blob: f7cae640015528aeeb211cf28ee64ef98f685d89 [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>
Jeff Garzik28555682007-10-11 17:12:35 -040036#include <linux/pci.h>
Tejun Heoece1d632006-04-02 18:51:53 +090037#include <scsi/scsi.h>
38#include <scsi/scsi_host.h>
39#include <scsi/scsi_eh.h>
40#include <scsi/scsi_device.h>
41#include <scsi/scsi_cmnd.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040042#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090043
44#include <linux/libata.h>
45
46#include "libata.h"
47
Tejun Heo7d47e8d2007-02-02 16:22:31 +090048enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090049 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090050 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
51 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
52 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090053 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090054
55 /* error flags */
56 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090057 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Tejun Heo3884f7b2007-11-27 19:28:56 +090058
59 /* error categories */
60 ATA_ECAT_NONE = 0,
61 ATA_ECAT_ATA_BUS = 1,
62 ATA_ECAT_TOUT_HSM = 2,
63 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090064 ATA_ECAT_DUBIOUS_NONE = 4,
65 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
66 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
67 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
68 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090069};
70
Tejun Heo31daabd2007-02-02 16:50:52 +090071/* Waiting in ->prereset can never be reliable. It's sometimes nice
72 * to wait there but it can't be depended upon; otherwise, we wouldn't
73 * be resetting. Just give it enough time for most drives to spin up.
74 */
75enum {
76 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
Tejun Heo5ddf24c2007-07-16 14:29:41 +090077 ATA_EH_FASTDRAIN_INTERVAL = 3 * HZ,
Tejun Heo31daabd2007-02-02 16:50:52 +090078};
79
80/* The following table determines how we sequence resets. Each entry
81 * represents timeout for that try. The first try can be soft or
82 * hardreset. All others are hardreset if available. In most cases
83 * the first reset w/ 10sec timeout should succeed. Following entries
84 * are mostly for error handling, hotplug and retarded devices.
85 */
86static const unsigned long ata_eh_reset_timeouts[] = {
87 10 * HZ, /* most drives spin up by 10sec */
88 10 * HZ, /* > 99% working drives spin up before 20sec */
89 35 * HZ, /* give > 30 secs of idleness for retarded devices */
90 5 * HZ, /* and sweet one last chance */
91 /* > 1 min has elapsed, give up */
92};
93
Tejun Heoad9e2762006-05-15 20:58:12 +090094static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090095#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +090096static void ata_eh_handle_port_suspend(struct ata_port *ap);
97static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +090098#else /* CONFIG_PM */
99static void ata_eh_handle_port_suspend(struct ata_port *ap)
100{ }
101
102static void ata_eh_handle_port_resume(struct ata_port *ap)
103{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900104#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900105
Tejun Heob64bbc32007-07-16 14:29:39 +0900106static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
107 va_list args)
108{
109 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
110 ATA_EH_DESC_LEN - ehi->desc_len,
111 fmt, args);
112}
113
114/**
115 * __ata_ehi_push_desc - push error description without adding separator
116 * @ehi: target EHI
117 * @fmt: printf format string
118 *
119 * Format string according to @fmt and append it to @ehi->desc.
120 *
121 * LOCKING:
122 * spin_lock_irqsave(host lock)
123 */
124void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
125{
126 va_list args;
127
128 va_start(args, fmt);
129 __ata_ehi_pushv_desc(ehi, fmt, args);
130 va_end(args);
131}
132
133/**
134 * ata_ehi_push_desc - push error description with separator
135 * @ehi: target EHI
136 * @fmt: printf format string
137 *
138 * Format string according to @fmt and append it to @ehi->desc.
139 * If @ehi->desc is not empty, ", " is added in-between.
140 *
141 * LOCKING:
142 * spin_lock_irqsave(host lock)
143 */
144void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
145{
146 va_list args;
147
148 if (ehi->desc_len)
149 __ata_ehi_push_desc(ehi, ", ");
150
151 va_start(args, fmt);
152 __ata_ehi_pushv_desc(ehi, fmt, args);
153 va_end(args);
154}
155
156/**
157 * ata_ehi_clear_desc - clean error description
158 * @ehi: target EHI
159 *
160 * Clear @ehi->desc.
161 *
162 * LOCKING:
163 * spin_lock_irqsave(host lock)
164 */
165void ata_ehi_clear_desc(struct ata_eh_info *ehi)
166{
167 ehi->desc[0] = '\0';
168 ehi->desc_len = 0;
169}
170
Tejun Heocbcdd872007-08-18 13:14:55 +0900171/**
172 * ata_port_desc - append port description
173 * @ap: target ATA port
174 * @fmt: printf format string
175 *
176 * Format string according to @fmt and append it to port
177 * description. If port description is not empty, " " is added
178 * in-between. This function is to be used while initializing
179 * ata_host. The description is printed on host registration.
180 *
181 * LOCKING:
182 * None.
183 */
184void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
185{
186 va_list args;
187
188 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
189
190 if (ap->link.eh_info.desc_len)
191 __ata_ehi_push_desc(&ap->link.eh_info, " ");
192
193 va_start(args, fmt);
194 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
195 va_end(args);
196}
197
198#ifdef CONFIG_PCI
199
200/**
201 * ata_port_pbar_desc - append PCI BAR description
202 * @ap: target ATA port
203 * @bar: target PCI BAR
204 * @offset: offset into PCI BAR
205 * @name: name of the area
206 *
207 * If @offset is negative, this function formats a string which
208 * contains the name, address, size and type of the BAR and
209 * appends it to the port description. If @offset is zero or
210 * positive, only name and offsetted address is appended.
211 *
212 * LOCKING:
213 * None.
214 */
215void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
216 const char *name)
217{
218 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
219 char *type = "";
220 unsigned long long start, len;
221
222 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
223 type = "m";
224 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
225 type = "i";
226
227 start = (unsigned long long)pci_resource_start(pdev, bar);
228 len = (unsigned long long)pci_resource_len(pdev, bar);
229
230 if (offset < 0)
231 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
232 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800233 ata_port_desc(ap, "%s 0x%llx", name,
234 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900235}
236
237#endif /* CONFIG_PCI */
238
Tejun Heo3884f7b2007-11-27 19:28:56 +0900239static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900240 unsigned int err_mask)
241{
242 struct ata_ering_entry *ent;
243
244 WARN_ON(!err_mask);
245
246 ering->cursor++;
247 ering->cursor %= ATA_ERING_SIZE;
248
249 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900250 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900251 ent->err_mask = err_mask;
252 ent->timestamp = get_jiffies_64();
253}
254
Tejun Heo76326ac2007-11-27 19:28:59 +0900255static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
256{
257 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
258
259 if (ent->err_mask)
260 return ent;
261 return NULL;
262}
263
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900264static void ata_ering_clear(struct ata_ering *ering)
Tejun Heo0c247c52006-05-15 20:58:19 +0900265{
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900266 memset(ering, 0, sizeof(*ering));
Tejun Heo0c247c52006-05-15 20:58:19 +0900267}
268
269static int ata_ering_map(struct ata_ering *ering,
270 int (*map_fn)(struct ata_ering_entry *, void *),
271 void *arg)
272{
273 int idx, rc = 0;
274 struct ata_ering_entry *ent;
275
276 idx = ering->cursor;
277 do {
278 ent = &ering->ring[idx];
279 if (!ent->err_mask)
280 break;
281 rc = map_fn(ent, arg);
282 if (rc)
283 break;
284 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
285 } while (idx != ering->cursor);
286
287 return rc;
288}
289
Tejun Heo64f65ca2006-06-24 20:30:18 +0900290static unsigned int ata_eh_dev_action(struct ata_device *dev)
291{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900292 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900293
294 return ehc->i.action | ehc->i.dev_action[dev->devno];
295}
296
Tejun Heof58229f2007-08-06 18:36:23 +0900297static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900298 struct ata_eh_info *ehi, unsigned int action)
299{
Tejun Heof58229f2007-08-06 18:36:23 +0900300 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900301
302 if (!dev) {
303 ehi->action &= ~action;
Tejun Heof58229f2007-08-06 18:36:23 +0900304 ata_link_for_each_dev(tdev, link)
305 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900306 } else {
307 /* doesn't make sense for port-wide EH actions */
308 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
309
310 /* break ehi->action into ehi->dev_action */
311 if (ehi->action & action) {
Tejun Heof58229f2007-08-06 18:36:23 +0900312 ata_link_for_each_dev(tdev, link)
313 ehi->dev_action[tdev->devno] |=
314 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900315 ehi->action &= ~action;
316 }
317
318 /* turn off the specified per-dev action */
319 ehi->dev_action[dev->devno] &= ~action;
320 }
321}
322
Tejun Heoece1d632006-04-02 18:51:53 +0900323/**
324 * ata_scsi_timed_out - SCSI layer time out callback
325 * @cmd: timed out SCSI command
326 *
327 * Handles SCSI layer timeout. We race with normal completion of
328 * the qc for @cmd. If the qc is already gone, we lose and let
329 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
330 * timed out and EH should be invoked. Prevent ata_qc_complete()
331 * from finishing it by setting EH_SCHEDULED and return
332 * EH_NOT_HANDLED.
333 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900334 * TODO: kill this function once old EH is gone.
335 *
Tejun Heoece1d632006-04-02 18:51:53 +0900336 * LOCKING:
337 * Called from timer context
338 *
339 * RETURNS:
340 * EH_HANDLED or EH_NOT_HANDLED
341 */
342enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
343{
344 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400345 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900346 unsigned long flags;
347 struct ata_queued_cmd *qc;
Tejun Heoad9e2762006-05-15 20:58:12 +0900348 enum scsi_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900349
350 DPRINTK("ENTER\n");
351
Tejun Heoad9e2762006-05-15 20:58:12 +0900352 if (ap->ops->error_handler) {
353 ret = EH_NOT_HANDLED;
354 goto out;
355 }
356
357 ret = EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400358 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900359 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900360 if (qc) {
361 WARN_ON(qc->scsicmd != cmd);
362 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
363 qc->err_mask |= AC_ERR_TIMEOUT;
364 ret = EH_NOT_HANDLED;
365 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400366 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900367
Tejun Heoad9e2762006-05-15 20:58:12 +0900368 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900369 DPRINTK("EXIT, ret=%d\n", ret);
370 return ret;
371}
372
373/**
374 * ata_scsi_error - SCSI layer error handler callback
375 * @host: SCSI host on which error occurred
376 *
377 * Handles SCSI-layer-thrown error events.
378 *
379 * LOCKING:
380 * Inherited from SCSI layer (none, can sleep)
381 *
382 * RETURNS:
383 * Zero.
384 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400385void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900386{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400387 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900388 int i;
Tejun Heoad9e2762006-05-15 20:58:12 +0900389 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900390
391 DPRINTK("ENTER\n");
392
Tejun Heoad9e2762006-05-15 20:58:12 +0900393 /* synchronize with port task */
Tejun Heoece1d632006-04-02 18:51:53 +0900394 ata_port_flush_task(ap);
395
Jeff Garzikcca39742006-08-24 03:19:22 -0400396 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900397
Tejun Heoad9e2762006-05-15 20:58:12 +0900398 /* For new EH, all qcs are finished in one of three ways -
399 * normal completion, error completion, and SCSI timeout.
400 * Both cmpletions can race against SCSI timeout. When normal
401 * completion wins, the qc never reaches EH. When error
402 * completion wins, the qc has ATA_QCFLAG_FAILED set.
403 *
404 * When SCSI timeout wins, things are a bit more complex.
405 * Normal or error completion can occur after the timeout but
406 * before this point. In such cases, both types of
407 * completions are honored. A scmd is determined to have
408 * timed out iff its associated qc is active and not failed.
409 */
410 if (ap->ops->error_handler) {
411 struct scsi_cmnd *scmd, *tmp;
412 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900413
Tejun Heoe30349d2006-07-03 03:02:15 +0900414 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900415
416 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
417 struct ata_queued_cmd *qc;
418
419 for (i = 0; i < ATA_MAX_QUEUE; i++) {
420 qc = __ata_qc_from_tag(ap, i);
421 if (qc->flags & ATA_QCFLAG_ACTIVE &&
422 qc->scsicmd == scmd)
423 break;
424 }
425
426 if (i < ATA_MAX_QUEUE) {
427 /* the scmd has an associated qc */
428 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
429 /* which hasn't failed yet, timeout */
430 qc->err_mask |= AC_ERR_TIMEOUT;
431 qc->flags |= ATA_QCFLAG_FAILED;
432 nr_timedout++;
433 }
434 } else {
435 /* Normal completion occurred after
436 * SCSI timeout but before this point.
437 * Successfully complete it.
438 */
439 scmd->retries = scmd->allowed;
440 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
441 }
442 }
443
444 /* If we have timed out qcs. They belong to EH from
445 * this point but the state of the controller is
446 * unknown. Freeze the port to make sure the IRQ
447 * handler doesn't diddle with those qcs. This must
448 * be done atomically w.r.t. setting QCFLAG_FAILED.
449 */
450 if (nr_timedout)
451 __ata_port_freeze(ap);
452
Tejun Heoe30349d2006-07-03 03:02:15 +0900453 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900454
455 /* initialize eh_tries */
456 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900457 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900458 spin_unlock_wait(ap->lock);
Tejun Heoad9e2762006-05-15 20:58:12 +0900459
460 repeat:
461 /* invoke error handler */
462 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900463 struct ata_link *link;
464
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900465 /* kill fast drain timer */
466 del_timer_sync(&ap->fastdrain_timer);
467
Tejun Heo500530f2006-07-03 16:07:27 +0900468 /* process port resume request */
469 ata_eh_handle_port_resume(ap);
470
Tejun Heof3e81b12006-05-15 20:58:21 +0900471 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900472 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900473
Tejun Heocf1b86c2007-08-06 18:36:23 +0900474 __ata_port_for_each_link(link, ap) {
Tejun Heo00115e02007-11-27 19:28:58 +0900475 struct ata_eh_context *ehc = &link->eh_context;
476 struct ata_device *dev;
477
Tejun Heocf1b86c2007-08-06 18:36:23 +0900478 memset(&link->eh_context, 0, sizeof(link->eh_context));
479 link->eh_context.i = link->eh_info;
480 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900481
482 ata_link_for_each_dev(dev, link) {
483 int devno = dev->devno;
484
485 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
486 if (ata_ncq_enabled(dev))
487 ehc->saved_ncq_enabled |= 1 << devno;
488 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900489 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900490
Tejun Heob51e9e52006-06-29 01:29:30 +0900491 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
492 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900493 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900494
Tejun Heoe30349d2006-07-03 03:02:15 +0900495 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900496
Tejun Heo500530f2006-07-03 16:07:27 +0900497 /* invoke EH, skip if unloading or suspended */
498 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900499 ap->ops->error_handler(ap);
500 else
501 ata_eh_finish(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900502
Tejun Heo500530f2006-07-03 16:07:27 +0900503 /* process port suspend request */
504 ata_eh_handle_port_suspend(ap);
505
Tejun Heoad9e2762006-05-15 20:58:12 +0900506 /* Exception might have happend after ->error_handler
507 * recovered the port but before this point. Repeat
508 * EH in such case.
509 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900510 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900511
Tejun Heob51e9e52006-06-29 01:29:30 +0900512 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900513 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900514 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900515 goto repeat;
516 }
517 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
Tejun Heoa1e10f72007-08-18 13:28:49 +0900518 "tries, giving up\n", ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900519 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900520 }
521
Tejun Heof3e81b12006-05-15 20:58:21 +0900522 /* this run is complete, make sure EH info is clear */
Tejun Heocf1b86c2007-08-06 18:36:23 +0900523 __ata_port_for_each_link(link, ap)
524 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900525
Tejun Heoe30349d2006-07-03 03:02:15 +0900526 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900527 * that if exception occurs after this point but
528 * before EH completion, SCSI midlayer will
529 * re-initiate EH.
530 */
531 host->host_eh_scheduled = 0;
532
Tejun Heoe30349d2006-07-03 03:02:15 +0900533 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900534 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900535 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900536 ap->ops->eng_timeout(ap);
537 }
538
539 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900540 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
541
542 scsi_eh_flush_done_q(&ap->eh_done_q);
543
Tejun Heoad9e2762006-05-15 20:58:12 +0900544 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900545 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900546
Tejun Heo1cdaf532006-07-03 16:07:26 +0900547 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900548 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900549 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
David Howells52bad642006-11-22 14:54:01 +0000550 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900551
552 if (ap->pflags & ATA_PFLAG_RECOVERED)
553 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900554
Tejun Heob51e9e52006-06-29 01:29:30 +0900555 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900556
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900557 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900558 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900559 wake_up_all(&ap->eh_wait_q);
560
Tejun Heoe30349d2006-07-03 03:02:15 +0900561 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900562
Tejun Heoece1d632006-04-02 18:51:53 +0900563 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900564}
565
566/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900567 * ata_port_wait_eh - Wait for the currently pending EH to complete
568 * @ap: Port to wait EH for
569 *
570 * Wait until the currently pending EH is complete.
571 *
572 * LOCKING:
573 * Kernel thread context (may sleep).
574 */
575void ata_port_wait_eh(struct ata_port *ap)
576{
577 unsigned long flags;
578 DEFINE_WAIT(wait);
579
580 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400581 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900582
Tejun Heob51e9e52006-06-29 01:29:30 +0900583 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900584 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400585 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900586 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400587 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900588 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900589 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900590
Jeff Garzikba6a1302006-06-22 23:46:10 -0400591 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900592
593 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400594 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900595 msleep(10);
596 goto retry;
597 }
598}
599
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900600static int ata_eh_nr_in_flight(struct ata_port *ap)
601{
602 unsigned int tag;
603 int nr = 0;
604
605 /* count only non-internal commands */
606 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
607 if (ata_qc_from_tag(ap, tag))
608 nr++;
609
610 return nr;
611}
612
613void ata_eh_fastdrain_timerfn(unsigned long arg)
614{
615 struct ata_port *ap = (void *)arg;
616 unsigned long flags;
617 int cnt;
618
619 spin_lock_irqsave(ap->lock, flags);
620
621 cnt = ata_eh_nr_in_flight(ap);
622
623 /* are we done? */
624 if (!cnt)
625 goto out_unlock;
626
627 if (cnt == ap->fastdrain_cnt) {
628 unsigned int tag;
629
630 /* No progress during the last interval, tag all
631 * in-flight qcs as timed out and freeze the port.
632 */
633 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
634 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
635 if (qc)
636 qc->err_mask |= AC_ERR_TIMEOUT;
637 }
638
639 ata_port_freeze(ap);
640 } else {
641 /* some qcs have finished, give it another chance */
642 ap->fastdrain_cnt = cnt;
643 ap->fastdrain_timer.expires =
644 jiffies + ATA_EH_FASTDRAIN_INTERVAL;
645 add_timer(&ap->fastdrain_timer);
646 }
647
648 out_unlock:
649 spin_unlock_irqrestore(ap->lock, flags);
650}
651
652/**
653 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
654 * @ap: target ATA port
655 * @fastdrain: activate fast drain
656 *
657 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
658 * is non-zero and EH wasn't pending before. Fast drain ensures
659 * that EH kicks in in timely manner.
660 *
661 * LOCKING:
662 * spin_lock_irqsave(host lock)
663 */
664static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
665{
666 int cnt;
667
668 /* already scheduled? */
669 if (ap->pflags & ATA_PFLAG_EH_PENDING)
670 return;
671
672 ap->pflags |= ATA_PFLAG_EH_PENDING;
673
674 if (!fastdrain)
675 return;
676
677 /* do we have in-flight qcs? */
678 cnt = ata_eh_nr_in_flight(ap);
679 if (!cnt)
680 return;
681
682 /* activate fast drain */
683 ap->fastdrain_cnt = cnt;
684 ap->fastdrain_timer.expires = jiffies + ATA_EH_FASTDRAIN_INTERVAL;
685 add_timer(&ap->fastdrain_timer);
686}
687
Tejun Heof686bcb2006-05-15 20:58:05 +0900688/**
689 * ata_qc_schedule_eh - schedule qc for error handling
690 * @qc: command to schedule error handling for
691 *
692 * Schedule error handling for @qc. EH will kick in as soon as
693 * other commands are drained.
694 *
695 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400696 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900697 */
698void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
699{
700 struct ata_port *ap = qc->ap;
701
702 WARN_ON(!ap->ops->error_handler);
703
704 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900705 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900706
707 /* The following will fail if timeout has already expired.
708 * ata_scsi_error() takes care of such scmds on EH entry.
709 * Note that ATA_QCFLAG_FAILED is unconditionally set after
710 * this function completes.
711 */
712 scsi_req_abort_cmd(qc->scsicmd);
713}
714
Tejun Heo7b70fc02006-05-15 20:58:07 +0900715/**
716 * ata_port_schedule_eh - schedule error handling without a qc
717 * @ap: ATA port to schedule EH for
718 *
719 * Schedule error handling for @ap. EH will kick in as soon as
720 * all commands are drained.
721 *
722 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400723 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900724 */
725void ata_port_schedule_eh(struct ata_port *ap)
726{
727 WARN_ON(!ap->ops->error_handler);
728
Tejun Heof4d6d002007-05-01 11:50:15 +0200729 if (ap->pflags & ATA_PFLAG_INITIALIZING)
730 return;
731
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900732 ata_eh_set_pending(ap, 1);
Jeff Garzikcca39742006-08-24 03:19:22 -0400733 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900734
735 DPRINTK("port EH scheduled\n");
736}
737
Tejun Heodbd82612007-08-06 18:36:23 +0900738static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900739{
740 int tag, nr_aborted = 0;
741
742 WARN_ON(!ap->ops->error_handler);
743
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900744 /* we're gonna abort all commands, no need for fast drain */
745 ata_eh_set_pending(ap, 0);
746
Tejun Heo7b70fc02006-05-15 20:58:07 +0900747 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
748 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
749
Tejun Heodbd82612007-08-06 18:36:23 +0900750 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +0900751 qc->flags |= ATA_QCFLAG_FAILED;
752 ata_qc_complete(qc);
753 nr_aborted++;
754 }
755 }
756
757 if (!nr_aborted)
758 ata_port_schedule_eh(ap);
759
760 return nr_aborted;
761}
762
Tejun Heoe3180492006-05-15 20:58:09 +0900763/**
Tejun Heodbd82612007-08-06 18:36:23 +0900764 * ata_link_abort - abort all qc's on the link
765 * @link: ATA link to abort qc's for
766 *
767 * Abort all active qc's active on @link and schedule EH.
768 *
769 * LOCKING:
770 * spin_lock_irqsave(host lock)
771 *
772 * RETURNS:
773 * Number of aborted qc's.
774 */
775int ata_link_abort(struct ata_link *link)
776{
777 return ata_do_link_abort(link->ap, link);
778}
779
780/**
781 * ata_port_abort - abort all qc's on the port
782 * @ap: ATA port to abort qc's for
783 *
784 * Abort all active qc's of @ap and schedule EH.
785 *
786 * LOCKING:
787 * spin_lock_irqsave(host_set lock)
788 *
789 * RETURNS:
790 * Number of aborted qc's.
791 */
792int ata_port_abort(struct ata_port *ap)
793{
794 return ata_do_link_abort(ap, NULL);
795}
796
797/**
Tejun Heoe3180492006-05-15 20:58:09 +0900798 * __ata_port_freeze - freeze port
799 * @ap: ATA port to freeze
800 *
801 * This function is called when HSM violation or some other
802 * condition disrupts normal operation of the port. Frozen port
803 * is not allowed to perform any operation until the port is
804 * thawed, which usually follows a successful reset.
805 *
806 * ap->ops->freeze() callback can be used for freezing the port
807 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
808 * port cannot be frozen hardware-wise, the interrupt handler
809 * must ack and clear interrupts unconditionally while the port
810 * is frozen.
811 *
812 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400813 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900814 */
815static void __ata_port_freeze(struct ata_port *ap)
816{
817 WARN_ON(!ap->ops->error_handler);
818
819 if (ap->ops->freeze)
820 ap->ops->freeze(ap);
821
Tejun Heob51e9e52006-06-29 01:29:30 +0900822 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900823
Tejun Heo44877b42007-02-21 01:06:51 +0900824 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900825}
826
827/**
828 * ata_port_freeze - abort & freeze port
829 * @ap: ATA port to freeze
830 *
831 * Abort and freeze @ap.
832 *
833 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400834 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900835 *
836 * RETURNS:
837 * Number of aborted commands.
838 */
839int ata_port_freeze(struct ata_port *ap)
840{
841 int nr_aborted;
842
843 WARN_ON(!ap->ops->error_handler);
844
845 nr_aborted = ata_port_abort(ap);
846 __ata_port_freeze(ap);
847
848 return nr_aborted;
849}
850
851/**
Tejun Heo7d77b242007-09-23 13:14:13 +0900852 * sata_async_notification - SATA async notification handler
853 * @ap: ATA port where async notification is received
854 *
855 * Handler to be called when async notification via SDB FIS is
856 * received. This function schedules EH if necessary.
857 *
858 * LOCKING:
859 * spin_lock_irqsave(host lock)
860 *
861 * RETURNS:
862 * 1 if EH is scheduled, 0 otherwise.
863 */
864int sata_async_notification(struct ata_port *ap)
865{
866 u32 sntf;
867 int rc;
868
869 if (!(ap->flags & ATA_FLAG_AN))
870 return 0;
871
872 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
873 if (rc == 0)
874 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
875
876 if (!ap->nr_pmp_links || rc) {
877 /* PMP is not attached or SNTF is not available */
878 if (!ap->nr_pmp_links) {
879 /* PMP is not attached. Check whether ATAPI
880 * AN is configured. If so, notify media
881 * change.
882 */
883 struct ata_device *dev = ap->link.device;
884
885 if ((dev->class == ATA_DEV_ATAPI) &&
886 (dev->flags & ATA_DFLAG_AN))
887 ata_scsi_media_change_notify(dev);
888 return 0;
889 } else {
890 /* PMP is attached but SNTF is not available.
891 * ATAPI async media change notification is
892 * not used. The PMP must be reporting PHY
893 * status change, schedule EH.
894 */
895 ata_port_schedule_eh(ap);
896 return 1;
897 }
898 } else {
899 /* PMP is attached and SNTF is available */
900 struct ata_link *link;
901
902 /* check and notify ATAPI AN */
903 ata_port_for_each_link(link, ap) {
904 if (!(sntf & (1 << link->pmp)))
905 continue;
906
907 if ((link->device->class == ATA_DEV_ATAPI) &&
908 (link->device->flags & ATA_DFLAG_AN))
909 ata_scsi_media_change_notify(link->device);
910 }
911
912 /* If PMP is reporting that PHY status of some
913 * downstream ports has changed, schedule EH.
914 */
915 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
916 ata_port_schedule_eh(ap);
917 return 1;
918 }
919
920 return 0;
921 }
922}
923
924/**
Tejun Heoe3180492006-05-15 20:58:09 +0900925 * ata_eh_freeze_port - EH helper to freeze port
926 * @ap: ATA port to freeze
927 *
928 * Freeze @ap.
929 *
930 * LOCKING:
931 * None.
932 */
933void ata_eh_freeze_port(struct ata_port *ap)
934{
935 unsigned long flags;
936
937 if (!ap->ops->error_handler)
938 return;
939
Jeff Garzikba6a1302006-06-22 23:46:10 -0400940 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900941 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400942 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900943}
944
945/**
946 * ata_port_thaw_port - EH helper to thaw port
947 * @ap: ATA port to thaw
948 *
949 * Thaw frozen port @ap.
950 *
951 * LOCKING:
952 * None.
953 */
954void ata_eh_thaw_port(struct ata_port *ap)
955{
956 unsigned long flags;
957
958 if (!ap->ops->error_handler)
959 return;
960
Jeff Garzikba6a1302006-06-22 23:46:10 -0400961 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900962
Tejun Heob51e9e52006-06-29 01:29:30 +0900963 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900964
965 if (ap->ops->thaw)
966 ap->ops->thaw(ap);
967
Jeff Garzikba6a1302006-06-22 23:46:10 -0400968 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +0900969
Tejun Heo44877b42007-02-21 01:06:51 +0900970 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900971}
972
Tejun Heoece1d632006-04-02 18:51:53 +0900973static void ata_eh_scsidone(struct scsi_cmnd *scmd)
974{
975 /* nada */
976}
977
978static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
979{
980 struct ata_port *ap = qc->ap;
981 struct scsi_cmnd *scmd = qc->scsicmd;
982 unsigned long flags;
983
Jeff Garzikba6a1302006-06-22 23:46:10 -0400984 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900985 qc->scsidone = ata_eh_scsidone;
986 __ata_qc_complete(qc);
987 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -0400988 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900989
990 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
991}
992
993/**
994 * ata_eh_qc_complete - Complete an active ATA command from EH
995 * @qc: Command to complete
996 *
997 * Indicate to the mid and upper layers that an ATA command has
998 * completed. To be used from EH.
999 */
1000void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1001{
1002 struct scsi_cmnd *scmd = qc->scsicmd;
1003 scmd->retries = scmd->allowed;
1004 __ata_eh_qc_complete(qc);
1005}
1006
1007/**
1008 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1009 * @qc: Command to retry
1010 *
1011 * Indicate to the mid and upper layers that an ATA command
1012 * should be retried. To be used from EH.
1013 *
1014 * SCSI midlayer limits the number of retries to scmd->allowed.
1015 * scmd->retries is decremented for commands which get retried
1016 * due to unrelated failures (qc->err_mask is zero).
1017 */
1018void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1019{
1020 struct scsi_cmnd *scmd = qc->scsicmd;
1021 if (!qc->err_mask && scmd->retries)
1022 scmd->retries--;
1023 __ata_eh_qc_complete(qc);
1024}
Tejun Heo022bdb02006-05-15 20:58:22 +09001025
1026/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001027 * ata_eh_detach_dev - detach ATA device
1028 * @dev: ATA device to detach
1029 *
1030 * Detach @dev.
1031 *
1032 * LOCKING:
1033 * None.
1034 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001035void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001036{
Tejun Heof58229f2007-08-06 18:36:23 +09001037 struct ata_link *link = dev->link;
1038 struct ata_port *ap = link->ap;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001039 unsigned long flags;
1040
1041 ata_dev_disable(dev);
1042
Jeff Garzikba6a1302006-06-22 23:46:10 -04001043 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001044
1045 dev->flags &= ~ATA_DFLAG_DETACH;
1046
1047 if (ata_scsi_offline_dev(dev)) {
1048 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001049 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001050 }
1051
Tejun Heobeb07c12006-06-24 20:30:19 +09001052 /* clear per-dev EH actions */
Tejun Heof58229f2007-08-06 18:36:23 +09001053 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1054 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heobeb07c12006-06-24 20:30:19 +09001055
Jeff Garzikba6a1302006-06-22 23:46:10 -04001056 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001057}
1058
1059/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001060 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001061 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001062 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001063 * @action: action about to be performed
1064 *
1065 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001066 * in @link->eh_info such that eh actions are not unnecessarily
1067 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001068 *
1069 * LOCKING:
1070 * None.
1071 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001072void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1073 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001074{
Tejun Heo955e57d2007-08-06 18:36:23 +09001075 struct ata_port *ap = link->ap;
1076 struct ata_eh_info *ehi = &link->eh_info;
1077 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001078 unsigned long flags;
1079
Jeff Garzikba6a1302006-06-22 23:46:10 -04001080 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001081
Tejun Heocf480622008-01-24 00:05:14 +09001082 /* suck in and clear reset modifier */
1083 if (action & ATA_EH_RESET) {
Tejun Heo13abf502006-07-10 23:18:46 +09001084 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
Tejun Heo13abf502006-07-10 23:18:46 +09001085 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
1086 }
1087
Tejun Heo955e57d2007-08-06 18:36:23 +09001088 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001089
1090 if (!(ehc->i.flags & ATA_EHI_QUIET))
Tejun Heo1cdaf532006-07-03 16:07:26 +09001091 ap->pflags |= ATA_PFLAG_RECOVERED;
1092
Jeff Garzikba6a1302006-06-22 23:46:10 -04001093 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001094}
1095
1096/**
Tejun Heo47005f22006-06-19 18:27:23 +09001097 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001098* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001099 * @dev: target ATA dev for per-dev action (can be NULL)
1100 * @action: action just completed
1101 *
1102 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001103 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001104 *
1105 * LOCKING:
1106 * None.
1107 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001108void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1109 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001110{
Tejun Heo955e57d2007-08-06 18:36:23 +09001111 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001112
Tejun Heocf480622008-01-24 00:05:14 +09001113 /* if reset is complete, clear reset modifier */
1114 if (action & ATA_EH_RESET)
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001115 ehc->i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
Tejun Heo13abf502006-07-10 23:18:46 +09001116
Tejun Heo955e57d2007-08-06 18:36:23 +09001117 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001118}
1119
1120/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001121 * ata_err_string - convert err_mask to descriptive string
1122 * @err_mask: error mask to convert to string
1123 *
1124 * Convert @err_mask to descriptive string. Errors are
1125 * prioritized according to severity and only the most severe
1126 * error is reported.
1127 *
1128 * LOCKING:
1129 * None.
1130 *
1131 * RETURNS:
1132 * Descriptive string for @err_mask
1133 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001134static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001135{
1136 if (err_mask & AC_ERR_HOST_BUS)
1137 return "host bus error";
1138 if (err_mask & AC_ERR_ATA_BUS)
1139 return "ATA bus error";
1140 if (err_mask & AC_ERR_TIMEOUT)
1141 return "timeout";
1142 if (err_mask & AC_ERR_HSM)
1143 return "HSM violation";
1144 if (err_mask & AC_ERR_SYSTEM)
1145 return "internal error";
1146 if (err_mask & AC_ERR_MEDIA)
1147 return "media error";
1148 if (err_mask & AC_ERR_INVALID)
1149 return "invalid argument";
1150 if (err_mask & AC_ERR_DEV)
1151 return "device error";
1152 return "unknown error";
1153}
1154
1155/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001156 * ata_read_log_page - read a specific log page
1157 * @dev: target device
1158 * @page: page to read
1159 * @buf: buffer to store read page
1160 * @sectors: number of sectors to read
1161 *
1162 * Read log page using READ_LOG_EXT command.
1163 *
1164 * LOCKING:
1165 * Kernel thread context (may sleep).
1166 *
1167 * RETURNS:
1168 * 0 on success, AC_ERR_* mask otherwise.
1169 */
1170static unsigned int ata_read_log_page(struct ata_device *dev,
1171 u8 page, void *buf, unsigned int sectors)
1172{
1173 struct ata_taskfile tf;
1174 unsigned int err_mask;
1175
1176 DPRINTK("read log page - page %d\n", page);
1177
1178 ata_tf_init(dev, &tf);
1179 tf.command = ATA_CMD_READ_LOG_EXT;
1180 tf.lbal = page;
1181 tf.nsect = sectors;
1182 tf.hob_nsect = sectors >> 8;
1183 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1184 tf.protocol = ATA_PROT_PIO;
1185
1186 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001187 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001188
1189 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1190 return err_mask;
1191}
1192
1193/**
1194 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1195 * @dev: Device to read log page 10h from
1196 * @tag: Resulting tag of the failed command
1197 * @tf: Resulting taskfile registers of the failed command
1198 *
1199 * Read log page 10h to obtain NCQ error details and clear error
1200 * condition.
1201 *
1202 * LOCKING:
1203 * Kernel thread context (may sleep).
1204 *
1205 * RETURNS:
1206 * 0 on success, -errno otherwise.
1207 */
1208static int ata_eh_read_log_10h(struct ata_device *dev,
1209 int *tag, struct ata_taskfile *tf)
1210{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001211 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001212 unsigned int err_mask;
1213 u8 csum;
1214 int i;
1215
1216 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1217 if (err_mask)
1218 return -EIO;
1219
1220 csum = 0;
1221 for (i = 0; i < ATA_SECT_SIZE; i++)
1222 csum += buf[i];
1223 if (csum)
1224 ata_dev_printk(dev, KERN_WARNING,
1225 "invalid checksum 0x%x on log page 10h\n", csum);
1226
1227 if (buf[0] & 0x80)
1228 return -ENOENT;
1229
1230 *tag = buf[0] & 0x1f;
1231
1232 tf->command = buf[2];
1233 tf->feature = buf[3];
1234 tf->lbal = buf[4];
1235 tf->lbam = buf[5];
1236 tf->lbah = buf[6];
1237 tf->device = buf[7];
1238 tf->hob_lbal = buf[8];
1239 tf->hob_lbam = buf[9];
1240 tf->hob_lbah = buf[10];
1241 tf->nsect = buf[12];
1242 tf->hob_nsect = buf[13];
1243
1244 return 0;
1245}
1246
1247/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001248 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1249 * @dev: device to perform REQUEST_SENSE to
1250 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1251 *
1252 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1253 * SENSE. This function is EH helper.
1254 *
1255 * LOCKING:
1256 * Kernel thread context (may sleep).
1257 *
1258 * RETURNS:
1259 * 0 on success, AC_ERR_* mask on failure
1260 */
Albert Lee56287762007-04-02 11:30:46 +08001261static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
Tejun Heo022bdb02006-05-15 20:58:22 +09001262{
Albert Lee56287762007-04-02 11:30:46 +08001263 struct ata_device *dev = qc->dev;
1264 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001265 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001266 struct ata_taskfile tf;
1267 u8 cdb[ATAPI_CDB_LEN];
1268
1269 DPRINTK("ATAPI request sense\n");
1270
Tejun Heo022bdb02006-05-15 20:58:22 +09001271 /* FIXME: is this needed? */
1272 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1273
Albert Lee56287762007-04-02 11:30:46 +08001274 /* initialize sense_buf with the error register,
1275 * for the case where they are -not- overwritten
1276 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001277 sense_buf[0] = 0x70;
Albert Lee56287762007-04-02 11:30:46 +08001278 sense_buf[2] = qc->result_tf.feature >> 4;
1279
Jeff Garzika617c092007-05-21 20:14:23 -04001280 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001281 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001282
1283 memset(cdb, 0, ATAPI_CDB_LEN);
1284 cdb[0] = REQUEST_SENSE;
1285 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1286
1287 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1288 tf.command = ATA_CMD_PACKET;
1289
1290 /* is it pointless to prefer PIO for "safety reasons"? */
1291 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001292 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001293 tf.feature |= ATAPI_PKT_DMA;
1294 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001295 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001296 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1297 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001298 }
1299
1300 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001301 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001302}
1303
1304/**
1305 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001306 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001307 *
1308 * Analyze SError if available and further determine cause of
1309 * failure.
1310 *
1311 * LOCKING:
1312 * None.
1313 */
Tejun Heo02607312007-08-06 18:36:23 +09001314static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001315{
Tejun Heo02607312007-08-06 18:36:23 +09001316 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001317 u32 serror = ehc->i.serror;
1318 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001319 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001320
1321 if (serror & SERR_PERSISTENT) {
1322 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001323 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001324 }
1325 if (serror &
1326 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1327 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001328 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001329 }
1330 if (serror & SERR_PROTOCOL) {
1331 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001332 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001333 }
1334 if (serror & SERR_INTERNAL) {
1335 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001336 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001337 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001338
1339 /* Determine whether a hotplug event has occurred. Both
1340 * SError.N/X are considered hotplug events for enabled or
1341 * host links. For disabled PMP links, only N bit is
1342 * considered as X bit is left at 1 for link plugging.
1343 */
1344 hotplug_mask = 0;
1345
1346 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1347 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1348 else
1349 hotplug_mask = SERR_PHYRDY_CHG;
1350
1351 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001352 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001353
1354 ehc->i.err_mask |= err_mask;
1355 ehc->i.action |= action;
1356}
1357
1358/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001359 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001360 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001361 *
1362 * Read log page 10h, determine the offending qc and acquire
1363 * error status TF. For NCQ device errors, all LLDDs have to do
1364 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1365 * care of the rest.
1366 *
1367 * LOCKING:
1368 * Kernel thread context (may sleep).
1369 */
Tejun Heo02607312007-08-06 18:36:23 +09001370static void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001371{
Tejun Heo02607312007-08-06 18:36:23 +09001372 struct ata_port *ap = link->ap;
1373 struct ata_eh_context *ehc = &link->eh_context;
1374 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001375 struct ata_queued_cmd *qc;
1376 struct ata_taskfile tf;
1377 int tag, rc;
1378
1379 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001380 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001381 return;
1382
1383 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001384 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001385 return;
1386
1387 /* has LLDD analyzed already? */
1388 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1389 qc = __ata_qc_from_tag(ap, tag);
1390
1391 if (!(qc->flags & ATA_QCFLAG_FAILED))
1392 continue;
1393
1394 if (qc->err_mask)
1395 return;
1396 }
1397
1398 /* okay, this error is ours */
1399 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1400 if (rc) {
Tejun Heo02607312007-08-06 18:36:23 +09001401 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001402 "(errno=%d)\n", rc);
1403 return;
1404 }
1405
Tejun Heo02607312007-08-06 18:36:23 +09001406 if (!(link->sactive & (1 << tag))) {
1407 ata_link_printk(link, KERN_ERR, "log page 10h reported "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001408 "inactive tag %d\n", tag);
1409 return;
1410 }
1411
1412 /* we've got the perpetrator, condemn it */
1413 qc = __ata_qc_from_tag(ap, tag);
1414 memcpy(&qc->result_tf, &tf, sizeof(tf));
Tejun Heo5335b722007-07-16 14:29:40 +09001415 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001416 ehc->i.err_mask &= ~AC_ERR_DEV;
1417}
1418
1419/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001420 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1421 * @qc: qc to analyze
1422 * @tf: Taskfile registers to analyze
1423 *
1424 * Analyze taskfile of @qc and further determine cause of
1425 * failure. This function also requests ATAPI sense data if
1426 * avaliable.
1427 *
1428 * LOCKING:
1429 * Kernel thread context (may sleep).
1430 *
1431 * RETURNS:
1432 * Determined recovery action
1433 */
1434static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1435 const struct ata_taskfile *tf)
1436{
1437 unsigned int tmp, action = 0;
1438 u8 stat = tf->command, err = tf->feature;
1439
1440 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1441 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001442 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001443 }
1444
Tejun Heoa51d6442007-03-20 15:24:11 +09001445 if (stat & (ATA_ERR | ATA_DF))
1446 qc->err_mask |= AC_ERR_DEV;
1447 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001448 return 0;
1449
1450 switch (qc->dev->class) {
1451 case ATA_DEV_ATA:
1452 if (err & ATA_ICRC)
1453 qc->err_mask |= AC_ERR_ATA_BUS;
1454 if (err & ATA_UNC)
1455 qc->err_mask |= AC_ERR_MEDIA;
1456 if (err & ATA_IDNF)
1457 qc->err_mask |= AC_ERR_INVALID;
1458 break;
1459
1460 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001461 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Albert Lee56287762007-04-02 11:30:46 +08001462 tmp = atapi_eh_request_sense(qc);
Tejun Heoa569a302006-11-21 10:40:51 +09001463 if (!tmp) {
1464 /* ATA_QCFLAG_SENSE_VALID is used to
1465 * tell atapi_qc_complete() that sense
1466 * data is already valid.
1467 *
1468 * TODO: interpret sense data and set
1469 * appropriate err_mask.
1470 */
1471 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1472 } else
1473 qc->err_mask |= tmp;
1474 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001475 }
1476
1477 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001478 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001479
1480 return action;
1481}
1482
Tejun Heo76326ac2007-11-27 19:28:59 +09001483static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1484 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001485{
Tejun Heo76326ac2007-11-27 19:28:59 +09001486 int base = 0;
1487
1488 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1489 *xfer_ok = 1;
1490
1491 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001492 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001493
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001494 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001495 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001496
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001497 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001498 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001499
Tejun Heo3884f7b2007-11-27 19:28:56 +09001500 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001501 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001502 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001503 if ((err_mask &
1504 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001505 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001506 }
1507
1508 return 0;
1509}
1510
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001511struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001512 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001513 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001514 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001515};
1516
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001517static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001518{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001519 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001520 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001521
1522 if (ent->timestamp < arg->since)
1523 return -1;
1524
Tejun Heo76326ac2007-11-27 19:28:59 +09001525 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1526 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001527 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001528
Tejun Heo022bdb02006-05-15 20:58:22 +09001529 return 0;
1530}
1531
1532/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001533 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001534 * @dev: Device of interest
1535 *
1536 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001537 * whether NCQ needs to be turned off, transfer speed should be
1538 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001539 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001540 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001541 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001542 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1543 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001544 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001545 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001546 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001547 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1548 * data transfer hasn't been verified.
1549 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001550 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001551 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001552 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001553 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001554 * SPEED_DOWN : Speed down transfer speed but don't fall back
1555 * to PIO.
1556 *
1557 * FALLBACK_TO_PIO : Fall back to PIO.
1558 *
1559 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001560 * taken per error. An action triggered by non-DUBIOUS errors
1561 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1562 * This is to expedite speed down decisions right after device is
1563 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001564 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001565 * The followings are speed down rules. #1 and #2 deal with
1566 * DUBIOUS errors.
1567 *
1568 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1569 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1570 *
1571 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1572 * occurred during last 5 mins, NCQ_OFF.
1573 *
1574 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Tejun Heo3884f7b2007-11-27 19:28:56 +09001575 * ocurred during last 5 mins, FALLBACK_TO_PIO
1576 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001577 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001578 * during last 10 mins, NCQ_OFF.
1579 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001580 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001581 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001582 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001583 * LOCKING:
1584 * Inherited from caller.
1585 *
1586 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001587 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001588 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001589static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001590{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001591 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1592 u64 j64 = get_jiffies_64();
1593 struct speed_down_verdict_arg arg;
1594 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001595
Tejun Heo3884f7b2007-11-27 19:28:56 +09001596 /* scan past 5 mins of error history */
1597 memset(&arg, 0, sizeof(arg));
1598 arg.since = j64 - min(j64, j5mins);
1599 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1600
Tejun Heo76326ac2007-11-27 19:28:59 +09001601 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1602 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1603 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1604 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1605
1606 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1607 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1608 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1609
Tejun Heo3884f7b2007-11-27 19:28:56 +09001610 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1611 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b82007-11-27 19:28:57 +09001612 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001613 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1614
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001615 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001616 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001617 arg.since = j64 - min(j64, j10mins);
1618 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001619
Tejun Heo3884f7b2007-11-27 19:28:56 +09001620 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1621 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001622 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001623
1624 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1625 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b82007-11-27 19:28:57 +09001626 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001627 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001628
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001629 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001630}
1631
1632/**
1633 * ata_eh_speed_down - record error and speed down if necessary
1634 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001635 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001636 * @err_mask: err_mask of the error
1637 *
1638 * Record error and examine error history to determine whether
1639 * adjusting transmission speed is necessary. It also sets
1640 * transmission limits appropriately if such adjustment is
1641 * necessary.
1642 *
1643 * LOCKING:
1644 * Kernel thread context (may sleep).
1645 *
1646 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001647 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001648 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001649static unsigned int ata_eh_speed_down(struct ata_device *dev,
1650 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001651{
Tejun Heo3884f7b2007-11-27 19:28:56 +09001652 struct ata_link *link = dev->link;
Tejun Heo76326ac2007-11-27 19:28:59 +09001653 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001654 unsigned int verdict;
1655 unsigned int action = 0;
1656
1657 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001658 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001659 return 0;
1660
1661 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001662 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001663 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001664
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001665 /* turn off NCQ? */
1666 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1667 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1668 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1669 dev->flags |= ATA_DFLAG_NCQ_OFF;
1670 ata_dev_printk(dev, KERN_WARNING,
1671 "NCQ disabled due to excessive errors\n");
1672 goto done;
1673 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001674
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001675 /* speed down? */
1676 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1677 /* speed down SATA link speed if possible */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001678 if (sata_down_spd_limit(link) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001679 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001680 goto done;
1681 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001682
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001683 /* lower transfer mode */
1684 if (dev->spdn_cnt < 2) {
1685 static const int dma_dnxfer_sel[] =
1686 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1687 static const int pio_dnxfer_sel[] =
1688 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1689 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001690
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001691 if (dev->xfer_shift != ATA_SHIFT_PIO)
1692 sel = dma_dnxfer_sel[dev->spdn_cnt];
1693 else
1694 sel = pio_dnxfer_sel[dev->spdn_cnt];
1695
1696 dev->spdn_cnt++;
1697
1698 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001699 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001700 goto done;
1701 }
1702 }
1703 }
1704
1705 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b82007-11-27 19:28:57 +09001706 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001707 */
1708 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b82007-11-27 19:28:57 +09001709 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001710 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1711 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1712 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09001713 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001714 goto done;
1715 }
1716 }
1717
Tejun Heo022bdb02006-05-15 20:58:22 +09001718 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001719 done:
1720 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09001721 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1722 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001723 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001724}
1725
1726/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001727 * ata_eh_link_autopsy - analyze error and determine recovery action
1728 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09001729 *
Tejun Heo02607312007-08-06 18:36:23 +09001730 * Analyze why @link failed and determine which recovery actions
1731 * are needed. This function also sets more detailed AC_ERR_*
1732 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09001733 *
1734 * LOCKING:
1735 * Kernel thread context (may sleep).
1736 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001737static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001738{
Tejun Heo02607312007-08-06 18:36:23 +09001739 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09001740 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09001741 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001742 unsigned int all_err_mask = 0, eflags = 0;
1743 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09001744 u32 serror;
1745 int rc;
1746
1747 DPRINTK("ENTER\n");
1748
Tejun Heo1cdaf532006-07-03 16:07:26 +09001749 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1750 return;
1751
Tejun Heo022bdb02006-05-15 20:58:22 +09001752 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09001753 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09001754 if (rc == 0) {
1755 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09001756 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09001757 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09001758 /* SError read failed, force reset and probing */
Tejun Heo4e57c512007-07-16 14:29:41 +09001759 ata_ehi_schedule_probe(&ehc->i);
Tejun Heocf480622008-01-24 00:05:14 +09001760 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09001761 ehc->i.err_mask |= AC_ERR_OTHER;
1762 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001763
Tejun Heoe8ee8452006-05-15 21:03:46 +09001764 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09001765 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001766
Tejun Heo022bdb02006-05-15 20:58:22 +09001767 /* any real error trumps AC_ERR_OTHER */
1768 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1769 ehc->i.err_mask &= ~AC_ERR_OTHER;
1770
1771 all_err_mask |= ehc->i.err_mask;
1772
1773 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1774 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1775
Tejun Heo02607312007-08-06 18:36:23 +09001776 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001777 continue;
1778
1779 /* inherit upper level err_mask */
1780 qc->err_mask |= ehc->i.err_mask;
1781
Tejun Heo022bdb02006-05-15 20:58:22 +09001782 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001783 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001784
1785 /* DEV errors are probably spurious in case of ATA_BUS error */
1786 if (qc->err_mask & AC_ERR_ATA_BUS)
1787 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1788 AC_ERR_INVALID);
1789
1790 /* any real error trumps unknown error */
1791 if (qc->err_mask & ~AC_ERR_OTHER)
1792 qc->err_mask &= ~AC_ERR_OTHER;
1793
1794 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09001795 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09001796 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09001797
1798 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001799 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09001800 all_err_mask |= qc->err_mask;
1801 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001802 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09001803 }
1804
Tejun Heoa20f33f2006-05-16 12:58:24 +09001805 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001806 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001807 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09001808 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001809 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1810 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001811 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001812
Tejun Heodfcc1732007-10-31 10:17:05 +09001813 /* If we have offending qcs and the associated failed device,
1814 * perform per-dev EH action only on the offending device.
1815 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001816 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09001817 ehc->i.dev_action[ehc->i.dev->devno] |=
1818 ehc->i.action & ATA_EH_PERDEV_MASK;
1819 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09001820 }
1821
Tejun Heo2695e362008-01-10 13:41:23 +09001822 /* propagate timeout to host link */
1823 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1824 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1825
1826 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09001827 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09001828 if (!dev && ((ata_link_max_devices(link) == 1 &&
1829 ata_dev_enabled(link->device))))
1830 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09001831
Tejun Heo76326ac2007-11-27 19:28:59 +09001832 if (dev) {
1833 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1834 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001835 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09001836 }
Tejun Heodfcc1732007-10-31 10:17:05 +09001837
Tejun Heo022bdb02006-05-15 20:58:22 +09001838 DPRINTK("EXIT\n");
1839}
1840
1841/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001842 * ata_eh_autopsy - analyze error and determine recovery action
1843 * @ap: host port to perform autopsy on
1844 *
1845 * Analyze all links of @ap and determine why they failed and
1846 * which recovery actions are needed.
1847 *
1848 * LOCKING:
1849 * Kernel thread context (may sleep).
1850 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001851void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09001852{
1853 struct ata_link *link;
1854
Tejun Heo2695e362008-01-10 13:41:23 +09001855 ata_port_for_each_link(link, ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09001856 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09001857
1858 /* Autopsy of fanout ports can affect host link autopsy.
1859 * Perform host link autopsy last.
1860 */
1861 if (ap->nr_pmp_links)
1862 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09001863}
1864
1865/**
1866 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09001867 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09001868 *
1869 * Report EH to user.
1870 *
1871 * LOCKING:
1872 * None.
1873 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001874static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001875{
Tejun Heo02607312007-08-06 18:36:23 +09001876 struct ata_port *ap = link->ap;
1877 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001878 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09001879 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09001880 int tag, nr_failed = 0;
1881
Tejun Heo94ff3d52007-10-09 14:57:56 +09001882 if (ehc->i.flags & ATA_EHI_QUIET)
1883 return;
1884
Tejun Heo022bdb02006-05-15 20:58:22 +09001885 desc = NULL;
1886 if (ehc->i.desc[0] != '\0')
1887 desc = ehc->i.desc;
1888
1889 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1890 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1891
Tejun Heoe027bd32007-10-26 16:19:26 +09001892 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link ||
1893 ((qc->flags & ATA_QCFLAG_QUIET) &&
1894 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09001895 continue;
1896 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1897 continue;
1898
1899 nr_failed++;
1900 }
1901
1902 if (!nr_failed && !ehc->i.err_mask)
1903 return;
1904
1905 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09001906 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09001907 frozen = " frozen";
1908
Tejun Heoa1e10f72007-08-18 13:28:49 +09001909 memset(tries_buf, 0, sizeof(tries_buf));
1910 if (ap->eh_tries < ATA_EH_MAX_TRIES)
1911 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
1912 ap->eh_tries);
1913
Tejun Heo022bdb02006-05-15 20:58:22 +09001914 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001915 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09001916 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1917 ehc->i.err_mask, link->sactive, ehc->i.serror,
1918 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001919 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09001920 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001921 } else {
Tejun Heo02607312007-08-06 18:36:23 +09001922 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09001923 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1924 ehc->i.err_mask, link->sactive, ehc->i.serror,
1925 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001926 if (desc)
Tejun Heo02607312007-08-06 18:36:23 +09001927 ata_link_printk(link, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001928 }
1929
Robert Hancock1333e192007-10-02 11:22:02 -04001930 if (ehc->i.serror)
1931 ata_port_printk(ap, KERN_ERR,
1932 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
1933 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
1934 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
1935 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
1936 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
1937 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
1938 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
1939 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
1940 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
1941 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
1942 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
1943 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
1944 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
1945 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
1946 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
1947 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
1948 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001949 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04001950
Tejun Heo022bdb02006-05-15 20:58:22 +09001951 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1952 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09001953 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09001954 const u8 *cdb = qc->cdb;
1955 char data_buf[20] = "";
1956 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09001957
Tejun Heo02607312007-08-06 18:36:23 +09001958 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1959 qc->dev->link != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001960 continue;
1961
Tejun Heoabb6a882007-11-28 23:16:09 +09001962 if (qc->dma_dir != DMA_NONE) {
1963 static const char *dma_str[] = {
1964 [DMA_BIDIRECTIONAL] = "bidi",
1965 [DMA_TO_DEVICE] = "out",
1966 [DMA_FROM_DEVICE] = "in",
1967 };
1968 static const char *prot_str[] = {
1969 [ATA_PROT_PIO] = "pio",
1970 [ATA_PROT_DMA] = "dma",
1971 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05001972 [ATAPI_PROT_PIO] = "pio",
1973 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09001974 };
1975
1976 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
1977 prot_str[qc->tf.protocol], qc->nbytes,
1978 dma_str[qc->dma_dir]);
1979 }
1980
Jeff Garzike39eec12007-12-01 18:05:39 -05001981 if (ata_is_atapi(qc->tf.protocol))
Tejun Heoabb6a882007-11-28 23:16:09 +09001982 snprintf(cdb_buf, sizeof(cdb_buf),
1983 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
1984 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
1985 cdb[0], cdb[1], cdb[2], cdb[3],
1986 cdb[4], cdb[5], cdb[6], cdb[7],
1987 cdb[8], cdb[9], cdb[10], cdb[11],
1988 cdb[12], cdb[13], cdb[14], cdb[15]);
1989
Tejun Heo8a937582006-11-14 22:36:12 +09001990 ata_dev_printk(qc->dev, KERN_ERR,
1991 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09001992 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09001993 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09001994 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09001995 cmd->command, cmd->feature, cmd->nsect,
1996 cmd->lbal, cmd->lbam, cmd->lbah,
1997 cmd->hob_feature, cmd->hob_nsect,
1998 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09001999 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002000 res->command, res->feature, res->nsect,
2001 res->lbal, res->lbam, res->lbah,
2002 res->hob_feature, res->hob_nsect,
2003 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002004 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2005 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002006
2007 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002008 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002009 if (res->command & ATA_BUSY)
2010 ata_dev_printk(qc->dev, KERN_ERR,
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002011 "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002012 else
2013 ata_dev_printk(qc->dev, KERN_ERR,
2014 "status: { %s%s%s%s}\n",
2015 res->command & ATA_DRDY ? "DRDY " : "",
2016 res->command & ATA_DF ? "DF " : "",
2017 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002018 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002019 }
2020
2021 if (cmd->command != ATA_CMD_PACKET &&
2022 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2023 ATA_ABORTED)))
2024 ata_dev_printk(qc->dev, KERN_ERR,
2025 "error: { %s%s%s%s}\n",
2026 res->feature & ATA_ICRC ? "ICRC " : "",
2027 res->feature & ATA_UNC ? "UNC " : "",
2028 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002029 res->feature & ATA_ABORTED ? "ABRT " : "");
Tejun Heo022bdb02006-05-15 20:58:22 +09002030 }
2031}
2032
Tejun Heo9b1e2652007-08-06 18:36:24 +09002033/**
2034 * ata_eh_report - report error handling to user
2035 * @ap: ATA port to report EH about
2036 *
2037 * Report EH to user.
2038 *
2039 * LOCKING:
2040 * None.
2041 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002042void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002043{
2044 struct ata_link *link;
2045
2046 __ata_port_for_each_link(link, ap)
2047 ata_eh_link_report(link);
2048}
2049
Tejun Heocc0680a2007-08-06 18:36:23 +09002050static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heod4b2bab2007-02-02 16:50:52 +09002051 unsigned int *classes, unsigned long deadline)
Tejun Heod87fa382006-05-31 18:28:24 +09002052{
Tejun Heof58229f2007-08-06 18:36:23 +09002053 struct ata_device *dev;
2054 int rc;
Tejun Heod87fa382006-05-31 18:28:24 +09002055
Tejun Heocc0680a2007-08-06 18:36:23 +09002056 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002057 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002058
Tejun Heocc0680a2007-08-06 18:36:23 +09002059 rc = reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002060 if (rc)
2061 return rc;
2062
2063 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
2064 * is complete and convert all ATA_DEV_UNKNOWN to
2065 * ATA_DEV_NONE.
2066 */
Tejun Heocc0680a2007-08-06 18:36:23 +09002067 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002068 if (classes[dev->devno] != ATA_DEV_UNKNOWN)
Tejun Heod87fa382006-05-31 18:28:24 +09002069 break;
2070
Tejun Heof58229f2007-08-06 18:36:23 +09002071 if (dev) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002072 ata_link_for_each_dev(dev, link) {
Tejun Heof58229f2007-08-06 18:36:23 +09002073 if (classes[dev->devno] == ATA_DEV_UNKNOWN)
2074 classes[dev->devno] = ATA_DEV_NONE;
2075 }
2076 }
Tejun Heod87fa382006-05-31 18:28:24 +09002077
2078 return 0;
2079}
2080
Tejun Heoae791c02007-09-23 13:14:12 +09002081static int ata_eh_followup_srst_needed(struct ata_link *link,
2082 int rc, int classify,
Tejun Heo664faf02006-05-31 18:27:50 +09002083 const unsigned int *classes)
2084{
Tejun Heoae791c02007-09-23 13:14:12 +09002085 if (link->flags & ATA_LFLAG_NO_SRST)
2086 return 0;
Tejun Heo664faf02006-05-31 18:27:50 +09002087 if (rc == -EAGAIN)
2088 return 1;
2089 if (rc != 0)
2090 return 0;
Tejun Heo3495de72007-09-23 13:19:53 +09002091 if ((link->ap->flags & ATA_FLAG_PMP) && ata_is_host_link(link))
2092 return 1;
Tejun Heoae791c02007-09-23 13:14:12 +09002093 if (classify && !(link->flags & ATA_LFLAG_ASSUME_CLASS) &&
2094 classes[0] == ATA_DEV_UNKNOWN)
Tejun Heo664faf02006-05-31 18:27:50 +09002095 return 1;
2096 return 0;
2097}
2098
Tejun Heofb7fd612007-09-23 13:14:12 +09002099int ata_eh_reset(struct ata_link *link, int classify,
2100 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2101 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002102{
Tejun Heo416dc9e2007-10-31 10:17:03 +09002103 const int max_tries = ARRAY_SIZE(ata_eh_reset_timeouts);
Tejun Heoafaa5c32007-10-09 15:06:10 +09002104 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09002105 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09002106 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002107 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002108 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heo31daabd2007-02-02 16:50:52 +09002109 int try = 0;
Tejun Heof58229f2007-08-06 18:36:23 +09002110 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002111 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002112 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002113 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002114 u32 sstatus;
Tejun Heof58229f2007-08-06 18:36:23 +09002115 int rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002116
Tejun Heo13abf502006-07-10 23:18:46 +09002117 /* about to reset */
Tejun Heoafaa5c32007-10-09 15:06:10 +09002118 spin_lock_irqsave(ap->lock, flags);
2119 ap->pflags |= ATA_PFLAG_RESETTING;
2120 spin_unlock_irqrestore(ap->lock, flags);
2121
Tejun Heocf480622008-01-24 00:05:14 +09002122 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002123
Tejun Heocdeab112007-10-29 16:41:09 +09002124 ata_link_for_each_dev(dev, link) {
2125 /* If we issue an SRST then an ATA drive (not ATAPI)
2126 * may change configuration and be in PIO0 timing. If
2127 * we do a hard reset (or are coming from power on)
2128 * this is true for ATA or ATAPI. Until we've set a
2129 * suitable controller mode we should not touch the
2130 * bus as we may be talking too fast.
2131 */
2132 dev->pio_mode = XFER_PIO_0;
2133
2134 /* If the controller has a pio mode setup function
2135 * then use it to set the chipset to rights. Don't
2136 * touch the DMA setup as that will be dealt with when
2137 * configuring devices.
2138 */
2139 if (ap->ops->set_piomode)
2140 ap->ops->set_piomode(ap, dev);
2141 }
2142
Tejun Heoeec59f72008-03-06 13:09:34 +09002143 if (!softreset && !hardreset) {
2144 if (verbose)
2145 ata_link_printk(link, KERN_INFO, "no reset method "
2146 "available, skipping reset\n");
2147 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2148 lflags |= ATA_LFLAG_ASSUME_ATA;
2149 goto done;
2150 }
2151
Tejun Heocf480622008-01-24 00:05:14 +09002152 /* prefer hardreset */
2153 ehc->i.action &= ~ATA_EH_RESET;
2154 if (hardreset) {
2155 reset = hardreset;
2156 ehc->i.action = ATA_EH_HARDRESET;
2157 } else {
2158 reset = softreset;
2159 ehc->i.action = ATA_EH_SOFTRESET;
2160 }
Tejun Heof5914a42006-05-31 18:27:48 +09002161
2162 if (prereset) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002163 rc = prereset(link, jiffies + ATA_EH_PRERESET_TIMEOUT);
Tejun Heof5914a42006-05-31 18:27:48 +09002164 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002165 if (rc == -ENOENT) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002166 ata_link_printk(link, KERN_DEBUG,
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002167 "port disabled. ignoring.\n");
Tejun Heocf480622008-01-24 00:05:14 +09002168 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002169
Tejun Heo936fd732007-08-06 18:36:23 +09002170 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002171 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002172
2173 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002174 } else
Tejun Heocc0680a2007-08-06 18:36:23 +09002175 ata_link_printk(link, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09002176 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002177 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002178 }
2179 }
2180
Tejun Heocf480622008-01-24 00:05:14 +09002181 /* prereset() might have cleared ATA_EH_RESET */
2182 if (!(ehc->i.action & ATA_EH_RESET)) {
Tejun Heof5914a42006-05-31 18:27:48 +09002183 /* prereset told us not to reset, bang classes and return */
Tejun Heo936fd732007-08-06 18:36:23 +09002184 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002185 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002186 rc = 0;
2187 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002188 }
2189
Tejun Heo022bdb02006-05-15 20:58:22 +09002190 retry:
Tejun Heo31daabd2007-02-02 16:50:52 +09002191 deadline = jiffies + ata_eh_reset_timeouts[try++];
2192
Tejun Heo3e706392006-05-31 18:28:11 +09002193 /* shut up during boot probing */
2194 if (verbose)
Tejun Heocc0680a2007-08-06 18:36:23 +09002195 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
Tejun Heo3e706392006-05-31 18:28:11 +09002196 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002197
Tejun Heo13abf502006-07-10 23:18:46 +09002198 /* mark that this EH session started with reset */
Tejun Heo0d64a232007-04-23 02:41:05 +09002199 if (reset == hardreset)
2200 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2201 else
2202 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002203
Tejun Heocc0680a2007-08-06 18:36:23 +09002204 rc = ata_do_reset(link, reset, classes, deadline);
Tejun Heo022bdb02006-05-15 20:58:22 +09002205
Tejun Heo664faf02006-05-31 18:27:50 +09002206 if (reset == hardreset &&
Tejun Heoae791c02007-09-23 13:14:12 +09002207 ata_eh_followup_srst_needed(link, rc, classify, classes)) {
Tejun Heo664faf02006-05-31 18:27:50 +09002208 /* okay, let's do follow-up softreset */
Tejun Heo664faf02006-05-31 18:27:50 +09002209 reset = softreset;
2210
2211 if (!reset) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002212 ata_link_printk(link, KERN_ERR,
Tejun Heo664faf02006-05-31 18:27:50 +09002213 "follow-up softreset required "
2214 "but no softreset avaliable\n");
Tejun Heofccb6ea2007-07-16 14:29:41 +09002215 rc = -EINVAL;
Tejun Heo08cf69d2007-10-31 10:17:04 +09002216 goto fail;
Tejun Heo664faf02006-05-31 18:27:50 +09002217 }
2218
Tejun Heocf480622008-01-24 00:05:14 +09002219 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heocc0680a2007-08-06 18:36:23 +09002220 rc = ata_do_reset(link, reset, classes, deadline);
Tejun Heo664faf02006-05-31 18:27:50 +09002221 }
2222
Tejun Heo416dc9e2007-10-31 10:17:03 +09002223 /* -EAGAIN can happen if we skipped followup SRST */
2224 if (rc && rc != -EAGAIN)
2225 goto fail;
Tejun Heoae791c02007-09-23 13:14:12 +09002226
Tejun Heo08cf69d2007-10-31 10:17:04 +09002227 /* was classification successful? */
2228 if (classify && classes[0] == ATA_DEV_UNKNOWN &&
2229 !(lflags & ATA_LFLAG_ASSUME_CLASS)) {
2230 if (try < max_tries) {
2231 ata_link_printk(link, KERN_WARNING,
2232 "classification failed\n");
2233 rc = -EINVAL;
2234 goto fail;
2235 }
2236
2237 ata_link_printk(link, KERN_WARNING,
2238 "classfication failed, assuming ATA\n");
2239 lflags |= ATA_LFLAG_ASSUME_ATA;
2240 }
2241
Tejun Heoeec59f72008-03-06 13:09:34 +09002242 done:
Tejun Heo416dc9e2007-10-31 10:17:03 +09002243 ata_link_for_each_dev(dev, link) {
2244 /* After the reset, the device state is PIO 0 and the
2245 * controller state is undefined. Reset also wakes up
2246 * drives from sleeping mode.
2247 */
2248 dev->pio_mode = XFER_PIO_0;
2249 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002250
Tejun Heo416dc9e2007-10-31 10:17:03 +09002251 if (ata_link_offline(link))
2252 continue;
Tejun Heo664faf02006-05-31 18:27:50 +09002253
Tejun Heo4ccd3322008-01-08 20:26:12 +09002254 /* apply class override */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002255 if (lflags & ATA_LFLAG_ASSUME_ATA)
2256 classes[dev->devno] = ATA_DEV_ATA;
2257 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2258 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
Tejun Heo022bdb02006-05-15 20:58:22 +09002259 }
2260
Tejun Heo416dc9e2007-10-31 10:17:03 +09002261 /* record current link speed */
2262 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2263 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002264
Tejun Heo416dc9e2007-10-31 10:17:03 +09002265 if (postreset)
2266 postreset(link, classes);
Tejun Heo20952b62006-05-31 18:27:23 +09002267
Tejun Heo416dc9e2007-10-31 10:17:03 +09002268 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002269 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002270 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heoae791c02007-09-23 13:14:12 +09002271
Tejun Heo416dc9e2007-10-31 10:17:03 +09002272 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002273 out:
2274 /* clear hotplug flag */
2275 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002276
2277 spin_lock_irqsave(ap->lock, flags);
2278 ap->pflags &= ~ATA_PFLAG_RESETTING;
2279 spin_unlock_irqrestore(ap->lock, flags);
2280
Tejun Heo022bdb02006-05-15 20:58:22 +09002281 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002282
2283 fail:
2284 if (rc == -ERESTART || try >= max_tries)
2285 goto out;
2286
2287 now = jiffies;
2288 if (time_before(now, deadline)) {
2289 unsigned long delta = deadline - now;
2290
2291 ata_link_printk(link, KERN_WARNING, "reset failed "
2292 "(errno=%d), retrying in %u secs\n",
2293 rc, (jiffies_to_msecs(delta) + 999) / 1000);
2294
2295 while (delta)
2296 delta = schedule_timeout_uninterruptible(delta);
2297 }
2298
2299 if (rc == -EPIPE || try == max_tries - 1)
2300 sata_down_spd_limit(link);
2301 if (hardreset)
2302 reset = hardreset;
2303 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002304}
2305
Tejun Heo02607312007-08-06 18:36:23 +09002306static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09002307 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09002308{
Tejun Heo02607312007-08-06 18:36:23 +09002309 struct ata_port *ap = link->ap;
2310 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002311 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002312 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002313 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09002314 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002315
2316 DPRINTK("ENTER\n");
2317
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002318 /* For PATA drive side cable detection to work, IDENTIFY must
2319 * be done backwards such that PDIAG- is released by the slave
2320 * device before the master device is identified.
2321 */
Tejun Heo02607312007-08-06 18:36:23 +09002322 ata_link_for_each_dev_reverse(dev, link) {
Tejun Heof58229f2007-08-06 18:36:23 +09002323 unsigned int action = ata_eh_dev_action(dev);
2324 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09002325
Tejun Heobff04642006-11-10 18:08:10 +09002326 if (ehc->i.flags & ATA_EHI_DID_RESET)
2327 readid_flags |= ATA_READID_POSTRESET;
2328
Tejun Heo9666f402007-05-04 21:27:47 +02002329 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a2007-09-23 13:19:54 +09002330 WARN_ON(dev->class == ATA_DEV_PMP);
2331
Tejun Heo02607312007-08-06 18:36:23 +09002332 if (ata_link_offline(link)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002333 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002334 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002335 }
2336
Tejun Heo02607312007-08-06 18:36:23 +09002337 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09002338 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2339 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09002340 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002341 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002342
Tejun Heo02607312007-08-06 18:36:23 +09002343 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09002344
Tejun Heobaa1e782006-11-01 18:39:27 +09002345 /* Configuration may have changed, reconfigure
2346 * transfer mode.
2347 */
2348 ehc->i.flags |= ATA_EHI_SETMODE;
2349
zhao, forrest3057ac32006-06-12 12:01:34 +08002350 /* schedule the scsi_rescan_device() here */
2351 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09002352 } else if (dev->class == ATA_DEV_UNKNOWN &&
2353 ehc->tries[dev->devno] &&
2354 ata_class_enabled(ehc->classes[dev->devno])) {
2355 dev->class = ehc->classes[dev->devno];
2356
Tejun Heo633273a2007-09-23 13:19:54 +09002357 if (dev->class == ATA_DEV_PMP)
2358 rc = sata_pmp_attach(dev);
2359 else
2360 rc = ata_dev_read_id(dev, &dev->class,
2361 readid_flags, dev->id);
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002362 switch (rc) {
2363 case 0:
Tejun Heof58229f2007-08-06 18:36:23 +09002364 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002365 break;
2366 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002367 /* IDENTIFY was issued to non-existent
2368 * device. No need to reset. Just
2369 * thaw and kill the device.
2370 */
2371 ata_eh_thaw_port(ap);
2372 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo084fe632006-05-31 18:28:03 +09002373 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002374 default:
2375 dev->class = ATA_DEV_UNKNOWN;
2376 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002377 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002378 }
2379 }
2380
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002381 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09002382 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2383 if (ap->ops->cable_detect)
2384 ap->cbl = ap->ops->cable_detect(ap);
2385 ata_force_cbl(ap);
2386 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002387
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002388 /* Configure new devices forward such that user doesn't see
2389 * device detection messages backwards.
2390 */
Tejun Heo02607312007-08-06 18:36:23 +09002391 ata_link_for_each_dev(dev, link) {
Tejun Heo633273a2007-09-23 13:19:54 +09002392 if (!(new_mask & (1 << dev->devno)) ||
2393 dev->class == ATA_DEV_PMP)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002394 continue;
2395
2396 ehc->i.flags |= ATA_EHI_PRINTINFO;
2397 rc = ata_dev_configure(dev);
2398 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2399 if (rc)
2400 goto err;
2401
2402 spin_lock_irqsave(ap->lock, flags);
2403 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2404 spin_unlock_irqrestore(ap->lock, flags);
2405
2406 /* new device discovered, configure xfermode */
2407 ehc->i.flags |= ATA_EHI_SETMODE;
2408 }
2409
2410 return 0;
2411
2412 err:
2413 *r_failed_dev = dev;
2414 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002415 return rc;
2416}
2417
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002418/**
2419 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2420 * @link: link on which timings will be programmed
2421 * @r_failed_dev: out paramter for failed device
2422 *
2423 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2424 * ata_set_mode() fails, pointer to the failing device is
2425 * returned in @r_failed_dev.
2426 *
2427 * LOCKING:
2428 * PCI/etc. bus probe sem.
2429 *
2430 * RETURNS:
2431 * 0 on success, negative errno otherwise
2432 */
2433int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2434{
2435 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09002436 struct ata_device *dev;
2437 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002438
Tejun Heo76326ac2007-11-27 19:28:59 +09002439 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2440 ata_link_for_each_dev(dev, link) {
2441 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2442 struct ata_ering_entry *ent;
2443
2444 ent = ata_ering_top(&dev->ering);
2445 if (ent)
2446 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2447 }
2448 }
2449
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002450 /* has private set_mode? */
2451 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09002452 rc = ap->ops->set_mode(link, r_failed_dev);
2453 else
2454 rc = ata_do_set_mode(link, r_failed_dev);
2455
2456 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2457 ata_link_for_each_dev(dev, link) {
2458 struct ata_eh_context *ehc = &link->eh_context;
2459 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2460 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2461
2462 if (dev->xfer_mode != saved_xfer_mode ||
2463 ata_ncq_enabled(dev) != saved_ncq)
2464 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2465 }
2466
2467 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002468}
2469
Tejun Heo02607312007-08-06 18:36:23 +09002470static int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002471{
Tejun Heof58229f2007-08-06 18:36:23 +09002472 struct ata_device *dev;
2473 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002474
Tejun Heo02607312007-08-06 18:36:23 +09002475 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002476 if (ata_dev_enabled(dev))
Tejun Heo022bdb02006-05-15 20:58:22 +09002477 cnt++;
2478 return cnt;
2479}
2480
Tejun Heo02607312007-08-06 18:36:23 +09002481static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002482{
Tejun Heof58229f2007-08-06 18:36:23 +09002483 struct ata_device *dev;
2484 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002485
Tejun Heo02607312007-08-06 18:36:23 +09002486 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002487 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09002488 cnt++;
2489 return cnt;
2490}
2491
Tejun Heo02607312007-08-06 18:36:23 +09002492static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002493{
Tejun Heo02607312007-08-06 18:36:23 +09002494 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09002495 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09002496
Tejun Heof9df58c2007-09-23 13:14:13 +09002497 /* skip disabled links */
2498 if (link->flags & ATA_LFLAG_DISABLED)
2499 return 1;
2500
Tejun Heo7c8c2cf2006-07-10 23:18:23 +09002501 /* thaw frozen port, resume link and recover failed devices */
Tejun Heo02607312007-08-06 18:36:23 +09002502 if ((link->ap->pflags & ATA_PFLAG_FROZEN) ||
2503 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_link_nr_enabled(link))
Tejun Heo084fe632006-05-31 18:28:03 +09002504 return 0;
2505
2506 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo02607312007-08-06 18:36:23 +09002507 ata_link_for_each_dev(dev, link) {
Tejun Heo084fe632006-05-31 18:28:03 +09002508 if (dev->class == ATA_DEV_UNKNOWN &&
2509 ehc->classes[dev->devno] != ATA_DEV_NONE)
2510 return 0;
2511 }
2512
2513 return 1;
2514}
2515
Tejun Heo02c05a22007-11-27 19:28:54 +09002516static int ata_eh_schedule_probe(struct ata_device *dev)
2517{
2518 struct ata_eh_context *ehc = &dev->link->eh_context;
2519
2520 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2521 (ehc->did_probe_mask & (1 << dev->devno)))
2522 return 0;
2523
2524 ata_eh_detach_dev(dev);
2525 ata_dev_init(dev);
2526 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09002527 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09002528 ehc->saved_xfer_mode[dev->devno] = 0;
2529 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09002530
2531 return 1;
2532}
2533
Tejun Heo9b1e2652007-08-06 18:36:24 +09002534static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09002535{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09002536 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09002537
2538 ehc->tries[dev->devno]--;
2539
2540 switch (err) {
2541 case -ENODEV:
2542 /* device missing or wrong IDENTIFY data, schedule probing */
2543 ehc->i.probe_mask |= (1 << dev->devno);
2544 case -EINVAL:
2545 /* give it just one more chance */
2546 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2547 case -EIO:
Tejun Heo4fb46152007-10-29 16:45:05 +09002548 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
Tejun Heofee7ca72007-07-01 19:05:58 +09002549 /* This is the last chance, better to slow
2550 * down than lose it.
2551 */
Tejun Heo936fd732007-08-06 18:36:23 +09002552 sata_down_spd_limit(dev->link);
Tejun Heofee7ca72007-07-01 19:05:58 +09002553 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2554 }
2555 }
2556
2557 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2558 /* disable device if it has used up all its chances */
2559 ata_dev_disable(dev);
2560
2561 /* detach if offline */
Tejun Heo936fd732007-08-06 18:36:23 +09002562 if (ata_link_offline(dev->link))
Tejun Heofee7ca72007-07-01 19:05:58 +09002563 ata_eh_detach_dev(dev);
2564
Tejun Heo02c05a22007-11-27 19:28:54 +09002565 /* schedule probe if necessary */
2566 if (ata_eh_schedule_probe(dev))
Tejun Heofee7ca72007-07-01 19:05:58 +09002567 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002568
2569 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09002570 } else {
Tejun Heocf480622008-01-24 00:05:14 +09002571 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002572 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09002573 }
2574}
2575
Tejun Heo022bdb02006-05-15 20:58:22 +09002576/**
2577 * ata_eh_recover - recover host port after error
2578 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09002579 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002580 * @softreset: softreset method (can be NULL)
2581 * @hardreset: hardreset method (can be NULL)
2582 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002583 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09002584 *
2585 * This is the alpha and omega, eum and yang, heart and soul of
2586 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09002587 * recover each link and hotplug requests are recorded in the
2588 * link's eh_context. This function executes all the operations
2589 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09002590 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09002591 *
2592 * LOCKING:
2593 * Kernel thread context (may sleep).
2594 *
2595 * RETURNS:
2596 * 0 on success, -errno on failure.
2597 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002598int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2599 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2600 ata_postreset_fn_t postreset,
2601 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002602{
Tejun Heo9b1e2652007-08-06 18:36:24 +09002603 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09002604 struct ata_device *dev;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002605 int nr_failed_devs, nr_disabled_devs;
2606 int reset, rc;
Tejun Heof9df58c2007-09-23 13:14:13 +09002607 unsigned long flags;
Tejun Heo022bdb02006-05-15 20:58:22 +09002608
2609 DPRINTK("ENTER\n");
2610
2611 /* prep for recovery */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002612 ata_port_for_each_link(link, ap) {
2613 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09002614
Tejun Heof9df58c2007-09-23 13:14:13 +09002615 /* re-enable link? */
2616 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
2617 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
2618 spin_lock_irqsave(ap->lock, flags);
2619 link->flags &= ~ATA_LFLAG_DISABLED;
2620 spin_unlock_irqrestore(ap->lock, flags);
2621 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
2622 }
2623
Tejun Heo9b1e2652007-08-06 18:36:24 +09002624 ata_link_for_each_dev(dev, link) {
Tejun Heofd995f72007-09-23 13:14:12 +09002625 if (link->flags & ATA_LFLAG_NO_RETRY)
2626 ehc->tries[dev->devno] = 1;
2627 else
2628 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09002629
Tejun Heo9b1e2652007-08-06 18:36:24 +09002630 /* collect port action mask recorded in dev actions */
2631 ehc->i.action |= ehc->i.dev_action[dev->devno] &
2632 ~ATA_EH_PERDEV_MASK;
2633 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09002634
Tejun Heo9b1e2652007-08-06 18:36:24 +09002635 /* process hotplug request */
2636 if (dev->flags & ATA_DFLAG_DETACH)
2637 ata_eh_detach_dev(dev);
2638
Tejun Heo02c05a22007-11-27 19:28:54 +09002639 /* schedule probe if necessary */
2640 if (!ata_dev_enabled(dev))
2641 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09002642 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002643 }
2644
2645 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09002646 rc = 0;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002647 nr_failed_devs = 0;
2648 nr_disabled_devs = 0;
2649 reset = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002650
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002651 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09002652 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002653 goto out;
2654
Tejun Heo9b1e2652007-08-06 18:36:24 +09002655 /* prep for EH */
2656 ata_port_for_each_link(link, ap) {
2657 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002658
Tejun Heo9b1e2652007-08-06 18:36:24 +09002659 /* skip EH if possible. */
2660 if (ata_eh_skip_recovery(link))
2661 ehc->i.action = 0;
2662
2663 /* do we need to reset? */
Tejun Heocf480622008-01-24 00:05:14 +09002664 if (ehc->i.action & ATA_EH_RESET)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002665 reset = 1;
2666
2667 ata_link_for_each_dev(dev, link)
2668 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
2669 }
Tejun Heo084fe632006-05-31 18:28:03 +09002670
Tejun Heo022bdb02006-05-15 20:58:22 +09002671 /* reset */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002672 if (reset) {
Tejun Heob06ce3e2007-10-09 15:06:48 +09002673 /* if PMP is attached, this function only deals with
2674 * downstream links, port should stay thawed.
2675 */
2676 if (!ap->nr_pmp_links)
2677 ata_eh_freeze_port(ap);
Tejun Heo022bdb02006-05-15 20:58:22 +09002678
Tejun Heo9b1e2652007-08-06 18:36:24 +09002679 ata_port_for_each_link(link, ap) {
2680 struct ata_eh_context *ehc = &link->eh_context;
2681
Tejun Heocf480622008-01-24 00:05:14 +09002682 if (!(ehc->i.action & ATA_EH_RESET))
Tejun Heo9b1e2652007-08-06 18:36:24 +09002683 continue;
2684
2685 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
2686 prereset, softreset, hardreset,
2687 postreset);
2688 if (rc) {
2689 ata_link_printk(link, KERN_ERR,
2690 "reset failed, giving up\n");
2691 goto out;
2692 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002693 }
2694
Tejun Heob06ce3e2007-10-09 15:06:48 +09002695 if (!ap->nr_pmp_links)
2696 ata_eh_thaw_port(ap);
Tejun Heo022bdb02006-05-15 20:58:22 +09002697 }
2698
Tejun Heo9b1e2652007-08-06 18:36:24 +09002699 /* the rest */
2700 ata_port_for_each_link(link, ap) {
2701 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002702
Tejun Heo9b1e2652007-08-06 18:36:24 +09002703 /* revalidate existing devices and attach new ones */
2704 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09002705 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09002706 goto dev_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002707
Tejun Heo633273a2007-09-23 13:19:54 +09002708 /* if PMP got attached, return, pmp EH will take care of it */
2709 if (link->device->class == ATA_DEV_PMP) {
2710 ehc->i.action = 0;
2711 return 0;
2712 }
2713
Tejun Heo9b1e2652007-08-06 18:36:24 +09002714 /* configure transfer mode if necessary */
2715 if (ehc->i.flags & ATA_EHI_SETMODE) {
2716 rc = ata_set_mode(link, &dev);
2717 if (rc)
2718 goto dev_fail;
2719 ehc->i.flags &= ~ATA_EHI_SETMODE;
2720 }
2721
Tejun Heo3ec25eb2008-03-27 18:37:14 +09002722 if (ehc->i.action & ATA_EH_LPM)
Kristen Carlson Accardica773292007-10-25 00:58:59 -04002723 ata_link_for_each_dev(dev, link)
2724 ata_dev_enable_pm(dev, ap->pm_policy);
2725
Tejun Heo9b1e2652007-08-06 18:36:24 +09002726 /* this link is okay now */
2727 ehc->i.flags = 0;
2728 continue;
2729
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002730dev_fail:
Tejun Heo9b1e2652007-08-06 18:36:24 +09002731 nr_failed_devs++;
2732 if (ata_eh_handle_dev_fail(dev, rc))
2733 nr_disabled_devs++;
2734
Tejun Heob06ce3e2007-10-09 15:06:48 +09002735 if (ap->pflags & ATA_PFLAG_FROZEN) {
2736 /* PMP reset requires working host port.
2737 * Can't retry if it's frozen.
2738 */
2739 if (ap->nr_pmp_links)
2740 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002741 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09002742 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002743 }
2744
Tejun Heo9b1e2652007-08-06 18:36:24 +09002745 if (nr_failed_devs) {
2746 if (nr_failed_devs != nr_disabled_devs) {
2747 ata_port_printk(ap, KERN_WARNING, "failed to recover "
2748 "some devices, retrying in 5 secs\n");
2749 ssleep(5);
2750 } else {
2751 /* no device left to recover, repeat fast */
2752 msleep(500);
2753 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002754
Tejun Heo9b1e2652007-08-06 18:36:24 +09002755 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002756 }
2757
Tejun Heo022bdb02006-05-15 20:58:22 +09002758 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09002759 if (rc && r_failed_link)
2760 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09002761
2762 DPRINTK("EXIT, rc=%d\n", rc);
2763 return rc;
2764}
2765
2766/**
2767 * ata_eh_finish - finish up EH
2768 * @ap: host port to finish EH for
2769 *
2770 * Recovery is complete. Clean up EH states and retry or finish
2771 * failed qcs.
2772 *
2773 * LOCKING:
2774 * None.
2775 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002776void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09002777{
2778 int tag;
2779
2780 /* retry or finish qcs */
2781 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2782 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2783
2784 if (!(qc->flags & ATA_QCFLAG_FAILED))
2785 continue;
2786
2787 if (qc->err_mask) {
2788 /* FIXME: Once EH migration is complete,
2789 * generate sense data in this function,
2790 * considering both err_mask and tf.
Tejun Heof90f0822007-10-26 16:12:41 +09002791 *
2792 * There's no point in retrying invalid
2793 * (detected by libata) and non-IO device
2794 * errors (rejected by device). Finish them
2795 * immediately.
Tejun Heo022bdb02006-05-15 20:58:22 +09002796 */
Tejun Heof90f0822007-10-26 16:12:41 +09002797 if ((qc->err_mask & AC_ERR_INVALID) ||
2798 (!(qc->flags & ATA_QCFLAG_IO) &&
2799 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002800 ata_eh_qc_complete(qc);
2801 else
2802 ata_eh_qc_retry(qc);
2803 } else {
2804 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2805 ata_eh_qc_complete(qc);
2806 } else {
2807 /* feed zero TF to sense generation */
2808 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2809 ata_eh_qc_retry(qc);
2810 }
2811 }
2812 }
Tejun Heoda917d62007-09-23 13:14:12 +09002813
2814 /* make sure nr_active_links is zero after EH */
2815 WARN_ON(ap->nr_active_links);
2816 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002817}
2818
2819/**
2820 * ata_do_eh - do standard error handling
2821 * @ap: host port to handle error for
Tejun Heof5914a42006-05-31 18:27:48 +09002822 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002823 * @softreset: softreset method (can be NULL)
2824 * @hardreset: hardreset method (can be NULL)
2825 * @postreset: postreset method (can be NULL)
2826 *
2827 * Perform standard error handling sequence.
2828 *
2829 * LOCKING:
2830 * Kernel thread context (may sleep).
2831 */
Tejun Heof5914a42006-05-31 18:27:48 +09002832void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2833 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2834 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002835{
Tejun Heo9b1e2652007-08-06 18:36:24 +09002836 struct ata_device *dev;
2837 int rc;
2838
2839 ata_eh_autopsy(ap);
2840 ata_eh_report(ap);
2841
2842 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
2843 NULL);
2844 if (rc) {
2845 ata_link_for_each_dev(dev, &ap->link)
2846 ata_dev_disable(dev);
2847 }
2848
Tejun Heo022bdb02006-05-15 20:58:22 +09002849 ata_eh_finish(ap);
2850}
Tejun Heo500530f2006-07-03 16:07:27 +09002851
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002852#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09002853/**
2854 * ata_eh_handle_port_suspend - perform port suspend operation
2855 * @ap: port to suspend
2856 *
2857 * Suspend @ap.
2858 *
2859 * LOCKING:
2860 * Kernel thread context (may sleep).
2861 */
2862static void ata_eh_handle_port_suspend(struct ata_port *ap)
2863{
2864 unsigned long flags;
2865 int rc = 0;
2866
2867 /* are we suspending? */
2868 spin_lock_irqsave(ap->lock, flags);
2869 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2870 ap->pm_mesg.event == PM_EVENT_ON) {
2871 spin_unlock_irqrestore(ap->lock, flags);
2872 return;
2873 }
2874 spin_unlock_irqrestore(ap->lock, flags);
2875
2876 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2877
Tejun Heo64578a3de2007-05-15 03:28:16 +09002878 /* tell ACPI we're suspending */
2879 rc = ata_acpi_on_suspend(ap);
2880 if (rc)
2881 goto out;
2882
Tejun Heo500530f2006-07-03 16:07:27 +09002883 /* suspend */
2884 ata_eh_freeze_port(ap);
2885
2886 if (ap->ops->port_suspend)
2887 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2888
Shaohua Libd3adca2007-11-02 09:32:38 +08002889 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a3de2007-05-15 03:28:16 +09002890 out:
Tejun Heo500530f2006-07-03 16:07:27 +09002891 /* report result */
2892 spin_lock_irqsave(ap->lock, flags);
2893
2894 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2895 if (rc == 0)
2896 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a3de2007-05-15 03:28:16 +09002897 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09002898 ata_port_schedule_eh(ap);
2899
2900 if (ap->pm_result) {
2901 *ap->pm_result = rc;
2902 ap->pm_result = NULL;
2903 }
2904
2905 spin_unlock_irqrestore(ap->lock, flags);
2906
2907 return;
2908}
2909
2910/**
2911 * ata_eh_handle_port_resume - perform port resume operation
2912 * @ap: port to resume
2913 *
2914 * Resume @ap.
2915 *
Tejun Heo500530f2006-07-03 16:07:27 +09002916 * LOCKING:
2917 * Kernel thread context (may sleep).
2918 */
2919static void ata_eh_handle_port_resume(struct ata_port *ap)
2920{
Tejun Heo500530f2006-07-03 16:07:27 +09002921 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02002922 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09002923
2924 /* are we resuming? */
2925 spin_lock_irqsave(ap->lock, flags);
2926 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2927 ap->pm_mesg.event != PM_EVENT_ON) {
2928 spin_unlock_irqrestore(ap->lock, flags);
2929 return;
2930 }
2931 spin_unlock_irqrestore(ap->lock, flags);
2932
Tejun Heo9666f402007-05-04 21:27:47 +02002933 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09002934
Shaohua Libd3adca2007-11-02 09:32:38 +08002935 ata_acpi_set_state(ap, PMSG_ON);
2936
Tejun Heo500530f2006-07-03 16:07:27 +09002937 if (ap->ops->port_resume)
2938 rc = ap->ops->port_resume(ap);
2939
Tejun Heo67465442007-05-15 03:28:16 +09002940 /* tell ACPI that we're resuming */
2941 ata_acpi_on_resume(ap);
2942
Tejun Heo9666f402007-05-04 21:27:47 +02002943 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09002944 spin_lock_irqsave(ap->lock, flags);
2945 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2946 if (ap->pm_result) {
2947 *ap->pm_result = rc;
2948 ap->pm_result = NULL;
2949 }
2950 spin_unlock_irqrestore(ap->lock, flags);
2951}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002952#endif /* CONFIG_PM */