1712 return;
1713 }
1714
1715 /* okay, this error is ours */
1716 memset(&tf, 0, sizeof(tf));
1717 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1718 if (rc) {
1719 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1720 "(errno=%d)\n", rc);
1721 return;
1722 }
1723
1724 if (!(link->sactive & (1 << tag))) {
1725 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1726 "inactive tag %d\n", tag);
1727 return;
1728 }
1729
1730 /* we've got the perpetrator, condemn it */
1731 qc = __ata_qc_from_tag(ap, tag);
1732 memcpy(&qc->result_tf, &tf, sizeof(tf));
1733 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1734 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1735 ehc->i.err_mask &= ~AC_ERR_DEV;
1736 }
1737
1738 /**
1739 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1740 * @qc: qc to analyze
1741 * @tf: Taskfile registers to analyze
1742 *
1743 * Analyze taskfile of @qc and further determine cause of
1744 * failure. This function also requests ATAPI sense data if
1745 * available.
1746 *
1747 * LOCKING:
1748 * Kernel thread context (may sleep).
1749 *
1750 * RETURNS:
1751 * Determined recovery action
1752 */
ata_eh_analyze_tf(struct ata_queued_cmd * qc,const struct ata_taskfile * tf)1753 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1754 const struct ata_taskfile *tf)
1755 {
1756 unsigned int tmp, action = 0;
1757 u8 stat = tf->command, err = tf->feature;
1758
1759 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1760 qc->err_mask |= AC_ERR_HSM;
1761 return ATA_EH_RESET;
1762 }
1763
1764 if (stat & (ATA_ERR | ATA_DF))
1765 qc->err_mask |= AC_ERR_DEV;
1766 else
1767 return 0;
1768
1769 switch (qc->dev->class) {
1770 case ATA_DEV_ATA:
1771 if (err & ATA_ICRC)
1772 qc->err_mask |= AC_ERR_ATA_BUS;
1773 if (err & ATA_UNC)
1774 qc->err_mask |= AC_ERR_MEDIA;
1775 if (err & ATA_IDNF)
1776 qc->err_mask |= AC_ERR_INVALID;
1777 break;
1778
1779 case ATA_DEV_ATAPI:
1780 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1781 tmp = atapi_eh_request_sense(qc->dev,
1782 qc->scsicmd->sense_buffer,
1783 qc->result_tf.feature >> 4);
1784 if (!tmp) {
1785 /* ATA_QCFLAG_SENSE_VALID is used to
1786 * tell atapi_qc_complete() that sense
1787 * data is already valid.
1788 *
1789 * TODO: interpret sense data and set
1790 * appropriate err_mask.
1791 */
1792 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1793 } else
1794 qc->err_mask |= tmp;
1795 }
1796 }
1797
1798 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1799 action |= ATA_EH_RESET;
1800
1801 return action;
1802 }
1803
ata_eh_categorize_error(unsigned int eflags,unsigned int err_mask,int * xfer_ok)1804 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1805 int *xfer_ok)
1806 {
1807 int base = 0;
1808
1809 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1810 *xfer_ok = 1;
1811
1812 if (!*xfer_ok)
1813 base = ATA_ECAT_DUBIOUS_NONE;
1814
1815 if (err_mask & AC_ERR_ATA_BUS)
1816 return base + ATA_ECAT_ATA_BUS;
1817
1818 if (err_mask & AC_ERR_TIMEOUT)
1819 return base + ATA_ECAT_TOUT_HSM;
1820
1821 if (eflags & ATA_EFLAG_IS_IO) {
1822 if (err_mask & AC_ERR_HSM)
1823 return base + ATA_ECAT_TOUT_HSM;
1824 if ((err_mask &
1825 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1826 return base + ATA_ECAT_UNK_DEV;
1827 }
1828
1829 return 0;
1830 }
1831
1832 struct speed_down_verdict_arg {
1833 u64 since;
1834 int xfer_ok;
1835 int nr_errors[ATA_ECAT_NR];
1836 };
1837
speed_down_verdict_cb(struct ata_ering_entry * ent,void * void_arg)1838 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1839 {
1840 struct speed_down_verdict_arg *arg = void_arg;
1841 int cat;
1842
1843 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1844 return -1;
1845
1846 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1847 &arg->xfer_ok);
1848 arg->nr_errors[cat]++;
1849
1850 return 0;
1851 }
1852
1853 /**
1854 * ata_eh_speed_down_verdict - Determine speed down verdict
1855 * @dev: Device of interest
1856 *
1857 * This function examines error ring of @dev and determines
1858 * whether NCQ needs to be turned off, transfer speed should be
1859 * stepped down, or falling back to PIO is necessary.
1860 *
1861 * ECAT_ATA_BUS : ATA_BUS error for any command
1862 *
1863 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1864 * IO commands
1865 *
1866 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1867 *
1868 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1869 * data transfer hasn't been verified.
1870 *
1871 * Verdicts are
1872 *
1873 * NCQ_OFF : Turn off NCQ.
1874 *
1875 * SPEED_DOWN : Speed down transfer speed but don't fall back
1876 * to PIO.
1877 *
1878 * FALLBACK_TO_PIO : Fall back to PIO.
1879 *
1880 * Even if multiple verdicts are returned, only one action is
1881 * taken per error. An action triggered by non-DUBIOUS errors
1882 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1883 * This is to expedite speed down decisions right after device is
1884 * initially configured.
1885 *
1886 * The followings are speed down rules. #1 and #2 deal with
1887 * DUBIOUS errors.
1888 *
1889 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1890 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1891 *
1892 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1893 * occurred during last 5 mins, NCQ_OFF.
1894 *
1895 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1896 * occurred during last 5 mins, FALLBACK_TO_PIO
1897 *
1898 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1899 * during last 10 mins, NCQ_OFF.
1900 *
1901 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1902 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1903 *
1904 * LOCKING:
1905 * Inherited from caller.
1906 *
1907 * RETURNS:
1908 * OR of ATA_EH_SPDN_* flags.
1909 */
ata_eh_speed_down_verdict(struct ata_device * dev)1910 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1911 {
1912 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1913 u64 j64 = get_jiffies_64();
1914 struct speed_down_verdict_arg arg;
1915 unsigned int verdict = 0;
1916
1917 /* scan past 5 mins of error history */
1918 memset(&arg, 0, sizeof(arg));
1919 arg.since = j64 - min(j64, j5mins);
1920 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1921
1922 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1923 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1924 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1925 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1926
1927 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1928 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1929 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1930
1931 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1932 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1933 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1934 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1935
1936 /* scan past 10 mins of error history */
1937 memset(&arg, 0, sizeof(arg));
1938 arg.since = j64 - min(j64, j10mins);
1939 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1940
1941 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1942 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1943 verdict |= ATA_EH_SPDN_NCQ_OFF;
1944
1945 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1946 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1947 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1948 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1949
1950 return verdict;
1951 }
1952
1953 /**
1954 * ata_eh_speed_down - record error and speed down if necessary
1955 * @dev: Failed device
1956 * @eflags: mask of ATA_EFLAG_* flags
1957 * @err_mask: err_mask of the error
1958 *
1959 * Record error and examine error history to determine whether
1960 * adjusting transmission speed is necessary. It also sets
1961 * transmission limits appropriately if such adjustment is
1962 * necessary.
1963 *
1964 * LOCKING:
1965 * Kernel thread context (may sleep).
1966 *
1967 * RETURNS:
1968 * Determined recovery action.
1969 */
ata_eh_speed_down(struct ata_device * dev,unsigned int eflags,unsigned int err_mask)1970 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1971 unsigned int eflags, unsigned int err_mask)
1972 {
1973 struct ata_link *link = ata_dev_phys_link(dev);
1974 int xfer_ok = 0;
1975 unsigned int verdict;
1976 unsigned int action = 0;
1977
1978 /* don't bother if Cat-0 error */
1979 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1980 return 0;
1981
1982 /* record error and determine whether speed down is necessary */
1983 ata_ering_record(&dev->ering, eflags, err_mask);
1984 verdict = ata_eh_speed_down_verdict(dev);
1985
1986 /* turn off NCQ? */
1987 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1988 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1989 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1990 dev->flags |= ATA_DFLAG_NCQ_OFF;
1991 ata_dev_printk(dev, KERN_WARNING,
1992 "NCQ disabled due to excessive errors\n");
1993 goto done;
1994 }
1995
1996 /* speed down? */
1997 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1998 /* speed down SATA link speed if possible */
1999 if (sata_down_spd_limit(link, 0) == 0) {
2000 action |= ATA_EH_RESET;
2001 goto done;
2002 }
2003
2004 /* lower transfer mode */
2005 if (dev->spdn_cnt < 2) {
2006 static const int dma_dnxfer_sel[] =
2007 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2008 static const int pio_dnxfer_sel[] =
2009 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2010 int sel;
2011
2012 if (dev->xfer_shift != ATA_SHIFT_PIO)
2013 sel = dma_dnxfer_sel[dev->spdn_cnt];
2014 else
2015 sel = pio_dnxfer_sel[dev->spdn_cnt];
2016
2017 dev->spdn_cnt++;
2018
2019 if (ata_down_xfermask_limit(dev, sel) == 0) {
2020 action |= ATA_EH_RESET;
2021 goto done;
2022 }
2023 }
2024 }
2025
2026 /* Fall back to PIO? Slowing down to PIO is meaningless for
2027 * SATA ATA devices. Consider it only for PATA and SATAPI.
2028 */
2029 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2030 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2031 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2032 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2033 dev->spdn_cnt = 0;
2034 action |= ATA_EH_RESET;
2035 goto done;
2036 }
2037 }
2038
2039 return 0;
2040 done:
2041 /* device has been slowed down, blow error history */
2042 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2043 ata_ering_clear(&dev->ering);
2044 return action;
2045 }
2046
2047 /**
2048 * ata_eh_link_autopsy - analyze error and determine recovery action
2049 * @link: host link to perform autopsy on
2050 *
2051 * Analyze why @link failed and determine which recovery actions
2052 * are needed. This function also sets more detailed AC_ERR_*
2053 * values and fills sense data for ATAPI CHECK SENSE.
2054 *
2055 * LOCKING:
2056 * Kernel thread context (may sleep).
2057 */
ata_eh_link_autopsy(struct ata_link * link)2058 static void ata_eh_link_autopsy(struct ata_link *link)
2059 {
2060 struct ata_port *ap = link->ap;
2061 struct ata_eh_context *ehc = &link->eh_context;
2062 struct ata_device *dev;
2063 unsigned int all_err_mask = 0, eflags = 0;
2064 int tag;
2065 u32 serror;
2066 int rc;
2067
2068 DPRINTK("ENTER\n");
2069
2070 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2071 return;
2072
2073 /* obtain and analyze SError */
2074 rc = sata_scr_read(link, SCR_ERROR, &serror);
2075 if (rc == 0) {
2076 ehc->i.serror |= serror;
2077 ata_eh_analyze_serror(link);
2078 } else if (rc != -EOPNOTSUPP) {
2079 /* SError read failed, force reset and probing */
2080 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2081 ehc->i.action |= ATA_EH_RESET;
2082 ehc->i.err_mask |= AC_ERR_OTHER;
2083 }
2084
2085 /* analyze NCQ failure */
2086 ata_eh_analyze_ncq_error(link);
2087
2088 /* any real error trumps AC_ERR_OTHER */
2089 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2090 ehc->i.err_mask &= ~AC_ERR_OTHER;
2091
2092 all_err_mask |= ehc->i.err_mask;
2093
2094 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2095 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2096
2097 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2098 ata_dev_phys_link(qc->dev) != link)
2099 continue;
2100
2101 /* inherit upper level err_mask */
2102 qc->err_mask |= ehc->i.err_mask;
2103
2104 /* analyze TF */
2105 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2106
2107 /* DEV errors are probably spurious in case of ATA_BUS error */
2108 if (qc->err_mask & AC_ERR_ATA_BUS)
2109 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2110 AC_ERR_INVALID);
2111
2112 /* any real error trumps unknown error */
2113 if (qc->err_mask & ~AC_ERR_OTHER)
2114 qc->err_mask &= ~AC_ERR_OTHER;
2115
2116 /* SENSE_VALID trumps dev/unknown error and revalidation */
2117 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2118 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2119
2120 /* determine whether the command is worth retrying */
2121 if (qc->flags & ATA_QCFLAG_IO ||
2122 (!(qc->err_mask & AC_ERR_INVALID) &&
2123 qc->err_mask != AC_ERR_DEV))
2124 qc->flags |= ATA_QCFLAG_RETRY;
2125
2126 /* accumulate error info */
2127 ehc->i.dev = qc->dev;
2128 all_err_mask |= qc->err_mask;
2129 if (qc->flags & ATA_QCFLAG_IO)
2130 eflags |= ATA_EFLAG_IS_IO;
2131 }
2132
2133 /* enforce default EH actions */
2134 if (ap->pflags & ATA_PFLAG_FROZEN ||
2135 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2136 ehc->i.action |= ATA_EH_RESET;
2137 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2138 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2139 ehc->i.action |= ATA_EH_REVALIDATE;
2140
2141 /* If we have offending qcs and the associated failed device,
2142 * perform per-dev EH action only on the offending device.
2143 */
2144 if (ehc->i.dev) {
2145 ehc->i.dev_action[ehc->i.dev->devno] |=
2146 ehc->i.action & ATA_EH_PERDEV_MASK;
2147 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2148 }
2149
2150 /* propagate timeout to host link */
2151 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2152 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2153
2154 /* record error and consider speeding down */
2155 dev = ehc->i.dev;
2156 if (!dev && ((ata_link_max_devices(link) == 1 &&
2157 ata_dev_enabled(link->device))))
2158 dev = link->device;
2159
2160 if (dev) {
2161 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2162 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2163 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2164 }
2165
2166 DPRINTK("EXIT\n");
2167 }
2168
2169 /**
2170 * ata_eh_autopsy - analyze error and determine recovery action
2171 * @ap: host port to perform autopsy on
2172 *
2173 * Analyze all links of @ap and determine why they failed and
2174 * which recovery actions are needed.
2175 *
2176 * LOCKING:
2177 * Kernel thread context (may sleep).
2178 */
ata_eh_autopsy(struct ata_port * ap)2179 void ata_eh_autopsy(struct ata_port *ap)
2180 {
2181 struct ata_link *link;
2182
2183 ata_for_each_link(link, ap, EDGE)
2184 ata_eh_link_autopsy(link);
2185
2186 /* Handle the frigging slave link. Autopsy is done similarly
2187 * but actions and flags are transferred over to the master
2188 * link and handled from there.
2189 */
2190 if (ap->slave_link) {
2191 struct ata_eh_context *mehc = &ap->link.eh_context;
2192 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2193
2194 /* transfer control flags from master to slave */
2195 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2196
2197 /* perform autopsy on the slave link */
2198 ata_eh_link_autopsy(ap->slave_link);
2199
2200 /* transfer actions from slave to master and clear slave */
2201 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2202 mehc->i.action |= sehc->i.action;
2203 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2204 mehc->i.flags |= sehc->i.flags;
2205 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2206 }
2207
2208 /* Autopsy of fanout ports can affect host link autopsy.
2209 * Perform host link autopsy last.
2210 */
2211 if (sata_pmp_attached(ap))
2212 ata_eh_link_autopsy(&ap->link);
2213 }
2214
2215 /**
2216 * ata_get_cmd_descript - get description for ATA command
2217 * @command: ATA command code to get description for
2218 *
2219 * Return a textual description of the given command, or NULL if the
2220 * command is not known.
2221 *
2222 * LOCKING:
2223 * None
2224 */
ata_get_cmd_descript(u8 command)2225 const char *ata_get_cmd_descript(u8 command)
2226 {
2227 #ifdef CONFIG_ATA_VERBOSE_ERROR
2228 static const struct
2229 {
2230 u8 command;
2231 const char *text;
2232 } cmd_descr[] = {
2233 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2234 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2235 { ATA_CMD_STANDBY, "STANDBY" },
2236 { ATA_CMD_IDLE, "IDLE" },
2237 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2238 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2239 { ATA_CMD_NOP, "NOP" },
2240 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2241 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2242 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2243 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2244 { ATA_CMD_SERVICE, "SERVICE" },
2245 { ATA_CMD_READ, "READ DMA" },
2246 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2247 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2248 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2249 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2250 { ATA_CMD_WRITE, "WRITE DMA" },
2251 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2252 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2253 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2254 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2255 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2256 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2257 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2258 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2259 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2260 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2261 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2262 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2263 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2264 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2265 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2266 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2267 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2268 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2269 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2270 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2271 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2272 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2273 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2274 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2275 { ATA_CMD_SLEEP, "SLEEP" },
2276 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2277 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2278 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2279 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2280 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2281 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2282 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2283 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2284 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2285 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2286 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2287 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2288 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2289 { ATA_CMD_PMP_READ, "READ BUFFER" },
2290 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2291 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2292 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2293 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2294 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2295 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2296 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2297 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2298 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2299 { ATA_CMD_SMART, "SMART" },
2300 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2301 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2302 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2303 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2304 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2305 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2306 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2307 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2308 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2309 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2310 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2311 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2312 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2313 { ATA_CMD_RESTORE, "RECALIBRATE" },
2314 { 0, NULL } /* terminate list */
2315 };
2316
2317 unsigned int i;
2318 for (i = 0; cmd_descr[i].text; i++)
2319 if (cmd_descr[i].command == command)
2320 return cmd_descr[i].text;
2321 #endif
2322
2323 return NULL;
2324 }
2325
2326 /**
2327 * ata_eh_link_report - report error handling to user
2328 * @link: ATA link EH is going on
2329 *
2330 * Report EH to user.
2331 *
2332 * LOCKING:
2333 * None.
2334 */
ata_eh_link_report(struct ata_link * link)2335 static void ata_eh_link_report(struct ata_link *link)
2336 {
2337 struct ata_port *ap = link->ap;
2338 struct ata_eh_context *ehc = &link->eh_context;
2339 const char *frozen, *desc;
2340 char tries_buf[6];
2341 int tag, nr_failed = 0;
2342
2343 if (ehc->i.flags & ATA_EHI_QUIET)
2344 return;
2345
2346 desc = NULL;
2347 if (ehc->i.desc[0] != '\0')
2348 desc = ehc->i.desc;
2349
2350 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2351 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2352
2353 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2354 ata_dev_phys_link(qc->dev) != link ||
2355 ((qc->flags & ATA_QCFLAG_QUIET) &&
2356 qc->err_mask == AC_ERR_DEV))
2357 continue;
2358 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2359 continue;
2360
2361 nr_failed++;
2362 }
2363
2364 if (!nr_failed && !ehc->i.err_mask)
2365 return;
2366
2367 frozen = "";
2368 if (ap->pflags & ATA_PFLAG_FROZEN)
2369 frozen = " frozen";
2370
2371 memset(tries_buf, 0, sizeof(tries_buf));
2372 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2373 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2374 ap->eh_tries);
2375
2376 if (ehc->i.dev) {
2377 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2378 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2379 ehc->i.err_mask, link->sactive, ehc->i.serror,
2380 ehc->i.action, frozen, tries_buf);
2381 if (desc)
2382 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2383 } else {
2384 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2385 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2386 ehc->i.err_mask, link->sactive, ehc->i.serror,
2387 ehc->i.action, frozen, tries_buf);
2388 if (desc)
2389 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2390 }
2391
2392 #ifdef CONFIG_ATA_VERBOSE_ERROR
2393 if (ehc->i.serror)
2394 ata_link_printk(link, KERN_ERR,
2395 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2396 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2397 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2398 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2399 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2400 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2401 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2402 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2403 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2404 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2405 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2406 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2407 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2408 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2409 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2410 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2411 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2412 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2413 #endif
2414
2415 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2416 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2417 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2418 const u8 *cdb = qc->cdb;
2419 char data_buf[20] = "";
2420 char cdb_buf[70] = "";
2421
2422 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
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 static const char *prot_str[] = {
2433 [ATA_PROT_PIO] = "pio",
2434 [ATA_PROT_DMA] = "dma",
2435 [ATA_PROT_NCQ] = "ncq",
2436 [ATAPI_PROT_PIO] = "pio",
2437 [ATAPI_PROT_DMA] = "dma",
2438 };
2439
2440 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2441 prot_str[qc->tf.protocol], qc->nbytes,
2442 dma_str[qc->dma_dir]);
2443 }
2444
2445 if (ata_is_atapi(qc->tf.protocol)) {
2446 if (qc->scsicmd)
2447 scsi_print_command(qc->scsicmd);
2448 else
2449 snprintf(cdb_buf, sizeof(cdb_buf),
2450 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2451 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2452 cdb[0], cdb[1], cdb[2], cdb[3],
2453 cdb[4], cdb[5], cdb[6], cdb[7],
2454 cdb[8], cdb[9], cdb[10], cdb[11],
2455 cdb[12], cdb[13], cdb[14], cdb[15]);
2456 } else {
2457 const char *descr = ata_get_cmd_descript(cmd->command);
2458 if (descr)
2459 ata_dev_printk(qc->dev, KERN_ERR,
2460 "failed command: %s\n", descr);
2461 }
2462
2463 ata_dev_printk(qc->dev, KERN_ERR,
2464 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2465 "tag %d%s\n %s"
2466 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2467 "Emask 0x%x (%s)%s\n",
2468 cmd->command, cmd->feature, cmd->nsect,
2469 cmd->lbal, cmd->lbam, cmd->lbah,
2470 cmd->hob_feature, cmd->hob_nsect,
2471 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2472 cmd->device, qc->tag, data_buf, cdb_buf,
2473 res->command, res->feature, res->nsect,
2474 res->lbal, res->lbam, res->lbah,
2475 res->hob_feature, res->hob_nsect,
2476 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2477 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2478 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2479
2480 #ifdef CONFIG_ATA_VERBOSE_ERROR
2481 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2482 ATA_ERR)) {
2483 if (res->command & ATA_BUSY)
2484 ata_dev_printk(qc->dev, KERN_ERR,
2485 "status: { Busy }\n");
2486 else
2487 ata_dev_printk(qc->dev, KERN_ERR,
2488 "status: { %s%s%s%s}\n",
2489 res->command & ATA_DRDY ? "DRDY " : "",
2490 res->command & ATA_DF ? "DF " : "",
2491 res->command & ATA_DRQ ? "DRQ " : "",
2492 res->command & ATA_ERR ? "ERR " : "");
2493 }
2494
2495 if (cmd->command != ATA_CMD_PACKET &&
2496 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2497 ATA_ABORTED)))
2498 ata_dev_printk(qc->dev, KERN_ERR,
2499 "error: { %s%s%s%s}\n",
2500 res->feature & ATA_ICRC ? "ICRC " : "",
2501 res->feature & ATA_UNC ? "UNC " : "",
2502 res->feature & ATA_IDNF ? "IDNF " : "",
2503 res->feature & ATA_ABORTED ? "ABRT " : "");
2504 #endif
2505 }
2506 }
2507
2508 /**
2509 * ata_eh_report - report error handling to user
2510 * @ap: ATA port to report EH about
2511 *
2512 * Report EH to user.
2513 *
2514 * LOCKING:
2515 * None.
2516 */
ata_eh_report(struct ata_port * ap)2517 void ata_eh_report(struct ata_port *ap)
2518 {
2519 struct ata_link *link;
2520
2521 ata_for_each_link(link, ap, HOST_FIRST)
2522 ata_eh_link_report(link);
2523 }
2524
ata_do_reset(struct ata_link * link,ata_reset_fn_t reset,unsigned int * classes,unsigned long deadline,bool clear_classes)2525 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2526 unsigned int *classes, unsigned long deadline,
2527 bool clear_classes)
2528 {
2529 struct ata_device *dev;
2530
2531 if (clear_classes)
2532 ata_for_each_dev(dev, link, ALL)
2533 classes[dev->devno] = ATA_DEV_UNKNOWN;
2534
2535 return reset(link, classes, deadline);
2536 }
2537
ata_eh_followup_srst_needed(struct ata_link * link,int rc,const unsigned int * classes)2538 static int ata_eh_followup_srst_needed(struct ata_link *link,
2539 int rc, const unsigned int *classes)
2540 {
2541 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2542 return 0;
2543 if (rc == -EAGAIN)
2544 return 1;
2545 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2546 return 1;
2547 return 0;
2548 }
2549
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)2550 int ata_eh_reset(struct ata_link *link, int classify,
2551 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2552 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2553 {
2554 struct ata_port *ap = link->ap;
2555 struct ata_link *slave = ap->slave_link;
2556 struct ata_eh_context *ehc = &link->eh_context;
2557 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2558 unsigned int *classes = ehc->classes;
2559 unsigned int lflags = link->flags;
2560 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2561 int max_tries = 0, try = 0;
2562 struct ata_link *failed_link;
2563 struct ata_device *dev;
2564 unsigned long deadline, now;
2565 ata_reset_fn_t reset;
2566 unsigned long flags;
2567 u32 sstatus;
2568 int nr_unknown, rc;
2569
2570 /*
2571 * Prepare to reset
2572 */
2573 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2574 max_tries++;
2575 if (link->flags & ATA_LFLAG_NO_HRST)
2576 hardreset = NULL;
2577 if (link->flags & ATA_LFLAG_NO_SRST)
2578 softreset = NULL;
2579
2580 /* make sure each reset attempt is at least COOL_DOWN apart */
2581 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2582 now = jiffies;
2583 WARN_ON(time_after(ehc->last_reset, now));
2584 deadline = ata_deadline(ehc->last_reset,
2585 ATA_EH_RESET_COOL_DOWN);
2586 if (time_before(now, deadline))
2587 schedule_timeout_uninterruptible(deadline - now);
2588 }
2589
2590 spin_lock_irqsave(ap->lock, flags);
2591 ap->pflags |= ATA_PFLAG_RESETTING;
2592 spin_unlock_irqrestore(ap->lock, flags);
2593
2594 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2595
2596 ata_for_each_dev(dev, link, ALL) {
2597 /* If we issue an SRST then an ATA drive (not ATAPI)
2598 * may change configuration and be in PIO0 timing. If
2599 * we do a hard reset (or are coming from power on)
2600 * this is true for ATA or ATAPI. Until we've set a
2601 * suitable controller mode we should not touch the
2602 * bus as we may be talking too fast.
2603 */
2604 dev->pio_mode = XFER_PIO_0;
2605
2606 /* If the controller has a pio mode setup function
2607 * then use it to set the chipset to rights. Don't
2608 * touch the DMA setup as that will be dealt with when
2609 * configuring devices.
2610 */
2611 if (ap->ops->set_piomode)
2612 ap->ops->set_piomode(ap, dev);
2613 }
2614
2615 /* prefer hardreset */
2616 reset = NULL;
2617 ehc->i.action &= ~ATA_EH_RESET;
2618 if (hardreset) {
2619 reset = hardreset;
2620 ehc->i.action |= ATA_EH_HARDRESET;
2621 } else if (softreset) {
2622 reset = softreset;
2623 ehc->i.action |= ATA_EH_SOFTRESET;
2624 }
2625
2626 if (prereset) {
2627 unsigned long deadline = ata_deadline(jiffies,
2628 ATA_EH_PRERESET_TIMEOUT);
2629
2630 if (slave) {
2631 sehc->i.action &= ~ATA_EH_RESET;
2632 sehc->i.action |= ehc->i.action;
2633 }
2634
2635 rc = prereset(link, deadline);
2636
2637 /* If present, do prereset on slave link too. Reset
2638 * is skipped iff both master and slave links report
2639 * -ENOENT or clear ATA_EH_RESET.
2640 */
2641 if (slave && (rc == 0 || rc == -ENOENT)) {
2642 int tmp;
2643
2644 tmp = prereset(slave, deadline);
2645 if (tmp != -ENOENT)
2646 rc = tmp;
2647
2648 ehc->i.action |= sehc->i.action;
2649 }
2650
2651 if (rc) {
2652 if (rc == -ENOENT) {
2653 ata_link_printk(link, KERN_DEBUG,
2654 "port disabled. ignoring.\n");
2655 ehc->i.action &= ~ATA_EH_RESET;
2656
2657 ata_for_each_dev(dev, link, ALL)
2658 classes[dev->devno] = ATA_DEV_NONE;
2659
2660 rc = 0;
2661 } else
2662 ata_link_printk(link, KERN_ERR,
2663 "prereset failed (errno=%d)\n", rc);
2664 goto out;
2665 }
2666
2667 /* prereset() might have cleared ATA_EH_RESET. If so,
2668 * bang classes, thaw and return.
2669 */
2670 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2671 ata_for_each_dev(dev, link, ALL)
2672 classes[dev->devno] = ATA_DEV_NONE;
2673 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2674 ata_is_host_link(link))
2675 ata_eh_thaw_port(ap);
2676 rc = 0;
2677 goto out;
2678 }
2679 }
2680
2681 retry:
2682 /*
2683 * Perform reset
2684 */
2685 if (ata_is_host_link(link))
2686 ata_eh_freeze_port(ap);
2687
2688 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2689
2690 if (reset) {
2691 if (verbose)
2692 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2693 reset == softreset ? "soft" : "hard");
2694
2695 /* mark that this EH session started with reset */
2696 ehc->last_reset = jiffies;
2697 if (reset == hardreset)
2698 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2699 else
2700 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2701
2702 rc = ata_do_reset(link, reset, classes, deadline, true);
2703 if (rc && rc != -EAGAIN) {
2704 failed_link = link;
2705 goto fail;
2706 }
2707
2708 /* hardreset slave link if existent */
2709 if (slave && reset == hardreset) {
2710 int tmp;
2711
2712 if (verbose)
2713 ata_link_printk(slave, KERN_INFO,
2714 "hard resetting link\n");
2715
2716 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2717 tmp = ata_do_reset(slave, reset, classes, deadline,
2718 false);
2719 switch (tmp) {
2720 case -EAGAIN:
2721 rc = -EAGAIN;
2722 case 0:
2723 break;
2724 default:
2725 failed_link = slave;
2726 rc = tmp;
2727 goto fail;
2728 }
2729 }
2730
2731 /* perform follow-up SRST if necessary */
2732 if (reset == hardreset &&
2733 ata_eh_followup_srst_needed(link, rc, classes)) {
2734 reset = softreset;
2735
2736 if (!reset) {
2737 ata_link_printk(link, KERN_ERR,
2738 "follow-up softreset required "
2739 "but no softreset available\n");
2740 failed_link = link;
2741 rc = -EINVAL;
2742 goto fail;
2743 }
2744
2745 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2746 rc = ata_do_reset(link, reset, classes, deadline, true);
2747 if (rc) {
2748 failed_link = link;
2749 goto fail;
2750 }
2751 }
2752 } else {
2753 if (verbose)
2754 ata_link_printk(link, KERN_INFO, "no reset method "
2755 "available, skipping reset\n");
2756 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2757 lflags |= ATA_LFLAG_ASSUME_ATA;
2758 }
2759
2760 /*
2761 * Post-reset processing
2762 */
2763 ata_for_each_dev(dev, link, ALL) {
2764 /* After the reset, the device state is PIO 0 and the
2765 * controller state is undefined. Reset also wakes up
2766 * drives from sleeping mode.
2767 */
2768 dev->pio_mode = XFER_PIO_0;
2769 dev->flags &= ~ATA_DFLAG_SLEEPING;
2770
2771 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2772 continue;
2773
2774 /* apply class override */
2775 if (lflags & ATA_LFLAG_ASSUME_ATA)
2776 classes[dev->devno] = ATA_DEV_ATA;
2777 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2778 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2779 }
2780
2781 /* record current link speed */
2782 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2783 link->sata_spd = (sstatus >> 4) & 0xf;
2784 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2785 slave->sata_spd = (sstatus >> 4) & 0xf;
2786
2787 /* thaw the port */
2788 if (ata_is_host_link(link))
2789 ata_eh_thaw_port(ap);
2790
2791 /* postreset() should clear hardware SError. Although SError
2792 * is cleared during link resume, clearing SError here is
2793 * necessary as some PHYs raise hotplug events after SRST.
2794 * This introduces race condition where hotplug occurs between
2795 * reset and here. This race is mediated by cross checking
2796 * link onlineness and classification result later.
2797 */
2798 if (postreset) {
2799 postreset(link, classes);
2800 if (slave)
2801 postreset(slave, classes);
2802 }
2803
2804 /*
2805 * Some controllers can't be frozen very well and may set
2806 * spuruious error conditions during reset. Clear accumulated
2807 * error information. As reset is the final recovery action,
2808 * nothing is lost by doing this.
2809 */
2810 spin_lock_irqsave(link->ap->lock, flags);
2811 memset(&link->eh_info, 0, sizeof(link->eh_info));
2812 if (slave)
2813 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2814 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2815 spin_unlock_irqrestore(link->ap->lock, flags);
2816
2817 /*
2818 * Make sure onlineness and classification result correspond.
2819 * Hotplug could have happened during reset and some
2820 * controllers fail to wait while a drive is spinning up after
2821 * being hotplugged causing misdetection. By cross checking
2822 * link on/offlineness and classification result, those
2823 * conditions can be reliably detected and retried.
2824 */
2825 nr_unknown = 0;
2826 ata_for_each_dev(dev, link, ALL) {
2827 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2828 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2829 ata_dev_printk(dev, KERN_DEBUG, "link online "
2830 "but device misclassifed\n");
2831 classes[dev->devno] = ATA_DEV_NONE;
2832 nr_unknown++;
2833 }
2834 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2835 if (ata_class_enabled(classes[dev->devno]))
2836 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2837 "clearing class %d to NONE\n",
2838 classes[dev->devno]);
2839 classes[dev->devno] = ATA_DEV_NONE;
2840 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2841 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2842 "clearing UNKNOWN to NONE\n");
2843 classes[dev->devno] = ATA_DEV_NONE;
2844 }
2845 }
2846
2847 if (classify && nr_unknown) {
2848 if (try < max_tries) {
2849 ata_link_printk(link, KERN_WARNING, "link online but "
2850 "%d devices misclassified, retrying\n",
2851 nr_unknown);
2852 failed_link = link;
2853 rc = -EAGAIN;
2854 goto fail;
2855 }
2856 ata_link_printk(link, KERN_WARNING,
2857 "link online but %d devices misclassified, "
2858 "device detection might fail\n", nr_unknown);
2859 }
2860
2861 /* reset successful, schedule revalidation */
2862 ata_eh_done(link, NULL, ATA_EH_RESET);
2863 if (slave)
2864 ata_eh_done(slave, NULL, ATA_EH_RESET);
2865 ehc->last_reset = jiffies; /* update to completion time */
2866 ehc->i.action |= ATA_EH_REVALIDATE;
2867 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2868
2869 rc = 0;
2870 out:
2871 /* clear hotplug flag */
2872 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2873 if (slave)
2874 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2875
2876 spin_lock_irqsave(ap->lock, flags);
2877 ap->pflags &= ~ATA_PFLAG_RESETTING;
2878 spin_unlock_irqrestore(ap->lock, flags);
2879
2880 return rc;
2881
2882 fail:
2883 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2884 if (!ata_is_host_link(link) &&
2885 sata_scr_read(link, SCR_STATUS, &sstatus))
2886 rc = -ERESTART;
2887
2888 if (rc == -ERESTART || try >= max_tries)
2889 goto out;
2890
2891 now = jiffies;
2892 if (time_before(now, deadline)) {
2893 unsigned long delta = deadline - now;
2894
2895 ata_link_printk(failed_link, KERN_WARNING,
2896 "reset failed (errno=%d), retrying in %u secs\n",
2897 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2898
2899 ata_eh_release(ap);
2900 while (delta)
2901 delta = schedule_timeout_uninterruptible(delta);
2902 ata_eh_acquire(ap);
2903 }
2904
2905 if (try == max_tries - 1) {
2906 sata_down_spd_limit(link, 0);
2907 if (slave)
2908 sata_down_spd_limit(slave, 0);
2909 } else if (rc == -EPIPE)
2910 sata_down_spd_limit(failed_link, 0);
2911
2912 if (hardreset)
2913 reset = hardreset;
2914 goto retry;
2915 }
2916
ata_eh_pull_park_action(struct ata_port * ap)2917 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2918 {
2919 struct ata_link *link;
2920 struct ata_device *dev;
2921 unsigned long flags;
2922
2923 /*
2924 * This function can be thought of as an extended version of
2925 * ata_eh_about_to_do() specially crafted to accommodate the
2926 * requirements of ATA_EH_PARK handling. Since the EH thread
2927 * does not leave the do {} while () loop in ata_eh_recover as
2928 * long as the timeout for a park request to *one* device on
2929 * the port has not expired, and since we still want to pick
2930 * up park requests to other devices on the same port or
2931 * timeout updates for the same device, we have to pull
2932 * ATA_EH_PARK actions from eh_info into eh_context.i
2933 * ourselves at the beginning of each pass over the loop.
2934 *
2935 * Additionally, all write accesses to &ap->park_req_pending
2936 * through INIT_COMPLETION() (see below) or complete_all()
2937 * (see ata_scsi_park_store()) are protected by the host lock.
2938 * As a result we have that park_req_pending.done is zero on
2939 * exit from this function, i.e. when ATA_EH_PARK actions for
2940 * *all* devices on port ap have been pulled into the
2941 * respective eh_context structs. If, and only if,
2942 * park_req_pending.done is non-zero by the time we reach
2943 * wait_for_completion_timeout(), another ATA_EH_PARK action
2944 * has been scheduled for at least one of the devices on port
2945 * ap and we have to cycle over the do {} while () loop in
2946 * ata_eh_recover() again.
2947 */
2948
2949 spin_lock_irqsave(ap->lock, flags);
2950 INIT_COMPLETION(ap->park_req_pending);
2951 ata_for_each_link(link, ap, EDGE) {
2952 ata_for_each_dev(dev, link, ALL) {
2953 struct ata_eh_info *ehi = &link->eh_info;
2954
2955 link->eh_context.i.dev_action[dev->devno] |=
2956 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2957 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2958 }
2959 }
2960 spin_unlock_irqrestore(ap->lock, flags);
2961 }
2962
ata_eh_park_issue_cmd(struct ata_device * dev,int park)2963 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2964 {
2965 struct ata_eh_context *ehc = &dev->link->eh_context;
2966 struct ata_taskfile tf;
2967 unsigned int err_mask;
2968
2969 ata_tf_init(dev, &tf);
2970 if (park) {
2971 ehc->unloaded_mask |= 1 << dev->devno;
2972 tf.command = ATA_CMD_IDLEIMMEDIATE;
2973 tf.feature = 0x44;
2974 tf.lbal = 0x4c;
2975 tf.lbam = 0x4e;
2976 tf.lbah = 0x55;
2977 } else {
2978 ehc->unloaded_mask &= ~(1 << dev->devno);
2979 tf.command = ATA_CMD_CHK_POWER;
2980 }
2981
2982 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2983 tf.protocol |= ATA_PROT_NODATA;
2984 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2985 if (park && (err_mask || tf.lbal != 0xc4)) {
2986 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2987 ehc->unloaded_mask &= ~(1 << dev->devno);
2988 }
2989 }
2990
ata_eh_revalidate_and_attach(struct ata_link * link,struct ata_device ** r_failed_dev)2991 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2992 struct ata_device **r_failed_dev)
2993 {
2994 struct ata_port *ap = link->ap;
2995 struct ata_eh_context *ehc = &link->eh_context;
2996 struct ata_device *dev;
2997 unsigned int new_mask = 0;
2998 unsigned long flags;
2999 int rc = 0;
3000
3001 DPRINTK("ENTER\n");
3002
3003 /* For PATA drive side cable detection to work, IDENTIFY must
3004 * be done backwards such that PDIAG- is released by the slave
3005 * device before the master device is identified.
3006 */
3007 ata_for_each_dev(dev, link, ALL_REVERSE) {
3008 unsigned int action = ata_eh_dev_action(dev);
3009 unsigned int readid_flags = 0;
3010
3011 if (ehc->i.flags & ATA_EHI_DID_RESET)
3012 readid_flags |= ATA_READID_POSTRESET;
3013
3014 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3015 WARN_ON(dev->class == ATA_DEV_PMP);
3016
3017 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3018 rc = -EIO;
3019 goto err;
3020 }
3021
3022 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3023 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3024 readid_flags);
3025 if (rc)
3026 goto err;
3027
3028 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3029
3030 /* Configuration may have changed, reconfigure
3031 * transfer mode.
3032 */
3033 ehc->i.flags |= ATA_EHI_SETMODE;
3034
3035 /* schedule the scsi_rescan_device() here */
3036 schedule_work(&(ap->scsi_rescan_task));
3037 } else if (dev->class == ATA_DEV_UNKNOWN &&
3038 ehc->tries[dev->devno] &&
3039 ata_class_enabled(ehc->classes[dev->devno])) {
3040 /* Temporarily set dev->class, it will be
3041 * permanently set once all configurations are
3042 * complete. This is necessary because new
3043 * device configuration is done in two
3044 * separate loops.
3045 */
3046 dev->class = ehc->classes[dev->devno];
3047
3048 if (dev->class == ATA_DEV_PMP)
3049 rc = sata_pmp_attach(dev);
3050 else
3051 rc = ata_dev_read_id(dev, &dev->class,
3052 readid_flags, dev->id);
3053
3054 /* read_id might have changed class, store and reset */
3055 ehc->classes[dev->devno] = dev->class;
3056 dev->class = ATA_DEV_UNKNOWN;
3057
3058 switch (rc) {
3059 case 0:
3060 /* clear error info accumulated during probe */
3061 ata_ering_clear(&dev->ering);
3062 new_mask |= 1 << dev->devno;
3063 break;
3064 case -ENOENT:
3065 /* IDENTIFY was issued to non-existent
3066 * device. No need to reset. Just
3067 * thaw and ignore the device.
3068 */
3069 ata_eh_thaw_port(ap);
3070 break;
3071 default:
3072 goto err;
3073 }
3074 }
3075 }
3076
3077 /* PDIAG- should have been released, ask cable type if post-reset */
3078 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3079 if (ap->ops->cable_detect)
3080 ap->cbl = ap->ops->cable_detect(ap);
3081 ata_force_cbl(ap);
3082 }
3083
3084 /* Configure new devices forward such that user doesn't see
3085 * device detection messages backwards.
3086 */
3087 ata_for_each_dev(dev, link, ALL) {
3088 if (!(new_mask & (1 << dev->devno)))
3089 continue;
3090
3091 dev->class = ehc->classes[dev->devno];
3092
3093 if (dev->class == ATA_DEV_PMP)
3094 continue;
3095
3096 ehc->i.flags |= ATA_EHI_PRINTINFO;
3097 rc = ata_dev_configure(dev);
3098 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3099 if (rc) {
3100 dev->class = ATA_DEV_UNKNOWN;
3101 goto err;
3102 }
3103
3104 spin_lock_irqsave(ap->lock, flags);
3105 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3106 spin_unlock_irqrestore(ap->lock, flags);
3107
3108 /* new device discovered, configure xfermode */
3109 ehc->i.flags |= ATA_EHI_SETMODE;
3110 }
3111
3112 return 0;
3113
3114 err:
3115 *r_failed_dev = dev;
3116 DPRINTK("EXIT rc=%d\n", rc);
3117 return rc;
3118 }
3119
3120 /**
3121 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3122 * @link: link on which timings will be programmed
3123 * @r_failed_dev: out parameter for failed device
3124 *
3125 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3126 * ata_set_mode() fails, pointer to the failing device is
3127 * returned in @r_failed_dev.
3128 *
3129 * LOCKING:
3130 * PCI/etc. bus probe sem.
3131 *
3132 * RETURNS:
3133 * 0 on success, negative errno otherwise
3134 */
ata_set_mode(struct ata_link * link,struct ata_device ** r_failed_dev)3135 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3136 {
3137 struct ata_port *ap = link->ap;
3138 struct ata_device *dev;
3139 int rc;
3140
3141 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3142 ata_for_each_dev(dev, link, ENABLED) {
3143 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3144 struct ata_ering_entry *ent;
3145
3146 ent = ata_ering_top(&dev->ering);
3147 if (ent)
3148 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3149 }
3150 }
3151
3152 /* has private set_mode? */
3153 if (ap->ops->set_mode)
3154 rc = ap->ops->set_mode(link, r_failed_dev);
3155 else
3156 rc = ata_do_set_mode(link, r_failed_dev);
3157
3158 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3159 ata_for_each_dev(dev, link, ENABLED) {
3160 struct ata_eh_context *ehc = &link->eh_context;
3161 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3162 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3163
3164 if (dev->xfer_mode != saved_xfer_mode ||
3165 ata_ncq_enabled(dev) != saved_ncq)
3166 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3167 }
3168
3169 return rc;
3170 }
3171
3172 /**
3173 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3174 * @dev: ATAPI device to clear UA for
3175 *
3176 * Resets and other operations can make an ATAPI device raise
3177 * UNIT ATTENTION which causes the next operation to fail. This
3178 * function clears UA.
3179 *
3180 * LOCKING:
3181 * EH context (may sleep).
3182 *
3183 * RETURNS:
3184 * 0 on success, -errno on failure.
3185 */
atapi_eh_clear_ua(struct ata_device * dev)3186 static int atapi_eh_clear_ua(struct ata_device *dev)
3187 {
3188 int i;
3189
3190 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3191 u8 *sense_buffer = dev->link->ap->sector_buf;
3192 u8 sense_key = 0;
3193 unsigned int err_mask;
3194
3195 err_mask = atapi_eh_tur(dev, &sense_key);
3196 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3197 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3198 "failed (err_mask=0x%x)\n", err_mask);
3199 return -EIO;
3200 }
3201
3202 if (!err_mask || sense_key != UNIT_ATTENTION)
3203 return 0;
3204
3205 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3206 if (err_mask) {
3207 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3208 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3209 return -EIO;
3210 }
3211 }
3212
3213 ata_dev_printk(dev, KERN_WARNING,
3214 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3215
3216 return 0;
3217 }
3218
3219 /**
3220 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3221 * @dev: ATA device which may need FLUSH retry
3222 *
3223 * If @dev failed FLUSH, it needs to be reported upper layer
3224 * immediately as it means that @dev failed to remap and already
3225 * lost at least a sector and further FLUSH retrials won't make
3226 * any difference to the lost sector. However, if FLUSH failed
3227 * for other reasons, for example transmission error, FLUSH needs
3228 * to be retried.
3229 *
3230 * This function determines whether FLUSH failure retry is
3231 * necessary and performs it if so.
3232 *
3233 * RETURNS:
3234 * 0 if EH can continue, -errno if EH needs to be repeated.
3235 */
ata_eh_maybe_retry_flush(struct ata_device * dev)3236 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3237 {
3238 struct ata_link *link = dev->link;
3239 struct ata_port *ap = link->ap;
3240 struct ata_queued_cmd *qc;
3241 struct ata_taskfile tf;
3242 unsigned int err_mask;
3243 int rc = 0;
3244
3245 /* did flush fail for this device? */
3246 if (!ata_tag_valid(link->active_tag))
3247 return 0;
3248
3249 qc = __ata_qc_from_tag(ap, link->active_tag);
3250 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3251 qc->tf.command != ATA_CMD_FLUSH))
3252 return 0;
3253
3254 /* if the device failed it, it should be reported to upper layers */
3255 if (qc->err_mask & AC_ERR_DEV)
3256 return 0;
3257
3258 /* flush failed for some other reason, give it another shot */
3259 ata_tf_init(dev, &tf);
3260
3261 tf.command = qc->tf.command;
3262 tf.flags |= ATA_TFLAG_DEVICE;
3263 tf.protocol = ATA_PROT_NODATA;
3264
3265 ata_dev_printk(dev, KERN_WARNING, "retrying FLUSH 0x%x Emask 0x%x\n",
3266 tf.command, qc->err_mask);
3267
3268 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3269 if (!err_mask) {
3270 /*
3271 * FLUSH is complete but there's no way to
3272 * successfully complete a failed command from EH.
3273 * Making sure retry is allowed at least once and
3274 * retrying it should do the trick - whatever was in
3275 * the cache is already on the platter and this won't
3276 * cause infinite loop.
3277 */
3278 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3279 } else {
3280 ata_dev_printk(dev, KERN_WARNING, "FLUSH failed Emask 0x%x\n",
3281 err_mask);
3282 rc = -EIO;
3283
3284 /* if device failed it, report it to upper layers */
3285 if (err_mask & AC_ERR_DEV) {
3286 qc->err_mask |= AC_ERR_DEV;
3287 qc->result_tf = tf;
3288 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3289 rc = 0;
3290 }
3291 }
3292 return rc;
3293 }
3294
3295 /**
3296 * ata_eh_set_lpm - configure SATA interface power management
3297 * @link: link to configure power management
3298 * @policy: the link power management policy
3299 * @r_failed_dev: out parameter for failed device
3300 *
3301 * Enable SATA Interface power management. This will enable
3302 * Device Interface Power Management (DIPM) for min_power
3303 * policy, and then call driver specific callbacks for
3304 * enabling Host Initiated Power management.
3305 *
3306 * LOCKING:
3307 * EH context.
3308 *
3309 * RETURNS:
3310 * 0 on success, -errno on failure.
3311 */
ata_eh_set_lpm(struct ata_link * link,enum ata_lpm_policy policy,struct ata_device ** r_failed_dev)3312 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3313 struct ata_device **r_failed_dev)
3314 {
3315 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3316 struct ata_eh_context *ehc = &link->eh_context;
3317 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3318 enum ata_lpm_policy old_policy = link->lpm_policy;
3319 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3320 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3321 unsigned int err_mask;
3322 int rc;
3323
3324 /* if the link or host doesn't do LPM, noop */
3325 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3326 return 0;
3327
3328 /*
3329 * DIPM is enabled only for MIN_POWER as some devices
3330 * misbehave when the host NACKs transition to SLUMBER. Order
3331 * device and link configurations such that the host always
3332 * allows DIPM requests.
3333 */
3334 ata_for_each_dev(dev, link, ENABLED) {
3335 bool hipm = ata_id_has_hipm(dev->id);
3336 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3337
3338 /* find the first enabled and LPM enabled devices */
3339 if (!link_dev)
3340 link_dev = dev;
3341
3342 if (!lpm_dev && (hipm || dipm))
3343 lpm_dev = dev;
3344
3345 hints &= ~ATA_LPM_EMPTY;
3346 if (!hipm)
3347 hints &= ~ATA_LPM_HIPM;
3348
3349 /* disable DIPM before changing link config */
3350 if (policy != ATA_LPM_MIN_POWER && dipm) {
3351 err_mask = ata_dev_set_feature(dev,
3352 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3353 if (err_mask && err_mask != AC_ERR_DEV) {
3354 ata_dev_printk(dev, KERN_WARNING,
3355 "failed to disable DIPM, Emask 0x%x\n",
3356 err_mask);
3357 rc = -EIO;
3358 goto fail;
3359 }
3360 }
3361 }
3362
3363 if (ap) {
3364 rc = ap->ops->set_lpm(link, policy, hints);
3365 if (!rc && ap->slave_link)
3366 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3367 } else
3368 rc = sata_pmp_set_lpm(link, policy, hints);
3369
3370 /*
3371 * Attribute link config failure to the first (LPM) enabled
3372 * device on the link.
3373 */
3374 if (rc) {
3375 if (rc == -EOPNOTSUPP) {
3376 link->flags |= ATA_LFLAG_NO_LPM;
3377 return 0;
3378 }
3379 dev = lpm_dev ? lpm_dev : link_dev;
3380 goto fail;
3381 }
3382
3383 /*
3384 * Low level driver acked the transition. Issue DIPM command
3385 * with the new policy set.
3386 */
3387 link->lpm_policy = policy;
3388 if (ap && ap->slave_link)
3389 ap->slave_link->lpm_policy = policy;
3390
3391 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3392 ata_for_each_dev(dev, link, ENABLED) {
3393 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3394 ata_id_has_dipm(dev->id)) {
3395 err_mask = ata_dev_set_feature(dev,
3396 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3397 if (err_mask && err_mask != AC_ERR_DEV) {
3398 ata_dev_printk(dev, KERN_WARNING,
3399 "failed to enable DIPM, Emask 0x%x\n",
3400 err_mask);
3401 rc = -EIO;
3402 goto fail;
3403 }
3404 }
3405 }
3406
3407 return 0;
3408
3409 fail:
3410 /* restore the old policy */
3411 link->lpm_policy = old_policy;
3412 if (ap && ap->slave_link)
3413 ap->slave_link->lpm_policy = old_policy;
3414
3415 /* if no device or only one more chance is left, disable LPM */
3416 if (!dev || ehc->tries[dev->devno] <= 2) {
3417 ata_link_printk(link, KERN_WARNING,
3418 "disabling LPM on the link\n");
3419 link->flags |= ATA_LFLAG_NO_LPM;
3420 }
3421 if (r_failed_dev)
3422 *r_failed_dev = dev;
3423 return rc;
3424 }
3425
ata_link_nr_enabled(struct ata_link * link)3426 static int ata_link_nr_enabled(struct ata_link *link)
3427 {
3428 struct ata_device *dev;
3429 int cnt = 0;
3430
3431 ata_for_each_dev(dev, link, ENABLED)
3432 cnt++;
3433 return cnt;
3434 }
3435
ata_link_nr_vacant(struct ata_link * link)3436 static int ata_link_nr_vacant(struct ata_link *link)
3437 {
3438 struct ata_device *dev;
3439 int cnt = 0;
3440
3441 ata_for_each_dev(dev, link, ALL)
3442 if (dev->class == ATA_DEV_UNKNOWN)
3443 cnt++;
3444 return cnt;
3445 }
3446
ata_eh_skip_recovery(struct ata_link * link)3447 static int ata_eh_skip_recovery(struct ata_link *link)
3448 {
3449 struct ata_port *ap = link->ap;
3450 struct ata_eh_context *ehc = &link->eh_context;
3451 struct ata_device *dev;
3452
3453 /* skip disabled links */
3454 if (link->flags & ATA_LFLAG_DISABLED)
3455 return 1;
3456
3457 /* skip if explicitly requested */
3458 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3459 return 1;
3460
3461 /* thaw frozen port and recover failed devices */
3462 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3463 return 0;
3464
3465 /* reset at least once if reset is requested */
3466 if ((ehc->i.action & ATA_EH_RESET) &&
3467 !(ehc->i.flags & ATA_EHI_DID_RESET))
3468 return 0;
3469
3470 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3471 ata_for_each_dev(dev, link, ALL) {
3472 if (dev->class == ATA_DEV_UNKNOWN &&
3473 ehc->classes[dev->devno] != ATA_DEV_NONE)
3474 return 0;
3475 }
3476
3477 return 1;
3478 }
3479
ata_count_probe_trials_cb(struct ata_ering_entry * ent,void * void_arg)3480 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3481 {
3482 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3483 u64 now = get_jiffies_64();
3484 int *trials = void_arg;
3485
3486 if (ent->timestamp < now - min(now, interval))
3487 return -1;
3488
3489 (*trials)++;
3490 return 0;
3491 }
3492
ata_eh_schedule_probe(struct ata_device * dev)3493 static int ata_eh_schedule_probe(struct ata_device *dev)
3494 {
3495 struct ata_eh_context *ehc = &dev->link->eh_context;
3496 struct ata_link *link = ata_dev_phys_link(dev);
3497 int trials = 0;
3498
3499 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3500 (ehc->did_probe_mask & (1 << dev->devno)))
3501 return 0;
3502
3503 ata_eh_detach_dev(dev);
3504 ata_dev_init(dev);
3505 ehc->did_probe_mask |= (1 << dev->devno);
3506 ehc->i.action |= ATA_EH_RESET;
3507 ehc->saved_xfer_mode[dev->devno] = 0;
3508 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3509
3510 /* the link maybe in a deep sleep, wake it up */
3511 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3512 if (ata_is_host_link(link))
3513 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3514 ATA_LPM_EMPTY);
3515 else
3516 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3517 ATA_LPM_EMPTY);
3518 }
3519
3520 /* Record and count probe trials on the ering. The specific
3521 * error mask used is irrelevant. Because a successful device
3522 * detection clears the ering, this count accumulates only if
3523 * there are consecutive failed probes.
3524 *
3525 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3526 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3527 * forced to 1.5Gbps.
3528 *
3529 * This is to work around cases where failed link speed
3530 * negotiation results in device misdetection leading to
3531 * infinite DEVXCHG or PHRDY CHG events.
3532 */
3533 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3534 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3535
3536 if (trials > ATA_EH_PROBE_TRIALS)
3537 sata_down_spd_limit(link, 1);
3538
3539 return 1;
3540 }
3541
ata_eh_handle_dev_fail(struct ata_device * dev,int err)3542 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3543 {
3544 struct ata_eh_context *ehc = &dev->link->eh_context;
3545
3546 /* -EAGAIN from EH routine indicates retry without prejudice.
3547 * The requester is responsible for ensuring forward progress.
3548 */
3549 if (err != -EAGAIN)
3550 ehc->tries[dev->devno]--;
3551
3552 switch (err) {
3553 case -ENODEV:
3554 /* device missing or wrong IDENTIFY data, schedule probing */
3555 ehc->i.probe_mask |= (1 << dev->devno);
3556 case -EINVAL:
3557 /* give it just one more chance */
3558 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3559 case -EIO:
3560 if (ehc->tries[dev->devno] == 1) {
3561 /* This is the last chance, better to slow
3562 * down than lose it.
3563 */
3564 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3565 if (dev->pio_mode > XFER_PIO_0)
3566 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3567 }
3568 }
3569
3570 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3571 /* disable device if it has used up all its chances */
3572 ata_dev_disable(dev);
3573
3574 /* detach if offline */
3575 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3576 ata_eh_detach_dev(dev);
3577
3578 /* schedule probe if necessary */
3579 if (ata_eh_schedule_probe(dev)) {
3580 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3581 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3582 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3583 }
3584
3585 return 1;
3586 } else {
3587 ehc->i.action |= ATA_EH_RESET;
3588 return 0;
3589 }
3590 }
3591
3592 /**
3593 * ata_eh_recover - recover host port after error
3594 * @ap: host port to recover
3595 * @prereset: prereset method (can be NULL)
3596 * @softreset: softreset method (can be NULL)
3597 * @hardreset: hardreset method (can be NULL)
3598 * @postreset: postreset method (can be NULL)
3599 * @r_failed_link: out parameter for failed link
3600 *
3601 * This is the alpha and omega, eum and yang, heart and soul of
3602 * libata exception handling. On entry, actions required to
3603 * recover each link and hotplug requests are recorded in the
3604 * link's eh_context. This function executes all the operations
3605 * with appropriate retrials and fallbacks to resurrect failed
3606 * devices, detach goners and greet newcomers.
3607 *
3608 * LOCKING:
3609 * Kernel thread context (may sleep).
3610 *
3611 * RETURNS:
3612 * 0 on success, -errno on failure.
3613 */
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)3614 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3615 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3616 ata_postreset_fn_t postreset,
3617 struct ata_link **r_failed_link)
3618 {
3619 struct ata_link *link;
3620 struct ata_device *dev;
3621 int rc, nr_fails;
3622 unsigned long flags, deadline;
3623
3624 DPRINTK("ENTER\n");
3625
3626 /* prep for recovery */
3627 ata_for_each_link(link, ap, EDGE) {
3628 struct ata_eh_context *ehc = &link->eh_context;
3629
3630 /* re-enable link? */
3631 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3632 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3633 spin_lock_irqsave(ap->lock, flags);
3634 link->flags &= ~ATA_LFLAG_DISABLED;
3635 spin_unlock_irqrestore(ap->lock, flags);
3636 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3637 }
3638
3639 ata_for_each_dev(dev, link, ALL) {
3640 if (link->flags & ATA_LFLAG_NO_RETRY)
3641 ehc->tries[dev->devno] = 1;
3642 else
3643 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3644
3645 /* collect port action mask recorded in dev actions */
3646 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3647 ~ATA_EH_PERDEV_MASK;
3648 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3649
3650 /* process hotplug request */
3651 if (dev->flags & ATA_DFLAG_DETACH)
3652 ata_eh_detach_dev(dev);
3653
3654 /* schedule probe if necessary */
3655 if (!ata_dev_enabled(dev))
3656 ata_eh_schedule_probe(dev);
3657 }
3658 }
3659
3660 retry:
3661 rc = 0;
3662
3663 /* if UNLOADING, finish immediately */
3664 if (ap->pflags & ATA_PFLAG_UNLOADING)
3665 goto out;
3666
3667 /* prep for EH */
3668 ata_for_each_link(link, ap, EDGE) {
3669 struct ata_eh_context *ehc = &link->eh_context;
3670
3671 /* skip EH if possible. */
3672 if (ata_eh_skip_recovery(link))
3673 ehc->i.action = 0;
3674
3675 ata_for_each_dev(dev, link, ALL)
3676 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3677 }
3678
3679 /* reset */
3680 ata_for_each_link(link, ap, EDGE) {
3681 struct ata_eh_context *ehc = &link->eh_context;
3682
3683 if (!(ehc->i.action & ATA_EH_RESET))
3684 continue;
3685
3686 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3687 prereset, softreset, hardreset, postreset);
3688 if (rc) {
3689 ata_link_printk(link, KERN_ERR,
3690 "reset failed, giving up\n");
3691 goto out;
3692 }
3693 }
3694
3695 do {
3696 unsigned long now;
3697
3698 /*
3699 * clears ATA_EH_PARK in eh_info and resets
3700 * ap->park_req_pending
3701 */
3702 ata_eh_pull_park_action(ap);
3703
3704 deadline = jiffies;
3705 ata_for_each_link(link, ap, EDGE) {
3706 ata_for_each_dev(dev, link, ALL) {
3707 struct ata_eh_context *ehc = &link->eh_context;
3708 unsigned long tmp;
3709
3710 if (dev->class != ATA_DEV_ATA)
3711 continue;
3712 if (!(ehc->i.dev_action[dev->devno] &
3713 ATA_EH_PARK))
3714 continue;
3715 tmp = dev->unpark_deadline;
3716 if (time_before(deadline, tmp))
3717 deadline = tmp;
3718 else if (time_before_eq(tmp, jiffies))
3719 continue;
3720 if (ehc->unloaded_mask & (1 << dev->devno))
3721 continue;
3722
3723 ata_eh_park_issue_cmd(dev, 1);
3724 }
3725 }
3726
3727 now = jiffies;
3728 if (time_before_eq(deadline, now))
3729 break;
3730
3731 ata_eh_release(ap);
3732 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3733 deadline - now);
3734 ata_eh_acquire(ap);
3735 } while (deadline);
3736 ata_for_each_link(link, ap, EDGE) {
3737 ata_for_each_dev(dev, link, ALL) {
3738 if (!(link->eh_context.unloaded_mask &
3739 (1 << dev->devno)))
3740 continue;
3741
3742 ata_eh_park_issue_cmd(dev, 0);
3743 ata_eh_done(link, dev, ATA_EH_PARK);
3744 }
3745 }
3746
3747 /* the rest */
3748 nr_fails = 0;
3749 ata_for_each_link(link, ap, PMP_FIRST) {
3750 struct ata_eh_context *ehc = &link->eh_context;
3751
3752 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3753 goto config_lpm;
3754
3755 /* revalidate existing devices and attach new ones */
3756 rc = ata_eh_revalidate_and_attach(link, &dev);
3757 if (rc)
3758 goto rest_fail;
3759
3760 /* if PMP got attached, return, pmp EH will take care of it */
3761 if (link->device->class == ATA_DEV_PMP) {
3762 ehc->i.action = 0;
3763 return 0;
3764 }
3765
3766 /* configure transfer mode if necessary */
3767 if (ehc->i.flags & ATA_EHI_SETMODE) {
3768 rc = ata_set_mode(link, &dev);
3769 if (rc)
3770 goto rest_fail;
3771 ehc->i.flags &= ~ATA_EHI_SETMODE;
3772 }
3773
3774 /* If reset has been issued, clear UA to avoid
3775 * disrupting the current users of the device.
3776 */
3777 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3778 ata_for_each_dev(dev, link, ALL) {
3779 if (dev->class != ATA_DEV_ATAPI)
3780 continue;
3781 rc = atapi_eh_clear_ua(dev);
3782 if (rc)
3783 goto rest_fail;
3784 }
3785 }
3786
3787 /* retry flush if necessary */
3788 ata_for_each_dev(dev, link, ALL) {
3789 if (dev->class != ATA_DEV_ATA)
3790 continue;
3791 rc = ata_eh_maybe_retry_flush(dev);
3792 if (rc)
3793 goto rest_fail;
3794 }
3795
3796 config_lpm:
3797 /* configure link power saving */
3798 if (link->lpm_policy != ap->target_lpm_policy) {
3799 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3800 if (rc)
3801 goto rest_fail;
3802 }
3803
3804 /* this link is okay now */
3805 ehc->i.flags = 0;
3806 continue;
3807
3808 rest_fail:
3809 nr_fails++;
3810 if (dev)
3811 ata_eh_handle_dev_fail(dev, rc);
3812
3813 if (ap->pflags & ATA_PFLAG_FROZEN) {
3814 /* PMP reset requires working host port.
3815 * Can't retry if it's frozen.
3816 */
3817 if (sata_pmp_attached(ap))
3818 goto out;
3819 break;
3820 }
3821 }
3822
3823 if (nr_fails)
3824 goto retry;
3825
3826 out:
3827 if (rc && r_failed_link)
3828 *r_failed_link = link;
3829
3830 DPRINTK("EXIT, rc=%d\n", rc);
3831 return rc;
3832 }
3833
3834 /**
3835 * ata_eh_finish - finish up EH
3836 * @ap: host port to finish EH for
3837 *
3838 * Recovery is complete. Clean up EH states and retry or finish
3839 * failed qcs.
3840 *
3841 * LOCKING:
3842 * None.
3843 */
ata_eh_finish(struct ata_port * ap)3844 void ata_eh_finish(struct ata_port *ap)
3845 {
3846 int tag;
3847
3848 /* retry or finish qcs */
3849 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3850 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3851
3852 if (!(qc->flags & ATA_QCFLAG_FAILED))
3853 continue;
3854
3855 if (qc->err_mask) {
3856 /* FIXME: Once EH migration is complete,
3857 * generate sense data in this function,
3858 * considering both err_mask and tf.
3859 */
3860 if (qc->flags & ATA_QCFLAG_RETRY)
3861 ata_eh_qc_retry(qc);
3862 else
3863 ata_eh_qc_complete(qc);
3864 } else {
3865 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3866 ata_eh_qc_complete(qc);
3867 } else {
3868 /* feed zero TF to sense generation */
3869 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3870 ata_eh_qc_retry(qc);
3871 }
3872 }
3873 }
3874
3875 /* make sure nr_active_links is zero after EH */
3876 WARN_ON(ap->nr_active_links);
3877 ap->nr_active_links = 0;
3878 }
3879
3880 /**
3881 * ata_do_eh - do standard error handling
3882 * @ap: host port to handle error for
3883 *
3884 * @prereset: prereset method (can be NULL)
3885 * @softreset: softreset method (can be NULL)
3886 * @hardreset: hardreset method (can be NULL)
3887 * @postreset: postreset method (can be NULL)
3888 *
3889 * Perform standard error handling sequence.
3890 *
3891 * LOCKING:
3892 * Kernel thread context (may sleep).
3893 */
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)3894 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3895 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3896 ata_postreset_fn_t postreset)
3897 {
3898 struct ata_device *dev;
3899 int rc;
3900
3901 ata_eh_autopsy(ap);
3902 ata_eh_report(ap);
3903
3904 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3905 NULL);
3906 if (rc) {
3907 ata_for_each_dev(dev, &ap->link, ALL)
3908 ata_dev_disable(dev);
3909 }
3910
3911 ata_eh_finish(ap);
3912 }
3913
3914 /**
3915 * ata_std_error_handler - standard error handler
3916 * @ap: host port to handle error for
3917 *
3918 * Standard error handler
3919 *
3920 * LOCKING:
3921 * Kernel thread context (may sleep).
3922 */
ata_std_error_handler(struct ata_port * ap)3923 void ata_std_error_handler(struct ata_port *ap)
3924 {
3925 struct ata_port_operations *ops = ap->ops;
3926 ata_reset_fn_t hardreset = ops->hardreset;
3927
3928 /* ignore built-in hardreset if SCR access is not available */
3929 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3930 hardreset = NULL;
3931
3932 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3933 }
3934
3935 #ifdef CONFIG_PM
3936 /**
3937 * ata_eh_handle_port_suspend - perform port suspend operation
3938 * @ap: port to suspend
3939 *
3940 * Suspend @ap.
3941 *
3942 * LOCKING:
3943 * Kernel thread context (may sleep).
3944 */
ata_eh_handle_port_suspend(struct ata_port * ap)3945 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3946 {
3947 unsigned long flags;
3948 int rc = 0;
3949
3950 /* are we suspending? */
3951 spin_lock_irqsave(ap->lock, flags);
3952 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3953 ap->pm_mesg.event == PM_EVENT_ON) {
3954 spin_unlock_irqrestore(ap->lock, flags);
3955 return;
3956 }
3957 spin_unlock_irqrestore(ap->lock, flags);
3958
3959 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3960
3961 /* tell ACPI we're suspending */
3962 rc = ata_acpi_on_suspend(ap);
3963 if (rc)
3964 goto out;
3965
3966 /* suspend */
3967 ata_eh_freeze_port(ap);
3968
3969 if (ap->ops->port_suspend)
3970 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3971
3972 ata_acpi_set_state(ap, PMSG_SUSPEND);
3973 out:
3974 /* report result */
3975 spin_lock_irqsave(ap->lock, flags);
3976
3977 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3978 if (rc == 0)
3979 ap->pflags |= ATA_PFLAG_SUSPENDED;
3980 else if (ap->pflags & ATA_PFLAG_FROZEN)
3981 ata_port_schedule_eh(ap);
3982
3983 if (ap->pm_result) {
3984 *ap->pm_result = rc;
3985 ap->pm_result = NULL;
3986 }
3987
3988 spin_unlock_irqrestore(ap->lock, flags);
3989
3990 return;
3991 }
3992
3993 /**
3994 * ata_eh_handle_port_resume - perform port resume operation
3995 * @ap: port to resume
3996 *
3997 * Resume @ap.
3998 *
3999 * LOCKING:
4000 * Kernel thread context (may sleep).
4001 */
ata_eh_handle_port_resume(struct ata_port * ap)4002 static void ata_eh_handle_port_resume(struct ata_port *ap)
4003 {
4004 struct ata_link *link;
4005 struct ata_device *dev;
4006 unsigned long flags;
4007 int rc = 0;
4008
4009 /* are we resuming? */
4010 spin_lock_irqsave(ap->lock, flags);
4011 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4012 ap->pm_mesg.event != PM_EVENT_ON) {
4013 spin_unlock_irqrestore(ap->lock, flags);
4014 return;
4015 }
4016 spin_unlock_irqrestore(ap->lock, flags);
4017
4018 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4019
4020 /*
4021 * Error timestamps are in jiffies which doesn't run while
4022 * suspended and PHY events during resume isn't too uncommon.
4023 * When the two are combined, it can lead to unnecessary speed
4024 * downs if the machine is suspended and resumed repeatedly.
4025 * Clear error history.
4026 */
4027 ata_for_each_link(link, ap, HOST_FIRST)
4028 ata_for_each_dev(dev, link, ALL)
4029 ata_ering_clear(&dev->ering);
4030
4031 ata_acpi_set_state(ap, PMSG_ON);
4032
4033 if (ap->ops->port_resume)
4034 rc = ap->ops->port_resume(ap);
4035
4036 /* tell ACPI that we're resuming */
4037 ata_acpi_on_resume(ap);
4038
4039 /* report result */
4040 spin_lock_irqsave(ap->lock, flags);
4041 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4042 if (ap->pm_result) {
4043 *ap->pm_result = rc;
4044 ap->pm_result = NULL;
4045 }
4046 spin_unlock_irqrestore(ap->lock, flags);
4047 }
4048 #endif /* CONFIG_PM */
4049