blob: 751dad0138aea6ffab3f062797e38d15df469166 [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
Tejun Heo071f44b2008-04-07 22:47:22 +0900876 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +0900877 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +0900878 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +0900879 /* 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 Heo955e57d2007-08-06 18:36:23 +09001082 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001083
1084 if (!(ehc->i.flags & ATA_EHI_QUIET))
Tejun Heo1cdaf532006-07-03 16:07:26 +09001085 ap->pflags |= ATA_PFLAG_RECOVERED;
1086
Jeff Garzikba6a1302006-06-22 23:46:10 -04001087 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001088}
1089
1090/**
Tejun Heo47005f22006-06-19 18:27:23 +09001091 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001092* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001093 * @dev: target ATA dev for per-dev action (can be NULL)
1094 * @action: action just completed
1095 *
1096 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001097 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001098 *
1099 * LOCKING:
1100 * None.
1101 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001102void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1103 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001104{
Tejun Heo955e57d2007-08-06 18:36:23 +09001105 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001106
Tejun Heo955e57d2007-08-06 18:36:23 +09001107 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001108}
1109
1110/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001111 * ata_err_string - convert err_mask to descriptive string
1112 * @err_mask: error mask to convert to string
1113 *
1114 * Convert @err_mask to descriptive string. Errors are
1115 * prioritized according to severity and only the most severe
1116 * error is reported.
1117 *
1118 * LOCKING:
1119 * None.
1120 *
1121 * RETURNS:
1122 * Descriptive string for @err_mask
1123 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001124static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001125{
1126 if (err_mask & AC_ERR_HOST_BUS)
1127 return "host bus error";
1128 if (err_mask & AC_ERR_ATA_BUS)
1129 return "ATA bus error";
1130 if (err_mask & AC_ERR_TIMEOUT)
1131 return "timeout";
1132 if (err_mask & AC_ERR_HSM)
1133 return "HSM violation";
1134 if (err_mask & AC_ERR_SYSTEM)
1135 return "internal error";
1136 if (err_mask & AC_ERR_MEDIA)
1137 return "media error";
1138 if (err_mask & AC_ERR_INVALID)
1139 return "invalid argument";
1140 if (err_mask & AC_ERR_DEV)
1141 return "device error";
1142 return "unknown error";
1143}
1144
1145/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001146 * ata_read_log_page - read a specific log page
1147 * @dev: target device
1148 * @page: page to read
1149 * @buf: buffer to store read page
1150 * @sectors: number of sectors to read
1151 *
1152 * Read log page using READ_LOG_EXT command.
1153 *
1154 * LOCKING:
1155 * Kernel thread context (may sleep).
1156 *
1157 * RETURNS:
1158 * 0 on success, AC_ERR_* mask otherwise.
1159 */
1160static unsigned int ata_read_log_page(struct ata_device *dev,
1161 u8 page, void *buf, unsigned int sectors)
1162{
1163 struct ata_taskfile tf;
1164 unsigned int err_mask;
1165
1166 DPRINTK("read log page - page %d\n", page);
1167
1168 ata_tf_init(dev, &tf);
1169 tf.command = ATA_CMD_READ_LOG_EXT;
1170 tf.lbal = page;
1171 tf.nsect = sectors;
1172 tf.hob_nsect = sectors >> 8;
1173 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1174 tf.protocol = ATA_PROT_PIO;
1175
1176 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001177 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001178
1179 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1180 return err_mask;
1181}
1182
1183/**
1184 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1185 * @dev: Device to read log page 10h from
1186 * @tag: Resulting tag of the failed command
1187 * @tf: Resulting taskfile registers of the failed command
1188 *
1189 * Read log page 10h to obtain NCQ error details and clear error
1190 * condition.
1191 *
1192 * LOCKING:
1193 * Kernel thread context (may sleep).
1194 *
1195 * RETURNS:
1196 * 0 on success, -errno otherwise.
1197 */
1198static int ata_eh_read_log_10h(struct ata_device *dev,
1199 int *tag, struct ata_taskfile *tf)
1200{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001201 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001202 unsigned int err_mask;
1203 u8 csum;
1204 int i;
1205
1206 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1207 if (err_mask)
1208 return -EIO;
1209
1210 csum = 0;
1211 for (i = 0; i < ATA_SECT_SIZE; i++)
1212 csum += buf[i];
1213 if (csum)
1214 ata_dev_printk(dev, KERN_WARNING,
1215 "invalid checksum 0x%x on log page 10h\n", csum);
1216
1217 if (buf[0] & 0x80)
1218 return -ENOENT;
1219
1220 *tag = buf[0] & 0x1f;
1221
1222 tf->command = buf[2];
1223 tf->feature = buf[3];
1224 tf->lbal = buf[4];
1225 tf->lbam = buf[5];
1226 tf->lbah = buf[6];
1227 tf->device = buf[7];
1228 tf->hob_lbal = buf[8];
1229 tf->hob_lbam = buf[9];
1230 tf->hob_lbah = buf[10];
1231 tf->nsect = buf[12];
1232 tf->hob_nsect = buf[13];
1233
1234 return 0;
1235}
1236
1237/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001238 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1239 * @dev: device to perform REQUEST_SENSE to
1240 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1241 *
1242 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1243 * SENSE. This function is EH helper.
1244 *
1245 * LOCKING:
1246 * Kernel thread context (may sleep).
1247 *
1248 * RETURNS:
1249 * 0 on success, AC_ERR_* mask on failure
1250 */
Albert Lee56287762007-04-02 11:30:46 +08001251static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
Tejun Heo022bdb02006-05-15 20:58:22 +09001252{
Albert Lee56287762007-04-02 11:30:46 +08001253 struct ata_device *dev = qc->dev;
1254 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001255 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001256 struct ata_taskfile tf;
1257 u8 cdb[ATAPI_CDB_LEN];
1258
1259 DPRINTK("ATAPI request sense\n");
1260
Tejun Heo022bdb02006-05-15 20:58:22 +09001261 /* FIXME: is this needed? */
1262 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1263
Albert Lee56287762007-04-02 11:30:46 +08001264 /* initialize sense_buf with the error register,
1265 * for the case where they are -not- overwritten
1266 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001267 sense_buf[0] = 0x70;
Albert Lee56287762007-04-02 11:30:46 +08001268 sense_buf[2] = qc->result_tf.feature >> 4;
1269
Jeff Garzika617c092007-05-21 20:14:23 -04001270 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001271 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001272
1273 memset(cdb, 0, ATAPI_CDB_LEN);
1274 cdb[0] = REQUEST_SENSE;
1275 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1276
1277 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1278 tf.command = ATA_CMD_PACKET;
1279
1280 /* is it pointless to prefer PIO for "safety reasons"? */
1281 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001282 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001283 tf.feature |= ATAPI_PKT_DMA;
1284 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001285 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001286 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1287 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001288 }
1289
1290 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001291 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001292}
1293
1294/**
1295 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001296 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001297 *
1298 * Analyze SError if available and further determine cause of
1299 * failure.
1300 *
1301 * LOCKING:
1302 * None.
1303 */
Tejun Heo02607312007-08-06 18:36:23 +09001304static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001305{
Tejun Heo02607312007-08-06 18:36:23 +09001306 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001307 u32 serror = ehc->i.serror;
1308 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001309 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001310
1311 if (serror & SERR_PERSISTENT) {
1312 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001313 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001314 }
1315 if (serror &
1316 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1317 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001318 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001319 }
1320 if (serror & SERR_PROTOCOL) {
1321 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001322 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001323 }
1324 if (serror & SERR_INTERNAL) {
1325 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001326 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001327 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001328
1329 /* Determine whether a hotplug event has occurred. Both
1330 * SError.N/X are considered hotplug events for enabled or
1331 * host links. For disabled PMP links, only N bit is
1332 * considered as X bit is left at 1 for link plugging.
1333 */
1334 hotplug_mask = 0;
1335
1336 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1337 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1338 else
1339 hotplug_mask = SERR_PHYRDY_CHG;
1340
1341 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001342 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001343
1344 ehc->i.err_mask |= err_mask;
1345 ehc->i.action |= action;
1346}
1347
1348/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001349 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001350 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001351 *
1352 * Read log page 10h, determine the offending qc and acquire
1353 * error status TF. For NCQ device errors, all LLDDs have to do
1354 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1355 * care of the rest.
1356 *
1357 * LOCKING:
1358 * Kernel thread context (may sleep).
1359 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001360void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001361{
Tejun Heo02607312007-08-06 18:36:23 +09001362 struct ata_port *ap = link->ap;
1363 struct ata_eh_context *ehc = &link->eh_context;
1364 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001365 struct ata_queued_cmd *qc;
1366 struct ata_taskfile tf;
1367 int tag, rc;
1368
1369 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001370 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001371 return;
1372
1373 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001374 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001375 return;
1376
1377 /* has LLDD analyzed already? */
1378 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1379 qc = __ata_qc_from_tag(ap, tag);
1380
1381 if (!(qc->flags & ATA_QCFLAG_FAILED))
1382 continue;
1383
1384 if (qc->err_mask)
1385 return;
1386 }
1387
1388 /* okay, this error is ours */
1389 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1390 if (rc) {
Tejun Heo02607312007-08-06 18:36:23 +09001391 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001392 "(errno=%d)\n", rc);
1393 return;
1394 }
1395
Tejun Heo02607312007-08-06 18:36:23 +09001396 if (!(link->sactive & (1 << tag))) {
1397 ata_link_printk(link, KERN_ERR, "log page 10h reported "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001398 "inactive tag %d\n", tag);
1399 return;
1400 }
1401
1402 /* we've got the perpetrator, condemn it */
1403 qc = __ata_qc_from_tag(ap, tag);
1404 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001405 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001406 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001407 ehc->i.err_mask &= ~AC_ERR_DEV;
1408}
1409
1410/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001411 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1412 * @qc: qc to analyze
1413 * @tf: Taskfile registers to analyze
1414 *
1415 * Analyze taskfile of @qc and further determine cause of
1416 * failure. This function also requests ATAPI sense data if
1417 * avaliable.
1418 *
1419 * LOCKING:
1420 * Kernel thread context (may sleep).
1421 *
1422 * RETURNS:
1423 * Determined recovery action
1424 */
1425static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1426 const struct ata_taskfile *tf)
1427{
1428 unsigned int tmp, action = 0;
1429 u8 stat = tf->command, err = tf->feature;
1430
1431 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1432 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001433 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001434 }
1435
Tejun Heoa51d6442007-03-20 15:24:11 +09001436 if (stat & (ATA_ERR | ATA_DF))
1437 qc->err_mask |= AC_ERR_DEV;
1438 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001439 return 0;
1440
1441 switch (qc->dev->class) {
1442 case ATA_DEV_ATA:
1443 if (err & ATA_ICRC)
1444 qc->err_mask |= AC_ERR_ATA_BUS;
1445 if (err & ATA_UNC)
1446 qc->err_mask |= AC_ERR_MEDIA;
1447 if (err & ATA_IDNF)
1448 qc->err_mask |= AC_ERR_INVALID;
1449 break;
1450
1451 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001452 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Albert Lee56287762007-04-02 11:30:46 +08001453 tmp = atapi_eh_request_sense(qc);
Tejun Heoa569a302006-11-21 10:40:51 +09001454 if (!tmp) {
1455 /* ATA_QCFLAG_SENSE_VALID is used to
1456 * tell atapi_qc_complete() that sense
1457 * data is already valid.
1458 *
1459 * TODO: interpret sense data and set
1460 * appropriate err_mask.
1461 */
1462 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1463 } else
1464 qc->err_mask |= tmp;
1465 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001466 }
1467
1468 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001469 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001470
1471 return action;
1472}
1473
Tejun Heo76326ac2007-11-27 19:28:59 +09001474static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1475 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001476{
Tejun Heo76326ac2007-11-27 19:28:59 +09001477 int base = 0;
1478
1479 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1480 *xfer_ok = 1;
1481
1482 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001483 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001484
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001485 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001486 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001487
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001488 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001489 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001490
Tejun Heo3884f7b2007-11-27 19:28:56 +09001491 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001492 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001493 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001494 if ((err_mask &
1495 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001496 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001497 }
1498
1499 return 0;
1500}
1501
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001502struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001503 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001504 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001505 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001506};
1507
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001508static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001509{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001510 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001511 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001512
1513 if (ent->timestamp < arg->since)
1514 return -1;
1515
Tejun Heo76326ac2007-11-27 19:28:59 +09001516 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1517 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001518 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001519
Tejun Heo022bdb02006-05-15 20:58:22 +09001520 return 0;
1521}
1522
1523/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001524 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001525 * @dev: Device of interest
1526 *
1527 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001528 * whether NCQ needs to be turned off, transfer speed should be
1529 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001530 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001531 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001532 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001533 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1534 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001535 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001536 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001537 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001538 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1539 * data transfer hasn't been verified.
1540 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001541 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001542 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001543 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001544 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001545 * SPEED_DOWN : Speed down transfer speed but don't fall back
1546 * to PIO.
1547 *
1548 * FALLBACK_TO_PIO : Fall back to PIO.
1549 *
1550 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001551 * taken per error. An action triggered by non-DUBIOUS errors
1552 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1553 * This is to expedite speed down decisions right after device is
1554 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001555 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001556 * The followings are speed down rules. #1 and #2 deal with
1557 * DUBIOUS errors.
1558 *
1559 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1560 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1561 *
1562 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1563 * occurred during last 5 mins, NCQ_OFF.
1564 *
1565 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Tejun Heo3884f7b2007-11-27 19:28:56 +09001566 * ocurred during last 5 mins, FALLBACK_TO_PIO
1567 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001568 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001569 * during last 10 mins, NCQ_OFF.
1570 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001571 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001572 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001573 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001574 * LOCKING:
1575 * Inherited from caller.
1576 *
1577 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001578 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001579 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001580static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001581{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001582 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1583 u64 j64 = get_jiffies_64();
1584 struct speed_down_verdict_arg arg;
1585 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001586
Tejun Heo3884f7b2007-11-27 19:28:56 +09001587 /* scan past 5 mins of error history */
1588 memset(&arg, 0, sizeof(arg));
1589 arg.since = j64 - min(j64, j5mins);
1590 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1591
Tejun Heo76326ac2007-11-27 19:28:59 +09001592 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1593 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1594 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1595 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1596
1597 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1598 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1599 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1600
Tejun Heo3884f7b2007-11-27 19:28:56 +09001601 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1602 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001603 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001604 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1605
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001606 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001607 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001608 arg.since = j64 - min(j64, j10mins);
1609 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001610
Tejun Heo3884f7b2007-11-27 19:28:56 +09001611 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1612 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001613 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001614
1615 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1616 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001617 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001618 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001619
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001620 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001621}
1622
1623/**
1624 * ata_eh_speed_down - record error and speed down if necessary
1625 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001626 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001627 * @err_mask: err_mask of the error
1628 *
1629 * Record error and examine error history to determine whether
1630 * adjusting transmission speed is necessary. It also sets
1631 * transmission limits appropriately if such adjustment is
1632 * necessary.
1633 *
1634 * LOCKING:
1635 * Kernel thread context (may sleep).
1636 *
1637 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001638 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001639 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001640static unsigned int ata_eh_speed_down(struct ata_device *dev,
1641 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001642{
Tejun Heo3884f7b2007-11-27 19:28:56 +09001643 struct ata_link *link = dev->link;
Tejun Heo76326ac2007-11-27 19:28:59 +09001644 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001645 unsigned int verdict;
1646 unsigned int action = 0;
1647
1648 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001649 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001650 return 0;
1651
1652 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001653 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001654 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001655
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001656 /* turn off NCQ? */
1657 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1658 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1659 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1660 dev->flags |= ATA_DFLAG_NCQ_OFF;
1661 ata_dev_printk(dev, KERN_WARNING,
1662 "NCQ disabled due to excessive errors\n");
1663 goto done;
1664 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001665
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001666 /* speed down? */
1667 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1668 /* speed down SATA link speed if possible */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001669 if (sata_down_spd_limit(link) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001670 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001671 goto done;
1672 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001673
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001674 /* lower transfer mode */
1675 if (dev->spdn_cnt < 2) {
1676 static const int dma_dnxfer_sel[] =
1677 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1678 static const int pio_dnxfer_sel[] =
1679 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1680 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001681
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001682 if (dev->xfer_shift != ATA_SHIFT_PIO)
1683 sel = dma_dnxfer_sel[dev->spdn_cnt];
1684 else
1685 sel = pio_dnxfer_sel[dev->spdn_cnt];
1686
1687 dev->spdn_cnt++;
1688
1689 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001690 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001691 goto done;
1692 }
1693 }
1694 }
1695
1696 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09001697 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001698 */
1699 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09001700 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001701 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1702 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1703 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09001704 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001705 goto done;
1706 }
1707 }
1708
Tejun Heo022bdb02006-05-15 20:58:22 +09001709 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001710 done:
1711 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09001712 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1713 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001714 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001715}
1716
1717/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001718 * ata_eh_link_autopsy - analyze error and determine recovery action
1719 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09001720 *
Tejun Heo02607312007-08-06 18:36:23 +09001721 * Analyze why @link failed and determine which recovery actions
1722 * are needed. This function also sets more detailed AC_ERR_*
1723 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09001724 *
1725 * LOCKING:
1726 * Kernel thread context (may sleep).
1727 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001728static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001729{
Tejun Heo02607312007-08-06 18:36:23 +09001730 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09001731 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09001732 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001733 unsigned int all_err_mask = 0, eflags = 0;
1734 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09001735 u32 serror;
1736 int rc;
1737
1738 DPRINTK("ENTER\n");
1739
Tejun Heo1cdaf532006-07-03 16:07:26 +09001740 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1741 return;
1742
Tejun Heo022bdb02006-05-15 20:58:22 +09001743 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09001744 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09001745 if (rc == 0) {
1746 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09001747 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09001748 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09001749 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09001750 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09001751 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09001752 ehc->i.err_mask |= AC_ERR_OTHER;
1753 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001754
Tejun Heoe8ee8452006-05-15 21:03:46 +09001755 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09001756 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001757
Tejun Heo022bdb02006-05-15 20:58:22 +09001758 /* any real error trumps AC_ERR_OTHER */
1759 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1760 ehc->i.err_mask &= ~AC_ERR_OTHER;
1761
1762 all_err_mask |= ehc->i.err_mask;
1763
1764 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1765 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1766
Tejun Heo02607312007-08-06 18:36:23 +09001767 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001768 continue;
1769
1770 /* inherit upper level err_mask */
1771 qc->err_mask |= ehc->i.err_mask;
1772
Tejun Heo022bdb02006-05-15 20:58:22 +09001773 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001774 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001775
1776 /* DEV errors are probably spurious in case of ATA_BUS error */
1777 if (qc->err_mask & AC_ERR_ATA_BUS)
1778 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1779 AC_ERR_INVALID);
1780
1781 /* any real error trumps unknown error */
1782 if (qc->err_mask & ~AC_ERR_OTHER)
1783 qc->err_mask &= ~AC_ERR_OTHER;
1784
1785 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09001786 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09001787 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09001788
Tejun Heo03faab72008-03-27 19:14:24 +09001789 /* determine whether the command is worth retrying */
1790 if (!(qc->err_mask & AC_ERR_INVALID) &&
1791 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1792 qc->flags |= ATA_QCFLAG_RETRY;
1793
Tejun Heo022bdb02006-05-15 20:58:22 +09001794 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001795 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09001796 all_err_mask |= qc->err_mask;
1797 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001798 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09001799 }
1800
Tejun Heoa20f33f2006-05-16 12:58:24 +09001801 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001802 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001803 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09001804 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001805 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1806 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001807 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001808
Tejun Heodfcc1732007-10-31 10:17:05 +09001809 /* If we have offending qcs and the associated failed device,
1810 * perform per-dev EH action only on the offending device.
1811 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001812 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09001813 ehc->i.dev_action[ehc->i.dev->devno] |=
1814 ehc->i.action & ATA_EH_PERDEV_MASK;
1815 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09001816 }
1817
Tejun Heo2695e362008-01-10 13:41:23 +09001818 /* propagate timeout to host link */
1819 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1820 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1821
1822 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09001823 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09001824 if (!dev && ((ata_link_max_devices(link) == 1 &&
1825 ata_dev_enabled(link->device))))
1826 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09001827
Tejun Heo76326ac2007-11-27 19:28:59 +09001828 if (dev) {
1829 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1830 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001831 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09001832 }
Tejun Heodfcc1732007-10-31 10:17:05 +09001833
Tejun Heo022bdb02006-05-15 20:58:22 +09001834 DPRINTK("EXIT\n");
1835}
1836
1837/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001838 * ata_eh_autopsy - analyze error and determine recovery action
1839 * @ap: host port to perform autopsy on
1840 *
1841 * Analyze all links of @ap and determine why they failed and
1842 * which recovery actions are needed.
1843 *
1844 * LOCKING:
1845 * Kernel thread context (may sleep).
1846 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001847void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09001848{
1849 struct ata_link *link;
1850
Tejun Heo2695e362008-01-10 13:41:23 +09001851 ata_port_for_each_link(link, ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09001852 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09001853
1854 /* Autopsy of fanout ports can affect host link autopsy.
1855 * Perform host link autopsy last.
1856 */
Tejun Heo071f44b2008-04-07 22:47:22 +09001857 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09001858 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09001859}
1860
1861/**
1862 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09001863 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09001864 *
1865 * Report EH to user.
1866 *
1867 * LOCKING:
1868 * None.
1869 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001870static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001871{
Tejun Heo02607312007-08-06 18:36:23 +09001872 struct ata_port *ap = link->ap;
1873 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001874 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09001875 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09001876 int tag, nr_failed = 0;
1877
Tejun Heo94ff3d52007-10-09 14:57:56 +09001878 if (ehc->i.flags & ATA_EHI_QUIET)
1879 return;
1880
Tejun Heo022bdb02006-05-15 20:58:22 +09001881 desc = NULL;
1882 if (ehc->i.desc[0] != '\0')
1883 desc = ehc->i.desc;
1884
1885 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1886 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1887
Tejun Heoe027bd32007-10-26 16:19:26 +09001888 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link ||
1889 ((qc->flags & ATA_QCFLAG_QUIET) &&
1890 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09001891 continue;
1892 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1893 continue;
1894
1895 nr_failed++;
1896 }
1897
1898 if (!nr_failed && !ehc->i.err_mask)
1899 return;
1900
1901 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09001902 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09001903 frozen = " frozen";
1904
Tejun Heoa1e10f72007-08-18 13:28:49 +09001905 memset(tries_buf, 0, sizeof(tries_buf));
1906 if (ap->eh_tries < ATA_EH_MAX_TRIES)
1907 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
1908 ap->eh_tries);
1909
Tejun Heo022bdb02006-05-15 20:58:22 +09001910 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09001911 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09001912 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1913 ehc->i.err_mask, link->sactive, ehc->i.serror,
1914 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001915 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09001916 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001917 } else {
Tejun Heo02607312007-08-06 18:36:23 +09001918 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09001919 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1920 ehc->i.err_mask, link->sactive, ehc->i.serror,
1921 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001922 if (desc)
Tejun Heo02607312007-08-06 18:36:23 +09001923 ata_link_printk(link, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09001924 }
1925
Robert Hancock1333e192007-10-02 11:22:02 -04001926 if (ehc->i.serror)
1927 ata_port_printk(ap, KERN_ERR,
1928 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
1929 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
1930 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
1931 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
1932 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
1933 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
1934 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
1935 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
1936 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
1937 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
1938 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
1939 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
1940 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
1941 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
1942 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
1943 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
1944 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001945 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04001946
Tejun Heo022bdb02006-05-15 20:58:22 +09001947 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1948 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09001949 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09001950 const u8 *cdb = qc->cdb;
1951 char data_buf[20] = "";
1952 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09001953
Tejun Heo02607312007-08-06 18:36:23 +09001954 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1955 qc->dev->link != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001956 continue;
1957
Tejun Heoabb6a882007-11-28 23:16:09 +09001958 if (qc->dma_dir != DMA_NONE) {
1959 static const char *dma_str[] = {
1960 [DMA_BIDIRECTIONAL] = "bidi",
1961 [DMA_TO_DEVICE] = "out",
1962 [DMA_FROM_DEVICE] = "in",
1963 };
1964 static const char *prot_str[] = {
1965 [ATA_PROT_PIO] = "pio",
1966 [ATA_PROT_DMA] = "dma",
1967 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05001968 [ATAPI_PROT_PIO] = "pio",
1969 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09001970 };
1971
1972 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
1973 prot_str[qc->tf.protocol], qc->nbytes,
1974 dma_str[qc->dma_dir]);
1975 }
1976
Jeff Garzike39eec12007-12-01 18:05:39 -05001977 if (ata_is_atapi(qc->tf.protocol))
Tejun Heoabb6a882007-11-28 23:16:09 +09001978 snprintf(cdb_buf, sizeof(cdb_buf),
1979 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
1980 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
1981 cdb[0], cdb[1], cdb[2], cdb[3],
1982 cdb[4], cdb[5], cdb[6], cdb[7],
1983 cdb[8], cdb[9], cdb[10], cdb[11],
1984 cdb[12], cdb[13], cdb[14], cdb[15]);
1985
Tejun Heo8a937582006-11-14 22:36:12 +09001986 ata_dev_printk(qc->dev, KERN_ERR,
1987 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09001988 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09001989 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09001990 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09001991 cmd->command, cmd->feature, cmd->nsect,
1992 cmd->lbal, cmd->lbam, cmd->lbah,
1993 cmd->hob_feature, cmd->hob_nsect,
1994 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09001995 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09001996 res->command, res->feature, res->nsect,
1997 res->lbal, res->lbam, res->lbah,
1998 res->hob_feature, res->hob_nsect,
1999 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002000 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2001 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002002
2003 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002004 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002005 if (res->command & ATA_BUSY)
2006 ata_dev_printk(qc->dev, KERN_ERR,
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002007 "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002008 else
2009 ata_dev_printk(qc->dev, KERN_ERR,
2010 "status: { %s%s%s%s}\n",
2011 res->command & ATA_DRDY ? "DRDY " : "",
2012 res->command & ATA_DF ? "DF " : "",
2013 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002014 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002015 }
2016
2017 if (cmd->command != ATA_CMD_PACKET &&
2018 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2019 ATA_ABORTED)))
2020 ata_dev_printk(qc->dev, KERN_ERR,
2021 "error: { %s%s%s%s}\n",
2022 res->feature & ATA_ICRC ? "ICRC " : "",
2023 res->feature & ATA_UNC ? "UNC " : "",
2024 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002025 res->feature & ATA_ABORTED ? "ABRT " : "");
Tejun Heo022bdb02006-05-15 20:58:22 +09002026 }
2027}
2028
Tejun Heo9b1e2652007-08-06 18:36:24 +09002029/**
2030 * ata_eh_report - report error handling to user
2031 * @ap: ATA port to report EH about
2032 *
2033 * Report EH to user.
2034 *
2035 * LOCKING:
2036 * None.
2037 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002038void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002039{
2040 struct ata_link *link;
2041
2042 __ata_port_for_each_link(link, ap)
2043 ata_eh_link_report(link);
2044}
2045
Tejun Heocc0680a2007-08-06 18:36:23 +09002046static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heod4b2bab2007-02-02 16:50:52 +09002047 unsigned int *classes, unsigned long deadline)
Tejun Heod87fa382006-05-31 18:28:24 +09002048{
Tejun Heof58229f2007-08-06 18:36:23 +09002049 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002050
Tejun Heocc0680a2007-08-06 18:36:23 +09002051 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002052 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002053
Tejun Heof0465192008-05-19 01:15:08 +09002054 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002055}
2056
Tejun Heoae791c02007-09-23 13:14:12 +09002057static int ata_eh_followup_srst_needed(struct ata_link *link,
2058 int rc, int classify,
Tejun Heo664faf02006-05-31 18:27:50 +09002059 const unsigned int *classes)
2060{
Tejun Heo45db2f62008-04-08 01:46:56 +09002061 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002062 return 0;
Tejun Heo305d2a12008-04-07 22:47:20 +09002063 if (rc == -EAGAIN) {
2064 if (classify)
2065 return 1;
2066 rc = 0;
2067 }
Tejun Heo664faf02006-05-31 18:27:50 +09002068 if (rc != 0)
2069 return 0;
Tejun Heo071f44b2008-04-07 22:47:22 +09002070 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002071 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002072 return 0;
2073}
2074
Tejun Heofb7fd612007-09-23 13:14:12 +09002075int ata_eh_reset(struct ata_link *link, int classify,
2076 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2077 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002078{
Tejun Heo416dc9e2007-10-31 10:17:03 +09002079 const int max_tries = ARRAY_SIZE(ata_eh_reset_timeouts);
Tejun Heoafaa5c32007-10-09 15:06:10 +09002080 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09002081 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09002082 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002083 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002084 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heo31daabd2007-02-02 16:50:52 +09002085 int try = 0;
Tejun Heof58229f2007-08-06 18:36:23 +09002086 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002087 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002088 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002089 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002090 u32 sstatus;
Tejun Heof0465192008-05-19 01:15:08 +09002091 int nr_known, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002092
Tejun Heo932648b2008-05-19 01:15:06 +09002093 /*
2094 * Prepare to reset
2095 */
Tejun Heoafaa5c32007-10-09 15:06:10 +09002096 spin_lock_irqsave(ap->lock, flags);
2097 ap->pflags |= ATA_PFLAG_RESETTING;
2098 spin_unlock_irqrestore(ap->lock, flags);
2099
Tejun Heocf480622008-01-24 00:05:14 +09002100 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002101
Tejun Heocdeab112007-10-29 16:41:09 +09002102 ata_link_for_each_dev(dev, link) {
2103 /* If we issue an SRST then an ATA drive (not ATAPI)
2104 * may change configuration and be in PIO0 timing. If
2105 * we do a hard reset (or are coming from power on)
2106 * this is true for ATA or ATAPI. Until we've set a
2107 * suitable controller mode we should not touch the
2108 * bus as we may be talking too fast.
2109 */
2110 dev->pio_mode = XFER_PIO_0;
2111
2112 /* If the controller has a pio mode setup function
2113 * then use it to set the chipset to rights. Don't
2114 * touch the DMA setup as that will be dealt with when
2115 * configuring devices.
2116 */
2117 if (ap->ops->set_piomode)
2118 ap->ops->set_piomode(ap, dev);
2119 }
2120
Tejun Heocf480622008-01-24 00:05:14 +09002121 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002122 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002123 ehc->i.action &= ~ATA_EH_RESET;
2124 if (hardreset) {
2125 reset = hardreset;
2126 ehc->i.action = ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002127 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002128 reset = softreset;
2129 ehc->i.action = ATA_EH_SOFTRESET;
2130 }
Tejun Heof5914a42006-05-31 18:27:48 +09002131
2132 if (prereset) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002133 rc = prereset(link, jiffies + ATA_EH_PRERESET_TIMEOUT);
Tejun Heof5914a42006-05-31 18:27:48 +09002134 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002135 if (rc == -ENOENT) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002136 ata_link_printk(link, KERN_DEBUG,
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002137 "port disabled. ignoring.\n");
Tejun Heocf480622008-01-24 00:05:14 +09002138 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002139
Tejun Heo936fd732007-08-06 18:36:23 +09002140 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002141 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002142
2143 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002144 } else
Tejun Heocc0680a2007-08-06 18:36:23 +09002145 ata_link_printk(link, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09002146 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002147 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002148 }
Tejun Heof5914a42006-05-31 18:27:48 +09002149
Tejun Heo932648b2008-05-19 01:15:06 +09002150 /* prereset() might have cleared ATA_EH_RESET. If so,
2151 * bang classes and return.
2152 */
2153 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2154 ata_link_for_each_dev(dev, link)
2155 classes[dev->devno] = ATA_DEV_NONE;
2156 rc = 0;
2157 goto out;
2158 }
Tejun Heof5914a42006-05-31 18:27:48 +09002159 }
2160
Tejun Heo022bdb02006-05-15 20:58:22 +09002161 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002162 /*
2163 * Perform reset
2164 */
Tejun Heodc98c322008-05-19 01:15:07 +09002165 if (ata_is_host_link(link))
2166 ata_eh_freeze_port(ap);
2167
Tejun Heo31daabd2007-02-02 16:50:52 +09002168 deadline = jiffies + ata_eh_reset_timeouts[try++];
2169
Tejun Heo932648b2008-05-19 01:15:06 +09002170 if (reset) {
2171 if (verbose)
2172 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2173 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002174
Tejun Heo932648b2008-05-19 01:15:06 +09002175 /* mark that this EH session started with reset */
2176 if (reset == hardreset)
2177 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2178 else
2179 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002180
Tejun Heo932648b2008-05-19 01:15:06 +09002181 rc = ata_do_reset(link, reset, classes, deadline);
Tejun Heo022bdb02006-05-15 20:58:22 +09002182
Tejun Heo932648b2008-05-19 01:15:06 +09002183 if (reset == hardreset &&
2184 ata_eh_followup_srst_needed(link, rc, classify, classes)) {
2185 /* okay, let's do follow-up softreset */
2186 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002187
Tejun Heo932648b2008-05-19 01:15:06 +09002188 if (!reset) {
2189 ata_link_printk(link, KERN_ERR,
2190 "follow-up softreset required "
2191 "but no softreset avaliable\n");
2192 rc = -EINVAL;
2193 goto fail;
2194 }
2195
2196 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2197 rc = ata_do_reset(link, reset, classes, deadline);
Tejun Heo664faf02006-05-31 18:27:50 +09002198 }
2199
Tejun Heo932648b2008-05-19 01:15:06 +09002200 /* -EAGAIN can happen if we skipped followup SRST */
2201 if (rc && rc != -EAGAIN)
2202 goto fail;
2203 } else {
2204 if (verbose)
2205 ata_link_printk(link, KERN_INFO, "no reset method "
2206 "available, skipping reset\n");
2207 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2208 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002209 }
2210
Tejun Heo932648b2008-05-19 01:15:06 +09002211 /*
2212 * Post-reset processing
2213 */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002214 ata_link_for_each_dev(dev, link) {
2215 /* After the reset, the device state is PIO 0 and the
2216 * controller state is undefined. Reset also wakes up
2217 * drives from sleeping mode.
2218 */
2219 dev->pio_mode = XFER_PIO_0;
2220 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002221
Tejun Heo416dc9e2007-10-31 10:17:03 +09002222 if (ata_link_offline(link))
2223 continue;
Tejun Heo664faf02006-05-31 18:27:50 +09002224
Tejun Heo4ccd3322008-01-08 20:26:12 +09002225 /* apply class override */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002226 if (lflags & ATA_LFLAG_ASSUME_ATA)
2227 classes[dev->devno] = ATA_DEV_ATA;
2228 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2229 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
Tejun Heo022bdb02006-05-15 20:58:22 +09002230 }
2231
Tejun Heo416dc9e2007-10-31 10:17:03 +09002232 /* record current link speed */
2233 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2234 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002235
Tejun Heodc98c322008-05-19 01:15:07 +09002236 /* thaw the port */
2237 if (ata_is_host_link(link))
2238 ata_eh_thaw_port(ap);
2239
Tejun Heof0465192008-05-19 01:15:08 +09002240 /* postreset() should clear hardware SError. Although SError
2241 * is cleared during link resume, clearing SError here is
2242 * necessary as some PHYs raise hotplug events after SRST.
2243 * This introduces race condition where hotplug occurs between
2244 * reset and here. This race is mediated by cross checking
2245 * link onlineness and classification result later.
2246 */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002247 if (postreset)
2248 postreset(link, classes);
Tejun Heo20952b62006-05-31 18:27:23 +09002249
Tejun Heof0465192008-05-19 01:15:08 +09002250 /* clear cached SError */
2251 spin_lock_irqsave(link->ap->lock, flags);
2252 link->eh_info.serror = 0;
2253 spin_unlock_irqrestore(link->ap->lock, flags);
2254
2255 /* Make sure onlineness and classification result correspond.
2256 * Hotplug could have happened during reset and some
2257 * controllers fail to wait while a drive is spinning up after
2258 * being hotplugged causing misdetection. By cross checking
2259 * link onlineness and classification result, those conditions
2260 * can be reliably detected and retried.
2261 */
2262 nr_known = 0;
2263 ata_link_for_each_dev(dev, link) {
2264 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2265 if (classes[dev->devno] == ATA_DEV_UNKNOWN)
2266 classes[dev->devno] = ATA_DEV_NONE;
2267 else
2268 nr_known++;
2269 }
2270
2271 if (classify && !nr_known && ata_link_online(link)) {
2272 if (try < max_tries) {
2273 ata_link_printk(link, KERN_WARNING, "link online but "
2274 "device misclassified, retrying\n");
2275 rc = -EAGAIN;
2276 goto fail;
2277 }
2278 ata_link_printk(link, KERN_WARNING,
2279 "link online but device misclassified, "
2280 "device detection might fail\n");
2281 }
2282
Tejun Heo416dc9e2007-10-31 10:17:03 +09002283 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002284 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heo416dc9e2007-10-31 10:17:03 +09002285 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heoae791c02007-09-23 13:14:12 +09002286
Tejun Heo416dc9e2007-10-31 10:17:03 +09002287 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002288 out:
2289 /* clear hotplug flag */
2290 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002291
2292 spin_lock_irqsave(ap->lock, flags);
2293 ap->pflags &= ~ATA_PFLAG_RESETTING;
2294 spin_unlock_irqrestore(ap->lock, flags);
2295
Tejun Heo022bdb02006-05-15 20:58:22 +09002296 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002297
2298 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002299 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2300 if (!ata_is_host_link(link) &&
2301 sata_scr_read(link, SCR_STATUS, &sstatus))
2302 rc = -ERESTART;
2303
Tejun Heo416dc9e2007-10-31 10:17:03 +09002304 if (rc == -ERESTART || try >= max_tries)
2305 goto out;
2306
2307 now = jiffies;
2308 if (time_before(now, deadline)) {
2309 unsigned long delta = deadline - now;
2310
2311 ata_link_printk(link, KERN_WARNING, "reset failed "
2312 "(errno=%d), retrying in %u secs\n",
2313 rc, (jiffies_to_msecs(delta) + 999) / 1000);
2314
2315 while (delta)
2316 delta = schedule_timeout_uninterruptible(delta);
2317 }
2318
2319 if (rc == -EPIPE || try == max_tries - 1)
2320 sata_down_spd_limit(link);
2321 if (hardreset)
2322 reset = hardreset;
2323 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002324}
2325
Tejun Heo02607312007-08-06 18:36:23 +09002326static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09002327 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09002328{
Tejun Heo02607312007-08-06 18:36:23 +09002329 struct ata_port *ap = link->ap;
2330 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002331 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002332 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002333 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09002334 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002335
2336 DPRINTK("ENTER\n");
2337
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002338 /* For PATA drive side cable detection to work, IDENTIFY must
2339 * be done backwards such that PDIAG- is released by the slave
2340 * device before the master device is identified.
2341 */
Tejun Heo02607312007-08-06 18:36:23 +09002342 ata_link_for_each_dev_reverse(dev, link) {
Tejun Heof58229f2007-08-06 18:36:23 +09002343 unsigned int action = ata_eh_dev_action(dev);
2344 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09002345
Tejun Heobff04642006-11-10 18:08:10 +09002346 if (ehc->i.flags & ATA_EHI_DID_RESET)
2347 readid_flags |= ATA_READID_POSTRESET;
2348
Tejun Heo9666f402007-05-04 21:27:47 +02002349 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a32007-09-23 13:19:54 +09002350 WARN_ON(dev->class == ATA_DEV_PMP);
2351
Tejun Heo02607312007-08-06 18:36:23 +09002352 if (ata_link_offline(link)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002353 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002354 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002355 }
2356
Tejun Heo02607312007-08-06 18:36:23 +09002357 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09002358 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2359 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09002360 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002361 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002362
Tejun Heo02607312007-08-06 18:36:23 +09002363 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09002364
Tejun Heobaa1e782006-11-01 18:39:27 +09002365 /* Configuration may have changed, reconfigure
2366 * transfer mode.
2367 */
2368 ehc->i.flags |= ATA_EHI_SETMODE;
2369
zhao, forrest3057ac32006-06-12 12:01:34 +08002370 /* schedule the scsi_rescan_device() here */
2371 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09002372 } else if (dev->class == ATA_DEV_UNKNOWN &&
2373 ehc->tries[dev->devno] &&
2374 ata_class_enabled(ehc->classes[dev->devno])) {
2375 dev->class = ehc->classes[dev->devno];
2376
Tejun Heo633273a32007-09-23 13:19:54 +09002377 if (dev->class == ATA_DEV_PMP)
2378 rc = sata_pmp_attach(dev);
2379 else
2380 rc = ata_dev_read_id(dev, &dev->class,
2381 readid_flags, dev->id);
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002382 switch (rc) {
2383 case 0:
Tejun Heof58229f2007-08-06 18:36:23 +09002384 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002385 break;
2386 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002387 /* IDENTIFY was issued to non-existent
2388 * device. No need to reset. Just
2389 * thaw and kill the device.
2390 */
2391 ata_eh_thaw_port(ap);
2392 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo084fe632006-05-31 18:28:03 +09002393 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002394 default:
2395 dev->class = ATA_DEV_UNKNOWN;
2396 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002397 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002398 }
2399 }
2400
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002401 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09002402 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2403 if (ap->ops->cable_detect)
2404 ap->cbl = ap->ops->cable_detect(ap);
2405 ata_force_cbl(ap);
2406 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002407
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002408 /* Configure new devices forward such that user doesn't see
2409 * device detection messages backwards.
2410 */
Tejun Heo02607312007-08-06 18:36:23 +09002411 ata_link_for_each_dev(dev, link) {
Tejun Heo633273a32007-09-23 13:19:54 +09002412 if (!(new_mask & (1 << dev->devno)) ||
2413 dev->class == ATA_DEV_PMP)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002414 continue;
2415
2416 ehc->i.flags |= ATA_EHI_PRINTINFO;
2417 rc = ata_dev_configure(dev);
2418 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2419 if (rc)
2420 goto err;
2421
2422 spin_lock_irqsave(ap->lock, flags);
2423 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2424 spin_unlock_irqrestore(ap->lock, flags);
2425
2426 /* new device discovered, configure xfermode */
2427 ehc->i.flags |= ATA_EHI_SETMODE;
2428 }
2429
2430 return 0;
2431
2432 err:
2433 *r_failed_dev = dev;
2434 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002435 return rc;
2436}
2437
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002438/**
2439 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2440 * @link: link on which timings will be programmed
2441 * @r_failed_dev: out paramter for failed device
2442 *
2443 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2444 * ata_set_mode() fails, pointer to the failing device is
2445 * returned in @r_failed_dev.
2446 *
2447 * LOCKING:
2448 * PCI/etc. bus probe sem.
2449 *
2450 * RETURNS:
2451 * 0 on success, negative errno otherwise
2452 */
2453int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2454{
2455 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09002456 struct ata_device *dev;
2457 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002458
Tejun Heo76326ac2007-11-27 19:28:59 +09002459 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2460 ata_link_for_each_dev(dev, link) {
2461 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2462 struct ata_ering_entry *ent;
2463
2464 ent = ata_ering_top(&dev->ering);
2465 if (ent)
2466 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2467 }
2468 }
2469
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002470 /* has private set_mode? */
2471 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09002472 rc = ap->ops->set_mode(link, r_failed_dev);
2473 else
2474 rc = ata_do_set_mode(link, r_failed_dev);
2475
2476 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2477 ata_link_for_each_dev(dev, link) {
2478 struct ata_eh_context *ehc = &link->eh_context;
2479 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2480 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2481
2482 if (dev->xfer_mode != saved_xfer_mode ||
2483 ata_ncq_enabled(dev) != saved_ncq)
2484 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2485 }
2486
2487 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002488}
2489
Tejun Heo02607312007-08-06 18:36:23 +09002490static int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002491{
Tejun Heof58229f2007-08-06 18:36:23 +09002492 struct ata_device *dev;
2493 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002494
Tejun Heo02607312007-08-06 18:36:23 +09002495 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002496 if (ata_dev_enabled(dev))
Tejun Heo022bdb02006-05-15 20:58:22 +09002497 cnt++;
2498 return cnt;
2499}
2500
Tejun Heo02607312007-08-06 18:36:23 +09002501static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002502{
Tejun Heof58229f2007-08-06 18:36:23 +09002503 struct ata_device *dev;
2504 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002505
Tejun Heo02607312007-08-06 18:36:23 +09002506 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002507 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09002508 cnt++;
2509 return cnt;
2510}
2511
Tejun Heo02607312007-08-06 18:36:23 +09002512static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002513{
Tejun Heo672b2d62008-01-24 00:05:14 +09002514 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09002515 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09002516 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09002517
Tejun Heof9df58c2007-09-23 13:14:13 +09002518 /* skip disabled links */
2519 if (link->flags & ATA_LFLAG_DISABLED)
2520 return 1;
2521
Tejun Heo672b2d62008-01-24 00:05:14 +09002522 /* thaw frozen port and recover failed devices */
2523 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2524 return 0;
2525
2526 /* reset at least once if reset is requested */
2527 if ((ehc->i.action & ATA_EH_RESET) &&
2528 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09002529 return 0;
2530
2531 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo02607312007-08-06 18:36:23 +09002532 ata_link_for_each_dev(dev, link) {
Tejun Heo084fe632006-05-31 18:28:03 +09002533 if (dev->class == ATA_DEV_UNKNOWN &&
2534 ehc->classes[dev->devno] != ATA_DEV_NONE)
2535 return 0;
2536 }
2537
2538 return 1;
2539}
2540
Tejun Heo02c05a22007-11-27 19:28:54 +09002541static int ata_eh_schedule_probe(struct ata_device *dev)
2542{
2543 struct ata_eh_context *ehc = &dev->link->eh_context;
2544
2545 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2546 (ehc->did_probe_mask & (1 << dev->devno)))
2547 return 0;
2548
2549 ata_eh_detach_dev(dev);
2550 ata_dev_init(dev);
2551 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09002552 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09002553 ehc->saved_xfer_mode[dev->devno] = 0;
2554 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09002555
2556 return 1;
2557}
2558
Tejun Heo9b1e2652007-08-06 18:36:24 +09002559static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09002560{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09002561 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09002562
2563 ehc->tries[dev->devno]--;
2564
2565 switch (err) {
2566 case -ENODEV:
2567 /* device missing or wrong IDENTIFY data, schedule probing */
2568 ehc->i.probe_mask |= (1 << dev->devno);
2569 case -EINVAL:
2570 /* give it just one more chance */
2571 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2572 case -EIO:
Tejun Heo4fb46152007-10-29 16:45:05 +09002573 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
Tejun Heofee7ca72007-07-01 19:05:58 +09002574 /* This is the last chance, better to slow
2575 * down than lose it.
2576 */
Tejun Heo936fd732007-08-06 18:36:23 +09002577 sata_down_spd_limit(dev->link);
Tejun Heofee7ca72007-07-01 19:05:58 +09002578 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2579 }
2580 }
2581
2582 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2583 /* disable device if it has used up all its chances */
2584 ata_dev_disable(dev);
2585
2586 /* detach if offline */
Tejun Heo936fd732007-08-06 18:36:23 +09002587 if (ata_link_offline(dev->link))
Tejun Heofee7ca72007-07-01 19:05:58 +09002588 ata_eh_detach_dev(dev);
2589
Tejun Heo02c05a22007-11-27 19:28:54 +09002590 /* schedule probe if necessary */
2591 if (ata_eh_schedule_probe(dev))
Tejun Heofee7ca72007-07-01 19:05:58 +09002592 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002593
2594 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09002595 } else {
Tejun Heocf480622008-01-24 00:05:14 +09002596 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002597 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09002598 }
2599}
2600
Tejun Heo022bdb02006-05-15 20:58:22 +09002601/**
2602 * ata_eh_recover - recover host port after error
2603 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09002604 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002605 * @softreset: softreset method (can be NULL)
2606 * @hardreset: hardreset method (can be NULL)
2607 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002608 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09002609 *
2610 * This is the alpha and omega, eum and yang, heart and soul of
2611 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09002612 * recover each link and hotplug requests are recorded in the
2613 * link's eh_context. This function executes all the operations
2614 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09002615 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09002616 *
2617 * LOCKING:
2618 * Kernel thread context (may sleep).
2619 *
2620 * RETURNS:
2621 * 0 on success, -errno on failure.
2622 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002623int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2624 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2625 ata_postreset_fn_t postreset,
2626 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002627{
Tejun Heo9b1e2652007-08-06 18:36:24 +09002628 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09002629 struct ata_device *dev;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002630 int nr_failed_devs, nr_disabled_devs;
Tejun Heodc98c322008-05-19 01:15:07 +09002631 int rc;
Tejun Heof9df58c2007-09-23 13:14:13 +09002632 unsigned long flags;
Tejun Heo022bdb02006-05-15 20:58:22 +09002633
2634 DPRINTK("ENTER\n");
2635
2636 /* prep for recovery */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002637 ata_port_for_each_link(link, ap) {
2638 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09002639
Tejun Heof9df58c2007-09-23 13:14:13 +09002640 /* re-enable link? */
2641 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
2642 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
2643 spin_lock_irqsave(ap->lock, flags);
2644 link->flags &= ~ATA_LFLAG_DISABLED;
2645 spin_unlock_irqrestore(ap->lock, flags);
2646 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
2647 }
2648
Tejun Heo9b1e2652007-08-06 18:36:24 +09002649 ata_link_for_each_dev(dev, link) {
Tejun Heofd995f72007-09-23 13:14:12 +09002650 if (link->flags & ATA_LFLAG_NO_RETRY)
2651 ehc->tries[dev->devno] = 1;
2652 else
2653 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09002654
Tejun Heo9b1e2652007-08-06 18:36:24 +09002655 /* collect port action mask recorded in dev actions */
2656 ehc->i.action |= ehc->i.dev_action[dev->devno] &
2657 ~ATA_EH_PERDEV_MASK;
2658 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09002659
Tejun Heo9b1e2652007-08-06 18:36:24 +09002660 /* process hotplug request */
2661 if (dev->flags & ATA_DFLAG_DETACH)
2662 ata_eh_detach_dev(dev);
2663
Tejun Heo02c05a22007-11-27 19:28:54 +09002664 /* schedule probe if necessary */
2665 if (!ata_dev_enabled(dev))
2666 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09002667 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002668 }
2669
2670 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09002671 rc = 0;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002672 nr_failed_devs = 0;
2673 nr_disabled_devs = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002674
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002675 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09002676 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09002677 goto out;
2678
Tejun Heo9b1e2652007-08-06 18:36:24 +09002679 /* prep for EH */
2680 ata_port_for_each_link(link, ap) {
2681 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002682
Tejun Heo9b1e2652007-08-06 18:36:24 +09002683 /* skip EH if possible. */
2684 if (ata_eh_skip_recovery(link))
2685 ehc->i.action = 0;
2686
Tejun Heo9b1e2652007-08-06 18:36:24 +09002687 ata_link_for_each_dev(dev, link)
2688 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
2689 }
Tejun Heo084fe632006-05-31 18:28:03 +09002690
Tejun Heo022bdb02006-05-15 20:58:22 +09002691 /* reset */
Tejun Heodc98c322008-05-19 01:15:07 +09002692 ata_port_for_each_link(link, ap) {
2693 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002694
Tejun Heodc98c322008-05-19 01:15:07 +09002695 if (!(ehc->i.action & ATA_EH_RESET))
2696 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002697
Tejun Heodc98c322008-05-19 01:15:07 +09002698 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
2699 prereset, softreset, hardreset, postreset);
2700 if (rc) {
2701 ata_link_printk(link, KERN_ERR,
2702 "reset failed, giving up\n");
2703 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09002704 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002705 }
2706
Tejun Heo9b1e2652007-08-06 18:36:24 +09002707 /* the rest */
2708 ata_port_for_each_link(link, ap) {
2709 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002710
Tejun Heo9b1e2652007-08-06 18:36:24 +09002711 /* revalidate existing devices and attach new ones */
2712 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09002713 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09002714 goto dev_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002715
Tejun Heo633273a32007-09-23 13:19:54 +09002716 /* if PMP got attached, return, pmp EH will take care of it */
2717 if (link->device->class == ATA_DEV_PMP) {
2718 ehc->i.action = 0;
2719 return 0;
2720 }
2721
Tejun Heo9b1e2652007-08-06 18:36:24 +09002722 /* configure transfer mode if necessary */
2723 if (ehc->i.flags & ATA_EHI_SETMODE) {
2724 rc = ata_set_mode(link, &dev);
2725 if (rc)
2726 goto dev_fail;
2727 ehc->i.flags &= ~ATA_EHI_SETMODE;
2728 }
2729
Tejun Heo3ec25eb2008-03-27 18:37:14 +09002730 if (ehc->i.action & ATA_EH_LPM)
Kristen Carlson Accardica773292007-10-25 00:58:59 -04002731 ata_link_for_each_dev(dev, link)
2732 ata_dev_enable_pm(dev, ap->pm_policy);
2733
Tejun Heo9b1e2652007-08-06 18:36:24 +09002734 /* this link is okay now */
2735 ehc->i.flags = 0;
2736 continue;
2737
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002738dev_fail:
Tejun Heo9b1e2652007-08-06 18:36:24 +09002739 nr_failed_devs++;
2740 if (ata_eh_handle_dev_fail(dev, rc))
2741 nr_disabled_devs++;
2742
Tejun Heob06ce3e2007-10-09 15:06:48 +09002743 if (ap->pflags & ATA_PFLAG_FROZEN) {
2744 /* PMP reset requires working host port.
2745 * Can't retry if it's frozen.
2746 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002747 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09002748 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002749 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09002750 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002751 }
2752
Tejun Heo9b1e2652007-08-06 18:36:24 +09002753 if (nr_failed_devs) {
2754 if (nr_failed_devs != nr_disabled_devs) {
2755 ata_port_printk(ap, KERN_WARNING, "failed to recover "
2756 "some devices, retrying in 5 secs\n");
2757 ssleep(5);
2758 } else {
2759 /* no device left to recover, repeat fast */
2760 msleep(500);
2761 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002762
Tejun Heo9b1e2652007-08-06 18:36:24 +09002763 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002764 }
2765
Tejun Heo022bdb02006-05-15 20:58:22 +09002766 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09002767 if (rc && r_failed_link)
2768 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09002769
2770 DPRINTK("EXIT, rc=%d\n", rc);
2771 return rc;
2772}
2773
2774/**
2775 * ata_eh_finish - finish up EH
2776 * @ap: host port to finish EH for
2777 *
2778 * Recovery is complete. Clean up EH states and retry or finish
2779 * failed qcs.
2780 *
2781 * LOCKING:
2782 * None.
2783 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002784void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09002785{
2786 int tag;
2787
2788 /* retry or finish qcs */
2789 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2790 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2791
2792 if (!(qc->flags & ATA_QCFLAG_FAILED))
2793 continue;
2794
2795 if (qc->err_mask) {
2796 /* FIXME: Once EH migration is complete,
2797 * generate sense data in this function,
2798 * considering both err_mask and tf.
2799 */
Tejun Heo03faab72008-03-27 19:14:24 +09002800 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09002801 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09002802 else
2803 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002804 } else {
2805 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2806 ata_eh_qc_complete(qc);
2807 } else {
2808 /* feed zero TF to sense generation */
2809 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2810 ata_eh_qc_retry(qc);
2811 }
2812 }
2813 }
Tejun Heoda917d62007-09-23 13:14:12 +09002814
2815 /* make sure nr_active_links is zero after EH */
2816 WARN_ON(ap->nr_active_links);
2817 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002818}
2819
2820/**
2821 * ata_do_eh - do standard error handling
2822 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09002823 *
Tejun Heof5914a42006-05-31 18:27:48 +09002824 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002825 * @softreset: softreset method (can be NULL)
2826 * @hardreset: hardreset method (can be NULL)
2827 * @postreset: postreset method (can be NULL)
2828 *
2829 * Perform standard error handling sequence.
2830 *
2831 * LOCKING:
2832 * Kernel thread context (may sleep).
2833 */
Tejun Heof5914a42006-05-31 18:27:48 +09002834void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2835 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2836 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002837{
Tejun Heo9b1e2652007-08-06 18:36:24 +09002838 struct ata_device *dev;
2839 int rc;
2840
2841 ata_eh_autopsy(ap);
2842 ata_eh_report(ap);
2843
2844 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
2845 NULL);
2846 if (rc) {
2847 ata_link_for_each_dev(dev, &ap->link)
2848 ata_dev_disable(dev);
2849 }
2850
Tejun Heo022bdb02006-05-15 20:58:22 +09002851 ata_eh_finish(ap);
2852}
Tejun Heo500530f2006-07-03 16:07:27 +09002853
Tejun Heoa1efdab2008-03-25 12:22:50 +09002854/**
2855 * ata_std_error_handler - standard error handler
2856 * @ap: host port to handle error for
2857 *
2858 * Standard error handler
2859 *
2860 * LOCKING:
2861 * Kernel thread context (may sleep).
2862 */
2863void ata_std_error_handler(struct ata_port *ap)
2864{
2865 struct ata_port_operations *ops = ap->ops;
2866 ata_reset_fn_t hardreset = ops->hardreset;
2867
Tejun Heo57c9efd2008-04-07 22:47:19 +09002868 /* ignore built-in hardreset if SCR access is not available */
2869 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09002870 hardreset = NULL;
2871
2872 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
2873}
2874
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002875#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09002876/**
2877 * ata_eh_handle_port_suspend - perform port suspend operation
2878 * @ap: port to suspend
2879 *
2880 * Suspend @ap.
2881 *
2882 * LOCKING:
2883 * Kernel thread context (may sleep).
2884 */
2885static void ata_eh_handle_port_suspend(struct ata_port *ap)
2886{
2887 unsigned long flags;
2888 int rc = 0;
2889
2890 /* are we suspending? */
2891 spin_lock_irqsave(ap->lock, flags);
2892 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2893 ap->pm_mesg.event == PM_EVENT_ON) {
2894 spin_unlock_irqrestore(ap->lock, flags);
2895 return;
2896 }
2897 spin_unlock_irqrestore(ap->lock, flags);
2898
2899 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2900
Tejun Heo64578a32007-05-15 03:28:16 +09002901 /* tell ACPI we're suspending */
2902 rc = ata_acpi_on_suspend(ap);
2903 if (rc)
2904 goto out;
2905
Tejun Heo500530f2006-07-03 16:07:27 +09002906 /* suspend */
2907 ata_eh_freeze_port(ap);
2908
2909 if (ap->ops->port_suspend)
2910 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2911
Shaohua Libd3adca2007-11-02 09:32:38 +08002912 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a32007-05-15 03:28:16 +09002913 out:
Tejun Heo500530f2006-07-03 16:07:27 +09002914 /* report result */
2915 spin_lock_irqsave(ap->lock, flags);
2916
2917 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2918 if (rc == 0)
2919 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09002920 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09002921 ata_port_schedule_eh(ap);
2922
2923 if (ap->pm_result) {
2924 *ap->pm_result = rc;
2925 ap->pm_result = NULL;
2926 }
2927
2928 spin_unlock_irqrestore(ap->lock, flags);
2929
2930 return;
2931}
2932
2933/**
2934 * ata_eh_handle_port_resume - perform port resume operation
2935 * @ap: port to resume
2936 *
2937 * Resume @ap.
2938 *
Tejun Heo500530f2006-07-03 16:07:27 +09002939 * LOCKING:
2940 * Kernel thread context (may sleep).
2941 */
2942static void ata_eh_handle_port_resume(struct ata_port *ap)
2943{
Tejun Heo500530f2006-07-03 16:07:27 +09002944 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02002945 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09002946
2947 /* are we resuming? */
2948 spin_lock_irqsave(ap->lock, flags);
2949 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2950 ap->pm_mesg.event != PM_EVENT_ON) {
2951 spin_unlock_irqrestore(ap->lock, flags);
2952 return;
2953 }
2954 spin_unlock_irqrestore(ap->lock, flags);
2955
Tejun Heo9666f402007-05-04 21:27:47 +02002956 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09002957
Shaohua Libd3adca2007-11-02 09:32:38 +08002958 ata_acpi_set_state(ap, PMSG_ON);
2959
Tejun Heo500530f2006-07-03 16:07:27 +09002960 if (ap->ops->port_resume)
2961 rc = ap->ops->port_resume(ap);
2962
Tejun Heo67465442007-05-15 03:28:16 +09002963 /* tell ACPI that we're resuming */
2964 ata_acpi_on_resume(ap);
2965
Tejun Heo9666f402007-05-04 21:27:47 +02002966 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09002967 spin_lock_irqsave(ap->lock, flags);
2968 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2969 if (ap->pm_result) {
2970 *ap->pm_result = rc;
2971 ap->pm_result = NULL;
2972 }
2973 spin_unlock_irqrestore(ap->lock, flags);
2974}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09002975#endif /* CONFIG_PM */