blob: c69fcce505c03d06c7b20ae333ff9708a228f869 [file] [log] [blame]
Tejun Heoece1d632006-04-02 18:51:53 +09001/*
2 * libata-eh.c - libata error handling
3 *
Tejun Heo8c3d3d42013-05-14 11:09:50 -07004 * Maintained by: Tejun Heo <tj@kernel.org>
Tejun Heoece1d632006-04-02 18:51:53 +09005 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
Tejun Heoece1d632006-04-02 18:51:53 +090035#include <linux/kernel.h>
Jens Axboe242f9dc2008-09-14 05:55:09 -070036#include <linux/blkdev.h>
Paul Gortmaker38789fd2011-07-17 15:33:58 -040037#include <linux/export.h>
Jeff Garzik28555682007-10-11 17:12:35 -040038#include <linux/pci.h>
Tejun Heoece1d632006-04-02 18:51:53 +090039#include <scsi/scsi.h>
40#include <scsi/scsi_host.h>
41#include <scsi/scsi_eh.h>
42#include <scsi/scsi_device.h>
43#include <scsi/scsi_cmnd.h>
Robert Hancock65211482009-07-14 20:43:39 -060044#include <scsi/scsi_dbg.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040045#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090046
47#include <linux/libata.h>
48
49#include "libata.h"
50
Tejun Heo7d47e8d2007-02-02 16:22:31 +090051enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090052 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090053 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
54 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
55 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090056 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090057
58 /* error flags */
59 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090060 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Gwendal Grignoud9027472010-05-25 12:31:38 -070061 ATA_EFLAG_OLD_ER = (1 << 31),
Tejun Heo3884f7b2007-11-27 19:28:56 +090062
63 /* error categories */
64 ATA_ECAT_NONE = 0,
65 ATA_ECAT_ATA_BUS = 1,
66 ATA_ECAT_TOUT_HSM = 2,
67 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090068 ATA_ECAT_DUBIOUS_NONE = 4,
69 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
70 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
71 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
72 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090073
Tejun Heo87fbc5a2008-05-20 02:17:54 +090074 ATA_EH_CMD_DFL_TIMEOUT = 5000,
75
Tejun Heo0a2c0f52008-05-20 02:17:52 +090076 /* always put at least this amount of time between resets */
77 ATA_EH_RESET_COOL_DOWN = 5000,
78
Tejun Heo341c2c92008-05-20 02:17:51 +090079 /* Waiting in ->prereset can never be reliable. It's
80 * sometimes nice to wait there but it can't be depended upon;
81 * otherwise, we wouldn't be resetting. Just give it enough
82 * time for most drives to spin up.
83 */
84 ATA_EH_PRERESET_TIMEOUT = 10000,
85 ATA_EH_FASTDRAIN_INTERVAL = 3000,
Tejun Heo11fc33d2008-08-30 14:20:01 +020086
87 ATA_EH_UA_TRIES = 5,
Tejun Heoc2c7a892009-01-29 20:31:34 +090088
89 /* probe speed down parameters, see ata_eh_schedule_probe() */
90 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
91 ATA_EH_PROBE_TRIALS = 2,
Tejun Heo31daabd2007-02-02 16:50:52 +090092};
93
94/* The following table determines how we sequence resets. Each entry
95 * represents timeout for that try. The first try can be soft or
96 * hardreset. All others are hardreset if available. In most cases
97 * the first reset w/ 10sec timeout should succeed. Following entries
98 * are mostly for error handling, hotplug and retarded devices.
99 */
100static const unsigned long ata_eh_reset_timeouts[] = {
Tejun Heo341c2c92008-05-20 02:17:51 +0900101 10000, /* most drives spin up by 10sec */
102 10000, /* > 99% working drives spin up before 20sec */
103 35000, /* give > 30 secs of idleness for retarded devices */
104 5000, /* and sweet one last chance */
Tejun Heod8af0eb2008-05-20 02:17:53 +0900105 ULONG_MAX, /* > 1 min has elapsed, give up */
Tejun Heo31daabd2007-02-02 16:50:52 +0900106};
107
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900108static const unsigned long ata_eh_identify_timeouts[] = {
109 5000, /* covers > 99% of successes and not too boring on failures */
110 10000, /* combined time till here is enough even for media access */
111 30000, /* for true idiots */
112 ULONG_MAX,
113};
114
Tejun Heo6013efd2009-11-19 15:36:45 +0900115static const unsigned long ata_eh_flush_timeouts[] = {
116 15000, /* be generous with flush */
117 15000, /* ditto */
118 30000, /* and even more generous */
119 ULONG_MAX,
120};
121
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900122static const unsigned long ata_eh_other_timeouts[] = {
123 5000, /* same rationale as identify timeout */
124 10000, /* ditto */
125 /* but no merciful 30sec for other commands, it just isn't worth it */
126 ULONG_MAX,
127};
128
129struct ata_eh_cmd_timeout_ent {
130 const u8 *commands;
131 const unsigned long *timeouts;
132};
133
134/* The following table determines timeouts to use for EH internal
135 * commands. Each table entry is a command class and matches the
136 * commands the entry applies to and the timeout table to use.
137 *
138 * On the retry after a command timed out, the next timeout value from
139 * the table is used. If the table doesn't contain further entries,
140 * the last value is used.
141 *
142 * ehc->cmd_timeout_idx keeps track of which timeout to use per
143 * command class, so if SET_FEATURES times out on the first try, the
144 * next try will use the second timeout value only for that class.
145 */
146#define CMDS(cmds...) (const u8 []){ cmds, 0 }
147static const struct ata_eh_cmd_timeout_ent
148ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
149 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
150 .timeouts = ata_eh_identify_timeouts, },
151 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
152 .timeouts = ata_eh_other_timeouts, },
153 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_FEATURES),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
158 .timeouts = ata_eh_other_timeouts, },
Tejun Heo6013efd2009-11-19 15:36:45 +0900159 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
160 .timeouts = ata_eh_flush_timeouts },
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900161};
162#undef CMDS
163
Tejun Heoad9e2762006-05-15 20:58:12 +0900164static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900165#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +0900166static void ata_eh_handle_port_suspend(struct ata_port *ap);
167static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900168#else /* CONFIG_PM */
169static void ata_eh_handle_port_suspend(struct ata_port *ap)
170{ }
171
172static void ata_eh_handle_port_resume(struct ata_port *ap)
173{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900174#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900175
Tejun Heob64bbc32007-07-16 14:29:39 +0900176static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
177 va_list args)
178{
179 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
180 ATA_EH_DESC_LEN - ehi->desc_len,
181 fmt, args);
182}
183
184/**
185 * __ata_ehi_push_desc - push error description without adding separator
186 * @ehi: target EHI
187 * @fmt: printf format string
188 *
189 * Format string according to @fmt and append it to @ehi->desc.
190 *
191 * LOCKING:
192 * spin_lock_irqsave(host lock)
193 */
194void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
195{
196 va_list args;
197
198 va_start(args, fmt);
199 __ata_ehi_pushv_desc(ehi, fmt, args);
200 va_end(args);
201}
202
203/**
204 * ata_ehi_push_desc - push error description with separator
205 * @ehi: target EHI
206 * @fmt: printf format string
207 *
208 * Format string according to @fmt and append it to @ehi->desc.
209 * If @ehi->desc is not empty, ", " is added in-between.
210 *
211 * LOCKING:
212 * spin_lock_irqsave(host lock)
213 */
214void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
215{
216 va_list args;
217
218 if (ehi->desc_len)
219 __ata_ehi_push_desc(ehi, ", ");
220
221 va_start(args, fmt);
222 __ata_ehi_pushv_desc(ehi, fmt, args);
223 va_end(args);
224}
225
226/**
227 * ata_ehi_clear_desc - clean error description
228 * @ehi: target EHI
229 *
230 * Clear @ehi->desc.
231 *
232 * LOCKING:
233 * spin_lock_irqsave(host lock)
234 */
235void ata_ehi_clear_desc(struct ata_eh_info *ehi)
236{
237 ehi->desc[0] = '\0';
238 ehi->desc_len = 0;
239}
240
Tejun Heocbcdd872007-08-18 13:14:55 +0900241/**
242 * ata_port_desc - append port description
243 * @ap: target ATA port
244 * @fmt: printf format string
245 *
246 * Format string according to @fmt and append it to port
247 * description. If port description is not empty, " " is added
248 * in-between. This function is to be used while initializing
249 * ata_host. The description is printed on host registration.
250 *
251 * LOCKING:
252 * None.
253 */
254void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
255{
256 va_list args;
257
258 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
259
260 if (ap->link.eh_info.desc_len)
261 __ata_ehi_push_desc(&ap->link.eh_info, " ");
262
263 va_start(args, fmt);
264 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
265 va_end(args);
266}
267
268#ifdef CONFIG_PCI
269
270/**
271 * ata_port_pbar_desc - append PCI BAR description
272 * @ap: target ATA port
273 * @bar: target PCI BAR
274 * @offset: offset into PCI BAR
275 * @name: name of the area
276 *
277 * If @offset is negative, this function formats a string which
278 * contains the name, address, size and type of the BAR and
279 * appends it to the port description. If @offset is zero or
280 * positive, only name and offsetted address is appended.
281 *
282 * LOCKING:
283 * None.
284 */
285void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
286 const char *name)
287{
288 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
289 char *type = "";
290 unsigned long long start, len;
291
292 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
293 type = "m";
294 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
295 type = "i";
296
297 start = (unsigned long long)pci_resource_start(pdev, bar);
298 len = (unsigned long long)pci_resource_len(pdev, bar);
299
300 if (offset < 0)
301 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
302 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800303 ata_port_desc(ap, "%s 0x%llx", name,
304 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900305}
306
307#endif /* CONFIG_PCI */
308
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900309static int ata_lookup_timeout_table(u8 cmd)
310{
311 int i;
312
313 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
314 const u8 *cur;
315
316 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
317 if (*cur == cmd)
318 return i;
319 }
320
321 return -1;
322}
323
324/**
325 * ata_internal_cmd_timeout - determine timeout for an internal command
326 * @dev: target device
327 * @cmd: internal command to be issued
328 *
329 * Determine timeout for internal command @cmd for @dev.
330 *
331 * LOCKING:
332 * EH context.
333 *
334 * RETURNS:
335 * Determined timeout.
336 */
337unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
338{
339 struct ata_eh_context *ehc = &dev->link->eh_context;
340 int ent = ata_lookup_timeout_table(cmd);
341 int idx;
342
343 if (ent < 0)
344 return ATA_EH_CMD_DFL_TIMEOUT;
345
346 idx = ehc->cmd_timeout_idx[dev->devno][ent];
347 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
348}
349
350/**
351 * ata_internal_cmd_timed_out - notification for internal command timeout
352 * @dev: target device
353 * @cmd: internal command which timed out
354 *
355 * Notify EH that internal command @cmd for @dev timed out. This
356 * function should be called only for commands whose timeouts are
357 * determined using ata_internal_cmd_timeout().
358 *
359 * LOCKING:
360 * EH context.
361 */
362void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
363{
364 struct ata_eh_context *ehc = &dev->link->eh_context;
365 int ent = ata_lookup_timeout_table(cmd);
366 int idx;
367
368 if (ent < 0)
369 return;
370
371 idx = ehc->cmd_timeout_idx[dev->devno][ent];
372 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
373 ehc->cmd_timeout_idx[dev->devno][ent]++;
374}
375
Tejun Heo3884f7b2007-11-27 19:28:56 +0900376static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900377 unsigned int err_mask)
378{
379 struct ata_ering_entry *ent;
380
381 WARN_ON(!err_mask);
382
383 ering->cursor++;
384 ering->cursor %= ATA_ERING_SIZE;
385
386 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900387 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900388 ent->err_mask = err_mask;
389 ent->timestamp = get_jiffies_64();
390}
391
Tejun Heo76326ac2007-11-27 19:28:59 +0900392static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
393{
394 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
395
396 if (ent->err_mask)
397 return ent;
398 return NULL;
399}
400
Gwendal Grignoud9027472010-05-25 12:31:38 -0700401int ata_ering_map(struct ata_ering *ering,
402 int (*map_fn)(struct ata_ering_entry *, void *),
403 void *arg)
Tejun Heo0c247c52006-05-15 20:58:19 +0900404{
405 int idx, rc = 0;
406 struct ata_ering_entry *ent;
407
408 idx = ering->cursor;
409 do {
410 ent = &ering->ring[idx];
411 if (!ent->err_mask)
412 break;
413 rc = map_fn(ent, arg);
414 if (rc)
415 break;
416 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
417 } while (idx != ering->cursor);
418
419 return rc;
420}
421
H Hartley Sweeten60428402012-04-12 15:40:37 -0700422static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
Gwendal Grignoud9027472010-05-25 12:31:38 -0700423{
424 ent->eflags |= ATA_EFLAG_OLD_ER;
425 return 0;
426}
427
428static void ata_ering_clear(struct ata_ering *ering)
429{
430 ata_ering_map(ering, ata_ering_clear_cb, NULL);
431}
432
Tejun Heo64f65ca2006-06-24 20:30:18 +0900433static unsigned int ata_eh_dev_action(struct ata_device *dev)
434{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900435 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900436
437 return ehc->i.action | ehc->i.dev_action[dev->devno];
438}
439
Tejun Heof58229f2007-08-06 18:36:23 +0900440static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900441 struct ata_eh_info *ehi, unsigned int action)
442{
Tejun Heof58229f2007-08-06 18:36:23 +0900443 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900444
445 if (!dev) {
446 ehi->action &= ~action;
Tejun Heo1eca4362008-11-03 20:03:17 +0900447 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900448 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900449 } else {
450 /* doesn't make sense for port-wide EH actions */
451 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
452
453 /* break ehi->action into ehi->dev_action */
454 if (ehi->action & action) {
Tejun Heo1eca4362008-11-03 20:03:17 +0900455 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900456 ehi->dev_action[tdev->devno] |=
457 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900458 ehi->action &= ~action;
459 }
460
461 /* turn off the specified per-dev action */
462 ehi->dev_action[dev->devno] &= ~action;
463 }
464}
465
Tejun Heoece1d632006-04-02 18:51:53 +0900466/**
Tejun Heoc0c362b2010-09-06 17:57:14 +0200467 * ata_eh_acquire - acquire EH ownership
468 * @ap: ATA port to acquire EH ownership for
469 *
470 * Acquire EH ownership for @ap. This is the basic exclusion
471 * mechanism for ports sharing a host. Only one port hanging off
472 * the same host can claim the ownership of EH.
473 *
474 * LOCKING:
475 * EH context.
476 */
477void ata_eh_acquire(struct ata_port *ap)
478{
479 mutex_lock(&ap->host->eh_mutex);
480 WARN_ON_ONCE(ap->host->eh_owner);
481 ap->host->eh_owner = current;
482}
483
484/**
485 * ata_eh_release - release EH ownership
486 * @ap: ATA port to release EH ownership for
487 *
488 * Release EH ownership for @ap if the caller. The caller must
489 * have acquired EH ownership using ata_eh_acquire() previously.
490 *
491 * LOCKING:
492 * EH context.
493 */
494void ata_eh_release(struct ata_port *ap)
495{
496 WARN_ON_ONCE(ap->host->eh_owner != current);
497 ap->host->eh_owner = NULL;
498 mutex_unlock(&ap->host->eh_mutex);
499}
500
501/**
Tejun Heoece1d632006-04-02 18:51:53 +0900502 * ata_scsi_timed_out - SCSI layer time out callback
503 * @cmd: timed out SCSI command
504 *
505 * Handles SCSI layer timeout. We race with normal completion of
506 * the qc for @cmd. If the qc is already gone, we lose and let
507 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
508 * timed out and EH should be invoked. Prevent ata_qc_complete()
509 * from finishing it by setting EH_SCHEDULED and return
510 * EH_NOT_HANDLED.
511 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900512 * TODO: kill this function once old EH is gone.
513 *
Tejun Heoece1d632006-04-02 18:51:53 +0900514 * LOCKING:
515 * Called from timer context
516 *
517 * RETURNS:
518 * EH_HANDLED or EH_NOT_HANDLED
519 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700520enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
Tejun Heoece1d632006-04-02 18:51:53 +0900521{
522 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400523 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900524 unsigned long flags;
525 struct ata_queued_cmd *qc;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700526 enum blk_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900527
528 DPRINTK("ENTER\n");
529
Tejun Heoad9e2762006-05-15 20:58:12 +0900530 if (ap->ops->error_handler) {
Jens Axboe242f9dc2008-09-14 05:55:09 -0700531 ret = BLK_EH_NOT_HANDLED;
Tejun Heoad9e2762006-05-15 20:58:12 +0900532 goto out;
533 }
534
Jens Axboe242f9dc2008-09-14 05:55:09 -0700535 ret = BLK_EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400536 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900537 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900538 if (qc) {
539 WARN_ON(qc->scsicmd != cmd);
540 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
541 qc->err_mask |= AC_ERR_TIMEOUT;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700542 ret = BLK_EH_NOT_HANDLED;
Tejun Heoece1d632006-04-02 18:51:53 +0900543 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400544 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900545
Tejun Heoad9e2762006-05-15 20:58:12 +0900546 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900547 DPRINTK("EXIT, ret=%d\n", ret);
548 return ret;
549}
550
Tejun Heoece180d2008-11-03 20:04:37 +0900551static void ata_eh_unload(struct ata_port *ap)
552{
553 struct ata_link *link;
554 struct ata_device *dev;
555 unsigned long flags;
556
557 /* Restore SControl IPM and SPD for the next driver and
558 * disable attached devices.
559 */
560 ata_for_each_link(link, ap, PMP_FIRST) {
561 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
562 ata_for_each_dev(dev, link, ALL)
563 ata_dev_disable(dev);
564 }
565
566 /* freeze and set UNLOADED */
567 spin_lock_irqsave(ap->lock, flags);
568
569 ata_port_freeze(ap); /* won't be thawed */
570 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
571 ap->pflags |= ATA_PFLAG_UNLOADED;
572
573 spin_unlock_irqrestore(ap->lock, flags);
574}
575
Tejun Heoece1d632006-04-02 18:51:53 +0900576/**
577 * ata_scsi_error - SCSI layer error handler callback
578 * @host: SCSI host on which error occurred
579 *
580 * Handles SCSI-layer-thrown error events.
581 *
582 * LOCKING:
583 * Inherited from SCSI layer (none, can sleep)
584 *
585 * RETURNS:
586 * Zero.
587 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400588void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900589{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400590 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoad9e2762006-05-15 20:58:12 +0900591 unsigned long flags;
James Bottomleyc34aeeb2011-01-23 08:31:14 -0600592 LIST_HEAD(eh_work_q);
Tejun Heoece1d632006-04-02 18:51:53 +0900593
594 DPRINTK("ENTER\n");
595
James Bottomleyc34aeeb2011-01-23 08:31:14 -0600596 spin_lock_irqsave(host->host_lock, flags);
597 list_splice_init(&host->eh_cmd_q, &eh_work_q);
598 spin_unlock_irqrestore(host->host_lock, flags);
599
James Bottomley0e0b494c2011-01-23 09:42:50 -0600600 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
601
602 /* If we timed raced normal completion and there is nothing to
603 recover nr_timedout == 0 why exactly are we doing error recovery ? */
604 ata_scsi_port_error_handler(host, ap);
605
606 /* finish or retry handled scmd's and clean up */
607 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
608
609 DPRINTK("EXIT\n");
610}
611
612/**
613 * ata_scsi_cmd_error_handler - error callback for a list of commands
614 * @host: scsi host containing the port
615 * @ap: ATA port within the host
616 * @eh_work_q: list of commands to process
617 *
618 * process the given list of commands and return those finished to the
619 * ap->eh_done_q. This function is the first part of the libata error
620 * handler which processes a given list of failed commands.
621 */
622void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
623 struct list_head *eh_work_q)
624{
625 int i;
626 unsigned long flags;
627
Tejun Heoc4291372010-05-10 21:41:38 +0200628 /* make sure sff pio task is not running */
629 ata_sff_flush_pio_task(ap);
Tejun Heoece1d632006-04-02 18:51:53 +0900630
Jeff Garzikcca39742006-08-24 03:19:22 -0400631 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900632
Tejun Heoad9e2762006-05-15 20:58:12 +0900633 /* For new EH, all qcs are finished in one of three ways -
634 * normal completion, error completion, and SCSI timeout.
Alan Coxc96f1732009-03-24 10:23:46 +0000635 * Both completions can race against SCSI timeout. When normal
Tejun Heoad9e2762006-05-15 20:58:12 +0900636 * completion wins, the qc never reaches EH. When error
637 * completion wins, the qc has ATA_QCFLAG_FAILED set.
638 *
639 * When SCSI timeout wins, things are a bit more complex.
640 * Normal or error completion can occur after the timeout but
641 * before this point. In such cases, both types of
642 * completions are honored. A scmd is determined to have
643 * timed out iff its associated qc is active and not failed.
644 */
645 if (ap->ops->error_handler) {
646 struct scsi_cmnd *scmd, *tmp;
647 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900648
Tejun Heoe30349d2006-07-03 03:02:15 +0900649 spin_lock_irqsave(ap->lock, flags);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700650
Alan Coxc96f1732009-03-24 10:23:46 +0000651 /* This must occur under the ap->lock as we don't want
652 a polled recovery to race the real interrupt handler
Gwendal Grignoud9027472010-05-25 12:31:38 -0700653
Alan Coxc96f1732009-03-24 10:23:46 +0000654 The lost_interrupt handler checks for any completed but
655 non-notified command and completes much like an IRQ handler.
Gwendal Grignoud9027472010-05-25 12:31:38 -0700656
Alan Coxc96f1732009-03-24 10:23:46 +0000657 We then fall into the error recovery code which will treat
658 this as if normal completion won the race */
Tejun Heoad9e2762006-05-15 20:58:12 +0900659
Alan Coxc96f1732009-03-24 10:23:46 +0000660 if (ap->ops->lost_interrupt)
661 ap->ops->lost_interrupt(ap);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700662
James Bottomley0e0b494c2011-01-23 09:42:50 -0600663 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
Tejun Heoad9e2762006-05-15 20:58:12 +0900664 struct ata_queued_cmd *qc;
665
666 for (i = 0; i < ATA_MAX_QUEUE; i++) {
667 qc = __ata_qc_from_tag(ap, i);
668 if (qc->flags & ATA_QCFLAG_ACTIVE &&
669 qc->scsicmd == scmd)
670 break;
671 }
672
673 if (i < ATA_MAX_QUEUE) {
674 /* the scmd has an associated qc */
675 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
676 /* which hasn't failed yet, timeout */
677 qc->err_mask |= AC_ERR_TIMEOUT;
678 qc->flags |= ATA_QCFLAG_FAILED;
679 nr_timedout++;
680 }
681 } else {
682 /* Normal completion occurred after
683 * SCSI timeout but before this point.
684 * Successfully complete it.
685 */
686 scmd->retries = scmd->allowed;
687 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
688 }
689 }
690
691 /* If we have timed out qcs. They belong to EH from
692 * this point but the state of the controller is
693 * unknown. Freeze the port to make sure the IRQ
694 * handler doesn't diddle with those qcs. This must
695 * be done atomically w.r.t. setting QCFLAG_FAILED.
696 */
697 if (nr_timedout)
698 __ata_port_freeze(ap);
699
Tejun Heoe30349d2006-07-03 03:02:15 +0900700 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900701
702 /* initialize eh_tries */
703 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900704 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900705 spin_unlock_wait(ap->lock);
Gwendal Grignoud9027472010-05-25 12:31:38 -0700706
James Bottomley0e0b494c2011-01-23 09:42:50 -0600707}
708EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
709
710/**
711 * ata_scsi_port_error_handler - recover the port after the commands
712 * @host: SCSI host containing the port
713 * @ap: the ATA port
714 *
715 * Handle the recovery of the port @ap after all the commands
716 * have been recovered.
717 */
718void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
719{
720 unsigned long flags;
Tejun Heoad9e2762006-05-15 20:58:12 +0900721
Tejun Heoad9e2762006-05-15 20:58:12 +0900722 /* invoke error handler */
723 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900724 struct ata_link *link;
725
Tejun Heoc0c362b2010-09-06 17:57:14 +0200726 /* acquire EH ownership */
727 ata_eh_acquire(ap);
728 repeat:
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900729 /* kill fast drain timer */
730 del_timer_sync(&ap->fastdrain_timer);
731
Tejun Heo500530f2006-07-03 16:07:27 +0900732 /* process port resume request */
733 ata_eh_handle_port_resume(ap);
734
Tejun Heof3e81b12006-05-15 20:58:21 +0900735 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900736 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900737
Tejun Heo1eca4362008-11-03 20:03:17 +0900738 ata_for_each_link(link, ap, HOST_FIRST) {
Tejun Heo00115e02007-11-27 19:28:58 +0900739 struct ata_eh_context *ehc = &link->eh_context;
740 struct ata_device *dev;
741
Tejun Heocf1b86c2007-08-06 18:36:23 +0900742 memset(&link->eh_context, 0, sizeof(link->eh_context));
743 link->eh_context.i = link->eh_info;
744 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900745
Tejun Heo1eca4362008-11-03 20:03:17 +0900746 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +0900747 int devno = dev->devno;
748
749 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
750 if (ata_ncq_enabled(dev))
751 ehc->saved_ncq_enabled |= 1 << devno;
752 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900753 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900754
Tejun Heob51e9e52006-06-29 01:29:30 +0900755 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
756 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900757 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900758
Tejun Heoe30349d2006-07-03 03:02:15 +0900759 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900760
Tejun Heo500530f2006-07-03 16:07:27 +0900761 /* invoke EH, skip if unloading or suspended */
762 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900763 ap->ops->error_handler(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900764 else {
765 /* if unloading, commence suicide */
766 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
767 !(ap->pflags & ATA_PFLAG_UNLOADED))
768 ata_eh_unload(ap);
Tejun Heo720ba122006-05-31 18:28:13 +0900769 ata_eh_finish(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900770 }
Tejun Heoad9e2762006-05-15 20:58:12 +0900771
Tejun Heo500530f2006-07-03 16:07:27 +0900772 /* process port suspend request */
773 ata_eh_handle_port_suspend(ap);
774
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300775 /* Exception might have happened after ->error_handler
Tejun Heoad9e2762006-05-15 20:58:12 +0900776 * recovered the port but before this point. Repeat
777 * EH in such case.
778 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900779 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900780
Tejun Heob51e9e52006-06-29 01:29:30 +0900781 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900782 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900783 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900784 goto repeat;
785 }
Joe Perchesa9a79df2011-04-15 15:51:59 -0700786 ata_port_err(ap,
787 "EH pending after %d tries, giving up\n",
788 ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900789 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900790 }
791
Tejun Heof3e81b12006-05-15 20:58:21 +0900792 /* this run is complete, make sure EH info is clear */
Tejun Heo1eca4362008-11-03 20:03:17 +0900793 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heocf1b86c2007-08-06 18:36:23 +0900794 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900795
Dan Williamse4a9c372012-06-21 23:25:27 -0700796 /* end eh (clear host_eh_scheduled) while holding
797 * ap->lock such that if exception occurs after this
798 * point but before EH completion, SCSI midlayer will
Tejun Heoad9e2762006-05-15 20:58:12 +0900799 * re-initiate EH.
800 */
Dan Williamse4a9c372012-06-21 23:25:27 -0700801 ap->ops->end_eh(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900802
Tejun Heoe30349d2006-07-03 03:02:15 +0900803 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc0c362b2010-09-06 17:57:14 +0200804 ata_eh_release(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900805 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900806 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900807 ap->ops->eng_timeout(ap);
808 }
809
Tejun Heoece1d632006-04-02 18:51:53 +0900810 scsi_eh_flush_done_q(&ap->eh_done_q);
811
Tejun Heoad9e2762006-05-15 20:58:12 +0900812 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900813 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900814
Tejun Heo1cdaf532006-07-03 16:07:26 +0900815 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900816 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900817 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
Tejun Heoad72cf92010-07-02 10:03:52 +0200818 schedule_delayed_work(&ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900819
820 if (ap->pflags & ATA_PFLAG_RECOVERED)
Joe Perchesa9a79df2011-04-15 15:51:59 -0700821 ata_port_info(ap, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900822
Tejun Heob51e9e52006-06-29 01:29:30 +0900823 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900824
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900825 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900826 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900827 wake_up_all(&ap->eh_wait_q);
828
Tejun Heoe30349d2006-07-03 03:02:15 +0900829 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900830}
James Bottomley0e0b494c2011-01-23 09:42:50 -0600831EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
Tejun Heoece1d632006-04-02 18:51:53 +0900832
833/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900834 * ata_port_wait_eh - Wait for the currently pending EH to complete
835 * @ap: Port to wait EH for
836 *
837 * Wait until the currently pending EH is complete.
838 *
839 * LOCKING:
840 * Kernel thread context (may sleep).
841 */
842void ata_port_wait_eh(struct ata_port *ap)
843{
844 unsigned long flags;
845 DEFINE_WAIT(wait);
846
847 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400848 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900849
Tejun Heob51e9e52006-06-29 01:29:30 +0900850 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900851 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400852 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900853 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400854 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900855 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900856 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900857
Jeff Garzikba6a1302006-06-22 23:46:10 -0400858 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900859
860 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400861 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heo97750ce2010-09-06 17:56:29 +0200862 ata_msleep(ap, 10);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900863 goto retry;
864 }
865}
Dan Williams81c757b2011-12-02 16:07:01 -0800866EXPORT_SYMBOL_GPL(ata_port_wait_eh);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900867
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900868static int ata_eh_nr_in_flight(struct ata_port *ap)
869{
870 unsigned int tag;
871 int nr = 0;
872
873 /* count only non-internal commands */
874 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
875 if (ata_qc_from_tag(ap, tag))
876 nr++;
877
878 return nr;
879}
880
881void ata_eh_fastdrain_timerfn(unsigned long arg)
882{
883 struct ata_port *ap = (void *)arg;
884 unsigned long flags;
885 int cnt;
886
887 spin_lock_irqsave(ap->lock, flags);
888
889 cnt = ata_eh_nr_in_flight(ap);
890
891 /* are we done? */
892 if (!cnt)
893 goto out_unlock;
894
895 if (cnt == ap->fastdrain_cnt) {
896 unsigned int tag;
897
898 /* No progress during the last interval, tag all
899 * in-flight qcs as timed out and freeze the port.
900 */
901 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
902 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
903 if (qc)
904 qc->err_mask |= AC_ERR_TIMEOUT;
905 }
906
907 ata_port_freeze(ap);
908 } else {
909 /* some qcs have finished, give it another chance */
910 ap->fastdrain_cnt = cnt;
911 ap->fastdrain_timer.expires =
Tejun Heo341c2c92008-05-20 02:17:51 +0900912 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900913 add_timer(&ap->fastdrain_timer);
914 }
915
916 out_unlock:
917 spin_unlock_irqrestore(ap->lock, flags);
918}
919
920/**
921 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
922 * @ap: target ATA port
923 * @fastdrain: activate fast drain
924 *
925 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
926 * is non-zero and EH wasn't pending before. Fast drain ensures
927 * that EH kicks in in timely manner.
928 *
929 * LOCKING:
930 * spin_lock_irqsave(host lock)
931 */
932static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
933{
934 int cnt;
935
936 /* already scheduled? */
937 if (ap->pflags & ATA_PFLAG_EH_PENDING)
938 return;
939
940 ap->pflags |= ATA_PFLAG_EH_PENDING;
941
942 if (!fastdrain)
943 return;
944
945 /* do we have in-flight qcs? */
946 cnt = ata_eh_nr_in_flight(ap);
947 if (!cnt)
948 return;
949
950 /* activate fast drain */
951 ap->fastdrain_cnt = cnt;
Tejun Heo341c2c92008-05-20 02:17:51 +0900952 ap->fastdrain_timer.expires =
953 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900954 add_timer(&ap->fastdrain_timer);
955}
956
Tejun Heof686bcb2006-05-15 20:58:05 +0900957/**
958 * ata_qc_schedule_eh - schedule qc for error handling
959 * @qc: command to schedule error handling for
960 *
961 * Schedule error handling for @qc. EH will kick in as soon as
962 * other commands are drained.
963 *
964 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400965 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900966 */
967void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
968{
969 struct ata_port *ap = qc->ap;
Tejun Heofa41efd2010-04-15 08:57:37 +0900970 struct request_queue *q = qc->scsicmd->device->request_queue;
971 unsigned long flags;
Tejun Heof686bcb2006-05-15 20:58:05 +0900972
973 WARN_ON(!ap->ops->error_handler);
974
975 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900976 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900977
978 /* The following will fail if timeout has already expired.
979 * ata_scsi_error() takes care of such scmds on EH entry.
980 * Note that ATA_QCFLAG_FAILED is unconditionally set after
981 * this function completes.
982 */
Tejun Heofa41efd2010-04-15 08:57:37 +0900983 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700984 blk_abort_request(qc->scsicmd->request);
Tejun Heofa41efd2010-04-15 08:57:37 +0900985 spin_unlock_irqrestore(q->queue_lock, flags);
Tejun Heof686bcb2006-05-15 20:58:05 +0900986}
987
Tejun Heo7b70fc02006-05-15 20:58:07 +0900988/**
Dan Williamse4a9c372012-06-21 23:25:27 -0700989 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
990 * @ap: ATA port to schedule EH for
991 *
992 * LOCKING: inherited from ata_port_schedule_eh
993 * spin_lock_irqsave(host lock)
994 */
995void ata_std_sched_eh(struct ata_port *ap)
996{
997 WARN_ON(!ap->ops->error_handler);
998
999 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1000 return;
1001
1002 ata_eh_set_pending(ap, 1);
1003 scsi_schedule_eh(ap->scsi_host);
1004
1005 DPRINTK("port EH scheduled\n");
1006}
1007EXPORT_SYMBOL_GPL(ata_std_sched_eh);
1008
1009/**
1010 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1011 * @ap: ATA port to end EH for
1012 *
1013 * In the libata object model there is a 1:1 mapping of ata_port to
1014 * shost, so host fields can be directly manipulated under ap->lock, in
1015 * the libsas case we need to hold a lock at the ha->level to coordinate
1016 * these events.
1017 *
1018 * LOCKING:
1019 * spin_lock_irqsave(host lock)
1020 */
1021void ata_std_end_eh(struct ata_port *ap)
1022{
1023 struct Scsi_Host *host = ap->scsi_host;
1024
1025 host->host_eh_scheduled = 0;
1026}
1027EXPORT_SYMBOL(ata_std_end_eh);
1028
1029
1030/**
Tejun Heo7b70fc02006-05-15 20:58:07 +09001031 * ata_port_schedule_eh - schedule error handling without a qc
1032 * @ap: ATA port to schedule EH for
1033 *
1034 * Schedule error handling for @ap. EH will kick in as soon as
1035 * all commands are drained.
1036 *
1037 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001038 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +09001039 */
1040void ata_port_schedule_eh(struct ata_port *ap)
1041{
Dan Williamse4a9c372012-06-21 23:25:27 -07001042 /* see: ata_std_sched_eh, unless you know better */
1043 ap->ops->sched_eh(ap);
Tejun Heo7b70fc02006-05-15 20:58:07 +09001044}
1045
Tejun Heodbd82612007-08-06 18:36:23 +09001046static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +09001047{
1048 int tag, nr_aborted = 0;
1049
1050 WARN_ON(!ap->ops->error_handler);
1051
Tejun Heo5ddf24c2007-07-16 14:29:41 +09001052 /* we're gonna abort all commands, no need for fast drain */
1053 ata_eh_set_pending(ap, 0);
1054
Tejun Heo7b70fc02006-05-15 20:58:07 +09001055 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1056 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1057
Tejun Heodbd82612007-08-06 18:36:23 +09001058 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +09001059 qc->flags |= ATA_QCFLAG_FAILED;
1060 ata_qc_complete(qc);
1061 nr_aborted++;
1062 }
1063 }
1064
1065 if (!nr_aborted)
1066 ata_port_schedule_eh(ap);
1067
1068 return nr_aborted;
1069}
1070
Tejun Heoe3180492006-05-15 20:58:09 +09001071/**
Tejun Heodbd82612007-08-06 18:36:23 +09001072 * ata_link_abort - abort all qc's on the link
1073 * @link: ATA link to abort qc's for
1074 *
1075 * Abort all active qc's active on @link and schedule EH.
1076 *
1077 * LOCKING:
1078 * spin_lock_irqsave(host lock)
1079 *
1080 * RETURNS:
1081 * Number of aborted qc's.
1082 */
1083int ata_link_abort(struct ata_link *link)
1084{
1085 return ata_do_link_abort(link->ap, link);
1086}
1087
1088/**
1089 * ata_port_abort - abort all qc's on the port
1090 * @ap: ATA port to abort qc's for
1091 *
1092 * Abort all active qc's of @ap and schedule EH.
1093 *
1094 * LOCKING:
1095 * spin_lock_irqsave(host_set lock)
1096 *
1097 * RETURNS:
1098 * Number of aborted qc's.
1099 */
1100int ata_port_abort(struct ata_port *ap)
1101{
1102 return ata_do_link_abort(ap, NULL);
1103}
1104
1105/**
Tejun Heoe3180492006-05-15 20:58:09 +09001106 * __ata_port_freeze - freeze port
1107 * @ap: ATA port to freeze
1108 *
1109 * This function is called when HSM violation or some other
1110 * condition disrupts normal operation of the port. Frozen port
1111 * is not allowed to perform any operation until the port is
1112 * thawed, which usually follows a successful reset.
1113 *
1114 * ap->ops->freeze() callback can be used for freezing the port
1115 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1116 * port cannot be frozen hardware-wise, the interrupt handler
1117 * must ack and clear interrupts unconditionally while the port
1118 * is frozen.
1119 *
1120 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001121 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001122 */
1123static void __ata_port_freeze(struct ata_port *ap)
1124{
1125 WARN_ON(!ap->ops->error_handler);
1126
1127 if (ap->ops->freeze)
1128 ap->ops->freeze(ap);
1129
Tejun Heob51e9e52006-06-29 01:29:30 +09001130 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001131
Tejun Heo44877b42007-02-21 01:06:51 +09001132 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001133}
1134
1135/**
1136 * ata_port_freeze - abort & freeze port
1137 * @ap: ATA port to freeze
1138 *
Jeff Garzik54c38442009-04-07 19:13:15 -04001139 * Abort and freeze @ap. The freeze operation must be called
1140 * first, because some hardware requires special operations
1141 * before the taskfile registers are accessible.
Tejun Heoe3180492006-05-15 20:58:09 +09001142 *
1143 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001144 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001145 *
1146 * RETURNS:
1147 * Number of aborted commands.
1148 */
1149int ata_port_freeze(struct ata_port *ap)
1150{
1151 int nr_aborted;
1152
1153 WARN_ON(!ap->ops->error_handler);
1154
Tejun Heoe3180492006-05-15 20:58:09 +09001155 __ata_port_freeze(ap);
Jeff Garzik54c38442009-04-07 19:13:15 -04001156 nr_aborted = ata_port_abort(ap);
Tejun Heoe3180492006-05-15 20:58:09 +09001157
1158 return nr_aborted;
1159}
1160
1161/**
Tejun Heo7d77b242007-09-23 13:14:13 +09001162 * sata_async_notification - SATA async notification handler
1163 * @ap: ATA port where async notification is received
1164 *
1165 * Handler to be called when async notification via SDB FIS is
1166 * received. This function schedules EH if necessary.
1167 *
1168 * LOCKING:
1169 * spin_lock_irqsave(host lock)
1170 *
1171 * RETURNS:
1172 * 1 if EH is scheduled, 0 otherwise.
1173 */
1174int sata_async_notification(struct ata_port *ap)
1175{
1176 u32 sntf;
1177 int rc;
1178
1179 if (!(ap->flags & ATA_FLAG_AN))
1180 return 0;
1181
1182 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1183 if (rc == 0)
1184 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1185
Tejun Heo071f44b2008-04-07 22:47:22 +09001186 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001187 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +09001188 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001189 /* PMP is not attached. Check whether ATAPI
1190 * AN is configured. If so, notify media
1191 * change.
1192 */
1193 struct ata_device *dev = ap->link.device;
1194
1195 if ((dev->class == ATA_DEV_ATAPI) &&
1196 (dev->flags & ATA_DFLAG_AN))
1197 ata_scsi_media_change_notify(dev);
1198 return 0;
1199 } else {
1200 /* PMP is attached but SNTF is not available.
1201 * ATAPI async media change notification is
1202 * not used. The PMP must be reporting PHY
1203 * status change, schedule EH.
1204 */
1205 ata_port_schedule_eh(ap);
1206 return 1;
1207 }
1208 } else {
1209 /* PMP is attached and SNTF is available */
1210 struct ata_link *link;
1211
1212 /* check and notify ATAPI AN */
Tejun Heo1eca4362008-11-03 20:03:17 +09001213 ata_for_each_link(link, ap, EDGE) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001214 if (!(sntf & (1 << link->pmp)))
1215 continue;
1216
1217 if ((link->device->class == ATA_DEV_ATAPI) &&
1218 (link->device->flags & ATA_DFLAG_AN))
1219 ata_scsi_media_change_notify(link->device);
1220 }
1221
1222 /* If PMP is reporting that PHY status of some
1223 * downstream ports has changed, schedule EH.
1224 */
1225 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1226 ata_port_schedule_eh(ap);
1227 return 1;
1228 }
1229
1230 return 0;
1231 }
1232}
1233
1234/**
Tejun Heoe3180492006-05-15 20:58:09 +09001235 * ata_eh_freeze_port - EH helper to freeze port
1236 * @ap: ATA port to freeze
1237 *
1238 * Freeze @ap.
1239 *
1240 * LOCKING:
1241 * None.
1242 */
1243void ata_eh_freeze_port(struct ata_port *ap)
1244{
1245 unsigned long flags;
1246
1247 if (!ap->ops->error_handler)
1248 return;
1249
Jeff Garzikba6a1302006-06-22 23:46:10 -04001250 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001251 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -04001252 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001253}
1254
1255/**
1256 * ata_port_thaw_port - EH helper to thaw port
1257 * @ap: ATA port to thaw
1258 *
1259 * Thaw frozen port @ap.
1260 *
1261 * LOCKING:
1262 * None.
1263 */
1264void ata_eh_thaw_port(struct ata_port *ap)
1265{
1266 unsigned long flags;
1267
1268 if (!ap->ops->error_handler)
1269 return;
1270
Jeff Garzikba6a1302006-06-22 23:46:10 -04001271 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001272
Tejun Heob51e9e52006-06-29 01:29:30 +09001273 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001274
1275 if (ap->ops->thaw)
1276 ap->ops->thaw(ap);
1277
Jeff Garzikba6a1302006-06-22 23:46:10 -04001278 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001279
Tejun Heo44877b42007-02-21 01:06:51 +09001280 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001281}
1282
Tejun Heoece1d632006-04-02 18:51:53 +09001283static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1284{
1285 /* nada */
1286}
1287
1288static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1289{
1290 struct ata_port *ap = qc->ap;
1291 struct scsi_cmnd *scmd = qc->scsicmd;
1292 unsigned long flags;
1293
Jeff Garzikba6a1302006-06-22 23:46:10 -04001294 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001295 qc->scsidone = ata_eh_scsidone;
1296 __ata_qc_complete(qc);
1297 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -04001298 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001299
1300 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1301}
1302
1303/**
1304 * ata_eh_qc_complete - Complete an active ATA command from EH
1305 * @qc: Command to complete
1306 *
1307 * Indicate to the mid and upper layers that an ATA command has
1308 * completed. To be used from EH.
1309 */
1310void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1311{
1312 struct scsi_cmnd *scmd = qc->scsicmd;
1313 scmd->retries = scmd->allowed;
1314 __ata_eh_qc_complete(qc);
1315}
1316
1317/**
1318 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1319 * @qc: Command to retry
1320 *
1321 * Indicate to the mid and upper layers that an ATA command
1322 * should be retried. To be used from EH.
1323 *
1324 * SCSI midlayer limits the number of retries to scmd->allowed.
1325 * scmd->retries is decremented for commands which get retried
1326 * due to unrelated failures (qc->err_mask is zero).
1327 */
1328void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1329{
1330 struct scsi_cmnd *scmd = qc->scsicmd;
1331 if (!qc->err_mask && scmd->retries)
1332 scmd->retries--;
1333 __ata_eh_qc_complete(qc);
1334}
Tejun Heo022bdb02006-05-15 20:58:22 +09001335
1336/**
Tejun Heo678afac2009-01-29 20:31:30 +09001337 * ata_dev_disable - disable ATA device
1338 * @dev: ATA device to disable
1339 *
1340 * Disable @dev.
1341 *
1342 * Locking:
1343 * EH context.
1344 */
1345void ata_dev_disable(struct ata_device *dev)
1346{
1347 if (!ata_dev_enabled(dev))
1348 return;
1349
1350 if (ata_msg_drv(dev->link->ap))
Joe Perchesa9a79df2011-04-15 15:51:59 -07001351 ata_dev_warn(dev, "disabled\n");
Tejun Heo678afac2009-01-29 20:31:30 +09001352 ata_acpi_on_disable(dev);
1353 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1354 dev->class++;
Tejun Heo99cf6102009-01-29 20:31:32 +09001355
1356 /* From now till the next successful probe, ering is used to
1357 * track probe failures. Clear accumulated device error info.
1358 */
1359 ata_ering_clear(&dev->ering);
Tejun Heo678afac2009-01-29 20:31:30 +09001360}
1361
1362/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001363 * ata_eh_detach_dev - detach ATA device
1364 * @dev: ATA device to detach
1365 *
1366 * Detach @dev.
1367 *
1368 * LOCKING:
1369 * None.
1370 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001371void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001372{
Tejun Heof58229f2007-08-06 18:36:23 +09001373 struct ata_link *link = dev->link;
1374 struct ata_port *ap = link->ap;
Tejun Heo90484eb2008-10-26 15:43:03 +09001375 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001376 unsigned long flags;
1377
1378 ata_dev_disable(dev);
1379
Jeff Garzikba6a1302006-06-22 23:46:10 -04001380 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001381
1382 dev->flags &= ~ATA_DFLAG_DETACH;
1383
1384 if (ata_scsi_offline_dev(dev)) {
1385 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001386 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001387 }
1388
Tejun Heo90484eb2008-10-26 15:43:03 +09001389 /* clear per-dev EH info */
Tejun Heof58229f2007-08-06 18:36:23 +09001390 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1391 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heo90484eb2008-10-26 15:43:03 +09001392 ehc->saved_xfer_mode[dev->devno] = 0;
1393 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heobeb07c12006-06-24 20:30:19 +09001394
Jeff Garzikba6a1302006-06-22 23:46:10 -04001395 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001396}
1397
1398/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001399 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001400 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001401 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001402 * @action: action about to be performed
1403 *
1404 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001405 * in @link->eh_info such that eh actions are not unnecessarily
1406 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001407 *
1408 * LOCKING:
1409 * None.
1410 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001411void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1412 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001413{
Tejun Heo955e57d2007-08-06 18:36:23 +09001414 struct ata_port *ap = link->ap;
1415 struct ata_eh_info *ehi = &link->eh_info;
1416 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001417 unsigned long flags;
1418
Jeff Garzikba6a1302006-06-22 23:46:10 -04001419 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001420
Tejun Heo955e57d2007-08-06 18:36:23 +09001421 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001422
Tejun Heoa568d1d2008-10-21 20:37:21 +09001423 /* About to take EH action, set RECOVERED. Ignore actions on
1424 * slave links as master will do them again.
1425 */
1426 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
Tejun Heo1cdaf532006-07-03 16:07:26 +09001427 ap->pflags |= ATA_PFLAG_RECOVERED;
1428
Jeff Garzikba6a1302006-06-22 23:46:10 -04001429 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001430}
1431
1432/**
Tejun Heo47005f22006-06-19 18:27:23 +09001433 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001434* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001435 * @dev: target ATA dev for per-dev action (can be NULL)
1436 * @action: action just completed
1437 *
1438 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001439 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001440 *
1441 * LOCKING:
1442 * None.
1443 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001444void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1445 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001446{
Tejun Heo955e57d2007-08-06 18:36:23 +09001447 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001448
Tejun Heo955e57d2007-08-06 18:36:23 +09001449 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001450}
1451
1452/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001453 * ata_err_string - convert err_mask to descriptive string
1454 * @err_mask: error mask to convert to string
1455 *
1456 * Convert @err_mask to descriptive string. Errors are
1457 * prioritized according to severity and only the most severe
1458 * error is reported.
1459 *
1460 * LOCKING:
1461 * None.
1462 *
1463 * RETURNS:
1464 * Descriptive string for @err_mask
1465 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001466static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001467{
1468 if (err_mask & AC_ERR_HOST_BUS)
1469 return "host bus error";
1470 if (err_mask & AC_ERR_ATA_BUS)
1471 return "ATA bus error";
1472 if (err_mask & AC_ERR_TIMEOUT)
1473 return "timeout";
1474 if (err_mask & AC_ERR_HSM)
1475 return "HSM violation";
1476 if (err_mask & AC_ERR_SYSTEM)
1477 return "internal error";
1478 if (err_mask & AC_ERR_MEDIA)
1479 return "media error";
1480 if (err_mask & AC_ERR_INVALID)
1481 return "invalid argument";
1482 if (err_mask & AC_ERR_DEV)
1483 return "device error";
1484 return "unknown error";
1485}
1486
1487/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001488 * ata_read_log_page - read a specific log page
1489 * @dev: target device
Shane Huang65fe1f02012-09-07 22:40:01 +08001490 * @log: log to read
Tejun Heoe8ee8452006-05-15 21:03:46 +09001491 * @page: page to read
1492 * @buf: buffer to store read page
1493 * @sectors: number of sectors to read
1494 *
1495 * Read log page using READ_LOG_EXT command.
1496 *
1497 * LOCKING:
1498 * Kernel thread context (may sleep).
1499 *
1500 * RETURNS:
1501 * 0 on success, AC_ERR_* mask otherwise.
1502 */
Shane Huang65fe1f02012-09-07 22:40:01 +08001503unsigned int ata_read_log_page(struct ata_device *dev, u8 log,
1504 u8 page, void *buf, unsigned int sectors)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001505{
1506 struct ata_taskfile tf;
1507 unsigned int err_mask;
1508
Shane Huang65fe1f02012-09-07 22:40:01 +08001509 DPRINTK("read log page - log 0x%x, page 0x%x\n", log, page);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001510
1511 ata_tf_init(dev, &tf);
1512 tf.command = ATA_CMD_READ_LOG_EXT;
Shane Huang65fe1f02012-09-07 22:40:01 +08001513 tf.lbal = log;
1514 tf.lbam = page;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001515 tf.nsect = sectors;
1516 tf.hob_nsect = sectors >> 8;
1517 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1518 tf.protocol = ATA_PROT_PIO;
1519
1520 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001521 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001522
1523 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1524 return err_mask;
1525}
1526
1527/**
1528 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1529 * @dev: Device to read log page 10h from
1530 * @tag: Resulting tag of the failed command
1531 * @tf: Resulting taskfile registers of the failed command
1532 *
1533 * Read log page 10h to obtain NCQ error details and clear error
1534 * condition.
1535 *
1536 * LOCKING:
1537 * Kernel thread context (may sleep).
1538 *
1539 * RETURNS:
1540 * 0 on success, -errno otherwise.
1541 */
1542static int ata_eh_read_log_10h(struct ata_device *dev,
1543 int *tag, struct ata_taskfile *tf)
1544{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001545 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001546 unsigned int err_mask;
1547 u8 csum;
1548 int i;
1549
Shane Huang65fe1f02012-09-07 22:40:01 +08001550 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001551 if (err_mask)
1552 return -EIO;
1553
1554 csum = 0;
1555 for (i = 0; i < ATA_SECT_SIZE; i++)
1556 csum += buf[i];
1557 if (csum)
Joe Perchesa9a79df2011-04-15 15:51:59 -07001558 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1559 csum);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001560
1561 if (buf[0] & 0x80)
1562 return -ENOENT;
1563
1564 *tag = buf[0] & 0x1f;
1565
1566 tf->command = buf[2];
1567 tf->feature = buf[3];
1568 tf->lbal = buf[4];
1569 tf->lbam = buf[5];
1570 tf->lbah = buf[6];
1571 tf->device = buf[7];
1572 tf->hob_lbal = buf[8];
1573 tf->hob_lbam = buf[9];
1574 tf->hob_lbah = buf[10];
1575 tf->nsect = buf[12];
1576 tf->hob_nsect = buf[13];
1577
1578 return 0;
1579}
1580
1581/**
Tejun Heo11fc33d2008-08-30 14:20:01 +02001582 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1583 * @dev: target ATAPI device
1584 * @r_sense_key: out parameter for sense_key
1585 *
1586 * Perform ATAPI TEST_UNIT_READY.
1587 *
1588 * LOCKING:
1589 * EH context (may sleep).
1590 *
1591 * RETURNS:
1592 * 0 on success, AC_ERR_* mask on failure.
1593 */
Aaron Lu3dc67442013-01-15 17:21:00 +08001594unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
Tejun Heo11fc33d2008-08-30 14:20:01 +02001595{
1596 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1597 struct ata_taskfile tf;
1598 unsigned int err_mask;
1599
1600 ata_tf_init(dev, &tf);
1601
1602 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1603 tf.command = ATA_CMD_PACKET;
1604 tf.protocol = ATAPI_PROT_NODATA;
1605
1606 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1607 if (err_mask == AC_ERR_DEV)
1608 *r_sense_key = tf.feature >> 4;
1609 return err_mask;
1610}
1611
1612/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001613 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1614 * @dev: device to perform REQUEST_SENSE to
1615 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
Tejun Heo3eabddb2008-06-10 18:28:05 +09001616 * @dfl_sense_key: default sense key to use
Tejun Heo022bdb02006-05-15 20:58:22 +09001617 *
1618 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1619 * SENSE. This function is EH helper.
1620 *
1621 * LOCKING:
1622 * Kernel thread context (may sleep).
1623 *
1624 * RETURNS:
1625 * 0 on success, AC_ERR_* mask on failure
1626 */
Aaron Lu3dc67442013-01-15 17:21:00 +08001627unsigned int atapi_eh_request_sense(struct ata_device *dev,
Tejun Heo3eabddb2008-06-10 18:28:05 +09001628 u8 *sense_buf, u8 dfl_sense_key)
Tejun Heo022bdb02006-05-15 20:58:22 +09001629{
Tejun Heo3eabddb2008-06-10 18:28:05 +09001630 u8 cdb[ATAPI_CDB_LEN] =
1631 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001632 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001633 struct ata_taskfile tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001634
1635 DPRINTK("ATAPI request sense\n");
1636
Tejun Heo022bdb02006-05-15 20:58:22 +09001637 /* FIXME: is this needed? */
1638 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1639
Albert Lee56287762007-04-02 11:30:46 +08001640 /* initialize sense_buf with the error register,
1641 * for the case where they are -not- overwritten
1642 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001643 sense_buf[0] = 0x70;
Tejun Heo3eabddb2008-06-10 18:28:05 +09001644 sense_buf[2] = dfl_sense_key;
Albert Lee56287762007-04-02 11:30:46 +08001645
Jeff Garzika617c092007-05-21 20:14:23 -04001646 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001647 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001648
Tejun Heo022bdb02006-05-15 20:58:22 +09001649 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1650 tf.command = ATA_CMD_PACKET;
1651
1652 /* is it pointless to prefer PIO for "safety reasons"? */
1653 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001654 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001655 tf.feature |= ATAPI_PKT_DMA;
1656 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001657 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001658 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1659 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001660 }
1661
1662 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001663 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001664}
1665
1666/**
1667 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001668 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001669 *
1670 * Analyze SError if available and further determine cause of
1671 * failure.
1672 *
1673 * LOCKING:
1674 * None.
1675 */
Tejun Heo02607312007-08-06 18:36:23 +09001676static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001677{
Tejun Heo02607312007-08-06 18:36:23 +09001678 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001679 u32 serror = ehc->i.serror;
1680 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001681 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001682
Tejun Heoe0614db2008-05-19 01:15:09 +09001683 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001684 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001685 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001686 }
1687 if (serror & SERR_PROTOCOL) {
1688 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001689 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001690 }
1691 if (serror & SERR_INTERNAL) {
1692 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001693 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001694 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001695
1696 /* Determine whether a hotplug event has occurred. Both
1697 * SError.N/X are considered hotplug events for enabled or
1698 * host links. For disabled PMP links, only N bit is
1699 * considered as X bit is left at 1 for link plugging.
1700 */
Tejun Heoeb0e85e2011-02-24 19:30:37 +01001701 if (link->lpm_policy > ATA_LPM_MAX_POWER)
Tejun Heo6b7ae952010-09-01 17:50:06 +02001702 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1703 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
Tejun Heof9df58c2007-09-23 13:14:13 +09001704 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1705 else
1706 hotplug_mask = SERR_PHYRDY_CHG;
1707
1708 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001709 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001710
1711 ehc->i.err_mask |= err_mask;
1712 ehc->i.action |= action;
1713}
1714
1715/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001716 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001717 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001718 *
1719 * Read log page 10h, determine the offending qc and acquire
1720 * error status TF. For NCQ device errors, all LLDDs have to do
1721 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1722 * care of the rest.
1723 *
1724 * LOCKING:
1725 * Kernel thread context (may sleep).
1726 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001727void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001728{
Tejun Heo02607312007-08-06 18:36:23 +09001729 struct ata_port *ap = link->ap;
1730 struct ata_eh_context *ehc = &link->eh_context;
1731 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001732 struct ata_queued_cmd *qc;
1733 struct ata_taskfile tf;
1734 int tag, rc;
1735
1736 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001737 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001738 return;
1739
1740 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001741 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001742 return;
1743
1744 /* has LLDD analyzed already? */
1745 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1746 qc = __ata_qc_from_tag(ap, tag);
1747
1748 if (!(qc->flags & ATA_QCFLAG_FAILED))
1749 continue;
1750
1751 if (qc->err_mask)
1752 return;
1753 }
1754
1755 /* okay, this error is ours */
Jeff Garzika09bf4c2010-04-22 21:59:13 -04001756 memset(&tf, 0, sizeof(tf));
Tejun Heoe8ee8452006-05-15 21:03:46 +09001757 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1758 if (rc) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07001759 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1760 rc);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001761 return;
1762 }
1763
Tejun Heo02607312007-08-06 18:36:23 +09001764 if (!(link->sactive & (1 << tag))) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07001765 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1766 tag);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001767 return;
1768 }
1769
1770 /* we've got the perpetrator, condemn it */
1771 qc = __ata_qc_from_tag(ap, tag);
1772 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001773 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001774 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001775 ehc->i.err_mask &= ~AC_ERR_DEV;
1776}
1777
1778/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001779 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1780 * @qc: qc to analyze
1781 * @tf: Taskfile registers to analyze
1782 *
1783 * Analyze taskfile of @qc and further determine cause of
1784 * failure. This function also requests ATAPI sense data if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001785 * available.
Tejun Heo022bdb02006-05-15 20:58:22 +09001786 *
1787 * LOCKING:
1788 * Kernel thread context (may sleep).
1789 *
1790 * RETURNS:
1791 * Determined recovery action
1792 */
1793static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1794 const struct ata_taskfile *tf)
1795{
1796 unsigned int tmp, action = 0;
1797 u8 stat = tf->command, err = tf->feature;
1798
1799 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1800 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001801 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001802 }
1803
Tejun Heoa51d6442007-03-20 15:24:11 +09001804 if (stat & (ATA_ERR | ATA_DF))
1805 qc->err_mask |= AC_ERR_DEV;
1806 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001807 return 0;
1808
1809 switch (qc->dev->class) {
1810 case ATA_DEV_ATA:
1811 if (err & ATA_ICRC)
1812 qc->err_mask |= AC_ERR_ATA_BUS;
1813 if (err & ATA_UNC)
1814 qc->err_mask |= AC_ERR_MEDIA;
1815 if (err & ATA_IDNF)
1816 qc->err_mask |= AC_ERR_INVALID;
1817 break;
1818
1819 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001820 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Tejun Heo3eabddb2008-06-10 18:28:05 +09001821 tmp = atapi_eh_request_sense(qc->dev,
1822 qc->scsicmd->sense_buffer,
1823 qc->result_tf.feature >> 4);
Tejun Heoa569a302006-11-21 10:40:51 +09001824 if (!tmp) {
1825 /* ATA_QCFLAG_SENSE_VALID is used to
1826 * tell atapi_qc_complete() that sense
1827 * data is already valid.
1828 *
1829 * TODO: interpret sense data and set
1830 * appropriate err_mask.
1831 */
1832 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1833 } else
1834 qc->err_mask |= tmp;
1835 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001836 }
1837
1838 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001839 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001840
1841 return action;
1842}
1843
Tejun Heo76326ac2007-11-27 19:28:59 +09001844static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1845 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001846{
Tejun Heo76326ac2007-11-27 19:28:59 +09001847 int base = 0;
1848
1849 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1850 *xfer_ok = 1;
1851
1852 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001853 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001854
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001855 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001856 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001857
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001858 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001859 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001860
Tejun Heo3884f7b2007-11-27 19:28:56 +09001861 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001862 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001863 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001864 if ((err_mask &
1865 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001866 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001867 }
1868
1869 return 0;
1870}
1871
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001872struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001873 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001874 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001875 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001876};
1877
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001878static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001879{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001880 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001881 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001882
Gwendal Grignoud9027472010-05-25 12:31:38 -07001883 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
Tejun Heo022bdb02006-05-15 20:58:22 +09001884 return -1;
1885
Tejun Heo76326ac2007-11-27 19:28:59 +09001886 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1887 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001888 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001889
Tejun Heo022bdb02006-05-15 20:58:22 +09001890 return 0;
1891}
1892
1893/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001894 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001895 * @dev: Device of interest
1896 *
1897 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001898 * whether NCQ needs to be turned off, transfer speed should be
1899 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001900 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001901 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001902 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001903 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1904 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001905 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001906 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001907 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001908 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1909 * data transfer hasn't been verified.
1910 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001911 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001912 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001913 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001914 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001915 * SPEED_DOWN : Speed down transfer speed but don't fall back
1916 * to PIO.
1917 *
1918 * FALLBACK_TO_PIO : Fall back to PIO.
1919 *
1920 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001921 * taken per error. An action triggered by non-DUBIOUS errors
1922 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1923 * This is to expedite speed down decisions right after device is
1924 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001925 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001926 * The followings are speed down rules. #1 and #2 deal with
1927 * DUBIOUS errors.
1928 *
1929 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1930 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1931 *
1932 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1933 * occurred during last 5 mins, NCQ_OFF.
1934 *
1935 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001936 * occurred during last 5 mins, FALLBACK_TO_PIO
Tejun Heo3884f7b2007-11-27 19:28:56 +09001937 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001938 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001939 * during last 10 mins, NCQ_OFF.
1940 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001941 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001942 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001943 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001944 * LOCKING:
1945 * Inherited from caller.
1946 *
1947 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001948 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001949 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001950static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001951{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001952 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1953 u64 j64 = get_jiffies_64();
1954 struct speed_down_verdict_arg arg;
1955 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001956
Tejun Heo3884f7b2007-11-27 19:28:56 +09001957 /* scan past 5 mins of error history */
1958 memset(&arg, 0, sizeof(arg));
1959 arg.since = j64 - min(j64, j5mins);
1960 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1961
Tejun Heo76326ac2007-11-27 19:28:59 +09001962 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1963 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1964 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1965 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1966
1967 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1968 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1969 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1970
Tejun Heo3884f7b2007-11-27 19:28:56 +09001971 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1972 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001973 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001974 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1975
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001976 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001977 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001978 arg.since = j64 - min(j64, j10mins);
1979 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001980
Tejun Heo3884f7b2007-11-27 19:28:56 +09001981 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1982 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001983 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001984
1985 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1986 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001987 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001988 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001989
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001990 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001991}
1992
1993/**
1994 * ata_eh_speed_down - record error and speed down if necessary
1995 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001996 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001997 * @err_mask: err_mask of the error
1998 *
1999 * Record error and examine error history to determine whether
2000 * adjusting transmission speed is necessary. It also sets
2001 * transmission limits appropriately if such adjustment is
2002 * necessary.
2003 *
2004 * LOCKING:
2005 * Kernel thread context (may sleep).
2006 *
2007 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002008 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09002009 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09002010static unsigned int ata_eh_speed_down(struct ata_device *dev,
2011 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002012{
Tejun Heob1c72912008-07-31 17:02:43 +09002013 struct ata_link *link = ata_dev_phys_link(dev);
Tejun Heo76326ac2007-11-27 19:28:59 +09002014 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002015 unsigned int verdict;
2016 unsigned int action = 0;
2017
2018 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09002019 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09002020 return 0;
2021
2022 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09002023 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002024 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09002025
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002026 /* turn off NCQ? */
2027 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2028 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2029 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2030 dev->flags |= ATA_DFLAG_NCQ_OFF;
Joe Perchesa9a79df2011-04-15 15:51:59 -07002031 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002032 goto done;
2033 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002034
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002035 /* speed down? */
2036 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2037 /* speed down SATA link speed if possible */
Tejun Heoa07d4992009-01-29 20:31:33 +09002038 if (sata_down_spd_limit(link, 0) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002039 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002040 goto done;
2041 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002042
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002043 /* lower transfer mode */
2044 if (dev->spdn_cnt < 2) {
2045 static const int dma_dnxfer_sel[] =
2046 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2047 static const int pio_dnxfer_sel[] =
2048 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2049 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09002050
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002051 if (dev->xfer_shift != ATA_SHIFT_PIO)
2052 sel = dma_dnxfer_sel[dev->spdn_cnt];
2053 else
2054 sel = pio_dnxfer_sel[dev->spdn_cnt];
2055
2056 dev->spdn_cnt++;
2057
2058 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002059 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002060 goto done;
2061 }
2062 }
2063 }
2064
2065 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09002066 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002067 */
2068 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09002069 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002070 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2071 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2072 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09002073 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002074 goto done;
2075 }
2076 }
2077
Tejun Heo022bdb02006-05-15 20:58:22 +09002078 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002079 done:
2080 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09002081 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2082 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002083 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09002084}
2085
2086/**
Mark Lord8d899e72012-05-02 15:22:52 -04002087 * ata_eh_worth_retry - analyze error and decide whether to retry
2088 * @qc: qc to possibly retry
2089 *
2090 * Look at the cause of the error and decide if a retry
2091 * might be useful or not. We don't want to retry media errors
2092 * because the drive itself has probably already taken 10-30 seconds
2093 * doing its own internal retries before reporting the failure.
2094 */
2095static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2096{
Bian Yu1eaca392012-12-12 22:26:58 -05002097 if (qc->err_mask & AC_ERR_MEDIA)
Mark Lord8d899e72012-05-02 15:22:52 -04002098 return 0; /* don't retry media errors */
2099 if (qc->flags & ATA_QCFLAG_IO)
2100 return 1; /* otherwise retry anything from fs stack */
2101 if (qc->err_mask & AC_ERR_INVALID)
2102 return 0; /* don't retry these */
2103 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2104}
2105
2106/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002107 * ata_eh_link_autopsy - analyze error and determine recovery action
2108 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09002109 *
Tejun Heo02607312007-08-06 18:36:23 +09002110 * Analyze why @link failed and determine which recovery actions
2111 * are needed. This function also sets more detailed AC_ERR_*
2112 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09002113 *
2114 * LOCKING:
2115 * Kernel thread context (may sleep).
2116 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002117static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002118{
Tejun Heo02607312007-08-06 18:36:23 +09002119 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09002120 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09002121 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002122 unsigned int all_err_mask = 0, eflags = 0;
2123 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09002124 u32 serror;
2125 int rc;
2126
2127 DPRINTK("ENTER\n");
2128
Tejun Heo1cdaf532006-07-03 16:07:26 +09002129 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2130 return;
2131
Tejun Heo022bdb02006-05-15 20:58:22 +09002132 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09002133 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09002134 if (rc == 0) {
2135 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09002136 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09002137 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09002138 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09002139 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09002140 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09002141 ehc->i.err_mask |= AC_ERR_OTHER;
2142 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002143
Tejun Heoe8ee8452006-05-15 21:03:46 +09002144 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09002145 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09002146
Tejun Heo022bdb02006-05-15 20:58:22 +09002147 /* any real error trumps AC_ERR_OTHER */
2148 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2149 ehc->i.err_mask &= ~AC_ERR_OTHER;
2150
2151 all_err_mask |= ehc->i.err_mask;
2152
2153 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2154 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2155
Tejun Heob1c72912008-07-31 17:02:43 +09002156 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2157 ata_dev_phys_link(qc->dev) != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002158 continue;
2159
2160 /* inherit upper level err_mask */
2161 qc->err_mask |= ehc->i.err_mask;
2162
Tejun Heo022bdb02006-05-15 20:58:22 +09002163 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002164 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002165
2166 /* DEV errors are probably spurious in case of ATA_BUS error */
2167 if (qc->err_mask & AC_ERR_ATA_BUS)
2168 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2169 AC_ERR_INVALID);
2170
2171 /* any real error trumps unknown error */
2172 if (qc->err_mask & ~AC_ERR_OTHER)
2173 qc->err_mask &= ~AC_ERR_OTHER;
2174
2175 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09002176 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09002177 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09002178
Tejun Heo03faab72008-03-27 19:14:24 +09002179 /* determine whether the command is worth retrying */
Mark Lord8d899e72012-05-02 15:22:52 -04002180 if (ata_eh_worth_retry(qc))
Tejun Heo03faab72008-03-27 19:14:24 +09002181 qc->flags |= ATA_QCFLAG_RETRY;
2182
Tejun Heo022bdb02006-05-15 20:58:22 +09002183 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002184 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09002185 all_err_mask |= qc->err_mask;
2186 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09002187 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09002188 }
2189
Tejun Heoa20f33f2006-05-16 12:58:24 +09002190 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09002191 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09002192 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09002193 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002194 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2195 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09002196 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002197
Tejun Heodfcc1732007-10-31 10:17:05 +09002198 /* If we have offending qcs and the associated failed device,
2199 * perform per-dev EH action only on the offending device.
2200 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002201 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09002202 ehc->i.dev_action[ehc->i.dev->devno] |=
2203 ehc->i.action & ATA_EH_PERDEV_MASK;
2204 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09002205 }
2206
Tejun Heo2695e362008-01-10 13:41:23 +09002207 /* propagate timeout to host link */
2208 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2209 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2210
2211 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09002212 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09002213 if (!dev && ((ata_link_max_devices(link) == 1 &&
2214 ata_dev_enabled(link->device))))
2215 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09002216
Tejun Heo76326ac2007-11-27 19:28:59 +09002217 if (dev) {
2218 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2219 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002220 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09002221 }
Tejun Heodfcc1732007-10-31 10:17:05 +09002222
Tejun Heo022bdb02006-05-15 20:58:22 +09002223 DPRINTK("EXIT\n");
2224}
2225
2226/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002227 * ata_eh_autopsy - analyze error and determine recovery action
2228 * @ap: host port to perform autopsy on
2229 *
2230 * Analyze all links of @ap and determine why they failed and
2231 * which recovery actions are needed.
2232 *
2233 * LOCKING:
2234 * Kernel thread context (may sleep).
2235 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002236void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002237{
2238 struct ata_link *link;
2239
Tejun Heo1eca4362008-11-03 20:03:17 +09002240 ata_for_each_link(link, ap, EDGE)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002241 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09002242
Tejun Heob1c72912008-07-31 17:02:43 +09002243 /* Handle the frigging slave link. Autopsy is done similarly
2244 * but actions and flags are transferred over to the master
2245 * link and handled from there.
2246 */
2247 if (ap->slave_link) {
2248 struct ata_eh_context *mehc = &ap->link.eh_context;
2249 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2250
Tejun Heo848e4c62008-10-21 14:26:39 +09002251 /* transfer control flags from master to slave */
2252 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2253
2254 /* perform autopsy on the slave link */
Tejun Heob1c72912008-07-31 17:02:43 +09002255 ata_eh_link_autopsy(ap->slave_link);
2256
Tejun Heo848e4c62008-10-21 14:26:39 +09002257 /* transfer actions from slave to master and clear slave */
Tejun Heob1c72912008-07-31 17:02:43 +09002258 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2259 mehc->i.action |= sehc->i.action;
2260 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2261 mehc->i.flags |= sehc->i.flags;
2262 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2263 }
2264
Tejun Heo2695e362008-01-10 13:41:23 +09002265 /* Autopsy of fanout ports can affect host link autopsy.
2266 * Perform host link autopsy last.
2267 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002268 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09002269 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09002270}
2271
2272/**
Robert Hancock65211482009-07-14 20:43:39 -06002273 * ata_get_cmd_descript - get description for ATA command
2274 * @command: ATA command code to get description for
2275 *
2276 * Return a textual description of the given command, or NULL if the
2277 * command is not known.
2278 *
2279 * LOCKING:
2280 * None
2281 */
2282const char *ata_get_cmd_descript(u8 command)
2283{
2284#ifdef CONFIG_ATA_VERBOSE_ERROR
2285 static const struct
2286 {
2287 u8 command;
2288 const char *text;
2289 } cmd_descr[] = {
2290 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2291 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2292 { ATA_CMD_STANDBY, "STANDBY" },
2293 { ATA_CMD_IDLE, "IDLE" },
2294 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2295 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2296 { ATA_CMD_NOP, "NOP" },
2297 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2298 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2299 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2300 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2301 { ATA_CMD_SERVICE, "SERVICE" },
2302 { ATA_CMD_READ, "READ DMA" },
2303 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2304 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2305 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2306 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2307 { ATA_CMD_WRITE, "WRITE DMA" },
2308 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2309 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2310 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2311 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2312 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2313 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2314 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2315 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2316 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2317 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2318 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2319 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2320 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2321 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2322 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2323 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2324 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2325 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2326 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2327 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2328 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2329 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2330 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2331 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2332 { ATA_CMD_SLEEP, "SLEEP" },
2333 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2334 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2335 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2336 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2337 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2338 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2339 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2340 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2341 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2342 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2343 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2344 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2345 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2346 { ATA_CMD_PMP_READ, "READ BUFFER" },
2347 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2348 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2349 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2350 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2351 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2352 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2353 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2354 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2355 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2356 { ATA_CMD_SMART, "SMART" },
2357 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2358 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
FUJITA Tomonoriacad762722010-07-05 15:45:20 +09002359 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
Robert Hancock65211482009-07-14 20:43:39 -06002360 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2361 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2362 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2363 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2364 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2365 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2366 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2367 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2368 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2369 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2370 { ATA_CMD_RESTORE, "RECALIBRATE" },
2371 { 0, NULL } /* terminate list */
2372 };
2373
2374 unsigned int i;
2375 for (i = 0; cmd_descr[i].text; i++)
2376 if (cmd_descr[i].command == command)
2377 return cmd_descr[i].text;
2378#endif
2379
2380 return NULL;
2381}
2382
2383/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002384 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09002385 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09002386 *
2387 * Report EH to user.
2388 *
2389 * LOCKING:
2390 * None.
2391 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002392static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002393{
Tejun Heo02607312007-08-06 18:36:23 +09002394 struct ata_port *ap = link->ap;
2395 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002396 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09002397 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09002398 int tag, nr_failed = 0;
2399
Tejun Heo94ff3d52007-10-09 14:57:56 +09002400 if (ehc->i.flags & ATA_EHI_QUIET)
2401 return;
2402
Tejun Heo022bdb02006-05-15 20:58:22 +09002403 desc = NULL;
2404 if (ehc->i.desc[0] != '\0')
2405 desc = ehc->i.desc;
2406
2407 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2408 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2409
Tejun Heob1c72912008-07-31 17:02:43 +09002410 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2411 ata_dev_phys_link(qc->dev) != link ||
Tejun Heoe027bd32007-10-26 16:19:26 +09002412 ((qc->flags & ATA_QCFLAG_QUIET) &&
2413 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002414 continue;
2415 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2416 continue;
2417
2418 nr_failed++;
2419 }
2420
2421 if (!nr_failed && !ehc->i.err_mask)
2422 return;
2423
2424 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09002425 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09002426 frozen = " frozen";
2427
Tejun Heoa1e10f72007-08-18 13:28:49 +09002428 memset(tries_buf, 0, sizeof(tries_buf));
2429 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2430 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2431 ap->eh_tries);
2432
Tejun Heo022bdb02006-05-15 20:58:22 +09002433 if (ehc->i.dev) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002434 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2435 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2436 ehc->i.err_mask, link->sactive, ehc->i.serror,
2437 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002438 if (desc)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002439 ata_dev_err(ehc->i.dev, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002440 } else {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002441 ata_link_err(link, "exception Emask 0x%x "
2442 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2443 ehc->i.err_mask, link->sactive, ehc->i.serror,
2444 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002445 if (desc)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002446 ata_link_err(link, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002447 }
2448
Robert Hancock65211482009-07-14 20:43:39 -06002449#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002450 if (ehc->i.serror)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002451 ata_link_err(link,
Robert Hancock1333e192007-10-02 11:22:02 -04002452 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2453 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2454 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2455 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2456 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2457 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2458 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2459 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2460 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2461 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2462 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2463 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2464 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2465 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2466 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2467 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2468 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002469 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002470#endif
Robert Hancock1333e192007-10-02 11:22:02 -04002471
Tejun Heo022bdb02006-05-15 20:58:22 +09002472 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2473 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09002474 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09002475 const u8 *cdb = qc->cdb;
2476 char data_buf[20] = "";
2477 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002478
Tejun Heo02607312007-08-06 18:36:23 +09002479 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
Tejun Heob1c72912008-07-31 17:02:43 +09002480 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002481 continue;
2482
Tejun Heoabb6a882007-11-28 23:16:09 +09002483 if (qc->dma_dir != DMA_NONE) {
2484 static const char *dma_str[] = {
2485 [DMA_BIDIRECTIONAL] = "bidi",
2486 [DMA_TO_DEVICE] = "out",
2487 [DMA_FROM_DEVICE] = "in",
2488 };
2489 static const char *prot_str[] = {
2490 [ATA_PROT_PIO] = "pio",
2491 [ATA_PROT_DMA] = "dma",
2492 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05002493 [ATAPI_PROT_PIO] = "pio",
2494 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09002495 };
2496
2497 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2498 prot_str[qc->tf.protocol], qc->nbytes,
2499 dma_str[qc->dma_dir]);
2500 }
2501
Robert Hancock65211482009-07-14 20:43:39 -06002502 if (ata_is_atapi(qc->tf.protocol)) {
2503 if (qc->scsicmd)
2504 scsi_print_command(qc->scsicmd);
2505 else
2506 snprintf(cdb_buf, sizeof(cdb_buf),
Tejun Heoabb6a882007-11-28 23:16:09 +09002507 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2508 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2509 cdb[0], cdb[1], cdb[2], cdb[3],
2510 cdb[4], cdb[5], cdb[6], cdb[7],
2511 cdb[8], cdb[9], cdb[10], cdb[11],
2512 cdb[12], cdb[13], cdb[14], cdb[15]);
Robert Hancock65211482009-07-14 20:43:39 -06002513 } else {
2514 const char *descr = ata_get_cmd_descript(cmd->command);
2515 if (descr)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002516 ata_dev_err(qc->dev, "failed command: %s\n",
2517 descr);
Robert Hancock65211482009-07-14 20:43:39 -06002518 }
Tejun Heoabb6a882007-11-28 23:16:09 +09002519
Joe Perchesa9a79df2011-04-15 15:51:59 -07002520 ata_dev_err(qc->dev,
Tejun Heo8a937582006-11-14 22:36:12 +09002521 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002522 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002523 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002524 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002525 cmd->command, cmd->feature, cmd->nsect,
2526 cmd->lbal, cmd->lbam, cmd->lbah,
2527 cmd->hob_feature, cmd->hob_nsect,
2528 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002529 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002530 res->command, res->feature, res->nsect,
2531 res->lbal, res->lbam, res->lbah,
2532 res->hob_feature, res->hob_nsect,
2533 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002534 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2535 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002536
Robert Hancock65211482009-07-14 20:43:39 -06002537#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002538 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002539 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002540 if (res->command & ATA_BUSY)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002541 ata_dev_err(qc->dev, "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002542 else
Joe Perchesa9a79df2011-04-15 15:51:59 -07002543 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
Robert Hancock1333e192007-10-02 11:22:02 -04002544 res->command & ATA_DRDY ? "DRDY " : "",
2545 res->command & ATA_DF ? "DF " : "",
2546 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002547 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002548 }
2549
2550 if (cmd->command != ATA_CMD_PACKET &&
2551 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2552 ATA_ABORTED)))
Joe Perchesa9a79df2011-04-15 15:51:59 -07002553 ata_dev_err(qc->dev, "error: { %s%s%s%s}\n",
Robert Hancock1333e192007-10-02 11:22:02 -04002554 res->feature & ATA_ICRC ? "ICRC " : "",
2555 res->feature & ATA_UNC ? "UNC " : "",
2556 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002557 res->feature & ATA_ABORTED ? "ABRT " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002558#endif
Tejun Heo022bdb02006-05-15 20:58:22 +09002559 }
2560}
2561
Tejun Heo9b1e2652007-08-06 18:36:24 +09002562/**
2563 * ata_eh_report - report error handling to user
2564 * @ap: ATA port to report EH about
2565 *
2566 * Report EH to user.
2567 *
2568 * LOCKING:
2569 * None.
2570 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002571void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002572{
2573 struct ata_link *link;
2574
Tejun Heo1eca4362008-11-03 20:03:17 +09002575 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002576 ata_eh_link_report(link);
2577}
2578
Tejun Heocc0680a2007-08-06 18:36:23 +09002579static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heob1c72912008-07-31 17:02:43 +09002580 unsigned int *classes, unsigned long deadline,
2581 bool clear_classes)
Tejun Heod87fa382006-05-31 18:28:24 +09002582{
Tejun Heof58229f2007-08-06 18:36:23 +09002583 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002584
Tejun Heob1c72912008-07-31 17:02:43 +09002585 if (clear_classes)
Tejun Heo1eca4362008-11-03 20:03:17 +09002586 ata_for_each_dev(dev, link, ALL)
Tejun Heob1c72912008-07-31 17:02:43 +09002587 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002588
Tejun Heof0465192008-05-19 01:15:08 +09002589 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002590}
2591
Sergei Shtylyove8411fb2011-08-09 23:25:50 +04002592static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
Tejun Heo664faf02006-05-31 18:27:50 +09002593{
Tejun Heo45db2f62008-04-08 01:46:56 +09002594 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002595 return 0;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002596 if (rc == -EAGAIN)
2597 return 1;
Tejun Heo071f44b2008-04-07 22:47:22 +09002598 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002599 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002600 return 0;
2601}
2602
Tejun Heofb7fd612007-09-23 13:14:12 +09002603int ata_eh_reset(struct ata_link *link, int classify,
2604 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2605 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002606{
Tejun Heoafaa5c32007-10-09 15:06:10 +09002607 struct ata_port *ap = link->ap;
Tejun Heob1c72912008-07-31 17:02:43 +09002608 struct ata_link *slave = ap->slave_link;
Tejun Heo936fd732007-08-06 18:36:23 +09002609 struct ata_eh_context *ehc = &link->eh_context;
Bartlomiej Zolnierkiewicz705d2012009-07-26 16:21:01 +02002610 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
Tejun Heo664faf02006-05-31 18:27:50 +09002611 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002612 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002613 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heod8af0eb2008-05-20 02:17:53 +09002614 int max_tries = 0, try = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002615 struct ata_link *failed_link;
Tejun Heof58229f2007-08-06 18:36:23 +09002616 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002617 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002618 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002619 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002620 u32 sstatus;
Tejun Heob1c72912008-07-31 17:02:43 +09002621 int nr_unknown, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002622
Tejun Heo932648b2008-05-19 01:15:06 +09002623 /*
2624 * Prepare to reset
2625 */
Tejun Heod8af0eb2008-05-20 02:17:53 +09002626 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2627 max_tries++;
Dan Williamsca6d43b2012-06-21 23:41:41 -07002628 if (link->flags & ATA_LFLAG_RST_ONCE)
2629 max_tries = 1;
Tejun Heo05944bd2008-08-13 20:19:09 +09002630 if (link->flags & ATA_LFLAG_NO_HRST)
2631 hardreset = NULL;
2632 if (link->flags & ATA_LFLAG_NO_SRST)
2633 softreset = NULL;
Tejun Heod8af0eb2008-05-20 02:17:53 +09002634
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002635 /* make sure each reset attempt is at least COOL_DOWN apart */
Tejun Heo19b72322008-11-04 17:08:40 +09002636 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2637 now = jiffies;
2638 WARN_ON(time_after(ehc->last_reset, now));
2639 deadline = ata_deadline(ehc->last_reset,
2640 ATA_EH_RESET_COOL_DOWN);
2641 if (time_before(now, deadline))
2642 schedule_timeout_uninterruptible(deadline - now);
2643 }
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002644
Tejun Heoafaa5c32007-10-09 15:06:10 +09002645 spin_lock_irqsave(ap->lock, flags);
2646 ap->pflags |= ATA_PFLAG_RESETTING;
2647 spin_unlock_irqrestore(ap->lock, flags);
2648
Tejun Heocf480622008-01-24 00:05:14 +09002649 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002650
Tejun Heo1eca4362008-11-03 20:03:17 +09002651 ata_for_each_dev(dev, link, ALL) {
Tejun Heocdeab112007-10-29 16:41:09 +09002652 /* If we issue an SRST then an ATA drive (not ATAPI)
2653 * may change configuration and be in PIO0 timing. If
2654 * we do a hard reset (or are coming from power on)
2655 * this is true for ATA or ATAPI. Until we've set a
2656 * suitable controller mode we should not touch the
2657 * bus as we may be talking too fast.
2658 */
2659 dev->pio_mode = XFER_PIO_0;
Aaron Lu54169122012-12-03 11:35:02 +08002660 dev->dma_mode = 0xff;
Tejun Heocdeab112007-10-29 16:41:09 +09002661
2662 /* If the controller has a pio mode setup function
2663 * then use it to set the chipset to rights. Don't
2664 * touch the DMA setup as that will be dealt with when
2665 * configuring devices.
2666 */
2667 if (ap->ops->set_piomode)
2668 ap->ops->set_piomode(ap, dev);
2669 }
2670
Tejun Heocf480622008-01-24 00:05:14 +09002671 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002672 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002673 ehc->i.action &= ~ATA_EH_RESET;
2674 if (hardreset) {
2675 reset = hardreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002676 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002677 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002678 reset = softreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002679 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heocf480622008-01-24 00:05:14 +09002680 }
Tejun Heof5914a42006-05-31 18:27:48 +09002681
2682 if (prereset) {
Tejun Heob1c72912008-07-31 17:02:43 +09002683 unsigned long deadline = ata_deadline(jiffies,
2684 ATA_EH_PRERESET_TIMEOUT);
2685
2686 if (slave) {
2687 sehc->i.action &= ~ATA_EH_RESET;
2688 sehc->i.action |= ehc->i.action;
2689 }
2690
2691 rc = prereset(link, deadline);
2692
2693 /* If present, do prereset on slave link too. Reset
2694 * is skipped iff both master and slave links report
2695 * -ENOENT or clear ATA_EH_RESET.
2696 */
2697 if (slave && (rc == 0 || rc == -ENOENT)) {
2698 int tmp;
2699
2700 tmp = prereset(slave, deadline);
2701 if (tmp != -ENOENT)
2702 rc = tmp;
2703
2704 ehc->i.action |= sehc->i.action;
2705 }
2706
Tejun Heof5914a42006-05-31 18:27:48 +09002707 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002708 if (rc == -ENOENT) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002709 ata_link_dbg(link, "port disabled--ignoring\n");
Tejun Heocf480622008-01-24 00:05:14 +09002710 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002711
Tejun Heo1eca4362008-11-03 20:03:17 +09002712 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09002713 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002714
2715 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002716 } else
Joe Perchesa9a79df2011-04-15 15:51:59 -07002717 ata_link_err(link,
2718 "prereset failed (errno=%d)\n",
2719 rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002720 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002721 }
Tejun Heof5914a42006-05-31 18:27:48 +09002722
Tejun Heo932648b2008-05-19 01:15:06 +09002723 /* prereset() might have cleared ATA_EH_RESET. If so,
Tejun Heod6515e62009-03-04 15:59:30 +09002724 * bang classes, thaw and return.
Tejun Heo932648b2008-05-19 01:15:06 +09002725 */
2726 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
Tejun Heo1eca4362008-11-03 20:03:17 +09002727 ata_for_each_dev(dev, link, ALL)
Tejun Heo932648b2008-05-19 01:15:06 +09002728 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heod6515e62009-03-04 15:59:30 +09002729 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2730 ata_is_host_link(link))
2731 ata_eh_thaw_port(ap);
Tejun Heo932648b2008-05-19 01:15:06 +09002732 rc = 0;
2733 goto out;
2734 }
Tejun Heof5914a42006-05-31 18:27:48 +09002735 }
2736
Tejun Heo022bdb02006-05-15 20:58:22 +09002737 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002738 /*
2739 * Perform reset
2740 */
Tejun Heodc98c322008-05-19 01:15:07 +09002741 if (ata_is_host_link(link))
2742 ata_eh_freeze_port(ap);
2743
Tejun Heo341c2c92008-05-20 02:17:51 +09002744 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
Tejun Heo31daabd2007-02-02 16:50:52 +09002745
Tejun Heo932648b2008-05-19 01:15:06 +09002746 if (reset) {
2747 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002748 ata_link_info(link, "%s resetting link\n",
2749 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002750
Tejun Heo932648b2008-05-19 01:15:06 +09002751 /* mark that this EH session started with reset */
Tejun Heo19b72322008-11-04 17:08:40 +09002752 ehc->last_reset = jiffies;
Tejun Heo932648b2008-05-19 01:15:06 +09002753 if (reset == hardreset)
2754 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2755 else
2756 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002757
Tejun Heob1c72912008-07-31 17:02:43 +09002758 rc = ata_do_reset(link, reset, classes, deadline, true);
2759 if (rc && rc != -EAGAIN) {
2760 failed_link = link;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002761 goto fail;
Tejun Heob1c72912008-07-31 17:02:43 +09002762 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002763
Tejun Heob1c72912008-07-31 17:02:43 +09002764 /* hardreset slave link if existent */
2765 if (slave && reset == hardreset) {
2766 int tmp;
2767
2768 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002769 ata_link_info(slave, "hard resetting link\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002770
2771 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2772 tmp = ata_do_reset(slave, reset, classes, deadline,
2773 false);
2774 switch (tmp) {
2775 case -EAGAIN:
2776 rc = -EAGAIN;
2777 case 0:
2778 break;
2779 default:
2780 failed_link = slave;
2781 rc = tmp;
2782 goto fail;
2783 }
2784 }
2785
2786 /* perform follow-up SRST if necessary */
Tejun Heo932648b2008-05-19 01:15:06 +09002787 if (reset == hardreset &&
Sergei Shtylyove8411fb2011-08-09 23:25:50 +04002788 ata_eh_followup_srst_needed(link, rc)) {
Tejun Heo932648b2008-05-19 01:15:06 +09002789 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002790
Tejun Heo932648b2008-05-19 01:15:06 +09002791 if (!reset) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002792 ata_link_err(link,
2793 "follow-up softreset required but no softreset available\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002794 failed_link = link;
Tejun Heo932648b2008-05-19 01:15:06 +09002795 rc = -EINVAL;
2796 goto fail;
2797 }
2798
2799 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002800 rc = ata_do_reset(link, reset, classes, deadline, true);
Tejun Heofe2c4d02009-07-08 12:16:37 +09002801 if (rc) {
2802 failed_link = link;
2803 goto fail;
2804 }
Tejun Heo664faf02006-05-31 18:27:50 +09002805 }
Tejun Heo932648b2008-05-19 01:15:06 +09002806 } else {
2807 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002808 ata_link_info(link,
2809 "no reset method available, skipping reset\n");
Tejun Heo932648b2008-05-19 01:15:06 +09002810 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2811 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002812 }
2813
Tejun Heo932648b2008-05-19 01:15:06 +09002814 /*
2815 * Post-reset processing
2816 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002817 ata_for_each_dev(dev, link, ALL) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002818 /* After the reset, the device state is PIO 0 and the
2819 * controller state is undefined. Reset also wakes up
2820 * drives from sleeping mode.
2821 */
2822 dev->pio_mode = XFER_PIO_0;
2823 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002824
Tejun Heo3b761d32009-10-06 17:08:40 +09002825 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2826 continue;
2827
2828 /* apply class override */
2829 if (lflags & ATA_LFLAG_ASSUME_ATA)
2830 classes[dev->devno] = ATA_DEV_ATA;
2831 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2832 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
Tejun Heo022bdb02006-05-15 20:58:22 +09002833 }
2834
Tejun Heo416dc9e2007-10-31 10:17:03 +09002835 /* record current link speed */
2836 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2837 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heob1c72912008-07-31 17:02:43 +09002838 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2839 slave->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002840
Tejun Heodc98c322008-05-19 01:15:07 +09002841 /* thaw the port */
2842 if (ata_is_host_link(link))
2843 ata_eh_thaw_port(ap);
2844
Tejun Heof0465192008-05-19 01:15:08 +09002845 /* postreset() should clear hardware SError. Although SError
2846 * is cleared during link resume, clearing SError here is
2847 * necessary as some PHYs raise hotplug events after SRST.
2848 * This introduces race condition where hotplug occurs between
2849 * reset and here. This race is mediated by cross checking
2850 * link onlineness and classification result later.
2851 */
Tejun Heob1c72912008-07-31 17:02:43 +09002852 if (postreset) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002853 postreset(link, classes);
Tejun Heob1c72912008-07-31 17:02:43 +09002854 if (slave)
2855 postreset(slave, classes);
2856 }
Tejun Heo20952b62006-05-31 18:27:23 +09002857
Tejun Heo1e641062009-07-17 11:58:33 +09002858 /*
Tejun Heo8c56cac2011-05-25 13:19:39 +02002859 * Some controllers can't be frozen very well and may set spurious
2860 * error conditions during reset. Clear accumulated error
2861 * information and re-thaw the port if frozen. As reset is the
2862 * final recovery action and we cross check link onlineness against
2863 * device classification later, no hotplug event is lost by this.
Tejun Heo1e641062009-07-17 11:58:33 +09002864 */
Tejun Heof0465192008-05-19 01:15:08 +09002865 spin_lock_irqsave(link->ap->lock, flags);
Tejun Heo1e641062009-07-17 11:58:33 +09002866 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heob1c72912008-07-31 17:02:43 +09002867 if (slave)
Tejun Heo1e641062009-07-17 11:58:33 +09002868 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2869 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof0465192008-05-19 01:15:08 +09002870 spin_unlock_irqrestore(link->ap->lock, flags);
2871
Tejun Heo8c56cac2011-05-25 13:19:39 +02002872 if (ap->pflags & ATA_PFLAG_FROZEN)
2873 ata_eh_thaw_port(ap);
2874
Tejun Heo3b761d32009-10-06 17:08:40 +09002875 /*
2876 * Make sure onlineness and classification result correspond.
Tejun Heof0465192008-05-19 01:15:08 +09002877 * Hotplug could have happened during reset and some
2878 * controllers fail to wait while a drive is spinning up after
2879 * being hotplugged causing misdetection. By cross checking
Tejun Heo3b761d32009-10-06 17:08:40 +09002880 * link on/offlineness and classification result, those
2881 * conditions can be reliably detected and retried.
Tejun Heof0465192008-05-19 01:15:08 +09002882 */
Tejun Heob1c72912008-07-31 17:02:43 +09002883 nr_unknown = 0;
Tejun Heo1eca4362008-11-03 20:03:17 +09002884 ata_for_each_dev(dev, link, ALL) {
Tejun Heo3b761d32009-10-06 17:08:40 +09002885 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2886 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002887 ata_dev_dbg(dev, "link online but device misclassified\n");
Tejun Heo3b761d32009-10-06 17:08:40 +09002888 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002889 nr_unknown++;
Tejun Heo3b761d32009-10-06 17:08:40 +09002890 }
2891 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2892 if (ata_class_enabled(classes[dev->devno]))
Joe Perchesa9a79df2011-04-15 15:51:59 -07002893 ata_dev_dbg(dev,
2894 "link offline, clearing class %d to NONE\n",
2895 classes[dev->devno]);
Tejun Heo3b761d32009-10-06 17:08:40 +09002896 classes[dev->devno] = ATA_DEV_NONE;
2897 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002898 ata_dev_dbg(dev,
2899 "link status unknown, clearing UNKNOWN to NONE\n");
Tejun Heo3b761d32009-10-06 17:08:40 +09002900 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002901 }
Tejun Heof0465192008-05-19 01:15:08 +09002902 }
2903
Tejun Heob1c72912008-07-31 17:02:43 +09002904 if (classify && nr_unknown) {
Tejun Heof0465192008-05-19 01:15:08 +09002905 if (try < max_tries) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002906 ata_link_warn(link,
2907 "link online but %d devices misclassified, retrying\n",
2908 nr_unknown);
Tejun Heob1c72912008-07-31 17:02:43 +09002909 failed_link = link;
Tejun Heof0465192008-05-19 01:15:08 +09002910 rc = -EAGAIN;
2911 goto fail;
2912 }
Joe Perchesa9a79df2011-04-15 15:51:59 -07002913 ata_link_warn(link,
2914 "link online but %d devices misclassified, "
2915 "device detection might fail\n", nr_unknown);
Tejun Heof0465192008-05-19 01:15:08 +09002916 }
2917
Tejun Heo416dc9e2007-10-31 10:17:03 +09002918 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002919 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002920 if (slave)
2921 ata_eh_done(slave, NULL, ATA_EH_RESET);
Tejun Heo6b7ae952010-09-01 17:50:06 +02002922 ehc->last_reset = jiffies; /* update to completion time */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002923 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo6b7ae952010-09-01 17:50:06 +02002924 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
Tejun Heoae791c02007-09-23 13:14:12 +09002925
Tejun Heo416dc9e2007-10-31 10:17:03 +09002926 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002927 out:
2928 /* clear hotplug flag */
2929 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heob1c72912008-07-31 17:02:43 +09002930 if (slave)
2931 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002932
2933 spin_lock_irqsave(ap->lock, flags);
2934 ap->pflags &= ~ATA_PFLAG_RESETTING;
2935 spin_unlock_irqrestore(ap->lock, flags);
2936
Tejun Heo022bdb02006-05-15 20:58:22 +09002937 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002938
2939 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002940 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2941 if (!ata_is_host_link(link) &&
2942 sata_scr_read(link, SCR_STATUS, &sstatus))
2943 rc = -ERESTART;
2944
Gwendal Grignou7a46c072011-10-19 17:17:02 -07002945 if (try >= max_tries) {
Tejun Heo8ea76452011-05-25 12:23:56 +02002946 /*
2947 * Thaw host port even if reset failed, so that the port
2948 * can be retried on the next phy event. This risks
2949 * repeated EH runs but seems to be a better tradeoff than
2950 * shutting down a port after a botched hotplug attempt.
2951 */
2952 if (ata_is_host_link(link))
2953 ata_eh_thaw_port(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002954 goto out;
Tejun Heo8ea76452011-05-25 12:23:56 +02002955 }
Tejun Heo416dc9e2007-10-31 10:17:03 +09002956
2957 now = jiffies;
2958 if (time_before(now, deadline)) {
2959 unsigned long delta = deadline - now;
2960
Joe Perchesa9a79df2011-04-15 15:51:59 -07002961 ata_link_warn(failed_link,
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002962 "reset failed (errno=%d), retrying in %u secs\n",
2963 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
Tejun Heo416dc9e2007-10-31 10:17:03 +09002964
Tejun Heoc0c362b2010-09-06 17:57:14 +02002965 ata_eh_release(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002966 while (delta)
2967 delta = schedule_timeout_uninterruptible(delta);
Tejun Heoc0c362b2010-09-06 17:57:14 +02002968 ata_eh_acquire(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002969 }
2970
Gwendal Grignou7a46c072011-10-19 17:17:02 -07002971 /*
2972 * While disks spinup behind PMP, some controllers fail sending SRST.
2973 * They need to be reset - as well as the PMP - before retrying.
2974 */
2975 if (rc == -ERESTART) {
2976 if (ata_is_host_link(link))
2977 ata_eh_thaw_port(ap);
2978 goto out;
2979 }
2980
Tejun Heob1c72912008-07-31 17:02:43 +09002981 if (try == max_tries - 1) {
Tejun Heoa07d4992009-01-29 20:31:33 +09002982 sata_down_spd_limit(link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002983 if (slave)
Tejun Heoa07d4992009-01-29 20:31:33 +09002984 sata_down_spd_limit(slave, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002985 } else if (rc == -EPIPE)
Tejun Heoa07d4992009-01-29 20:31:33 +09002986 sata_down_spd_limit(failed_link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002987
Tejun Heo416dc9e2007-10-31 10:17:03 +09002988 if (hardreset)
2989 reset = hardreset;
2990 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002991}
2992
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002993static inline void ata_eh_pull_park_action(struct ata_port *ap)
2994{
2995 struct ata_link *link;
2996 struct ata_device *dev;
2997 unsigned long flags;
2998
2999 /*
3000 * This function can be thought of as an extended version of
3001 * ata_eh_about_to_do() specially crafted to accommodate the
3002 * requirements of ATA_EH_PARK handling. Since the EH thread
3003 * does not leave the do {} while () loop in ata_eh_recover as
3004 * long as the timeout for a park request to *one* device on
3005 * the port has not expired, and since we still want to pick
3006 * up park requests to other devices on the same port or
3007 * timeout updates for the same device, we have to pull
3008 * ATA_EH_PARK actions from eh_info into eh_context.i
3009 * ourselves at the beginning of each pass over the loop.
3010 *
3011 * Additionally, all write accesses to &ap->park_req_pending
3012 * through INIT_COMPLETION() (see below) or complete_all()
3013 * (see ata_scsi_park_store()) are protected by the host lock.
3014 * As a result we have that park_req_pending.done is zero on
3015 * exit from this function, i.e. when ATA_EH_PARK actions for
3016 * *all* devices on port ap have been pulled into the
3017 * respective eh_context structs. If, and only if,
3018 * park_req_pending.done is non-zero by the time we reach
3019 * wait_for_completion_timeout(), another ATA_EH_PARK action
3020 * has been scheduled for at least one of the devices on port
3021 * ap and we have to cycle over the do {} while () loop in
3022 * ata_eh_recover() again.
3023 */
3024
3025 spin_lock_irqsave(ap->lock, flags);
3026 INIT_COMPLETION(ap->park_req_pending);
Tejun Heo1eca4362008-11-03 20:03:17 +09003027 ata_for_each_link(link, ap, EDGE) {
3028 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003029 struct ata_eh_info *ehi = &link->eh_info;
3030
3031 link->eh_context.i.dev_action[dev->devno] |=
3032 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3033 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3034 }
3035 }
3036 spin_unlock_irqrestore(ap->lock, flags);
3037}
3038
3039static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3040{
3041 struct ata_eh_context *ehc = &dev->link->eh_context;
3042 struct ata_taskfile tf;
3043 unsigned int err_mask;
3044
3045 ata_tf_init(dev, &tf);
3046 if (park) {
3047 ehc->unloaded_mask |= 1 << dev->devno;
3048 tf.command = ATA_CMD_IDLEIMMEDIATE;
3049 tf.feature = 0x44;
3050 tf.lbal = 0x4c;
3051 tf.lbam = 0x4e;
3052 tf.lbah = 0x55;
3053 } else {
3054 ehc->unloaded_mask &= ~(1 << dev->devno);
3055 tf.command = ATA_CMD_CHK_POWER;
3056 }
3057
3058 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3059 tf.protocol |= ATA_PROT_NODATA;
3060 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3061 if (park && (err_mask || tf.lbal != 0xc4)) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003062 ata_dev_err(dev, "head unload failed!\n");
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003063 ehc->unloaded_mask &= ~(1 << dev->devno);
3064 }
3065}
3066
Tejun Heo02607312007-08-06 18:36:23 +09003067static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09003068 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09003069{
Tejun Heo02607312007-08-06 18:36:23 +09003070 struct ata_port *ap = link->ap;
3071 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003072 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003073 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003074 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09003075 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003076
3077 DPRINTK("ENTER\n");
3078
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003079 /* For PATA drive side cable detection to work, IDENTIFY must
3080 * be done backwards such that PDIAG- is released by the slave
3081 * device before the master device is identified.
3082 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003083 ata_for_each_dev(dev, link, ALL_REVERSE) {
Tejun Heof58229f2007-08-06 18:36:23 +09003084 unsigned int action = ata_eh_dev_action(dev);
3085 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09003086
Tejun Heobff04642006-11-10 18:08:10 +09003087 if (ehc->i.flags & ATA_EHI_DID_RESET)
3088 readid_flags |= ATA_READID_POSTRESET;
3089
Tejun Heo9666f402007-05-04 21:27:47 +02003090 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a2007-09-23 13:19:54 +09003091 WARN_ON(dev->class == ATA_DEV_PMP);
3092
Tejun Heob1c72912008-07-31 17:02:43 +09003093 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
Tejun Heo022bdb02006-05-15 20:58:22 +09003094 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003095 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003096 }
3097
Tejun Heo02607312007-08-06 18:36:23 +09003098 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09003099 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3100 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09003101 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003102 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003103
Tejun Heo02607312007-08-06 18:36:23 +09003104 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09003105
Tejun Heobaa1e782006-11-01 18:39:27 +09003106 /* Configuration may have changed, reconfigure
3107 * transfer mode.
3108 */
3109 ehc->i.flags |= ATA_EHI_SETMODE;
3110
zhao, forrest3057ac32006-06-12 12:01:34 +08003111 /* schedule the scsi_rescan_device() here */
Tejun Heoad72cf92010-07-02 10:03:52 +02003112 schedule_work(&(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09003113 } else if (dev->class == ATA_DEV_UNKNOWN &&
3114 ehc->tries[dev->devno] &&
3115 ata_class_enabled(ehc->classes[dev->devno])) {
Tejun Heo842faa62009-05-10 01:06:54 +09003116 /* Temporarily set dev->class, it will be
3117 * permanently set once all configurations are
3118 * complete. This is necessary because new
3119 * device configuration is done in two
3120 * separate loops.
3121 */
Tejun Heo084fe632006-05-31 18:28:03 +09003122 dev->class = ehc->classes[dev->devno];
3123
Tejun Heo633273a2007-09-23 13:19:54 +09003124 if (dev->class == ATA_DEV_PMP)
3125 rc = sata_pmp_attach(dev);
3126 else
3127 rc = ata_dev_read_id(dev, &dev->class,
3128 readid_flags, dev->id);
Tejun Heo842faa62009-05-10 01:06:54 +09003129
3130 /* read_id might have changed class, store and reset */
3131 ehc->classes[dev->devno] = dev->class;
3132 dev->class = ATA_DEV_UNKNOWN;
3133
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003134 switch (rc) {
3135 case 0:
Tejun Heo99cf6102009-01-29 20:31:32 +09003136 /* clear error info accumulated during probe */
3137 ata_ering_clear(&dev->ering);
Tejun Heof58229f2007-08-06 18:36:23 +09003138 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003139 break;
3140 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003141 /* IDENTIFY was issued to non-existent
3142 * device. No need to reset. Just
Tejun Heo842faa62009-05-10 01:06:54 +09003143 * thaw and ignore the device.
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003144 */
3145 ata_eh_thaw_port(ap);
Tejun Heo084fe632006-05-31 18:28:03 +09003146 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003147 default:
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003148 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003149 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003150 }
3151 }
3152
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003153 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09003154 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3155 if (ap->ops->cable_detect)
3156 ap->cbl = ap->ops->cable_detect(ap);
3157 ata_force_cbl(ap);
3158 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003159
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003160 /* Configure new devices forward such that user doesn't see
3161 * device detection messages backwards.
3162 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003163 ata_for_each_dev(dev, link, ALL) {
Tejun Heo4f7c2872009-10-15 23:37:32 +09003164 if (!(new_mask & (1 << dev->devno)))
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003165 continue;
3166
Tejun Heo842faa62009-05-10 01:06:54 +09003167 dev->class = ehc->classes[dev->devno];
3168
Tejun Heo4f7c2872009-10-15 23:37:32 +09003169 if (dev->class == ATA_DEV_PMP)
3170 continue;
3171
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003172 ehc->i.flags |= ATA_EHI_PRINTINFO;
3173 rc = ata_dev_configure(dev);
3174 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
Tejun Heo842faa62009-05-10 01:06:54 +09003175 if (rc) {
3176 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003177 goto err;
Tejun Heo842faa62009-05-10 01:06:54 +09003178 }
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003179
3180 spin_lock_irqsave(ap->lock, flags);
3181 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3182 spin_unlock_irqrestore(ap->lock, flags);
3183
3184 /* new device discovered, configure xfermode */
3185 ehc->i.flags |= ATA_EHI_SETMODE;
3186 }
3187
3188 return 0;
3189
3190 err:
3191 *r_failed_dev = dev;
3192 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003193 return rc;
3194}
3195
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003196/**
3197 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3198 * @link: link on which timings will be programmed
Martin Olsson98a17082009-04-22 18:21:29 +02003199 * @r_failed_dev: out parameter for failed device
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003200 *
3201 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3202 * ata_set_mode() fails, pointer to the failing device is
3203 * returned in @r_failed_dev.
3204 *
3205 * LOCKING:
3206 * PCI/etc. bus probe sem.
3207 *
3208 * RETURNS:
3209 * 0 on success, negative errno otherwise
3210 */
3211int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3212{
3213 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09003214 struct ata_device *dev;
3215 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003216
Tejun Heo76326ac2007-11-27 19:28:59 +09003217 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
Tejun Heo1eca4362008-11-03 20:03:17 +09003218 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo76326ac2007-11-27 19:28:59 +09003219 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3220 struct ata_ering_entry *ent;
3221
3222 ent = ata_ering_top(&dev->ering);
3223 if (ent)
3224 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3225 }
3226 }
3227
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003228 /* has private set_mode? */
3229 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09003230 rc = ap->ops->set_mode(link, r_failed_dev);
3231 else
3232 rc = ata_do_set_mode(link, r_failed_dev);
3233
3234 /* if transfer mode has changed, set DUBIOUS_XFER on device */
Tejun Heo1eca4362008-11-03 20:03:17 +09003235 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +09003236 struct ata_eh_context *ehc = &link->eh_context;
3237 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3238 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3239
3240 if (dev->xfer_mode != saved_xfer_mode ||
3241 ata_ncq_enabled(dev) != saved_ncq)
3242 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3243 }
3244
3245 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003246}
3247
Tejun Heo11fc33d2008-08-30 14:20:01 +02003248/**
3249 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3250 * @dev: ATAPI device to clear UA for
3251 *
3252 * Resets and other operations can make an ATAPI device raise
3253 * UNIT ATTENTION which causes the next operation to fail. This
3254 * function clears UA.
3255 *
3256 * LOCKING:
3257 * EH context (may sleep).
3258 *
3259 * RETURNS:
3260 * 0 on success, -errno on failure.
3261 */
3262static int atapi_eh_clear_ua(struct ata_device *dev)
3263{
3264 int i;
3265
3266 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
Tejun Heob5357082009-03-02 18:55:16 +09003267 u8 *sense_buffer = dev->link->ap->sector_buf;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003268 u8 sense_key = 0;
3269 unsigned int err_mask;
3270
3271 err_mask = atapi_eh_tur(dev, &sense_key);
3272 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003273 ata_dev_warn(dev,
3274 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3275 err_mask);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003276 return -EIO;
3277 }
3278
3279 if (!err_mask || sense_key != UNIT_ATTENTION)
3280 return 0;
3281
3282 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3283 if (err_mask) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003284 ata_dev_warn(dev, "failed to clear "
Tejun Heo11fc33d2008-08-30 14:20:01 +02003285 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3286 return -EIO;
3287 }
3288 }
3289
Joe Perchesa9a79df2011-04-15 15:51:59 -07003290 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3291 ATA_EH_UA_TRIES);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003292
3293 return 0;
3294}
3295
Tejun Heo6013efd2009-11-19 15:36:45 +09003296/**
3297 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3298 * @dev: ATA device which may need FLUSH retry
3299 *
3300 * If @dev failed FLUSH, it needs to be reported upper layer
3301 * immediately as it means that @dev failed to remap and already
3302 * lost at least a sector and further FLUSH retrials won't make
3303 * any difference to the lost sector. However, if FLUSH failed
3304 * for other reasons, for example transmission error, FLUSH needs
3305 * to be retried.
3306 *
3307 * This function determines whether FLUSH failure retry is
3308 * necessary and performs it if so.
3309 *
3310 * RETURNS:
3311 * 0 if EH can continue, -errno if EH needs to be repeated.
3312 */
3313static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3314{
3315 struct ata_link *link = dev->link;
3316 struct ata_port *ap = link->ap;
3317 struct ata_queued_cmd *qc;
3318 struct ata_taskfile tf;
3319 unsigned int err_mask;
3320 int rc = 0;
3321
3322 /* did flush fail for this device? */
3323 if (!ata_tag_valid(link->active_tag))
3324 return 0;
3325
3326 qc = __ata_qc_from_tag(ap, link->active_tag);
3327 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3328 qc->tf.command != ATA_CMD_FLUSH))
3329 return 0;
3330
3331 /* if the device failed it, it should be reported to upper layers */
3332 if (qc->err_mask & AC_ERR_DEV)
3333 return 0;
3334
3335 /* flush failed for some other reason, give it another shot */
3336 ata_tf_init(dev, &tf);
3337
3338 tf.command = qc->tf.command;
3339 tf.flags |= ATA_TFLAG_DEVICE;
3340 tf.protocol = ATA_PROT_NODATA;
3341
Joe Perchesa9a79df2011-04-15 15:51:59 -07003342 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
Tejun Heo6013efd2009-11-19 15:36:45 +09003343 tf.command, qc->err_mask);
3344
3345 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3346 if (!err_mask) {
3347 /*
3348 * FLUSH is complete but there's no way to
3349 * successfully complete a failed command from EH.
3350 * Making sure retry is allowed at least once and
3351 * retrying it should do the trick - whatever was in
3352 * the cache is already on the platter and this won't
3353 * cause infinite loop.
3354 */
3355 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3356 } else {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003357 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
Tejun Heo6013efd2009-11-19 15:36:45 +09003358 err_mask);
3359 rc = -EIO;
3360
3361 /* if device failed it, report it to upper layers */
3362 if (err_mask & AC_ERR_DEV) {
3363 qc->err_mask |= AC_ERR_DEV;
3364 qc->result_tf = tf;
3365 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3366 rc = 0;
3367 }
3368 }
3369 return rc;
3370}
3371
Tejun Heo6b7ae952010-09-01 17:50:06 +02003372/**
3373 * ata_eh_set_lpm - configure SATA interface power management
3374 * @link: link to configure power management
3375 * @policy: the link power management policy
3376 * @r_failed_dev: out parameter for failed device
3377 *
3378 * Enable SATA Interface power management. This will enable
3379 * Device Interface Power Management (DIPM) for min_power
3380 * policy, and then call driver specific callbacks for
3381 * enabling Host Initiated Power management.
3382 *
3383 * LOCKING:
3384 * EH context.
3385 *
3386 * RETURNS:
3387 * 0 on success, -errno on failure.
3388 */
3389static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3390 struct ata_device **r_failed_dev)
3391{
Tejun Heo6c8ea892010-09-01 17:50:07 +02003392 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003393 struct ata_eh_context *ehc = &link->eh_context;
3394 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
Tejun Heoe5005b12010-12-09 16:13:15 +01003395 enum ata_lpm_policy old_policy = link->lpm_policy;
Tejun Heo5f6f12c2011-05-09 16:04:11 +02003396 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003397 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3398 unsigned int err_mask;
3399 int rc;
3400
3401 /* if the link or host doesn't do LPM, noop */
3402 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3403 return 0;
3404
3405 /*
3406 * DIPM is enabled only for MIN_POWER as some devices
3407 * misbehave when the host NACKs transition to SLUMBER. Order
3408 * device and link configurations such that the host always
3409 * allows DIPM requests.
3410 */
3411 ata_for_each_dev(dev, link, ENABLED) {
3412 bool hipm = ata_id_has_hipm(dev->id);
Tejun Heoae01b242011-03-16 11:14:55 +01003413 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003414
3415 /* find the first enabled and LPM enabled devices */
3416 if (!link_dev)
3417 link_dev = dev;
3418
3419 if (!lpm_dev && (hipm || dipm))
3420 lpm_dev = dev;
3421
3422 hints &= ~ATA_LPM_EMPTY;
3423 if (!hipm)
3424 hints &= ~ATA_LPM_HIPM;
3425
3426 /* disable DIPM before changing link config */
3427 if (policy != ATA_LPM_MIN_POWER && dipm) {
3428 err_mask = ata_dev_set_feature(dev,
3429 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3430 if (err_mask && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003431 ata_dev_warn(dev,
3432 "failed to disable DIPM, Emask 0x%x\n",
3433 err_mask);
Tejun Heo6b7ae952010-09-01 17:50:06 +02003434 rc = -EIO;
3435 goto fail;
3436 }
3437 }
3438 }
3439
Tejun Heo6c8ea892010-09-01 17:50:07 +02003440 if (ap) {
3441 rc = ap->ops->set_lpm(link, policy, hints);
3442 if (!rc && ap->slave_link)
3443 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3444 } else
3445 rc = sata_pmp_set_lpm(link, policy, hints);
Tejun Heo6b7ae952010-09-01 17:50:06 +02003446
3447 /*
3448 * Attribute link config failure to the first (LPM) enabled
3449 * device on the link.
3450 */
3451 if (rc) {
3452 if (rc == -EOPNOTSUPP) {
3453 link->flags |= ATA_LFLAG_NO_LPM;
3454 return 0;
3455 }
3456 dev = lpm_dev ? lpm_dev : link_dev;
3457 goto fail;
3458 }
3459
Tejun Heoe5005b12010-12-09 16:13:15 +01003460 /*
3461 * Low level driver acked the transition. Issue DIPM command
3462 * with the new policy set.
3463 */
3464 link->lpm_policy = policy;
3465 if (ap && ap->slave_link)
3466 ap->slave_link->lpm_policy = policy;
3467
Tejun Heo6b7ae952010-09-01 17:50:06 +02003468 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3469 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heoae01b242011-03-16 11:14:55 +01003470 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3471 ata_id_has_dipm(dev->id)) {
Tejun Heo6b7ae952010-09-01 17:50:06 +02003472 err_mask = ata_dev_set_feature(dev,
3473 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3474 if (err_mask && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003475 ata_dev_warn(dev,
Tejun Heo6b7ae952010-09-01 17:50:06 +02003476 "failed to enable DIPM, Emask 0x%x\n",
3477 err_mask);
3478 rc = -EIO;
3479 goto fail;
3480 }
3481 }
3482 }
3483
Tejun Heo6b7ae952010-09-01 17:50:06 +02003484 return 0;
3485
3486fail:
Tejun Heoe5005b12010-12-09 16:13:15 +01003487 /* restore the old policy */
3488 link->lpm_policy = old_policy;
3489 if (ap && ap->slave_link)
3490 ap->slave_link->lpm_policy = old_policy;
3491
Tejun Heo6b7ae952010-09-01 17:50:06 +02003492 /* if no device or only one more chance is left, disable LPM */
3493 if (!dev || ehc->tries[dev->devno] <= 2) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003494 ata_link_warn(link, "disabling LPM on the link\n");
Tejun Heo6b7ae952010-09-01 17:50:06 +02003495 link->flags |= ATA_LFLAG_NO_LPM;
3496 }
3497 if (r_failed_dev)
3498 *r_failed_dev = dev;
3499 return rc;
3500}
3501
Kristen Carlson Accardi8a745f12011-03-04 10:24:11 -08003502int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003503{
Tejun Heof58229f2007-08-06 18:36:23 +09003504 struct ata_device *dev;
3505 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003506
Tejun Heo1eca4362008-11-03 20:03:17 +09003507 ata_for_each_dev(dev, link, ENABLED)
3508 cnt++;
Tejun Heo022bdb02006-05-15 20:58:22 +09003509 return cnt;
3510}
3511
Tejun Heo02607312007-08-06 18:36:23 +09003512static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003513{
Tejun Heof58229f2007-08-06 18:36:23 +09003514 struct ata_device *dev;
3515 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003516
Tejun Heo1eca4362008-11-03 20:03:17 +09003517 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09003518 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09003519 cnt++;
3520 return cnt;
3521}
3522
Tejun Heo02607312007-08-06 18:36:23 +09003523static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003524{
Tejun Heo672b2d62008-01-24 00:05:14 +09003525 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09003526 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09003527 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09003528
Tejun Heof9df58c2007-09-23 13:14:13 +09003529 /* skip disabled links */
3530 if (link->flags & ATA_LFLAG_DISABLED)
3531 return 1;
3532
Tejun Heoe2f3d752010-09-07 14:05:31 +02003533 /* skip if explicitly requested */
3534 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3535 return 1;
3536
Tejun Heo672b2d62008-01-24 00:05:14 +09003537 /* thaw frozen port and recover failed devices */
3538 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3539 return 0;
3540
3541 /* reset at least once if reset is requested */
3542 if ((ehc->i.action & ATA_EH_RESET) &&
3543 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09003544 return 0;
3545
3546 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo1eca4362008-11-03 20:03:17 +09003547 ata_for_each_dev(dev, link, ALL) {
Tejun Heo084fe632006-05-31 18:28:03 +09003548 if (dev->class == ATA_DEV_UNKNOWN &&
3549 ehc->classes[dev->devno] != ATA_DEV_NONE)
3550 return 0;
3551 }
3552
3553 return 1;
3554}
3555
Tejun Heoc2c7a892009-01-29 20:31:34 +09003556static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3557{
3558 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3559 u64 now = get_jiffies_64();
3560 int *trials = void_arg;
3561
Lin Ming68682252012-05-03 22:15:07 +08003562 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3563 (ent->timestamp < now - min(now, interval)))
Tejun Heoc2c7a892009-01-29 20:31:34 +09003564 return -1;
3565
3566 (*trials)++;
3567 return 0;
3568}
3569
Tejun Heo02c05a22007-11-27 19:28:54 +09003570static int ata_eh_schedule_probe(struct ata_device *dev)
3571{
3572 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heoc2c7a892009-01-29 20:31:34 +09003573 struct ata_link *link = ata_dev_phys_link(dev);
3574 int trials = 0;
Tejun Heo02c05a22007-11-27 19:28:54 +09003575
3576 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3577 (ehc->did_probe_mask & (1 << dev->devno)))
3578 return 0;
3579
3580 ata_eh_detach_dev(dev);
3581 ata_dev_init(dev);
3582 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09003583 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09003584 ehc->saved_xfer_mode[dev->devno] = 0;
3585 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09003586
Tejun Heo6b7ae952010-09-01 17:50:06 +02003587 /* the link maybe in a deep sleep, wake it up */
Tejun Heo6c8ea892010-09-01 17:50:07 +02003588 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3589 if (ata_is_host_link(link))
3590 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3591 ATA_LPM_EMPTY);
3592 else
3593 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3594 ATA_LPM_EMPTY);
3595 }
Tejun Heo6b7ae952010-09-01 17:50:06 +02003596
Tejun Heoc2c7a892009-01-29 20:31:34 +09003597 /* Record and count probe trials on the ering. The specific
3598 * error mask used is irrelevant. Because a successful device
3599 * detection clears the ering, this count accumulates only if
3600 * there are consecutive failed probes.
3601 *
3602 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3603 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3604 * forced to 1.5Gbps.
3605 *
3606 * This is to work around cases where failed link speed
3607 * negotiation results in device misdetection leading to
3608 * infinite DEVXCHG or PHRDY CHG events.
3609 */
3610 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3611 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3612
3613 if (trials > ATA_EH_PROBE_TRIALS)
3614 sata_down_spd_limit(link, 1);
3615
Tejun Heo02c05a22007-11-27 19:28:54 +09003616 return 1;
3617}
3618
Tejun Heo9b1e2652007-08-06 18:36:24 +09003619static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09003620{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09003621 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09003622
Tejun Heocf9a5902009-01-29 20:31:35 +09003623 /* -EAGAIN from EH routine indicates retry without prejudice.
3624 * The requester is responsible for ensuring forward progress.
3625 */
3626 if (err != -EAGAIN)
3627 ehc->tries[dev->devno]--;
Tejun Heofee7ca72007-07-01 19:05:58 +09003628
3629 switch (err) {
3630 case -ENODEV:
3631 /* device missing or wrong IDENTIFY data, schedule probing */
3632 ehc->i.probe_mask |= (1 << dev->devno);
3633 case -EINVAL:
3634 /* give it just one more chance */
3635 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3636 case -EIO:
Tejun Heod89293a2009-01-29 20:31:29 +09003637 if (ehc->tries[dev->devno] == 1) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003638 /* This is the last chance, better to slow
3639 * down than lose it.
3640 */
Tejun Heoa07d4992009-01-29 20:31:33 +09003641 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
Tejun Heod89293a2009-01-29 20:31:29 +09003642 if (dev->pio_mode > XFER_PIO_0)
3643 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
Tejun Heofee7ca72007-07-01 19:05:58 +09003644 }
3645 }
3646
3647 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3648 /* disable device if it has used up all its chances */
3649 ata_dev_disable(dev);
3650
3651 /* detach if offline */
Tejun Heob1c72912008-07-31 17:02:43 +09003652 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
Tejun Heofee7ca72007-07-01 19:05:58 +09003653 ata_eh_detach_dev(dev);
3654
Tejun Heo02c05a22007-11-27 19:28:54 +09003655 /* schedule probe if necessary */
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003656 if (ata_eh_schedule_probe(dev)) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003657 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003658 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3659 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3660 }
Tejun Heo9b1e2652007-08-06 18:36:24 +09003661
3662 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09003663 } else {
Tejun Heocf480622008-01-24 00:05:14 +09003664 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003665 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09003666 }
3667}
3668
Tejun Heo022bdb02006-05-15 20:58:22 +09003669/**
3670 * ata_eh_recover - recover host port after error
3671 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09003672 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003673 * @softreset: softreset method (can be NULL)
3674 * @hardreset: hardreset method (can be NULL)
3675 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003676 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09003677 *
3678 * This is the alpha and omega, eum and yang, heart and soul of
3679 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09003680 * recover each link and hotplug requests are recorded in the
3681 * link's eh_context. This function executes all the operations
3682 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09003683 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09003684 *
3685 * LOCKING:
3686 * Kernel thread context (may sleep).
3687 *
3688 * RETURNS:
3689 * 0 on success, -errno on failure.
3690 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003691int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3692 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3693 ata_postreset_fn_t postreset,
3694 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003695{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003696 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003697 struct ata_device *dev;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003698 int rc, nr_fails;
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003699 unsigned long flags, deadline;
Tejun Heo022bdb02006-05-15 20:58:22 +09003700
3701 DPRINTK("ENTER\n");
3702
3703 /* prep for recovery */
Tejun Heo1eca4362008-11-03 20:03:17 +09003704 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003705 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09003706
Tejun Heof9df58c2007-09-23 13:14:13 +09003707 /* re-enable link? */
3708 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3709 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3710 spin_lock_irqsave(ap->lock, flags);
3711 link->flags &= ~ATA_LFLAG_DISABLED;
3712 spin_unlock_irqrestore(ap->lock, flags);
3713 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3714 }
3715
Tejun Heo1eca4362008-11-03 20:03:17 +09003716 ata_for_each_dev(dev, link, ALL) {
Tejun Heofd995f72007-09-23 13:14:12 +09003717 if (link->flags & ATA_LFLAG_NO_RETRY)
3718 ehc->tries[dev->devno] = 1;
3719 else
3720 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09003721
Tejun Heo9b1e2652007-08-06 18:36:24 +09003722 /* collect port action mask recorded in dev actions */
3723 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3724 ~ATA_EH_PERDEV_MASK;
3725 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09003726
Tejun Heo9b1e2652007-08-06 18:36:24 +09003727 /* process hotplug request */
3728 if (dev->flags & ATA_DFLAG_DETACH)
3729 ata_eh_detach_dev(dev);
3730
Tejun Heo02c05a22007-11-27 19:28:54 +09003731 /* schedule probe if necessary */
3732 if (!ata_dev_enabled(dev))
3733 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09003734 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003735 }
3736
3737 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09003738 rc = 0;
3739
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003740 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09003741 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003742 goto out;
3743
Tejun Heo9b1e2652007-08-06 18:36:24 +09003744 /* prep for EH */
Tejun Heo1eca4362008-11-03 20:03:17 +09003745 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003746 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003747
Tejun Heo9b1e2652007-08-06 18:36:24 +09003748 /* skip EH if possible. */
3749 if (ata_eh_skip_recovery(link))
3750 ehc->i.action = 0;
3751
Tejun Heo1eca4362008-11-03 20:03:17 +09003752 ata_for_each_dev(dev, link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003753 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3754 }
Tejun Heo084fe632006-05-31 18:28:03 +09003755
Tejun Heo022bdb02006-05-15 20:58:22 +09003756 /* reset */
Tejun Heo1eca4362008-11-03 20:03:17 +09003757 ata_for_each_link(link, ap, EDGE) {
Tejun Heodc98c322008-05-19 01:15:07 +09003758 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003759
Tejun Heodc98c322008-05-19 01:15:07 +09003760 if (!(ehc->i.action & ATA_EH_RESET))
3761 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003762
Tejun Heodc98c322008-05-19 01:15:07 +09003763 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3764 prereset, softreset, hardreset, postreset);
3765 if (rc) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003766 ata_link_err(link, "reset failed, giving up\n");
Tejun Heodc98c322008-05-19 01:15:07 +09003767 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09003768 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003769 }
3770
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003771 do {
3772 unsigned long now;
3773
3774 /*
3775 * clears ATA_EH_PARK in eh_info and resets
3776 * ap->park_req_pending
3777 */
3778 ata_eh_pull_park_action(ap);
3779
3780 deadline = jiffies;
Tejun Heo1eca4362008-11-03 20:03:17 +09003781 ata_for_each_link(link, ap, EDGE) {
3782 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003783 struct ata_eh_context *ehc = &link->eh_context;
3784 unsigned long tmp;
3785
3786 if (dev->class != ATA_DEV_ATA)
3787 continue;
3788 if (!(ehc->i.dev_action[dev->devno] &
3789 ATA_EH_PARK))
3790 continue;
3791 tmp = dev->unpark_deadline;
3792 if (time_before(deadline, tmp))
3793 deadline = tmp;
3794 else if (time_before_eq(tmp, jiffies))
3795 continue;
3796 if (ehc->unloaded_mask & (1 << dev->devno))
3797 continue;
3798
3799 ata_eh_park_issue_cmd(dev, 1);
3800 }
3801 }
3802
3803 now = jiffies;
3804 if (time_before_eq(deadline, now))
3805 break;
3806
Tejun Heoc0c362b2010-09-06 17:57:14 +02003807 ata_eh_release(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003808 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3809 deadline - now);
Tejun Heoc0c362b2010-09-06 17:57:14 +02003810 ata_eh_acquire(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003811 } while (deadline);
Tejun Heo1eca4362008-11-03 20:03:17 +09003812 ata_for_each_link(link, ap, EDGE) {
3813 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003814 if (!(link->eh_context.unloaded_mask &
3815 (1 << dev->devno)))
3816 continue;
3817
3818 ata_eh_park_issue_cmd(dev, 0);
3819 ata_eh_done(link, dev, ATA_EH_PARK);
3820 }
3821 }
3822
Tejun Heo9b1e2652007-08-06 18:36:24 +09003823 /* the rest */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003824 nr_fails = 0;
3825 ata_for_each_link(link, ap, PMP_FIRST) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003826 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003827
Tejun Heo6b7ae952010-09-01 17:50:06 +02003828 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3829 goto config_lpm;
3830
Tejun Heo9b1e2652007-08-06 18:36:24 +09003831 /* revalidate existing devices and attach new ones */
3832 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09003833 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003834 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003835
Tejun Heo633273a2007-09-23 13:19:54 +09003836 /* if PMP got attached, return, pmp EH will take care of it */
3837 if (link->device->class == ATA_DEV_PMP) {
3838 ehc->i.action = 0;
3839 return 0;
3840 }
3841
Tejun Heo9b1e2652007-08-06 18:36:24 +09003842 /* configure transfer mode if necessary */
3843 if (ehc->i.flags & ATA_EHI_SETMODE) {
3844 rc = ata_set_mode(link, &dev);
3845 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003846 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003847 ehc->i.flags &= ~ATA_EHI_SETMODE;
3848 }
3849
Tejun Heo11fc33d2008-08-30 14:20:01 +02003850 /* If reset has been issued, clear UA to avoid
3851 * disrupting the current users of the device.
3852 */
3853 if (ehc->i.flags & ATA_EHI_DID_RESET) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003854 ata_for_each_dev(dev, link, ALL) {
Tejun Heo11fc33d2008-08-30 14:20:01 +02003855 if (dev->class != ATA_DEV_ATAPI)
3856 continue;
3857 rc = atapi_eh_clear_ua(dev);
3858 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003859 goto rest_fail;
Aaron Lu21334202013-01-15 17:21:01 +08003860 if (zpodd_dev_enabled(dev))
3861 zpodd_post_poweron(dev);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003862 }
3863 }
3864
Tejun Heo6013efd2009-11-19 15:36:45 +09003865 /* retry flush if necessary */
3866 ata_for_each_dev(dev, link, ALL) {
3867 if (dev->class != ATA_DEV_ATA)
3868 continue;
3869 rc = ata_eh_maybe_retry_flush(dev);
3870 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003871 goto rest_fail;
Tejun Heo6013efd2009-11-19 15:36:45 +09003872 }
3873
Tejun Heo6b7ae952010-09-01 17:50:06 +02003874 config_lpm:
Tejun Heo11fc33d2008-08-30 14:20:01 +02003875 /* configure link power saving */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003876 if (link->lpm_policy != ap->target_lpm_policy) {
3877 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3878 if (rc)
3879 goto rest_fail;
3880 }
Kristen Carlson Accardica773292007-10-25 00:58:59 -04003881
Tejun Heo9b1e2652007-08-06 18:36:24 +09003882 /* this link is okay now */
3883 ehc->i.flags = 0;
3884 continue;
3885
Tejun Heo6b7ae952010-09-01 17:50:06 +02003886 rest_fail:
3887 nr_fails++;
3888 if (dev)
3889 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo9b1e2652007-08-06 18:36:24 +09003890
Tejun Heob06ce3e2007-10-09 15:06:48 +09003891 if (ap->pflags & ATA_PFLAG_FROZEN) {
3892 /* PMP reset requires working host port.
3893 * Can't retry if it's frozen.
3894 */
Tejun Heo071f44b2008-04-07 22:47:22 +09003895 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09003896 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003897 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09003898 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003899 }
3900
Tejun Heo6b7ae952010-09-01 17:50:06 +02003901 if (nr_fails)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003902 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003903
Tejun Heo022bdb02006-05-15 20:58:22 +09003904 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003905 if (rc && r_failed_link)
3906 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003907
3908 DPRINTK("EXIT, rc=%d\n", rc);
3909 return rc;
3910}
3911
3912/**
3913 * ata_eh_finish - finish up EH
3914 * @ap: host port to finish EH for
3915 *
3916 * Recovery is complete. Clean up EH states and retry or finish
3917 * failed qcs.
3918 *
3919 * LOCKING:
3920 * None.
3921 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003922void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09003923{
3924 int tag;
3925
3926 /* retry or finish qcs */
3927 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3928 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3929
3930 if (!(qc->flags & ATA_QCFLAG_FAILED))
3931 continue;
3932
3933 if (qc->err_mask) {
3934 /* FIXME: Once EH migration is complete,
3935 * generate sense data in this function,
3936 * considering both err_mask and tf.
3937 */
Tejun Heo03faab72008-03-27 19:14:24 +09003938 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09003939 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09003940 else
3941 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003942 } else {
3943 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3944 ata_eh_qc_complete(qc);
3945 } else {
3946 /* feed zero TF to sense generation */
3947 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3948 ata_eh_qc_retry(qc);
3949 }
3950 }
3951 }
Tejun Heoda917d62007-09-23 13:14:12 +09003952
3953 /* make sure nr_active_links is zero after EH */
3954 WARN_ON(ap->nr_active_links);
3955 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003956}
3957
3958/**
3959 * ata_do_eh - do standard error handling
3960 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09003961 *
Tejun Heof5914a42006-05-31 18:27:48 +09003962 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003963 * @softreset: softreset method (can be NULL)
3964 * @hardreset: hardreset method (can be NULL)
3965 * @postreset: postreset method (can be NULL)
3966 *
3967 * Perform standard error handling sequence.
3968 *
3969 * LOCKING:
3970 * Kernel thread context (may sleep).
3971 */
Tejun Heof5914a42006-05-31 18:27:48 +09003972void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3973 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3974 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09003975{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003976 struct ata_device *dev;
3977 int rc;
3978
3979 ata_eh_autopsy(ap);
3980 ata_eh_report(ap);
3981
3982 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3983 NULL);
3984 if (rc) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003985 ata_for_each_dev(dev, &ap->link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003986 ata_dev_disable(dev);
3987 }
3988
Tejun Heo022bdb02006-05-15 20:58:22 +09003989 ata_eh_finish(ap);
3990}
Tejun Heo500530f2006-07-03 16:07:27 +09003991
Tejun Heoa1efdab2008-03-25 12:22:50 +09003992/**
3993 * ata_std_error_handler - standard error handler
3994 * @ap: host port to handle error for
3995 *
3996 * Standard error handler
3997 *
3998 * LOCKING:
3999 * Kernel thread context (may sleep).
4000 */
4001void ata_std_error_handler(struct ata_port *ap)
4002{
4003 struct ata_port_operations *ops = ap->ops;
4004 ata_reset_fn_t hardreset = ops->hardreset;
4005
Tejun Heo57c9efd2008-04-07 22:47:19 +09004006 /* ignore built-in hardreset if SCR access is not available */
Tejun Heofe06e5f2010-05-10 21:41:39 +02004007 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09004008 hardreset = NULL;
4009
4010 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4011}
4012
Tejun Heo6ffa01d2007-03-02 17:32:47 +09004013#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09004014/**
4015 * ata_eh_handle_port_suspend - perform port suspend operation
4016 * @ap: port to suspend
4017 *
4018 * Suspend @ap.
4019 *
4020 * LOCKING:
4021 * Kernel thread context (may sleep).
4022 */
4023static void ata_eh_handle_port_suspend(struct ata_port *ap)
4024{
4025 unsigned long flags;
4026 int rc = 0;
Aaron Lu3dc67442013-01-15 17:21:00 +08004027 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09004028
4029 /* are we suspending? */
4030 spin_lock_irqsave(ap->lock, flags);
4031 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
Aaron Lua7ff60d2013-01-25 14:29:35 +08004032 ap->pm_mesg.event & PM_EVENT_RESUME) {
Tejun Heo500530f2006-07-03 16:07:27 +09004033 spin_unlock_irqrestore(ap->lock, flags);
4034 return;
4035 }
4036 spin_unlock_irqrestore(ap->lock, flags);
4037
4038 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4039
Aaron Lu3dc67442013-01-15 17:21:00 +08004040 /*
4041 * If we have a ZPODD attached, check its zero
4042 * power ready status before the port is frozen.
Aaron Lua7ff60d2013-01-25 14:29:35 +08004043 * Only needed for runtime suspend.
Aaron Lu3dc67442013-01-15 17:21:00 +08004044 */
Aaron Lua7ff60d2013-01-25 14:29:35 +08004045 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4046 ata_for_each_dev(dev, &ap->link, ENABLED) {
4047 if (zpodd_dev_enabled(dev))
4048 zpodd_on_suspend(dev);
4049 }
Aaron Lu3dc67442013-01-15 17:21:00 +08004050 }
4051
Tejun Heo64578a32007-05-15 03:28:16 +09004052 /* tell ACPI we're suspending */
4053 rc = ata_acpi_on_suspend(ap);
4054 if (rc)
4055 goto out;
4056
Tejun Heo500530f2006-07-03 16:07:27 +09004057 /* suspend */
4058 ata_eh_freeze_port(ap);
4059
4060 if (ap->ops->port_suspend)
4061 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4062
Aaron Lua7ff60d2013-01-25 14:29:35 +08004063 ata_acpi_set_state(ap, ap->pm_mesg);
Tejun Heo64578a32007-05-15 03:28:16 +09004064 out:
Tejun Heo500530f2006-07-03 16:07:27 +09004065 /* report result */
4066 spin_lock_irqsave(ap->lock, flags);
4067
4068 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4069 if (rc == 0)
4070 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09004071 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09004072 ata_port_schedule_eh(ap);
4073
4074 if (ap->pm_result) {
4075 *ap->pm_result = rc;
4076 ap->pm_result = NULL;
4077 }
4078
4079 spin_unlock_irqrestore(ap->lock, flags);
4080
4081 return;
4082}
4083
4084/**
4085 * ata_eh_handle_port_resume - perform port resume operation
4086 * @ap: port to resume
4087 *
4088 * Resume @ap.
4089 *
Tejun Heo500530f2006-07-03 16:07:27 +09004090 * LOCKING:
4091 * Kernel thread context (may sleep).
4092 */
4093static void ata_eh_handle_port_resume(struct ata_port *ap)
4094{
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004095 struct ata_link *link;
4096 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09004097 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02004098 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09004099
4100 /* are we resuming? */
4101 spin_lock_irqsave(ap->lock, flags);
4102 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
Aaron Lua7ff60d2013-01-25 14:29:35 +08004103 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
Tejun Heo500530f2006-07-03 16:07:27 +09004104 spin_unlock_irqrestore(ap->lock, flags);
4105 return;
4106 }
4107 spin_unlock_irqrestore(ap->lock, flags);
4108
Tejun Heo9666f402007-05-04 21:27:47 +02004109 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09004110
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004111 /*
4112 * Error timestamps are in jiffies which doesn't run while
4113 * suspended and PHY events during resume isn't too uncommon.
4114 * When the two are combined, it can lead to unnecessary speed
4115 * downs if the machine is suspended and resumed repeatedly.
4116 * Clear error history.
4117 */
4118 ata_for_each_link(link, ap, HOST_FIRST)
4119 ata_for_each_dev(dev, link, ALL)
4120 ata_ering_clear(&dev->ering);
4121
Aaron Lua7ff60d2013-01-25 14:29:35 +08004122 ata_acpi_set_state(ap, ap->pm_mesg);
Shaohua Libd3adca2007-11-02 09:32:38 +08004123
Tejun Heo500530f2006-07-03 16:07:27 +09004124 if (ap->ops->port_resume)
4125 rc = ap->ops->port_resume(ap);
4126
Tejun Heo67465442007-05-15 03:28:16 +09004127 /* tell ACPI that we're resuming */
4128 ata_acpi_on_resume(ap);
4129
Tejun Heo9666f402007-05-04 21:27:47 +02004130 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09004131 spin_lock_irqsave(ap->lock, flags);
4132 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4133 if (ap->pm_result) {
4134 *ap->pm_result = rc;
4135 ap->pm_result = NULL;
4136 }
4137 spin_unlock_irqrestore(ap->lock, flags);
4138}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09004139#endif /* CONFIG_PM */