blob: c61316e9d2f7fe0aa45cf542093be2fd811eec05 [file] [log] [blame]
Tejun Heoece1d632006-04-02 18:51:53 +09001/*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
Tejun Heoece1d632006-04-02 18:51:53 +090035#include <linux/kernel.h>
Jens Axboe242f9dc2008-09-14 05:55:09 -070036#include <linux/blkdev.h>
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
Gwendal Grignoud9027472010-05-25 12:31:38 -0700422int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
423{
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
Tejun Heoe30349d2006-07-03 03:02:15 +0900796 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900797 * that if exception occurs after this point but
798 * before EH completion, SCSI midlayer will
799 * re-initiate EH.
800 */
801 host->host_eh_scheduled = 0;
802
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/**
989 * ata_port_schedule_eh - schedule error handling without a qc
990 * @ap: ATA port to schedule EH for
991 *
992 * Schedule error handling for @ap. EH will kick in as soon as
993 * all commands are drained.
994 *
995 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400996 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900997 */
998void ata_port_schedule_eh(struct ata_port *ap)
999{
1000 WARN_ON(!ap->ops->error_handler);
1001
Tejun Heof4d6d002007-05-01 11:50:15 +02001002 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1003 return;
1004
Tejun Heo5ddf24c2007-07-16 14:29:41 +09001005 ata_eh_set_pending(ap, 1);
Jeff Garzikcca39742006-08-24 03:19:22 -04001006 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +09001007
1008 DPRINTK("port EH scheduled\n");
1009}
1010
Tejun Heodbd82612007-08-06 18:36:23 +09001011static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +09001012{
1013 int tag, nr_aborted = 0;
1014
1015 WARN_ON(!ap->ops->error_handler);
1016
Tejun Heo5ddf24c2007-07-16 14:29:41 +09001017 /* we're gonna abort all commands, no need for fast drain */
1018 ata_eh_set_pending(ap, 0);
1019
Tejun Heo7b70fc02006-05-15 20:58:07 +09001020 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1021 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1022
Tejun Heodbd82612007-08-06 18:36:23 +09001023 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +09001024 qc->flags |= ATA_QCFLAG_FAILED;
1025 ata_qc_complete(qc);
1026 nr_aborted++;
1027 }
1028 }
1029
1030 if (!nr_aborted)
1031 ata_port_schedule_eh(ap);
1032
1033 return nr_aborted;
1034}
1035
Tejun Heoe3180492006-05-15 20:58:09 +09001036/**
Tejun Heodbd82612007-08-06 18:36:23 +09001037 * ata_link_abort - abort all qc's on the link
1038 * @link: ATA link to abort qc's for
1039 *
1040 * Abort all active qc's active on @link and schedule EH.
1041 *
1042 * LOCKING:
1043 * spin_lock_irqsave(host lock)
1044 *
1045 * RETURNS:
1046 * Number of aborted qc's.
1047 */
1048int ata_link_abort(struct ata_link *link)
1049{
1050 return ata_do_link_abort(link->ap, link);
1051}
1052
1053/**
1054 * ata_port_abort - abort all qc's on the port
1055 * @ap: ATA port to abort qc's for
1056 *
1057 * Abort all active qc's of @ap and schedule EH.
1058 *
1059 * LOCKING:
1060 * spin_lock_irqsave(host_set lock)
1061 *
1062 * RETURNS:
1063 * Number of aborted qc's.
1064 */
1065int ata_port_abort(struct ata_port *ap)
1066{
1067 return ata_do_link_abort(ap, NULL);
1068}
1069
1070/**
Tejun Heoe3180492006-05-15 20:58:09 +09001071 * __ata_port_freeze - freeze port
1072 * @ap: ATA port to freeze
1073 *
1074 * This function is called when HSM violation or some other
1075 * condition disrupts normal operation of the port. Frozen port
1076 * is not allowed to perform any operation until the port is
1077 * thawed, which usually follows a successful reset.
1078 *
1079 * ap->ops->freeze() callback can be used for freezing the port
1080 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1081 * port cannot be frozen hardware-wise, the interrupt handler
1082 * must ack and clear interrupts unconditionally while the port
1083 * is frozen.
1084 *
1085 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001086 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001087 */
1088static void __ata_port_freeze(struct ata_port *ap)
1089{
1090 WARN_ON(!ap->ops->error_handler);
1091
1092 if (ap->ops->freeze)
1093 ap->ops->freeze(ap);
1094
Tejun Heob51e9e52006-06-29 01:29:30 +09001095 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001096
Tejun Heo44877b42007-02-21 01:06:51 +09001097 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001098}
1099
1100/**
1101 * ata_port_freeze - abort & freeze port
1102 * @ap: ATA port to freeze
1103 *
Jeff Garzik54c38442009-04-07 19:13:15 -04001104 * Abort and freeze @ap. The freeze operation must be called
1105 * first, because some hardware requires special operations
1106 * before the taskfile registers are accessible.
Tejun Heoe3180492006-05-15 20:58:09 +09001107 *
1108 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001109 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001110 *
1111 * RETURNS:
1112 * Number of aborted commands.
1113 */
1114int ata_port_freeze(struct ata_port *ap)
1115{
1116 int nr_aborted;
1117
1118 WARN_ON(!ap->ops->error_handler);
1119
Tejun Heoe3180492006-05-15 20:58:09 +09001120 __ata_port_freeze(ap);
Jeff Garzik54c38442009-04-07 19:13:15 -04001121 nr_aborted = ata_port_abort(ap);
Tejun Heoe3180492006-05-15 20:58:09 +09001122
1123 return nr_aborted;
1124}
1125
1126/**
Tejun Heo7d77b242007-09-23 13:14:13 +09001127 * sata_async_notification - SATA async notification handler
1128 * @ap: ATA port where async notification is received
1129 *
1130 * Handler to be called when async notification via SDB FIS is
1131 * received. This function schedules EH if necessary.
1132 *
1133 * LOCKING:
1134 * spin_lock_irqsave(host lock)
1135 *
1136 * RETURNS:
1137 * 1 if EH is scheduled, 0 otherwise.
1138 */
1139int sata_async_notification(struct ata_port *ap)
1140{
1141 u32 sntf;
1142 int rc;
1143
1144 if (!(ap->flags & ATA_FLAG_AN))
1145 return 0;
1146
1147 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1148 if (rc == 0)
1149 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1150
Tejun Heo071f44b2008-04-07 22:47:22 +09001151 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001152 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +09001153 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001154 /* PMP is not attached. Check whether ATAPI
1155 * AN is configured. If so, notify media
1156 * change.
1157 */
1158 struct ata_device *dev = ap->link.device;
1159
1160 if ((dev->class == ATA_DEV_ATAPI) &&
1161 (dev->flags & ATA_DFLAG_AN))
1162 ata_scsi_media_change_notify(dev);
1163 return 0;
1164 } else {
1165 /* PMP is attached but SNTF is not available.
1166 * ATAPI async media change notification is
1167 * not used. The PMP must be reporting PHY
1168 * status change, schedule EH.
1169 */
1170 ata_port_schedule_eh(ap);
1171 return 1;
1172 }
1173 } else {
1174 /* PMP is attached and SNTF is available */
1175 struct ata_link *link;
1176
1177 /* check and notify ATAPI AN */
Tejun Heo1eca4362008-11-03 20:03:17 +09001178 ata_for_each_link(link, ap, EDGE) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001179 if (!(sntf & (1 << link->pmp)))
1180 continue;
1181
1182 if ((link->device->class == ATA_DEV_ATAPI) &&
1183 (link->device->flags & ATA_DFLAG_AN))
1184 ata_scsi_media_change_notify(link->device);
1185 }
1186
1187 /* If PMP is reporting that PHY status of some
1188 * downstream ports has changed, schedule EH.
1189 */
1190 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1191 ata_port_schedule_eh(ap);
1192 return 1;
1193 }
1194
1195 return 0;
1196 }
1197}
1198
1199/**
Tejun Heoe3180492006-05-15 20:58:09 +09001200 * ata_eh_freeze_port - EH helper to freeze port
1201 * @ap: ATA port to freeze
1202 *
1203 * Freeze @ap.
1204 *
1205 * LOCKING:
1206 * None.
1207 */
1208void ata_eh_freeze_port(struct ata_port *ap)
1209{
1210 unsigned long flags;
1211
1212 if (!ap->ops->error_handler)
1213 return;
1214
Jeff Garzikba6a1302006-06-22 23:46:10 -04001215 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001216 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -04001217 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001218}
1219
1220/**
1221 * ata_port_thaw_port - EH helper to thaw port
1222 * @ap: ATA port to thaw
1223 *
1224 * Thaw frozen port @ap.
1225 *
1226 * LOCKING:
1227 * None.
1228 */
1229void ata_eh_thaw_port(struct ata_port *ap)
1230{
1231 unsigned long flags;
1232
1233 if (!ap->ops->error_handler)
1234 return;
1235
Jeff Garzikba6a1302006-06-22 23:46:10 -04001236 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001237
Tejun Heob51e9e52006-06-29 01:29:30 +09001238 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001239
1240 if (ap->ops->thaw)
1241 ap->ops->thaw(ap);
1242
Jeff Garzikba6a1302006-06-22 23:46:10 -04001243 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001244
Tejun Heo44877b42007-02-21 01:06:51 +09001245 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001246}
1247
Tejun Heoece1d632006-04-02 18:51:53 +09001248static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1249{
1250 /* nada */
1251}
1252
1253static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1254{
1255 struct ata_port *ap = qc->ap;
1256 struct scsi_cmnd *scmd = qc->scsicmd;
1257 unsigned long flags;
1258
Jeff Garzikba6a1302006-06-22 23:46:10 -04001259 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001260 qc->scsidone = ata_eh_scsidone;
1261 __ata_qc_complete(qc);
1262 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -04001263 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001264
1265 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1266}
1267
1268/**
1269 * ata_eh_qc_complete - Complete an active ATA command from EH
1270 * @qc: Command to complete
1271 *
1272 * Indicate to the mid and upper layers that an ATA command has
1273 * completed. To be used from EH.
1274 */
1275void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1276{
1277 struct scsi_cmnd *scmd = qc->scsicmd;
1278 scmd->retries = scmd->allowed;
1279 __ata_eh_qc_complete(qc);
1280}
1281
1282/**
1283 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1284 * @qc: Command to retry
1285 *
1286 * Indicate to the mid and upper layers that an ATA command
1287 * should be retried. To be used from EH.
1288 *
1289 * SCSI midlayer limits the number of retries to scmd->allowed.
1290 * scmd->retries is decremented for commands which get retried
1291 * due to unrelated failures (qc->err_mask is zero).
1292 */
1293void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1294{
1295 struct scsi_cmnd *scmd = qc->scsicmd;
1296 if (!qc->err_mask && scmd->retries)
1297 scmd->retries--;
1298 __ata_eh_qc_complete(qc);
1299}
Tejun Heo022bdb02006-05-15 20:58:22 +09001300
1301/**
Tejun Heo678afac2009-01-29 20:31:30 +09001302 * ata_dev_disable - disable ATA device
1303 * @dev: ATA device to disable
1304 *
1305 * Disable @dev.
1306 *
1307 * Locking:
1308 * EH context.
1309 */
1310void ata_dev_disable(struct ata_device *dev)
1311{
1312 if (!ata_dev_enabled(dev))
1313 return;
1314
1315 if (ata_msg_drv(dev->link->ap))
Joe Perchesa9a79df2011-04-15 15:51:59 -07001316 ata_dev_warn(dev, "disabled\n");
Tejun Heo678afac2009-01-29 20:31:30 +09001317 ata_acpi_on_disable(dev);
1318 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1319 dev->class++;
Tejun Heo99cf6102009-01-29 20:31:32 +09001320
1321 /* From now till the next successful probe, ering is used to
1322 * track probe failures. Clear accumulated device error info.
1323 */
1324 ata_ering_clear(&dev->ering);
Tejun Heo678afac2009-01-29 20:31:30 +09001325}
1326
1327/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001328 * ata_eh_detach_dev - detach ATA device
1329 * @dev: ATA device to detach
1330 *
1331 * Detach @dev.
1332 *
1333 * LOCKING:
1334 * None.
1335 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001336void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001337{
Tejun Heof58229f2007-08-06 18:36:23 +09001338 struct ata_link *link = dev->link;
1339 struct ata_port *ap = link->ap;
Tejun Heo90484eb2008-10-26 15:43:03 +09001340 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001341 unsigned long flags;
1342
1343 ata_dev_disable(dev);
1344
Jeff Garzikba6a1302006-06-22 23:46:10 -04001345 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001346
1347 dev->flags &= ~ATA_DFLAG_DETACH;
1348
1349 if (ata_scsi_offline_dev(dev)) {
1350 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001351 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001352 }
1353
Tejun Heo90484eb2008-10-26 15:43:03 +09001354 /* clear per-dev EH info */
Tejun Heof58229f2007-08-06 18:36:23 +09001355 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1356 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heo90484eb2008-10-26 15:43:03 +09001357 ehc->saved_xfer_mode[dev->devno] = 0;
1358 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heobeb07c12006-06-24 20:30:19 +09001359
Jeff Garzikba6a1302006-06-22 23:46:10 -04001360 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001361}
1362
1363/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001364 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001365 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001366 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001367 * @action: action about to be performed
1368 *
1369 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001370 * in @link->eh_info such that eh actions are not unnecessarily
1371 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001372 *
1373 * LOCKING:
1374 * None.
1375 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001376void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1377 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001378{
Tejun Heo955e57d2007-08-06 18:36:23 +09001379 struct ata_port *ap = link->ap;
1380 struct ata_eh_info *ehi = &link->eh_info;
1381 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001382 unsigned long flags;
1383
Jeff Garzikba6a1302006-06-22 23:46:10 -04001384 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001385
Tejun Heo955e57d2007-08-06 18:36:23 +09001386 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001387
Tejun Heoa568d1d2008-10-21 20:37:21 +09001388 /* About to take EH action, set RECOVERED. Ignore actions on
1389 * slave links as master will do them again.
1390 */
1391 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
Tejun Heo1cdaf532006-07-03 16:07:26 +09001392 ap->pflags |= ATA_PFLAG_RECOVERED;
1393
Jeff Garzikba6a1302006-06-22 23:46:10 -04001394 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001395}
1396
1397/**
Tejun Heo47005f22006-06-19 18:27:23 +09001398 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001399* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001400 * @dev: target ATA dev for per-dev action (can be NULL)
1401 * @action: action just completed
1402 *
1403 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001404 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001405 *
1406 * LOCKING:
1407 * None.
1408 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001409void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1410 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001411{
Tejun Heo955e57d2007-08-06 18:36:23 +09001412 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001413
Tejun Heo955e57d2007-08-06 18:36:23 +09001414 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001415}
1416
1417/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001418 * ata_err_string - convert err_mask to descriptive string
1419 * @err_mask: error mask to convert to string
1420 *
1421 * Convert @err_mask to descriptive string. Errors are
1422 * prioritized according to severity and only the most severe
1423 * error is reported.
1424 *
1425 * LOCKING:
1426 * None.
1427 *
1428 * RETURNS:
1429 * Descriptive string for @err_mask
1430 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001431static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001432{
1433 if (err_mask & AC_ERR_HOST_BUS)
1434 return "host bus error";
1435 if (err_mask & AC_ERR_ATA_BUS)
1436 return "ATA bus error";
1437 if (err_mask & AC_ERR_TIMEOUT)
1438 return "timeout";
1439 if (err_mask & AC_ERR_HSM)
1440 return "HSM violation";
1441 if (err_mask & AC_ERR_SYSTEM)
1442 return "internal error";
1443 if (err_mask & AC_ERR_MEDIA)
1444 return "media error";
1445 if (err_mask & AC_ERR_INVALID)
1446 return "invalid argument";
1447 if (err_mask & AC_ERR_DEV)
1448 return "device error";
1449 return "unknown error";
1450}
1451
1452/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001453 * ata_read_log_page - read a specific log page
1454 * @dev: target device
1455 * @page: page to read
1456 * @buf: buffer to store read page
1457 * @sectors: number of sectors to read
1458 *
1459 * Read log page using READ_LOG_EXT command.
1460 *
1461 * LOCKING:
1462 * Kernel thread context (may sleep).
1463 *
1464 * RETURNS:
1465 * 0 on success, AC_ERR_* mask otherwise.
1466 */
1467static unsigned int ata_read_log_page(struct ata_device *dev,
1468 u8 page, void *buf, unsigned int sectors)
1469{
1470 struct ata_taskfile tf;
1471 unsigned int err_mask;
1472
1473 DPRINTK("read log page - page %d\n", page);
1474
1475 ata_tf_init(dev, &tf);
1476 tf.command = ATA_CMD_READ_LOG_EXT;
1477 tf.lbal = page;
1478 tf.nsect = sectors;
1479 tf.hob_nsect = sectors >> 8;
1480 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1481 tf.protocol = ATA_PROT_PIO;
1482
1483 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001484 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001485
1486 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1487 return err_mask;
1488}
1489
1490/**
1491 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1492 * @dev: Device to read log page 10h from
1493 * @tag: Resulting tag of the failed command
1494 * @tf: Resulting taskfile registers of the failed command
1495 *
1496 * Read log page 10h to obtain NCQ error details and clear error
1497 * condition.
1498 *
1499 * LOCKING:
1500 * Kernel thread context (may sleep).
1501 *
1502 * RETURNS:
1503 * 0 on success, -errno otherwise.
1504 */
1505static int ata_eh_read_log_10h(struct ata_device *dev,
1506 int *tag, struct ata_taskfile *tf)
1507{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001508 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001509 unsigned int err_mask;
1510 u8 csum;
1511 int i;
1512
1513 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1514 if (err_mask)
1515 return -EIO;
1516
1517 csum = 0;
1518 for (i = 0; i < ATA_SECT_SIZE; i++)
1519 csum += buf[i];
1520 if (csum)
Joe Perchesa9a79df2011-04-15 15:51:59 -07001521 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1522 csum);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001523
1524 if (buf[0] & 0x80)
1525 return -ENOENT;
1526
1527 *tag = buf[0] & 0x1f;
1528
1529 tf->command = buf[2];
1530 tf->feature = buf[3];
1531 tf->lbal = buf[4];
1532 tf->lbam = buf[5];
1533 tf->lbah = buf[6];
1534 tf->device = buf[7];
1535 tf->hob_lbal = buf[8];
1536 tf->hob_lbam = buf[9];
1537 tf->hob_lbah = buf[10];
1538 tf->nsect = buf[12];
1539 tf->hob_nsect = buf[13];
1540
1541 return 0;
1542}
1543
1544/**
Tejun Heo11fc33d2008-08-30 14:20:01 +02001545 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1546 * @dev: target ATAPI device
1547 * @r_sense_key: out parameter for sense_key
1548 *
1549 * Perform ATAPI TEST_UNIT_READY.
1550 *
1551 * LOCKING:
1552 * EH context (may sleep).
1553 *
1554 * RETURNS:
1555 * 0 on success, AC_ERR_* mask on failure.
1556 */
1557static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1558{
1559 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1560 struct ata_taskfile tf;
1561 unsigned int err_mask;
1562
1563 ata_tf_init(dev, &tf);
1564
1565 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1566 tf.command = ATA_CMD_PACKET;
1567 tf.protocol = ATAPI_PROT_NODATA;
1568
1569 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1570 if (err_mask == AC_ERR_DEV)
1571 *r_sense_key = tf.feature >> 4;
1572 return err_mask;
1573}
1574
1575/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001576 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1577 * @dev: device to perform REQUEST_SENSE to
1578 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
Tejun Heo3eabddb2008-06-10 18:28:05 +09001579 * @dfl_sense_key: default sense key to use
Tejun Heo022bdb02006-05-15 20:58:22 +09001580 *
1581 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1582 * SENSE. This function is EH helper.
1583 *
1584 * LOCKING:
1585 * Kernel thread context (may sleep).
1586 *
1587 * RETURNS:
1588 * 0 on success, AC_ERR_* mask on failure
1589 */
Tejun Heo3eabddb2008-06-10 18:28:05 +09001590static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1591 u8 *sense_buf, u8 dfl_sense_key)
Tejun Heo022bdb02006-05-15 20:58:22 +09001592{
Tejun Heo3eabddb2008-06-10 18:28:05 +09001593 u8 cdb[ATAPI_CDB_LEN] =
1594 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001595 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001596 struct ata_taskfile tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001597
1598 DPRINTK("ATAPI request sense\n");
1599
Tejun Heo022bdb02006-05-15 20:58:22 +09001600 /* FIXME: is this needed? */
1601 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1602
Albert Lee56287762007-04-02 11:30:46 +08001603 /* initialize sense_buf with the error register,
1604 * for the case where they are -not- overwritten
1605 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001606 sense_buf[0] = 0x70;
Tejun Heo3eabddb2008-06-10 18:28:05 +09001607 sense_buf[2] = dfl_sense_key;
Albert Lee56287762007-04-02 11:30:46 +08001608
Jeff Garzika617c092007-05-21 20:14:23 -04001609 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001610 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001611
Tejun Heo022bdb02006-05-15 20:58:22 +09001612 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1613 tf.command = ATA_CMD_PACKET;
1614
1615 /* is it pointless to prefer PIO for "safety reasons"? */
1616 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001617 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001618 tf.feature |= ATAPI_PKT_DMA;
1619 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001620 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001621 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1622 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001623 }
1624
1625 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001626 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001627}
1628
1629/**
1630 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001631 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001632 *
1633 * Analyze SError if available and further determine cause of
1634 * failure.
1635 *
1636 * LOCKING:
1637 * None.
1638 */
Tejun Heo02607312007-08-06 18:36:23 +09001639static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001640{
Tejun Heo02607312007-08-06 18:36:23 +09001641 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001642 u32 serror = ehc->i.serror;
1643 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001644 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001645
Tejun Heoe0614db2008-05-19 01:15:09 +09001646 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001647 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001648 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001649 }
1650 if (serror & SERR_PROTOCOL) {
1651 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001652 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001653 }
1654 if (serror & SERR_INTERNAL) {
1655 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001656 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001657 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001658
1659 /* Determine whether a hotplug event has occurred. Both
1660 * SError.N/X are considered hotplug events for enabled or
1661 * host links. For disabled PMP links, only N bit is
1662 * considered as X bit is left at 1 for link plugging.
1663 */
Tejun Heoeb0e85e2011-02-24 19:30:37 +01001664 if (link->lpm_policy > ATA_LPM_MAX_POWER)
Tejun Heo6b7ae952010-09-01 17:50:06 +02001665 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1666 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
Tejun Heof9df58c2007-09-23 13:14:13 +09001667 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1668 else
1669 hotplug_mask = SERR_PHYRDY_CHG;
1670
1671 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001672 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001673
1674 ehc->i.err_mask |= err_mask;
1675 ehc->i.action |= action;
1676}
1677
1678/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001679 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001680 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001681 *
1682 * Read log page 10h, determine the offending qc and acquire
1683 * error status TF. For NCQ device errors, all LLDDs have to do
1684 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1685 * care of the rest.
1686 *
1687 * LOCKING:
1688 * Kernel thread context (may sleep).
1689 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001690void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001691{
Tejun Heo02607312007-08-06 18:36:23 +09001692 struct ata_port *ap = link->ap;
1693 struct ata_eh_context *ehc = &link->eh_context;
1694 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001695 struct ata_queued_cmd *qc;
1696 struct ata_taskfile tf;
1697 int tag, rc;
1698
1699 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001700 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001701 return;
1702
1703 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001704 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001705 return;
1706
1707 /* has LLDD analyzed already? */
1708 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1709 qc = __ata_qc_from_tag(ap, tag);
1710
1711 if (!(qc->flags & ATA_QCFLAG_FAILED))
1712 continue;
1713
1714 if (qc->err_mask)
1715 return;
1716 }
1717
1718 /* okay, this error is ours */
Jeff Garzika09bf4c2010-04-22 21:59:13 -04001719 memset(&tf, 0, sizeof(tf));
Tejun Heoe8ee8452006-05-15 21:03:46 +09001720 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1721 if (rc) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07001722 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1723 rc);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001724 return;
1725 }
1726
Tejun Heo02607312007-08-06 18:36:23 +09001727 if (!(link->sactive & (1 << tag))) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07001728 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1729 tag);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001730 return;
1731 }
1732
1733 /* we've got the perpetrator, condemn it */
1734 qc = __ata_qc_from_tag(ap, tag);
1735 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001736 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001737 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001738 ehc->i.err_mask &= ~AC_ERR_DEV;
1739}
1740
1741/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001742 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1743 * @qc: qc to analyze
1744 * @tf: Taskfile registers to analyze
1745 *
1746 * Analyze taskfile of @qc and further determine cause of
1747 * failure. This function also requests ATAPI sense data if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001748 * available.
Tejun Heo022bdb02006-05-15 20:58:22 +09001749 *
1750 * LOCKING:
1751 * Kernel thread context (may sleep).
1752 *
1753 * RETURNS:
1754 * Determined recovery action
1755 */
1756static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1757 const struct ata_taskfile *tf)
1758{
1759 unsigned int tmp, action = 0;
1760 u8 stat = tf->command, err = tf->feature;
1761
1762 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1763 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001764 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001765 }
1766
Tejun Heoa51d6442007-03-20 15:24:11 +09001767 if (stat & (ATA_ERR | ATA_DF))
1768 qc->err_mask |= AC_ERR_DEV;
1769 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001770 return 0;
1771
1772 switch (qc->dev->class) {
1773 case ATA_DEV_ATA:
1774 if (err & ATA_ICRC)
1775 qc->err_mask |= AC_ERR_ATA_BUS;
1776 if (err & ATA_UNC)
1777 qc->err_mask |= AC_ERR_MEDIA;
1778 if (err & ATA_IDNF)
1779 qc->err_mask |= AC_ERR_INVALID;
1780 break;
1781
1782 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001783 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Tejun Heo3eabddb2008-06-10 18:28:05 +09001784 tmp = atapi_eh_request_sense(qc->dev,
1785 qc->scsicmd->sense_buffer,
1786 qc->result_tf.feature >> 4);
Tejun Heoa569a302006-11-21 10:40:51 +09001787 if (!tmp) {
1788 /* ATA_QCFLAG_SENSE_VALID is used to
1789 * tell atapi_qc_complete() that sense
1790 * data is already valid.
1791 *
1792 * TODO: interpret sense data and set
1793 * appropriate err_mask.
1794 */
1795 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1796 } else
1797 qc->err_mask |= tmp;
1798 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001799 }
1800
1801 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001802 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001803
1804 return action;
1805}
1806
Tejun Heo76326ac2007-11-27 19:28:59 +09001807static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1808 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001809{
Tejun Heo76326ac2007-11-27 19:28:59 +09001810 int base = 0;
1811
1812 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1813 *xfer_ok = 1;
1814
1815 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001816 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001817
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001818 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001819 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001820
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001821 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001822 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001823
Tejun Heo3884f7b2007-11-27 19:28:56 +09001824 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001825 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001826 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001827 if ((err_mask &
1828 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001829 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001830 }
1831
1832 return 0;
1833}
1834
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001835struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001836 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001837 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001838 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001839};
1840
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001841static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001842{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001843 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001844 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001845
Gwendal Grignoud9027472010-05-25 12:31:38 -07001846 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
Tejun Heo022bdb02006-05-15 20:58:22 +09001847 return -1;
1848
Tejun Heo76326ac2007-11-27 19:28:59 +09001849 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1850 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001851 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001852
Tejun Heo022bdb02006-05-15 20:58:22 +09001853 return 0;
1854}
1855
1856/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001857 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001858 * @dev: Device of interest
1859 *
1860 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001861 * whether NCQ needs to be turned off, transfer speed should be
1862 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001863 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001864 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001865 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001866 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1867 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001868 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001869 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001870 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001871 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1872 * data transfer hasn't been verified.
1873 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001874 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001875 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001876 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001877 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001878 * SPEED_DOWN : Speed down transfer speed but don't fall back
1879 * to PIO.
1880 *
1881 * FALLBACK_TO_PIO : Fall back to PIO.
1882 *
1883 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001884 * taken per error. An action triggered by non-DUBIOUS errors
1885 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1886 * This is to expedite speed down decisions right after device is
1887 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001888 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001889 * The followings are speed down rules. #1 and #2 deal with
1890 * DUBIOUS errors.
1891 *
1892 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1893 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1894 *
1895 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1896 * occurred during last 5 mins, NCQ_OFF.
1897 *
1898 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001899 * occurred during last 5 mins, FALLBACK_TO_PIO
Tejun Heo3884f7b2007-11-27 19:28:56 +09001900 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001901 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001902 * during last 10 mins, NCQ_OFF.
1903 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001904 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001905 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001906 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001907 * LOCKING:
1908 * Inherited from caller.
1909 *
1910 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001911 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001912 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001913static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001914{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001915 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1916 u64 j64 = get_jiffies_64();
1917 struct speed_down_verdict_arg arg;
1918 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001919
Tejun Heo3884f7b2007-11-27 19:28:56 +09001920 /* scan past 5 mins of error history */
1921 memset(&arg, 0, sizeof(arg));
1922 arg.since = j64 - min(j64, j5mins);
1923 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1924
Tejun Heo76326ac2007-11-27 19:28:59 +09001925 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1926 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1927 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1928 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1929
1930 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1931 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1932 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1933
Tejun Heo3884f7b2007-11-27 19:28:56 +09001934 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1935 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001936 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001937 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1938
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001939 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001940 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001941 arg.since = j64 - min(j64, j10mins);
1942 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001943
Tejun Heo3884f7b2007-11-27 19:28:56 +09001944 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1945 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001946 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001947
1948 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1949 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001950 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001951 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001952
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001953 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001954}
1955
1956/**
1957 * ata_eh_speed_down - record error and speed down if necessary
1958 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001959 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001960 * @err_mask: err_mask of the error
1961 *
1962 * Record error and examine error history to determine whether
1963 * adjusting transmission speed is necessary. It also sets
1964 * transmission limits appropriately if such adjustment is
1965 * necessary.
1966 *
1967 * LOCKING:
1968 * Kernel thread context (may sleep).
1969 *
1970 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001971 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001972 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001973static unsigned int ata_eh_speed_down(struct ata_device *dev,
1974 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001975{
Tejun Heob1c72912008-07-31 17:02:43 +09001976 struct ata_link *link = ata_dev_phys_link(dev);
Tejun Heo76326ac2007-11-27 19:28:59 +09001977 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001978 unsigned int verdict;
1979 unsigned int action = 0;
1980
1981 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001982 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001983 return 0;
1984
1985 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001986 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001987 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001988
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001989 /* turn off NCQ? */
1990 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1991 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1992 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1993 dev->flags |= ATA_DFLAG_NCQ_OFF;
Joe Perchesa9a79df2011-04-15 15:51:59 -07001994 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001995 goto done;
1996 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001997
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001998 /* speed down? */
1999 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2000 /* speed down SATA link speed if possible */
Tejun Heoa07d4992009-01-29 20:31:33 +09002001 if (sata_down_spd_limit(link, 0) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002002 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002003 goto done;
2004 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002005
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002006 /* lower transfer mode */
2007 if (dev->spdn_cnt < 2) {
2008 static const int dma_dnxfer_sel[] =
2009 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2010 static const int pio_dnxfer_sel[] =
2011 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2012 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09002013
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002014 if (dev->xfer_shift != ATA_SHIFT_PIO)
2015 sel = dma_dnxfer_sel[dev->spdn_cnt];
2016 else
2017 sel = pio_dnxfer_sel[dev->spdn_cnt];
2018
2019 dev->spdn_cnt++;
2020
2021 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09002022 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002023 goto done;
2024 }
2025 }
2026 }
2027
2028 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09002029 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002030 */
2031 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09002032 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002033 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2034 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2035 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09002036 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002037 goto done;
2038 }
2039 }
2040
Tejun Heo022bdb02006-05-15 20:58:22 +09002041 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002042 done:
2043 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09002044 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2045 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09002046 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09002047}
2048
2049/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002050 * ata_eh_link_autopsy - analyze error and determine recovery action
2051 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09002052 *
Tejun Heo02607312007-08-06 18:36:23 +09002053 * Analyze why @link failed and determine which recovery actions
2054 * are needed. This function also sets more detailed AC_ERR_*
2055 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09002056 *
2057 * LOCKING:
2058 * Kernel thread context (may sleep).
2059 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002060static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002061{
Tejun Heo02607312007-08-06 18:36:23 +09002062 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09002063 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09002064 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002065 unsigned int all_err_mask = 0, eflags = 0;
2066 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09002067 u32 serror;
2068 int rc;
2069
2070 DPRINTK("ENTER\n");
2071
Tejun Heo1cdaf532006-07-03 16:07:26 +09002072 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2073 return;
2074
Tejun Heo022bdb02006-05-15 20:58:22 +09002075 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09002076 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09002077 if (rc == 0) {
2078 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09002079 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09002080 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09002081 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09002082 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09002083 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09002084 ehc->i.err_mask |= AC_ERR_OTHER;
2085 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002086
Tejun Heoe8ee8452006-05-15 21:03:46 +09002087 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09002088 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09002089
Tejun Heo022bdb02006-05-15 20:58:22 +09002090 /* any real error trumps AC_ERR_OTHER */
2091 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2092 ehc->i.err_mask &= ~AC_ERR_OTHER;
2093
2094 all_err_mask |= ehc->i.err_mask;
2095
2096 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2097 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2098
Tejun Heob1c72912008-07-31 17:02:43 +09002099 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2100 ata_dev_phys_link(qc->dev) != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002101 continue;
2102
2103 /* inherit upper level err_mask */
2104 qc->err_mask |= ehc->i.err_mask;
2105
Tejun Heo022bdb02006-05-15 20:58:22 +09002106 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002107 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002108
2109 /* DEV errors are probably spurious in case of ATA_BUS error */
2110 if (qc->err_mask & AC_ERR_ATA_BUS)
2111 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2112 AC_ERR_INVALID);
2113
2114 /* any real error trumps unknown error */
2115 if (qc->err_mask & ~AC_ERR_OTHER)
2116 qc->err_mask &= ~AC_ERR_OTHER;
2117
2118 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09002119 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09002120 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09002121
Tejun Heo03faab72008-03-27 19:14:24 +09002122 /* determine whether the command is worth retrying */
Tejun Heo534ead72010-01-14 16:18:09 +09002123 if (qc->flags & ATA_QCFLAG_IO ||
2124 (!(qc->err_mask & AC_ERR_INVALID) &&
2125 qc->err_mask != AC_ERR_DEV))
Tejun Heo03faab72008-03-27 19:14:24 +09002126 qc->flags |= ATA_QCFLAG_RETRY;
2127
Tejun Heo022bdb02006-05-15 20:58:22 +09002128 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002129 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09002130 all_err_mask |= qc->err_mask;
2131 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09002132 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09002133 }
2134
Tejun Heoa20f33f2006-05-16 12:58:24 +09002135 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09002136 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09002137 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09002138 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002139 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2140 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09002141 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002142
Tejun Heodfcc1732007-10-31 10:17:05 +09002143 /* If we have offending qcs and the associated failed device,
2144 * perform per-dev EH action only on the offending device.
2145 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002146 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09002147 ehc->i.dev_action[ehc->i.dev->devno] |=
2148 ehc->i.action & ATA_EH_PERDEV_MASK;
2149 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09002150 }
2151
Tejun Heo2695e362008-01-10 13:41:23 +09002152 /* propagate timeout to host link */
2153 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2154 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2155
2156 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09002157 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09002158 if (!dev && ((ata_link_max_devices(link) == 1 &&
2159 ata_dev_enabled(link->device))))
2160 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09002161
Tejun Heo76326ac2007-11-27 19:28:59 +09002162 if (dev) {
2163 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2164 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002165 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09002166 }
Tejun Heodfcc1732007-10-31 10:17:05 +09002167
Tejun Heo022bdb02006-05-15 20:58:22 +09002168 DPRINTK("EXIT\n");
2169}
2170
2171/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002172 * ata_eh_autopsy - analyze error and determine recovery action
2173 * @ap: host port to perform autopsy on
2174 *
2175 * Analyze all links of @ap and determine why they failed and
2176 * which recovery actions are needed.
2177 *
2178 * LOCKING:
2179 * Kernel thread context (may sleep).
2180 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002181void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002182{
2183 struct ata_link *link;
2184
Tejun Heo1eca4362008-11-03 20:03:17 +09002185 ata_for_each_link(link, ap, EDGE)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002186 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09002187
Tejun Heob1c72912008-07-31 17:02:43 +09002188 /* Handle the frigging slave link. Autopsy is done similarly
2189 * but actions and flags are transferred over to the master
2190 * link and handled from there.
2191 */
2192 if (ap->slave_link) {
2193 struct ata_eh_context *mehc = &ap->link.eh_context;
2194 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2195
Tejun Heo848e4c62008-10-21 14:26:39 +09002196 /* transfer control flags from master to slave */
2197 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2198
2199 /* perform autopsy on the slave link */
Tejun Heob1c72912008-07-31 17:02:43 +09002200 ata_eh_link_autopsy(ap->slave_link);
2201
Tejun Heo848e4c62008-10-21 14:26:39 +09002202 /* transfer actions from slave to master and clear slave */
Tejun Heob1c72912008-07-31 17:02:43 +09002203 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2204 mehc->i.action |= sehc->i.action;
2205 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2206 mehc->i.flags |= sehc->i.flags;
2207 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2208 }
2209
Tejun Heo2695e362008-01-10 13:41:23 +09002210 /* Autopsy of fanout ports can affect host link autopsy.
2211 * Perform host link autopsy last.
2212 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002213 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09002214 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09002215}
2216
2217/**
Robert Hancock65211482009-07-14 20:43:39 -06002218 * ata_get_cmd_descript - get description for ATA command
2219 * @command: ATA command code to get description for
2220 *
2221 * Return a textual description of the given command, or NULL if the
2222 * command is not known.
2223 *
2224 * LOCKING:
2225 * None
2226 */
2227const char *ata_get_cmd_descript(u8 command)
2228{
2229#ifdef CONFIG_ATA_VERBOSE_ERROR
2230 static const struct
2231 {
2232 u8 command;
2233 const char *text;
2234 } cmd_descr[] = {
2235 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2236 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2237 { ATA_CMD_STANDBY, "STANDBY" },
2238 { ATA_CMD_IDLE, "IDLE" },
2239 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2240 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2241 { ATA_CMD_NOP, "NOP" },
2242 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2243 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2244 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2245 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2246 { ATA_CMD_SERVICE, "SERVICE" },
2247 { ATA_CMD_READ, "READ DMA" },
2248 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2249 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2250 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2251 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2252 { ATA_CMD_WRITE, "WRITE DMA" },
2253 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2254 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2255 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2256 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2257 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2258 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2259 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2260 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2261 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2262 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2263 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2264 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2265 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2266 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2267 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2268 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2269 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2270 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2271 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2272 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2273 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2274 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2275 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2276 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2277 { ATA_CMD_SLEEP, "SLEEP" },
2278 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2279 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2280 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2281 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2282 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2283 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2284 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2285 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2286 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2287 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2288 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2289 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2290 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2291 { ATA_CMD_PMP_READ, "READ BUFFER" },
2292 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2293 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2294 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2295 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2296 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2297 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2298 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2299 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2300 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2301 { ATA_CMD_SMART, "SMART" },
2302 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2303 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
FUJITA Tomonoriacad762722010-07-05 15:45:20 +09002304 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
Robert Hancock65211482009-07-14 20:43:39 -06002305 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2306 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2307 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2308 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2309 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2310 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2311 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2312 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2313 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2314 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2315 { ATA_CMD_RESTORE, "RECALIBRATE" },
2316 { 0, NULL } /* terminate list */
2317 };
2318
2319 unsigned int i;
2320 for (i = 0; cmd_descr[i].text; i++)
2321 if (cmd_descr[i].command == command)
2322 return cmd_descr[i].text;
2323#endif
2324
2325 return NULL;
2326}
2327
2328/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002329 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09002330 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09002331 *
2332 * Report EH to user.
2333 *
2334 * LOCKING:
2335 * None.
2336 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002337static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002338{
Tejun Heo02607312007-08-06 18:36:23 +09002339 struct ata_port *ap = link->ap;
2340 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002341 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09002342 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09002343 int tag, nr_failed = 0;
2344
Tejun Heo94ff3d52007-10-09 14:57:56 +09002345 if (ehc->i.flags & ATA_EHI_QUIET)
2346 return;
2347
Tejun Heo022bdb02006-05-15 20:58:22 +09002348 desc = NULL;
2349 if (ehc->i.desc[0] != '\0')
2350 desc = ehc->i.desc;
2351
2352 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2353 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2354
Tejun Heob1c72912008-07-31 17:02:43 +09002355 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2356 ata_dev_phys_link(qc->dev) != link ||
Tejun Heoe027bd32007-10-26 16:19:26 +09002357 ((qc->flags & ATA_QCFLAG_QUIET) &&
2358 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002359 continue;
2360 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2361 continue;
2362
2363 nr_failed++;
2364 }
2365
2366 if (!nr_failed && !ehc->i.err_mask)
2367 return;
2368
2369 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09002370 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09002371 frozen = " frozen";
2372
Tejun Heoa1e10f72007-08-18 13:28:49 +09002373 memset(tries_buf, 0, sizeof(tries_buf));
2374 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2375 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2376 ap->eh_tries);
2377
Tejun Heo022bdb02006-05-15 20:58:22 +09002378 if (ehc->i.dev) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002379 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2380 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2381 ehc->i.err_mask, link->sactive, ehc->i.serror,
2382 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002383 if (desc)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002384 ata_dev_err(ehc->i.dev, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002385 } else {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002386 ata_link_err(link, "exception Emask 0x%x "
2387 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2388 ehc->i.err_mask, link->sactive, ehc->i.serror,
2389 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002390 if (desc)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002391 ata_link_err(link, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002392 }
2393
Robert Hancock65211482009-07-14 20:43:39 -06002394#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002395 if (ehc->i.serror)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002396 ata_link_err(link,
Robert Hancock1333e192007-10-02 11:22:02 -04002397 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2398 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2399 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2400 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2401 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2402 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2403 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2404 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2405 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2406 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2407 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2408 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2409 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2410 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2411 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2412 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2413 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002414 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002415#endif
Robert Hancock1333e192007-10-02 11:22:02 -04002416
Tejun Heo022bdb02006-05-15 20:58:22 +09002417 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2418 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09002419 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09002420 const u8 *cdb = qc->cdb;
2421 char data_buf[20] = "";
2422 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002423
Tejun Heo02607312007-08-06 18:36:23 +09002424 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
Tejun Heob1c72912008-07-31 17:02:43 +09002425 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002426 continue;
2427
Tejun Heoabb6a882007-11-28 23:16:09 +09002428 if (qc->dma_dir != DMA_NONE) {
2429 static const char *dma_str[] = {
2430 [DMA_BIDIRECTIONAL] = "bidi",
2431 [DMA_TO_DEVICE] = "out",
2432 [DMA_FROM_DEVICE] = "in",
2433 };
2434 static const char *prot_str[] = {
2435 [ATA_PROT_PIO] = "pio",
2436 [ATA_PROT_DMA] = "dma",
2437 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05002438 [ATAPI_PROT_PIO] = "pio",
2439 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09002440 };
2441
2442 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2443 prot_str[qc->tf.protocol], qc->nbytes,
2444 dma_str[qc->dma_dir]);
2445 }
2446
Robert Hancock65211482009-07-14 20:43:39 -06002447 if (ata_is_atapi(qc->tf.protocol)) {
2448 if (qc->scsicmd)
2449 scsi_print_command(qc->scsicmd);
2450 else
2451 snprintf(cdb_buf, sizeof(cdb_buf),
Tejun Heoabb6a882007-11-28 23:16:09 +09002452 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2453 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2454 cdb[0], cdb[1], cdb[2], cdb[3],
2455 cdb[4], cdb[5], cdb[6], cdb[7],
2456 cdb[8], cdb[9], cdb[10], cdb[11],
2457 cdb[12], cdb[13], cdb[14], cdb[15]);
Robert Hancock65211482009-07-14 20:43:39 -06002458 } else {
2459 const char *descr = ata_get_cmd_descript(cmd->command);
2460 if (descr)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002461 ata_dev_err(qc->dev, "failed command: %s\n",
2462 descr);
Robert Hancock65211482009-07-14 20:43:39 -06002463 }
Tejun Heoabb6a882007-11-28 23:16:09 +09002464
Joe Perchesa9a79df2011-04-15 15:51:59 -07002465 ata_dev_err(qc->dev,
Tejun Heo8a937582006-11-14 22:36:12 +09002466 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002467 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002468 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002469 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002470 cmd->command, cmd->feature, cmd->nsect,
2471 cmd->lbal, cmd->lbam, cmd->lbah,
2472 cmd->hob_feature, cmd->hob_nsect,
2473 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002474 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002475 res->command, res->feature, res->nsect,
2476 res->lbal, res->lbam, res->lbah,
2477 res->hob_feature, res->hob_nsect,
2478 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002479 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2480 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002481
Robert Hancock65211482009-07-14 20:43:39 -06002482#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002483 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002484 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002485 if (res->command & ATA_BUSY)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002486 ata_dev_err(qc->dev, "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002487 else
Joe Perchesa9a79df2011-04-15 15:51:59 -07002488 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
Robert Hancock1333e192007-10-02 11:22:02 -04002489 res->command & ATA_DRDY ? "DRDY " : "",
2490 res->command & ATA_DF ? "DF " : "",
2491 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002492 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002493 }
2494
2495 if (cmd->command != ATA_CMD_PACKET &&
2496 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2497 ATA_ABORTED)))
Joe Perchesa9a79df2011-04-15 15:51:59 -07002498 ata_dev_err(qc->dev, "error: { %s%s%s%s}\n",
Robert Hancock1333e192007-10-02 11:22:02 -04002499 res->feature & ATA_ICRC ? "ICRC " : "",
2500 res->feature & ATA_UNC ? "UNC " : "",
2501 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002502 res->feature & ATA_ABORTED ? "ABRT " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002503#endif
Tejun Heo022bdb02006-05-15 20:58:22 +09002504 }
2505}
2506
Tejun Heo9b1e2652007-08-06 18:36:24 +09002507/**
2508 * ata_eh_report - report error handling to user
2509 * @ap: ATA port to report EH about
2510 *
2511 * Report EH to user.
2512 *
2513 * LOCKING:
2514 * None.
2515 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002516void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002517{
2518 struct ata_link *link;
2519
Tejun Heo1eca4362008-11-03 20:03:17 +09002520 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002521 ata_eh_link_report(link);
2522}
2523
Tejun Heocc0680a2007-08-06 18:36:23 +09002524static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heob1c72912008-07-31 17:02:43 +09002525 unsigned int *classes, unsigned long deadline,
2526 bool clear_classes)
Tejun Heod87fa382006-05-31 18:28:24 +09002527{
Tejun Heof58229f2007-08-06 18:36:23 +09002528 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002529
Tejun Heob1c72912008-07-31 17:02:43 +09002530 if (clear_classes)
Tejun Heo1eca4362008-11-03 20:03:17 +09002531 ata_for_each_dev(dev, link, ALL)
Tejun Heob1c72912008-07-31 17:02:43 +09002532 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002533
Tejun Heof0465192008-05-19 01:15:08 +09002534 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002535}
2536
Sergei Shtylyove8411fb2011-08-09 23:25:50 +04002537static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
Tejun Heo664faf02006-05-31 18:27:50 +09002538{
Tejun Heo45db2f62008-04-08 01:46:56 +09002539 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002540 return 0;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002541 if (rc == -EAGAIN)
2542 return 1;
Tejun Heo071f44b2008-04-07 22:47:22 +09002543 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002544 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002545 return 0;
2546}
2547
Tejun Heofb7fd612007-09-23 13:14:12 +09002548int ata_eh_reset(struct ata_link *link, int classify,
2549 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2550 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002551{
Tejun Heoafaa5c32007-10-09 15:06:10 +09002552 struct ata_port *ap = link->ap;
Tejun Heob1c72912008-07-31 17:02:43 +09002553 struct ata_link *slave = ap->slave_link;
Tejun Heo936fd732007-08-06 18:36:23 +09002554 struct ata_eh_context *ehc = &link->eh_context;
Bartlomiej Zolnierkiewicz705d2012009-07-26 16:21:01 +02002555 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
Tejun Heo664faf02006-05-31 18:27:50 +09002556 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002557 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002558 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heod8af0eb2008-05-20 02:17:53 +09002559 int max_tries = 0, try = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002560 struct ata_link *failed_link;
Tejun Heof58229f2007-08-06 18:36:23 +09002561 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002562 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002563 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002564 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002565 u32 sstatus;
Tejun Heob1c72912008-07-31 17:02:43 +09002566 int nr_unknown, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002567
Tejun Heo932648b2008-05-19 01:15:06 +09002568 /*
2569 * Prepare to reset
2570 */
Tejun Heod8af0eb2008-05-20 02:17:53 +09002571 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2572 max_tries++;
Tejun Heo05944bd2008-08-13 20:19:09 +09002573 if (link->flags & ATA_LFLAG_NO_HRST)
2574 hardreset = NULL;
2575 if (link->flags & ATA_LFLAG_NO_SRST)
2576 softreset = NULL;
Tejun Heod8af0eb2008-05-20 02:17:53 +09002577
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002578 /* make sure each reset attempt is at least COOL_DOWN apart */
Tejun Heo19b72322008-11-04 17:08:40 +09002579 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2580 now = jiffies;
2581 WARN_ON(time_after(ehc->last_reset, now));
2582 deadline = ata_deadline(ehc->last_reset,
2583 ATA_EH_RESET_COOL_DOWN);
2584 if (time_before(now, deadline))
2585 schedule_timeout_uninterruptible(deadline - now);
2586 }
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002587
Tejun Heoafaa5c32007-10-09 15:06:10 +09002588 spin_lock_irqsave(ap->lock, flags);
2589 ap->pflags |= ATA_PFLAG_RESETTING;
2590 spin_unlock_irqrestore(ap->lock, flags);
2591
Tejun Heocf480622008-01-24 00:05:14 +09002592 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002593
Tejun Heo1eca4362008-11-03 20:03:17 +09002594 ata_for_each_dev(dev, link, ALL) {
Tejun Heocdeab112007-10-29 16:41:09 +09002595 /* If we issue an SRST then an ATA drive (not ATAPI)
2596 * may change configuration and be in PIO0 timing. If
2597 * we do a hard reset (or are coming from power on)
2598 * this is true for ATA or ATAPI. Until we've set a
2599 * suitable controller mode we should not touch the
2600 * bus as we may be talking too fast.
2601 */
2602 dev->pio_mode = XFER_PIO_0;
2603
2604 /* If the controller has a pio mode setup function
2605 * then use it to set the chipset to rights. Don't
2606 * touch the DMA setup as that will be dealt with when
2607 * configuring devices.
2608 */
2609 if (ap->ops->set_piomode)
2610 ap->ops->set_piomode(ap, dev);
2611 }
2612
Tejun Heocf480622008-01-24 00:05:14 +09002613 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002614 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002615 ehc->i.action &= ~ATA_EH_RESET;
2616 if (hardreset) {
2617 reset = hardreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002618 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002619 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002620 reset = softreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002621 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heocf480622008-01-24 00:05:14 +09002622 }
Tejun Heof5914a42006-05-31 18:27:48 +09002623
2624 if (prereset) {
Tejun Heob1c72912008-07-31 17:02:43 +09002625 unsigned long deadline = ata_deadline(jiffies,
2626 ATA_EH_PRERESET_TIMEOUT);
2627
2628 if (slave) {
2629 sehc->i.action &= ~ATA_EH_RESET;
2630 sehc->i.action |= ehc->i.action;
2631 }
2632
2633 rc = prereset(link, deadline);
2634
2635 /* If present, do prereset on slave link too. Reset
2636 * is skipped iff both master and slave links report
2637 * -ENOENT or clear ATA_EH_RESET.
2638 */
2639 if (slave && (rc == 0 || rc == -ENOENT)) {
2640 int tmp;
2641
2642 tmp = prereset(slave, deadline);
2643 if (tmp != -ENOENT)
2644 rc = tmp;
2645
2646 ehc->i.action |= sehc->i.action;
2647 }
2648
Tejun Heof5914a42006-05-31 18:27:48 +09002649 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002650 if (rc == -ENOENT) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002651 ata_link_dbg(link, "port disabled--ignoring\n");
Tejun Heocf480622008-01-24 00:05:14 +09002652 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002653
Tejun Heo1eca4362008-11-03 20:03:17 +09002654 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09002655 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002656
2657 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002658 } else
Joe Perchesa9a79df2011-04-15 15:51:59 -07002659 ata_link_err(link,
2660 "prereset failed (errno=%d)\n",
2661 rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002662 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002663 }
Tejun Heof5914a42006-05-31 18:27:48 +09002664
Tejun Heo932648b2008-05-19 01:15:06 +09002665 /* prereset() might have cleared ATA_EH_RESET. If so,
Tejun Heod6515e62009-03-04 15:59:30 +09002666 * bang classes, thaw and return.
Tejun Heo932648b2008-05-19 01:15:06 +09002667 */
2668 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
Tejun Heo1eca4362008-11-03 20:03:17 +09002669 ata_for_each_dev(dev, link, ALL)
Tejun Heo932648b2008-05-19 01:15:06 +09002670 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heod6515e62009-03-04 15:59:30 +09002671 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2672 ata_is_host_link(link))
2673 ata_eh_thaw_port(ap);
Tejun Heo932648b2008-05-19 01:15:06 +09002674 rc = 0;
2675 goto out;
2676 }
Tejun Heof5914a42006-05-31 18:27:48 +09002677 }
2678
Tejun Heo022bdb02006-05-15 20:58:22 +09002679 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002680 /*
2681 * Perform reset
2682 */
Tejun Heodc98c322008-05-19 01:15:07 +09002683 if (ata_is_host_link(link))
2684 ata_eh_freeze_port(ap);
2685
Tejun Heo341c2c92008-05-20 02:17:51 +09002686 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
Tejun Heo31daabd2007-02-02 16:50:52 +09002687
Tejun Heo932648b2008-05-19 01:15:06 +09002688 if (reset) {
2689 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002690 ata_link_info(link, "%s resetting link\n",
2691 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002692
Tejun Heo932648b2008-05-19 01:15:06 +09002693 /* mark that this EH session started with reset */
Tejun Heo19b72322008-11-04 17:08:40 +09002694 ehc->last_reset = jiffies;
Tejun Heo932648b2008-05-19 01:15:06 +09002695 if (reset == hardreset)
2696 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2697 else
2698 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002699
Tejun Heob1c72912008-07-31 17:02:43 +09002700 rc = ata_do_reset(link, reset, classes, deadline, true);
2701 if (rc && rc != -EAGAIN) {
2702 failed_link = link;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002703 goto fail;
Tejun Heob1c72912008-07-31 17:02:43 +09002704 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002705
Tejun Heob1c72912008-07-31 17:02:43 +09002706 /* hardreset slave link if existent */
2707 if (slave && reset == hardreset) {
2708 int tmp;
2709
2710 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002711 ata_link_info(slave, "hard resetting link\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002712
2713 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2714 tmp = ata_do_reset(slave, reset, classes, deadline,
2715 false);
2716 switch (tmp) {
2717 case -EAGAIN:
2718 rc = -EAGAIN;
2719 case 0:
2720 break;
2721 default:
2722 failed_link = slave;
2723 rc = tmp;
2724 goto fail;
2725 }
2726 }
2727
2728 /* perform follow-up SRST if necessary */
Tejun Heo932648b2008-05-19 01:15:06 +09002729 if (reset == hardreset &&
Sergei Shtylyove8411fb2011-08-09 23:25:50 +04002730 ata_eh_followup_srst_needed(link, rc)) {
Tejun Heo932648b2008-05-19 01:15:06 +09002731 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002732
Tejun Heo932648b2008-05-19 01:15:06 +09002733 if (!reset) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002734 ata_link_err(link,
2735 "follow-up softreset required but no softreset available\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002736 failed_link = link;
Tejun Heo932648b2008-05-19 01:15:06 +09002737 rc = -EINVAL;
2738 goto fail;
2739 }
2740
2741 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002742 rc = ata_do_reset(link, reset, classes, deadline, true);
Tejun Heofe2c4d02009-07-08 12:16:37 +09002743 if (rc) {
2744 failed_link = link;
2745 goto fail;
2746 }
Tejun Heo664faf02006-05-31 18:27:50 +09002747 }
Tejun Heo932648b2008-05-19 01:15:06 +09002748 } else {
2749 if (verbose)
Joe Perchesa9a79df2011-04-15 15:51:59 -07002750 ata_link_info(link,
2751 "no reset method available, skipping reset\n");
Tejun Heo932648b2008-05-19 01:15:06 +09002752 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2753 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002754 }
2755
Tejun Heo932648b2008-05-19 01:15:06 +09002756 /*
2757 * Post-reset processing
2758 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002759 ata_for_each_dev(dev, link, ALL) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002760 /* After the reset, the device state is PIO 0 and the
2761 * controller state is undefined. Reset also wakes up
2762 * drives from sleeping mode.
2763 */
2764 dev->pio_mode = XFER_PIO_0;
2765 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002766
Tejun Heo3b761d32009-10-06 17:08:40 +09002767 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2768 continue;
2769
2770 /* apply class override */
2771 if (lflags & ATA_LFLAG_ASSUME_ATA)
2772 classes[dev->devno] = ATA_DEV_ATA;
2773 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2774 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
Tejun Heo022bdb02006-05-15 20:58:22 +09002775 }
2776
Tejun Heo416dc9e2007-10-31 10:17:03 +09002777 /* record current link speed */
2778 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2779 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heob1c72912008-07-31 17:02:43 +09002780 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2781 slave->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002782
Tejun Heodc98c322008-05-19 01:15:07 +09002783 /* thaw the port */
2784 if (ata_is_host_link(link))
2785 ata_eh_thaw_port(ap);
2786
Tejun Heof0465192008-05-19 01:15:08 +09002787 /* postreset() should clear hardware SError. Although SError
2788 * is cleared during link resume, clearing SError here is
2789 * necessary as some PHYs raise hotplug events after SRST.
2790 * This introduces race condition where hotplug occurs between
2791 * reset and here. This race is mediated by cross checking
2792 * link onlineness and classification result later.
2793 */
Tejun Heob1c72912008-07-31 17:02:43 +09002794 if (postreset) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002795 postreset(link, classes);
Tejun Heob1c72912008-07-31 17:02:43 +09002796 if (slave)
2797 postreset(slave, classes);
2798 }
Tejun Heo20952b62006-05-31 18:27:23 +09002799
Tejun Heo1e641062009-07-17 11:58:33 +09002800 /*
Tejun Heo8c56cac2011-05-25 13:19:39 +02002801 * Some controllers can't be frozen very well and may set spurious
2802 * error conditions during reset. Clear accumulated error
2803 * information and re-thaw the port if frozen. As reset is the
2804 * final recovery action and we cross check link onlineness against
2805 * device classification later, no hotplug event is lost by this.
Tejun Heo1e641062009-07-17 11:58:33 +09002806 */
Tejun Heof0465192008-05-19 01:15:08 +09002807 spin_lock_irqsave(link->ap->lock, flags);
Tejun Heo1e641062009-07-17 11:58:33 +09002808 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heob1c72912008-07-31 17:02:43 +09002809 if (slave)
Tejun Heo1e641062009-07-17 11:58:33 +09002810 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2811 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof0465192008-05-19 01:15:08 +09002812 spin_unlock_irqrestore(link->ap->lock, flags);
2813
Tejun Heo8c56cac2011-05-25 13:19:39 +02002814 if (ap->pflags & ATA_PFLAG_FROZEN)
2815 ata_eh_thaw_port(ap);
2816
Tejun Heo3b761d32009-10-06 17:08:40 +09002817 /*
2818 * Make sure onlineness and classification result correspond.
Tejun Heof0465192008-05-19 01:15:08 +09002819 * Hotplug could have happened during reset and some
2820 * controllers fail to wait while a drive is spinning up after
2821 * being hotplugged causing misdetection. By cross checking
Tejun Heo3b761d32009-10-06 17:08:40 +09002822 * link on/offlineness and classification result, those
2823 * conditions can be reliably detected and retried.
Tejun Heof0465192008-05-19 01:15:08 +09002824 */
Tejun Heob1c72912008-07-31 17:02:43 +09002825 nr_unknown = 0;
Tejun Heo1eca4362008-11-03 20:03:17 +09002826 ata_for_each_dev(dev, link, ALL) {
Tejun Heo3b761d32009-10-06 17:08:40 +09002827 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2828 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002829 ata_dev_dbg(dev, "link online but device misclassified\n");
Tejun Heo3b761d32009-10-06 17:08:40 +09002830 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002831 nr_unknown++;
Tejun Heo3b761d32009-10-06 17:08:40 +09002832 }
2833 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2834 if (ata_class_enabled(classes[dev->devno]))
Joe Perchesa9a79df2011-04-15 15:51:59 -07002835 ata_dev_dbg(dev,
2836 "link offline, clearing class %d to NONE\n",
2837 classes[dev->devno]);
Tejun Heo3b761d32009-10-06 17:08:40 +09002838 classes[dev->devno] = ATA_DEV_NONE;
2839 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002840 ata_dev_dbg(dev,
2841 "link status unknown, clearing UNKNOWN to NONE\n");
Tejun Heo3b761d32009-10-06 17:08:40 +09002842 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002843 }
Tejun Heof0465192008-05-19 01:15:08 +09002844 }
2845
Tejun Heob1c72912008-07-31 17:02:43 +09002846 if (classify && nr_unknown) {
Tejun Heof0465192008-05-19 01:15:08 +09002847 if (try < max_tries) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07002848 ata_link_warn(link,
2849 "link online but %d devices misclassified, retrying\n",
2850 nr_unknown);
Tejun Heob1c72912008-07-31 17:02:43 +09002851 failed_link = link;
Tejun Heof0465192008-05-19 01:15:08 +09002852 rc = -EAGAIN;
2853 goto fail;
2854 }
Joe Perchesa9a79df2011-04-15 15:51:59 -07002855 ata_link_warn(link,
2856 "link online but %d devices misclassified, "
2857 "device detection might fail\n", nr_unknown);
Tejun Heof0465192008-05-19 01:15:08 +09002858 }
2859
Tejun Heo416dc9e2007-10-31 10:17:03 +09002860 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002861 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002862 if (slave)
2863 ata_eh_done(slave, NULL, ATA_EH_RESET);
Tejun Heo6b7ae952010-09-01 17:50:06 +02002864 ehc->last_reset = jiffies; /* update to completion time */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002865 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo6b7ae952010-09-01 17:50:06 +02002866 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
Tejun Heoae791c02007-09-23 13:14:12 +09002867
Tejun Heo416dc9e2007-10-31 10:17:03 +09002868 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002869 out:
2870 /* clear hotplug flag */
2871 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heob1c72912008-07-31 17:02:43 +09002872 if (slave)
2873 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002874
2875 spin_lock_irqsave(ap->lock, flags);
2876 ap->pflags &= ~ATA_PFLAG_RESETTING;
2877 spin_unlock_irqrestore(ap->lock, flags);
2878
Tejun Heo022bdb02006-05-15 20:58:22 +09002879 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002880
2881 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002882 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2883 if (!ata_is_host_link(link) &&
2884 sata_scr_read(link, SCR_STATUS, &sstatus))
2885 rc = -ERESTART;
2886
Gwendal Grignou7a46c072011-10-19 17:17:02 -07002887 if (try >= max_tries) {
Tejun Heo8ea76452011-05-25 12:23:56 +02002888 /*
2889 * Thaw host port even if reset failed, so that the port
2890 * can be retried on the next phy event. This risks
2891 * repeated EH runs but seems to be a better tradeoff than
2892 * shutting down a port after a botched hotplug attempt.
2893 */
2894 if (ata_is_host_link(link))
2895 ata_eh_thaw_port(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002896 goto out;
Tejun Heo8ea76452011-05-25 12:23:56 +02002897 }
Tejun Heo416dc9e2007-10-31 10:17:03 +09002898
2899 now = jiffies;
2900 if (time_before(now, deadline)) {
2901 unsigned long delta = deadline - now;
2902
Joe Perchesa9a79df2011-04-15 15:51:59 -07002903 ata_link_warn(failed_link,
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002904 "reset failed (errno=%d), retrying in %u secs\n",
2905 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
Tejun Heo416dc9e2007-10-31 10:17:03 +09002906
Tejun Heoc0c362b2010-09-06 17:57:14 +02002907 ata_eh_release(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002908 while (delta)
2909 delta = schedule_timeout_uninterruptible(delta);
Tejun Heoc0c362b2010-09-06 17:57:14 +02002910 ata_eh_acquire(ap);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002911 }
2912
Gwendal Grignou7a46c072011-10-19 17:17:02 -07002913 /*
2914 * While disks spinup behind PMP, some controllers fail sending SRST.
2915 * They need to be reset - as well as the PMP - before retrying.
2916 */
2917 if (rc == -ERESTART) {
2918 if (ata_is_host_link(link))
2919 ata_eh_thaw_port(ap);
2920 goto out;
2921 }
2922
Tejun Heob1c72912008-07-31 17:02:43 +09002923 if (try == max_tries - 1) {
Tejun Heoa07d4992009-01-29 20:31:33 +09002924 sata_down_spd_limit(link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002925 if (slave)
Tejun Heoa07d4992009-01-29 20:31:33 +09002926 sata_down_spd_limit(slave, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002927 } else if (rc == -EPIPE)
Tejun Heoa07d4992009-01-29 20:31:33 +09002928 sata_down_spd_limit(failed_link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002929
Tejun Heo416dc9e2007-10-31 10:17:03 +09002930 if (hardreset)
2931 reset = hardreset;
2932 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002933}
2934
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002935static inline void ata_eh_pull_park_action(struct ata_port *ap)
2936{
2937 struct ata_link *link;
2938 struct ata_device *dev;
2939 unsigned long flags;
2940
2941 /*
2942 * This function can be thought of as an extended version of
2943 * ata_eh_about_to_do() specially crafted to accommodate the
2944 * requirements of ATA_EH_PARK handling. Since the EH thread
2945 * does not leave the do {} while () loop in ata_eh_recover as
2946 * long as the timeout for a park request to *one* device on
2947 * the port has not expired, and since we still want to pick
2948 * up park requests to other devices on the same port or
2949 * timeout updates for the same device, we have to pull
2950 * ATA_EH_PARK actions from eh_info into eh_context.i
2951 * ourselves at the beginning of each pass over the loop.
2952 *
2953 * Additionally, all write accesses to &ap->park_req_pending
2954 * through INIT_COMPLETION() (see below) or complete_all()
2955 * (see ata_scsi_park_store()) are protected by the host lock.
2956 * As a result we have that park_req_pending.done is zero on
2957 * exit from this function, i.e. when ATA_EH_PARK actions for
2958 * *all* devices on port ap have been pulled into the
2959 * respective eh_context structs. If, and only if,
2960 * park_req_pending.done is non-zero by the time we reach
2961 * wait_for_completion_timeout(), another ATA_EH_PARK action
2962 * has been scheduled for at least one of the devices on port
2963 * ap and we have to cycle over the do {} while () loop in
2964 * ata_eh_recover() again.
2965 */
2966
2967 spin_lock_irqsave(ap->lock, flags);
2968 INIT_COMPLETION(ap->park_req_pending);
Tejun Heo1eca4362008-11-03 20:03:17 +09002969 ata_for_each_link(link, ap, EDGE) {
2970 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002971 struct ata_eh_info *ehi = &link->eh_info;
2972
2973 link->eh_context.i.dev_action[dev->devno] |=
2974 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2975 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2976 }
2977 }
2978 spin_unlock_irqrestore(ap->lock, flags);
2979}
2980
2981static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2982{
2983 struct ata_eh_context *ehc = &dev->link->eh_context;
2984 struct ata_taskfile tf;
2985 unsigned int err_mask;
2986
2987 ata_tf_init(dev, &tf);
2988 if (park) {
2989 ehc->unloaded_mask |= 1 << dev->devno;
2990 tf.command = ATA_CMD_IDLEIMMEDIATE;
2991 tf.feature = 0x44;
2992 tf.lbal = 0x4c;
2993 tf.lbam = 0x4e;
2994 tf.lbah = 0x55;
2995 } else {
2996 ehc->unloaded_mask &= ~(1 << dev->devno);
2997 tf.command = ATA_CMD_CHK_POWER;
2998 }
2999
3000 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3001 tf.protocol |= ATA_PROT_NODATA;
3002 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3003 if (park && (err_mask || tf.lbal != 0xc4)) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003004 ata_dev_err(dev, "head unload failed!\n");
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003005 ehc->unloaded_mask &= ~(1 << dev->devno);
3006 }
3007}
3008
Tejun Heo02607312007-08-06 18:36:23 +09003009static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09003010 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09003011{
Tejun Heo02607312007-08-06 18:36:23 +09003012 struct ata_port *ap = link->ap;
3013 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003014 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003015 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003016 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09003017 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003018
3019 DPRINTK("ENTER\n");
3020
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003021 /* For PATA drive side cable detection to work, IDENTIFY must
3022 * be done backwards such that PDIAG- is released by the slave
3023 * device before the master device is identified.
3024 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003025 ata_for_each_dev(dev, link, ALL_REVERSE) {
Tejun Heof58229f2007-08-06 18:36:23 +09003026 unsigned int action = ata_eh_dev_action(dev);
3027 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09003028
Tejun Heobff04642006-11-10 18:08:10 +09003029 if (ehc->i.flags & ATA_EHI_DID_RESET)
3030 readid_flags |= ATA_READID_POSTRESET;
3031
Tejun Heo9666f402007-05-04 21:27:47 +02003032 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a32007-09-23 13:19:54 +09003033 WARN_ON(dev->class == ATA_DEV_PMP);
3034
Tejun Heob1c72912008-07-31 17:02:43 +09003035 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
Tejun Heo022bdb02006-05-15 20:58:22 +09003036 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003037 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003038 }
3039
Tejun Heo02607312007-08-06 18:36:23 +09003040 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09003041 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3042 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09003043 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003044 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09003045
Tejun Heo02607312007-08-06 18:36:23 +09003046 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09003047
Tejun Heobaa1e782006-11-01 18:39:27 +09003048 /* Configuration may have changed, reconfigure
3049 * transfer mode.
3050 */
3051 ehc->i.flags |= ATA_EHI_SETMODE;
3052
zhao, forrest3057ac32006-06-12 12:01:34 +08003053 /* schedule the scsi_rescan_device() here */
Tejun Heoad72cf92010-07-02 10:03:52 +02003054 schedule_work(&(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09003055 } else if (dev->class == ATA_DEV_UNKNOWN &&
3056 ehc->tries[dev->devno] &&
3057 ata_class_enabled(ehc->classes[dev->devno])) {
Tejun Heo842faa62009-05-10 01:06:54 +09003058 /* Temporarily set dev->class, it will be
3059 * permanently set once all configurations are
3060 * complete. This is necessary because new
3061 * device configuration is done in two
3062 * separate loops.
3063 */
Tejun Heo084fe632006-05-31 18:28:03 +09003064 dev->class = ehc->classes[dev->devno];
3065
Tejun Heo633273a32007-09-23 13:19:54 +09003066 if (dev->class == ATA_DEV_PMP)
3067 rc = sata_pmp_attach(dev);
3068 else
3069 rc = ata_dev_read_id(dev, &dev->class,
3070 readid_flags, dev->id);
Tejun Heo842faa62009-05-10 01:06:54 +09003071
3072 /* read_id might have changed class, store and reset */
3073 ehc->classes[dev->devno] = dev->class;
3074 dev->class = ATA_DEV_UNKNOWN;
3075
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003076 switch (rc) {
3077 case 0:
Tejun Heo99cf6102009-01-29 20:31:32 +09003078 /* clear error info accumulated during probe */
3079 ata_ering_clear(&dev->ering);
Tejun Heof58229f2007-08-06 18:36:23 +09003080 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003081 break;
3082 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003083 /* IDENTIFY was issued to non-existent
3084 * device. No need to reset. Just
Tejun Heo842faa62009-05-10 01:06:54 +09003085 * thaw and ignore the device.
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003086 */
3087 ata_eh_thaw_port(ap);
Tejun Heo084fe632006-05-31 18:28:03 +09003088 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003089 default:
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003090 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09003091 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003092 }
3093 }
3094
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003095 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09003096 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3097 if (ap->ops->cable_detect)
3098 ap->cbl = ap->ops->cable_detect(ap);
3099 ata_force_cbl(ap);
3100 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09003101
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003102 /* Configure new devices forward such that user doesn't see
3103 * device detection messages backwards.
3104 */
Tejun Heo1eca4362008-11-03 20:03:17 +09003105 ata_for_each_dev(dev, link, ALL) {
Tejun Heo4f7c2872009-10-15 23:37:32 +09003106 if (!(new_mask & (1 << dev->devno)))
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003107 continue;
3108
Tejun Heo842faa62009-05-10 01:06:54 +09003109 dev->class = ehc->classes[dev->devno];
3110
Tejun Heo4f7c2872009-10-15 23:37:32 +09003111 if (dev->class == ATA_DEV_PMP)
3112 continue;
3113
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003114 ehc->i.flags |= ATA_EHI_PRINTINFO;
3115 rc = ata_dev_configure(dev);
3116 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
Tejun Heo842faa62009-05-10 01:06:54 +09003117 if (rc) {
3118 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003119 goto err;
Tejun Heo842faa62009-05-10 01:06:54 +09003120 }
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003121
3122 spin_lock_irqsave(ap->lock, flags);
3123 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3124 spin_unlock_irqrestore(ap->lock, flags);
3125
3126 /* new device discovered, configure xfermode */
3127 ehc->i.flags |= ATA_EHI_SETMODE;
3128 }
3129
3130 return 0;
3131
3132 err:
3133 *r_failed_dev = dev;
3134 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003135 return rc;
3136}
3137
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003138/**
3139 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3140 * @link: link on which timings will be programmed
Martin Olsson98a17082009-04-22 18:21:29 +02003141 * @r_failed_dev: out parameter for failed device
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003142 *
3143 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3144 * ata_set_mode() fails, pointer to the failing device is
3145 * returned in @r_failed_dev.
3146 *
3147 * LOCKING:
3148 * PCI/etc. bus probe sem.
3149 *
3150 * RETURNS:
3151 * 0 on success, negative errno otherwise
3152 */
3153int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3154{
3155 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09003156 struct ata_device *dev;
3157 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003158
Tejun Heo76326ac2007-11-27 19:28:59 +09003159 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
Tejun Heo1eca4362008-11-03 20:03:17 +09003160 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo76326ac2007-11-27 19:28:59 +09003161 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3162 struct ata_ering_entry *ent;
3163
3164 ent = ata_ering_top(&dev->ering);
3165 if (ent)
3166 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3167 }
3168 }
3169
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003170 /* has private set_mode? */
3171 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09003172 rc = ap->ops->set_mode(link, r_failed_dev);
3173 else
3174 rc = ata_do_set_mode(link, r_failed_dev);
3175
3176 /* if transfer mode has changed, set DUBIOUS_XFER on device */
Tejun Heo1eca4362008-11-03 20:03:17 +09003177 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +09003178 struct ata_eh_context *ehc = &link->eh_context;
3179 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3180 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3181
3182 if (dev->xfer_mode != saved_xfer_mode ||
3183 ata_ncq_enabled(dev) != saved_ncq)
3184 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3185 }
3186
3187 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003188}
3189
Tejun Heo11fc33d2008-08-30 14:20:01 +02003190/**
3191 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3192 * @dev: ATAPI device to clear UA for
3193 *
3194 * Resets and other operations can make an ATAPI device raise
3195 * UNIT ATTENTION which causes the next operation to fail. This
3196 * function clears UA.
3197 *
3198 * LOCKING:
3199 * EH context (may sleep).
3200 *
3201 * RETURNS:
3202 * 0 on success, -errno on failure.
3203 */
3204static int atapi_eh_clear_ua(struct ata_device *dev)
3205{
3206 int i;
3207
3208 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
Tejun Heob5357082009-03-02 18:55:16 +09003209 u8 *sense_buffer = dev->link->ap->sector_buf;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003210 u8 sense_key = 0;
3211 unsigned int err_mask;
3212
3213 err_mask = atapi_eh_tur(dev, &sense_key);
3214 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003215 ata_dev_warn(dev,
3216 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3217 err_mask);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003218 return -EIO;
3219 }
3220
3221 if (!err_mask || sense_key != UNIT_ATTENTION)
3222 return 0;
3223
3224 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3225 if (err_mask) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003226 ata_dev_warn(dev, "failed to clear "
Tejun Heo11fc33d2008-08-30 14:20:01 +02003227 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3228 return -EIO;
3229 }
3230 }
3231
Joe Perchesa9a79df2011-04-15 15:51:59 -07003232 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3233 ATA_EH_UA_TRIES);
Tejun Heo11fc33d2008-08-30 14:20:01 +02003234
3235 return 0;
3236}
3237
Tejun Heo6013efd2009-11-19 15:36:45 +09003238/**
3239 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3240 * @dev: ATA device which may need FLUSH retry
3241 *
3242 * If @dev failed FLUSH, it needs to be reported upper layer
3243 * immediately as it means that @dev failed to remap and already
3244 * lost at least a sector and further FLUSH retrials won't make
3245 * any difference to the lost sector. However, if FLUSH failed
3246 * for other reasons, for example transmission error, FLUSH needs
3247 * to be retried.
3248 *
3249 * This function determines whether FLUSH failure retry is
3250 * necessary and performs it if so.
3251 *
3252 * RETURNS:
3253 * 0 if EH can continue, -errno if EH needs to be repeated.
3254 */
3255static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3256{
3257 struct ata_link *link = dev->link;
3258 struct ata_port *ap = link->ap;
3259 struct ata_queued_cmd *qc;
3260 struct ata_taskfile tf;
3261 unsigned int err_mask;
3262 int rc = 0;
3263
3264 /* did flush fail for this device? */
3265 if (!ata_tag_valid(link->active_tag))
3266 return 0;
3267
3268 qc = __ata_qc_from_tag(ap, link->active_tag);
3269 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3270 qc->tf.command != ATA_CMD_FLUSH))
3271 return 0;
3272
3273 /* if the device failed it, it should be reported to upper layers */
3274 if (qc->err_mask & AC_ERR_DEV)
3275 return 0;
3276
3277 /* flush failed for some other reason, give it another shot */
3278 ata_tf_init(dev, &tf);
3279
3280 tf.command = qc->tf.command;
3281 tf.flags |= ATA_TFLAG_DEVICE;
3282 tf.protocol = ATA_PROT_NODATA;
3283
Joe Perchesa9a79df2011-04-15 15:51:59 -07003284 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
Tejun Heo6013efd2009-11-19 15:36:45 +09003285 tf.command, qc->err_mask);
3286
3287 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3288 if (!err_mask) {
3289 /*
3290 * FLUSH is complete but there's no way to
3291 * successfully complete a failed command from EH.
3292 * Making sure retry is allowed at least once and
3293 * retrying it should do the trick - whatever was in
3294 * the cache is already on the platter and this won't
3295 * cause infinite loop.
3296 */
3297 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3298 } else {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003299 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
Tejun Heo6013efd2009-11-19 15:36:45 +09003300 err_mask);
3301 rc = -EIO;
3302
3303 /* if device failed it, report it to upper layers */
3304 if (err_mask & AC_ERR_DEV) {
3305 qc->err_mask |= AC_ERR_DEV;
3306 qc->result_tf = tf;
3307 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3308 rc = 0;
3309 }
3310 }
3311 return rc;
3312}
3313
Tejun Heo6b7ae952010-09-01 17:50:06 +02003314/**
3315 * ata_eh_set_lpm - configure SATA interface power management
3316 * @link: link to configure power management
3317 * @policy: the link power management policy
3318 * @r_failed_dev: out parameter for failed device
3319 *
3320 * Enable SATA Interface power management. This will enable
3321 * Device Interface Power Management (DIPM) for min_power
3322 * policy, and then call driver specific callbacks for
3323 * enabling Host Initiated Power management.
3324 *
3325 * LOCKING:
3326 * EH context.
3327 *
3328 * RETURNS:
3329 * 0 on success, -errno on failure.
3330 */
3331static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3332 struct ata_device **r_failed_dev)
3333{
Tejun Heo6c8ea892010-09-01 17:50:07 +02003334 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003335 struct ata_eh_context *ehc = &link->eh_context;
3336 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
Tejun Heoe5005b12010-12-09 16:13:15 +01003337 enum ata_lpm_policy old_policy = link->lpm_policy;
Tejun Heo5f6f12c2011-05-09 16:04:11 +02003338 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003339 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3340 unsigned int err_mask;
3341 int rc;
3342
3343 /* if the link or host doesn't do LPM, noop */
3344 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3345 return 0;
3346
3347 /*
3348 * DIPM is enabled only for MIN_POWER as some devices
3349 * misbehave when the host NACKs transition to SLUMBER. Order
3350 * device and link configurations such that the host always
3351 * allows DIPM requests.
3352 */
3353 ata_for_each_dev(dev, link, ENABLED) {
3354 bool hipm = ata_id_has_hipm(dev->id);
Tejun Heoae01b242011-03-16 11:14:55 +01003355 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003356
3357 /* find the first enabled and LPM enabled devices */
3358 if (!link_dev)
3359 link_dev = dev;
3360
3361 if (!lpm_dev && (hipm || dipm))
3362 lpm_dev = dev;
3363
3364 hints &= ~ATA_LPM_EMPTY;
3365 if (!hipm)
3366 hints &= ~ATA_LPM_HIPM;
3367
3368 /* disable DIPM before changing link config */
3369 if (policy != ATA_LPM_MIN_POWER && dipm) {
3370 err_mask = ata_dev_set_feature(dev,
3371 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3372 if (err_mask && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003373 ata_dev_warn(dev,
3374 "failed to disable DIPM, Emask 0x%x\n",
3375 err_mask);
Tejun Heo6b7ae952010-09-01 17:50:06 +02003376 rc = -EIO;
3377 goto fail;
3378 }
3379 }
3380 }
3381
Tejun Heo6c8ea892010-09-01 17:50:07 +02003382 if (ap) {
3383 rc = ap->ops->set_lpm(link, policy, hints);
3384 if (!rc && ap->slave_link)
3385 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3386 } else
3387 rc = sata_pmp_set_lpm(link, policy, hints);
Tejun Heo6b7ae952010-09-01 17:50:06 +02003388
3389 /*
3390 * Attribute link config failure to the first (LPM) enabled
3391 * device on the link.
3392 */
3393 if (rc) {
3394 if (rc == -EOPNOTSUPP) {
3395 link->flags |= ATA_LFLAG_NO_LPM;
3396 return 0;
3397 }
3398 dev = lpm_dev ? lpm_dev : link_dev;
3399 goto fail;
3400 }
3401
Tejun Heoe5005b12010-12-09 16:13:15 +01003402 /*
3403 * Low level driver acked the transition. Issue DIPM command
3404 * with the new policy set.
3405 */
3406 link->lpm_policy = policy;
3407 if (ap && ap->slave_link)
3408 ap->slave_link->lpm_policy = policy;
3409
Tejun Heo6b7ae952010-09-01 17:50:06 +02003410 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3411 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heoae01b242011-03-16 11:14:55 +01003412 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3413 ata_id_has_dipm(dev->id)) {
Tejun Heo6b7ae952010-09-01 17:50:06 +02003414 err_mask = ata_dev_set_feature(dev,
3415 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3416 if (err_mask && err_mask != AC_ERR_DEV) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003417 ata_dev_warn(dev,
Tejun Heo6b7ae952010-09-01 17:50:06 +02003418 "failed to enable DIPM, Emask 0x%x\n",
3419 err_mask);
3420 rc = -EIO;
3421 goto fail;
3422 }
3423 }
3424 }
3425
Tejun Heo6b7ae952010-09-01 17:50:06 +02003426 return 0;
3427
3428fail:
Tejun Heoe5005b12010-12-09 16:13:15 +01003429 /* restore the old policy */
3430 link->lpm_policy = old_policy;
3431 if (ap && ap->slave_link)
3432 ap->slave_link->lpm_policy = old_policy;
3433
Tejun Heo6b7ae952010-09-01 17:50:06 +02003434 /* if no device or only one more chance is left, disable LPM */
3435 if (!dev || ehc->tries[dev->devno] <= 2) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003436 ata_link_warn(link, "disabling LPM on the link\n");
Tejun Heo6b7ae952010-09-01 17:50:06 +02003437 link->flags |= ATA_LFLAG_NO_LPM;
3438 }
3439 if (r_failed_dev)
3440 *r_failed_dev = dev;
3441 return rc;
3442}
3443
Kristen Carlson Accardi8a745f12011-03-04 10:24:11 -08003444int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003445{
Tejun Heof58229f2007-08-06 18:36:23 +09003446 struct ata_device *dev;
3447 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003448
Tejun Heo1eca4362008-11-03 20:03:17 +09003449 ata_for_each_dev(dev, link, ENABLED)
3450 cnt++;
Tejun Heo022bdb02006-05-15 20:58:22 +09003451 return cnt;
3452}
3453
Tejun Heo02607312007-08-06 18:36:23 +09003454static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003455{
Tejun Heof58229f2007-08-06 18:36:23 +09003456 struct ata_device *dev;
3457 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003458
Tejun Heo1eca4362008-11-03 20:03:17 +09003459 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09003460 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09003461 cnt++;
3462 return cnt;
3463}
3464
Tejun Heo02607312007-08-06 18:36:23 +09003465static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003466{
Tejun Heo672b2d62008-01-24 00:05:14 +09003467 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09003468 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09003469 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09003470
Tejun Heof9df58c2007-09-23 13:14:13 +09003471 /* skip disabled links */
3472 if (link->flags & ATA_LFLAG_DISABLED)
3473 return 1;
3474
Tejun Heoe2f3d752010-09-07 14:05:31 +02003475 /* skip if explicitly requested */
3476 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3477 return 1;
3478
Tejun Heo672b2d62008-01-24 00:05:14 +09003479 /* thaw frozen port and recover failed devices */
3480 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3481 return 0;
3482
3483 /* reset at least once if reset is requested */
3484 if ((ehc->i.action & ATA_EH_RESET) &&
3485 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09003486 return 0;
3487
3488 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo1eca4362008-11-03 20:03:17 +09003489 ata_for_each_dev(dev, link, ALL) {
Tejun Heo084fe632006-05-31 18:28:03 +09003490 if (dev->class == ATA_DEV_UNKNOWN &&
3491 ehc->classes[dev->devno] != ATA_DEV_NONE)
3492 return 0;
3493 }
3494
3495 return 1;
3496}
3497
Tejun Heoc2c7a892009-01-29 20:31:34 +09003498static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3499{
3500 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3501 u64 now = get_jiffies_64();
3502 int *trials = void_arg;
3503
3504 if (ent->timestamp < now - min(now, interval))
3505 return -1;
3506
3507 (*trials)++;
3508 return 0;
3509}
3510
Tejun Heo02c05a22007-11-27 19:28:54 +09003511static int ata_eh_schedule_probe(struct ata_device *dev)
3512{
3513 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heoc2c7a892009-01-29 20:31:34 +09003514 struct ata_link *link = ata_dev_phys_link(dev);
3515 int trials = 0;
Tejun Heo02c05a22007-11-27 19:28:54 +09003516
3517 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3518 (ehc->did_probe_mask & (1 << dev->devno)))
3519 return 0;
3520
3521 ata_eh_detach_dev(dev);
3522 ata_dev_init(dev);
3523 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09003524 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09003525 ehc->saved_xfer_mode[dev->devno] = 0;
3526 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09003527
Tejun Heo6b7ae952010-09-01 17:50:06 +02003528 /* the link maybe in a deep sleep, wake it up */
Tejun Heo6c8ea892010-09-01 17:50:07 +02003529 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3530 if (ata_is_host_link(link))
3531 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3532 ATA_LPM_EMPTY);
3533 else
3534 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3535 ATA_LPM_EMPTY);
3536 }
Tejun Heo6b7ae952010-09-01 17:50:06 +02003537
Tejun Heoc2c7a892009-01-29 20:31:34 +09003538 /* Record and count probe trials on the ering. The specific
3539 * error mask used is irrelevant. Because a successful device
3540 * detection clears the ering, this count accumulates only if
3541 * there are consecutive failed probes.
3542 *
3543 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3544 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3545 * forced to 1.5Gbps.
3546 *
3547 * This is to work around cases where failed link speed
3548 * negotiation results in device misdetection leading to
3549 * infinite DEVXCHG or PHRDY CHG events.
3550 */
3551 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3552 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3553
3554 if (trials > ATA_EH_PROBE_TRIALS)
3555 sata_down_spd_limit(link, 1);
3556
Tejun Heo02c05a22007-11-27 19:28:54 +09003557 return 1;
3558}
3559
Tejun Heo9b1e2652007-08-06 18:36:24 +09003560static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09003561{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09003562 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09003563
Tejun Heocf9a5902009-01-29 20:31:35 +09003564 /* -EAGAIN from EH routine indicates retry without prejudice.
3565 * The requester is responsible for ensuring forward progress.
3566 */
3567 if (err != -EAGAIN)
3568 ehc->tries[dev->devno]--;
Tejun Heofee7ca72007-07-01 19:05:58 +09003569
3570 switch (err) {
3571 case -ENODEV:
3572 /* device missing or wrong IDENTIFY data, schedule probing */
3573 ehc->i.probe_mask |= (1 << dev->devno);
3574 case -EINVAL:
3575 /* give it just one more chance */
3576 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3577 case -EIO:
Tejun Heod89293a2009-01-29 20:31:29 +09003578 if (ehc->tries[dev->devno] == 1) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003579 /* This is the last chance, better to slow
3580 * down than lose it.
3581 */
Tejun Heoa07d4992009-01-29 20:31:33 +09003582 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
Tejun Heod89293a2009-01-29 20:31:29 +09003583 if (dev->pio_mode > XFER_PIO_0)
3584 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
Tejun Heofee7ca72007-07-01 19:05:58 +09003585 }
3586 }
3587
3588 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3589 /* disable device if it has used up all its chances */
3590 ata_dev_disable(dev);
3591
3592 /* detach if offline */
Tejun Heob1c72912008-07-31 17:02:43 +09003593 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
Tejun Heofee7ca72007-07-01 19:05:58 +09003594 ata_eh_detach_dev(dev);
3595
Tejun Heo02c05a22007-11-27 19:28:54 +09003596 /* schedule probe if necessary */
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003597 if (ata_eh_schedule_probe(dev)) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003598 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003599 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3600 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3601 }
Tejun Heo9b1e2652007-08-06 18:36:24 +09003602
3603 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09003604 } else {
Tejun Heocf480622008-01-24 00:05:14 +09003605 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003606 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09003607 }
3608}
3609
Tejun Heo022bdb02006-05-15 20:58:22 +09003610/**
3611 * ata_eh_recover - recover host port after error
3612 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09003613 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003614 * @softreset: softreset method (can be NULL)
3615 * @hardreset: hardreset method (can be NULL)
3616 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003617 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09003618 *
3619 * This is the alpha and omega, eum and yang, heart and soul of
3620 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09003621 * recover each link and hotplug requests are recorded in the
3622 * link's eh_context. This function executes all the operations
3623 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09003624 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09003625 *
3626 * LOCKING:
3627 * Kernel thread context (may sleep).
3628 *
3629 * RETURNS:
3630 * 0 on success, -errno on failure.
3631 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003632int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3633 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3634 ata_postreset_fn_t postreset,
3635 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003636{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003637 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003638 struct ata_device *dev;
Tejun Heo6b7ae952010-09-01 17:50:06 +02003639 int rc, nr_fails;
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003640 unsigned long flags, deadline;
Tejun Heo022bdb02006-05-15 20:58:22 +09003641
3642 DPRINTK("ENTER\n");
3643
3644 /* prep for recovery */
Tejun Heo1eca4362008-11-03 20:03:17 +09003645 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003646 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09003647
Tejun Heof9df58c2007-09-23 13:14:13 +09003648 /* re-enable link? */
3649 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3650 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3651 spin_lock_irqsave(ap->lock, flags);
3652 link->flags &= ~ATA_LFLAG_DISABLED;
3653 spin_unlock_irqrestore(ap->lock, flags);
3654 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3655 }
3656
Tejun Heo1eca4362008-11-03 20:03:17 +09003657 ata_for_each_dev(dev, link, ALL) {
Tejun Heofd995f72007-09-23 13:14:12 +09003658 if (link->flags & ATA_LFLAG_NO_RETRY)
3659 ehc->tries[dev->devno] = 1;
3660 else
3661 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09003662
Tejun Heo9b1e2652007-08-06 18:36:24 +09003663 /* collect port action mask recorded in dev actions */
3664 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3665 ~ATA_EH_PERDEV_MASK;
3666 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09003667
Tejun Heo9b1e2652007-08-06 18:36:24 +09003668 /* process hotplug request */
3669 if (dev->flags & ATA_DFLAG_DETACH)
3670 ata_eh_detach_dev(dev);
3671
Tejun Heo02c05a22007-11-27 19:28:54 +09003672 /* schedule probe if necessary */
3673 if (!ata_dev_enabled(dev))
3674 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09003675 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003676 }
3677
3678 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09003679 rc = 0;
3680
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003681 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09003682 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003683 goto out;
3684
Tejun Heo9b1e2652007-08-06 18:36:24 +09003685 /* prep for EH */
Tejun Heo1eca4362008-11-03 20:03:17 +09003686 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003687 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003688
Tejun Heo9b1e2652007-08-06 18:36:24 +09003689 /* skip EH if possible. */
3690 if (ata_eh_skip_recovery(link))
3691 ehc->i.action = 0;
3692
Tejun Heo1eca4362008-11-03 20:03:17 +09003693 ata_for_each_dev(dev, link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003694 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3695 }
Tejun Heo084fe632006-05-31 18:28:03 +09003696
Tejun Heo022bdb02006-05-15 20:58:22 +09003697 /* reset */
Tejun Heo1eca4362008-11-03 20:03:17 +09003698 ata_for_each_link(link, ap, EDGE) {
Tejun Heodc98c322008-05-19 01:15:07 +09003699 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003700
Tejun Heodc98c322008-05-19 01:15:07 +09003701 if (!(ehc->i.action & ATA_EH_RESET))
3702 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003703
Tejun Heodc98c322008-05-19 01:15:07 +09003704 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3705 prereset, softreset, hardreset, postreset);
3706 if (rc) {
Joe Perchesa9a79df2011-04-15 15:51:59 -07003707 ata_link_err(link, "reset failed, giving up\n");
Tejun Heodc98c322008-05-19 01:15:07 +09003708 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09003709 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003710 }
3711
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003712 do {
3713 unsigned long now;
3714
3715 /*
3716 * clears ATA_EH_PARK in eh_info and resets
3717 * ap->park_req_pending
3718 */
3719 ata_eh_pull_park_action(ap);
3720
3721 deadline = jiffies;
Tejun Heo1eca4362008-11-03 20:03:17 +09003722 ata_for_each_link(link, ap, EDGE) {
3723 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003724 struct ata_eh_context *ehc = &link->eh_context;
3725 unsigned long tmp;
3726
3727 if (dev->class != ATA_DEV_ATA)
3728 continue;
3729 if (!(ehc->i.dev_action[dev->devno] &
3730 ATA_EH_PARK))
3731 continue;
3732 tmp = dev->unpark_deadline;
3733 if (time_before(deadline, tmp))
3734 deadline = tmp;
3735 else if (time_before_eq(tmp, jiffies))
3736 continue;
3737 if (ehc->unloaded_mask & (1 << dev->devno))
3738 continue;
3739
3740 ata_eh_park_issue_cmd(dev, 1);
3741 }
3742 }
3743
3744 now = jiffies;
3745 if (time_before_eq(deadline, now))
3746 break;
3747
Tejun Heoc0c362b2010-09-06 17:57:14 +02003748 ata_eh_release(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003749 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3750 deadline - now);
Tejun Heoc0c362b2010-09-06 17:57:14 +02003751 ata_eh_acquire(ap);
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003752 } while (deadline);
Tejun Heo1eca4362008-11-03 20:03:17 +09003753 ata_for_each_link(link, ap, EDGE) {
3754 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003755 if (!(link->eh_context.unloaded_mask &
3756 (1 << dev->devno)))
3757 continue;
3758
3759 ata_eh_park_issue_cmd(dev, 0);
3760 ata_eh_done(link, dev, ATA_EH_PARK);
3761 }
3762 }
3763
Tejun Heo9b1e2652007-08-06 18:36:24 +09003764 /* the rest */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003765 nr_fails = 0;
3766 ata_for_each_link(link, ap, PMP_FIRST) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003767 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003768
Tejun Heo6b7ae952010-09-01 17:50:06 +02003769 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3770 goto config_lpm;
3771
Tejun Heo9b1e2652007-08-06 18:36:24 +09003772 /* revalidate existing devices and attach new ones */
3773 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09003774 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003775 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003776
Tejun Heo633273a32007-09-23 13:19:54 +09003777 /* if PMP got attached, return, pmp EH will take care of it */
3778 if (link->device->class == ATA_DEV_PMP) {
3779 ehc->i.action = 0;
3780 return 0;
3781 }
3782
Tejun Heo9b1e2652007-08-06 18:36:24 +09003783 /* configure transfer mode if necessary */
3784 if (ehc->i.flags & ATA_EHI_SETMODE) {
3785 rc = ata_set_mode(link, &dev);
3786 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003787 goto rest_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003788 ehc->i.flags &= ~ATA_EHI_SETMODE;
3789 }
3790
Tejun Heo11fc33d2008-08-30 14:20:01 +02003791 /* If reset has been issued, clear UA to avoid
3792 * disrupting the current users of the device.
3793 */
3794 if (ehc->i.flags & ATA_EHI_DID_RESET) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003795 ata_for_each_dev(dev, link, ALL) {
Tejun Heo11fc33d2008-08-30 14:20:01 +02003796 if (dev->class != ATA_DEV_ATAPI)
3797 continue;
3798 rc = atapi_eh_clear_ua(dev);
3799 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003800 goto rest_fail;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003801 }
3802 }
3803
Tejun Heo6013efd2009-11-19 15:36:45 +09003804 /* retry flush if necessary */
3805 ata_for_each_dev(dev, link, ALL) {
3806 if (dev->class != ATA_DEV_ATA)
3807 continue;
3808 rc = ata_eh_maybe_retry_flush(dev);
3809 if (rc)
Tejun Heo6b7ae952010-09-01 17:50:06 +02003810 goto rest_fail;
Tejun Heo6013efd2009-11-19 15:36:45 +09003811 }
3812
Tejun Heo6b7ae952010-09-01 17:50:06 +02003813 config_lpm:
Tejun Heo11fc33d2008-08-30 14:20:01 +02003814 /* configure link power saving */
Tejun Heo6b7ae952010-09-01 17:50:06 +02003815 if (link->lpm_policy != ap->target_lpm_policy) {
3816 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3817 if (rc)
3818 goto rest_fail;
3819 }
Kristen Carlson Accardica773292007-10-25 00:58:59 -04003820
Tejun Heo9b1e2652007-08-06 18:36:24 +09003821 /* this link is okay now */
3822 ehc->i.flags = 0;
3823 continue;
3824
Tejun Heo6b7ae952010-09-01 17:50:06 +02003825 rest_fail:
3826 nr_fails++;
3827 if (dev)
3828 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo9b1e2652007-08-06 18:36:24 +09003829
Tejun Heob06ce3e2007-10-09 15:06:48 +09003830 if (ap->pflags & ATA_PFLAG_FROZEN) {
3831 /* PMP reset requires working host port.
3832 * Can't retry if it's frozen.
3833 */
Tejun Heo071f44b2008-04-07 22:47:22 +09003834 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09003835 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003836 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09003837 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003838 }
3839
Tejun Heo6b7ae952010-09-01 17:50:06 +02003840 if (nr_fails)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003841 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003842
Tejun Heo022bdb02006-05-15 20:58:22 +09003843 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003844 if (rc && r_failed_link)
3845 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003846
3847 DPRINTK("EXIT, rc=%d\n", rc);
3848 return rc;
3849}
3850
3851/**
3852 * ata_eh_finish - finish up EH
3853 * @ap: host port to finish EH for
3854 *
3855 * Recovery is complete. Clean up EH states and retry or finish
3856 * failed qcs.
3857 *
3858 * LOCKING:
3859 * None.
3860 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003861void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09003862{
3863 int tag;
3864
3865 /* retry or finish qcs */
3866 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3867 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3868
3869 if (!(qc->flags & ATA_QCFLAG_FAILED))
3870 continue;
3871
3872 if (qc->err_mask) {
3873 /* FIXME: Once EH migration is complete,
3874 * generate sense data in this function,
3875 * considering both err_mask and tf.
3876 */
Tejun Heo03faab72008-03-27 19:14:24 +09003877 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09003878 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09003879 else
3880 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003881 } else {
3882 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3883 ata_eh_qc_complete(qc);
3884 } else {
3885 /* feed zero TF to sense generation */
3886 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3887 ata_eh_qc_retry(qc);
3888 }
3889 }
3890 }
Tejun Heoda917d62007-09-23 13:14:12 +09003891
3892 /* make sure nr_active_links is zero after EH */
3893 WARN_ON(ap->nr_active_links);
3894 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003895}
3896
3897/**
3898 * ata_do_eh - do standard error handling
3899 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09003900 *
Tejun Heof5914a42006-05-31 18:27:48 +09003901 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003902 * @softreset: softreset method (can be NULL)
3903 * @hardreset: hardreset method (can be NULL)
3904 * @postreset: postreset method (can be NULL)
3905 *
3906 * Perform standard error handling sequence.
3907 *
3908 * LOCKING:
3909 * Kernel thread context (may sleep).
3910 */
Tejun Heof5914a42006-05-31 18:27:48 +09003911void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3912 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3913 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09003914{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003915 struct ata_device *dev;
3916 int rc;
3917
3918 ata_eh_autopsy(ap);
3919 ata_eh_report(ap);
3920
3921 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3922 NULL);
3923 if (rc) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003924 ata_for_each_dev(dev, &ap->link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003925 ata_dev_disable(dev);
3926 }
3927
Tejun Heo022bdb02006-05-15 20:58:22 +09003928 ata_eh_finish(ap);
3929}
Tejun Heo500530f2006-07-03 16:07:27 +09003930
Tejun Heoa1efdab2008-03-25 12:22:50 +09003931/**
3932 * ata_std_error_handler - standard error handler
3933 * @ap: host port to handle error for
3934 *
3935 * Standard error handler
3936 *
3937 * LOCKING:
3938 * Kernel thread context (may sleep).
3939 */
3940void ata_std_error_handler(struct ata_port *ap)
3941{
3942 struct ata_port_operations *ops = ap->ops;
3943 ata_reset_fn_t hardreset = ops->hardreset;
3944
Tejun Heo57c9efd2008-04-07 22:47:19 +09003945 /* ignore built-in hardreset if SCR access is not available */
Tejun Heofe06e5f2010-05-10 21:41:39 +02003946 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09003947 hardreset = NULL;
3948
3949 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3950}
3951
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003952#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09003953/**
3954 * ata_eh_handle_port_suspend - perform port suspend operation
3955 * @ap: port to suspend
3956 *
3957 * Suspend @ap.
3958 *
3959 * LOCKING:
3960 * Kernel thread context (may sleep).
3961 */
3962static void ata_eh_handle_port_suspend(struct ata_port *ap)
3963{
3964 unsigned long flags;
3965 int rc = 0;
3966
3967 /* are we suspending? */
3968 spin_lock_irqsave(ap->lock, flags);
3969 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3970 ap->pm_mesg.event == PM_EVENT_ON) {
3971 spin_unlock_irqrestore(ap->lock, flags);
3972 return;
3973 }
3974 spin_unlock_irqrestore(ap->lock, flags);
3975
3976 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3977
Tejun Heo64578a32007-05-15 03:28:16 +09003978 /* tell ACPI we're suspending */
3979 rc = ata_acpi_on_suspend(ap);
3980 if (rc)
3981 goto out;
3982
Tejun Heo500530f2006-07-03 16:07:27 +09003983 /* suspend */
3984 ata_eh_freeze_port(ap);
3985
3986 if (ap->ops->port_suspend)
3987 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3988
Shaohua Libd3adca2007-11-02 09:32:38 +08003989 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a32007-05-15 03:28:16 +09003990 out:
Tejun Heo500530f2006-07-03 16:07:27 +09003991 /* report result */
3992 spin_lock_irqsave(ap->lock, flags);
3993
3994 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3995 if (rc == 0)
3996 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09003997 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09003998 ata_port_schedule_eh(ap);
3999
4000 if (ap->pm_result) {
4001 *ap->pm_result = rc;
4002 ap->pm_result = NULL;
4003 }
4004
4005 spin_unlock_irqrestore(ap->lock, flags);
4006
4007 return;
4008}
4009
4010/**
4011 * ata_eh_handle_port_resume - perform port resume operation
4012 * @ap: port to resume
4013 *
4014 * Resume @ap.
4015 *
Tejun Heo500530f2006-07-03 16:07:27 +09004016 * LOCKING:
4017 * Kernel thread context (may sleep).
4018 */
4019static void ata_eh_handle_port_resume(struct ata_port *ap)
4020{
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004021 struct ata_link *link;
4022 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09004023 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02004024 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09004025
4026 /* are we resuming? */
4027 spin_lock_irqsave(ap->lock, flags);
4028 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4029 ap->pm_mesg.event != PM_EVENT_ON) {
4030 spin_unlock_irqrestore(ap->lock, flags);
4031 return;
4032 }
4033 spin_unlock_irqrestore(ap->lock, flags);
4034
Tejun Heo9666f402007-05-04 21:27:47 +02004035 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09004036
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09004037 /*
4038 * Error timestamps are in jiffies which doesn't run while
4039 * suspended and PHY events during resume isn't too uncommon.
4040 * When the two are combined, it can lead to unnecessary speed
4041 * downs if the machine is suspended and resumed repeatedly.
4042 * Clear error history.
4043 */
4044 ata_for_each_link(link, ap, HOST_FIRST)
4045 ata_for_each_dev(dev, link, ALL)
4046 ata_ering_clear(&dev->ering);
4047
Shaohua Libd3adca2007-11-02 09:32:38 +08004048 ata_acpi_set_state(ap, PMSG_ON);
4049
Tejun Heo500530f2006-07-03 16:07:27 +09004050 if (ap->ops->port_resume)
4051 rc = ap->ops->port_resume(ap);
4052
Tejun Heo67465442007-05-15 03:28:16 +09004053 /* tell ACPI that we're resuming */
4054 ata_acpi_on_resume(ap);
4055
Tejun Heo9666f402007-05-04 21:27:47 +02004056 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09004057 spin_lock_irqsave(ap->lock, flags);
4058 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4059 if (ap->pm_result) {
4060 *ap->pm_result = rc;
4061 ap->pm_result = NULL;
4062 }
4063 spin_unlock_irqrestore(ap->lock, flags);
4064}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09004065#endif /* CONFIG_PM */