blob: dad83df555c49d94534a544483e20aa0a9767635 [file] [log] [blame]
Tejun Heoece1d632006-04-02 18:51:53 +09001/*
2 * libata-eh.c - libata error handling
3 *
Tejun Heo8c3d3d42013-05-14 11:09:50 -07004 * Maintained by: Tejun Heo <tj@kernel.org>
Tejun Heoece1d632006-04-02 18:51:53 +09005 * 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
Tejun Heoece1d632006-04-02 18:51:53 +090035#include <linux/kernel.h>
Jens Axboe242f9dc2008-09-14 05:55:09 -070036#include <linux/blkdev.h>
Paul Gortmaker38789fd2011-07-17 15:33:58 -040037#include <linux/export.h>
Jeff Garzik28555682007-10-11 17:12:35 -040038#include <linux/pci.h>
Tejun Heoece1d632006-04-02 18:51:53 +090039#include <scsi/scsi.h>
40#include <scsi/scsi_host.h>
41#include <scsi/scsi_eh.h>
42#include <scsi/scsi_device.h>
43#include <scsi/scsi_cmnd.h>
Robert Hancock65211482009-07-14 20:43:39 -060044#include <scsi/scsi_dbg.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040045#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090046
47#include <linux/libata.h>
48
49#include "libata.h"
50
Tejun Heo7d47e8d2007-02-02 16:22:31 +090051enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090052 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090053 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
54 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
55 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090056 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090057
58 /* error flags */
59 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090060 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Gwendal Grignoud9027472010-05-25 12:31:38 -070061 ATA_EFLAG_OLD_ER = (1 << 31),
Tejun Heo3884f7b2007-11-27 19:28:56 +090062
63 /* error categories */
64 ATA_ECAT_NONE = 0,
65 ATA_ECAT_ATA_BUS = 1,
66 ATA_ECAT_TOUT_HSM = 2,
67 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090068 ATA_ECAT_DUBIOUS_NONE = 4,
69 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
70 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
71 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
72 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090073
Tejun Heo87fbc5a2008-05-20 02:17:54 +090074 ATA_EH_CMD_DFL_TIMEOUT = 5000,
75
Tejun Heo0a2c0f52008-05-20 02:17:52 +090076 /* always put at least this amount of time between resets */
77 ATA_EH_RESET_COOL_DOWN = 5000,
78
Tejun Heo341c2c92008-05-20 02:17:51 +090079 /* Waiting in ->prereset can never be reliable. It's
80 * sometimes nice to wait there but it can't be depended upon;
81 * otherwise, we wouldn't be resetting. Just give it enough
82 * time for most drives to spin up.
83 */
84 ATA_EH_PRERESET_TIMEOUT = 10000,
85 ATA_EH_FASTDRAIN_INTERVAL = 3000,
Tejun Heo11fc33d2008-08-30 14:20:01 +020086
87 ATA_EH_UA_TRIES = 5,
Tejun Heoc2c7a892009-01-29 20:31:34 +090088
89 /* probe speed down parameters, see ata_eh_schedule_probe() */
90 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
91 ATA_EH_PROBE_TRIALS = 2,
Tejun Heo31daabd2007-02-02 16:50:52 +090092};
93
94/* The following table determines how we sequence resets. Each entry
95 * represents timeout for that try. The first try can be soft or
96 * hardreset. All others are hardreset if available. In most cases
97 * the first reset w/ 10sec timeout should succeed. Following entries
Dan Williams35bf8822014-03-07 14:22:48 -080098 * are mostly for error handling, hotplug and those outlier devices that
99 * take an exceptionally long time to recover from reset.
Tejun Heo31daabd2007-02-02 16:50:52 +0900100 */
101static const unsigned long ata_eh_reset_timeouts[] = {
Tejun Heo341c2c92008-05-20 02:17:51 +0900102 10000, /* most drives spin up by 10sec */
103 10000, /* > 99% working drives spin up before 20sec */
Dan Williams35bf8822014-03-07 14:22:48 -0800104 35000, /* give > 30 secs of idleness for outlier devices */
Tejun Heo341c2c92008-05-20 02:17:51 +0900105 5000, /* and sweet one last chance */
Tejun Heod8af0eb2008-05-20 02:17:53 +0900106 ULONG_MAX, /* > 1 min has elapsed, give up */
Tejun Heo31daabd2007-02-02 16:50:52 +0900107};
108
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900109static const unsigned long ata_eh_identify_timeouts[] = {
110 5000, /* covers > 99% of successes and not too boring on failures */
111 10000, /* combined time till here is enough even for media access */
112 30000, /* for true idiots */
113 ULONG_MAX,
114};
115
Tejun Heo6013efd2009-11-19 15:36:45 +0900116static const unsigned long ata_eh_flush_timeouts[] = {
117 15000, /* be generous with flush */
118 15000, /* ditto */
119 30000, /* and even more generous */
120 ULONG_MAX,
121};
122
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900123static const unsigned long ata_eh_other_timeouts[] = {
124 5000, /* same rationale as identify timeout */
125 10000, /* ditto */
126 /* but no merciful 30sec for other commands, it just isn't worth it */
127 ULONG_MAX,
128};
129
130struct ata_eh_cmd_timeout_ent {
131 const u8 *commands;
132 const unsigned long *timeouts;
133};
134
135/* The following table determines timeouts to use for EH internal
136 * commands. Each table entry is a command class and matches the
137 * commands the entry applies to and the timeout table to use.
138 *
139 * On the retry after a command timed out, the next timeout value from
140 * the table is used. If the table doesn't contain further entries,
141 * the last value is used.
142 *
143 * ehc->cmd_timeout_idx keeps track of which timeout to use per
144 * command class, so if SET_FEATURES times out on the first try, the
145 * next try will use the second timeout value only for that class.
146 */
147#define CMDS(cmds...) (const u8 []){ cmds, 0 }
148static const struct ata_eh_cmd_timeout_ent
149ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
150 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
151 .timeouts = ata_eh_identify_timeouts, },
152 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
153 .timeouts = ata_eh_other_timeouts, },
154 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
155 .timeouts = ata_eh_other_timeouts, },
156 { .commands = CMDS(ATA_CMD_SET_FEATURES),
157 .timeouts = ata_eh_other_timeouts, },
158 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
159 .timeouts = ata_eh_other_timeouts, },
Tejun Heo6013efd2009-11-19 15:36:45 +0900160 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
161 .timeouts = ata_eh_flush_timeouts },
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900162};
163#undef CMDS
164
Tejun Heoad9e2762006-05-15 20:58:12 +0900165static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900166#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +0900167static void ata_eh_handle_port_suspend(struct ata_port *ap);
168static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900169#else /* CONFIG_PM */
170static void ata_eh_handle_port_suspend(struct ata_port *ap)
171{ }
172
173static void ata_eh_handle_port_resume(struct ata_port *ap)
174{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900175#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900176
Tejun Heob64bbc32007-07-16 14:29:39 +0900177static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
178 va_list args)
179{
180 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
181 ATA_EH_DESC_LEN - ehi->desc_len,
182 fmt, args);
183}
184
185/**
186 * __ata_ehi_push_desc - push error description without adding separator
187 * @ehi: target EHI
188 * @fmt: printf format string
189 *
190 * Format string according to @fmt and append it to @ehi->desc.
191 *
192 * LOCKING:
193 * spin_lock_irqsave(host lock)
194 */
195void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
196{
197 va_list args;
198
199 va_start(args, fmt);
200 __ata_ehi_pushv_desc(ehi, fmt, args);
201 va_end(args);
202}
203
204/**
205 * ata_ehi_push_desc - push error description with separator
206 * @ehi: target EHI
207 * @fmt: printf format string
208 *
209 * Format string according to @fmt and append it to @ehi->desc.
210 * If @ehi->desc is not empty, ", " is added in-between.
211 *
212 * LOCKING:
213 * spin_lock_irqsave(host lock)
214 */
215void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
216{
217 va_list args;
218
219 if (ehi->desc_len)
220 __ata_ehi_push_desc(ehi, ", ");
221
222 va_start(args, fmt);
223 __ata_ehi_pushv_desc(ehi, fmt, args);
224 va_end(args);
225}
226
227/**
228 * ata_ehi_clear_desc - clean error description
229 * @ehi: target EHI
230 *
231 * Clear @ehi->desc.
232 *
233 * LOCKING:
234 * spin_lock_irqsave(host lock)
235 */
236void ata_ehi_clear_desc(struct ata_eh_info *ehi)
237{
238 ehi->desc[0] = '\0';
239 ehi->desc_len = 0;
240}
241
Tejun Heocbcdd872007-08-18 13:14:55 +0900242/**
243 * ata_port_desc - append port description
244 * @ap: target ATA port
245 * @fmt: printf format string
246 *
247 * Format string according to @fmt and append it to port
248 * description. If port description is not empty, " " is added
249 * in-between. This function is to be used while initializing
250 * ata_host. The description is printed on host registration.
251 *
252 * LOCKING:
253 * None.
254 */
255void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
256{
257 va_list args;
258
259 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
260
261 if (ap->link.eh_info.desc_len)
262 __ata_ehi_push_desc(&ap->link.eh_info, " ");
263
264 va_start(args, fmt);
265 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
266 va_end(args);
267}
268
269#ifdef CONFIG_PCI
270
271/**
272 * ata_port_pbar_desc - append PCI BAR description
273 * @ap: target ATA port
274 * @bar: target PCI BAR
275 * @offset: offset into PCI BAR
276 * @name: name of the area
277 *
278 * If @offset is negative, this function formats a string which
279 * contains the name, address, size and type of the BAR and
280 * appends it to the port description. If @offset is zero or
281 * positive, only name and offsetted address is appended.
282 *
283 * LOCKING:
284 * None.
285 */
286void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
287 const char *name)
288{
289 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
290 char *type = "";
291 unsigned long long start, len;
292
293 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
294 type = "m";
295 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
296 type = "i";
297
298 start = (unsigned long long)pci_resource_start(pdev, bar);
299 len = (unsigned long long)pci_resource_len(pdev, bar);
300
301 if (offset < 0)
302 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
303 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800304 ata_port_desc(ap, "%s 0x%llx", name,
305 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900306}
307
308#endif /* CONFIG_PCI */
309
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900310static int ata_lookup_timeout_table(u8 cmd)
311{
312 int i;
313
314 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
315 const u8 *cur;
316
317 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
318 if (*cur == cmd)
319 return i;
320 }
321
322 return -1;
323}
324
325/**
326 * ata_internal_cmd_timeout - determine timeout for an internal command
327 * @dev: target device
328 * @cmd: internal command to be issued
329 *
330 * Determine timeout for internal command @cmd for @dev.
331 *
332 * LOCKING:
333 * EH context.
334 *
335 * RETURNS:
336 * Determined timeout.
337 */
338unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
339{
340 struct ata_eh_context *ehc = &dev->link->eh_context;
341 int ent = ata_lookup_timeout_table(cmd);
342 int idx;
343
344 if (ent < 0)
345 return ATA_EH_CMD_DFL_TIMEOUT;
346
347 idx = ehc->cmd_timeout_idx[dev->devno][ent];
348 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
349}
350
351/**
352 * ata_internal_cmd_timed_out - notification for internal command timeout
353 * @dev: target device
354 * @cmd: internal command which timed out
355 *
356 * Notify EH that internal command @cmd for @dev timed out. This
357 * function should be called only for commands whose timeouts are
358 * determined using ata_internal_cmd_timeout().
359 *
360 * LOCKING:
361 * EH context.
362 */
363void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
364{
365 struct ata_eh_context *ehc = &dev->link->eh_context;
366 int ent = ata_lookup_timeout_table(cmd);
367 int idx;
368
369 if (ent < 0)
370 return;
371
372 idx = ehc->cmd_timeout_idx[dev->devno][ent];
373 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
374 ehc->cmd_timeout_idx[dev->devno][ent]++;
375}
376
Tejun Heo3884f7b2007-11-27 19:28:56 +0900377static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900378 unsigned int err_mask)
379{
380 struct ata_ering_entry *ent;
381
382 WARN_ON(!err_mask);
383
384 ering->cursor++;
385 ering->cursor %= ATA_ERING_SIZE;
386
387 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900388 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900389 ent->err_mask = err_mask;
390 ent->timestamp = get_jiffies_64();
391}
392
Tejun Heo76326ac2007-11-27 19:28:59 +0900393static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
394{
395 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
396
397 if (ent->err_mask)
398 return ent;
399 return NULL;
400}
401
Gwendal Grignoud9027472010-05-25 12:31:38 -0700402int ata_ering_map(struct ata_ering *ering,
403 int (*map_fn)(struct ata_ering_entry *, void *),
404 void *arg)
Tejun Heo0c247c52006-05-15 20:58:19 +0900405{
406 int idx, rc = 0;
407 struct ata_ering_entry *ent;
408
409 idx = ering->cursor;
410 do {
411 ent = &ering->ring[idx];
412 if (!ent->err_mask)
413 break;
414 rc = map_fn(ent, arg);
415 if (rc)
416 break;
417 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
418 } while (idx != ering->cursor);
419
420 return rc;
421}
422
H Hartley Sweeten60428402012-04-12 15:40:37 -0700423static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
Gwendal Grignoud9027472010-05-25 12:31:38 -0700424{
425 ent->eflags |= ATA_EFLAG_OLD_ER;
426 return 0;
427}
428
429static void ata_ering_clear(struct ata_ering *ering)
430{
431 ata_ering_map(ering, ata_ering_clear_cb, NULL);
432}
433
Tejun Heo64f65ca2006-06-24 20:30:18 +0900434static unsigned int ata_eh_dev_action(struct ata_device *dev)
435{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900436 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900437
438 return ehc->i.action | ehc->i.dev_action[dev->devno];
439}
440
Tejun Heof58229f2007-08-06 18:36:23 +0900441static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900442 struct ata_eh_info *ehi, unsigned int action)
443{
Tejun Heof58229f2007-08-06 18:36:23 +0900444 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900445
446 if (!dev) {
447 ehi->action &= ~action;
Tejun Heo1eca4362008-11-03 20:03:17 +0900448 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900449 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900450 } else {
451 /* doesn't make sense for port-wide EH actions */
452 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
453
454 /* break ehi->action into ehi->dev_action */
455 if (ehi->action & action) {
Tejun Heo1eca4362008-11-03 20:03:17 +0900456 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900457 ehi->dev_action[tdev->devno] |=
458 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900459 ehi->action &= ~action;
460 }
461
462 /* turn off the specified per-dev action */
463 ehi->dev_action[dev->devno] &= ~action;
464 }
465}
466
Tejun Heoece1d632006-04-02 18:51:53 +0900467/**
Tejun Heoc0c362b2010-09-06 17:57:14 +0200468 * ata_eh_acquire - acquire EH ownership
469 * @ap: ATA port to acquire EH ownership for
470 *
471 * Acquire EH ownership for @ap. This is the basic exclusion
472 * mechanism for ports sharing a host. Only one port hanging off
473 * the same host can claim the ownership of EH.
474 *
475 * LOCKING:
476 * EH context.
477 */
478void ata_eh_acquire(struct ata_port *ap)
479{
480 mutex_lock(&ap->host->eh_mutex);
481 WARN_ON_ONCE(ap->host->eh_owner);
482 ap->host->eh_owner = current;
483}
484
485/**
486 * ata_eh_release - release EH ownership
487 * @ap: ATA port to release EH ownership for
488 *
489 * Release EH ownership for @ap if the caller. The caller must
490 * have acquired EH ownership using ata_eh_acquire() previously.
491 *
492 * LOCKING:
493 * EH context.
494 */
495void ata_eh_release(struct ata_port *ap)
496{
497 WARN_ON_ONCE(ap->host->eh_owner != current);
498 ap->host->eh_owner = NULL;
499 mutex_unlock(&ap->host->eh_mutex);
500}
501
502/**
Tejun Heoece1d632006-04-02 18:51:53 +0900503 * ata_scsi_timed_out - SCSI layer time out callback
504 * @cmd: timed out SCSI command
505 *
506 * Handles SCSI layer timeout. We race with normal completion of
507 * the qc for @cmd. If the qc is already gone, we lose and let
508 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
509 * timed out and EH should be invoked. Prevent ata_qc_complete()
510 * from finishing it by setting EH_SCHEDULED and return
511 * EH_NOT_HANDLED.
512 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900513 * TODO: kill this function once old EH is gone.
514 *
Tejun Heoece1d632006-04-02 18:51:53 +0900515 * LOCKING:
516 * Called from timer context
517 *
518 * RETURNS:
519 * EH_HANDLED or EH_NOT_HANDLED
520 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700521enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
Tejun Heoece1d632006-04-02 18:51:53 +0900522{
523 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400524 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900525 unsigned long flags;
526 struct ata_queued_cmd *qc;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700527 enum blk_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900528
529 DPRINTK("ENTER\n");
530
Tejun Heoad9e2762006-05-15 20:58:12 +0900531 if (ap->ops->error_handler) {
Jens Axboe242f9dc2008-09-14 05:55:09 -0700532 ret = BLK_EH_NOT_HANDLED;
Tejun Heoad9e2762006-05-15 20:58:12 +0900533 goto out;
534 }
535
Jens Axboe242f9dc2008-09-14 05:55:09 -0700536 ret = BLK_EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400537 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900538 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900539 if (qc) {
540 WARN_ON(qc->scsicmd != cmd);
541 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
542 qc->err_mask |= AC_ERR_TIMEOUT;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700543 ret = BLK_EH_NOT_HANDLED;
Tejun Heoece1d632006-04-02 18:51:53 +0900544 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400545 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900546
Tejun Heoad9e2762006-05-15 20:58:12 +0900547 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900548 DPRINTK("EXIT, ret=%d\n", ret);
549 return ret;
550}
551
Tejun Heoece180d2008-11-03 20:04:37 +0900552static void ata_eh_unload(struct ata_port *ap)
553{
554 struct ata_link *link;
555 struct ata_device *dev;
556 unsigned long flags;
557
558 /* Restore SControl IPM and SPD for the next driver and
559 * disable attached devices.
560 */
561 ata_for_each_link(link, ap, PMP_FIRST) {
562 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
563 ata_for_each_dev(dev, link, ALL)
564 ata_dev_disable(dev);
565 }
566
567 /* freeze and set UNLOADED */
568 spin_lock_irqsave(ap->lock, flags);
569
570 ata_port_freeze(ap); /* won't be thawed */
571 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
572 ap->pflags |= ATA_PFLAG_UNLOADED;
573
574 spin_unlock_irqrestore(ap->lock, flags);
575}
576
Tejun Heoece1d632006-04-02 18:51:53 +0900577/**
578 * ata_scsi_error - SCSI layer error handler callback
579 * @host: SCSI host on which error occurred
580 *
581 * Handles SCSI-layer-thrown error events.
582 *
583 * LOCKING:
584 * Inherited from SCSI layer (none, can sleep)
585 *
586 * RETURNS:
587 * Zero.
588 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400589void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900590{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400591 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoad9e2762006-05-15 20:58:12 +0900592 unsigned long flags;
James Bottomleyc34aeeb2011-01-23 08:31:14 -0600593 LIST_HEAD(eh_work_q);
Tejun Heoece1d632006-04-02 18:51:53 +0900594
595 DPRINTK("ENTER\n");
596
James Bottomleyc34aeeb2011-01-23 08:31:14 -0600597 spin_lock_irqsave(host->host_lock, flags);
598 list_splice_init(&host->eh_cmd_q, &eh_work_q);
599 spin_unlock_irqrestore(host->host_lock, flags);
600
James Bottomley0e0b494c2011-01-23 09:42:50 -0600601 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
602
603 /* If we timed raced normal completion and there is nothing to
604 recover nr_timedout == 0 why exactly are we doing error recovery ? */
605 ata_scsi_port_error_handler(host, ap);
606
607 /* finish or retry handled scmd's and clean up */
608 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
609
610 DPRINTK("EXIT\n");
611}
612
613/**
614 * ata_scsi_cmd_error_handler - error callback for a list of commands
615 * @host: scsi host containing the port
616 * @ap: ATA port within the host
617 * @eh_work_q: list of commands to process
618 *
619 * process the given list of commands and return those finished to the
620 * ap->eh_done_q. This function is the first part of the libata error
621 * handler which processes a given list of failed commands.
622 */
623void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
624 struct list_head *eh_work_q)
625{
626 int i;
627 unsigned long flags;
628
Tejun Heoc4291372010-05-10 21:41:38 +0200629 /* make sure sff pio task is not running */
630 ata_sff_flush_pio_task(ap);
Tejun Heoece1d632006-04-02 18:51:53 +0900631
Jeff Garzikcca39742006-08-24 03:19:22 -0400632 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900633
Tejun Heoad9e2762006-05-15 20:58:12 +0900634 /* For new EH, all qcs are finished in one of three ways -
635 * normal completion, error completion, and SCSI timeout.
Alan Coxc96f1732009-03-24 10:23:46 +0000636 * Both completions can race against SCSI timeout. When normal
Tejun Heoad9e2762006-05-15 20:58:12 +0900637 * completion wins, the qc never reaches EH. When error
638 * completion wins, the qc has ATA_QCFLAG_FAILED set.
639 *
640 * When SCSI timeout wins, things are a bit more complex.
641 * Normal or error completion can occur after the timeout but
642 * before this point. In such cases, both types of
643 * completions are honored. A scmd is determined to have
644 * timed out iff its associated qc is active and not failed.
645 */
646 if (ap->ops->error_handler) {
647 struct scsi_cmnd *scmd, *tmp;
648 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900649
Tejun Heoe30349d2006-07-03 03:02:15 +0900650 spin_lock_irqsave(ap->lock, flags);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700651
Alan Coxc96f1732009-03-24 10:23:46 +0000652 /* This must occur under the ap->lock as we don't want
653 a polled recovery to race the real interrupt handler
Gwendal Grignoud9027472010-05-25 12:31:38 -0700654
Alan Coxc96f1732009-03-24 10:23:46 +0000655 The lost_interrupt handler checks for any completed but
656 non-notified command and completes much like an IRQ handler.
Gwendal Grignoud9027472010-05-25 12:31:38 -0700657
Alan Coxc96f1732009-03-24 10:23:46 +0000658 We then fall into the error recovery code which will treat
659 this as if normal completion won the race */
Tejun Heoad9e2762006-05-15 20:58:12 +0900660
Alan Coxc96f1732009-03-24 10:23:46 +0000661 if (ap->ops->lost_interrupt)
662 ap->ops->lost_interrupt(ap);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700663
James Bottomley0e0b494c2011-01-23 09:42:50 -0600664 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
Tejun Heoad9e2762006-05-15 20:58:12 +0900665 struct ata_queued_cmd *qc;
666
667 for (i = 0; i < ATA_MAX_QUEUE; i++) {
668 qc = __ata_qc_from_tag(ap, i);
669 if (qc->flags & ATA_QCFLAG_ACTIVE &&
670 qc->scsicmd == scmd)
671 break;
672 }
673
674 if (i < ATA_MAX_QUEUE) {
675 /* the scmd has an associated qc */
676 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
677 /* which hasn't failed yet, timeout */
678 qc->err_mask |= AC_ERR_TIMEOUT;
679 qc->flags |= ATA_QCFLAG_FAILED;
680 nr_timedout++;
681 }
682 } else {
683 /* Normal completion occurred after
684 * SCSI timeout but before this point.
685 * Successfully complete it.
686 */
687 scmd->retries = scmd->allowed;
688 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
689 }
690 }
691
692 /* If we have timed out qcs. They belong to EH from
693 * this point but the state of the controller is
694 * unknown. Freeze the port to make sure the IRQ
695 * handler doesn't diddle with those qcs. This must
696 * be done atomically w.r.t. setting QCFLAG_FAILED.
697 */
698 if (nr_timedout)
699 __ata_port_freeze(ap);
700
Tejun Heoe30349d2006-07-03 03:02:15 +0900701 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900702
703 /* initialize eh_tries */
704 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900705 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900706 spin_unlock_wait(ap->lock);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700707
James Bottomley0e0b494c2011-01-23 09:42:50 -0600708}
709EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
710
711/**
712 * ata_scsi_port_error_handler - recover the port after the commands
713 * @host: SCSI host containing the port
714 * @ap: the ATA port
715 *
716 * Handle the recovery of the port @ap after all the commands
717 * have been recovered.
718 */
719void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
720{
721 unsigned long flags;
Tejun Heoad9e2762006-05-15 20:58:12 +0900722
Tejun Heoad9e2762006-05-15 20:58:12 +0900723 /* invoke error handler */
724 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900725 struct ata_link *link;
726
Tejun Heoc0c362b2010-09-06 17:57:14 +0200727 /* acquire EH ownership */
728 ata_eh_acquire(ap);
729 repeat:
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900730 /* kill fast drain timer */
731 del_timer_sync(&ap->fastdrain_timer);
732
Tejun Heo500530f2006-07-03 16:07:27 +0900733 /* process port resume request */
734 ata_eh_handle_port_resume(ap);
735
Tejun Heof3e81b12006-05-15 20:58:21 +0900736 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900737 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900738
Tejun Heo1eca4362008-11-03 20:03:17 +0900739 ata_for_each_link(link, ap, HOST_FIRST) {
Tejun Heo00115e02007-11-27 19:28:58 +0900740 struct ata_eh_context *ehc = &link->eh_context;
741 struct ata_device *dev;
742
Tejun Heocf1b86c2007-08-06 18:36:23 +0900743 memset(&link->eh_context, 0, sizeof(link->eh_context));
744 link->eh_context.i = link->eh_info;
745 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900746
Tejun Heo1eca4362008-11-03 20:03:17 +0900747 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +0900748 int devno = dev->devno;
749
750 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
751 if (ata_ncq_enabled(dev))
752 ehc->saved_ncq_enabled |= 1 << devno;
753 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900754 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900755
Tejun Heob51e9e52006-06-29 01:29:30 +0900756 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
757 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900758 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900759
Tejun Heoe30349d2006-07-03 03:02:15 +0900760 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900761
Tejun Heo500530f2006-07-03 16:07:27 +0900762 /* invoke EH, skip if unloading or suspended */
763 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900764 ap->ops->error_handler(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900765 else {
766 /* if unloading, commence suicide */
767 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
768 !(ap->pflags & ATA_PFLAG_UNLOADED))
769 ata_eh_unload(ap);
Tejun Heo720ba122006-05-31 18:28:13 +0900770 ata_eh_finish(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900771 }
Tejun Heoad9e2762006-05-15 20:58:12 +0900772
Tejun Heo500530f2006-07-03 16:07:27 +0900773 /* process port suspend request */
774 ata_eh_handle_port_suspend(ap);
775
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300776 /* Exception might have happened after ->error_handler
Tejun Heoad9e2762006-05-15 20:58:12 +0900777 * recovered the port but before this point. Repeat
778 * EH in such case.
779 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900780 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900781
Tejun Heob51e9e52006-06-29 01:29:30 +0900782 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900783 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900784 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900785 goto repeat;
786 }
Joe Perchesa9a79df2011-04-15 15:51:59 -0700787 ata_port_err(ap,
788 "EH pending after %d tries, giving up\n",
789 ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900790 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900791 }
792
Tejun Heof3e81b12006-05-15 20:58:21 +0900793 /* this run is complete, make sure EH info is clear */
Tejun Heo1eca4362008-11-03 20:03:17 +0900794 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heocf1b86c2007-08-06 18:36:23 +0900795 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900796
Dan Williamse4a9c372012-06-21 23:25:27 -0700797 /* end eh (clear host_eh_scheduled) while holding
798 * ap->lock such that if exception occurs after this
799 * point but before EH completion, SCSI midlayer will
Tejun Heoad9e2762006-05-15 20:58:12 +0900800 * re-initiate EH.
801 */
Dan Williamse4a9c372012-06-21 23:25:27 -0700802 ap->ops->end_eh(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900803
Tejun Heoe30349d2006-07-03 03:02:15 +0900804 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc0c362b2010-09-06 17:57:14 +0200805 ata_eh_release(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900806 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900807 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900808 ap->ops->eng_timeout(ap);
809 }
810
Tejun Heoece1d632006-04-02 18:51:53 +0900811 scsi_eh_flush_done_q(&ap->eh_done_q);
812
Tejun Heoad9e2762006-05-15 20:58:12 +0900813 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900814 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900815
Tejun Heo1cdaf532006-07-03 16:07:26 +0900816 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900817 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900818 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
Tejun Heoad72cf92010-07-02 10:03:52 +0200819 schedule_delayed_work(&ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900820
821 if (ap->pflags & ATA_PFLAG_RECOVERED)
Joe Perchesa9a79df2011-04-15 15:51:59 -0700822 ata_port_info(ap, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900823
Tejun Heob51e9e52006-06-29 01:29:30 +0900824 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900825
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900826 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900827 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900828 wake_up_all(&ap->eh_wait_q);
829
Tejun Heoe30349d2006-07-03 03:02:15 +0900830 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900831}
James Bottomley0e0b494c2011-01-23 09:42:50 -0600832EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
Tejun Heoece1d632006-04-02 18:51:53 +0900833
834/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900835 * ata_port_wait_eh - Wait for the currently pending EH to complete
836 * @ap: Port to wait EH for
837 *
838 * Wait until the currently pending EH is complete.
839 *
840 * LOCKING:
841 * Kernel thread context (may sleep).
842 */
843void ata_port_wait_eh(struct ata_port *ap)
844{
845 unsigned long flags;
846 DEFINE_WAIT(wait);
847
848 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400849 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900850
Tejun Heob51e9e52006-06-29 01:29:30 +0900851 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900852 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400853 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900854 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400855 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900856 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900857 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900858
Jeff Garzikba6a1302006-06-22 23:46:10 -0400859 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900860
861 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400862 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heo97750ce2010-09-06 17:56:29 +0200863 ata_msleep(ap, 10);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900864 goto retry;
865 }
866}
Dan Williams81c757b2011-12-02 16:07:01 -0800867EXPORT_SYMBOL_GPL(ata_port_wait_eh);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900868
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900869static int ata_eh_nr_in_flight(struct ata_port *ap)
870{
871 unsigned int tag;
872 int nr = 0;
873
874 /* count only non-internal commands */
875 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
876 if (ata_qc_from_tag(ap, tag))
877 nr++;
878
879 return nr;
880}
881
882void ata_eh_fastdrain_timerfn(unsigned long arg)
883{
884 struct ata_port *ap = (void *)arg;
885 unsigned long flags;
886 int cnt;
887
888 spin_lock_irqsave(ap->lock, flags);
889
890 cnt = ata_eh_nr_in_flight(ap);
891
892 /* are we done? */
893 if (!cnt)
894 goto out_unlock;
895
896 if (cnt == ap->fastdrain_cnt) {
897 unsigned int tag;
898
899 /* No progress during the last interval, tag all
900 * in-flight qcs as timed out and freeze the port.
901 */
902 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
903 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
904 if (qc)
905 qc->err_mask |= AC_ERR_TIMEOUT;
906 }
907
908 ata_port_freeze(ap);
909 } else {
910 /* some qcs have finished, give it another chance */
911 ap->fastdrain_cnt = cnt;
912 ap->fastdrain_timer.expires =
Tejun Heo341c2c92008-05-20 02:17:51 +0900913 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900914 add_timer(&ap->fastdrain_timer);
915 }
916
917 out_unlock:
918 spin_unlock_irqrestore(ap->lock, flags);
919}
920
921/**
922 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
923 * @ap: target ATA port
924 * @fastdrain: activate fast drain
925 *
926 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
927 * is non-zero and EH wasn't pending before. Fast drain ensures
928 * that EH kicks in in timely manner.
929 *
930 * LOCKING:
931 * spin_lock_irqsave(host lock)
932 */
933static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
934{
935 int cnt;
936
937 /* already scheduled? */
938 if (ap->pflags & ATA_PFLAG_EH_PENDING)
939 return;
940
941 ap->pflags |= ATA_PFLAG_EH_PENDING;
942
943 if (!fastdrain)
944 return;
945
946 /* do we have in-flight qcs? */
947 cnt = ata_eh_nr_in_flight(ap);
948 if (!cnt)
949 return;
950
951 /* activate fast drain */
952 ap->fastdrain_cnt = cnt;
Tejun Heo341c2c92008-05-20 02:17:51 +0900953 ap->fastdrain_timer.expires =
954 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900955 add_timer(&ap->fastdrain_timer);
956}
957
Tejun Heof686bcb2006-05-15 20:58:05 +0900958/**
959 * ata_qc_schedule_eh - schedule qc for error handling
960 * @qc: command to schedule error handling for
961 *
962 * Schedule error handling for @qc. EH will kick in as soon as
963 * other commands are drained.
964 *
965 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400966 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900967 */
968void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
969{
970 struct ata_port *ap = qc->ap;
Tejun Heofa41efd2010-04-15 08:57:37 +0900971 struct request_queue *q = qc->scsicmd->device->request_queue;
972 unsigned long flags;
Tejun Heof686bcb2006-05-15 20:58:05 +0900973
974 WARN_ON(!ap->ops->error_handler);
975
976 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900977 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900978
979 /* The following will fail if timeout has already expired.
980 * ata_scsi_error() takes care of such scmds on EH entry.
981 * Note that ATA_QCFLAG_FAILED is unconditionally set after
982 * this function completes.
983 */
Tejun Heofa41efd2010-04-15 08:57:37 +0900984 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700985 blk_abort_request(qc->scsicmd->request);
Tejun Heofa41efd2010-04-15 08:57:37 +0900986 spin_unlock_irqrestore(q->queue_lock, flags);
Tejun Heof686bcb2006-05-15 20:58:05 +0900987}
988
Tejun Heo7b70fc02006-05-15 20:58:07 +0900989/**
Dan Williamse4a9c372012-06-21 23:25:27 -0700990 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
991 * @ap: ATA port to schedule EH for
992 *
993 * LOCKING: inherited from ata_port_schedule_eh
994 * spin_lock_irqsave(host lock)
995 */
996void ata_std_sched_eh(struct ata_port *ap)
997{
998 WARN_ON(!ap->ops->error_handler);
999
1000 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1001 return;
1002
1003 ata_eh_set_pending(ap, 1);
1004 scsi_schedule_eh(ap->scsi_host);
1005
1006 DPRINTK("port EH scheduled\n");
1007}
1008EXPORT_SYMBOL_GPL(ata_std_sched_eh);
1009
1010/**
1011 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1012 * @ap: ATA port to end EH for
1013 *
1014 * In the libata object model there is a 1:1 mapping of ata_port to
1015 * shost, so host fields can be directly manipulated under ap->lock, in
1016 * the libsas case we need to hold a lock at the ha->level to coordinate
1017 * these events.
1018 *
1019 * LOCKING:
1020 * spin_lock_irqsave(host lock)
1021 */
1022void ata_std_end_eh(struct ata_port *ap)
1023{
1024 struct Scsi_Host *host = ap->scsi_host;
1025
1026 host->host_eh_scheduled = 0;
1027}
1028EXPORT_SYMBOL(ata_std_end_eh);
1029
1030
1031/**
Tejun Heo7b70fc02006-05-15 20:58:07 +09001032 * ata_port_schedule_eh - schedule error handling without a qc
1033 * @ap: ATA port to schedule EH for
1034 *
1035 * Schedule error handling for @ap. EH will kick in as soon as
1036 * all commands are drained.
1037 *
1038 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001039 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +09001040 */
1041void ata_port_schedule_eh(struct ata_port *ap)
1042{
Dan Williamse4a9c372012-06-21 23:25:27 -07001043 /* see: ata_std_sched_eh, unless you know better */
1044 ap->ops->sched_eh(ap);
Tejun Heo7b70fc02006-05-15 20:58:07 +09001045}
1046
Tejun Heodbd82612007-08-06 18:36:23 +09001047static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +09001048{
1049 int tag, nr_aborted = 0;
1050
1051 WARN_ON(!ap->ops->error_handler);
1052
Tejun Heo5ddf24c2007-07-16 14:29:41 +09001053 /* we're gonna abort all commands, no need for fast drain */
1054 ata_eh_set_pending(ap, 0);
1055
Tejun Heo7b70fc02006-05-15 20:58:07 +09001056 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1057 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1058
Tejun Heodbd82612007-08-06 18:36:23 +09001059 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +09001060 qc->flags |= ATA_QCFLAG_FAILED;
1061 ata_qc_complete(qc);
1062 nr_aborted++;
1063 }
1064 }
1065
1066 if (!nr_aborted)
1067 ata_port_schedule_eh(ap);
1068
1069 return nr_aborted;
1070}
1071
Tejun Heoe3180492006-05-15 20:58:09 +09001072/**
Tejun Heodbd82612007-08-06 18:36:23 +09001073 * ata_link_abort - abort all qc's on the link
1074 * @link: ATA link to abort qc's for
1075 *
1076 * Abort all active qc's active on @link and schedule EH.
1077 *
1078 * LOCKING:
1079 * spin_lock_irqsave(host lock)
1080 *
1081 * RETURNS:
1082 * Number of aborted qc's.
1083 */
1084int ata_link_abort(struct ata_link *link)
1085{
1086 return ata_do_link_abort(link->ap, link);
1087}
1088
1089/**
1090 * ata_port_abort - abort all qc's on the port
1091 * @ap: ATA port to abort qc's for
1092 *
1093 * Abort all active qc's of @ap and schedule EH.
1094 *
1095 * LOCKING:
1096 * spin_lock_irqsave(host_set lock)
1097 *
1098 * RETURNS:
1099 * Number of aborted qc's.
1100 */
1101int ata_port_abort(struct ata_port *ap)
1102{
1103 return ata_do_link_abort(ap, NULL);
1104}
1105
1106/**
Tejun Heoe3180492006-05-15 20:58:09 +09001107 * __ata_port_freeze - freeze port
1108 * @ap: ATA port to freeze
1109 *
1110 * This function is called when HSM violation or some other
1111 * condition disrupts normal operation of the port. Frozen port
1112 * is not allowed to perform any operation until the port is
1113 * thawed, which usually follows a successful reset.
1114 *
1115 * ap->ops->freeze() callback can be used for freezing the port
1116 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1117 * port cannot be frozen hardware-wise, the interrupt handler
1118 * must ack and clear interrupts unconditionally while the port
1119 * is frozen.
1120 *
1121 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001122 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001123 */
1124static void __ata_port_freeze(struct ata_port *ap)
1125{
1126 WARN_ON(!ap->ops->error_handler);
1127
1128 if (ap->ops->freeze)
1129 ap->ops->freeze(ap);
1130
Tejun Heob51e9e52006-06-29 01:29:30 +09001131 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001132
Tejun Heo44877b42007-02-21 01:06:51 +09001133 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001134}
1135
1136/**
1137 * ata_port_freeze - abort & freeze port
1138 * @ap: ATA port to freeze
1139 *
Jeff Garzik54c38442009-04-07 19:13:15 -04001140 * Abort and freeze @ap. The freeze operation must be called
1141 * first, because some hardware requires special operations
1142 * before the taskfile registers are accessible.
Tejun Heoe3180492006-05-15 20:58:09 +09001143 *
1144 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001145 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001146 *
1147 * RETURNS:
1148 * Number of aborted commands.
1149 */
1150int ata_port_freeze(struct ata_port *ap)
1151{
1152 int nr_aborted;
1153
1154 WARN_ON(!ap->ops->error_handler);
1155
Tejun Heoe3180492006-05-15 20:58:09 +09001156 __ata_port_freeze(ap);
Jeff Garzik54c38442009-04-07 19:13:15 -04001157 nr_aborted = ata_port_abort(ap);
Tejun Heoe3180492006-05-15 20:58:09 +09001158
1159 return nr_aborted;
1160}
1161
1162/**
Tejun Heo7d77b242007-09-23 13:14:13 +09001163 * sata_async_notification - SATA async notification handler
1164 * @ap: ATA port where async notification is received
1165 *
1166 * Handler to be called when async notification via SDB FIS is
1167 * received. This function schedules EH if necessary.
1168 *
1169 * LOCKING:
1170 * spin_lock_irqsave(host lock)
1171 *
1172 * RETURNS:
1173 * 1 if EH is scheduled, 0 otherwise.
1174 */
1175int sata_async_notification(struct ata_port *ap)
1176{
1177 u32 sntf;
1178 int rc;
1179
1180 if (!(ap->flags & ATA_FLAG_AN))
1181 return 0;
1182
1183 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1184 if (rc == 0)
1185 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1186
Tejun Heo071f44b2008-04-07 22:47:22 +09001187 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001188 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +09001189 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001190 /* PMP is not attached. Check whether ATAPI
1191 * AN is configured. If so, notify media
1192 * change.
1193 */
1194 struct ata_device *dev = ap->link.device;
1195
1196 if ((dev->class == ATA_DEV_ATAPI) &&
1197 (dev->flags & ATA_DFLAG_AN))
1198 ata_scsi_media_change_notify(dev);
1199 return 0;
1200 } else {
1201 /* PMP is attached but SNTF is not available.
1202 * ATAPI async media change notification is
1203 * not used. The PMP must be reporting PHY
1204 * status change, schedule EH.
1205 */
1206 ata_port_schedule_eh(ap);
1207 return 1;
1208 }
1209 } else {
1210 /* PMP is attached and SNTF is available */
1211 struct ata_link *link;
1212
1213 /* check and notify ATAPI AN */
Tejun Heo1eca4362008-11-03 20:03:17 +09001214 ata_for_each_link(link, ap, EDGE) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001215 if (!(sntf & (1 << link->pmp)))
1216 continue;
1217
1218 if ((link->device->class == ATA_DEV_ATAPI) &&
1219 (link->device->flags & ATA_DFLAG_AN))
1220 ata_scsi_media_change_notify(link->device);
1221 }
1222
1223 /* If PMP is reporting that PHY status of some
1224 * downstream ports has changed, schedule EH.
1225 */
1226 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1227 ata_port_schedule_eh(ap);
1228 return 1;
1229 }
1230
1231 return 0;
1232 }
1233}
1234
1235/**
Tejun Heoe3180492006-05-15 20:58:09 +09001236 * ata_eh_freeze_port - EH helper to freeze port
1237 * @ap: ATA port to freeze
1238 *
1239 * Freeze @ap.
1240 *
1241 * LOCKING:
1242 * None.
1243 */
1244void ata_eh_freeze_port(struct ata_port *ap)
1245{
1246 unsigned long flags;
1247
1248 if (!ap->ops->error_handler)
1249 return;
1250
Jeff Garzikba6a1302006-06-22 23:46:10 -04001251 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001252 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -04001253 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001254}
1255
1256/**
1257 * ata_port_thaw_port - EH helper to thaw port
1258 * @ap: ATA port to thaw
1259 *
1260 * Thaw frozen port @ap.
1261 *
1262 * LOCKING:
1263 * None.
1264 */
1265void ata_eh_thaw_port(struct ata_port *ap)
1266{
1267 unsigned long flags;
1268
1269 if (!ap->ops->error_handler)
1270 return;
1271
Jeff Garzikba6a1302006-06-22 23:46:10 -04001272 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001273
Tejun Heob51e9e52006-06-29 01:29:30 +09001274 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001275
1276 if (ap->ops->thaw)
1277 ap->ops->thaw(ap);
1278
Jeff Garzikba6a1302006-06-22 23:46:10 -04001279 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001280
Tejun Heo44877b42007-02-21 01:06:51 +09001281 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001282}
1283
Tejun Heoece1d632006-04-02 18:51:53 +09001284static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1285{
1286 /* nada */
1287}
1288
1289static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1290{
1291 struct ata_port *ap = qc->ap;
1292 struct scsi_cmnd *scmd = qc->scsicmd;
1293 unsigned long flags;
1294
Jeff Garzikba6a1302006-06-22 23:46:10 -04001295 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001296 qc->scsidone = ata_eh_scsidone;
1297 __ata_qc_complete(qc);
1298 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -04001299 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001300
1301 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1302}
1303
1304/**
1305 * ata_eh_qc_complete - Complete an active ATA command from EH
1306 * @qc: Command to complete
1307 *
1308 * Indicate to the mid and upper layers that an ATA command has
1309 * completed. To be used from EH.
1310 */
1311void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1312{
1313 struct scsi_cmnd *scmd = qc->scsicmd;
1314 scmd->retries = scmd->allowed;
1315 __ata_eh_qc_complete(qc);
1316}
1317
1318/**
1319 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1320 * @qc: Command to retry
1321 *
1322 * Indicate to the mid and upper layers that an ATA command
1323 * should be retried. To be used from EH.
1324 *
1325 * SCSI midlayer limits the number of retries to scmd->allowed.
Gwendal Grignouf13e2202009-08-07 16:17:49 -07001326 * scmd->allowed is incremented for commands which get retried
Tejun Heoece1d632006-04-02 18:51:53 +09001327 * due to unrelated failures (qc->err_mask is zero).
1328 */
1329void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1330{
1331 struct scsi_cmnd *scmd = qc->scsicmd;
Gwendal Grignouf13e2202009-08-07 16:17:49 -07001332 if (!qc->err_mask)
1333 scmd->allowed++;
Tejun Heoece1d632006-04-02 18:51:53 +09001334 __ata_eh_qc_complete(qc);
1335}
Tejun Heo022bdb02006-05-15 20:58:22 +09001336
1337/**
Tejun Heo678afac2009-01-29 20:31:30 +09001338 * ata_dev_disable - disable ATA device
1339 * @dev: ATA device to disable
1340 *
1341 * Disable @dev.
1342 *
1343 * Locking:
1344 * EH context.
1345 */
1346void ata_dev_disable(struct ata_device *dev)
1347{
1348 if (!ata_dev_enabled(dev))
1349 return;
1350
1351 if (ata_msg_drv(dev->link->ap))
Joe Perchesa9a79df2011-04-15 15:51:59 -07001352 ata_dev_warn(dev, "disabled\n");
Tejun Heo678afac2009-01-29 20:31:30 +09001353 ata_acpi_on_disable(dev);
1354 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1355 dev->class++;
Tejun Heo99cf6102009-01-29 20:31:32 +09001356
1357 /* From now till the next successful probe, ering is used to
1358 * track probe failures. Clear accumulated device error info.
1359 */
1360 ata_ering_clear(&dev->ering);
Tejun Heo678afac2009-01-29 20:31:30 +09001361}
1362
1363/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001364 * ata_eh_detach_dev - detach ATA device
1365 * @dev: ATA device to detach
1366 *
1367 * Detach @dev.
1368 *
1369 * LOCKING:
1370 * None.
1371 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001372void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001373{
Tejun Heof58229f2007-08-06 18:36:23 +09001374 struct ata_link *link = dev->link;
1375 struct ata_port *ap = link->ap;
Tejun Heo90484eb2008-10-26 15:43:03 +09001376 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001377 unsigned long flags;
1378
1379 ata_dev_disable(dev);
1380
Jeff Garzikba6a1302006-06-22 23:46:10 -04001381 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001382
1383 dev->flags &= ~ATA_DFLAG_DETACH;
1384
1385 if (ata_scsi_offline_dev(dev)) {
1386 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001387 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001388 }
1389
Tejun Heo90484eb2008-10-26 15:43:03 +09001390 /* clear per-dev EH info */
Tejun Heof58229f2007-08-06 18:36:23 +09001391 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1392 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heo90484eb2008-10-26 15:43:03 +09001393 ehc->saved_xfer_mode[dev->devno] = 0;
1394 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heobeb07c12006-06-24 20:30:19 +09001395
Jeff Garzikba6a1302006-06-22 23:46:10 -04001396 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001397}
1398
1399/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001400 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001401 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001402 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001403 * @action: action about to be performed
1404 *
1405 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001406 * in @link->eh_info such that eh actions are not unnecessarily
1407 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001408 *
1409 * LOCKING:
1410 * None.
1411 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001412void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1413 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001414{
Tejun Heo955e57d2007-08-06 18:36:23 +09001415 struct ata_port *ap = link->ap;
1416 struct ata_eh_info *ehi = &link->eh_info;
1417 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001418 unsigned long flags;
1419
Jeff Garzikba6a1302006-06-22 23:46:10 -04001420 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001421
Tejun Heo955e57d2007-08-06 18:36:23 +09001422 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001423
Tejun Heoa568d1d2008-10-21 20:37:21 +09001424 /* About to take EH action, set RECOVERED. Ignore actions on
1425 * slave links as master will do them again.
1426 */
1427 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
Tejun Heo1cdaf532006-07-03 16:07:26 +09001428 ap->pflags |= ATA_PFLAG_RECOVERED;
1429
Jeff Garzikba6a1302006-06-22 23:46:10 -04001430 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001431}
1432
1433/**
Tejun Heo47005f22006-06-19 18:27:23 +09001434 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001435* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001436 * @dev: target ATA dev for per-dev action (can be NULL)
1437 * @action: action just completed
1438 *
1439 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001440 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001441 *
1442 * LOCKING:
1443 * None.
1444 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001445void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1446 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001447{
Tejun Heo955e57d2007-08-06 18:36:23 +09001448 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001449
Tejun Heo955e57d2007-08-06 18:36:23 +09001450 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001451}
1452
1453/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001454 * ata_err_string - convert err_mask to descriptive string
1455 * @err_mask: error mask to convert to string
1456 *
1457 * Convert @err_mask to descriptive string. Errors are
1458 * prioritized according to severity and only the most severe
1459 * error is reported.
1460 *
1461 * LOCKING:
1462 * None.
1463 *
1464 * RETURNS:
1465 * Descriptive string for @err_mask
1466 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001467static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001468{
1469 if (err_mask & AC_ERR_HOST_BUS)
1470 return "host bus error";
1471 if (err_mask & AC_ERR_ATA_BUS)
1472 return "ATA bus error";
1473 if (err_mask & AC_ERR_TIMEOUT)
1474 return "timeout";
1475 if (err_mask & AC_ERR_HSM)
1476 return "HSM violation";
1477 if (err_mask & AC_ERR_SYSTEM)
1478 return "internal error";
1479 if (err_mask & AC_ERR_MEDIA)
1480 return "media error";
1481 if (err_mask & AC_ERR_INVALID)
1482 return "invalid argument";
1483 if (err_mask & AC_ERR_DEV)
1484 return "device error";
1485 return "unknown error";
1486}
1487
1488/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001489 * ata_read_log_page - read a specific log page
1490 * @dev: target device
Shane Huang65fe1f02012-09-07 22:40:01 +08001491 * @log: log to read
Tejun Heoe8ee8452006-05-15 21:03:46 +09001492 * @page: page to read
1493 * @buf: buffer to store read page
1494 * @sectors: number of sectors to read
1495 *
1496 * Read log page using READ_LOG_EXT command.
1497 *
1498 * LOCKING:
1499 * Kernel thread context (may sleep).
1500 *
1501 * RETURNS:
1502 * 0 on success, AC_ERR_* mask otherwise.
1503 */
Shane Huang65fe1f02012-09-07 22:40:01 +08001504unsigned int ata_read_log_page(struct ata_device *dev, u8 log,
1505 u8 page, void *buf, unsigned int sectors)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001506{
1507 struct ata_taskfile tf;
1508 unsigned int err_mask;
1509
Shane Huang65fe1f02012-09-07 22:40:01 +08001510 DPRINTK("read log page - log 0x%x, page 0x%x\n", log, page);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001511
1512 ata_tf_init(dev, &tf);
1513 tf.command = ATA_CMD_READ_LOG_EXT;
Shane Huang65fe1f02012-09-07 22:40:01 +08001514 tf.lbal = log;
1515 tf.lbam = page;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001516 tf.nsect = sectors;
1517 tf.hob_nsect = sectors >> 8;
1518 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1519 tf.protocol = ATA_PROT_PIO;
1520
1521 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001522 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001523
1524 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1525 return err_mask;
1526}
1527
1528/**
1529 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1530 * @dev: Device to read log page 10h from
1531 * @tag: Resulting tag of the failed command
1532 * @tf: Resulting taskfile registers of the failed command
1533 *
1534 * Read log page 10h to obtain NCQ error details and clear error
1535 * condition.
1536 *
1537 * LOCKING:
1538 * Kernel thread context (may sleep).
1539 *
1540 * RETURNS:
1541 * 0 on success, -errno otherwise.
1542 */
1543static int ata_eh_read_log_10h(struct ata_device *dev,
1544 int *tag, struct ata_taskfile *tf)
1545{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001546 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001547 unsigned int err_mask;
1548 u8 csum;
1549 int i;
1550
Shane Huang65fe1f02012-09-07 22:40:01 +08001551 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001552 if (err_mask)
1553 return -EIO;
1554
1555 csum = 0;
1556 for (i = 0; i < ATA_SECT_SIZE; i++)
1557 csum += buf[i];
1558 if (csum)
Joe Perchesa9a79df2011-04-15 15:51:59 -07001559 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1560 csum);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001561
1562 if (buf[0] & 0x80)
1563 return -ENOENT;
1564
1565 *tag = buf[0] & 0x1f;
1566
1567 tf->command = buf[2];
1568 tf->feature = buf[3];
1569 tf->lbal = buf[4];
1570 tf->lbam = buf[5];
1571 tf->lbah = buf[6];
1572 tf->device = buf[7];
1573 tf->hob_lbal = buf[8];
1574 tf->hob_lbam = buf[9];
1575 tf->hob_lbah = buf[10];
1576 tf->nsect = buf[12];
1577 tf->hob_nsect = buf[13];
1578
1579 return 0;
1580}
1581
1582/**
Tejun Heo11fc33d2008-08-30 14:20:01 +02001583 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1584 * @dev: target ATAPI device
1585 * @r_sense_key: out parameter for sense_key
1586 *
1587 * Perform ATAPI TEST_UNIT_READY.
1588 *
1589 * LOCKING:
1590 * EH context (may sleep).
1591 *
1592 * RETURNS:
1593 * 0 on success, AC_ERR_* mask on failure.
1594 */
Aaron Lu3dc67442013-01-15 17:21:00 +08001595unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
Tejun Heo11fc33d2008-08-30 14:20:01 +02001596{
1597 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1598 struct ata_taskfile tf;
1599 unsigned int err_mask;
1600
1601 ata_tf_init(dev, &tf);
1602
1603 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1604 tf.command = ATA_CMD_PACKET;
1605 tf.protocol = ATAPI_PROT_NODATA;
1606
1607 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1608 if (err_mask == AC_ERR_DEV)
1609 *r_sense_key = tf.feature >> 4;
1610 return err_mask;
1611}
1612
1613/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001614 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1615 * @dev: device to perform REQUEST_SENSE to
1616 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
Tejun Heo3eabddb2008-06-10 18:28:05 +09001617 * @dfl_sense_key: default sense key to use
Tejun Heo022bdb02006-05-15 20:58:22 +09001618 *
1619 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1620 * SENSE. This function is EH helper.
1621 *
1622 * LOCKING:
1623 * Kernel thread context (may sleep).
1624 *
1625 * RETURNS:
1626 * 0 on success, AC_ERR_* mask on failure
1627 */
Aaron Lu3dc67442013-01-15 17:21:00 +08001628unsigned int atapi_eh_request_sense(struct ata_device *dev,
Tejun Heo3eabddb2008-06-10 18:28:05 +09001629 u8 *sense_buf, u8 dfl_sense_key)
Tejun Heo022bdb02006-05-15 20:58:22 +09001630{
Tejun Heo3eabddb2008-06-10 18:28:05 +09001631 u8 cdb[ATAPI_CDB_LEN] =
1632 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001633 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001634 struct ata_taskfile tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001635
1636 DPRINTK("ATAPI request sense\n");
1637
Tejun Heo022bdb02006-05-15 20:58:22 +09001638 /* FIXME: is this needed? */
1639 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1640
Albert Lee56287762007-04-02 11:30:46 +08001641 /* initialize sense_buf with the error register,
1642 * for the case where they are -not- overwritten
1643 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001644 sense_buf[0] = 0x70;
Tejun Heo3eabddb2008-06-10 18:28:05 +09001645 sense_buf[2] = dfl_sense_key;
Albert Lee56287762007-04-02 11:30:46 +08001646
Jeff Garzika617c092007-05-21 20:14:23 -04001647 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001648 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001649
Tejun Heo022bdb02006-05-15 20:58:22 +09001650 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1651 tf.command = ATA_CMD_PACKET;
1652
1653 /* is it pointless to prefer PIO for "safety reasons"? */
1654 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001655 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001656 tf.feature |= ATAPI_PKT_DMA;
1657 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001658 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001659 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1660 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001661 }
1662
1663 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001664 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001665}
1666
1667/**
1668 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001669 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001670 *
1671 * Analyze SError if available and further determine cause of
1672 * failure.
1673 *
1674 * LOCKING:
1675 * None.
1676 */
Tejun Heo02607312007-08-06 18:36:23 +09001677static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001678{
Tejun Heo02607312007-08-06 18:36:23 +09001679 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001680 u32 serror = ehc->i.serror;
1681 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001682 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001683
Tejun Heoe0614db2008-05-19 01:15:09 +09001684 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001685 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001686 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001687 }
1688 if (serror & SERR_PROTOCOL) {
1689 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001690 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001691 }
1692 if (serror & SERR_INTERNAL) {
1693 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001694 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001695 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001696
1697 /* Determine whether a hotplug event has occurred. Both
1698 * SError.N/X are considered hotplug events for enabled or
1699 * host links. For disabled PMP links, only N bit is
1700 * considered as X bit is left at 1 for link plugging.
1701 */
Tejun Heoeb0e85e2011-02-24 19:30:37 +01001702 if (link->lpm_policy > ATA_LPM_MAX_POWER)
Tejun Heo6b7ae952010-09-01 17:50:06 +02001703 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1704 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
Tejun Heof9df58c2007-09-23 13:14:13 +09001705 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1706 else
1707 hotplug_mask = SERR_PHYRDY_CHG;
1708
1709 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001710 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001711
1712 ehc->i.err_mask |= err_mask;
1713 ehc->i.action |= action;
1714}
1715
1716/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001717 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001718 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001719 *
1720 * Read log page 10h, determine the offending qc and acquire
1721 * error status TF. For NCQ device errors, all LLDDs have to do
1722 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1723 * care of the rest.
1724 *
1725 * LOCKING:
1726 * Kernel thread context (may sleep).
1727 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001728void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001729{
Tejun Heo02607312007-08-06 18:36:23 +09001730 struct ata_port *ap = link->ap;
1731 struct ata_eh_context *ehc = &link->eh_context;
1732 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001733 struct ata_queued_cmd *qc;
1734 struct ata_taskfile tf;
1735 int tag, rc;
1736
1737 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001738 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001739 return;
1740
1741 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001742 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001743 return;
1744
1745 /* has LLDD analyzed already? */
1746 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1747 qc = __ata_qc_from_tag(ap, tag);
1748
1749 if (!(qc->flags & ATA_QCFLAG_FAILED))
1750 continue;
1751
1752 if (qc->err_mask)
1753 return;
1754 }
1755
1756 /* okay, this error is ours */
Jeff Garzika09bf4c2010-04-22 21:59:13 -04001757 memset(&tf, 0, sizeof(tf));
Tejun Heoe8ee8452006-05-15 21:03:46 +09001758 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1759 if (rc) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07001760 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1761 rc);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001762 return;
1763 }
1764
Tejun Heo02607312007-08-06 18:36:23 +09001765 if (!(link->sactive & (1 << tag))) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07001766 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1767 tag);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001768 return;
1769 }
1770
1771 /* we've got the perpetrator, condemn it */
1772 qc = __ata_qc_from_tag(ap, tag);
1773 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001774 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001775 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001776 ehc->i.err_mask &= ~AC_ERR_DEV;
1777}
1778
1779/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001780 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1781 * @qc: qc to analyze
1782 * @tf: Taskfile registers to analyze
1783 *
1784 * Analyze taskfile of @qc and further determine cause of
1785 * failure. This function also requests ATAPI sense data if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001786 * available.
Tejun Heo022bdb02006-05-15 20:58:22 +09001787 *
1788 * LOCKING:
1789 * Kernel thread context (may sleep).
1790 *
1791 * RETURNS:
1792 * Determined recovery action
1793 */
1794static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1795 const struct ata_taskfile *tf)
1796{
1797 unsigned int tmp, action = 0;
1798 u8 stat = tf->command, err = tf->feature;
1799
1800 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1801 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001802 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001803 }
1804
Tejun Heoa51d6442007-03-20 15:24:11 +09001805 if (stat & (ATA_ERR | ATA_DF))
1806 qc->err_mask |= AC_ERR_DEV;
1807 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001808 return 0;
1809
1810 switch (qc->dev->class) {
1811 case ATA_DEV_ATA:
1812 if (err & ATA_ICRC)
1813 qc->err_mask |= AC_ERR_ATA_BUS;
Alexey Asemoveec7e1c2014-07-15 10:28:42 +04001814 if (err & (ATA_UNC | ATA_AMNF))
Tejun Heo022bdb02006-05-15 20:58:22 +09001815 qc->err_mask |= AC_ERR_MEDIA;
1816 if (err & ATA_IDNF)
1817 qc->err_mask |= AC_ERR_INVALID;
1818 break;
1819
1820 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001821 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Tejun Heo3eabddb2008-06-10 18:28:05 +09001822 tmp = atapi_eh_request_sense(qc->dev,
1823 qc->scsicmd->sense_buffer,
1824 qc->result_tf.feature >> 4);
Tejun Heoa569a302006-11-21 10:40:51 +09001825 if (!tmp) {
1826 /* ATA_QCFLAG_SENSE_VALID is used to
1827 * tell atapi_qc_complete() that sense
1828 * data is already valid.
1829 *
1830 * TODO: interpret sense data and set
1831 * appropriate err_mask.
1832 */
1833 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1834 } else
1835 qc->err_mask |= tmp;
1836 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001837 }
1838
1839 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001840 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001841
1842 return action;
1843}
1844
Tejun Heo76326ac2007-11-27 19:28:59 +09001845static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1846 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001847{
Tejun Heo76326ac2007-11-27 19:28:59 +09001848 int base = 0;
1849
1850 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1851 *xfer_ok = 1;
1852
1853 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001854 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001855
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001856 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001857 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001858
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001859 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001860 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001861
Tejun Heo3884f7b2007-11-27 19:28:56 +09001862 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001863 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001864 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001865 if ((err_mask &
1866 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001867 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001868 }
1869
1870 return 0;
1871}
1872
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001873struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001874 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001875 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001876 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001877};
1878
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001879static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001880{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001881 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001882 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001883
Gwendal Grignoud9027472010-05-25 12:31:38 -07001884 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
Tejun Heo022bdb02006-05-15 20:58:22 +09001885 return -1;
1886
Tejun Heo76326ac2007-11-27 19:28:59 +09001887 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1888 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001889 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001890
Tejun Heo022bdb02006-05-15 20:58:22 +09001891 return 0;
1892}
1893
1894/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001895 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001896 * @dev: Device of interest
1897 *
1898 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001899 * whether NCQ needs to be turned off, transfer speed should be
1900 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001901 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001902 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001903 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001904 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1905 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001906 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001907 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001908 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001909 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1910 * data transfer hasn't been verified.
1911 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001912 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001913 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001914 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001915 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001916 * SPEED_DOWN : Speed down transfer speed but don't fall back
1917 * to PIO.
1918 *
1919 * FALLBACK_TO_PIO : Fall back to PIO.
1920 *
1921 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001922 * taken per error. An action triggered by non-DUBIOUS errors
1923 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1924 * This is to expedite speed down decisions right after device is
1925 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001926 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001927 * The followings are speed down rules. #1 and #2 deal with
1928 * DUBIOUS errors.
1929 *
1930 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1931 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1932 *
1933 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1934 * occurred during last 5 mins, NCQ_OFF.
1935 *
1936 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001937 * occurred during last 5 mins, FALLBACK_TO_PIO
Tejun Heo3884f7b2007-11-27 19:28:56 +09001938 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001939 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001940 * during last 10 mins, NCQ_OFF.
1941 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001942 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001943 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001944 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001945 * LOCKING:
1946 * Inherited from caller.
1947 *
1948 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001949 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001950 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001951static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001952{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001953 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1954 u64 j64 = get_jiffies_64();
1955 struct speed_down_verdict_arg arg;
1956 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001957
Tejun Heo3884f7b2007-11-27 19:28:56 +09001958 /* scan past 5 mins of error history */
1959 memset(&arg, 0, sizeof(arg));
1960 arg.since = j64 - min(j64, j5mins);
1961 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1962
Tejun Heo76326ac2007-11-27 19:28:59 +09001963 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1964 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1965 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1966 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1967
1968 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1969 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1970 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1971
Tejun Heo3884f7b2007-11-27 19:28:56 +09001972 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1973 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001974 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001975 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1976
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001977 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001978 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001979 arg.since = j64 - min(j64, j10mins);
1980 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001981
Tejun Heo3884f7b2007-11-27 19:28:56 +09001982 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1983 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001984 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001985
1986 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1987 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001988 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001989 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001990
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001991 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001992}
1993
1994/**
1995 * ata_eh_speed_down - record error and speed down if necessary
1996 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001997 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001998 * @err_mask: err_mask of the error
1999 *
2000 * Record error and examine error history to determine whether
2001 * adjusting transmission speed is necessary. It also sets
2002 * transmission limits appropriately if such adjustment is
2003 * necessary.
2004 *
2005 * LOCKING:
2006 * Kernel thread context (may sleep).
2007 *
2008 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002009 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09002010 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09002011static unsigned int ata_eh_speed_down(struct ata_device *dev,
2012 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002013{
Tejun Heob1c72912008-07-31 17:02:43 +09002014 struct ata_link *link = ata_dev_phys_link(dev);
Tejun Heo76326ac2007-11-27 19:28:59 +09002015 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002016 unsigned int verdict;
2017 unsigned int action = 0;
2018
2019 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09002020 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09002021 return 0;
2022
2023 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09002024 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002025 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09002026
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002027 /* turn off NCQ? */
2028 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2029 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2030 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2031 dev->flags |= ATA_DFLAG_NCQ_OFF;
Joe Perchesa9a79df2011-04-15 15:51:59 -07002032 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002033 goto done;
2034 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002035
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002036 /* speed down? */
2037 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2038 /* speed down SATA link speed if possible */
Tejun Heoa07d4992009-01-29 20:31:33 +09002039 if (sata_down_spd_limit(link, 0) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002040 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002041 goto done;
2042 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002043
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002044 /* lower transfer mode */
2045 if (dev->spdn_cnt < 2) {
2046 static const int dma_dnxfer_sel[] =
2047 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2048 static const int pio_dnxfer_sel[] =
2049 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2050 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09002051
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002052 if (dev->xfer_shift != ATA_SHIFT_PIO)
2053 sel = dma_dnxfer_sel[dev->spdn_cnt];
2054 else
2055 sel = pio_dnxfer_sel[dev->spdn_cnt];
2056
2057 dev->spdn_cnt++;
2058
2059 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002060 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002061 goto done;
2062 }
2063 }
2064 }
2065
2066 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09002067 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002068 */
2069 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09002070 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002071 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2072 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2073 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09002074 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002075 goto done;
2076 }
2077 }
2078
Tejun Heo022bdb02006-05-15 20:58:22 +09002079 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002080 done:
2081 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09002082 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2083 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002084 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09002085}
2086
2087/**
Mark Lord8d899e72012-05-02 15:22:52 -04002088 * ata_eh_worth_retry - analyze error and decide whether to retry
2089 * @qc: qc to possibly retry
2090 *
2091 * Look at the cause of the error and decide if a retry
2092 * might be useful or not. We don't want to retry media errors
2093 * because the drive itself has probably already taken 10-30 seconds
2094 * doing its own internal retries before reporting the failure.
2095 */
2096static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2097{
Bian Yu1eaca392012-12-12 22:26:58 -05002098 if (qc->err_mask & AC_ERR_MEDIA)
Mark Lord8d899e72012-05-02 15:22:52 -04002099 return 0; /* don't retry media errors */
2100 if (qc->flags & ATA_QCFLAG_IO)
2101 return 1; /* otherwise retry anything from fs stack */
2102 if (qc->err_mask & AC_ERR_INVALID)
2103 return 0; /* don't retry these */
2104 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2105}
2106
2107/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002108 * ata_eh_link_autopsy - analyze error and determine recovery action
2109 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09002110 *
Tejun Heo02607312007-08-06 18:36:23 +09002111 * Analyze why @link failed and determine which recovery actions
2112 * are needed. This function also sets more detailed AC_ERR_*
2113 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09002114 *
2115 * LOCKING:
2116 * Kernel thread context (may sleep).
2117 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002118static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002119{
Tejun Heo02607312007-08-06 18:36:23 +09002120 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09002121 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09002122 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002123 unsigned int all_err_mask = 0, eflags = 0;
2124 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09002125 u32 serror;
2126 int rc;
2127
2128 DPRINTK("ENTER\n");
2129
Tejun Heo1cdaf532006-07-03 16:07:26 +09002130 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2131 return;
2132
Tejun Heo022bdb02006-05-15 20:58:22 +09002133 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09002134 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09002135 if (rc == 0) {
2136 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09002137 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09002138 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09002139 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09002140 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09002141 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09002142 ehc->i.err_mask |= AC_ERR_OTHER;
2143 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002144
Tejun Heoe8ee8452006-05-15 21:03:46 +09002145 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09002146 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09002147
Tejun Heo022bdb02006-05-15 20:58:22 +09002148 /* any real error trumps AC_ERR_OTHER */
2149 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2150 ehc->i.err_mask &= ~AC_ERR_OTHER;
2151
2152 all_err_mask |= ehc->i.err_mask;
2153
2154 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2155 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2156
Tejun Heob1c72912008-07-31 17:02:43 +09002157 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2158 ata_dev_phys_link(qc->dev) != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002159 continue;
2160
2161 /* inherit upper level err_mask */
2162 qc->err_mask |= ehc->i.err_mask;
2163
Tejun Heo022bdb02006-05-15 20:58:22 +09002164 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002165 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002166
2167 /* DEV errors are probably spurious in case of ATA_BUS error */
2168 if (qc->err_mask & AC_ERR_ATA_BUS)
2169 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2170 AC_ERR_INVALID);
2171
2172 /* any real error trumps unknown error */
2173 if (qc->err_mask & ~AC_ERR_OTHER)
2174 qc->err_mask &= ~AC_ERR_OTHER;
2175
2176 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09002177 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09002178 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09002179
Tejun Heo03faab72008-03-27 19:14:24 +09002180 /* determine whether the command is worth retrying */
Mark Lord8d899e72012-05-02 15:22:52 -04002181 if (ata_eh_worth_retry(qc))
Tejun Heo03faab72008-03-27 19:14:24 +09002182 qc->flags |= ATA_QCFLAG_RETRY;
2183
Tejun Heo022bdb02006-05-15 20:58:22 +09002184 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002185 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09002186 all_err_mask |= qc->err_mask;
2187 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09002188 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09002189 }
2190
Tejun Heoa20f33f2006-05-16 12:58:24 +09002191 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09002192 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09002193 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09002194 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002195 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2196 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09002197 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002198
Tejun Heodfcc1732007-10-31 10:17:05 +09002199 /* If we have offending qcs and the associated failed device,
2200 * perform per-dev EH action only on the offending device.
2201 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002202 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09002203 ehc->i.dev_action[ehc->i.dev->devno] |=
2204 ehc->i.action & ATA_EH_PERDEV_MASK;
2205 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09002206 }
2207
Tejun Heo2695e362008-01-10 13:41:23 +09002208 /* propagate timeout to host link */
2209 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2210 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2211
2212 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09002213 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09002214 if (!dev && ((ata_link_max_devices(link) == 1 &&
2215 ata_dev_enabled(link->device))))
2216 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09002217
Tejun Heo76326ac2007-11-27 19:28:59 +09002218 if (dev) {
2219 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2220 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002221 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09002222 }
Tejun Heodfcc1732007-10-31 10:17:05 +09002223
Tejun Heo022bdb02006-05-15 20:58:22 +09002224 DPRINTK("EXIT\n");
2225}
2226
2227/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002228 * ata_eh_autopsy - analyze error and determine recovery action
2229 * @ap: host port to perform autopsy on
2230 *
2231 * Analyze all links of @ap and determine why they failed and
2232 * which recovery actions are needed.
2233 *
2234 * LOCKING:
2235 * Kernel thread context (may sleep).
2236 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002237void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002238{
2239 struct ata_link *link;
2240
Tejun Heo1eca4362008-11-03 20:03:17 +09002241 ata_for_each_link(link, ap, EDGE)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002242 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09002243
Tejun Heob1c72912008-07-31 17:02:43 +09002244 /* Handle the frigging slave link. Autopsy is done similarly
2245 * but actions and flags are transferred over to the master
2246 * link and handled from there.
2247 */
2248 if (ap->slave_link) {
2249 struct ata_eh_context *mehc = &ap->link.eh_context;
2250 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2251
Tejun Heo848e4c62008-10-21 14:26:39 +09002252 /* transfer control flags from master to slave */
2253 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2254
2255 /* perform autopsy on the slave link */
Tejun Heob1c72912008-07-31 17:02:43 +09002256 ata_eh_link_autopsy(ap->slave_link);
2257
Tejun Heo848e4c62008-10-21 14:26:39 +09002258 /* transfer actions from slave to master and clear slave */
Tejun Heob1c72912008-07-31 17:02:43 +09002259 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2260 mehc->i.action |= sehc->i.action;
2261 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2262 mehc->i.flags |= sehc->i.flags;
2263 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2264 }
2265
Tejun Heo2695e362008-01-10 13:41:23 +09002266 /* Autopsy of fanout ports can affect host link autopsy.
2267 * Perform host link autopsy last.
2268 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002269 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09002270 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09002271}
2272
2273/**
Robert Hancock65211482009-07-14 20:43:39 -06002274 * ata_get_cmd_descript - get description for ATA command
2275 * @command: ATA command code to get description for
2276 *
2277 * Return a textual description of the given command, or NULL if the
2278 * command is not known.
2279 *
2280 * LOCKING:
2281 * None
2282 */
2283const char *ata_get_cmd_descript(u8 command)
2284{
2285#ifdef CONFIG_ATA_VERBOSE_ERROR
2286 static const struct
2287 {
2288 u8 command;
2289 const char *text;
2290 } cmd_descr[] = {
2291 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2292 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2293 { ATA_CMD_STANDBY, "STANDBY" },
2294 { ATA_CMD_IDLE, "IDLE" },
2295 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2296 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
Robert Hancock3915c3b2013-10-21 19:26:30 -06002297 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
Robert Hancock65211482009-07-14 20:43:39 -06002298 { ATA_CMD_NOP, "NOP" },
2299 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2300 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2301 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2302 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2303 { ATA_CMD_SERVICE, "SERVICE" },
2304 { ATA_CMD_READ, "READ DMA" },
2305 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2306 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2307 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2308 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2309 { ATA_CMD_WRITE, "WRITE DMA" },
2310 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2311 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2312 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2313 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2314 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2315 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2316 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2317 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
Robert Hancock3915c3b2013-10-21 19:26:30 -06002318 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2319 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
Robert Hancock65211482009-07-14 20:43:39 -06002320 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2321 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2322 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2323 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2324 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2325 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2326 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2327 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2328 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2329 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2330 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2331 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2332 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2333 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2334 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2335 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2336 { ATA_CMD_SLEEP, "SLEEP" },
2337 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2338 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2339 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2340 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2341 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2342 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2343 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2344 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2345 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
Robert Hancock3915c3b2013-10-21 19:26:30 -06002346 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
Robert Hancock65211482009-07-14 20:43:39 -06002347 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2348 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2349 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2350 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2351 { ATA_CMD_PMP_READ, "READ BUFFER" },
Robert Hancock3915c3b2013-10-21 19:26:30 -06002352 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
Robert Hancock65211482009-07-14 20:43:39 -06002353 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
Robert Hancock3915c3b2013-10-21 19:26:30 -06002354 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
Robert Hancock65211482009-07-14 20:43:39 -06002355 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2356 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2357 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2358 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2359 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2360 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2361 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2362 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2363 { ATA_CMD_SMART, "SMART" },
2364 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2365 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
FUJITA Tomonoriacad762722010-07-05 15:45:20 +09002366 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
Robert Hancock65211482009-07-14 20:43:39 -06002367 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2368 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2369 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2370 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2371 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2372 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
Robert Hancock3915c3b2013-10-21 19:26:30 -06002373 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2374 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
Robert Hancock65211482009-07-14 20:43:39 -06002375 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2376 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2377 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2378 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2379 { ATA_CMD_RESTORE, "RECALIBRATE" },
2380 { 0, NULL } /* terminate list */
2381 };
2382
2383 unsigned int i;
2384 for (i = 0; cmd_descr[i].text; i++)
2385 if (cmd_descr[i].command == command)
2386 return cmd_descr[i].text;
2387#endif
2388
2389 return NULL;
2390}
2391
2392/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002393 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09002394 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09002395 *
2396 * Report EH to user.
2397 *
2398 * LOCKING:
2399 * None.
2400 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002401static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002402{
Tejun Heo02607312007-08-06 18:36:23 +09002403 struct ata_port *ap = link->ap;
2404 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002405 const char *frozen, *desc;
Levente Kurusa462098b2013-10-29 20:01:46 +01002406 char tries_buf[6] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002407 int tag, nr_failed = 0;
2408
Tejun Heo94ff3d52007-10-09 14:57:56 +09002409 if (ehc->i.flags & ATA_EHI_QUIET)
2410 return;
2411
Tejun Heo022bdb02006-05-15 20:58:22 +09002412 desc = NULL;
2413 if (ehc->i.desc[0] != '\0')
2414 desc = ehc->i.desc;
2415
2416 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2417 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2418
Tejun Heob1c72912008-07-31 17:02:43 +09002419 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2420 ata_dev_phys_link(qc->dev) != link ||
Tejun Heoe027bd32007-10-26 16:19:26 +09002421 ((qc->flags & ATA_QCFLAG_QUIET) &&
2422 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002423 continue;
2424 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2425 continue;
2426
2427 nr_failed++;
2428 }
2429
2430 if (!nr_failed && !ehc->i.err_mask)
2431 return;
2432
2433 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09002434 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09002435 frozen = " frozen";
2436
Tejun Heoa1e10f72007-08-18 13:28:49 +09002437 if (ap->eh_tries < ATA_EH_MAX_TRIES)
Levente Kurusa462098b2013-10-29 20:01:46 +01002438 snprintf(tries_buf, sizeof(tries_buf), " t%d",
Tejun Heoa1e10f72007-08-18 13:28:49 +09002439 ap->eh_tries);
2440
Tejun Heo022bdb02006-05-15 20:58:22 +09002441 if (ehc->i.dev) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002442 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2443 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2444 ehc->i.err_mask, link->sactive, ehc->i.serror,
2445 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002446 if (desc)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002447 ata_dev_err(ehc->i.dev, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002448 } else {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002449 ata_link_err(link, "exception Emask 0x%x "
2450 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2451 ehc->i.err_mask, link->sactive, ehc->i.serror,
2452 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002453 if (desc)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002454 ata_link_err(link, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002455 }
2456
Robert Hancock65211482009-07-14 20:43:39 -06002457#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002458 if (ehc->i.serror)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002459 ata_link_err(link,
Robert Hancock1333e192007-10-02 11:22:02 -04002460 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2461 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2462 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2463 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2464 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2465 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2466 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2467 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2468 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2469 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2470 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2471 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2472 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2473 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2474 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2475 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2476 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002477 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002478#endif
Robert Hancock1333e192007-10-02 11:22:02 -04002479
Tejun Heo022bdb02006-05-15 20:58:22 +09002480 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2481 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09002482 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09002483 const u8 *cdb = qc->cdb;
2484 char data_buf[20] = "";
2485 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002486
Tejun Heo02607312007-08-06 18:36:23 +09002487 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
Tejun Heob1c72912008-07-31 17:02:43 +09002488 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002489 continue;
2490
Tejun Heoabb6a882007-11-28 23:16:09 +09002491 if (qc->dma_dir != DMA_NONE) {
2492 static const char *dma_str[] = {
2493 [DMA_BIDIRECTIONAL] = "bidi",
2494 [DMA_TO_DEVICE] = "out",
2495 [DMA_FROM_DEVICE] = "in",
2496 };
2497 static const char *prot_str[] = {
2498 [ATA_PROT_PIO] = "pio",
2499 [ATA_PROT_DMA] = "dma",
2500 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05002501 [ATAPI_PROT_PIO] = "pio",
2502 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09002503 };
2504
2505 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2506 prot_str[qc->tf.protocol], qc->nbytes,
2507 dma_str[qc->dma_dir]);
2508 }
2509
Robert Hancock65211482009-07-14 20:43:39 -06002510 if (ata_is_atapi(qc->tf.protocol)) {
2511 if (qc->scsicmd)
2512 scsi_print_command(qc->scsicmd);
2513 else
2514 snprintf(cdb_buf, sizeof(cdb_buf),
Tejun Heoabb6a882007-11-28 23:16:09 +09002515 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2516 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2517 cdb[0], cdb[1], cdb[2], cdb[3],
2518 cdb[4], cdb[5], cdb[6], cdb[7],
2519 cdb[8], cdb[9], cdb[10], cdb[11],
2520 cdb[12], cdb[13], cdb[14], cdb[15]);
Robert Hancock65211482009-07-14 20:43:39 -06002521 } else {
2522 const char *descr = ata_get_cmd_descript(cmd->command);
2523 if (descr)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002524 ata_dev_err(qc->dev, "failed command: %s\n",
2525 descr);
Robert Hancock65211482009-07-14 20:43:39 -06002526 }
Tejun Heoabb6a882007-11-28 23:16:09 +09002527
Joe Perchesa9a79df2011-04-15 15:51:59 -07002528 ata_dev_err(qc->dev,
Tejun Heo8a937582006-11-14 22:36:12 +09002529 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002530 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002531 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002532 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002533 cmd->command, cmd->feature, cmd->nsect,
2534 cmd->lbal, cmd->lbam, cmd->lbah,
2535 cmd->hob_feature, cmd->hob_nsect,
2536 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002537 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002538 res->command, res->feature, res->nsect,
2539 res->lbal, res->lbam, res->lbah,
2540 res->hob_feature, res->hob_nsect,
2541 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002542 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2543 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002544
Robert Hancock65211482009-07-14 20:43:39 -06002545#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002546 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002547 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002548 if (res->command & ATA_BUSY)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002549 ata_dev_err(qc->dev, "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002550 else
Joe Perchesa9a79df2011-04-15 15:51:59 -07002551 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
Robert Hancock1333e192007-10-02 11:22:02 -04002552 res->command & ATA_DRDY ? "DRDY " : "",
2553 res->command & ATA_DF ? "DF " : "",
2554 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002555 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002556 }
2557
2558 if (cmd->command != ATA_CMD_PACKET &&
Alexey Asemoveec7e1c2014-07-15 10:28:42 +04002559 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2560 ATA_IDNF | ATA_ABORTED)))
2561 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
Robert Hancock1333e192007-10-02 11:22:02 -04002562 res->feature & ATA_ICRC ? "ICRC " : "",
2563 res->feature & ATA_UNC ? "UNC " : "",
Alexey Asemoveec7e1c2014-07-15 10:28:42 +04002564 res->feature & ATA_AMNF ? "AMNF " : "",
Robert Hancock1333e192007-10-02 11:22:02 -04002565 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002566 res->feature & ATA_ABORTED ? "ABRT " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002567#endif
Tejun Heo022bdb02006-05-15 20:58:22 +09002568 }
2569}
2570
Tejun Heo9b1e2652007-08-06 18:36:24 +09002571/**
2572 * ata_eh_report - report error handling to user
2573 * @ap: ATA port to report EH about
2574 *
2575 * Report EH to user.
2576 *
2577 * LOCKING:
2578 * None.
2579 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002580void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002581{
2582 struct ata_link *link;
2583
Tejun Heo1eca4362008-11-03 20:03:17 +09002584 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002585 ata_eh_link_report(link);
2586}
2587
Tejun Heocc0680a2007-08-06 18:36:23 +09002588static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heob1c72912008-07-31 17:02:43 +09002589 unsigned int *classes, unsigned long deadline,
2590 bool clear_classes)
Tejun Heod87fa382006-05-31 18:28:24 +09002591{
Tejun Heof58229f2007-08-06 18:36:23 +09002592 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002593
Tejun Heob1c72912008-07-31 17:02:43 +09002594 if (clear_classes)
Tejun Heo1eca4362008-11-03 20:03:17 +09002595 ata_for_each_dev(dev, link, ALL)
Tejun Heob1c72912008-07-31 17:02:43 +09002596 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002597
Tejun Heof0465192008-05-19 01:15:08 +09002598 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002599}
2600
Sergei Shtylyove8411fb2011-08-09 23:25:50 +04002601static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
Tejun Heo664faf02006-05-31 18:27:50 +09002602{
Tejun Heo45db2f62008-04-08 01:46:56 +09002603 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002604 return 0;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002605 if (rc == -EAGAIN)
2606 return 1;
Tejun Heo071f44b2008-04-07 22:47:22 +09002607 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002608 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002609 return 0;
2610}
2611
Tejun Heofb7fd612007-09-23 13:14:12 +09002612int ata_eh_reset(struct ata_link *link, int classify,
2613 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2614 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002615{
Tejun Heoafaa5c32007-10-09 15:06:10 +09002616 struct ata_port *ap = link->ap;
Tejun Heob1c72912008-07-31 17:02:43 +09002617 struct ata_link *slave = ap->slave_link;
Tejun Heo936fd732007-08-06 18:36:23 +09002618 struct ata_eh_context *ehc = &link->eh_context;
Bartlomiej Zolnierkiewicz705d2012009-07-26 16:21:01 +02002619 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
Tejun Heo664faf02006-05-31 18:27:50 +09002620 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002621 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002622 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heod8af0eb2008-05-20 02:17:53 +09002623 int max_tries = 0, try = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002624 struct ata_link *failed_link;
Tejun Heof58229f2007-08-06 18:36:23 +09002625 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002626 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002627 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002628 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002629 u32 sstatus;
Tejun Heob1c72912008-07-31 17:02:43 +09002630 int nr_unknown, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002631
Tejun Heo932648b2008-05-19 01:15:06 +09002632 /*
2633 * Prepare to reset
2634 */
Tejun Heod8af0eb2008-05-20 02:17:53 +09002635 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2636 max_tries++;
Dan Williamsca6d43b2012-06-21 23:41:41 -07002637 if (link->flags & ATA_LFLAG_RST_ONCE)
2638 max_tries = 1;
Tejun Heo05944bd2008-08-13 20:19:09 +09002639 if (link->flags & ATA_LFLAG_NO_HRST)
2640 hardreset = NULL;
2641 if (link->flags & ATA_LFLAG_NO_SRST)
2642 softreset = NULL;
Tejun Heod8af0eb2008-05-20 02:17:53 +09002643
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002644 /* make sure each reset attempt is at least COOL_DOWN apart */
Tejun Heo19b72322008-11-04 17:08:40 +09002645 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2646 now = jiffies;
2647 WARN_ON(time_after(ehc->last_reset, now));
2648 deadline = ata_deadline(ehc->last_reset,
2649 ATA_EH_RESET_COOL_DOWN);
2650 if (time_before(now, deadline))
2651 schedule_timeout_uninterruptible(deadline - now);
2652 }
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002653
Tejun Heoafaa5c32007-10-09 15:06:10 +09002654 spin_lock_irqsave(ap->lock, flags);
2655 ap->pflags |= ATA_PFLAG_RESETTING;
2656 spin_unlock_irqrestore(ap->lock, flags);
2657
Tejun Heocf480622008-01-24 00:05:14 +09002658 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002659
Tejun Heo1eca4362008-11-03 20:03:17 +09002660 ata_for_each_dev(dev, link, ALL) {
Tejun Heocdeab112007-10-29 16:41:09 +09002661 /* If we issue an SRST then an ATA drive (not ATAPI)
2662 * may change configuration and be in PIO0 timing. If
2663 * we do a hard reset (or are coming from power on)
2664 * this is true for ATA or ATAPI. Until we've set a
2665 * suitable controller mode we should not touch the
2666 * bus as we may be talking too fast.
2667 */
2668 dev->pio_mode = XFER_PIO_0;
Aaron Lu54169122012-12-03 11:35:02 +08002669 dev->dma_mode = 0xff;
Tejun Heocdeab112007-10-29 16:41:09 +09002670
2671 /* If the controller has a pio mode setup function
2672 * then use it to set the chipset to rights. Don't
2673 * touch the DMA setup as that will be dealt with when
2674 * configuring devices.
2675 */
2676 if (ap->ops->set_piomode)
2677 ap->ops->set_piomode(ap, dev);
2678 }
2679
Tejun Heocf480622008-01-24 00:05:14 +09002680 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002681 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002682 ehc->i.action &= ~ATA_EH_RESET;
2683 if (hardreset) {
2684 reset = hardreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002685 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002686 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002687 reset = softreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002688 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heocf480622008-01-24 00:05:14 +09002689 }
Tejun Heof5914a42006-05-31 18:27:48 +09002690
2691 if (prereset) {
Tejun Heob1c72912008-07-31 17:02:43 +09002692 unsigned long deadline = ata_deadline(jiffies,
2693 ATA_EH_PRERESET_TIMEOUT);
2694
2695 if (slave) {
2696 sehc->i.action &= ~ATA_EH_RESET;
2697 sehc->i.action |= ehc->i.action;
2698 }
2699
2700 rc = prereset(link, deadline);
2701
2702 /* If present, do prereset on slave link too. Reset
2703 * is skipped iff both master and slave links report
2704 * -ENOENT or clear ATA_EH_RESET.
2705 */
2706 if (slave && (rc == 0 || rc == -ENOENT)) {
2707 int tmp;
2708
2709 tmp = prereset(slave, deadline);
2710 if (tmp != -ENOENT)
2711 rc = tmp;
2712
2713 ehc->i.action |= sehc->i.action;
2714 }
2715
Tejun Heof5914a42006-05-31 18:27:48 +09002716 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002717 if (rc == -ENOENT) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002718 ata_link_dbg(link, "port disabled--ignoring\n");
Tejun Heocf480622008-01-24 00:05:14 +09002719 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002720
Tejun Heo1eca4362008-11-03 20:03:17 +09002721 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09002722 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002723
2724 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002725 } else
Joe Perchesa9a79df2011-04-15 15:51:59 -07002726 ata_link_err(link,
2727 "prereset failed (errno=%d)\n",
2728 rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002729 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002730 }
Tejun Heof5914a42006-05-31 18:27:48 +09002731
Tejun Heo932648b2008-05-19 01:15:06 +09002732 /* prereset() might have cleared ATA_EH_RESET. If so,
Tejun Heod6515e62009-03-04 15:59:30 +09002733 * bang classes, thaw and return.
Tejun Heo932648b2008-05-19 01:15:06 +09002734 */
2735 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
Tejun Heo1eca4362008-11-03 20:03:17 +09002736 ata_for_each_dev(dev, link, ALL)
Tejun Heo932648b2008-05-19 01:15:06 +09002737 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heod6515e62009-03-04 15:59:30 +09002738 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2739 ata_is_host_link(link))
2740 ata_eh_thaw_port(ap);
Tejun Heo932648b2008-05-19 01:15:06 +09002741 rc = 0;
2742 goto out;
2743 }
Tejun Heof5914a42006-05-31 18:27:48 +09002744 }
2745
Tejun Heo022bdb02006-05-15 20:58:22 +09002746 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002747 /*
2748 * Perform reset
2749 */
Tejun Heodc98c322008-05-19 01:15:07 +09002750 if (ata_is_host_link(link))
2751 ata_eh_freeze_port(ap);
2752
Tejun Heo341c2c92008-05-20 02:17:51 +09002753 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
Tejun Heo31daabd2007-02-02 16:50:52 +09002754
Tejun Heo932648b2008-05-19 01:15:06 +09002755 if (reset) {
2756 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002757 ata_link_info(link, "%s resetting link\n",
2758 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002759
Tejun Heo932648b2008-05-19 01:15:06 +09002760 /* mark that this EH session started with reset */
Tejun Heo19b72322008-11-04 17:08:40 +09002761 ehc->last_reset = jiffies;
Tejun Heo932648b2008-05-19 01:15:06 +09002762 if (reset == hardreset)
2763 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2764 else
2765 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002766
Tejun Heob1c72912008-07-31 17:02:43 +09002767 rc = ata_do_reset(link, reset, classes, deadline, true);
2768 if (rc && rc != -EAGAIN) {
2769 failed_link = link;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002770 goto fail;
Tejun Heob1c72912008-07-31 17:02:43 +09002771 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002772
Tejun Heob1c72912008-07-31 17:02:43 +09002773 /* hardreset slave link if existent */
2774 if (slave && reset == hardreset) {
2775 int tmp;
2776
2777 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002778 ata_link_info(slave, "hard resetting link\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002779
2780 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2781 tmp = ata_do_reset(slave, reset, classes, deadline,
2782 false);
2783 switch (tmp) {
2784 case -EAGAIN:
2785 rc = -EAGAIN;
2786 case 0:
2787 break;
2788 default:
2789 failed_link = slave;
2790 rc = tmp;
2791 goto fail;
2792 }
2793 }
2794
2795 /* perform follow-up SRST if necessary */
Tejun Heo932648b2008-05-19 01:15:06 +09002796 if (reset == hardreset &&
Sergei Shtylyove8411fb2011-08-09 23:25:50 +04002797 ata_eh_followup_srst_needed(link, rc)) {
Tejun Heo932648b2008-05-19 01:15:06 +09002798 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002799
Tejun Heo932648b2008-05-19 01:15:06 +09002800 if (!reset) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002801 ata_link_err(link,
2802 "follow-up softreset required but no softreset available\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002803 failed_link = link;
Tejun Heo932648b2008-05-19 01:15:06 +09002804 rc = -EINVAL;
2805 goto fail;
2806 }
2807
2808 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002809 rc = ata_do_reset(link, reset, classes, deadline, true);
Tejun Heofe2c4d02009-07-08 12:16:37 +09002810 if (rc) {
2811 failed_link = link;
2812 goto fail;
2813 }
Tejun Heo664faf02006-05-31 18:27:50 +09002814 }
Tejun Heo932648b2008-05-19 01:15:06 +09002815 } else {
2816 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002817 ata_link_info(link,
2818 "no reset method available, skipping reset\n");
Tejun Heo932648b2008-05-19 01:15:06 +09002819 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2820 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002821 }
2822
Tejun Heo932648b2008-05-19 01:15:06 +09002823 /*
2824 * Post-reset processing
2825 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002826 ata_for_each_dev(dev, link, ALL) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002827 /* After the reset, the device state is PIO 0 and the
2828 * controller state is undefined. Reset also wakes up
2829 * drives from sleeping mode.
2830 */
2831 dev->pio_mode = XFER_PIO_0;
2832 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002833
Tejun Heo3b761d32009-10-06 17:08:40 +09002834 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2835 continue;
2836
2837 /* apply class override */
2838 if (lflags & ATA_LFLAG_ASSUME_ATA)
2839 classes[dev->devno] = ATA_DEV_ATA;
2840 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2841 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
Tejun Heo022bdb02006-05-15 20:58:22 +09002842 }
2843
Tejun Heo416dc9e2007-10-31 10:17:03 +09002844 /* record current link speed */
2845 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2846 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heob1c72912008-07-31 17:02:43 +09002847 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2848 slave->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002849
Tejun Heodc98c322008-05-19 01:15:07 +09002850 /* thaw the port */
2851 if (ata_is_host_link(link))
2852 ata_eh_thaw_port(ap);
2853
Tejun Heof0465192008-05-19 01:15:08 +09002854 /* postreset() should clear hardware SError. Although SError
2855 * is cleared during link resume, clearing SError here is
2856 * necessary as some PHYs raise hotplug events after SRST.
2857 * This introduces race condition where hotplug occurs between
2858 * reset and here. This race is mediated by cross checking
2859 * link onlineness and classification result later.
2860 */
Tejun Heob1c72912008-07-31 17:02:43 +09002861 if (postreset) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002862 postreset(link, classes);
Tejun Heob1c72912008-07-31 17:02:43 +09002863 if (slave)
2864 postreset(slave, classes);
2865 }
Tejun Heo20952b62006-05-31 18:27:23 +09002866
Tejun Heo1e641062009-07-17 11:58:33 +09002867 /*
Tejun Heo8c56cac2011-05-25 13:19:39 +02002868 * Some controllers can't be frozen very well and may set spurious
2869 * error conditions during reset. Clear accumulated error
2870 * information and re-thaw the port if frozen. As reset is the
2871 * final recovery action and we cross check link onlineness against
2872 * device classification later, no hotplug event is lost by this.
Tejun Heo1e641062009-07-17 11:58:33 +09002873 */
Tejun Heof0465192008-05-19 01:15:08 +09002874 spin_lock_irqsave(link->ap->lock, flags);
Tejun Heo1e641062009-07-17 11:58:33 +09002875 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heob1c72912008-07-31 17:02:43 +09002876 if (slave)
Tejun Heo1e641062009-07-17 11:58:33 +09002877 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2878 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof0465192008-05-19 01:15:08 +09002879 spin_unlock_irqrestore(link->ap->lock, flags);
2880
Tejun Heo8c56cac2011-05-25 13:19:39 +02002881 if (ap->pflags & ATA_PFLAG_FROZEN)
2882 ata_eh_thaw_port(ap);
2883
Tejun Heo3b761d32009-10-06 17:08:40 +09002884 /*
2885 * Make sure onlineness and classification result correspond.
Tejun Heof0465192008-05-19 01:15:08 +09002886 * Hotplug could have happened during reset and some
2887 * controllers fail to wait while a drive is spinning up after
2888 * being hotplugged causing misdetection. By cross checking
Tejun Heo3b761d32009-10-06 17:08:40 +09002889 * link on/offlineness and classification result, those
2890 * conditions can be reliably detected and retried.
Tejun Heof0465192008-05-19 01:15:08 +09002891 */
Tejun Heob1c72912008-07-31 17:02:43 +09002892 nr_unknown = 0;
Tejun Heo1eca4362008-11-03 20:03:17 +09002893 ata_for_each_dev(dev, link, ALL) {
Tejun Heo3b761d32009-10-06 17:08:40 +09002894 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2895 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002896 ata_dev_dbg(dev, "link online but device misclassified\n");
Tejun Heo3b761d32009-10-06 17:08:40 +09002897 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002898 nr_unknown++;
Tejun Heo3b761d32009-10-06 17:08:40 +09002899 }
2900 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2901 if (ata_class_enabled(classes[dev->devno]))
Joe Perchesa9a79df2011-04-15 15:51:59 -07002902 ata_dev_dbg(dev,
2903 "link offline, clearing class %d to NONE\n",
2904 classes[dev->devno]);
Tejun Heo3b761d32009-10-06 17:08:40 +09002905 classes[dev->devno] = ATA_DEV_NONE;
2906 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002907 ata_dev_dbg(dev,
2908 "link status unknown, clearing UNKNOWN to NONE\n");
Tejun Heo3b761d32009-10-06 17:08:40 +09002909 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002910 }
Tejun Heof0465192008-05-19 01:15:08 +09002911 }
2912
Tejun Heob1c72912008-07-31 17:02:43 +09002913 if (classify && nr_unknown) {
Tejun Heof0465192008-05-19 01:15:08 +09002914 if (try < max_tries) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002915 ata_link_warn(link,
2916 "link online but %d devices misclassified, retrying\n",
2917 nr_unknown);
Tejun Heob1c72912008-07-31 17:02:43 +09002918 failed_link = link;
Tejun Heof0465192008-05-19 01:15:08 +09002919 rc = -EAGAIN;
2920 goto fail;
2921 }
Joe Perchesa9a79df2011-04-15 15:51:59 -07002922 ata_link_warn(link,
2923 "link online but %d devices misclassified, "
2924 "device detection might fail\n", nr_unknown);
Tejun Heof0465192008-05-19 01:15:08 +09002925 }
2926
Tejun Heo416dc9e2007-10-31 10:17:03 +09002927 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002928 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002929 if (slave)
2930 ata_eh_done(slave, NULL, ATA_EH_RESET);
Tejun Heo6b7ae952010-09-01 17:50:06 +02002931 ehc->last_reset = jiffies; /* update to completion time */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002932 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo6b7ae952010-09-01 17:50:06 +02002933 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
Tejun Heoae791c02007-09-23 13:14:12 +09002934
Tejun Heo416dc9e2007-10-31 10:17:03 +09002935 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002936 out:
2937 /* clear hotplug flag */
2938 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heob1c72912008-07-31 17:02:43 +09002939 if (slave)
2940 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002941
2942 spin_lock_irqsave(ap->lock, flags);
2943 ap->pflags &= ~ATA_PFLAG_RESETTING;
2944 spin_unlock_irqrestore(ap->lock, flags);
2945
Tejun Heo022bdb02006-05-15 20:58:22 +09002946 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002947
2948 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002949 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2950 if (!ata_is_host_link(link) &&
2951 sata_scr_read(link, SCR_STATUS, &sstatus))
2952 rc = -ERESTART;
2953
Gwendal Grignou7a46c072011-10-19 17:17:02 -07002954 if (try >= max_tries) {
Tejun Heo8ea76452011-05-25 12:23:56 +02002955 /*
2956 * Thaw host port even if reset failed, so that the port
2957 * can be retried on the next phy event. This risks
2958 * repeated EH runs but seems to be a better tradeoff than
2959 * shutting down a port after a botched hotplug attempt.
2960 */
2961 if (ata_is_host_link(link))
2962 ata_eh_thaw_port(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002963 goto out;
Tejun Heo8ea76452011-05-25 12:23:56 +02002964 }
Tejun Heo416dc9e2007-10-31 10:17:03 +09002965
2966 now = jiffies;
2967 if (time_before(now, deadline)) {
2968 unsigned long delta = deadline - now;
2969
Joe Perchesa9a79df2011-04-15 15:51:59 -07002970 ata_link_warn(failed_link,
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002971 "reset failed (errno=%d), retrying in %u secs\n",
2972 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
Tejun Heo416dc9e2007-10-31 10:17:03 +09002973
Tejun Heoc0c362b2010-09-06 17:57:14 +02002974 ata_eh_release(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002975 while (delta)
2976 delta = schedule_timeout_uninterruptible(delta);
Tejun Heoc0c362b2010-09-06 17:57:14 +02002977 ata_eh_acquire(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002978 }
2979
Gwendal Grignou7a46c072011-10-19 17:17:02 -07002980 /*
2981 * While disks spinup behind PMP, some controllers fail sending SRST.
2982 * They need to be reset - as well as the PMP - before retrying.
2983 */
2984 if (rc == -ERESTART) {
2985 if (ata_is_host_link(link))
2986 ata_eh_thaw_port(ap);
2987 goto out;
2988 }
2989
Tejun Heob1c72912008-07-31 17:02:43 +09002990 if (try == max_tries - 1) {
Tejun Heoa07d4992009-01-29 20:31:33 +09002991 sata_down_spd_limit(link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002992 if (slave)
Tejun Heoa07d4992009-01-29 20:31:33 +09002993 sata_down_spd_limit(slave, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002994 } else if (rc == -EPIPE)
Tejun Heoa07d4992009-01-29 20:31:33 +09002995 sata_down_spd_limit(failed_link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002996
Tejun Heo416dc9e2007-10-31 10:17:03 +09002997 if (hardreset)
2998 reset = hardreset;
2999 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003000}
3001
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003002static inline void ata_eh_pull_park_action(struct ata_port *ap)
3003{
3004 struct ata_link *link;
3005 struct ata_device *dev;
3006 unsigned long flags;
3007
3008 /*
3009 * This function can be thought of as an extended version of
3010 * ata_eh_about_to_do() specially crafted to accommodate the
3011 * requirements of ATA_EH_PARK handling. Since the EH thread
3012 * does not leave the do {} while () loop in ata_eh_recover as
3013 * long as the timeout for a park request to *one* device on
3014 * the port has not expired, and since we still want to pick
3015 * up park requests to other devices on the same port or
3016 * timeout updates for the same device, we have to pull
3017 * ATA_EH_PARK actions from eh_info into eh_context.i
3018 * ourselves at the beginning of each pass over the loop.
3019 *
3020 * Additionally, all write accesses to &ap->park_req_pending
Wolfram Sang16735d02013-11-14 14:32:02 -08003021 * through reinit_completion() (see below) or complete_all()
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003022 * (see ata_scsi_park_store()) are protected by the host lock.
3023 * As a result we have that park_req_pending.done is zero on
3024 * exit from this function, i.e. when ATA_EH_PARK actions for
3025 * *all* devices on port ap have been pulled into the
3026 * respective eh_context structs. If, and only if,
3027 * park_req_pending.done is non-zero by the time we reach
3028 * wait_for_completion_timeout(), another ATA_EH_PARK action
3029 * has been scheduled for at least one of the devices on port
3030 * ap and we have to cycle over the do {} while () loop in
3031 * ata_eh_recover() again.
3032 */
3033
3034 spin_lock_irqsave(ap->lock, flags);
Wolfram Sang16735d02013-11-14 14:32:02 -08003035 reinit_completion(&ap->park_req_pending);
Tejun Heo1eca4362008-11-03 20:03:17 +09003036 ata_for_each_link(link, ap, EDGE) {
3037 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003038 struct ata_eh_info *ehi = &link->eh_info;
3039
3040 link->eh_context.i.dev_action[dev->devno] |=
3041 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3042 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3043 }
3044 }
3045 spin_unlock_irqrestore(ap->lock, flags);
3046}
3047
3048static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3049{
3050 struct ata_eh_context *ehc = &dev->link->eh_context;
3051 struct ata_taskfile tf;
3052 unsigned int err_mask;
3053
3054 ata_tf_init(dev, &tf);
3055 if (park) {
3056 ehc->unloaded_mask |= 1 << dev->devno;
3057 tf.command = ATA_CMD_IDLEIMMEDIATE;
3058 tf.feature = 0x44;
3059 tf.lbal = 0x4c;
3060 tf.lbam = 0x4e;
3061 tf.lbah = 0x55;
3062 } else {
3063 ehc->unloaded_mask &= ~(1 << dev->devno);
3064 tf.command = ATA_CMD_CHK_POWER;
3065 }
3066
3067 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3068 tf.protocol |= ATA_PROT_NODATA;
3069 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3070 if (park && (err_mask || tf.lbal != 0xc4)) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003071 ata_dev_err(dev, "head unload failed!\n");
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003072 ehc->unloaded_mask &= ~(1 << dev->devno);
3073 }
3074}
3075
Tejun Heo02607312007-08-06 18:36:23 +09003076static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09003077 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09003078{
Tejun Heo02607312007-08-06 18:36:23 +09003079 struct ata_port *ap = link->ap;
3080 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003081 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003082 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003083 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09003084 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003085
3086 DPRINTK("ENTER\n");
3087
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003088 /* For PATA drive side cable detection to work, IDENTIFY must
3089 * be done backwards such that PDIAG- is released by the slave
3090 * device before the master device is identified.
3091 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003092 ata_for_each_dev(dev, link, ALL_REVERSE) {
Tejun Heof58229f2007-08-06 18:36:23 +09003093 unsigned int action = ata_eh_dev_action(dev);
3094 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09003095
Tejun Heobff04642006-11-10 18:08:10 +09003096 if (ehc->i.flags & ATA_EHI_DID_RESET)
3097 readid_flags |= ATA_READID_POSTRESET;
3098
Tejun Heo9666f402007-05-04 21:27:47 +02003099 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a32007-09-23 13:19:54 +09003100 WARN_ON(dev->class == ATA_DEV_PMP);
3101
Tejun Heob1c72912008-07-31 17:02:43 +09003102 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
Tejun Heo022bdb02006-05-15 20:58:22 +09003103 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003104 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003105 }
3106
Tejun Heo02607312007-08-06 18:36:23 +09003107 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09003108 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3109 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09003110 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003111 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003112
Tejun Heo02607312007-08-06 18:36:23 +09003113 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09003114
Tejun Heobaa1e782006-11-01 18:39:27 +09003115 /* Configuration may have changed, reconfigure
3116 * transfer mode.
3117 */
3118 ehc->i.flags |= ATA_EHI_SETMODE;
3119
zhao, forrest3057ac32006-06-12 12:01:34 +08003120 /* schedule the scsi_rescan_device() here */
Tejun Heoad72cf92010-07-02 10:03:52 +02003121 schedule_work(&(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09003122 } else if (dev->class == ATA_DEV_UNKNOWN &&
3123 ehc->tries[dev->devno] &&
3124 ata_class_enabled(ehc->classes[dev->devno])) {
Tejun Heo842faa62009-05-10 01:06:54 +09003125 /* Temporarily set dev->class, it will be
3126 * permanently set once all configurations are
3127 * complete. This is necessary because new
3128 * device configuration is done in two
3129 * separate loops.
3130 */
Tejun Heo084fe632006-05-31 18:28:03 +09003131 dev->class = ehc->classes[dev->devno];
3132
Tejun Heo633273a32007-09-23 13:19:54 +09003133 if (dev->class == ATA_DEV_PMP)
3134 rc = sata_pmp_attach(dev);
3135 else
3136 rc = ata_dev_read_id(dev, &dev->class,
3137 readid_flags, dev->id);
Tejun Heo842faa62009-05-10 01:06:54 +09003138
3139 /* read_id might have changed class, store and reset */
3140 ehc->classes[dev->devno] = dev->class;
3141 dev->class = ATA_DEV_UNKNOWN;
3142
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003143 switch (rc) {
3144 case 0:
Tejun Heo99cf6102009-01-29 20:31:32 +09003145 /* clear error info accumulated during probe */
3146 ata_ering_clear(&dev->ering);
Tejun Heof58229f2007-08-06 18:36:23 +09003147 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003148 break;
3149 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003150 /* IDENTIFY was issued to non-existent
3151 * device. No need to reset. Just
Tejun Heo842faa62009-05-10 01:06:54 +09003152 * thaw and ignore the device.
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003153 */
3154 ata_eh_thaw_port(ap);
Tejun Heo084fe632006-05-31 18:28:03 +09003155 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003156 default:
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003157 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003158 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003159 }
3160 }
3161
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003162 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09003163 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3164 if (ap->ops->cable_detect)
3165 ap->cbl = ap->ops->cable_detect(ap);
3166 ata_force_cbl(ap);
3167 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003168
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003169 /* Configure new devices forward such that user doesn't see
3170 * device detection messages backwards.
3171 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003172 ata_for_each_dev(dev, link, ALL) {
Tejun Heo4f7c2872009-10-15 23:37:32 +09003173 if (!(new_mask & (1 << dev->devno)))
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003174 continue;
3175
Tejun Heo842faa62009-05-10 01:06:54 +09003176 dev->class = ehc->classes[dev->devno];
3177
Tejun Heo4f7c2872009-10-15 23:37:32 +09003178 if (dev->class == ATA_DEV_PMP)
3179 continue;
3180
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003181 ehc->i.flags |= ATA_EHI_PRINTINFO;
3182 rc = ata_dev_configure(dev);
3183 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
Tejun Heo842faa62009-05-10 01:06:54 +09003184 if (rc) {
3185 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003186 goto err;
Tejun Heo842faa62009-05-10 01:06:54 +09003187 }
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003188
3189 spin_lock_irqsave(ap->lock, flags);
3190 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3191 spin_unlock_irqrestore(ap->lock, flags);
3192
3193 /* new device discovered, configure xfermode */
3194 ehc->i.flags |= ATA_EHI_SETMODE;
3195 }
3196
3197 return 0;
3198
3199 err:
3200 *r_failed_dev = dev;
3201 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003202 return rc;
3203}
3204
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003205/**
3206 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3207 * @link: link on which timings will be programmed
Martin Olsson98a17082009-04-22 18:21:29 +02003208 * @r_failed_dev: out parameter for failed device
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003209 *
3210 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3211 * ata_set_mode() fails, pointer to the failing device is
3212 * returned in @r_failed_dev.
3213 *
3214 * LOCKING:
3215 * PCI/etc. bus probe sem.
3216 *
3217 * RETURNS:
3218 * 0 on success, negative errno otherwise
3219 */
3220int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3221{
3222 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09003223 struct ata_device *dev;
3224 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003225
Tejun Heo76326ac2007-11-27 19:28:59 +09003226 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
Tejun Heo1eca4362008-11-03 20:03:17 +09003227 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo76326ac2007-11-27 19:28:59 +09003228 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3229 struct ata_ering_entry *ent;
3230
3231 ent = ata_ering_top(&dev->ering);
3232 if (ent)
3233 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3234 }
3235 }
3236
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003237 /* has private set_mode? */
3238 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09003239 rc = ap->ops->set_mode(link, r_failed_dev);
3240 else
3241 rc = ata_do_set_mode(link, r_failed_dev);
3242
3243 /* if transfer mode has changed, set DUBIOUS_XFER on device */
Tejun Heo1eca4362008-11-03 20:03:17 +09003244 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +09003245 struct ata_eh_context *ehc = &link->eh_context;
3246 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3247 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3248
3249 if (dev->xfer_mode != saved_xfer_mode ||
3250 ata_ncq_enabled(dev) != saved_ncq)
3251 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3252 }
3253
3254 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003255}
3256
Tejun Heo11fc33d2008-08-30 14:20:01 +02003257/**
3258 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3259 * @dev: ATAPI device to clear UA for
3260 *
3261 * Resets and other operations can make an ATAPI device raise
3262 * UNIT ATTENTION which causes the next operation to fail. This
3263 * function clears UA.
3264 *
3265 * LOCKING:
3266 * EH context (may sleep).
3267 *
3268 * RETURNS:
3269 * 0 on success, -errno on failure.
3270 */
3271static int atapi_eh_clear_ua(struct ata_device *dev)
3272{
3273 int i;
3274
3275 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
Tejun Heob5357082009-03-02 18:55:16 +09003276 u8 *sense_buffer = dev->link->ap->sector_buf;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003277 u8 sense_key = 0;
3278 unsigned int err_mask;
3279
3280 err_mask = atapi_eh_tur(dev, &sense_key);
3281 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003282 ata_dev_warn(dev,
3283 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3284 err_mask);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003285 return -EIO;
3286 }
3287
3288 if (!err_mask || sense_key != UNIT_ATTENTION)
3289 return 0;
3290
3291 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3292 if (err_mask) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003293 ata_dev_warn(dev, "failed to clear "
Tejun Heo11fc33d2008-08-30 14:20:01 +02003294 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3295 return -EIO;
3296 }
3297 }
3298
Joe Perchesa9a79df2011-04-15 15:51:59 -07003299 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3300 ATA_EH_UA_TRIES);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003301
3302 return 0;
3303}
3304
Tejun Heo6013efd2009-11-19 15:36:45 +09003305/**
3306 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3307 * @dev: ATA device which may need FLUSH retry
3308 *
3309 * If @dev failed FLUSH, it needs to be reported upper layer
3310 * immediately as it means that @dev failed to remap and already
3311 * lost at least a sector and further FLUSH retrials won't make
3312 * any difference to the lost sector. However, if FLUSH failed
3313 * for other reasons, for example transmission error, FLUSH needs
3314 * to be retried.
3315 *
3316 * This function determines whether FLUSH failure retry is
3317 * necessary and performs it if so.
3318 *
3319 * RETURNS:
3320 * 0 if EH can continue, -errno if EH needs to be repeated.
3321 */
3322static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3323{
3324 struct ata_link *link = dev->link;
3325 struct ata_port *ap = link->ap;
3326 struct ata_queued_cmd *qc;
3327 struct ata_taskfile tf;
3328 unsigned int err_mask;
3329 int rc = 0;
3330
3331 /* did flush fail for this device? */
3332 if (!ata_tag_valid(link->active_tag))
3333 return 0;
3334
3335 qc = __ata_qc_from_tag(ap, link->active_tag);
3336 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3337 qc->tf.command != ATA_CMD_FLUSH))
3338 return 0;
3339
3340 /* if the device failed it, it should be reported to upper layers */
3341 if (qc->err_mask & AC_ERR_DEV)
3342 return 0;
3343
3344 /* flush failed for some other reason, give it another shot */
3345 ata_tf_init(dev, &tf);
3346
3347 tf.command = qc->tf.command;
3348 tf.flags |= ATA_TFLAG_DEVICE;
3349 tf.protocol = ATA_PROT_NODATA;
3350
Joe Perchesa9a79df2011-04-15 15:51:59 -07003351 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
Tejun Heo6013efd2009-11-19 15:36:45 +09003352 tf.command, qc->err_mask);
3353
3354 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3355 if (!err_mask) {
3356 /*
3357 * FLUSH is complete but there's no way to
3358 * successfully complete a failed command from EH.
3359 * Making sure retry is allowed at least once and
3360 * retrying it should do the trick - whatever was in
3361 * the cache is already on the platter and this won't
3362 * cause infinite loop.
3363 */
3364 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3365 } else {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003366 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
Tejun Heo6013efd2009-11-19 15:36:45 +09003367 err_mask);
3368 rc = -EIO;
3369
3370 /* if device failed it, report it to upper layers */
3371 if (err_mask & AC_ERR_DEV) {
3372 qc->err_mask |= AC_ERR_DEV;
3373 qc->result_tf = tf;
3374 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3375 rc = 0;
3376 }
3377 }
3378 return rc;
3379}
3380
Tejun Heo6b7ae952010-09-01 17:50:06 +02003381/**
3382 * ata_eh_set_lpm - configure SATA interface power management
3383 * @link: link to configure power management
3384 * @policy: the link power management policy
3385 * @r_failed_dev: out parameter for failed device
3386 *
3387 * Enable SATA Interface power management. This will enable
3388 * Device Interface Power Management (DIPM) for min_power
3389 * policy, and then call driver specific callbacks for
3390 * enabling Host Initiated Power management.
3391 *
3392 * LOCKING:
3393 * EH context.
3394 *
3395 * RETURNS:
3396 * 0 on success, -errno on failure.
3397 */
3398static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3399 struct ata_device **r_failed_dev)
3400{
Tejun Heo6c8ea892010-09-01 17:50:07 +02003401 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003402 struct ata_eh_context *ehc = &link->eh_context;
3403 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
Tejun Heoe5005b12010-12-09 16:13:15 +01003404 enum ata_lpm_policy old_policy = link->lpm_policy;
Tejun Heo5f6f12c2011-05-09 16:04:11 +02003405 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003406 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3407 unsigned int err_mask;
3408 int rc;
3409
3410 /* if the link or host doesn't do LPM, noop */
3411 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3412 return 0;
3413
3414 /*
3415 * DIPM is enabled only for MIN_POWER as some devices
3416 * misbehave when the host NACKs transition to SLUMBER. Order
3417 * device and link configurations such that the host always
3418 * allows DIPM requests.
3419 */
3420 ata_for_each_dev(dev, link, ENABLED) {
3421 bool hipm = ata_id_has_hipm(dev->id);
Tejun Heoae01b242011-03-16 11:14:55 +01003422 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003423
3424 /* find the first enabled and LPM enabled devices */
3425 if (!link_dev)
3426 link_dev = dev;
3427
3428 if (!lpm_dev && (hipm || dipm))
3429 lpm_dev = dev;
3430
3431 hints &= ~ATA_LPM_EMPTY;
3432 if (!hipm)
3433 hints &= ~ATA_LPM_HIPM;
3434
3435 /* disable DIPM before changing link config */
3436 if (policy != ATA_LPM_MIN_POWER && dipm) {
3437 err_mask = ata_dev_set_feature(dev,
3438 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3439 if (err_mask && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003440 ata_dev_warn(dev,
3441 "failed to disable DIPM, Emask 0x%x\n",
3442 err_mask);
Tejun Heo6b7ae952010-09-01 17:50:06 +02003443 rc = -EIO;
3444 goto fail;
3445 }
3446 }
3447 }
3448
Tejun Heo6c8ea892010-09-01 17:50:07 +02003449 if (ap) {
3450 rc = ap->ops->set_lpm(link, policy, hints);
3451 if (!rc && ap->slave_link)
3452 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3453 } else
3454 rc = sata_pmp_set_lpm(link, policy, hints);
Tejun Heo6b7ae952010-09-01 17:50:06 +02003455
3456 /*
3457 * Attribute link config failure to the first (LPM) enabled
3458 * device on the link.
3459 */
3460 if (rc) {
3461 if (rc == -EOPNOTSUPP) {
3462 link->flags |= ATA_LFLAG_NO_LPM;
3463 return 0;
3464 }
3465 dev = lpm_dev ? lpm_dev : link_dev;
3466 goto fail;
3467 }
3468
Tejun Heoe5005b12010-12-09 16:13:15 +01003469 /*
3470 * Low level driver acked the transition. Issue DIPM command
3471 * with the new policy set.
3472 */
3473 link->lpm_policy = policy;
3474 if (ap && ap->slave_link)
3475 ap->slave_link->lpm_policy = policy;
3476
Tejun Heo6b7ae952010-09-01 17:50:06 +02003477 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3478 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heoae01b242011-03-16 11:14:55 +01003479 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3480 ata_id_has_dipm(dev->id)) {
Tejun Heo6b7ae952010-09-01 17:50:06 +02003481 err_mask = ata_dev_set_feature(dev,
3482 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3483 if (err_mask && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003484 ata_dev_warn(dev,
Tejun Heo6b7ae952010-09-01 17:50:06 +02003485 "failed to enable DIPM, Emask 0x%x\n",
3486 err_mask);
3487 rc = -EIO;
3488 goto fail;
3489 }
3490 }
3491 }
3492
Tejun Heo6b7ae952010-09-01 17:50:06 +02003493 return 0;
3494
3495fail:
Tejun Heoe5005b12010-12-09 16:13:15 +01003496 /* restore the old policy */
3497 link->lpm_policy = old_policy;
3498 if (ap && ap->slave_link)
3499 ap->slave_link->lpm_policy = old_policy;
3500
Tejun Heo6b7ae952010-09-01 17:50:06 +02003501 /* if no device or only one more chance is left, disable LPM */
3502 if (!dev || ehc->tries[dev->devno] <= 2) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003503 ata_link_warn(link, "disabling LPM on the link\n");
Tejun Heo6b7ae952010-09-01 17:50:06 +02003504 link->flags |= ATA_LFLAG_NO_LPM;
3505 }
3506 if (r_failed_dev)
3507 *r_failed_dev = dev;
3508 return rc;
3509}
3510
Kristen Carlson Accardi8a745f12011-03-04 10:24:11 -08003511int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003512{
Tejun Heof58229f2007-08-06 18:36:23 +09003513 struct ata_device *dev;
3514 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003515
Tejun Heo1eca4362008-11-03 20:03:17 +09003516 ata_for_each_dev(dev, link, ENABLED)
3517 cnt++;
Tejun Heo022bdb02006-05-15 20:58:22 +09003518 return cnt;
3519}
3520
Tejun Heo02607312007-08-06 18:36:23 +09003521static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003522{
Tejun Heof58229f2007-08-06 18:36:23 +09003523 struct ata_device *dev;
3524 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003525
Tejun Heo1eca4362008-11-03 20:03:17 +09003526 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09003527 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09003528 cnt++;
3529 return cnt;
3530}
3531
Tejun Heo02607312007-08-06 18:36:23 +09003532static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003533{
Tejun Heo672b2d62008-01-24 00:05:14 +09003534 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09003535 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09003536 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09003537
Tejun Heof9df58c2007-09-23 13:14:13 +09003538 /* skip disabled links */
3539 if (link->flags & ATA_LFLAG_DISABLED)
3540 return 1;
3541
Tejun Heoe2f3d752010-09-07 14:05:31 +02003542 /* skip if explicitly requested */
3543 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3544 return 1;
3545
Tejun Heo672b2d62008-01-24 00:05:14 +09003546 /* thaw frozen port and recover failed devices */
3547 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3548 return 0;
3549
3550 /* reset at least once if reset is requested */
3551 if ((ehc->i.action & ATA_EH_RESET) &&
3552 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09003553 return 0;
3554
3555 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo1eca4362008-11-03 20:03:17 +09003556 ata_for_each_dev(dev, link, ALL) {
Tejun Heo084fe632006-05-31 18:28:03 +09003557 if (dev->class == ATA_DEV_UNKNOWN &&
3558 ehc->classes[dev->devno] != ATA_DEV_NONE)
3559 return 0;
3560 }
3561
3562 return 1;
3563}
3564
Tejun Heoc2c7a892009-01-29 20:31:34 +09003565static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3566{
3567 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3568 u64 now = get_jiffies_64();
3569 int *trials = void_arg;
3570
Lin Ming68682252012-05-03 22:15:07 +08003571 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3572 (ent->timestamp < now - min(now, interval)))
Tejun Heoc2c7a892009-01-29 20:31:34 +09003573 return -1;
3574
3575 (*trials)++;
3576 return 0;
3577}
3578
Tejun Heo02c05a22007-11-27 19:28:54 +09003579static int ata_eh_schedule_probe(struct ata_device *dev)
3580{
3581 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heoc2c7a892009-01-29 20:31:34 +09003582 struct ata_link *link = ata_dev_phys_link(dev);
3583 int trials = 0;
Tejun Heo02c05a22007-11-27 19:28:54 +09003584
3585 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3586 (ehc->did_probe_mask & (1 << dev->devno)))
3587 return 0;
3588
3589 ata_eh_detach_dev(dev);
3590 ata_dev_init(dev);
3591 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09003592 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09003593 ehc->saved_xfer_mode[dev->devno] = 0;
3594 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09003595
Tejun Heo6b7ae952010-09-01 17:50:06 +02003596 /* the link maybe in a deep sleep, wake it up */
Tejun Heo6c8ea892010-09-01 17:50:07 +02003597 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3598 if (ata_is_host_link(link))
3599 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3600 ATA_LPM_EMPTY);
3601 else
3602 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3603 ATA_LPM_EMPTY);
3604 }
Tejun Heo6b7ae952010-09-01 17:50:06 +02003605
Tejun Heoc2c7a892009-01-29 20:31:34 +09003606 /* Record and count probe trials on the ering. The specific
3607 * error mask used is irrelevant. Because a successful device
3608 * detection clears the ering, this count accumulates only if
3609 * there are consecutive failed probes.
3610 *
3611 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3612 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3613 * forced to 1.5Gbps.
3614 *
3615 * This is to work around cases where failed link speed
3616 * negotiation results in device misdetection leading to
3617 * infinite DEVXCHG or PHRDY CHG events.
3618 */
3619 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3620 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3621
3622 if (trials > ATA_EH_PROBE_TRIALS)
3623 sata_down_spd_limit(link, 1);
3624
Tejun Heo02c05a22007-11-27 19:28:54 +09003625 return 1;
3626}
3627
Tejun Heo9b1e2652007-08-06 18:36:24 +09003628static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09003629{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09003630 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09003631
Tejun Heocf9a5902009-01-29 20:31:35 +09003632 /* -EAGAIN from EH routine indicates retry without prejudice.
3633 * The requester is responsible for ensuring forward progress.
3634 */
3635 if (err != -EAGAIN)
3636 ehc->tries[dev->devno]--;
Tejun Heofee7ca72007-07-01 19:05:58 +09003637
3638 switch (err) {
3639 case -ENODEV:
3640 /* device missing or wrong IDENTIFY data, schedule probing */
3641 ehc->i.probe_mask |= (1 << dev->devno);
3642 case -EINVAL:
3643 /* give it just one more chance */
3644 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3645 case -EIO:
Tejun Heod89293a2009-01-29 20:31:29 +09003646 if (ehc->tries[dev->devno] == 1) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003647 /* This is the last chance, better to slow
3648 * down than lose it.
3649 */
Tejun Heoa07d4992009-01-29 20:31:33 +09003650 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
Tejun Heod89293a2009-01-29 20:31:29 +09003651 if (dev->pio_mode > XFER_PIO_0)
3652 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
Tejun Heofee7ca72007-07-01 19:05:58 +09003653 }
3654 }
3655
3656 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3657 /* disable device if it has used up all its chances */
3658 ata_dev_disable(dev);
3659
3660 /* detach if offline */
Tejun Heob1c72912008-07-31 17:02:43 +09003661 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
Tejun Heofee7ca72007-07-01 19:05:58 +09003662 ata_eh_detach_dev(dev);
3663
Tejun Heo02c05a22007-11-27 19:28:54 +09003664 /* schedule probe if necessary */
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003665 if (ata_eh_schedule_probe(dev)) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003666 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003667 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3668 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3669 }
Tejun Heo9b1e2652007-08-06 18:36:24 +09003670
3671 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09003672 } else {
Tejun Heocf480622008-01-24 00:05:14 +09003673 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003674 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09003675 }
3676}
3677
Tejun Heo022bdb02006-05-15 20:58:22 +09003678/**
3679 * ata_eh_recover - recover host port after error
3680 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09003681 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003682 * @softreset: softreset method (can be NULL)
3683 * @hardreset: hardreset method (can be NULL)
3684 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003685 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09003686 *
3687 * This is the alpha and omega, eum and yang, heart and soul of
3688 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09003689 * recover each link and hotplug requests are recorded in the
3690 * link's eh_context. This function executes all the operations
3691 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09003692 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09003693 *
3694 * LOCKING:
3695 * Kernel thread context (may sleep).
3696 *
3697 * RETURNS:
3698 * 0 on success, -errno on failure.
3699 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003700int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3701 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3702 ata_postreset_fn_t postreset,
3703 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003704{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003705 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003706 struct ata_device *dev;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003707 int rc, nr_fails;
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003708 unsigned long flags, deadline;
Tejun Heo022bdb02006-05-15 20:58:22 +09003709
3710 DPRINTK("ENTER\n");
3711
3712 /* prep for recovery */
Tejun Heo1eca4362008-11-03 20:03:17 +09003713 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003714 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09003715
Tejun Heof9df58c2007-09-23 13:14:13 +09003716 /* re-enable link? */
3717 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3718 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3719 spin_lock_irqsave(ap->lock, flags);
3720 link->flags &= ~ATA_LFLAG_DISABLED;
3721 spin_unlock_irqrestore(ap->lock, flags);
3722 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3723 }
3724
Tejun Heo1eca4362008-11-03 20:03:17 +09003725 ata_for_each_dev(dev, link, ALL) {
Tejun Heofd995f72007-09-23 13:14:12 +09003726 if (link->flags & ATA_LFLAG_NO_RETRY)
3727 ehc->tries[dev->devno] = 1;
3728 else
3729 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09003730
Tejun Heo9b1e2652007-08-06 18:36:24 +09003731 /* collect port action mask recorded in dev actions */
3732 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3733 ~ATA_EH_PERDEV_MASK;
3734 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09003735
Tejun Heo9b1e2652007-08-06 18:36:24 +09003736 /* process hotplug request */
3737 if (dev->flags & ATA_DFLAG_DETACH)
3738 ata_eh_detach_dev(dev);
3739
Tejun Heo02c05a22007-11-27 19:28:54 +09003740 /* schedule probe if necessary */
3741 if (!ata_dev_enabled(dev))
3742 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09003743 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003744 }
3745
3746 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09003747 rc = 0;
3748
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003749 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09003750 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003751 goto out;
3752
Tejun Heo9b1e2652007-08-06 18:36:24 +09003753 /* prep for EH */
Tejun Heo1eca4362008-11-03 20:03:17 +09003754 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003755 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003756
Tejun Heo9b1e2652007-08-06 18:36:24 +09003757 /* skip EH if possible. */
3758 if (ata_eh_skip_recovery(link))
3759 ehc->i.action = 0;
3760
Tejun Heo1eca4362008-11-03 20:03:17 +09003761 ata_for_each_dev(dev, link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003762 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3763 }
Tejun Heo084fe632006-05-31 18:28:03 +09003764
Tejun Heo022bdb02006-05-15 20:58:22 +09003765 /* reset */
Tejun Heo1eca4362008-11-03 20:03:17 +09003766 ata_for_each_link(link, ap, EDGE) {
Tejun Heodc98c322008-05-19 01:15:07 +09003767 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003768
Tejun Heodc98c322008-05-19 01:15:07 +09003769 if (!(ehc->i.action & ATA_EH_RESET))
3770 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003771
Tejun Heodc98c322008-05-19 01:15:07 +09003772 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3773 prereset, softreset, hardreset, postreset);
3774 if (rc) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003775 ata_link_err(link, "reset failed, giving up\n");
Tejun Heodc98c322008-05-19 01:15:07 +09003776 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09003777 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003778 }
3779
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003780 do {
3781 unsigned long now;
3782
3783 /*
3784 * clears ATA_EH_PARK in eh_info and resets
3785 * ap->park_req_pending
3786 */
3787 ata_eh_pull_park_action(ap);
3788
3789 deadline = jiffies;
Tejun Heo1eca4362008-11-03 20:03:17 +09003790 ata_for_each_link(link, ap, EDGE) {
3791 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003792 struct ata_eh_context *ehc = &link->eh_context;
3793 unsigned long tmp;
3794
3795 if (dev->class != ATA_DEV_ATA)
3796 continue;
3797 if (!(ehc->i.dev_action[dev->devno] &
3798 ATA_EH_PARK))
3799 continue;
3800 tmp = dev->unpark_deadline;
3801 if (time_before(deadline, tmp))
3802 deadline = tmp;
3803 else if (time_before_eq(tmp, jiffies))
3804 continue;
3805 if (ehc->unloaded_mask & (1 << dev->devno))
3806 continue;
3807
3808 ata_eh_park_issue_cmd(dev, 1);
3809 }
3810 }
3811
3812 now = jiffies;
3813 if (time_before_eq(deadline, now))
3814 break;
3815
Tejun Heoc0c362b2010-09-06 17:57:14 +02003816 ata_eh_release(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003817 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3818 deadline - now);
Tejun Heoc0c362b2010-09-06 17:57:14 +02003819 ata_eh_acquire(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003820 } while (deadline);
Tejun Heo1eca4362008-11-03 20:03:17 +09003821 ata_for_each_link(link, ap, EDGE) {
3822 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003823 if (!(link->eh_context.unloaded_mask &
3824 (1 << dev->devno)))
3825 continue;
3826
3827 ata_eh_park_issue_cmd(dev, 0);
3828 ata_eh_done(link, dev, ATA_EH_PARK);
3829 }
3830 }
3831
Tejun Heo9b1e2652007-08-06 18:36:24 +09003832 /* the rest */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003833 nr_fails = 0;
3834 ata_for_each_link(link, ap, PMP_FIRST) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003835 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003836
Tejun Heo6b7ae952010-09-01 17:50:06 +02003837 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3838 goto config_lpm;
3839
Tejun Heo9b1e2652007-08-06 18:36:24 +09003840 /* revalidate existing devices and attach new ones */
3841 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09003842 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003843 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003844
Tejun Heo633273a32007-09-23 13:19:54 +09003845 /* if PMP got attached, return, pmp EH will take care of it */
3846 if (link->device->class == ATA_DEV_PMP) {
3847 ehc->i.action = 0;
3848 return 0;
3849 }
3850
Tejun Heo9b1e2652007-08-06 18:36:24 +09003851 /* configure transfer mode if necessary */
3852 if (ehc->i.flags & ATA_EHI_SETMODE) {
3853 rc = ata_set_mode(link, &dev);
3854 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003855 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003856 ehc->i.flags &= ~ATA_EHI_SETMODE;
3857 }
3858
Tejun Heo11fc33d2008-08-30 14:20:01 +02003859 /* If reset has been issued, clear UA to avoid
3860 * disrupting the current users of the device.
3861 */
3862 if (ehc->i.flags & ATA_EHI_DID_RESET) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003863 ata_for_each_dev(dev, link, ALL) {
Tejun Heo11fc33d2008-08-30 14:20:01 +02003864 if (dev->class != ATA_DEV_ATAPI)
3865 continue;
3866 rc = atapi_eh_clear_ua(dev);
3867 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003868 goto rest_fail;
Aaron Lu21334202013-01-15 17:21:01 +08003869 if (zpodd_dev_enabled(dev))
3870 zpodd_post_poweron(dev);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003871 }
3872 }
3873
Tejun Heo6013efd2009-11-19 15:36:45 +09003874 /* retry flush if necessary */
3875 ata_for_each_dev(dev, link, ALL) {
3876 if (dev->class != ATA_DEV_ATA)
3877 continue;
3878 rc = ata_eh_maybe_retry_flush(dev);
3879 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003880 goto rest_fail;
Tejun Heo6013efd2009-11-19 15:36:45 +09003881 }
3882
Tejun Heo6b7ae952010-09-01 17:50:06 +02003883 config_lpm:
Tejun Heo11fc33d2008-08-30 14:20:01 +02003884 /* configure link power saving */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003885 if (link->lpm_policy != ap->target_lpm_policy) {
3886 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3887 if (rc)
3888 goto rest_fail;
3889 }
Kristen Carlson Accardica773292007-10-25 00:58:59 -04003890
Tejun Heo9b1e2652007-08-06 18:36:24 +09003891 /* this link is okay now */
3892 ehc->i.flags = 0;
3893 continue;
3894
Tejun Heo6b7ae952010-09-01 17:50:06 +02003895 rest_fail:
3896 nr_fails++;
3897 if (dev)
3898 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo9b1e2652007-08-06 18:36:24 +09003899
Tejun Heob06ce3e2007-10-09 15:06:48 +09003900 if (ap->pflags & ATA_PFLAG_FROZEN) {
3901 /* PMP reset requires working host port.
3902 * Can't retry if it's frozen.
3903 */
Tejun Heo071f44b2008-04-07 22:47:22 +09003904 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09003905 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003906 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09003907 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003908 }
3909
Tejun Heo6b7ae952010-09-01 17:50:06 +02003910 if (nr_fails)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003911 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003912
Tejun Heo022bdb02006-05-15 20:58:22 +09003913 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003914 if (rc && r_failed_link)
3915 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003916
3917 DPRINTK("EXIT, rc=%d\n", rc);
3918 return rc;
3919}
3920
3921/**
3922 * ata_eh_finish - finish up EH
3923 * @ap: host port to finish EH for
3924 *
3925 * Recovery is complete. Clean up EH states and retry or finish
3926 * failed qcs.
3927 *
3928 * LOCKING:
3929 * None.
3930 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003931void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09003932{
3933 int tag;
3934
3935 /* retry or finish qcs */
3936 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3937 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3938
3939 if (!(qc->flags & ATA_QCFLAG_FAILED))
3940 continue;
3941
3942 if (qc->err_mask) {
3943 /* FIXME: Once EH migration is complete,
3944 * generate sense data in this function,
3945 * considering both err_mask and tf.
3946 */
Tejun Heo03faab72008-03-27 19:14:24 +09003947 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09003948 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09003949 else
3950 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003951 } else {
3952 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3953 ata_eh_qc_complete(qc);
3954 } else {
3955 /* feed zero TF to sense generation */
3956 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3957 ata_eh_qc_retry(qc);
3958 }
3959 }
3960 }
Tejun Heoda917d62007-09-23 13:14:12 +09003961
3962 /* make sure nr_active_links is zero after EH */
3963 WARN_ON(ap->nr_active_links);
3964 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003965}
3966
3967/**
3968 * ata_do_eh - do standard error handling
3969 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09003970 *
Tejun Heof5914a42006-05-31 18:27:48 +09003971 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003972 * @softreset: softreset method (can be NULL)
3973 * @hardreset: hardreset method (can be NULL)
3974 * @postreset: postreset method (can be NULL)
3975 *
3976 * Perform standard error handling sequence.
3977 *
3978 * LOCKING:
3979 * Kernel thread context (may sleep).
3980 */
Tejun Heof5914a42006-05-31 18:27:48 +09003981void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3982 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3983 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09003984{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003985 struct ata_device *dev;
3986 int rc;
3987
3988 ata_eh_autopsy(ap);
3989 ata_eh_report(ap);
3990
3991 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3992 NULL);
3993 if (rc) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003994 ata_for_each_dev(dev, &ap->link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003995 ata_dev_disable(dev);
3996 }
3997
Tejun Heo022bdb02006-05-15 20:58:22 +09003998 ata_eh_finish(ap);
3999}
Tejun Heo500530f2006-07-03 16:07:27 +09004000
Tejun Heoa1efdab2008-03-25 12:22:50 +09004001/**
4002 * ata_std_error_handler - standard error handler
4003 * @ap: host port to handle error for
4004 *
4005 * Standard error handler
4006 *
4007 * LOCKING:
4008 * Kernel thread context (may sleep).
4009 */
4010void ata_std_error_handler(struct ata_port *ap)
4011{
4012 struct ata_port_operations *ops = ap->ops;
4013 ata_reset_fn_t hardreset = ops->hardreset;
4014
Tejun Heo57c9efd2008-04-07 22:47:19 +09004015 /* ignore built-in hardreset if SCR access is not available */
Tejun Heofe06e5f2010-05-10 21:41:39 +02004016 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09004017 hardreset = NULL;
4018
4019 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4020}
4021
Tejun Heo6ffa01d2007-03-02 17:32:47 +09004022#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09004023/**
4024 * ata_eh_handle_port_suspend - perform port suspend operation
4025 * @ap: port to suspend
4026 *
4027 * Suspend @ap.
4028 *
4029 * LOCKING:
4030 * Kernel thread context (may sleep).
4031 */
4032static void ata_eh_handle_port_suspend(struct ata_port *ap)
4033{
4034 unsigned long flags;
4035 int rc = 0;
Aaron Lu3dc67442013-01-15 17:21:00 +08004036 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09004037
4038 /* are we suspending? */
4039 spin_lock_irqsave(ap->lock, flags);
4040 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
Aaron Lua7ff60d2013-01-25 14:29:35 +08004041 ap->pm_mesg.event & PM_EVENT_RESUME) {
Tejun Heo500530f2006-07-03 16:07:27 +09004042 spin_unlock_irqrestore(ap->lock, flags);
4043 return;
4044 }
4045 spin_unlock_irqrestore(ap->lock, flags);
4046
4047 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4048
Aaron Lu3dc67442013-01-15 17:21:00 +08004049 /*
4050 * If we have a ZPODD attached, check its zero
4051 * power ready status before the port is frozen.
Aaron Lua7ff60d2013-01-25 14:29:35 +08004052 * Only needed for runtime suspend.
Aaron Lu3dc67442013-01-15 17:21:00 +08004053 */
Aaron Lua7ff60d2013-01-25 14:29:35 +08004054 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4055 ata_for_each_dev(dev, &ap->link, ENABLED) {
4056 if (zpodd_dev_enabled(dev))
4057 zpodd_on_suspend(dev);
4058 }
Aaron Lu3dc67442013-01-15 17:21:00 +08004059 }
4060
Tejun Heo64578a32007-05-15 03:28:16 +09004061 /* tell ACPI we're suspending */
4062 rc = ata_acpi_on_suspend(ap);
4063 if (rc)
4064 goto out;
4065
Tejun Heo500530f2006-07-03 16:07:27 +09004066 /* suspend */
4067 ata_eh_freeze_port(ap);
4068
4069 if (ap->ops->port_suspend)
4070 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4071
Aaron Lua7ff60d2013-01-25 14:29:35 +08004072 ata_acpi_set_state(ap, ap->pm_mesg);
Tejun Heo64578a32007-05-15 03:28:16 +09004073 out:
Dan Williamsbc6e7c42014-03-14 13:52:48 -07004074 /* update the flags */
Tejun Heo500530f2006-07-03 16:07:27 +09004075 spin_lock_irqsave(ap->lock, flags);
4076
4077 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4078 if (rc == 0)
4079 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09004080 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09004081 ata_port_schedule_eh(ap);
4082
Tejun Heo500530f2006-07-03 16:07:27 +09004083 spin_unlock_irqrestore(ap->lock, flags);
4084
4085 return;
4086}
4087
4088/**
4089 * ata_eh_handle_port_resume - perform port resume operation
4090 * @ap: port to resume
4091 *
4092 * Resume @ap.
4093 *
Tejun Heo500530f2006-07-03 16:07:27 +09004094 * LOCKING:
4095 * Kernel thread context (may sleep).
4096 */
4097static void ata_eh_handle_port_resume(struct ata_port *ap)
4098{
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004099 struct ata_link *link;
4100 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09004101 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02004102 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09004103
4104 /* are we resuming? */
4105 spin_lock_irqsave(ap->lock, flags);
4106 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
Aaron Lua7ff60d2013-01-25 14:29:35 +08004107 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
Tejun Heo500530f2006-07-03 16:07:27 +09004108 spin_unlock_irqrestore(ap->lock, flags);
4109 return;
4110 }
4111 spin_unlock_irqrestore(ap->lock, flags);
4112
Tejun Heo9666f402007-05-04 21:27:47 +02004113 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09004114
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004115 /*
4116 * Error timestamps are in jiffies which doesn't run while
4117 * suspended and PHY events during resume isn't too uncommon.
4118 * When the two are combined, it can lead to unnecessary speed
4119 * downs if the machine is suspended and resumed repeatedly.
4120 * Clear error history.
4121 */
4122 ata_for_each_link(link, ap, HOST_FIRST)
4123 ata_for_each_dev(dev, link, ALL)
4124 ata_ering_clear(&dev->ering);
4125
Aaron Lua7ff60d2013-01-25 14:29:35 +08004126 ata_acpi_set_state(ap, ap->pm_mesg);
Shaohua Libd3adca2007-11-02 09:32:38 +08004127
Tejun Heo500530f2006-07-03 16:07:27 +09004128 if (ap->ops->port_resume)
4129 rc = ap->ops->port_resume(ap);
4130
Tejun Heo67465442007-05-15 03:28:16 +09004131 /* tell ACPI that we're resuming */
4132 ata_acpi_on_resume(ap);
4133
Dan Williamsbc6e7c42014-03-14 13:52:48 -07004134 /* update the flags */
Tejun Heo500530f2006-07-03 16:07:27 +09004135 spin_lock_irqsave(ap->lock, flags);
4136 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
Tejun Heo500530f2006-07-03 16:07:27 +09004137 spin_unlock_irqrestore(ap->lock, flags);
4138}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09004139#endif /* CONFIG_PM */