blob: 7f099d6e4e0bca601f26f83597e01ec185f03d54 [file] [log] [blame]
Tejun Heoece1d632006-04-02 18:51:53 +09001/*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
Tejun Heoece1d632006-04-02 18:51:53 +090035#include <linux/kernel.h>
Jens Axboe242f9dc2008-09-14 05:55:09 -070036#include <linux/blkdev.h>
Jeff Garzik28555682007-10-11 17:12:35 -040037#include <linux/pci.h>
Tejun Heoece1d632006-04-02 18:51:53 +090038#include <scsi/scsi.h>
39#include <scsi/scsi_host.h>
40#include <scsi/scsi_eh.h>
41#include <scsi/scsi_device.h>
42#include <scsi/scsi_cmnd.h>
Robert Hancock65211482009-07-14 20:43:39 -060043#include <scsi/scsi_dbg.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040044#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090045
46#include <linux/libata.h>
47
48#include "libata.h"
49
Tejun Heo7d47e8d2007-02-02 16:22:31 +090050enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090051 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090052 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090055 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090056
57 /* error flags */
58 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090059 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Gwendal Grignoud9027472010-05-25 12:31:38 -070060 ATA_EFLAG_OLD_ER = (1 << 31),
Tejun Heo3884f7b2007-11-27 19:28:56 +090061
62 /* error categories */
63 ATA_ECAT_NONE = 0,
64 ATA_ECAT_ATA_BUS = 1,
65 ATA_ECAT_TOUT_HSM = 2,
66 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090067 ATA_ECAT_DUBIOUS_NONE = 4,
68 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
69 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
70 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090072
Tejun Heo87fbc5a2008-05-20 02:17:54 +090073 ATA_EH_CMD_DFL_TIMEOUT = 5000,
74
Tejun Heo0a2c0f52008-05-20 02:17:52 +090075 /* always put at least this amount of time between resets */
76 ATA_EH_RESET_COOL_DOWN = 5000,
77
Tejun Heo341c2c92008-05-20 02:17:51 +090078 /* Waiting in ->prereset can never be reliable. It's
79 * sometimes nice to wait there but it can't be depended upon;
80 * otherwise, we wouldn't be resetting. Just give it enough
81 * time for most drives to spin up.
82 */
83 ATA_EH_PRERESET_TIMEOUT = 10000,
84 ATA_EH_FASTDRAIN_INTERVAL = 3000,
Tejun Heo11fc33d2008-08-30 14:20:01 +020085
86 ATA_EH_UA_TRIES = 5,
Tejun Heoc2c7a892009-01-29 20:31:34 +090087
88 /* probe speed down parameters, see ata_eh_schedule_probe() */
89 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
90 ATA_EH_PROBE_TRIALS = 2,
Tejun Heo31daabd2007-02-02 16:50:52 +090091};
92
93/* The following table determines how we sequence resets. Each entry
94 * represents timeout for that try. The first try can be soft or
95 * hardreset. All others are hardreset if available. In most cases
96 * the first reset w/ 10sec timeout should succeed. Following entries
97 * are mostly for error handling, hotplug and retarded devices.
98 */
99static const unsigned long ata_eh_reset_timeouts[] = {
Tejun Heo341c2c92008-05-20 02:17:51 +0900100 10000, /* most drives spin up by 10sec */
101 10000, /* > 99% working drives spin up before 20sec */
102 35000, /* give > 30 secs of idleness for retarded devices */
103 5000, /* and sweet one last chance */
Tejun Heod8af0eb2008-05-20 02:17:53 +0900104 ULONG_MAX, /* > 1 min has elapsed, give up */
Tejun Heo31daabd2007-02-02 16:50:52 +0900105};
106
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900107static const unsigned long ata_eh_identify_timeouts[] = {
108 5000, /* covers > 99% of successes and not too boring on failures */
109 10000, /* combined time till here is enough even for media access */
110 30000, /* for true idiots */
111 ULONG_MAX,
112};
113
Tejun Heo6013efd2009-11-19 15:36:45 +0900114static const unsigned long ata_eh_flush_timeouts[] = {
115 15000, /* be generous with flush */
116 15000, /* ditto */
117 30000, /* and even more generous */
118 ULONG_MAX,
119};
120
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900121static const unsigned long ata_eh_other_timeouts[] = {
122 5000, /* same rationale as identify timeout */
123 10000, /* ditto */
124 /* but no merciful 30sec for other commands, it just isn't worth it */
125 ULONG_MAX,
126};
127
128struct ata_eh_cmd_timeout_ent {
129 const u8 *commands;
130 const unsigned long *timeouts;
131};
132
133/* The following table determines timeouts to use for EH internal
134 * commands. Each table entry is a command class and matches the
135 * commands the entry applies to and the timeout table to use.
136 *
137 * On the retry after a command timed out, the next timeout value from
138 * the table is used. If the table doesn't contain further entries,
139 * the last value is used.
140 *
141 * ehc->cmd_timeout_idx keeps track of which timeout to use per
142 * command class, so if SET_FEATURES times out on the first try, the
143 * next try will use the second timeout value only for that class.
144 */
145#define CMDS(cmds...) (const u8 []){ cmds, 0 }
146static const struct ata_eh_cmd_timeout_ent
147ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
148 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
149 .timeouts = ata_eh_identify_timeouts, },
150 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
151 .timeouts = ata_eh_other_timeouts, },
152 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
153 .timeouts = ata_eh_other_timeouts, },
154 { .commands = CMDS(ATA_CMD_SET_FEATURES),
155 .timeouts = ata_eh_other_timeouts, },
156 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
157 .timeouts = ata_eh_other_timeouts, },
Tejun Heo6013efd2009-11-19 15:36:45 +0900158 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
159 .timeouts = ata_eh_flush_timeouts },
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900160};
161#undef CMDS
162
Tejun Heoad9e2762006-05-15 20:58:12 +0900163static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900164#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +0900165static void ata_eh_handle_port_suspend(struct ata_port *ap);
166static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900167#else /* CONFIG_PM */
168static void ata_eh_handle_port_suspend(struct ata_port *ap)
169{ }
170
171static void ata_eh_handle_port_resume(struct ata_port *ap)
172{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900173#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900174
Tejun Heob64bbc32007-07-16 14:29:39 +0900175static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
176 va_list args)
177{
178 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
179 ATA_EH_DESC_LEN - ehi->desc_len,
180 fmt, args);
181}
182
183/**
184 * __ata_ehi_push_desc - push error description without adding separator
185 * @ehi: target EHI
186 * @fmt: printf format string
187 *
188 * Format string according to @fmt and append it to @ehi->desc.
189 *
190 * LOCKING:
191 * spin_lock_irqsave(host lock)
192 */
193void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
194{
195 va_list args;
196
197 va_start(args, fmt);
198 __ata_ehi_pushv_desc(ehi, fmt, args);
199 va_end(args);
200}
201
202/**
203 * ata_ehi_push_desc - push error description with separator
204 * @ehi: target EHI
205 * @fmt: printf format string
206 *
207 * Format string according to @fmt and append it to @ehi->desc.
208 * If @ehi->desc is not empty, ", " is added in-between.
209 *
210 * LOCKING:
211 * spin_lock_irqsave(host lock)
212 */
213void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
214{
215 va_list args;
216
217 if (ehi->desc_len)
218 __ata_ehi_push_desc(ehi, ", ");
219
220 va_start(args, fmt);
221 __ata_ehi_pushv_desc(ehi, fmt, args);
222 va_end(args);
223}
224
225/**
226 * ata_ehi_clear_desc - clean error description
227 * @ehi: target EHI
228 *
229 * Clear @ehi->desc.
230 *
231 * LOCKING:
232 * spin_lock_irqsave(host lock)
233 */
234void ata_ehi_clear_desc(struct ata_eh_info *ehi)
235{
236 ehi->desc[0] = '\0';
237 ehi->desc_len = 0;
238}
239
Tejun Heocbcdd872007-08-18 13:14:55 +0900240/**
241 * ata_port_desc - append port description
242 * @ap: target ATA port
243 * @fmt: printf format string
244 *
245 * Format string according to @fmt and append it to port
246 * description. If port description is not empty, " " is added
247 * in-between. This function is to be used while initializing
248 * ata_host. The description is printed on host registration.
249 *
250 * LOCKING:
251 * None.
252 */
253void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
254{
255 va_list args;
256
257 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
258
259 if (ap->link.eh_info.desc_len)
260 __ata_ehi_push_desc(&ap->link.eh_info, " ");
261
262 va_start(args, fmt);
263 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
264 va_end(args);
265}
266
267#ifdef CONFIG_PCI
268
269/**
270 * ata_port_pbar_desc - append PCI BAR description
271 * @ap: target ATA port
272 * @bar: target PCI BAR
273 * @offset: offset into PCI BAR
274 * @name: name of the area
275 *
276 * If @offset is negative, this function formats a string which
277 * contains the name, address, size and type of the BAR and
278 * appends it to the port description. If @offset is zero or
279 * positive, only name and offsetted address is appended.
280 *
281 * LOCKING:
282 * None.
283 */
284void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
285 const char *name)
286{
287 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
288 char *type = "";
289 unsigned long long start, len;
290
291 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
292 type = "m";
293 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
294 type = "i";
295
296 start = (unsigned long long)pci_resource_start(pdev, bar);
297 len = (unsigned long long)pci_resource_len(pdev, bar);
298
299 if (offset < 0)
300 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
301 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800302 ata_port_desc(ap, "%s 0x%llx", name,
303 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900304}
305
306#endif /* CONFIG_PCI */
307
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900308static int ata_lookup_timeout_table(u8 cmd)
309{
310 int i;
311
312 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
313 const u8 *cur;
314
315 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
316 if (*cur == cmd)
317 return i;
318 }
319
320 return -1;
321}
322
323/**
324 * ata_internal_cmd_timeout - determine timeout for an internal command
325 * @dev: target device
326 * @cmd: internal command to be issued
327 *
328 * Determine timeout for internal command @cmd for @dev.
329 *
330 * LOCKING:
331 * EH context.
332 *
333 * RETURNS:
334 * Determined timeout.
335 */
336unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
337{
338 struct ata_eh_context *ehc = &dev->link->eh_context;
339 int ent = ata_lookup_timeout_table(cmd);
340 int idx;
341
342 if (ent < 0)
343 return ATA_EH_CMD_DFL_TIMEOUT;
344
345 idx = ehc->cmd_timeout_idx[dev->devno][ent];
346 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
347}
348
349/**
350 * ata_internal_cmd_timed_out - notification for internal command timeout
351 * @dev: target device
352 * @cmd: internal command which timed out
353 *
354 * Notify EH that internal command @cmd for @dev timed out. This
355 * function should be called only for commands whose timeouts are
356 * determined using ata_internal_cmd_timeout().
357 *
358 * LOCKING:
359 * EH context.
360 */
361void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
362{
363 struct ata_eh_context *ehc = &dev->link->eh_context;
364 int ent = ata_lookup_timeout_table(cmd);
365 int idx;
366
367 if (ent < 0)
368 return;
369
370 idx = ehc->cmd_timeout_idx[dev->devno][ent];
371 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
372 ehc->cmd_timeout_idx[dev->devno][ent]++;
373}
374
Tejun Heo3884f7b2007-11-27 19:28:56 +0900375static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900376 unsigned int err_mask)
377{
378 struct ata_ering_entry *ent;
379
380 WARN_ON(!err_mask);
381
382 ering->cursor++;
383 ering->cursor %= ATA_ERING_SIZE;
384
385 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900386 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900387 ent->err_mask = err_mask;
388 ent->timestamp = get_jiffies_64();
389}
390
Tejun Heo76326ac2007-11-27 19:28:59 +0900391static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
392{
393 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
394
395 if (ent->err_mask)
396 return ent;
397 return NULL;
398}
399
Gwendal Grignoud9027472010-05-25 12:31:38 -0700400int ata_ering_map(struct ata_ering *ering,
401 int (*map_fn)(struct ata_ering_entry *, void *),
402 void *arg)
Tejun Heo0c247c52006-05-15 20:58:19 +0900403{
404 int idx, rc = 0;
405 struct ata_ering_entry *ent;
406
407 idx = ering->cursor;
408 do {
409 ent = &ering->ring[idx];
410 if (!ent->err_mask)
411 break;
412 rc = map_fn(ent, arg);
413 if (rc)
414 break;
415 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
416 } while (idx != ering->cursor);
417
418 return rc;
419}
420
Gwendal Grignoud9027472010-05-25 12:31:38 -0700421int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
422{
423 ent->eflags |= ATA_EFLAG_OLD_ER;
424 return 0;
425}
426
427static void ata_ering_clear(struct ata_ering *ering)
428{
429 ata_ering_map(ering, ata_ering_clear_cb, NULL);
430}
431
Tejun Heo64f65ca2006-06-24 20:30:18 +0900432static unsigned int ata_eh_dev_action(struct ata_device *dev)
433{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900434 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900435
436 return ehc->i.action | ehc->i.dev_action[dev->devno];
437}
438
Tejun Heof58229f2007-08-06 18:36:23 +0900439static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900440 struct ata_eh_info *ehi, unsigned int action)
441{
Tejun Heof58229f2007-08-06 18:36:23 +0900442 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900443
444 if (!dev) {
445 ehi->action &= ~action;
Tejun Heo1eca4362008-11-03 20:03:17 +0900446 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900447 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900448 } else {
449 /* doesn't make sense for port-wide EH actions */
450 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
451
452 /* break ehi->action into ehi->dev_action */
453 if (ehi->action & action) {
Tejun Heo1eca4362008-11-03 20:03:17 +0900454 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900455 ehi->dev_action[tdev->devno] |=
456 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900457 ehi->action &= ~action;
458 }
459
460 /* turn off the specified per-dev action */
461 ehi->dev_action[dev->devno] &= ~action;
462 }
463}
464
Tejun Heoece1d632006-04-02 18:51:53 +0900465/**
Tejun Heoc0c362b2010-09-06 17:57:14 +0200466 * ata_eh_acquire - acquire EH ownership
467 * @ap: ATA port to acquire EH ownership for
468 *
469 * Acquire EH ownership for @ap. This is the basic exclusion
470 * mechanism for ports sharing a host. Only one port hanging off
471 * the same host can claim the ownership of EH.
472 *
473 * LOCKING:
474 * EH context.
475 */
476void ata_eh_acquire(struct ata_port *ap)
477{
478 mutex_lock(&ap->host->eh_mutex);
479 WARN_ON_ONCE(ap->host->eh_owner);
480 ap->host->eh_owner = current;
481}
482
483/**
484 * ata_eh_release - release EH ownership
485 * @ap: ATA port to release EH ownership for
486 *
487 * Release EH ownership for @ap if the caller. The caller must
488 * have acquired EH ownership using ata_eh_acquire() previously.
489 *
490 * LOCKING:
491 * EH context.
492 */
493void ata_eh_release(struct ata_port *ap)
494{
495 WARN_ON_ONCE(ap->host->eh_owner != current);
496 ap->host->eh_owner = NULL;
497 mutex_unlock(&ap->host->eh_mutex);
498}
499
500/**
Tejun Heoece1d632006-04-02 18:51:53 +0900501 * ata_scsi_timed_out - SCSI layer time out callback
502 * @cmd: timed out SCSI command
503 *
504 * Handles SCSI layer timeout. We race with normal completion of
505 * the qc for @cmd. If the qc is already gone, we lose and let
506 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
507 * timed out and EH should be invoked. Prevent ata_qc_complete()
508 * from finishing it by setting EH_SCHEDULED and return
509 * EH_NOT_HANDLED.
510 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900511 * TODO: kill this function once old EH is gone.
512 *
Tejun Heoece1d632006-04-02 18:51:53 +0900513 * LOCKING:
514 * Called from timer context
515 *
516 * RETURNS:
517 * EH_HANDLED or EH_NOT_HANDLED
518 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700519enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
Tejun Heoece1d632006-04-02 18:51:53 +0900520{
521 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400522 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900523 unsigned long flags;
524 struct ata_queued_cmd *qc;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700525 enum blk_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900526
527 DPRINTK("ENTER\n");
528
Tejun Heoad9e2762006-05-15 20:58:12 +0900529 if (ap->ops->error_handler) {
Jens Axboe242f9dc2008-09-14 05:55:09 -0700530 ret = BLK_EH_NOT_HANDLED;
Tejun Heoad9e2762006-05-15 20:58:12 +0900531 goto out;
532 }
533
Jens Axboe242f9dc2008-09-14 05:55:09 -0700534 ret = BLK_EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400535 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900536 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900537 if (qc) {
538 WARN_ON(qc->scsicmd != cmd);
539 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
540 qc->err_mask |= AC_ERR_TIMEOUT;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700541 ret = BLK_EH_NOT_HANDLED;
Tejun Heoece1d632006-04-02 18:51:53 +0900542 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400543 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900544
Tejun Heoad9e2762006-05-15 20:58:12 +0900545 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900546 DPRINTK("EXIT, ret=%d\n", ret);
547 return ret;
548}
549
Tejun Heoece180d2008-11-03 20:04:37 +0900550static void ata_eh_unload(struct ata_port *ap)
551{
552 struct ata_link *link;
553 struct ata_device *dev;
554 unsigned long flags;
555
556 /* Restore SControl IPM and SPD for the next driver and
557 * disable attached devices.
558 */
559 ata_for_each_link(link, ap, PMP_FIRST) {
560 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
561 ata_for_each_dev(dev, link, ALL)
562 ata_dev_disable(dev);
563 }
564
565 /* freeze and set UNLOADED */
566 spin_lock_irqsave(ap->lock, flags);
567
568 ata_port_freeze(ap); /* won't be thawed */
569 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
570 ap->pflags |= ATA_PFLAG_UNLOADED;
571
572 spin_unlock_irqrestore(ap->lock, flags);
573}
574
Tejun Heoece1d632006-04-02 18:51:53 +0900575/**
576 * ata_scsi_error - SCSI layer error handler callback
577 * @host: SCSI host on which error occurred
578 *
579 * Handles SCSI-layer-thrown error events.
580 *
581 * LOCKING:
582 * Inherited from SCSI layer (none, can sleep)
583 *
584 * RETURNS:
585 * Zero.
586 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400587void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900588{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400589 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoad9e2762006-05-15 20:58:12 +0900590 unsigned long flags;
James Bottomleyc34aeeb2011-01-23 08:31:14 -0600591 LIST_HEAD(eh_work_q);
Tejun Heoece1d632006-04-02 18:51:53 +0900592
593 DPRINTK("ENTER\n");
594
James Bottomleyc34aeeb2011-01-23 08:31:14 -0600595 spin_lock_irqsave(host->host_lock, flags);
596 list_splice_init(&host->eh_cmd_q, &eh_work_q);
597 spin_unlock_irqrestore(host->host_lock, flags);
598
James Bottomley0e0b494c2011-01-23 09:42:50 -0600599 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
600
601 /* If we timed raced normal completion and there is nothing to
602 recover nr_timedout == 0 why exactly are we doing error recovery ? */
603 ata_scsi_port_error_handler(host, ap);
604
605 /* finish or retry handled scmd's and clean up */
606 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
607
608 DPRINTK("EXIT\n");
609}
610
611/**
612 * ata_scsi_cmd_error_handler - error callback for a list of commands
613 * @host: scsi host containing the port
614 * @ap: ATA port within the host
615 * @eh_work_q: list of commands to process
616 *
617 * process the given list of commands and return those finished to the
618 * ap->eh_done_q. This function is the first part of the libata error
619 * handler which processes a given list of failed commands.
620 */
621void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
622 struct list_head *eh_work_q)
623{
624 int i;
625 unsigned long flags;
626
Tejun Heoc4291372010-05-10 21:41:38 +0200627 /* make sure sff pio task is not running */
628 ata_sff_flush_pio_task(ap);
Tejun Heoece1d632006-04-02 18:51:53 +0900629
Jeff Garzikcca39742006-08-24 03:19:22 -0400630 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900631
Tejun Heoad9e2762006-05-15 20:58:12 +0900632 /* For new EH, all qcs are finished in one of three ways -
633 * normal completion, error completion, and SCSI timeout.
Alan Coxc96f1732009-03-24 10:23:46 +0000634 * Both completions can race against SCSI timeout. When normal
Tejun Heoad9e2762006-05-15 20:58:12 +0900635 * completion wins, the qc never reaches EH. When error
636 * completion wins, the qc has ATA_QCFLAG_FAILED set.
637 *
638 * When SCSI timeout wins, things are a bit more complex.
639 * Normal or error completion can occur after the timeout but
640 * before this point. In such cases, both types of
641 * completions are honored. A scmd is determined to have
642 * timed out iff its associated qc is active and not failed.
643 */
644 if (ap->ops->error_handler) {
645 struct scsi_cmnd *scmd, *tmp;
646 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900647
Tejun Heoe30349d2006-07-03 03:02:15 +0900648 spin_lock_irqsave(ap->lock, flags);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700649
Alan Coxc96f1732009-03-24 10:23:46 +0000650 /* This must occur under the ap->lock as we don't want
651 a polled recovery to race the real interrupt handler
Gwendal Grignoud9027472010-05-25 12:31:38 -0700652
Alan Coxc96f1732009-03-24 10:23:46 +0000653 The lost_interrupt handler checks for any completed but
654 non-notified command and completes much like an IRQ handler.
Gwendal Grignoud9027472010-05-25 12:31:38 -0700655
Alan Coxc96f1732009-03-24 10:23:46 +0000656 We then fall into the error recovery code which will treat
657 this as if normal completion won the race */
Tejun Heoad9e2762006-05-15 20:58:12 +0900658
Alan Coxc96f1732009-03-24 10:23:46 +0000659 if (ap->ops->lost_interrupt)
660 ap->ops->lost_interrupt(ap);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700661
James Bottomley0e0b494c2011-01-23 09:42:50 -0600662 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
Tejun Heoad9e2762006-05-15 20:58:12 +0900663 struct ata_queued_cmd *qc;
664
665 for (i = 0; i < ATA_MAX_QUEUE; i++) {
666 qc = __ata_qc_from_tag(ap, i);
667 if (qc->flags & ATA_QCFLAG_ACTIVE &&
668 qc->scsicmd == scmd)
669 break;
670 }
671
672 if (i < ATA_MAX_QUEUE) {
673 /* the scmd has an associated qc */
674 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
675 /* which hasn't failed yet, timeout */
676 qc->err_mask |= AC_ERR_TIMEOUT;
677 qc->flags |= ATA_QCFLAG_FAILED;
678 nr_timedout++;
679 }
680 } else {
681 /* Normal completion occurred after
682 * SCSI timeout but before this point.
683 * Successfully complete it.
684 */
685 scmd->retries = scmd->allowed;
686 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
687 }
688 }
689
690 /* If we have timed out qcs. They belong to EH from
691 * this point but the state of the controller is
692 * unknown. Freeze the port to make sure the IRQ
693 * handler doesn't diddle with those qcs. This must
694 * be done atomically w.r.t. setting QCFLAG_FAILED.
695 */
696 if (nr_timedout)
697 __ata_port_freeze(ap);
698
Tejun Heoe30349d2006-07-03 03:02:15 +0900699 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900700
701 /* initialize eh_tries */
702 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900703 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900704 spin_unlock_wait(ap->lock);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700705
James Bottomley0e0b494c2011-01-23 09:42:50 -0600706}
707EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
708
709/**
710 * ata_scsi_port_error_handler - recover the port after the commands
711 * @host: SCSI host containing the port
712 * @ap: the ATA port
713 *
714 * Handle the recovery of the port @ap after all the commands
715 * have been recovered.
716 */
717void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
718{
719 unsigned long flags;
Tejun Heoad9e2762006-05-15 20:58:12 +0900720
Tejun Heoad9e2762006-05-15 20:58:12 +0900721 /* invoke error handler */
722 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900723 struct ata_link *link;
724
Tejun Heoc0c362b2010-09-06 17:57:14 +0200725 /* acquire EH ownership */
726 ata_eh_acquire(ap);
727 repeat:
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900728 /* kill fast drain timer */
729 del_timer_sync(&ap->fastdrain_timer);
730
Tejun Heo500530f2006-07-03 16:07:27 +0900731 /* process port resume request */
732 ata_eh_handle_port_resume(ap);
733
Tejun Heof3e81b12006-05-15 20:58:21 +0900734 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900735 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900736
Tejun Heo1eca4362008-11-03 20:03:17 +0900737 ata_for_each_link(link, ap, HOST_FIRST) {
Tejun Heo00115e02007-11-27 19:28:58 +0900738 struct ata_eh_context *ehc = &link->eh_context;
739 struct ata_device *dev;
740
Tejun Heocf1b86c2007-08-06 18:36:23 +0900741 memset(&link->eh_context, 0, sizeof(link->eh_context));
742 link->eh_context.i = link->eh_info;
743 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900744
Tejun Heo1eca4362008-11-03 20:03:17 +0900745 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +0900746 int devno = dev->devno;
747
748 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
749 if (ata_ncq_enabled(dev))
750 ehc->saved_ncq_enabled |= 1 << devno;
751 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900752 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900753
Tejun Heob51e9e52006-06-29 01:29:30 +0900754 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
755 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900756 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900757
Tejun Heoe30349d2006-07-03 03:02:15 +0900758 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900759
Tejun Heo500530f2006-07-03 16:07:27 +0900760 /* invoke EH, skip if unloading or suspended */
761 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900762 ap->ops->error_handler(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900763 else {
764 /* if unloading, commence suicide */
765 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
766 !(ap->pflags & ATA_PFLAG_UNLOADED))
767 ata_eh_unload(ap);
Tejun Heo720ba122006-05-31 18:28:13 +0900768 ata_eh_finish(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900769 }
Tejun Heoad9e2762006-05-15 20:58:12 +0900770
Tejun Heo500530f2006-07-03 16:07:27 +0900771 /* process port suspend request */
772 ata_eh_handle_port_suspend(ap);
773
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300774 /* Exception might have happened after ->error_handler
Tejun Heoad9e2762006-05-15 20:58:12 +0900775 * recovered the port but before this point. Repeat
776 * EH in such case.
777 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900778 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900779
Tejun Heob51e9e52006-06-29 01:29:30 +0900780 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900781 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900782 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900783 goto repeat;
784 }
785 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
Tejun Heoa1e10f72007-08-18 13:28:49 +0900786 "tries, giving up\n", ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900787 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900788 }
789
Tejun Heof3e81b12006-05-15 20:58:21 +0900790 /* this run is complete, make sure EH info is clear */
Tejun Heo1eca4362008-11-03 20:03:17 +0900791 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heocf1b86c2007-08-06 18:36:23 +0900792 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900793
Tejun Heoe30349d2006-07-03 03:02:15 +0900794 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900795 * that if exception occurs after this point but
796 * before EH completion, SCSI midlayer will
797 * re-initiate EH.
798 */
799 host->host_eh_scheduled = 0;
800
Tejun Heoe30349d2006-07-03 03:02:15 +0900801 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc0c362b2010-09-06 17:57:14 +0200802 ata_eh_release(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900803 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900804 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900805 ap->ops->eng_timeout(ap);
806 }
807
Tejun Heoece1d632006-04-02 18:51:53 +0900808 scsi_eh_flush_done_q(&ap->eh_done_q);
809
Tejun Heoad9e2762006-05-15 20:58:12 +0900810 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900811 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900812
Tejun Heo1cdaf532006-07-03 16:07:26 +0900813 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900814 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900815 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
Tejun Heoad72cf92010-07-02 10:03:52 +0200816 schedule_delayed_work(&ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900817
818 if (ap->pflags & ATA_PFLAG_RECOVERED)
819 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900820
Tejun Heob51e9e52006-06-29 01:29:30 +0900821 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900822
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900823 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900824 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900825 wake_up_all(&ap->eh_wait_q);
826
Tejun Heoe30349d2006-07-03 03:02:15 +0900827 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900828}
James Bottomley0e0b494c2011-01-23 09:42:50 -0600829EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
Tejun Heoece1d632006-04-02 18:51:53 +0900830
831/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900832 * ata_port_wait_eh - Wait for the currently pending EH to complete
833 * @ap: Port to wait EH for
834 *
835 * Wait until the currently pending EH is complete.
836 *
837 * LOCKING:
838 * Kernel thread context (may sleep).
839 */
840void ata_port_wait_eh(struct ata_port *ap)
841{
842 unsigned long flags;
843 DEFINE_WAIT(wait);
844
845 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400846 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900847
Tejun Heob51e9e52006-06-29 01:29:30 +0900848 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900849 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400850 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900851 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400852 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900853 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900854 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900855
Jeff Garzikba6a1302006-06-22 23:46:10 -0400856 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900857
858 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400859 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heo97750ce2010-09-06 17:56:29 +0200860 ata_msleep(ap, 10);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900861 goto retry;
862 }
863}
864
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900865static int ata_eh_nr_in_flight(struct ata_port *ap)
866{
867 unsigned int tag;
868 int nr = 0;
869
870 /* count only non-internal commands */
871 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
872 if (ata_qc_from_tag(ap, tag))
873 nr++;
874
875 return nr;
876}
877
878void ata_eh_fastdrain_timerfn(unsigned long arg)
879{
880 struct ata_port *ap = (void *)arg;
881 unsigned long flags;
882 int cnt;
883
884 spin_lock_irqsave(ap->lock, flags);
885
886 cnt = ata_eh_nr_in_flight(ap);
887
888 /* are we done? */
889 if (!cnt)
890 goto out_unlock;
891
892 if (cnt == ap->fastdrain_cnt) {
893 unsigned int tag;
894
895 /* No progress during the last interval, tag all
896 * in-flight qcs as timed out and freeze the port.
897 */
898 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
899 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
900 if (qc)
901 qc->err_mask |= AC_ERR_TIMEOUT;
902 }
903
904 ata_port_freeze(ap);
905 } else {
906 /* some qcs have finished, give it another chance */
907 ap->fastdrain_cnt = cnt;
908 ap->fastdrain_timer.expires =
Tejun Heo341c2c92008-05-20 02:17:51 +0900909 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900910 add_timer(&ap->fastdrain_timer);
911 }
912
913 out_unlock:
914 spin_unlock_irqrestore(ap->lock, flags);
915}
916
917/**
918 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
919 * @ap: target ATA port
920 * @fastdrain: activate fast drain
921 *
922 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
923 * is non-zero and EH wasn't pending before. Fast drain ensures
924 * that EH kicks in in timely manner.
925 *
926 * LOCKING:
927 * spin_lock_irqsave(host lock)
928 */
929static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
930{
931 int cnt;
932
933 /* already scheduled? */
934 if (ap->pflags & ATA_PFLAG_EH_PENDING)
935 return;
936
937 ap->pflags |= ATA_PFLAG_EH_PENDING;
938
939 if (!fastdrain)
940 return;
941
942 /* do we have in-flight qcs? */
943 cnt = ata_eh_nr_in_flight(ap);
944 if (!cnt)
945 return;
946
947 /* activate fast drain */
948 ap->fastdrain_cnt = cnt;
Tejun Heo341c2c92008-05-20 02:17:51 +0900949 ap->fastdrain_timer.expires =
950 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900951 add_timer(&ap->fastdrain_timer);
952}
953
Tejun Heof686bcb2006-05-15 20:58:05 +0900954/**
955 * ata_qc_schedule_eh - schedule qc for error handling
956 * @qc: command to schedule error handling for
957 *
958 * Schedule error handling for @qc. EH will kick in as soon as
959 * other commands are drained.
960 *
961 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400962 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900963 */
964void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
965{
966 struct ata_port *ap = qc->ap;
Tejun Heofa41efd2010-04-15 08:57:37 +0900967 struct request_queue *q = qc->scsicmd->device->request_queue;
968 unsigned long flags;
Tejun Heof686bcb2006-05-15 20:58:05 +0900969
970 WARN_ON(!ap->ops->error_handler);
971
972 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900973 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900974
975 /* The following will fail if timeout has already expired.
976 * ata_scsi_error() takes care of such scmds on EH entry.
977 * Note that ATA_QCFLAG_FAILED is unconditionally set after
978 * this function completes.
979 */
Tejun Heofa41efd2010-04-15 08:57:37 +0900980 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700981 blk_abort_request(qc->scsicmd->request);
Tejun Heofa41efd2010-04-15 08:57:37 +0900982 spin_unlock_irqrestore(q->queue_lock, flags);
Tejun Heof686bcb2006-05-15 20:58:05 +0900983}
984
Tejun Heo7b70fc02006-05-15 20:58:07 +0900985/**
986 * ata_port_schedule_eh - schedule error handling without a qc
987 * @ap: ATA port to schedule EH for
988 *
989 * Schedule error handling for @ap. EH will kick in as soon as
990 * all commands are drained.
991 *
992 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400993 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900994 */
995void ata_port_schedule_eh(struct ata_port *ap)
996{
997 WARN_ON(!ap->ops->error_handler);
998
Tejun Heof4d6d002007-05-01 11:50:15 +0200999 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1000 return;
1001
Tejun Heo5ddf24c2007-07-16 14:29:41 +09001002 ata_eh_set_pending(ap, 1);
Jeff Garzikcca39742006-08-24 03:19:22 -04001003 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +09001004
1005 DPRINTK("port EH scheduled\n");
1006}
1007
Tejun Heodbd82612007-08-06 18:36:23 +09001008static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +09001009{
1010 int tag, nr_aborted = 0;
1011
1012 WARN_ON(!ap->ops->error_handler);
1013
Tejun Heo5ddf24c2007-07-16 14:29:41 +09001014 /* we're gonna abort all commands, no need for fast drain */
1015 ata_eh_set_pending(ap, 0);
1016
Tejun Heo7b70fc02006-05-15 20:58:07 +09001017 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1018 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1019
Tejun Heodbd82612007-08-06 18:36:23 +09001020 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +09001021 qc->flags |= ATA_QCFLAG_FAILED;
1022 ata_qc_complete(qc);
1023 nr_aborted++;
1024 }
1025 }
1026
1027 if (!nr_aborted)
1028 ata_port_schedule_eh(ap);
1029
1030 return nr_aborted;
1031}
1032
Tejun Heoe3180492006-05-15 20:58:09 +09001033/**
Tejun Heodbd82612007-08-06 18:36:23 +09001034 * ata_link_abort - abort all qc's on the link
1035 * @link: ATA link to abort qc's for
1036 *
1037 * Abort all active qc's active on @link and schedule EH.
1038 *
1039 * LOCKING:
1040 * spin_lock_irqsave(host lock)
1041 *
1042 * RETURNS:
1043 * Number of aborted qc's.
1044 */
1045int ata_link_abort(struct ata_link *link)
1046{
1047 return ata_do_link_abort(link->ap, link);
1048}
1049
1050/**
1051 * ata_port_abort - abort all qc's on the port
1052 * @ap: ATA port to abort qc's for
1053 *
1054 * Abort all active qc's of @ap and schedule EH.
1055 *
1056 * LOCKING:
1057 * spin_lock_irqsave(host_set lock)
1058 *
1059 * RETURNS:
1060 * Number of aborted qc's.
1061 */
1062int ata_port_abort(struct ata_port *ap)
1063{
1064 return ata_do_link_abort(ap, NULL);
1065}
1066
1067/**
Tejun Heoe3180492006-05-15 20:58:09 +09001068 * __ata_port_freeze - freeze port
1069 * @ap: ATA port to freeze
1070 *
1071 * This function is called when HSM violation or some other
1072 * condition disrupts normal operation of the port. Frozen port
1073 * is not allowed to perform any operation until the port is
1074 * thawed, which usually follows a successful reset.
1075 *
1076 * ap->ops->freeze() callback can be used for freezing the port
1077 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1078 * port cannot be frozen hardware-wise, the interrupt handler
1079 * must ack and clear interrupts unconditionally while the port
1080 * is frozen.
1081 *
1082 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001083 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001084 */
1085static void __ata_port_freeze(struct ata_port *ap)
1086{
1087 WARN_ON(!ap->ops->error_handler);
1088
1089 if (ap->ops->freeze)
1090 ap->ops->freeze(ap);
1091
Tejun Heob51e9e52006-06-29 01:29:30 +09001092 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001093
Tejun Heo44877b42007-02-21 01:06:51 +09001094 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001095}
1096
1097/**
1098 * ata_port_freeze - abort & freeze port
1099 * @ap: ATA port to freeze
1100 *
Jeff Garzik54c38442009-04-07 19:13:15 -04001101 * Abort and freeze @ap. The freeze operation must be called
1102 * first, because some hardware requires special operations
1103 * before the taskfile registers are accessible.
Tejun Heoe3180492006-05-15 20:58:09 +09001104 *
1105 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001106 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001107 *
1108 * RETURNS:
1109 * Number of aborted commands.
1110 */
1111int ata_port_freeze(struct ata_port *ap)
1112{
1113 int nr_aborted;
1114
1115 WARN_ON(!ap->ops->error_handler);
1116
Tejun Heoe3180492006-05-15 20:58:09 +09001117 __ata_port_freeze(ap);
Jeff Garzik54c38442009-04-07 19:13:15 -04001118 nr_aborted = ata_port_abort(ap);
Tejun Heoe3180492006-05-15 20:58:09 +09001119
1120 return nr_aborted;
1121}
1122
1123/**
Tejun Heo7d77b242007-09-23 13:14:13 +09001124 * sata_async_notification - SATA async notification handler
1125 * @ap: ATA port where async notification is received
1126 *
1127 * Handler to be called when async notification via SDB FIS is
1128 * received. This function schedules EH if necessary.
1129 *
1130 * LOCKING:
1131 * spin_lock_irqsave(host lock)
1132 *
1133 * RETURNS:
1134 * 1 if EH is scheduled, 0 otherwise.
1135 */
1136int sata_async_notification(struct ata_port *ap)
1137{
1138 u32 sntf;
1139 int rc;
1140
1141 if (!(ap->flags & ATA_FLAG_AN))
1142 return 0;
1143
1144 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1145 if (rc == 0)
1146 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1147
Tejun Heo071f44b2008-04-07 22:47:22 +09001148 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001149 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +09001150 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001151 /* PMP is not attached. Check whether ATAPI
1152 * AN is configured. If so, notify media
1153 * change.
1154 */
1155 struct ata_device *dev = ap->link.device;
1156
1157 if ((dev->class == ATA_DEV_ATAPI) &&
1158 (dev->flags & ATA_DFLAG_AN))
1159 ata_scsi_media_change_notify(dev);
1160 return 0;
1161 } else {
1162 /* PMP is attached but SNTF is not available.
1163 * ATAPI async media change notification is
1164 * not used. The PMP must be reporting PHY
1165 * status change, schedule EH.
1166 */
1167 ata_port_schedule_eh(ap);
1168 return 1;
1169 }
1170 } else {
1171 /* PMP is attached and SNTF is available */
1172 struct ata_link *link;
1173
1174 /* check and notify ATAPI AN */
Tejun Heo1eca4362008-11-03 20:03:17 +09001175 ata_for_each_link(link, ap, EDGE) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001176 if (!(sntf & (1 << link->pmp)))
1177 continue;
1178
1179 if ((link->device->class == ATA_DEV_ATAPI) &&
1180 (link->device->flags & ATA_DFLAG_AN))
1181 ata_scsi_media_change_notify(link->device);
1182 }
1183
1184 /* If PMP is reporting that PHY status of some
1185 * downstream ports has changed, schedule EH.
1186 */
1187 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1188 ata_port_schedule_eh(ap);
1189 return 1;
1190 }
1191
1192 return 0;
1193 }
1194}
1195
1196/**
Tejun Heoe3180492006-05-15 20:58:09 +09001197 * ata_eh_freeze_port - EH helper to freeze port
1198 * @ap: ATA port to freeze
1199 *
1200 * Freeze @ap.
1201 *
1202 * LOCKING:
1203 * None.
1204 */
1205void ata_eh_freeze_port(struct ata_port *ap)
1206{
1207 unsigned long flags;
1208
1209 if (!ap->ops->error_handler)
1210 return;
1211
Jeff Garzikba6a1302006-06-22 23:46:10 -04001212 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001213 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -04001214 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001215}
1216
1217/**
1218 * ata_port_thaw_port - EH helper to thaw port
1219 * @ap: ATA port to thaw
1220 *
1221 * Thaw frozen port @ap.
1222 *
1223 * LOCKING:
1224 * None.
1225 */
1226void ata_eh_thaw_port(struct ata_port *ap)
1227{
1228 unsigned long flags;
1229
1230 if (!ap->ops->error_handler)
1231 return;
1232
Jeff Garzikba6a1302006-06-22 23:46:10 -04001233 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001234
Tejun Heob51e9e52006-06-29 01:29:30 +09001235 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001236
1237 if (ap->ops->thaw)
1238 ap->ops->thaw(ap);
1239
Jeff Garzikba6a1302006-06-22 23:46:10 -04001240 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001241
Tejun Heo44877b42007-02-21 01:06:51 +09001242 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001243}
1244
Tejun Heoece1d632006-04-02 18:51:53 +09001245static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1246{
1247 /* nada */
1248}
1249
1250static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1251{
1252 struct ata_port *ap = qc->ap;
1253 struct scsi_cmnd *scmd = qc->scsicmd;
1254 unsigned long flags;
1255
Jeff Garzikba6a1302006-06-22 23:46:10 -04001256 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001257 qc->scsidone = ata_eh_scsidone;
1258 __ata_qc_complete(qc);
1259 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -04001260 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001261
1262 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1263}
1264
1265/**
1266 * ata_eh_qc_complete - Complete an active ATA command from EH
1267 * @qc: Command to complete
1268 *
1269 * Indicate to the mid and upper layers that an ATA command has
1270 * completed. To be used from EH.
1271 */
1272void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1273{
1274 struct scsi_cmnd *scmd = qc->scsicmd;
1275 scmd->retries = scmd->allowed;
1276 __ata_eh_qc_complete(qc);
1277}
1278
1279/**
1280 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1281 * @qc: Command to retry
1282 *
1283 * Indicate to the mid and upper layers that an ATA command
1284 * should be retried. To be used from EH.
1285 *
1286 * SCSI midlayer limits the number of retries to scmd->allowed.
1287 * scmd->retries is decremented for commands which get retried
1288 * due to unrelated failures (qc->err_mask is zero).
1289 */
1290void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1291{
1292 struct scsi_cmnd *scmd = qc->scsicmd;
1293 if (!qc->err_mask && scmd->retries)
1294 scmd->retries--;
1295 __ata_eh_qc_complete(qc);
1296}
Tejun Heo022bdb02006-05-15 20:58:22 +09001297
1298/**
Tejun Heo678afac2009-01-29 20:31:30 +09001299 * ata_dev_disable - disable ATA device
1300 * @dev: ATA device to disable
1301 *
1302 * Disable @dev.
1303 *
1304 * Locking:
1305 * EH context.
1306 */
1307void ata_dev_disable(struct ata_device *dev)
1308{
1309 if (!ata_dev_enabled(dev))
1310 return;
1311
1312 if (ata_msg_drv(dev->link->ap))
1313 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1314 ata_acpi_on_disable(dev);
1315 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1316 dev->class++;
Tejun Heo99cf6102009-01-29 20:31:32 +09001317
1318 /* From now till the next successful probe, ering is used to
1319 * track probe failures. Clear accumulated device error info.
1320 */
1321 ata_ering_clear(&dev->ering);
Tejun Heo678afac2009-01-29 20:31:30 +09001322}
1323
1324/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001325 * ata_eh_detach_dev - detach ATA device
1326 * @dev: ATA device to detach
1327 *
1328 * Detach @dev.
1329 *
1330 * LOCKING:
1331 * None.
1332 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001333void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001334{
Tejun Heof58229f2007-08-06 18:36:23 +09001335 struct ata_link *link = dev->link;
1336 struct ata_port *ap = link->ap;
Tejun Heo90484eb2008-10-26 15:43:03 +09001337 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001338 unsigned long flags;
1339
1340 ata_dev_disable(dev);
1341
Jeff Garzikba6a1302006-06-22 23:46:10 -04001342 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001343
1344 dev->flags &= ~ATA_DFLAG_DETACH;
1345
1346 if (ata_scsi_offline_dev(dev)) {
1347 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001348 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001349 }
1350
Tejun Heo90484eb2008-10-26 15:43:03 +09001351 /* clear per-dev EH info */
Tejun Heof58229f2007-08-06 18:36:23 +09001352 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1353 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heo90484eb2008-10-26 15:43:03 +09001354 ehc->saved_xfer_mode[dev->devno] = 0;
1355 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heobeb07c12006-06-24 20:30:19 +09001356
Jeff Garzikba6a1302006-06-22 23:46:10 -04001357 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001358}
1359
1360/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001361 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001362 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001363 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001364 * @action: action about to be performed
1365 *
1366 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001367 * in @link->eh_info such that eh actions are not unnecessarily
1368 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001369 *
1370 * LOCKING:
1371 * None.
1372 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001373void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1374 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001375{
Tejun Heo955e57d2007-08-06 18:36:23 +09001376 struct ata_port *ap = link->ap;
1377 struct ata_eh_info *ehi = &link->eh_info;
1378 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001379 unsigned long flags;
1380
Jeff Garzikba6a1302006-06-22 23:46:10 -04001381 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001382
Tejun Heo955e57d2007-08-06 18:36:23 +09001383 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001384
Tejun Heoa568d1d2008-10-21 20:37:21 +09001385 /* About to take EH action, set RECOVERED. Ignore actions on
1386 * slave links as master will do them again.
1387 */
1388 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
Tejun Heo1cdaf532006-07-03 16:07:26 +09001389 ap->pflags |= ATA_PFLAG_RECOVERED;
1390
Jeff Garzikba6a1302006-06-22 23:46:10 -04001391 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001392}
1393
1394/**
Tejun Heo47005f22006-06-19 18:27:23 +09001395 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001396* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001397 * @dev: target ATA dev for per-dev action (can be NULL)
1398 * @action: action just completed
1399 *
1400 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001401 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001402 *
1403 * LOCKING:
1404 * None.
1405 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001406void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1407 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001408{
Tejun Heo955e57d2007-08-06 18:36:23 +09001409 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001410
Tejun Heo955e57d2007-08-06 18:36:23 +09001411 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001412}
1413
1414/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001415 * ata_err_string - convert err_mask to descriptive string
1416 * @err_mask: error mask to convert to string
1417 *
1418 * Convert @err_mask to descriptive string. Errors are
1419 * prioritized according to severity and only the most severe
1420 * error is reported.
1421 *
1422 * LOCKING:
1423 * None.
1424 *
1425 * RETURNS:
1426 * Descriptive string for @err_mask
1427 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001428static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001429{
1430 if (err_mask & AC_ERR_HOST_BUS)
1431 return "host bus error";
1432 if (err_mask & AC_ERR_ATA_BUS)
1433 return "ATA bus error";
1434 if (err_mask & AC_ERR_TIMEOUT)
1435 return "timeout";
1436 if (err_mask & AC_ERR_HSM)
1437 return "HSM violation";
1438 if (err_mask & AC_ERR_SYSTEM)
1439 return "internal error";
1440 if (err_mask & AC_ERR_MEDIA)
1441 return "media error";
1442 if (err_mask & AC_ERR_INVALID)
1443 return "invalid argument";
1444 if (err_mask & AC_ERR_DEV)
1445 return "device error";
1446 return "unknown error";
1447}
1448
1449/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001450 * ata_read_log_page - read a specific log page
1451 * @dev: target device
1452 * @page: page to read
1453 * @buf: buffer to store read page
1454 * @sectors: number of sectors to read
1455 *
1456 * Read log page using READ_LOG_EXT command.
1457 *
1458 * LOCKING:
1459 * Kernel thread context (may sleep).
1460 *
1461 * RETURNS:
1462 * 0 on success, AC_ERR_* mask otherwise.
1463 */
1464static unsigned int ata_read_log_page(struct ata_device *dev,
1465 u8 page, void *buf, unsigned int sectors)
1466{
1467 struct ata_taskfile tf;
1468 unsigned int err_mask;
1469
1470 DPRINTK("read log page - page %d\n", page);
1471
1472 ata_tf_init(dev, &tf);
1473 tf.command = ATA_CMD_READ_LOG_EXT;
1474 tf.lbal = page;
1475 tf.nsect = sectors;
1476 tf.hob_nsect = sectors >> 8;
1477 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1478 tf.protocol = ATA_PROT_PIO;
1479
1480 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001481 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001482
1483 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1484 return err_mask;
1485}
1486
1487/**
1488 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1489 * @dev: Device to read log page 10h from
1490 * @tag: Resulting tag of the failed command
1491 * @tf: Resulting taskfile registers of the failed command
1492 *
1493 * Read log page 10h to obtain NCQ error details and clear error
1494 * condition.
1495 *
1496 * LOCKING:
1497 * Kernel thread context (may sleep).
1498 *
1499 * RETURNS:
1500 * 0 on success, -errno otherwise.
1501 */
1502static int ata_eh_read_log_10h(struct ata_device *dev,
1503 int *tag, struct ata_taskfile *tf)
1504{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001505 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001506 unsigned int err_mask;
1507 u8 csum;
1508 int i;
1509
1510 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1511 if (err_mask)
1512 return -EIO;
1513
1514 csum = 0;
1515 for (i = 0; i < ATA_SECT_SIZE; i++)
1516 csum += buf[i];
1517 if (csum)
1518 ata_dev_printk(dev, KERN_WARNING,
1519 "invalid checksum 0x%x on log page 10h\n", csum);
1520
1521 if (buf[0] & 0x80)
1522 return -ENOENT;
1523
1524 *tag = buf[0] & 0x1f;
1525
1526 tf->command = buf[2];
1527 tf->feature = buf[3];
1528 tf->lbal = buf[4];
1529 tf->lbam = buf[5];
1530 tf->lbah = buf[6];
1531 tf->device = buf[7];
1532 tf->hob_lbal = buf[8];
1533 tf->hob_lbam = buf[9];
1534 tf->hob_lbah = buf[10];
1535 tf->nsect = buf[12];
1536 tf->hob_nsect = buf[13];
1537
1538 return 0;
1539}
1540
1541/**
Tejun Heo11fc33d2008-08-30 14:20:01 +02001542 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1543 * @dev: target ATAPI device
1544 * @r_sense_key: out parameter for sense_key
1545 *
1546 * Perform ATAPI TEST_UNIT_READY.
1547 *
1548 * LOCKING:
1549 * EH context (may sleep).
1550 *
1551 * RETURNS:
1552 * 0 on success, AC_ERR_* mask on failure.
1553 */
1554static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1555{
1556 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1557 struct ata_taskfile tf;
1558 unsigned int err_mask;
1559
1560 ata_tf_init(dev, &tf);
1561
1562 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1563 tf.command = ATA_CMD_PACKET;
1564 tf.protocol = ATAPI_PROT_NODATA;
1565
1566 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1567 if (err_mask == AC_ERR_DEV)
1568 *r_sense_key = tf.feature >> 4;
1569 return err_mask;
1570}
1571
1572/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001573 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1574 * @dev: device to perform REQUEST_SENSE to
1575 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
Tejun Heo3eabddb2008-06-10 18:28:05 +09001576 * @dfl_sense_key: default sense key to use
Tejun Heo022bdb02006-05-15 20:58:22 +09001577 *
1578 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1579 * SENSE. This function is EH helper.
1580 *
1581 * LOCKING:
1582 * Kernel thread context (may sleep).
1583 *
1584 * RETURNS:
1585 * 0 on success, AC_ERR_* mask on failure
1586 */
Tejun Heo3eabddb2008-06-10 18:28:05 +09001587static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1588 u8 *sense_buf, u8 dfl_sense_key)
Tejun Heo022bdb02006-05-15 20:58:22 +09001589{
Tejun Heo3eabddb2008-06-10 18:28:05 +09001590 u8 cdb[ATAPI_CDB_LEN] =
1591 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001592 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001593 struct ata_taskfile tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001594
1595 DPRINTK("ATAPI request sense\n");
1596
Tejun Heo022bdb02006-05-15 20:58:22 +09001597 /* FIXME: is this needed? */
1598 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1599
Albert Lee56287762007-04-02 11:30:46 +08001600 /* initialize sense_buf with the error register,
1601 * for the case where they are -not- overwritten
1602 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001603 sense_buf[0] = 0x70;
Tejun Heo3eabddb2008-06-10 18:28:05 +09001604 sense_buf[2] = dfl_sense_key;
Albert Lee56287762007-04-02 11:30:46 +08001605
Jeff Garzika617c092007-05-21 20:14:23 -04001606 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001607 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001608
Tejun Heo022bdb02006-05-15 20:58:22 +09001609 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1610 tf.command = ATA_CMD_PACKET;
1611
1612 /* is it pointless to prefer PIO for "safety reasons"? */
1613 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001614 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001615 tf.feature |= ATAPI_PKT_DMA;
1616 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001617 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001618 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1619 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001620 }
1621
1622 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001623 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001624}
1625
1626/**
1627 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001628 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001629 *
1630 * Analyze SError if available and further determine cause of
1631 * failure.
1632 *
1633 * LOCKING:
1634 * None.
1635 */
Tejun Heo02607312007-08-06 18:36:23 +09001636static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001637{
Tejun Heo02607312007-08-06 18:36:23 +09001638 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001639 u32 serror = ehc->i.serror;
1640 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001641 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001642
Tejun Heoe0614db2008-05-19 01:15:09 +09001643 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001644 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001645 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001646 }
1647 if (serror & SERR_PROTOCOL) {
1648 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001649 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001650 }
1651 if (serror & SERR_INTERNAL) {
1652 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001653 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001654 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001655
1656 /* Determine whether a hotplug event has occurred. Both
1657 * SError.N/X are considered hotplug events for enabled or
1658 * host links. For disabled PMP links, only N bit is
1659 * considered as X bit is left at 1 for link plugging.
1660 */
Tejun Heoeb0e85e2011-02-24 19:30:37 +01001661 if (link->lpm_policy > ATA_LPM_MAX_POWER)
Tejun Heo6b7ae952010-09-01 17:50:06 +02001662 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1663 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
Tejun Heof9df58c2007-09-23 13:14:13 +09001664 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1665 else
1666 hotplug_mask = SERR_PHYRDY_CHG;
1667
1668 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001669 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001670
1671 ehc->i.err_mask |= err_mask;
1672 ehc->i.action |= action;
1673}
1674
1675/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001676 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001677 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001678 *
1679 * Read log page 10h, determine the offending qc and acquire
1680 * error status TF. For NCQ device errors, all LLDDs have to do
1681 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1682 * care of the rest.
1683 *
1684 * LOCKING:
1685 * Kernel thread context (may sleep).
1686 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001687void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001688{
Tejun Heo02607312007-08-06 18:36:23 +09001689 struct ata_port *ap = link->ap;
1690 struct ata_eh_context *ehc = &link->eh_context;
1691 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001692 struct ata_queued_cmd *qc;
1693 struct ata_taskfile tf;
1694 int tag, rc;
1695
1696 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001697 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001698 return;
1699
1700 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001701 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001702 return;
1703
1704 /* has LLDD analyzed already? */
1705 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1706 qc = __ata_qc_from_tag(ap, tag);
1707
1708 if (!(qc->flags & ATA_QCFLAG_FAILED))
1709 continue;
1710
1711 if (qc->err_mask)
1712 return;
1713 }
1714
1715 /* okay, this error is ours */
Jeff Garzika09bf4c2010-04-22 21:59:13 -04001716 memset(&tf, 0, sizeof(tf));
Tejun Heoe8ee8452006-05-15 21:03:46 +09001717 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1718 if (rc) {
Tejun Heo02607312007-08-06 18:36:23 +09001719 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001720 "(errno=%d)\n", rc);
1721 return;
1722 }
1723
Tejun Heo02607312007-08-06 18:36:23 +09001724 if (!(link->sactive & (1 << tag))) {
1725 ata_link_printk(link, KERN_ERR, "log page 10h reported "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001726 "inactive tag %d\n", tag);
1727 return;
1728 }
1729
1730 /* we've got the perpetrator, condemn it */
1731 qc = __ata_qc_from_tag(ap, tag);
1732 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001733 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001734 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001735 ehc->i.err_mask &= ~AC_ERR_DEV;
1736}
1737
1738/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001739 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1740 * @qc: qc to analyze
1741 * @tf: Taskfile registers to analyze
1742 *
1743 * Analyze taskfile of @qc and further determine cause of
1744 * failure. This function also requests ATAPI sense data if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001745 * available.
Tejun Heo022bdb02006-05-15 20:58:22 +09001746 *
1747 * LOCKING:
1748 * Kernel thread context (may sleep).
1749 *
1750 * RETURNS:
1751 * Determined recovery action
1752 */
1753static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1754 const struct ata_taskfile *tf)
1755{
1756 unsigned int tmp, action = 0;
1757 u8 stat = tf->command, err = tf->feature;
1758
1759 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1760 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001761 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001762 }
1763
Tejun Heoa51d6442007-03-20 15:24:11 +09001764 if (stat & (ATA_ERR | ATA_DF))
1765 qc->err_mask |= AC_ERR_DEV;
1766 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001767 return 0;
1768
1769 switch (qc->dev->class) {
1770 case ATA_DEV_ATA:
1771 if (err & ATA_ICRC)
1772 qc->err_mask |= AC_ERR_ATA_BUS;
1773 if (err & ATA_UNC)
1774 qc->err_mask |= AC_ERR_MEDIA;
1775 if (err & ATA_IDNF)
1776 qc->err_mask |= AC_ERR_INVALID;
1777 break;
1778
1779 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001780 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Tejun Heo3eabddb2008-06-10 18:28:05 +09001781 tmp = atapi_eh_request_sense(qc->dev,
1782 qc->scsicmd->sense_buffer,
1783 qc->result_tf.feature >> 4);
Tejun Heoa569a302006-11-21 10:40:51 +09001784 if (!tmp) {
1785 /* ATA_QCFLAG_SENSE_VALID is used to
1786 * tell atapi_qc_complete() that sense
1787 * data is already valid.
1788 *
1789 * TODO: interpret sense data and set
1790 * appropriate err_mask.
1791 */
1792 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1793 } else
1794 qc->err_mask |= tmp;
1795 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001796 }
1797
1798 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001799 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001800
1801 return action;
1802}
1803
Tejun Heo76326ac2007-11-27 19:28:59 +09001804static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1805 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001806{
Tejun Heo76326ac2007-11-27 19:28:59 +09001807 int base = 0;
1808
1809 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1810 *xfer_ok = 1;
1811
1812 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001813 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001814
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001815 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001816 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001817
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001818 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001819 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001820
Tejun Heo3884f7b2007-11-27 19:28:56 +09001821 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001822 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001823 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001824 if ((err_mask &
1825 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001826 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001827 }
1828
1829 return 0;
1830}
1831
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001832struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001833 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001834 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001835 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001836};
1837
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001838static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001839{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001840 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001841 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001842
Gwendal Grignoud9027472010-05-25 12:31:38 -07001843 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
Tejun Heo022bdb02006-05-15 20:58:22 +09001844 return -1;
1845
Tejun Heo76326ac2007-11-27 19:28:59 +09001846 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1847 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001848 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001849
Tejun Heo022bdb02006-05-15 20:58:22 +09001850 return 0;
1851}
1852
1853/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001854 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001855 * @dev: Device of interest
1856 *
1857 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001858 * whether NCQ needs to be turned off, transfer speed should be
1859 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001860 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001861 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001862 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001863 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1864 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001865 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001866 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001867 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001868 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1869 * data transfer hasn't been verified.
1870 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001871 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001872 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001873 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001874 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001875 * SPEED_DOWN : Speed down transfer speed but don't fall back
1876 * to PIO.
1877 *
1878 * FALLBACK_TO_PIO : Fall back to PIO.
1879 *
1880 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001881 * taken per error. An action triggered by non-DUBIOUS errors
1882 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1883 * This is to expedite speed down decisions right after device is
1884 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001885 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001886 * The followings are speed down rules. #1 and #2 deal with
1887 * DUBIOUS errors.
1888 *
1889 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1890 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1891 *
1892 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1893 * occurred during last 5 mins, NCQ_OFF.
1894 *
1895 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001896 * occurred during last 5 mins, FALLBACK_TO_PIO
Tejun Heo3884f7b2007-11-27 19:28:56 +09001897 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001898 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001899 * during last 10 mins, NCQ_OFF.
1900 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001901 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001902 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001903 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001904 * LOCKING:
1905 * Inherited from caller.
1906 *
1907 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001908 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001909 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001910static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001911{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001912 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1913 u64 j64 = get_jiffies_64();
1914 struct speed_down_verdict_arg arg;
1915 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001916
Tejun Heo3884f7b2007-11-27 19:28:56 +09001917 /* scan past 5 mins of error history */
1918 memset(&arg, 0, sizeof(arg));
1919 arg.since = j64 - min(j64, j5mins);
1920 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1921
Tejun Heo76326ac2007-11-27 19:28:59 +09001922 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1923 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1924 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1925 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1926
1927 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1928 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1929 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1930
Tejun Heo3884f7b2007-11-27 19:28:56 +09001931 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1932 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001933 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001934 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1935
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001936 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001937 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001938 arg.since = j64 - min(j64, j10mins);
1939 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001940
Tejun Heo3884f7b2007-11-27 19:28:56 +09001941 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1942 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001943 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001944
1945 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1946 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001947 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001948 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001949
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001950 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001951}
1952
1953/**
1954 * ata_eh_speed_down - record error and speed down if necessary
1955 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001956 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001957 * @err_mask: err_mask of the error
1958 *
1959 * Record error and examine error history to determine whether
1960 * adjusting transmission speed is necessary. It also sets
1961 * transmission limits appropriately if such adjustment is
1962 * necessary.
1963 *
1964 * LOCKING:
1965 * Kernel thread context (may sleep).
1966 *
1967 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001968 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001969 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001970static unsigned int ata_eh_speed_down(struct ata_device *dev,
1971 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001972{
Tejun Heob1c72912008-07-31 17:02:43 +09001973 struct ata_link *link = ata_dev_phys_link(dev);
Tejun Heo76326ac2007-11-27 19:28:59 +09001974 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001975 unsigned int verdict;
1976 unsigned int action = 0;
1977
1978 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001979 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001980 return 0;
1981
1982 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001983 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001984 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001985
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001986 /* turn off NCQ? */
1987 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1988 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1989 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1990 dev->flags |= ATA_DFLAG_NCQ_OFF;
1991 ata_dev_printk(dev, KERN_WARNING,
1992 "NCQ disabled due to excessive errors\n");
1993 goto done;
1994 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001995
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001996 /* speed down? */
1997 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1998 /* speed down SATA link speed if possible */
Tejun Heoa07d4992009-01-29 20:31:33 +09001999 if (sata_down_spd_limit(link, 0) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002000 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002001 goto done;
2002 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002003
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002004 /* lower transfer mode */
2005 if (dev->spdn_cnt < 2) {
2006 static const int dma_dnxfer_sel[] =
2007 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2008 static const int pio_dnxfer_sel[] =
2009 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2010 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09002011
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002012 if (dev->xfer_shift != ATA_SHIFT_PIO)
2013 sel = dma_dnxfer_sel[dev->spdn_cnt];
2014 else
2015 sel = pio_dnxfer_sel[dev->spdn_cnt];
2016
2017 dev->spdn_cnt++;
2018
2019 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002020 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002021 goto done;
2022 }
2023 }
2024 }
2025
2026 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09002027 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002028 */
2029 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09002030 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002031 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2032 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2033 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09002034 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002035 goto done;
2036 }
2037 }
2038
Tejun Heo022bdb02006-05-15 20:58:22 +09002039 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002040 done:
2041 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09002042 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2043 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002044 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09002045}
2046
2047/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002048 * ata_eh_link_autopsy - analyze error and determine recovery action
2049 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09002050 *
Tejun Heo02607312007-08-06 18:36:23 +09002051 * Analyze why @link failed and determine which recovery actions
2052 * are needed. This function also sets more detailed AC_ERR_*
2053 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09002054 *
2055 * LOCKING:
2056 * Kernel thread context (may sleep).
2057 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002058static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002059{
Tejun Heo02607312007-08-06 18:36:23 +09002060 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09002061 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09002062 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002063 unsigned int all_err_mask = 0, eflags = 0;
2064 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09002065 u32 serror;
2066 int rc;
2067
2068 DPRINTK("ENTER\n");
2069
Tejun Heo1cdaf532006-07-03 16:07:26 +09002070 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2071 return;
2072
Tejun Heo022bdb02006-05-15 20:58:22 +09002073 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09002074 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09002075 if (rc == 0) {
2076 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09002077 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09002078 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09002079 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09002080 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09002081 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09002082 ehc->i.err_mask |= AC_ERR_OTHER;
2083 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002084
Tejun Heoe8ee8452006-05-15 21:03:46 +09002085 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09002086 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09002087
Tejun Heo022bdb02006-05-15 20:58:22 +09002088 /* any real error trumps AC_ERR_OTHER */
2089 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2090 ehc->i.err_mask &= ~AC_ERR_OTHER;
2091
2092 all_err_mask |= ehc->i.err_mask;
2093
2094 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2095 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2096
Tejun Heob1c72912008-07-31 17:02:43 +09002097 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2098 ata_dev_phys_link(qc->dev) != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002099 continue;
2100
2101 /* inherit upper level err_mask */
2102 qc->err_mask |= ehc->i.err_mask;
2103
Tejun Heo022bdb02006-05-15 20:58:22 +09002104 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002105 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002106
2107 /* DEV errors are probably spurious in case of ATA_BUS error */
2108 if (qc->err_mask & AC_ERR_ATA_BUS)
2109 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2110 AC_ERR_INVALID);
2111
2112 /* any real error trumps unknown error */
2113 if (qc->err_mask & ~AC_ERR_OTHER)
2114 qc->err_mask &= ~AC_ERR_OTHER;
2115
2116 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09002117 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09002118 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09002119
Tejun Heo03faab72008-03-27 19:14:24 +09002120 /* determine whether the command is worth retrying */
Tejun Heo534ead72010-01-14 16:18:09 +09002121 if (qc->flags & ATA_QCFLAG_IO ||
2122 (!(qc->err_mask & AC_ERR_INVALID) &&
2123 qc->err_mask != AC_ERR_DEV))
Tejun Heo03faab72008-03-27 19:14:24 +09002124 qc->flags |= ATA_QCFLAG_RETRY;
2125
Tejun Heo022bdb02006-05-15 20:58:22 +09002126 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002127 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09002128 all_err_mask |= qc->err_mask;
2129 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09002130 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09002131 }
2132
Tejun Heoa20f33f2006-05-16 12:58:24 +09002133 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09002134 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09002135 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09002136 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002137 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2138 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09002139 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002140
Tejun Heodfcc1732007-10-31 10:17:05 +09002141 /* If we have offending qcs and the associated failed device,
2142 * perform per-dev EH action only on the offending device.
2143 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002144 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09002145 ehc->i.dev_action[ehc->i.dev->devno] |=
2146 ehc->i.action & ATA_EH_PERDEV_MASK;
2147 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09002148 }
2149
Tejun Heo2695e362008-01-10 13:41:23 +09002150 /* propagate timeout to host link */
2151 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2152 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2153
2154 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09002155 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09002156 if (!dev && ((ata_link_max_devices(link) == 1 &&
2157 ata_dev_enabled(link->device))))
2158 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09002159
Tejun Heo76326ac2007-11-27 19:28:59 +09002160 if (dev) {
2161 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2162 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002163 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09002164 }
Tejun Heodfcc1732007-10-31 10:17:05 +09002165
Tejun Heo022bdb02006-05-15 20:58:22 +09002166 DPRINTK("EXIT\n");
2167}
2168
2169/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002170 * ata_eh_autopsy - analyze error and determine recovery action
2171 * @ap: host port to perform autopsy on
2172 *
2173 * Analyze all links of @ap and determine why they failed and
2174 * which recovery actions are needed.
2175 *
2176 * LOCKING:
2177 * Kernel thread context (may sleep).
2178 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002179void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002180{
2181 struct ata_link *link;
2182
Tejun Heo1eca4362008-11-03 20:03:17 +09002183 ata_for_each_link(link, ap, EDGE)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002184 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09002185
Tejun Heob1c72912008-07-31 17:02:43 +09002186 /* Handle the frigging slave link. Autopsy is done similarly
2187 * but actions and flags are transferred over to the master
2188 * link and handled from there.
2189 */
2190 if (ap->slave_link) {
2191 struct ata_eh_context *mehc = &ap->link.eh_context;
2192 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2193
Tejun Heo848e4c62008-10-21 14:26:39 +09002194 /* transfer control flags from master to slave */
2195 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2196
2197 /* perform autopsy on the slave link */
Tejun Heob1c72912008-07-31 17:02:43 +09002198 ata_eh_link_autopsy(ap->slave_link);
2199
Tejun Heo848e4c62008-10-21 14:26:39 +09002200 /* transfer actions from slave to master and clear slave */
Tejun Heob1c72912008-07-31 17:02:43 +09002201 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2202 mehc->i.action |= sehc->i.action;
2203 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2204 mehc->i.flags |= sehc->i.flags;
2205 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2206 }
2207
Tejun Heo2695e362008-01-10 13:41:23 +09002208 /* Autopsy of fanout ports can affect host link autopsy.
2209 * Perform host link autopsy last.
2210 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002211 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09002212 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09002213}
2214
2215/**
Robert Hancock65211482009-07-14 20:43:39 -06002216 * ata_get_cmd_descript - get description for ATA command
2217 * @command: ATA command code to get description for
2218 *
2219 * Return a textual description of the given command, or NULL if the
2220 * command is not known.
2221 *
2222 * LOCKING:
2223 * None
2224 */
2225const char *ata_get_cmd_descript(u8 command)
2226{
2227#ifdef CONFIG_ATA_VERBOSE_ERROR
2228 static const struct
2229 {
2230 u8 command;
2231 const char *text;
2232 } cmd_descr[] = {
2233 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2234 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2235 { ATA_CMD_STANDBY, "STANDBY" },
2236 { ATA_CMD_IDLE, "IDLE" },
2237 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2238 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2239 { ATA_CMD_NOP, "NOP" },
2240 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2241 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2242 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2243 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2244 { ATA_CMD_SERVICE, "SERVICE" },
2245 { ATA_CMD_READ, "READ DMA" },
2246 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2247 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2248 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2249 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2250 { ATA_CMD_WRITE, "WRITE DMA" },
2251 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2252 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2253 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2254 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2255 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2256 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2257 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2258 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2259 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2260 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2261 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2262 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2263 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2264 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2265 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2266 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2267 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2268 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2269 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2270 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2271 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2272 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2273 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2274 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2275 { ATA_CMD_SLEEP, "SLEEP" },
2276 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2277 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2278 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2279 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2280 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2281 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2282 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2283 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2284 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2285 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2286 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2287 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2288 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2289 { ATA_CMD_PMP_READ, "READ BUFFER" },
2290 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2291 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2292 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2293 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2294 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2295 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2296 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2297 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2298 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2299 { ATA_CMD_SMART, "SMART" },
2300 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2301 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
FUJITA Tomonoriacad762722010-07-05 15:45:20 +09002302 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
Robert Hancock65211482009-07-14 20:43:39 -06002303 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2304 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2305 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2306 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2307 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2308 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2309 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2310 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2311 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2312 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2313 { ATA_CMD_RESTORE, "RECALIBRATE" },
2314 { 0, NULL } /* terminate list */
2315 };
2316
2317 unsigned int i;
2318 for (i = 0; cmd_descr[i].text; i++)
2319 if (cmd_descr[i].command == command)
2320 return cmd_descr[i].text;
2321#endif
2322
2323 return NULL;
2324}
2325
2326/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002327 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09002328 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09002329 *
2330 * Report EH to user.
2331 *
2332 * LOCKING:
2333 * None.
2334 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002335static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002336{
Tejun Heo02607312007-08-06 18:36:23 +09002337 struct ata_port *ap = link->ap;
2338 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002339 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09002340 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09002341 int tag, nr_failed = 0;
2342
Tejun Heo94ff3d52007-10-09 14:57:56 +09002343 if (ehc->i.flags & ATA_EHI_QUIET)
2344 return;
2345
Tejun Heo022bdb02006-05-15 20:58:22 +09002346 desc = NULL;
2347 if (ehc->i.desc[0] != '\0')
2348 desc = ehc->i.desc;
2349
2350 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2351 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2352
Tejun Heob1c72912008-07-31 17:02:43 +09002353 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2354 ata_dev_phys_link(qc->dev) != link ||
Tejun Heoe027bd32007-10-26 16:19:26 +09002355 ((qc->flags & ATA_QCFLAG_QUIET) &&
2356 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002357 continue;
2358 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2359 continue;
2360
2361 nr_failed++;
2362 }
2363
2364 if (!nr_failed && !ehc->i.err_mask)
2365 return;
2366
2367 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09002368 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09002369 frozen = " frozen";
2370
Tejun Heoa1e10f72007-08-18 13:28:49 +09002371 memset(tries_buf, 0, sizeof(tries_buf));
2372 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2373 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2374 ap->eh_tries);
2375
Tejun Heo022bdb02006-05-15 20:58:22 +09002376 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09002377 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002378 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2379 ehc->i.err_mask, link->sactive, ehc->i.serror,
2380 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002381 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09002382 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002383 } else {
Tejun Heo02607312007-08-06 18:36:23 +09002384 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002385 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2386 ehc->i.err_mask, link->sactive, ehc->i.serror,
2387 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002388 if (desc)
Tejun Heo02607312007-08-06 18:36:23 +09002389 ata_link_printk(link, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002390 }
2391
Robert Hancock65211482009-07-14 20:43:39 -06002392#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002393 if (ehc->i.serror)
Tejun Heoda0e21d2008-07-31 16:08:38 +09002394 ata_link_printk(link, KERN_ERR,
Robert Hancock1333e192007-10-02 11:22:02 -04002395 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2396 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2397 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2398 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2399 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2400 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2401 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2402 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2403 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2404 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2405 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2406 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2407 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2408 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2409 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2410 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2411 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002412 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002413#endif
Robert Hancock1333e192007-10-02 11:22:02 -04002414
Tejun Heo022bdb02006-05-15 20:58:22 +09002415 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2416 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09002417 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09002418 const u8 *cdb = qc->cdb;
2419 char data_buf[20] = "";
2420 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002421
Tejun Heo02607312007-08-06 18:36:23 +09002422 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
Tejun Heob1c72912008-07-31 17:02:43 +09002423 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002424 continue;
2425
Tejun Heoabb6a882007-11-28 23:16:09 +09002426 if (qc->dma_dir != DMA_NONE) {
2427 static const char *dma_str[] = {
2428 [DMA_BIDIRECTIONAL] = "bidi",
2429 [DMA_TO_DEVICE] = "out",
2430 [DMA_FROM_DEVICE] = "in",
2431 };
2432 static const char *prot_str[] = {
2433 [ATA_PROT_PIO] = "pio",
2434 [ATA_PROT_DMA] = "dma",
2435 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05002436 [ATAPI_PROT_PIO] = "pio",
2437 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09002438 };
2439
2440 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2441 prot_str[qc->tf.protocol], qc->nbytes,
2442 dma_str[qc->dma_dir]);
2443 }
2444
Robert Hancock65211482009-07-14 20:43:39 -06002445 if (ata_is_atapi(qc->tf.protocol)) {
2446 if (qc->scsicmd)
2447 scsi_print_command(qc->scsicmd);
2448 else
2449 snprintf(cdb_buf, sizeof(cdb_buf),
Tejun Heoabb6a882007-11-28 23:16:09 +09002450 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2451 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2452 cdb[0], cdb[1], cdb[2], cdb[3],
2453 cdb[4], cdb[5], cdb[6], cdb[7],
2454 cdb[8], cdb[9], cdb[10], cdb[11],
2455 cdb[12], cdb[13], cdb[14], cdb[15]);
Robert Hancock65211482009-07-14 20:43:39 -06002456 } else {
2457 const char *descr = ata_get_cmd_descript(cmd->command);
2458 if (descr)
2459 ata_dev_printk(qc->dev, KERN_ERR,
2460 "failed command: %s\n", descr);
2461 }
Tejun Heoabb6a882007-11-28 23:16:09 +09002462
Tejun Heo8a937582006-11-14 22:36:12 +09002463 ata_dev_printk(qc->dev, KERN_ERR,
2464 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002465 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002466 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002467 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002468 cmd->command, cmd->feature, cmd->nsect,
2469 cmd->lbal, cmd->lbam, cmd->lbah,
2470 cmd->hob_feature, cmd->hob_nsect,
2471 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002472 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002473 res->command, res->feature, res->nsect,
2474 res->lbal, res->lbam, res->lbah,
2475 res->hob_feature, res->hob_nsect,
2476 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002477 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2478 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002479
Robert Hancock65211482009-07-14 20:43:39 -06002480#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002481 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002482 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002483 if (res->command & ATA_BUSY)
2484 ata_dev_printk(qc->dev, KERN_ERR,
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002485 "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002486 else
2487 ata_dev_printk(qc->dev, KERN_ERR,
2488 "status: { %s%s%s%s}\n",
2489 res->command & ATA_DRDY ? "DRDY " : "",
2490 res->command & ATA_DF ? "DF " : "",
2491 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002492 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002493 }
2494
2495 if (cmd->command != ATA_CMD_PACKET &&
2496 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2497 ATA_ABORTED)))
2498 ata_dev_printk(qc->dev, KERN_ERR,
2499 "error: { %s%s%s%s}\n",
2500 res->feature & ATA_ICRC ? "ICRC " : "",
2501 res->feature & ATA_UNC ? "UNC " : "",
2502 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002503 res->feature & ATA_ABORTED ? "ABRT " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002504#endif
Tejun Heo022bdb02006-05-15 20:58:22 +09002505 }
2506}
2507
Tejun Heo9b1e2652007-08-06 18:36:24 +09002508/**
2509 * ata_eh_report - report error handling to user
2510 * @ap: ATA port to report EH about
2511 *
2512 * Report EH to user.
2513 *
2514 * LOCKING:
2515 * None.
2516 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002517void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002518{
2519 struct ata_link *link;
2520
Tejun Heo1eca4362008-11-03 20:03:17 +09002521 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002522 ata_eh_link_report(link);
2523}
2524
Tejun Heocc0680a2007-08-06 18:36:23 +09002525static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heob1c72912008-07-31 17:02:43 +09002526 unsigned int *classes, unsigned long deadline,
2527 bool clear_classes)
Tejun Heod87fa382006-05-31 18:28:24 +09002528{
Tejun Heof58229f2007-08-06 18:36:23 +09002529 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002530
Tejun Heob1c72912008-07-31 17:02:43 +09002531 if (clear_classes)
Tejun Heo1eca4362008-11-03 20:03:17 +09002532 ata_for_each_dev(dev, link, ALL)
Tejun Heob1c72912008-07-31 17:02:43 +09002533 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002534
Tejun Heof0465192008-05-19 01:15:08 +09002535 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002536}
2537
Tejun Heoae791c02007-09-23 13:14:12 +09002538static int ata_eh_followup_srst_needed(struct ata_link *link,
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002539 int rc, const unsigned int *classes)
Tejun Heo664faf02006-05-31 18:27:50 +09002540{
Tejun Heo45db2f62008-04-08 01:46:56 +09002541 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002542 return 0;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002543 if (rc == -EAGAIN)
2544 return 1;
Tejun Heo071f44b2008-04-07 22:47:22 +09002545 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002546 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002547 return 0;
2548}
2549
Tejun Heofb7fd612007-09-23 13:14:12 +09002550int ata_eh_reset(struct ata_link *link, int classify,
2551 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2552 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002553{
Tejun Heoafaa5c32007-10-09 15:06:10 +09002554 struct ata_port *ap = link->ap;
Tejun Heob1c72912008-07-31 17:02:43 +09002555 struct ata_link *slave = ap->slave_link;
Tejun Heo936fd732007-08-06 18:36:23 +09002556 struct ata_eh_context *ehc = &link->eh_context;
Bartlomiej Zolnierkiewicz705d2012009-07-26 16:21:01 +02002557 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
Tejun Heo664faf02006-05-31 18:27:50 +09002558 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002559 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002560 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heod8af0eb2008-05-20 02:17:53 +09002561 int max_tries = 0, try = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002562 struct ata_link *failed_link;
Tejun Heof58229f2007-08-06 18:36:23 +09002563 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002564 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002565 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002566 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002567 u32 sstatus;
Tejun Heob1c72912008-07-31 17:02:43 +09002568 int nr_unknown, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002569
Tejun Heo932648b2008-05-19 01:15:06 +09002570 /*
2571 * Prepare to reset
2572 */
Tejun Heod8af0eb2008-05-20 02:17:53 +09002573 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2574 max_tries++;
Tejun Heo05944bd2008-08-13 20:19:09 +09002575 if (link->flags & ATA_LFLAG_NO_HRST)
2576 hardreset = NULL;
2577 if (link->flags & ATA_LFLAG_NO_SRST)
2578 softreset = NULL;
Tejun Heod8af0eb2008-05-20 02:17:53 +09002579
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002580 /* make sure each reset attempt is at least COOL_DOWN apart */
Tejun Heo19b72322008-11-04 17:08:40 +09002581 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2582 now = jiffies;
2583 WARN_ON(time_after(ehc->last_reset, now));
2584 deadline = ata_deadline(ehc->last_reset,
2585 ATA_EH_RESET_COOL_DOWN);
2586 if (time_before(now, deadline))
2587 schedule_timeout_uninterruptible(deadline - now);
2588 }
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002589
Tejun Heoafaa5c32007-10-09 15:06:10 +09002590 spin_lock_irqsave(ap->lock, flags);
2591 ap->pflags |= ATA_PFLAG_RESETTING;
2592 spin_unlock_irqrestore(ap->lock, flags);
2593
Tejun Heocf480622008-01-24 00:05:14 +09002594 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002595
Tejun Heo1eca4362008-11-03 20:03:17 +09002596 ata_for_each_dev(dev, link, ALL) {
Tejun Heocdeab112007-10-29 16:41:09 +09002597 /* If we issue an SRST then an ATA drive (not ATAPI)
2598 * may change configuration and be in PIO0 timing. If
2599 * we do a hard reset (or are coming from power on)
2600 * this is true for ATA or ATAPI. Until we've set a
2601 * suitable controller mode we should not touch the
2602 * bus as we may be talking too fast.
2603 */
2604 dev->pio_mode = XFER_PIO_0;
2605
2606 /* If the controller has a pio mode setup function
2607 * then use it to set the chipset to rights. Don't
2608 * touch the DMA setup as that will be dealt with when
2609 * configuring devices.
2610 */
2611 if (ap->ops->set_piomode)
2612 ap->ops->set_piomode(ap, dev);
2613 }
2614
Tejun Heocf480622008-01-24 00:05:14 +09002615 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002616 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002617 ehc->i.action &= ~ATA_EH_RESET;
2618 if (hardreset) {
2619 reset = hardreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002620 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002621 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002622 reset = softreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002623 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heocf480622008-01-24 00:05:14 +09002624 }
Tejun Heof5914a42006-05-31 18:27:48 +09002625
2626 if (prereset) {
Tejun Heob1c72912008-07-31 17:02:43 +09002627 unsigned long deadline = ata_deadline(jiffies,
2628 ATA_EH_PRERESET_TIMEOUT);
2629
2630 if (slave) {
2631 sehc->i.action &= ~ATA_EH_RESET;
2632 sehc->i.action |= ehc->i.action;
2633 }
2634
2635 rc = prereset(link, deadline);
2636
2637 /* If present, do prereset on slave link too. Reset
2638 * is skipped iff both master and slave links report
2639 * -ENOENT or clear ATA_EH_RESET.
2640 */
2641 if (slave && (rc == 0 || rc == -ENOENT)) {
2642 int tmp;
2643
2644 tmp = prereset(slave, deadline);
2645 if (tmp != -ENOENT)
2646 rc = tmp;
2647
2648 ehc->i.action |= sehc->i.action;
2649 }
2650
Tejun Heof5914a42006-05-31 18:27:48 +09002651 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002652 if (rc == -ENOENT) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002653 ata_link_printk(link, KERN_DEBUG,
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002654 "port disabled. ignoring.\n");
Tejun Heocf480622008-01-24 00:05:14 +09002655 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002656
Tejun Heo1eca4362008-11-03 20:03:17 +09002657 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09002658 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002659
2660 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002661 } else
Tejun Heocc0680a2007-08-06 18:36:23 +09002662 ata_link_printk(link, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09002663 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002664 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002665 }
Tejun Heof5914a42006-05-31 18:27:48 +09002666
Tejun Heo932648b2008-05-19 01:15:06 +09002667 /* prereset() might have cleared ATA_EH_RESET. If so,
Tejun Heod6515e62009-03-04 15:59:30 +09002668 * bang classes, thaw and return.
Tejun Heo932648b2008-05-19 01:15:06 +09002669 */
2670 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
Tejun Heo1eca4362008-11-03 20:03:17 +09002671 ata_for_each_dev(dev, link, ALL)
Tejun Heo932648b2008-05-19 01:15:06 +09002672 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heod6515e62009-03-04 15:59:30 +09002673 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2674 ata_is_host_link(link))
2675 ata_eh_thaw_port(ap);
Tejun Heo932648b2008-05-19 01:15:06 +09002676 rc = 0;
2677 goto out;
2678 }
Tejun Heof5914a42006-05-31 18:27:48 +09002679 }
2680
Tejun Heo022bdb02006-05-15 20:58:22 +09002681 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002682 /*
2683 * Perform reset
2684 */
Tejun Heodc98c322008-05-19 01:15:07 +09002685 if (ata_is_host_link(link))
2686 ata_eh_freeze_port(ap);
2687
Tejun Heo341c2c92008-05-20 02:17:51 +09002688 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
Tejun Heo31daabd2007-02-02 16:50:52 +09002689
Tejun Heo932648b2008-05-19 01:15:06 +09002690 if (reset) {
2691 if (verbose)
2692 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2693 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002694
Tejun Heo932648b2008-05-19 01:15:06 +09002695 /* mark that this EH session started with reset */
Tejun Heo19b72322008-11-04 17:08:40 +09002696 ehc->last_reset = jiffies;
Tejun Heo932648b2008-05-19 01:15:06 +09002697 if (reset == hardreset)
2698 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2699 else
2700 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002701
Tejun Heob1c72912008-07-31 17:02:43 +09002702 rc = ata_do_reset(link, reset, classes, deadline, true);
2703 if (rc && rc != -EAGAIN) {
2704 failed_link = link;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002705 goto fail;
Tejun Heob1c72912008-07-31 17:02:43 +09002706 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002707
Tejun Heob1c72912008-07-31 17:02:43 +09002708 /* hardreset slave link if existent */
2709 if (slave && reset == hardreset) {
2710 int tmp;
2711
2712 if (verbose)
2713 ata_link_printk(slave, KERN_INFO,
2714 "hard resetting link\n");
2715
2716 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2717 tmp = ata_do_reset(slave, reset, classes, deadline,
2718 false);
2719 switch (tmp) {
2720 case -EAGAIN:
2721 rc = -EAGAIN;
2722 case 0:
2723 break;
2724 default:
2725 failed_link = slave;
2726 rc = tmp;
2727 goto fail;
2728 }
2729 }
2730
2731 /* perform follow-up SRST if necessary */
Tejun Heo932648b2008-05-19 01:15:06 +09002732 if (reset == hardreset &&
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002733 ata_eh_followup_srst_needed(link, rc, classes)) {
Tejun Heo932648b2008-05-19 01:15:06 +09002734 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002735
Tejun Heo932648b2008-05-19 01:15:06 +09002736 if (!reset) {
2737 ata_link_printk(link, KERN_ERR,
2738 "follow-up softreset required "
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002739 "but no softreset available\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002740 failed_link = link;
Tejun Heo932648b2008-05-19 01:15:06 +09002741 rc = -EINVAL;
2742 goto fail;
2743 }
2744
2745 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002746 rc = ata_do_reset(link, reset, classes, deadline, true);
Tejun Heofe2c4d02009-07-08 12:16:37 +09002747 if (rc) {
2748 failed_link = link;
2749 goto fail;
2750 }
Tejun Heo664faf02006-05-31 18:27:50 +09002751 }
Tejun Heo932648b2008-05-19 01:15:06 +09002752 } else {
2753 if (verbose)
2754 ata_link_printk(link, KERN_INFO, "no reset method "
2755 "available, skipping reset\n");
2756 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2757 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002758 }
2759
Tejun Heo932648b2008-05-19 01:15:06 +09002760 /*
2761 * Post-reset processing
2762 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002763 ata_for_each_dev(dev, link, ALL) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002764 /* After the reset, the device state is PIO 0 and the
2765 * controller state is undefined. Reset also wakes up
2766 * drives from sleeping mode.
2767 */
2768 dev->pio_mode = XFER_PIO_0;
2769 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002770
Tejun Heo3b761d32009-10-06 17:08:40 +09002771 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2772 continue;
2773
2774 /* apply class override */
2775 if (lflags & ATA_LFLAG_ASSUME_ATA)
2776 classes[dev->devno] = ATA_DEV_ATA;
2777 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2778 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
Tejun Heo022bdb02006-05-15 20:58:22 +09002779 }
2780
Tejun Heo416dc9e2007-10-31 10:17:03 +09002781 /* record current link speed */
2782 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2783 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heob1c72912008-07-31 17:02:43 +09002784 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2785 slave->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002786
Tejun Heodc98c322008-05-19 01:15:07 +09002787 /* thaw the port */
2788 if (ata_is_host_link(link))
2789 ata_eh_thaw_port(ap);
2790
Tejun Heof0465192008-05-19 01:15:08 +09002791 /* postreset() should clear hardware SError. Although SError
2792 * is cleared during link resume, clearing SError here is
2793 * necessary as some PHYs raise hotplug events after SRST.
2794 * This introduces race condition where hotplug occurs between
2795 * reset and here. This race is mediated by cross checking
2796 * link onlineness and classification result later.
2797 */
Tejun Heob1c72912008-07-31 17:02:43 +09002798 if (postreset) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002799 postreset(link, classes);
Tejun Heob1c72912008-07-31 17:02:43 +09002800 if (slave)
2801 postreset(slave, classes);
2802 }
Tejun Heo20952b62006-05-31 18:27:23 +09002803
Tejun Heo1e641062009-07-17 11:58:33 +09002804 /*
Tejun Heo8c56cac2011-05-25 13:19:39 +02002805 * Some controllers can't be frozen very well and may set spurious
2806 * error conditions during reset. Clear accumulated error
2807 * information and re-thaw the port if frozen. As reset is the
2808 * final recovery action and we cross check link onlineness against
2809 * device classification later, no hotplug event is lost by this.
Tejun Heo1e641062009-07-17 11:58:33 +09002810 */
Tejun Heof0465192008-05-19 01:15:08 +09002811 spin_lock_irqsave(link->ap->lock, flags);
Tejun Heo1e641062009-07-17 11:58:33 +09002812 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heob1c72912008-07-31 17:02:43 +09002813 if (slave)
Tejun Heo1e641062009-07-17 11:58:33 +09002814 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2815 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof0465192008-05-19 01:15:08 +09002816 spin_unlock_irqrestore(link->ap->lock, flags);
2817
Tejun Heo8c56cac2011-05-25 13:19:39 +02002818 if (ap->pflags & ATA_PFLAG_FROZEN)
2819 ata_eh_thaw_port(ap);
2820
Tejun Heo3b761d32009-10-06 17:08:40 +09002821 /*
2822 * Make sure onlineness and classification result correspond.
Tejun Heof0465192008-05-19 01:15:08 +09002823 * Hotplug could have happened during reset and some
2824 * controllers fail to wait while a drive is spinning up after
2825 * being hotplugged causing misdetection. By cross checking
Tejun Heo3b761d32009-10-06 17:08:40 +09002826 * link on/offlineness and classification result, those
2827 * conditions can be reliably detected and retried.
Tejun Heof0465192008-05-19 01:15:08 +09002828 */
Tejun Heob1c72912008-07-31 17:02:43 +09002829 nr_unknown = 0;
Tejun Heo1eca4362008-11-03 20:03:17 +09002830 ata_for_each_dev(dev, link, ALL) {
Tejun Heo3b761d32009-10-06 17:08:40 +09002831 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2832 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2833 ata_dev_printk(dev, KERN_DEBUG, "link online "
2834 "but device misclassifed\n");
2835 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002836 nr_unknown++;
Tejun Heo3b761d32009-10-06 17:08:40 +09002837 }
2838 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2839 if (ata_class_enabled(classes[dev->devno]))
2840 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2841 "clearing class %d to NONE\n",
2842 classes[dev->devno]);
2843 classes[dev->devno] = ATA_DEV_NONE;
2844 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2845 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2846 "clearing UNKNOWN to NONE\n");
2847 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002848 }
Tejun Heof0465192008-05-19 01:15:08 +09002849 }
2850
Tejun Heob1c72912008-07-31 17:02:43 +09002851 if (classify && nr_unknown) {
Tejun Heof0465192008-05-19 01:15:08 +09002852 if (try < max_tries) {
2853 ata_link_printk(link, KERN_WARNING, "link online but "
Tejun Heo3b761d32009-10-06 17:08:40 +09002854 "%d devices misclassified, retrying\n",
2855 nr_unknown);
Tejun Heob1c72912008-07-31 17:02:43 +09002856 failed_link = link;
Tejun Heof0465192008-05-19 01:15:08 +09002857 rc = -EAGAIN;
2858 goto fail;
2859 }
2860 ata_link_printk(link, KERN_WARNING,
Tejun Heo3b761d32009-10-06 17:08:40 +09002861 "link online but %d devices misclassified, "
2862 "device detection might fail\n", nr_unknown);
Tejun Heof0465192008-05-19 01:15:08 +09002863 }
2864
Tejun Heo416dc9e2007-10-31 10:17:03 +09002865 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002866 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002867 if (slave)
2868 ata_eh_done(slave, NULL, ATA_EH_RESET);
Tejun Heo6b7ae952010-09-01 17:50:06 +02002869 ehc->last_reset = jiffies; /* update to completion time */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002870 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo6b7ae952010-09-01 17:50:06 +02002871 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
Tejun Heoae791c02007-09-23 13:14:12 +09002872
Tejun Heo416dc9e2007-10-31 10:17:03 +09002873 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002874 out:
2875 /* clear hotplug flag */
2876 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heob1c72912008-07-31 17:02:43 +09002877 if (slave)
2878 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002879
2880 spin_lock_irqsave(ap->lock, flags);
2881 ap->pflags &= ~ATA_PFLAG_RESETTING;
2882 spin_unlock_irqrestore(ap->lock, flags);
2883
Tejun Heo022bdb02006-05-15 20:58:22 +09002884 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002885
2886 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002887 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2888 if (!ata_is_host_link(link) &&
2889 sata_scr_read(link, SCR_STATUS, &sstatus))
2890 rc = -ERESTART;
2891
Tejun Heo416dc9e2007-10-31 10:17:03 +09002892 if (rc == -ERESTART || try >= max_tries)
2893 goto out;
2894
2895 now = jiffies;
2896 if (time_before(now, deadline)) {
2897 unsigned long delta = deadline - now;
2898
Tejun Heob1c72912008-07-31 17:02:43 +09002899 ata_link_printk(failed_link, KERN_WARNING,
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002900 "reset failed (errno=%d), retrying in %u secs\n",
2901 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
Tejun Heo416dc9e2007-10-31 10:17:03 +09002902
Tejun Heoc0c362b2010-09-06 17:57:14 +02002903 ata_eh_release(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002904 while (delta)
2905 delta = schedule_timeout_uninterruptible(delta);
Tejun Heoc0c362b2010-09-06 17:57:14 +02002906 ata_eh_acquire(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002907 }
2908
Tejun Heob1c72912008-07-31 17:02:43 +09002909 if (try == max_tries - 1) {
Tejun Heoa07d4992009-01-29 20:31:33 +09002910 sata_down_spd_limit(link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002911 if (slave)
Tejun Heoa07d4992009-01-29 20:31:33 +09002912 sata_down_spd_limit(slave, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002913 } else if (rc == -EPIPE)
Tejun Heoa07d4992009-01-29 20:31:33 +09002914 sata_down_spd_limit(failed_link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002915
Tejun Heo416dc9e2007-10-31 10:17:03 +09002916 if (hardreset)
2917 reset = hardreset;
2918 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002919}
2920
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002921static inline void ata_eh_pull_park_action(struct ata_port *ap)
2922{
2923 struct ata_link *link;
2924 struct ata_device *dev;
2925 unsigned long flags;
2926
2927 /*
2928 * This function can be thought of as an extended version of
2929 * ata_eh_about_to_do() specially crafted to accommodate the
2930 * requirements of ATA_EH_PARK handling. Since the EH thread
2931 * does not leave the do {} while () loop in ata_eh_recover as
2932 * long as the timeout for a park request to *one* device on
2933 * the port has not expired, and since we still want to pick
2934 * up park requests to other devices on the same port or
2935 * timeout updates for the same device, we have to pull
2936 * ATA_EH_PARK actions from eh_info into eh_context.i
2937 * ourselves at the beginning of each pass over the loop.
2938 *
2939 * Additionally, all write accesses to &ap->park_req_pending
2940 * through INIT_COMPLETION() (see below) or complete_all()
2941 * (see ata_scsi_park_store()) are protected by the host lock.
2942 * As a result we have that park_req_pending.done is zero on
2943 * exit from this function, i.e. when ATA_EH_PARK actions for
2944 * *all* devices on port ap have been pulled into the
2945 * respective eh_context structs. If, and only if,
2946 * park_req_pending.done is non-zero by the time we reach
2947 * wait_for_completion_timeout(), another ATA_EH_PARK action
2948 * has been scheduled for at least one of the devices on port
2949 * ap and we have to cycle over the do {} while () loop in
2950 * ata_eh_recover() again.
2951 */
2952
2953 spin_lock_irqsave(ap->lock, flags);
2954 INIT_COMPLETION(ap->park_req_pending);
Tejun Heo1eca4362008-11-03 20:03:17 +09002955 ata_for_each_link(link, ap, EDGE) {
2956 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002957 struct ata_eh_info *ehi = &link->eh_info;
2958
2959 link->eh_context.i.dev_action[dev->devno] |=
2960 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2961 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2962 }
2963 }
2964 spin_unlock_irqrestore(ap->lock, flags);
2965}
2966
2967static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2968{
2969 struct ata_eh_context *ehc = &dev->link->eh_context;
2970 struct ata_taskfile tf;
2971 unsigned int err_mask;
2972
2973 ata_tf_init(dev, &tf);
2974 if (park) {
2975 ehc->unloaded_mask |= 1 << dev->devno;
2976 tf.command = ATA_CMD_IDLEIMMEDIATE;
2977 tf.feature = 0x44;
2978 tf.lbal = 0x4c;
2979 tf.lbam = 0x4e;
2980 tf.lbah = 0x55;
2981 } else {
2982 ehc->unloaded_mask &= ~(1 << dev->devno);
2983 tf.command = ATA_CMD_CHK_POWER;
2984 }
2985
2986 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2987 tf.protocol |= ATA_PROT_NODATA;
2988 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2989 if (park && (err_mask || tf.lbal != 0xc4)) {
2990 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2991 ehc->unloaded_mask &= ~(1 << dev->devno);
2992 }
2993}
2994
Tejun Heo02607312007-08-06 18:36:23 +09002995static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09002996 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09002997{
Tejun Heo02607312007-08-06 18:36:23 +09002998 struct ata_port *ap = link->ap;
2999 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003000 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003001 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003002 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09003003 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003004
3005 DPRINTK("ENTER\n");
3006
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003007 /* For PATA drive side cable detection to work, IDENTIFY must
3008 * be done backwards such that PDIAG- is released by the slave
3009 * device before the master device is identified.
3010 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003011 ata_for_each_dev(dev, link, ALL_REVERSE) {
Tejun Heof58229f2007-08-06 18:36:23 +09003012 unsigned int action = ata_eh_dev_action(dev);
3013 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09003014
Tejun Heobff04642006-11-10 18:08:10 +09003015 if (ehc->i.flags & ATA_EHI_DID_RESET)
3016 readid_flags |= ATA_READID_POSTRESET;
3017
Tejun Heo9666f402007-05-04 21:27:47 +02003018 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a32007-09-23 13:19:54 +09003019 WARN_ON(dev->class == ATA_DEV_PMP);
3020
Tejun Heob1c72912008-07-31 17:02:43 +09003021 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
Tejun Heo022bdb02006-05-15 20:58:22 +09003022 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003023 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003024 }
3025
Tejun Heo02607312007-08-06 18:36:23 +09003026 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09003027 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3028 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09003029 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003030 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003031
Tejun Heo02607312007-08-06 18:36:23 +09003032 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09003033
Tejun Heobaa1e782006-11-01 18:39:27 +09003034 /* Configuration may have changed, reconfigure
3035 * transfer mode.
3036 */
3037 ehc->i.flags |= ATA_EHI_SETMODE;
3038
zhao, forrest3057ac32006-06-12 12:01:34 +08003039 /* schedule the scsi_rescan_device() here */
Tejun Heoad72cf92010-07-02 10:03:52 +02003040 schedule_work(&(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09003041 } else if (dev->class == ATA_DEV_UNKNOWN &&
3042 ehc->tries[dev->devno] &&
3043 ata_class_enabled(ehc->classes[dev->devno])) {
Tejun Heo842faa62009-05-10 01:06:54 +09003044 /* Temporarily set dev->class, it will be
3045 * permanently set once all configurations are
3046 * complete. This is necessary because new
3047 * device configuration is done in two
3048 * separate loops.
3049 */
Tejun Heo084fe632006-05-31 18:28:03 +09003050 dev->class = ehc->classes[dev->devno];
3051
Tejun Heo633273a32007-09-23 13:19:54 +09003052 if (dev->class == ATA_DEV_PMP)
3053 rc = sata_pmp_attach(dev);
3054 else
3055 rc = ata_dev_read_id(dev, &dev->class,
3056 readid_flags, dev->id);
Tejun Heo842faa62009-05-10 01:06:54 +09003057
3058 /* read_id might have changed class, store and reset */
3059 ehc->classes[dev->devno] = dev->class;
3060 dev->class = ATA_DEV_UNKNOWN;
3061
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003062 switch (rc) {
3063 case 0:
Tejun Heo99cf6102009-01-29 20:31:32 +09003064 /* clear error info accumulated during probe */
3065 ata_ering_clear(&dev->ering);
Tejun Heof58229f2007-08-06 18:36:23 +09003066 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003067 break;
3068 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003069 /* IDENTIFY was issued to non-existent
3070 * device. No need to reset. Just
Tejun Heo842faa62009-05-10 01:06:54 +09003071 * thaw and ignore the device.
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003072 */
3073 ata_eh_thaw_port(ap);
Tejun Heo084fe632006-05-31 18:28:03 +09003074 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003075 default:
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003076 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003077 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003078 }
3079 }
3080
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003081 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09003082 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3083 if (ap->ops->cable_detect)
3084 ap->cbl = ap->ops->cable_detect(ap);
3085 ata_force_cbl(ap);
3086 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003087
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003088 /* Configure new devices forward such that user doesn't see
3089 * device detection messages backwards.
3090 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003091 ata_for_each_dev(dev, link, ALL) {
Tejun Heo4f7c2872009-10-15 23:37:32 +09003092 if (!(new_mask & (1 << dev->devno)))
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003093 continue;
3094
Tejun Heo842faa62009-05-10 01:06:54 +09003095 dev->class = ehc->classes[dev->devno];
3096
Tejun Heo4f7c2872009-10-15 23:37:32 +09003097 if (dev->class == ATA_DEV_PMP)
3098 continue;
3099
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003100 ehc->i.flags |= ATA_EHI_PRINTINFO;
3101 rc = ata_dev_configure(dev);
3102 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
Tejun Heo842faa62009-05-10 01:06:54 +09003103 if (rc) {
3104 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003105 goto err;
Tejun Heo842faa62009-05-10 01:06:54 +09003106 }
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003107
3108 spin_lock_irqsave(ap->lock, flags);
3109 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3110 spin_unlock_irqrestore(ap->lock, flags);
3111
3112 /* new device discovered, configure xfermode */
3113 ehc->i.flags |= ATA_EHI_SETMODE;
3114 }
3115
3116 return 0;
3117
3118 err:
3119 *r_failed_dev = dev;
3120 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003121 return rc;
3122}
3123
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003124/**
3125 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3126 * @link: link on which timings will be programmed
Martin Olsson98a17082009-04-22 18:21:29 +02003127 * @r_failed_dev: out parameter for failed device
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003128 *
3129 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3130 * ata_set_mode() fails, pointer to the failing device is
3131 * returned in @r_failed_dev.
3132 *
3133 * LOCKING:
3134 * PCI/etc. bus probe sem.
3135 *
3136 * RETURNS:
3137 * 0 on success, negative errno otherwise
3138 */
3139int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3140{
3141 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09003142 struct ata_device *dev;
3143 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003144
Tejun Heo76326ac2007-11-27 19:28:59 +09003145 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
Tejun Heo1eca4362008-11-03 20:03:17 +09003146 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo76326ac2007-11-27 19:28:59 +09003147 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3148 struct ata_ering_entry *ent;
3149
3150 ent = ata_ering_top(&dev->ering);
3151 if (ent)
3152 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3153 }
3154 }
3155
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003156 /* has private set_mode? */
3157 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09003158 rc = ap->ops->set_mode(link, r_failed_dev);
3159 else
3160 rc = ata_do_set_mode(link, r_failed_dev);
3161
3162 /* if transfer mode has changed, set DUBIOUS_XFER on device */
Tejun Heo1eca4362008-11-03 20:03:17 +09003163 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +09003164 struct ata_eh_context *ehc = &link->eh_context;
3165 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3166 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3167
3168 if (dev->xfer_mode != saved_xfer_mode ||
3169 ata_ncq_enabled(dev) != saved_ncq)
3170 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3171 }
3172
3173 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003174}
3175
Tejun Heo11fc33d2008-08-30 14:20:01 +02003176/**
3177 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3178 * @dev: ATAPI device to clear UA for
3179 *
3180 * Resets and other operations can make an ATAPI device raise
3181 * UNIT ATTENTION which causes the next operation to fail. This
3182 * function clears UA.
3183 *
3184 * LOCKING:
3185 * EH context (may sleep).
3186 *
3187 * RETURNS:
3188 * 0 on success, -errno on failure.
3189 */
3190static int atapi_eh_clear_ua(struct ata_device *dev)
3191{
3192 int i;
3193
3194 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
Tejun Heob5357082009-03-02 18:55:16 +09003195 u8 *sense_buffer = dev->link->ap->sector_buf;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003196 u8 sense_key = 0;
3197 unsigned int err_mask;
3198
3199 err_mask = atapi_eh_tur(dev, &sense_key);
3200 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3201 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3202 "failed (err_mask=0x%x)\n", err_mask);
3203 return -EIO;
3204 }
3205
3206 if (!err_mask || sense_key != UNIT_ATTENTION)
3207 return 0;
3208
3209 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3210 if (err_mask) {
3211 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3212 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3213 return -EIO;
3214 }
3215 }
3216
3217 ata_dev_printk(dev, KERN_WARNING,
3218 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3219
3220 return 0;
3221}
3222
Tejun Heo6013efd2009-11-19 15:36:45 +09003223/**
3224 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3225 * @dev: ATA device which may need FLUSH retry
3226 *
3227 * If @dev failed FLUSH, it needs to be reported upper layer
3228 * immediately as it means that @dev failed to remap and already
3229 * lost at least a sector and further FLUSH retrials won't make
3230 * any difference to the lost sector. However, if FLUSH failed
3231 * for other reasons, for example transmission error, FLUSH needs
3232 * to be retried.
3233 *
3234 * This function determines whether FLUSH failure retry is
3235 * necessary and performs it if so.
3236 *
3237 * RETURNS:
3238 * 0 if EH can continue, -errno if EH needs to be repeated.
3239 */
3240static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3241{
3242 struct ata_link *link = dev->link;
3243 struct ata_port *ap = link->ap;
3244 struct ata_queued_cmd *qc;
3245 struct ata_taskfile tf;
3246 unsigned int err_mask;
3247 int rc = 0;
3248
3249 /* did flush fail for this device? */
3250 if (!ata_tag_valid(link->active_tag))
3251 return 0;
3252
3253 qc = __ata_qc_from_tag(ap, link->active_tag);
3254 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3255 qc->tf.command != ATA_CMD_FLUSH))
3256 return 0;
3257
3258 /* if the device failed it, it should be reported to upper layers */
3259 if (qc->err_mask & AC_ERR_DEV)
3260 return 0;
3261
3262 /* flush failed for some other reason, give it another shot */
3263 ata_tf_init(dev, &tf);
3264
3265 tf.command = qc->tf.command;
3266 tf.flags |= ATA_TFLAG_DEVICE;
3267 tf.protocol = ATA_PROT_NODATA;
3268
3269 ata_dev_printk(dev, KERN_WARNING, "retrying FLUSH 0x%x Emask 0x%x\n",
3270 tf.command, qc->err_mask);
3271
3272 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3273 if (!err_mask) {
3274 /*
3275 * FLUSH is complete but there's no way to
3276 * successfully complete a failed command from EH.
3277 * Making sure retry is allowed at least once and
3278 * retrying it should do the trick - whatever was in
3279 * the cache is already on the platter and this won't
3280 * cause infinite loop.
3281 */
3282 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3283 } else {
3284 ata_dev_printk(dev, KERN_WARNING, "FLUSH failed Emask 0x%x\n",
3285 err_mask);
3286 rc = -EIO;
3287
3288 /* if device failed it, report it to upper layers */
3289 if (err_mask & AC_ERR_DEV) {
3290 qc->err_mask |= AC_ERR_DEV;
3291 qc->result_tf = tf;
3292 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3293 rc = 0;
3294 }
3295 }
3296 return rc;
3297}
3298
Tejun Heo6b7ae952010-09-01 17:50:06 +02003299/**
3300 * ata_eh_set_lpm - configure SATA interface power management
3301 * @link: link to configure power management
3302 * @policy: the link power management policy
3303 * @r_failed_dev: out parameter for failed device
3304 *
3305 * Enable SATA Interface power management. This will enable
3306 * Device Interface Power Management (DIPM) for min_power
3307 * policy, and then call driver specific callbacks for
3308 * enabling Host Initiated Power management.
3309 *
3310 * LOCKING:
3311 * EH context.
3312 *
3313 * RETURNS:
3314 * 0 on success, -errno on failure.
3315 */
3316static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3317 struct ata_device **r_failed_dev)
3318{
Tejun Heo6c8ea892010-09-01 17:50:07 +02003319 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003320 struct ata_eh_context *ehc = &link->eh_context;
3321 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
Tejun Heoe5005b12010-12-09 16:13:15 +01003322 enum ata_lpm_policy old_policy = link->lpm_policy;
Tejun Heo5f6f12c2011-05-09 16:04:11 +02003323 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003324 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3325 unsigned int err_mask;
3326 int rc;
3327
3328 /* if the link or host doesn't do LPM, noop */
3329 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3330 return 0;
3331
3332 /*
3333 * DIPM is enabled only for MIN_POWER as some devices
3334 * misbehave when the host NACKs transition to SLUMBER. Order
3335 * device and link configurations such that the host always
3336 * allows DIPM requests.
3337 */
3338 ata_for_each_dev(dev, link, ENABLED) {
3339 bool hipm = ata_id_has_hipm(dev->id);
Tejun Heoae01b242011-03-16 11:14:55 +01003340 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003341
3342 /* find the first enabled and LPM enabled devices */
3343 if (!link_dev)
3344 link_dev = dev;
3345
3346 if (!lpm_dev && (hipm || dipm))
3347 lpm_dev = dev;
3348
3349 hints &= ~ATA_LPM_EMPTY;
3350 if (!hipm)
3351 hints &= ~ATA_LPM_HIPM;
3352
3353 /* disable DIPM before changing link config */
3354 if (policy != ATA_LPM_MIN_POWER && dipm) {
3355 err_mask = ata_dev_set_feature(dev,
3356 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3357 if (err_mask && err_mask != AC_ERR_DEV) {
3358 ata_dev_printk(dev, KERN_WARNING,
3359 "failed to disable DIPM, Emask 0x%x\n",
3360 err_mask);
3361 rc = -EIO;
3362 goto fail;
3363 }
3364 }
3365 }
3366
Tejun Heo6c8ea892010-09-01 17:50:07 +02003367 if (ap) {
3368 rc = ap->ops->set_lpm(link, policy, hints);
3369 if (!rc && ap->slave_link)
3370 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3371 } else
3372 rc = sata_pmp_set_lpm(link, policy, hints);
Tejun Heo6b7ae952010-09-01 17:50:06 +02003373
3374 /*
3375 * Attribute link config failure to the first (LPM) enabled
3376 * device on the link.
3377 */
3378 if (rc) {
3379 if (rc == -EOPNOTSUPP) {
3380 link->flags |= ATA_LFLAG_NO_LPM;
3381 return 0;
3382 }
3383 dev = lpm_dev ? lpm_dev : link_dev;
3384 goto fail;
3385 }
3386
Tejun Heoe5005b12010-12-09 16:13:15 +01003387 /*
3388 * Low level driver acked the transition. Issue DIPM command
3389 * with the new policy set.
3390 */
3391 link->lpm_policy = policy;
3392 if (ap && ap->slave_link)
3393 ap->slave_link->lpm_policy = policy;
3394
Tejun Heo6b7ae952010-09-01 17:50:06 +02003395 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3396 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heoae01b242011-03-16 11:14:55 +01003397 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3398 ata_id_has_dipm(dev->id)) {
Tejun Heo6b7ae952010-09-01 17:50:06 +02003399 err_mask = ata_dev_set_feature(dev,
3400 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3401 if (err_mask && err_mask != AC_ERR_DEV) {
3402 ata_dev_printk(dev, KERN_WARNING,
3403 "failed to enable DIPM, Emask 0x%x\n",
3404 err_mask);
3405 rc = -EIO;
3406 goto fail;
3407 }
3408 }
3409 }
3410
Tejun Heo6b7ae952010-09-01 17:50:06 +02003411 return 0;
3412
3413fail:
Tejun Heoe5005b12010-12-09 16:13:15 +01003414 /* restore the old policy */
3415 link->lpm_policy = old_policy;
3416 if (ap && ap->slave_link)
3417 ap->slave_link->lpm_policy = old_policy;
3418
Tejun Heo6b7ae952010-09-01 17:50:06 +02003419 /* if no device or only one more chance is left, disable LPM */
3420 if (!dev || ehc->tries[dev->devno] <= 2) {
3421 ata_link_printk(link, KERN_WARNING,
3422 "disabling LPM on the link\n");
3423 link->flags |= ATA_LFLAG_NO_LPM;
3424 }
3425 if (r_failed_dev)
3426 *r_failed_dev = dev;
3427 return rc;
3428}
3429
Kristen Carlson Accardi8a745f12011-03-04 10:24:11 -08003430int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003431{
Tejun Heof58229f2007-08-06 18:36:23 +09003432 struct ata_device *dev;
3433 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003434
Tejun Heo1eca4362008-11-03 20:03:17 +09003435 ata_for_each_dev(dev, link, ENABLED)
3436 cnt++;
Tejun Heo022bdb02006-05-15 20:58:22 +09003437 return cnt;
3438}
3439
Tejun Heo02607312007-08-06 18:36:23 +09003440static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003441{
Tejun Heof58229f2007-08-06 18:36:23 +09003442 struct ata_device *dev;
3443 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003444
Tejun Heo1eca4362008-11-03 20:03:17 +09003445 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09003446 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09003447 cnt++;
3448 return cnt;
3449}
3450
Tejun Heo02607312007-08-06 18:36:23 +09003451static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003452{
Tejun Heo672b2d62008-01-24 00:05:14 +09003453 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09003454 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09003455 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09003456
Tejun Heof9df58c2007-09-23 13:14:13 +09003457 /* skip disabled links */
3458 if (link->flags & ATA_LFLAG_DISABLED)
3459 return 1;
3460
Tejun Heoe2f3d752010-09-07 14:05:31 +02003461 /* skip if explicitly requested */
3462 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3463 return 1;
3464
Tejun Heo672b2d62008-01-24 00:05:14 +09003465 /* thaw frozen port and recover failed devices */
3466 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3467 return 0;
3468
3469 /* reset at least once if reset is requested */
3470 if ((ehc->i.action & ATA_EH_RESET) &&
3471 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09003472 return 0;
3473
3474 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo1eca4362008-11-03 20:03:17 +09003475 ata_for_each_dev(dev, link, ALL) {
Tejun Heo084fe632006-05-31 18:28:03 +09003476 if (dev->class == ATA_DEV_UNKNOWN &&
3477 ehc->classes[dev->devno] != ATA_DEV_NONE)
3478 return 0;
3479 }
3480
3481 return 1;
3482}
3483
Tejun Heoc2c7a892009-01-29 20:31:34 +09003484static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3485{
3486 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3487 u64 now = get_jiffies_64();
3488 int *trials = void_arg;
3489
3490 if (ent->timestamp < now - min(now, interval))
3491 return -1;
3492
3493 (*trials)++;
3494 return 0;
3495}
3496
Tejun Heo02c05a22007-11-27 19:28:54 +09003497static int ata_eh_schedule_probe(struct ata_device *dev)
3498{
3499 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heoc2c7a892009-01-29 20:31:34 +09003500 struct ata_link *link = ata_dev_phys_link(dev);
3501 int trials = 0;
Tejun Heo02c05a22007-11-27 19:28:54 +09003502
3503 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3504 (ehc->did_probe_mask & (1 << dev->devno)))
3505 return 0;
3506
3507 ata_eh_detach_dev(dev);
3508 ata_dev_init(dev);
3509 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09003510 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09003511 ehc->saved_xfer_mode[dev->devno] = 0;
3512 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09003513
Tejun Heo6b7ae952010-09-01 17:50:06 +02003514 /* the link maybe in a deep sleep, wake it up */
Tejun Heo6c8ea892010-09-01 17:50:07 +02003515 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3516 if (ata_is_host_link(link))
3517 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3518 ATA_LPM_EMPTY);
3519 else
3520 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3521 ATA_LPM_EMPTY);
3522 }
Tejun Heo6b7ae952010-09-01 17:50:06 +02003523
Tejun Heoc2c7a892009-01-29 20:31:34 +09003524 /* Record and count probe trials on the ering. The specific
3525 * error mask used is irrelevant. Because a successful device
3526 * detection clears the ering, this count accumulates only if
3527 * there are consecutive failed probes.
3528 *
3529 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3530 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3531 * forced to 1.5Gbps.
3532 *
3533 * This is to work around cases where failed link speed
3534 * negotiation results in device misdetection leading to
3535 * infinite DEVXCHG or PHRDY CHG events.
3536 */
3537 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3538 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3539
3540 if (trials > ATA_EH_PROBE_TRIALS)
3541 sata_down_spd_limit(link, 1);
3542
Tejun Heo02c05a22007-11-27 19:28:54 +09003543 return 1;
3544}
3545
Tejun Heo9b1e2652007-08-06 18:36:24 +09003546static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09003547{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09003548 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09003549
Tejun Heocf9a5902009-01-29 20:31:35 +09003550 /* -EAGAIN from EH routine indicates retry without prejudice.
3551 * The requester is responsible for ensuring forward progress.
3552 */
3553 if (err != -EAGAIN)
3554 ehc->tries[dev->devno]--;
Tejun Heofee7ca72007-07-01 19:05:58 +09003555
3556 switch (err) {
3557 case -ENODEV:
3558 /* device missing or wrong IDENTIFY data, schedule probing */
3559 ehc->i.probe_mask |= (1 << dev->devno);
3560 case -EINVAL:
3561 /* give it just one more chance */
3562 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3563 case -EIO:
Tejun Heod89293a2009-01-29 20:31:29 +09003564 if (ehc->tries[dev->devno] == 1) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003565 /* This is the last chance, better to slow
3566 * down than lose it.
3567 */
Tejun Heoa07d4992009-01-29 20:31:33 +09003568 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
Tejun Heod89293a2009-01-29 20:31:29 +09003569 if (dev->pio_mode > XFER_PIO_0)
3570 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
Tejun Heofee7ca72007-07-01 19:05:58 +09003571 }
3572 }
3573
3574 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3575 /* disable device if it has used up all its chances */
3576 ata_dev_disable(dev);
3577
3578 /* detach if offline */
Tejun Heob1c72912008-07-31 17:02:43 +09003579 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
Tejun Heofee7ca72007-07-01 19:05:58 +09003580 ata_eh_detach_dev(dev);
3581
Tejun Heo02c05a22007-11-27 19:28:54 +09003582 /* schedule probe if necessary */
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003583 if (ata_eh_schedule_probe(dev)) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003584 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003585 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3586 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3587 }
Tejun Heo9b1e2652007-08-06 18:36:24 +09003588
3589 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09003590 } else {
Tejun Heocf480622008-01-24 00:05:14 +09003591 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003592 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09003593 }
3594}
3595
Tejun Heo022bdb02006-05-15 20:58:22 +09003596/**
3597 * ata_eh_recover - recover host port after error
3598 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09003599 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003600 * @softreset: softreset method (can be NULL)
3601 * @hardreset: hardreset method (can be NULL)
3602 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003603 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09003604 *
3605 * This is the alpha and omega, eum and yang, heart and soul of
3606 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09003607 * recover each link and hotplug requests are recorded in the
3608 * link's eh_context. This function executes all the operations
3609 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09003610 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09003611 *
3612 * LOCKING:
3613 * Kernel thread context (may sleep).
3614 *
3615 * RETURNS:
3616 * 0 on success, -errno on failure.
3617 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003618int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3619 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3620 ata_postreset_fn_t postreset,
3621 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003622{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003623 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003624 struct ata_device *dev;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003625 int rc, nr_fails;
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003626 unsigned long flags, deadline;
Tejun Heo022bdb02006-05-15 20:58:22 +09003627
3628 DPRINTK("ENTER\n");
3629
3630 /* prep for recovery */
Tejun Heo1eca4362008-11-03 20:03:17 +09003631 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003632 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09003633
Tejun Heof9df58c2007-09-23 13:14:13 +09003634 /* re-enable link? */
3635 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3636 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3637 spin_lock_irqsave(ap->lock, flags);
3638 link->flags &= ~ATA_LFLAG_DISABLED;
3639 spin_unlock_irqrestore(ap->lock, flags);
3640 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3641 }
3642
Tejun Heo1eca4362008-11-03 20:03:17 +09003643 ata_for_each_dev(dev, link, ALL) {
Tejun Heofd995f72007-09-23 13:14:12 +09003644 if (link->flags & ATA_LFLAG_NO_RETRY)
3645 ehc->tries[dev->devno] = 1;
3646 else
3647 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09003648
Tejun Heo9b1e2652007-08-06 18:36:24 +09003649 /* collect port action mask recorded in dev actions */
3650 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3651 ~ATA_EH_PERDEV_MASK;
3652 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09003653
Tejun Heo9b1e2652007-08-06 18:36:24 +09003654 /* process hotplug request */
3655 if (dev->flags & ATA_DFLAG_DETACH)
3656 ata_eh_detach_dev(dev);
3657
Tejun Heo02c05a22007-11-27 19:28:54 +09003658 /* schedule probe if necessary */
3659 if (!ata_dev_enabled(dev))
3660 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09003661 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003662 }
3663
3664 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09003665 rc = 0;
3666
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003667 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09003668 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003669 goto out;
3670
Tejun Heo9b1e2652007-08-06 18:36:24 +09003671 /* prep for EH */
Tejun Heo1eca4362008-11-03 20:03:17 +09003672 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003673 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003674
Tejun Heo9b1e2652007-08-06 18:36:24 +09003675 /* skip EH if possible. */
3676 if (ata_eh_skip_recovery(link))
3677 ehc->i.action = 0;
3678
Tejun Heo1eca4362008-11-03 20:03:17 +09003679 ata_for_each_dev(dev, link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003680 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3681 }
Tejun Heo084fe632006-05-31 18:28:03 +09003682
Tejun Heo022bdb02006-05-15 20:58:22 +09003683 /* reset */
Tejun Heo1eca4362008-11-03 20:03:17 +09003684 ata_for_each_link(link, ap, EDGE) {
Tejun Heodc98c322008-05-19 01:15:07 +09003685 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003686
Tejun Heodc98c322008-05-19 01:15:07 +09003687 if (!(ehc->i.action & ATA_EH_RESET))
3688 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003689
Tejun Heodc98c322008-05-19 01:15:07 +09003690 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3691 prereset, softreset, hardreset, postreset);
3692 if (rc) {
3693 ata_link_printk(link, KERN_ERR,
3694 "reset failed, giving up\n");
3695 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09003696 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003697 }
3698
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003699 do {
3700 unsigned long now;
3701
3702 /*
3703 * clears ATA_EH_PARK in eh_info and resets
3704 * ap->park_req_pending
3705 */
3706 ata_eh_pull_park_action(ap);
3707
3708 deadline = jiffies;
Tejun Heo1eca4362008-11-03 20:03:17 +09003709 ata_for_each_link(link, ap, EDGE) {
3710 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003711 struct ata_eh_context *ehc = &link->eh_context;
3712 unsigned long tmp;
3713
3714 if (dev->class != ATA_DEV_ATA)
3715 continue;
3716 if (!(ehc->i.dev_action[dev->devno] &
3717 ATA_EH_PARK))
3718 continue;
3719 tmp = dev->unpark_deadline;
3720 if (time_before(deadline, tmp))
3721 deadline = tmp;
3722 else if (time_before_eq(tmp, jiffies))
3723 continue;
3724 if (ehc->unloaded_mask & (1 << dev->devno))
3725 continue;
3726
3727 ata_eh_park_issue_cmd(dev, 1);
3728 }
3729 }
3730
3731 now = jiffies;
3732 if (time_before_eq(deadline, now))
3733 break;
3734
Tejun Heoc0c362b2010-09-06 17:57:14 +02003735 ata_eh_release(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003736 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3737 deadline - now);
Tejun Heoc0c362b2010-09-06 17:57:14 +02003738 ata_eh_acquire(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003739 } while (deadline);
Tejun Heo1eca4362008-11-03 20:03:17 +09003740 ata_for_each_link(link, ap, EDGE) {
3741 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003742 if (!(link->eh_context.unloaded_mask &
3743 (1 << dev->devno)))
3744 continue;
3745
3746 ata_eh_park_issue_cmd(dev, 0);
3747 ata_eh_done(link, dev, ATA_EH_PARK);
3748 }
3749 }
3750
Tejun Heo9b1e2652007-08-06 18:36:24 +09003751 /* the rest */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003752 nr_fails = 0;
3753 ata_for_each_link(link, ap, PMP_FIRST) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003754 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003755
Tejun Heo6b7ae952010-09-01 17:50:06 +02003756 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3757 goto config_lpm;
3758
Tejun Heo9b1e2652007-08-06 18:36:24 +09003759 /* revalidate existing devices and attach new ones */
3760 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09003761 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003762 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003763
Tejun Heo633273a32007-09-23 13:19:54 +09003764 /* if PMP got attached, return, pmp EH will take care of it */
3765 if (link->device->class == ATA_DEV_PMP) {
3766 ehc->i.action = 0;
3767 return 0;
3768 }
3769
Tejun Heo9b1e2652007-08-06 18:36:24 +09003770 /* configure transfer mode if necessary */
3771 if (ehc->i.flags & ATA_EHI_SETMODE) {
3772 rc = ata_set_mode(link, &dev);
3773 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003774 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003775 ehc->i.flags &= ~ATA_EHI_SETMODE;
3776 }
3777
Tejun Heo11fc33d2008-08-30 14:20:01 +02003778 /* If reset has been issued, clear UA to avoid
3779 * disrupting the current users of the device.
3780 */
3781 if (ehc->i.flags & ATA_EHI_DID_RESET) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003782 ata_for_each_dev(dev, link, ALL) {
Tejun Heo11fc33d2008-08-30 14:20:01 +02003783 if (dev->class != ATA_DEV_ATAPI)
3784 continue;
3785 rc = atapi_eh_clear_ua(dev);
3786 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003787 goto rest_fail;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003788 }
3789 }
3790
Tejun Heo6013efd2009-11-19 15:36:45 +09003791 /* retry flush if necessary */
3792 ata_for_each_dev(dev, link, ALL) {
3793 if (dev->class != ATA_DEV_ATA)
3794 continue;
3795 rc = ata_eh_maybe_retry_flush(dev);
3796 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003797 goto rest_fail;
Tejun Heo6013efd2009-11-19 15:36:45 +09003798 }
3799
Tejun Heo6b7ae952010-09-01 17:50:06 +02003800 config_lpm:
Tejun Heo11fc33d2008-08-30 14:20:01 +02003801 /* configure link power saving */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003802 if (link->lpm_policy != ap->target_lpm_policy) {
3803 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3804 if (rc)
3805 goto rest_fail;
3806 }
Kristen Carlson Accardica773292007-10-25 00:58:59 -04003807
Tejun Heo9b1e2652007-08-06 18:36:24 +09003808 /* this link is okay now */
3809 ehc->i.flags = 0;
3810 continue;
3811
Tejun Heo6b7ae952010-09-01 17:50:06 +02003812 rest_fail:
3813 nr_fails++;
3814 if (dev)
3815 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo9b1e2652007-08-06 18:36:24 +09003816
Tejun Heob06ce3e2007-10-09 15:06:48 +09003817 if (ap->pflags & ATA_PFLAG_FROZEN) {
3818 /* PMP reset requires working host port.
3819 * Can't retry if it's frozen.
3820 */
Tejun Heo071f44b2008-04-07 22:47:22 +09003821 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09003822 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003823 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09003824 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003825 }
3826
Tejun Heo6b7ae952010-09-01 17:50:06 +02003827 if (nr_fails)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003828 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003829
Tejun Heo022bdb02006-05-15 20:58:22 +09003830 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003831 if (rc && r_failed_link)
3832 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003833
3834 DPRINTK("EXIT, rc=%d\n", rc);
3835 return rc;
3836}
3837
3838/**
3839 * ata_eh_finish - finish up EH
3840 * @ap: host port to finish EH for
3841 *
3842 * Recovery is complete. Clean up EH states and retry or finish
3843 * failed qcs.
3844 *
3845 * LOCKING:
3846 * None.
3847 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003848void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09003849{
3850 int tag;
3851
3852 /* retry or finish qcs */
3853 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3854 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3855
3856 if (!(qc->flags & ATA_QCFLAG_FAILED))
3857 continue;
3858
3859 if (qc->err_mask) {
3860 /* FIXME: Once EH migration is complete,
3861 * generate sense data in this function,
3862 * considering both err_mask and tf.
3863 */
Tejun Heo03faab72008-03-27 19:14:24 +09003864 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09003865 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09003866 else
3867 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003868 } else {
3869 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3870 ata_eh_qc_complete(qc);
3871 } else {
3872 /* feed zero TF to sense generation */
3873 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3874 ata_eh_qc_retry(qc);
3875 }
3876 }
3877 }
Tejun Heoda917d62007-09-23 13:14:12 +09003878
3879 /* make sure nr_active_links is zero after EH */
3880 WARN_ON(ap->nr_active_links);
3881 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003882}
3883
3884/**
3885 * ata_do_eh - do standard error handling
3886 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09003887 *
Tejun Heof5914a42006-05-31 18:27:48 +09003888 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003889 * @softreset: softreset method (can be NULL)
3890 * @hardreset: hardreset method (can be NULL)
3891 * @postreset: postreset method (can be NULL)
3892 *
3893 * Perform standard error handling sequence.
3894 *
3895 * LOCKING:
3896 * Kernel thread context (may sleep).
3897 */
Tejun Heof5914a42006-05-31 18:27:48 +09003898void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3899 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3900 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09003901{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003902 struct ata_device *dev;
3903 int rc;
3904
3905 ata_eh_autopsy(ap);
3906 ata_eh_report(ap);
3907
3908 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3909 NULL);
3910 if (rc) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003911 ata_for_each_dev(dev, &ap->link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003912 ata_dev_disable(dev);
3913 }
3914
Tejun Heo022bdb02006-05-15 20:58:22 +09003915 ata_eh_finish(ap);
3916}
Tejun Heo500530f2006-07-03 16:07:27 +09003917
Tejun Heoa1efdab2008-03-25 12:22:50 +09003918/**
3919 * ata_std_error_handler - standard error handler
3920 * @ap: host port to handle error for
3921 *
3922 * Standard error handler
3923 *
3924 * LOCKING:
3925 * Kernel thread context (may sleep).
3926 */
3927void ata_std_error_handler(struct ata_port *ap)
3928{
3929 struct ata_port_operations *ops = ap->ops;
3930 ata_reset_fn_t hardreset = ops->hardreset;
3931
Tejun Heo57c9efd2008-04-07 22:47:19 +09003932 /* ignore built-in hardreset if SCR access is not available */
Tejun Heofe06e5f2010-05-10 21:41:39 +02003933 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09003934 hardreset = NULL;
3935
3936 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3937}
3938
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003939#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09003940/**
3941 * ata_eh_handle_port_suspend - perform port suspend operation
3942 * @ap: port to suspend
3943 *
3944 * Suspend @ap.
3945 *
3946 * LOCKING:
3947 * Kernel thread context (may sleep).
3948 */
3949static void ata_eh_handle_port_suspend(struct ata_port *ap)
3950{
3951 unsigned long flags;
3952 int rc = 0;
3953
3954 /* are we suspending? */
3955 spin_lock_irqsave(ap->lock, flags);
3956 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3957 ap->pm_mesg.event == PM_EVENT_ON) {
3958 spin_unlock_irqrestore(ap->lock, flags);
3959 return;
3960 }
3961 spin_unlock_irqrestore(ap->lock, flags);
3962
3963 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3964
Tejun Heo64578a32007-05-15 03:28:16 +09003965 /* tell ACPI we're suspending */
3966 rc = ata_acpi_on_suspend(ap);
3967 if (rc)
3968 goto out;
3969
Tejun Heo500530f2006-07-03 16:07:27 +09003970 /* suspend */
3971 ata_eh_freeze_port(ap);
3972
3973 if (ap->ops->port_suspend)
3974 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3975
Shaohua Libd3adca2007-11-02 09:32:38 +08003976 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a32007-05-15 03:28:16 +09003977 out:
Tejun Heo500530f2006-07-03 16:07:27 +09003978 /* report result */
3979 spin_lock_irqsave(ap->lock, flags);
3980
3981 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3982 if (rc == 0)
3983 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09003984 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09003985 ata_port_schedule_eh(ap);
3986
3987 if (ap->pm_result) {
3988 *ap->pm_result = rc;
3989 ap->pm_result = NULL;
3990 }
3991
3992 spin_unlock_irqrestore(ap->lock, flags);
3993
3994 return;
3995}
3996
3997/**
3998 * ata_eh_handle_port_resume - perform port resume operation
3999 * @ap: port to resume
4000 *
4001 * Resume @ap.
4002 *
Tejun Heo500530f2006-07-03 16:07:27 +09004003 * LOCKING:
4004 * Kernel thread context (may sleep).
4005 */
4006static void ata_eh_handle_port_resume(struct ata_port *ap)
4007{
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004008 struct ata_link *link;
4009 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09004010 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02004011 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09004012
4013 /* are we resuming? */
4014 spin_lock_irqsave(ap->lock, flags);
4015 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4016 ap->pm_mesg.event != PM_EVENT_ON) {
4017 spin_unlock_irqrestore(ap->lock, flags);
4018 return;
4019 }
4020 spin_unlock_irqrestore(ap->lock, flags);
4021
Tejun Heo9666f402007-05-04 21:27:47 +02004022 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09004023
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004024 /*
4025 * Error timestamps are in jiffies which doesn't run while
4026 * suspended and PHY events during resume isn't too uncommon.
4027 * When the two are combined, it can lead to unnecessary speed
4028 * downs if the machine is suspended and resumed repeatedly.
4029 * Clear error history.
4030 */
4031 ata_for_each_link(link, ap, HOST_FIRST)
4032 ata_for_each_dev(dev, link, ALL)
4033 ata_ering_clear(&dev->ering);
4034
Shaohua Libd3adca2007-11-02 09:32:38 +08004035 ata_acpi_set_state(ap, PMSG_ON);
4036
Tejun Heo500530f2006-07-03 16:07:27 +09004037 if (ap->ops->port_resume)
4038 rc = ap->ops->port_resume(ap);
4039
Tejun Heo67465442007-05-15 03:28:16 +09004040 /* tell ACPI that we're resuming */
4041 ata_acpi_on_resume(ap);
4042
Tejun Heo9666f402007-05-04 21:27:47 +02004043 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09004044 spin_lock_irqsave(ap->lock, flags);
4045 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4046 if (ap->pm_result) {
4047 *ap->pm_result = rc;
4048 ap->pm_result = NULL;
4049 }
4050 spin_unlock_irqrestore(ap->lock, flags);
4051}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09004052#endif /* CONFIG_PM */