blob: 32da9a93ce44ae221caa63b753c225c81b257f5d [file] [log] [blame]
Tejun Heoece1d632006-04-02 18:51:53 +09001/*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
Tejun Heoece1d632006-04-02 18:51:53 +090035#include <linux/kernel.h>
Jens Axboe242f9dc2008-09-14 05:55:09 -070036#include <linux/blkdev.h>
Jeff Garzik28555682007-10-11 17:12:35 -040037#include <linux/pci.h>
Tejun Heoece1d632006-04-02 18:51:53 +090038#include <scsi/scsi.h>
39#include <scsi/scsi_host.h>
40#include <scsi/scsi_eh.h>
41#include <scsi/scsi_device.h>
42#include <scsi/scsi_cmnd.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040043#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090044
45#include <linux/libata.h>
46
47#include "libata.h"
48
Tejun Heo7d47e8d2007-02-02 16:22:31 +090049enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090050 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090051 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090054 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090055
56 /* error flags */
57 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090058 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Tejun Heo3884f7b2007-11-27 19:28:56 +090059
60 /* error categories */
61 ATA_ECAT_NONE = 0,
62 ATA_ECAT_ATA_BUS = 1,
63 ATA_ECAT_TOUT_HSM = 2,
64 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090065 ATA_ECAT_DUBIOUS_NONE = 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
69 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090070
Tejun Heo87fbc5a2008-05-20 02:17:54 +090071 ATA_EH_CMD_DFL_TIMEOUT = 5000,
72
Tejun Heo0a2c0f52008-05-20 02:17:52 +090073 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN = 5000,
75
Tejun Heo341c2c92008-05-20 02:17:51 +090076 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
80 */
81 ATA_EH_PRERESET_TIMEOUT = 10000,
82 ATA_EH_FASTDRAIN_INTERVAL = 3000,
Tejun Heo11fc33d2008-08-30 14:20:01 +020083
84 ATA_EH_UA_TRIES = 5,
Tejun Heo31daabd2007-02-02 16:50:52 +090085};
86
87/* The following table determines how we sequence resets. Each entry
88 * represents timeout for that try. The first try can be soft or
89 * hardreset. All others are hardreset if available. In most cases
90 * the first reset w/ 10sec timeout should succeed. Following entries
91 * are mostly for error handling, hotplug and retarded devices.
92 */
93static const unsigned long ata_eh_reset_timeouts[] = {
Tejun Heo341c2c92008-05-20 02:17:51 +090094 10000, /* most drives spin up by 10sec */
95 10000, /* > 99% working drives spin up before 20sec */
96 35000, /* give > 30 secs of idleness for retarded devices */
97 5000, /* and sweet one last chance */
Tejun Heod8af0eb2008-05-20 02:17:53 +090098 ULONG_MAX, /* > 1 min has elapsed, give up */
Tejun Heo31daabd2007-02-02 16:50:52 +090099};
100
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900101static const unsigned long ata_eh_identify_timeouts[] = {
102 5000, /* covers > 99% of successes and not too boring on failures */
103 10000, /* combined time till here is enough even for media access */
104 30000, /* for true idiots */
105 ULONG_MAX,
106};
107
108static const unsigned long ata_eh_other_timeouts[] = {
109 5000, /* same rationale as identify timeout */
110 10000, /* ditto */
111 /* but no merciful 30sec for other commands, it just isn't worth it */
112 ULONG_MAX,
113};
114
115struct ata_eh_cmd_timeout_ent {
116 const u8 *commands;
117 const unsigned long *timeouts;
118};
119
120/* The following table determines timeouts to use for EH internal
121 * commands. Each table entry is a command class and matches the
122 * commands the entry applies to and the timeout table to use.
123 *
124 * On the retry after a command timed out, the next timeout value from
125 * the table is used. If the table doesn't contain further entries,
126 * the last value is used.
127 *
128 * ehc->cmd_timeout_idx keeps track of which timeout to use per
129 * command class, so if SET_FEATURES times out on the first try, the
130 * next try will use the second timeout value only for that class.
131 */
132#define CMDS(cmds...) (const u8 []){ cmds, 0 }
133static const struct ata_eh_cmd_timeout_ent
134ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
135 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
136 .timeouts = ata_eh_identify_timeouts, },
137 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
138 .timeouts = ata_eh_other_timeouts, },
139 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
140 .timeouts = ata_eh_other_timeouts, },
141 { .commands = CMDS(ATA_CMD_SET_FEATURES),
142 .timeouts = ata_eh_other_timeouts, },
143 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
144 .timeouts = ata_eh_other_timeouts, },
145};
146#undef CMDS
147
Tejun Heoad9e2762006-05-15 20:58:12 +0900148static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900149#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +0900150static void ata_eh_handle_port_suspend(struct ata_port *ap);
151static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900152#else /* CONFIG_PM */
153static void ata_eh_handle_port_suspend(struct ata_port *ap)
154{ }
155
156static void ata_eh_handle_port_resume(struct ata_port *ap)
157{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900158#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900159
Tejun Heob64bbc32007-07-16 14:29:39 +0900160static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
161 va_list args)
162{
163 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
164 ATA_EH_DESC_LEN - ehi->desc_len,
165 fmt, args);
166}
167
168/**
169 * __ata_ehi_push_desc - push error description without adding separator
170 * @ehi: target EHI
171 * @fmt: printf format string
172 *
173 * Format string according to @fmt and append it to @ehi->desc.
174 *
175 * LOCKING:
176 * spin_lock_irqsave(host lock)
177 */
178void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
179{
180 va_list args;
181
182 va_start(args, fmt);
183 __ata_ehi_pushv_desc(ehi, fmt, args);
184 va_end(args);
185}
186
187/**
188 * ata_ehi_push_desc - push error description with separator
189 * @ehi: target EHI
190 * @fmt: printf format string
191 *
192 * Format string according to @fmt and append it to @ehi->desc.
193 * If @ehi->desc is not empty, ", " is added in-between.
194 *
195 * LOCKING:
196 * spin_lock_irqsave(host lock)
197 */
198void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
199{
200 va_list args;
201
202 if (ehi->desc_len)
203 __ata_ehi_push_desc(ehi, ", ");
204
205 va_start(args, fmt);
206 __ata_ehi_pushv_desc(ehi, fmt, args);
207 va_end(args);
208}
209
210/**
211 * ata_ehi_clear_desc - clean error description
212 * @ehi: target EHI
213 *
214 * Clear @ehi->desc.
215 *
216 * LOCKING:
217 * spin_lock_irqsave(host lock)
218 */
219void ata_ehi_clear_desc(struct ata_eh_info *ehi)
220{
221 ehi->desc[0] = '\0';
222 ehi->desc_len = 0;
223}
224
Tejun Heocbcdd872007-08-18 13:14:55 +0900225/**
226 * ata_port_desc - append port description
227 * @ap: target ATA port
228 * @fmt: printf format string
229 *
230 * Format string according to @fmt and append it to port
231 * description. If port description is not empty, " " is added
232 * in-between. This function is to be used while initializing
233 * ata_host. The description is printed on host registration.
234 *
235 * LOCKING:
236 * None.
237 */
238void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
239{
240 va_list args;
241
242 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
243
244 if (ap->link.eh_info.desc_len)
245 __ata_ehi_push_desc(&ap->link.eh_info, " ");
246
247 va_start(args, fmt);
248 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
249 va_end(args);
250}
251
252#ifdef CONFIG_PCI
253
254/**
255 * ata_port_pbar_desc - append PCI BAR description
256 * @ap: target ATA port
257 * @bar: target PCI BAR
258 * @offset: offset into PCI BAR
259 * @name: name of the area
260 *
261 * If @offset is negative, this function formats a string which
262 * contains the name, address, size and type of the BAR and
263 * appends it to the port description. If @offset is zero or
264 * positive, only name and offsetted address is appended.
265 *
266 * LOCKING:
267 * None.
268 */
269void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
270 const char *name)
271{
272 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
273 char *type = "";
274 unsigned long long start, len;
275
276 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
277 type = "m";
278 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
279 type = "i";
280
281 start = (unsigned long long)pci_resource_start(pdev, bar);
282 len = (unsigned long long)pci_resource_len(pdev, bar);
283
284 if (offset < 0)
285 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
286 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800287 ata_port_desc(ap, "%s 0x%llx", name,
288 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900289}
290
291#endif /* CONFIG_PCI */
292
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900293static int ata_lookup_timeout_table(u8 cmd)
294{
295 int i;
296
297 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
298 const u8 *cur;
299
300 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
301 if (*cur == cmd)
302 return i;
303 }
304
305 return -1;
306}
307
308/**
309 * ata_internal_cmd_timeout - determine timeout for an internal command
310 * @dev: target device
311 * @cmd: internal command to be issued
312 *
313 * Determine timeout for internal command @cmd for @dev.
314 *
315 * LOCKING:
316 * EH context.
317 *
318 * RETURNS:
319 * Determined timeout.
320 */
321unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
322{
323 struct ata_eh_context *ehc = &dev->link->eh_context;
324 int ent = ata_lookup_timeout_table(cmd);
325 int idx;
326
327 if (ent < 0)
328 return ATA_EH_CMD_DFL_TIMEOUT;
329
330 idx = ehc->cmd_timeout_idx[dev->devno][ent];
331 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
332}
333
334/**
335 * ata_internal_cmd_timed_out - notification for internal command timeout
336 * @dev: target device
337 * @cmd: internal command which timed out
338 *
339 * Notify EH that internal command @cmd for @dev timed out. This
340 * function should be called only for commands whose timeouts are
341 * determined using ata_internal_cmd_timeout().
342 *
343 * LOCKING:
344 * EH context.
345 */
346void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
347{
348 struct ata_eh_context *ehc = &dev->link->eh_context;
349 int ent = ata_lookup_timeout_table(cmd);
350 int idx;
351
352 if (ent < 0)
353 return;
354
355 idx = ehc->cmd_timeout_idx[dev->devno][ent];
356 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
357 ehc->cmd_timeout_idx[dev->devno][ent]++;
358}
359
Tejun Heo3884f7b2007-11-27 19:28:56 +0900360static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900361 unsigned int err_mask)
362{
363 struct ata_ering_entry *ent;
364
365 WARN_ON(!err_mask);
366
367 ering->cursor++;
368 ering->cursor %= ATA_ERING_SIZE;
369
370 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900371 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900372 ent->err_mask = err_mask;
373 ent->timestamp = get_jiffies_64();
374}
375
Tejun Heo76326ac2007-11-27 19:28:59 +0900376static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
377{
378 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
379
380 if (ent->err_mask)
381 return ent;
382 return NULL;
383}
384
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900385static void ata_ering_clear(struct ata_ering *ering)
Tejun Heo0c247c52006-05-15 20:58:19 +0900386{
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900387 memset(ering, 0, sizeof(*ering));
Tejun Heo0c247c52006-05-15 20:58:19 +0900388}
389
390static int ata_ering_map(struct ata_ering *ering,
391 int (*map_fn)(struct ata_ering_entry *, void *),
392 void *arg)
393{
394 int idx, rc = 0;
395 struct ata_ering_entry *ent;
396
397 idx = ering->cursor;
398 do {
399 ent = &ering->ring[idx];
400 if (!ent->err_mask)
401 break;
402 rc = map_fn(ent, arg);
403 if (rc)
404 break;
405 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
406 } while (idx != ering->cursor);
407
408 return rc;
409}
410
Tejun Heo64f65ca2006-06-24 20:30:18 +0900411static unsigned int ata_eh_dev_action(struct ata_device *dev)
412{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900413 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900414
415 return ehc->i.action | ehc->i.dev_action[dev->devno];
416}
417
Tejun Heof58229f2007-08-06 18:36:23 +0900418static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900419 struct ata_eh_info *ehi, unsigned int action)
420{
Tejun Heof58229f2007-08-06 18:36:23 +0900421 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900422
423 if (!dev) {
424 ehi->action &= ~action;
Tejun Heof58229f2007-08-06 18:36:23 +0900425 ata_link_for_each_dev(tdev, link)
426 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900427 } else {
428 /* doesn't make sense for port-wide EH actions */
429 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
430
431 /* break ehi->action into ehi->dev_action */
432 if (ehi->action & action) {
Tejun Heof58229f2007-08-06 18:36:23 +0900433 ata_link_for_each_dev(tdev, link)
434 ehi->dev_action[tdev->devno] |=
435 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900436 ehi->action &= ~action;
437 }
438
439 /* turn off the specified per-dev action */
440 ehi->dev_action[dev->devno] &= ~action;
441 }
442}
443
Tejun Heoece1d632006-04-02 18:51:53 +0900444/**
445 * ata_scsi_timed_out - SCSI layer time out callback
446 * @cmd: timed out SCSI command
447 *
448 * Handles SCSI layer timeout. We race with normal completion of
449 * the qc for @cmd. If the qc is already gone, we lose and let
450 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
451 * timed out and EH should be invoked. Prevent ata_qc_complete()
452 * from finishing it by setting EH_SCHEDULED and return
453 * EH_NOT_HANDLED.
454 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900455 * TODO: kill this function once old EH is gone.
456 *
Tejun Heoece1d632006-04-02 18:51:53 +0900457 * LOCKING:
458 * Called from timer context
459 *
460 * RETURNS:
461 * EH_HANDLED or EH_NOT_HANDLED
462 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700463enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
Tejun Heoece1d632006-04-02 18:51:53 +0900464{
465 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400466 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900467 unsigned long flags;
468 struct ata_queued_cmd *qc;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700469 enum blk_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900470
471 DPRINTK("ENTER\n");
472
Tejun Heoad9e2762006-05-15 20:58:12 +0900473 if (ap->ops->error_handler) {
Jens Axboe242f9dc2008-09-14 05:55:09 -0700474 ret = BLK_EH_NOT_HANDLED;
Tejun Heoad9e2762006-05-15 20:58:12 +0900475 goto out;
476 }
477
Jens Axboe242f9dc2008-09-14 05:55:09 -0700478 ret = BLK_EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400479 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900480 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900481 if (qc) {
482 WARN_ON(qc->scsicmd != cmd);
483 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
484 qc->err_mask |= AC_ERR_TIMEOUT;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700485 ret = BLK_EH_NOT_HANDLED;
Tejun Heoece1d632006-04-02 18:51:53 +0900486 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400487 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900488
Tejun Heoad9e2762006-05-15 20:58:12 +0900489 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900490 DPRINTK("EXIT, ret=%d\n", ret);
491 return ret;
492}
493
494/**
495 * ata_scsi_error - SCSI layer error handler callback
496 * @host: SCSI host on which error occurred
497 *
498 * Handles SCSI-layer-thrown error events.
499 *
500 * LOCKING:
501 * Inherited from SCSI layer (none, can sleep)
502 *
503 * RETURNS:
504 * Zero.
505 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400506void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900507{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400508 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900509 int i;
Tejun Heoad9e2762006-05-15 20:58:12 +0900510 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900511
512 DPRINTK("ENTER\n");
513
Tejun Heoad9e2762006-05-15 20:58:12 +0900514 /* synchronize with port task */
Tejun Heoece1d632006-04-02 18:51:53 +0900515 ata_port_flush_task(ap);
516
Jeff Garzikcca39742006-08-24 03:19:22 -0400517 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900518
Tejun Heoad9e2762006-05-15 20:58:12 +0900519 /* For new EH, all qcs are finished in one of three ways -
520 * normal completion, error completion, and SCSI timeout.
521 * Both cmpletions can race against SCSI timeout. When normal
522 * completion wins, the qc never reaches EH. When error
523 * completion wins, the qc has ATA_QCFLAG_FAILED set.
524 *
525 * When SCSI timeout wins, things are a bit more complex.
526 * Normal or error completion can occur after the timeout but
527 * before this point. In such cases, both types of
528 * completions are honored. A scmd is determined to have
529 * timed out iff its associated qc is active and not failed.
530 */
531 if (ap->ops->error_handler) {
532 struct scsi_cmnd *scmd, *tmp;
533 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900534
Tejun Heoe30349d2006-07-03 03:02:15 +0900535 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900536
537 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
538 struct ata_queued_cmd *qc;
539
540 for (i = 0; i < ATA_MAX_QUEUE; i++) {
541 qc = __ata_qc_from_tag(ap, i);
542 if (qc->flags & ATA_QCFLAG_ACTIVE &&
543 qc->scsicmd == scmd)
544 break;
545 }
546
547 if (i < ATA_MAX_QUEUE) {
548 /* the scmd has an associated qc */
549 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
550 /* which hasn't failed yet, timeout */
551 qc->err_mask |= AC_ERR_TIMEOUT;
552 qc->flags |= ATA_QCFLAG_FAILED;
553 nr_timedout++;
554 }
555 } else {
556 /* Normal completion occurred after
557 * SCSI timeout but before this point.
558 * Successfully complete it.
559 */
560 scmd->retries = scmd->allowed;
561 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
562 }
563 }
564
565 /* If we have timed out qcs. They belong to EH from
566 * this point but the state of the controller is
567 * unknown. Freeze the port to make sure the IRQ
568 * handler doesn't diddle with those qcs. This must
569 * be done atomically w.r.t. setting QCFLAG_FAILED.
570 */
571 if (nr_timedout)
572 __ata_port_freeze(ap);
573
Tejun Heoe30349d2006-07-03 03:02:15 +0900574 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900575
576 /* initialize eh_tries */
577 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900578 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900579 spin_unlock_wait(ap->lock);
Tejun Heoad9e2762006-05-15 20:58:12 +0900580
581 repeat:
582 /* invoke error handler */
583 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900584 struct ata_link *link;
585
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900586 /* kill fast drain timer */
587 del_timer_sync(&ap->fastdrain_timer);
588
Tejun Heo500530f2006-07-03 16:07:27 +0900589 /* process port resume request */
590 ata_eh_handle_port_resume(ap);
591
Tejun Heof3e81b12006-05-15 20:58:21 +0900592 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900593 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900594
Tejun Heocf1b86c2007-08-06 18:36:23 +0900595 __ata_port_for_each_link(link, ap) {
Tejun Heo00115e02007-11-27 19:28:58 +0900596 struct ata_eh_context *ehc = &link->eh_context;
597 struct ata_device *dev;
598
Tejun Heocf1b86c2007-08-06 18:36:23 +0900599 memset(&link->eh_context, 0, sizeof(link->eh_context));
600 link->eh_context.i = link->eh_info;
601 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900602
603 ata_link_for_each_dev(dev, link) {
604 int devno = dev->devno;
605
Tejun Heo4a9c7b32008-10-27 19:59:23 +0900606 if (!ata_dev_enabled(dev))
607 continue;
608
Tejun Heo00115e02007-11-27 19:28:58 +0900609 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
610 if (ata_ncq_enabled(dev))
611 ehc->saved_ncq_enabled |= 1 << devno;
612 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900613 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900614
Tejun Heob51e9e52006-06-29 01:29:30 +0900615 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
616 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900617 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900618
Tejun Heoe30349d2006-07-03 03:02:15 +0900619 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900620
Tejun Heo500530f2006-07-03 16:07:27 +0900621 /* invoke EH, skip if unloading or suspended */
622 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900623 ap->ops->error_handler(ap);
624 else
625 ata_eh_finish(ap);
Tejun Heoad9e2762006-05-15 20:58:12 +0900626
Tejun Heo500530f2006-07-03 16:07:27 +0900627 /* process port suspend request */
628 ata_eh_handle_port_suspend(ap);
629
Tejun Heoad9e2762006-05-15 20:58:12 +0900630 /* Exception might have happend after ->error_handler
631 * recovered the port but before this point. Repeat
632 * EH in such case.
633 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900634 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900635
Tejun Heob51e9e52006-06-29 01:29:30 +0900636 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900637 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900638 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900639 goto repeat;
640 }
641 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
Tejun Heoa1e10f72007-08-18 13:28:49 +0900642 "tries, giving up\n", ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900643 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900644 }
645
Tejun Heof3e81b12006-05-15 20:58:21 +0900646 /* this run is complete, make sure EH info is clear */
Tejun Heocf1b86c2007-08-06 18:36:23 +0900647 __ata_port_for_each_link(link, ap)
648 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900649
Tejun Heoe30349d2006-07-03 03:02:15 +0900650 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900651 * that if exception occurs after this point but
652 * before EH completion, SCSI midlayer will
653 * re-initiate EH.
654 */
655 host->host_eh_scheduled = 0;
656
Tejun Heoe30349d2006-07-03 03:02:15 +0900657 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900658 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900659 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900660 ap->ops->eng_timeout(ap);
661 }
662
663 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900664 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
665
666 scsi_eh_flush_done_q(&ap->eh_done_q);
667
Tejun Heoad9e2762006-05-15 20:58:12 +0900668 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900669 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900670
Tejun Heo1cdaf532006-07-03 16:07:26 +0900671 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900672 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900673 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
David Howells52bad642006-11-22 14:54:01 +0000674 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900675
676 if (ap->pflags & ATA_PFLAG_RECOVERED)
677 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900678
Tejun Heob51e9e52006-06-29 01:29:30 +0900679 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900680
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900681 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900682 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900683 wake_up_all(&ap->eh_wait_q);
684
Tejun Heoe30349d2006-07-03 03:02:15 +0900685 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900686
Tejun Heoece1d632006-04-02 18:51:53 +0900687 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900688}
689
690/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900691 * ata_port_wait_eh - Wait for the currently pending EH to complete
692 * @ap: Port to wait EH for
693 *
694 * Wait until the currently pending EH is complete.
695 *
696 * LOCKING:
697 * Kernel thread context (may sleep).
698 */
699void ata_port_wait_eh(struct ata_port *ap)
700{
701 unsigned long flags;
702 DEFINE_WAIT(wait);
703
704 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400705 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900706
Tejun Heob51e9e52006-06-29 01:29:30 +0900707 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900708 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400709 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900710 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400711 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900712 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900713 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900714
Jeff Garzikba6a1302006-06-22 23:46:10 -0400715 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900716
717 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400718 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900719 msleep(10);
720 goto retry;
721 }
722}
723
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900724static int ata_eh_nr_in_flight(struct ata_port *ap)
725{
726 unsigned int tag;
727 int nr = 0;
728
729 /* count only non-internal commands */
730 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
731 if (ata_qc_from_tag(ap, tag))
732 nr++;
733
734 return nr;
735}
736
737void ata_eh_fastdrain_timerfn(unsigned long arg)
738{
739 struct ata_port *ap = (void *)arg;
740 unsigned long flags;
741 int cnt;
742
743 spin_lock_irqsave(ap->lock, flags);
744
745 cnt = ata_eh_nr_in_flight(ap);
746
747 /* are we done? */
748 if (!cnt)
749 goto out_unlock;
750
751 if (cnt == ap->fastdrain_cnt) {
752 unsigned int tag;
753
754 /* No progress during the last interval, tag all
755 * in-flight qcs as timed out and freeze the port.
756 */
757 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
758 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
759 if (qc)
760 qc->err_mask |= AC_ERR_TIMEOUT;
761 }
762
763 ata_port_freeze(ap);
764 } else {
765 /* some qcs have finished, give it another chance */
766 ap->fastdrain_cnt = cnt;
767 ap->fastdrain_timer.expires =
Tejun Heo341c2c92008-05-20 02:17:51 +0900768 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900769 add_timer(&ap->fastdrain_timer);
770 }
771
772 out_unlock:
773 spin_unlock_irqrestore(ap->lock, flags);
774}
775
776/**
777 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
778 * @ap: target ATA port
779 * @fastdrain: activate fast drain
780 *
781 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
782 * is non-zero and EH wasn't pending before. Fast drain ensures
783 * that EH kicks in in timely manner.
784 *
785 * LOCKING:
786 * spin_lock_irqsave(host lock)
787 */
788static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
789{
790 int cnt;
791
792 /* already scheduled? */
793 if (ap->pflags & ATA_PFLAG_EH_PENDING)
794 return;
795
796 ap->pflags |= ATA_PFLAG_EH_PENDING;
797
798 if (!fastdrain)
799 return;
800
801 /* do we have in-flight qcs? */
802 cnt = ata_eh_nr_in_flight(ap);
803 if (!cnt)
804 return;
805
806 /* activate fast drain */
807 ap->fastdrain_cnt = cnt;
Tejun Heo341c2c92008-05-20 02:17:51 +0900808 ap->fastdrain_timer.expires =
809 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900810 add_timer(&ap->fastdrain_timer);
811}
812
Tejun Heof686bcb2006-05-15 20:58:05 +0900813/**
814 * ata_qc_schedule_eh - schedule qc for error handling
815 * @qc: command to schedule error handling for
816 *
817 * Schedule error handling for @qc. EH will kick in as soon as
818 * other commands are drained.
819 *
820 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400821 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900822 */
823void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
824{
825 struct ata_port *ap = qc->ap;
826
827 WARN_ON(!ap->ops->error_handler);
828
829 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900830 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900831
832 /* The following will fail if timeout has already expired.
833 * ata_scsi_error() takes care of such scmds on EH entry.
834 * Note that ATA_QCFLAG_FAILED is unconditionally set after
835 * this function completes.
836 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700837 blk_abort_request(qc->scsicmd->request);
Tejun Heof686bcb2006-05-15 20:58:05 +0900838}
839
Tejun Heo7b70fc02006-05-15 20:58:07 +0900840/**
841 * ata_port_schedule_eh - schedule error handling without a qc
842 * @ap: ATA port to schedule EH for
843 *
844 * Schedule error handling for @ap. EH will kick in as soon as
845 * all commands are drained.
846 *
847 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400848 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900849 */
850void ata_port_schedule_eh(struct ata_port *ap)
851{
852 WARN_ON(!ap->ops->error_handler);
853
Tejun Heof4d6d002007-05-01 11:50:15 +0200854 if (ap->pflags & ATA_PFLAG_INITIALIZING)
855 return;
856
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900857 ata_eh_set_pending(ap, 1);
Jeff Garzikcca39742006-08-24 03:19:22 -0400858 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900859
860 DPRINTK("port EH scheduled\n");
861}
862
Tejun Heodbd82612007-08-06 18:36:23 +0900863static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900864{
865 int tag, nr_aborted = 0;
866
867 WARN_ON(!ap->ops->error_handler);
868
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900869 /* we're gonna abort all commands, no need for fast drain */
870 ata_eh_set_pending(ap, 0);
871
Tejun Heo7b70fc02006-05-15 20:58:07 +0900872 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
873 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
874
Tejun Heodbd82612007-08-06 18:36:23 +0900875 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +0900876 qc->flags |= ATA_QCFLAG_FAILED;
877 ata_qc_complete(qc);
878 nr_aborted++;
879 }
880 }
881
882 if (!nr_aborted)
883 ata_port_schedule_eh(ap);
884
885 return nr_aborted;
886}
887
Tejun Heoe3180492006-05-15 20:58:09 +0900888/**
Tejun Heodbd82612007-08-06 18:36:23 +0900889 * ata_link_abort - abort all qc's on the link
890 * @link: ATA link to abort qc's for
891 *
892 * Abort all active qc's active on @link and schedule EH.
893 *
894 * LOCKING:
895 * spin_lock_irqsave(host lock)
896 *
897 * RETURNS:
898 * Number of aborted qc's.
899 */
900int ata_link_abort(struct ata_link *link)
901{
902 return ata_do_link_abort(link->ap, link);
903}
904
905/**
906 * ata_port_abort - abort all qc's on the port
907 * @ap: ATA port to abort qc's for
908 *
909 * Abort all active qc's of @ap and schedule EH.
910 *
911 * LOCKING:
912 * spin_lock_irqsave(host_set lock)
913 *
914 * RETURNS:
915 * Number of aborted qc's.
916 */
917int ata_port_abort(struct ata_port *ap)
918{
919 return ata_do_link_abort(ap, NULL);
920}
921
922/**
Tejun Heoe3180492006-05-15 20:58:09 +0900923 * __ata_port_freeze - freeze port
924 * @ap: ATA port to freeze
925 *
926 * This function is called when HSM violation or some other
927 * condition disrupts normal operation of the port. Frozen port
928 * is not allowed to perform any operation until the port is
929 * thawed, which usually follows a successful reset.
930 *
931 * ap->ops->freeze() callback can be used for freezing the port
932 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
933 * port cannot be frozen hardware-wise, the interrupt handler
934 * must ack and clear interrupts unconditionally while the port
935 * is frozen.
936 *
937 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400938 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900939 */
940static void __ata_port_freeze(struct ata_port *ap)
941{
942 WARN_ON(!ap->ops->error_handler);
943
944 if (ap->ops->freeze)
945 ap->ops->freeze(ap);
946
Tejun Heob51e9e52006-06-29 01:29:30 +0900947 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +0900948
Tejun Heo44877b42007-02-21 01:06:51 +0900949 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +0900950}
951
952/**
953 * ata_port_freeze - abort & freeze port
954 * @ap: ATA port to freeze
955 *
956 * Abort and freeze @ap.
957 *
958 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400959 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900960 *
961 * RETURNS:
962 * Number of aborted commands.
963 */
964int ata_port_freeze(struct ata_port *ap)
965{
966 int nr_aborted;
967
968 WARN_ON(!ap->ops->error_handler);
969
970 nr_aborted = ata_port_abort(ap);
971 __ata_port_freeze(ap);
972
973 return nr_aborted;
974}
975
976/**
Tejun Heo7d77b242007-09-23 13:14:13 +0900977 * sata_async_notification - SATA async notification handler
978 * @ap: ATA port where async notification is received
979 *
980 * Handler to be called when async notification via SDB FIS is
981 * received. This function schedules EH if necessary.
982 *
983 * LOCKING:
984 * spin_lock_irqsave(host lock)
985 *
986 * RETURNS:
987 * 1 if EH is scheduled, 0 otherwise.
988 */
989int sata_async_notification(struct ata_port *ap)
990{
991 u32 sntf;
992 int rc;
993
994 if (!(ap->flags & ATA_FLAG_AN))
995 return 0;
996
997 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
998 if (rc == 0)
999 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1000
Tejun Heo071f44b2008-04-07 22:47:22 +09001001 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001002 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +09001003 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001004 /* PMP is not attached. Check whether ATAPI
1005 * AN is configured. If so, notify media
1006 * change.
1007 */
1008 struct ata_device *dev = ap->link.device;
1009
1010 if ((dev->class == ATA_DEV_ATAPI) &&
1011 (dev->flags & ATA_DFLAG_AN))
1012 ata_scsi_media_change_notify(dev);
1013 return 0;
1014 } else {
1015 /* PMP is attached but SNTF is not available.
1016 * ATAPI async media change notification is
1017 * not used. The PMP must be reporting PHY
1018 * status change, schedule EH.
1019 */
1020 ata_port_schedule_eh(ap);
1021 return 1;
1022 }
1023 } else {
1024 /* PMP is attached and SNTF is available */
1025 struct ata_link *link;
1026
1027 /* check and notify ATAPI AN */
1028 ata_port_for_each_link(link, ap) {
1029 if (!(sntf & (1 << link->pmp)))
1030 continue;
1031
1032 if ((link->device->class == ATA_DEV_ATAPI) &&
1033 (link->device->flags & ATA_DFLAG_AN))
1034 ata_scsi_media_change_notify(link->device);
1035 }
1036
1037 /* If PMP is reporting that PHY status of some
1038 * downstream ports has changed, schedule EH.
1039 */
1040 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1041 ata_port_schedule_eh(ap);
1042 return 1;
1043 }
1044
1045 return 0;
1046 }
1047}
1048
1049/**
Tejun Heoe3180492006-05-15 20:58:09 +09001050 * ata_eh_freeze_port - EH helper to freeze port
1051 * @ap: ATA port to freeze
1052 *
1053 * Freeze @ap.
1054 *
1055 * LOCKING:
1056 * None.
1057 */
1058void ata_eh_freeze_port(struct ata_port *ap)
1059{
1060 unsigned long flags;
1061
1062 if (!ap->ops->error_handler)
1063 return;
1064
Jeff Garzikba6a1302006-06-22 23:46:10 -04001065 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001066 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -04001067 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001068}
1069
1070/**
1071 * ata_port_thaw_port - EH helper to thaw port
1072 * @ap: ATA port to thaw
1073 *
1074 * Thaw frozen port @ap.
1075 *
1076 * LOCKING:
1077 * None.
1078 */
1079void ata_eh_thaw_port(struct ata_port *ap)
1080{
1081 unsigned long flags;
1082
1083 if (!ap->ops->error_handler)
1084 return;
1085
Jeff Garzikba6a1302006-06-22 23:46:10 -04001086 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001087
Tejun Heob51e9e52006-06-29 01:29:30 +09001088 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001089
1090 if (ap->ops->thaw)
1091 ap->ops->thaw(ap);
1092
Jeff Garzikba6a1302006-06-22 23:46:10 -04001093 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001094
Tejun Heo44877b42007-02-21 01:06:51 +09001095 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001096}
1097
Tejun Heoece1d632006-04-02 18:51:53 +09001098static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1099{
1100 /* nada */
1101}
1102
1103static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1104{
1105 struct ata_port *ap = qc->ap;
1106 struct scsi_cmnd *scmd = qc->scsicmd;
1107 unsigned long flags;
1108
Jeff Garzikba6a1302006-06-22 23:46:10 -04001109 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001110 qc->scsidone = ata_eh_scsidone;
1111 __ata_qc_complete(qc);
1112 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -04001113 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001114
1115 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1116}
1117
1118/**
1119 * ata_eh_qc_complete - Complete an active ATA command from EH
1120 * @qc: Command to complete
1121 *
1122 * Indicate to the mid and upper layers that an ATA command has
1123 * completed. To be used from EH.
1124 */
1125void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1126{
1127 struct scsi_cmnd *scmd = qc->scsicmd;
1128 scmd->retries = scmd->allowed;
1129 __ata_eh_qc_complete(qc);
1130}
1131
1132/**
1133 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1134 * @qc: Command to retry
1135 *
1136 * Indicate to the mid and upper layers that an ATA command
1137 * should be retried. To be used from EH.
1138 *
1139 * SCSI midlayer limits the number of retries to scmd->allowed.
1140 * scmd->retries is decremented for commands which get retried
1141 * due to unrelated failures (qc->err_mask is zero).
1142 */
1143void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1144{
1145 struct scsi_cmnd *scmd = qc->scsicmd;
1146 if (!qc->err_mask && scmd->retries)
1147 scmd->retries--;
1148 __ata_eh_qc_complete(qc);
1149}
Tejun Heo022bdb02006-05-15 20:58:22 +09001150
1151/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001152 * ata_eh_detach_dev - detach ATA device
1153 * @dev: ATA device to detach
1154 *
1155 * Detach @dev.
1156 *
1157 * LOCKING:
1158 * None.
1159 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001160void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001161{
Tejun Heof58229f2007-08-06 18:36:23 +09001162 struct ata_link *link = dev->link;
1163 struct ata_port *ap = link->ap;
Tejun Heo90484eb2008-10-26 15:43:03 +09001164 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001165 unsigned long flags;
1166
1167 ata_dev_disable(dev);
1168
Jeff Garzikba6a1302006-06-22 23:46:10 -04001169 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001170
1171 dev->flags &= ~ATA_DFLAG_DETACH;
1172
1173 if (ata_scsi_offline_dev(dev)) {
1174 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001175 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001176 }
1177
Tejun Heo90484eb2008-10-26 15:43:03 +09001178 /* clear per-dev EH info */
Tejun Heof58229f2007-08-06 18:36:23 +09001179 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1180 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heo90484eb2008-10-26 15:43:03 +09001181 ehc->saved_xfer_mode[dev->devno] = 0;
1182 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heobeb07c12006-06-24 20:30:19 +09001183
Jeff Garzikba6a1302006-06-22 23:46:10 -04001184 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001185}
1186
1187/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001188 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001189 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001190 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001191 * @action: action about to be performed
1192 *
1193 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001194 * in @link->eh_info such that eh actions are not unnecessarily
1195 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001196 *
1197 * LOCKING:
1198 * None.
1199 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001200void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1201 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001202{
Tejun Heo955e57d2007-08-06 18:36:23 +09001203 struct ata_port *ap = link->ap;
1204 struct ata_eh_info *ehi = &link->eh_info;
1205 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001206 unsigned long flags;
1207
Jeff Garzikba6a1302006-06-22 23:46:10 -04001208 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001209
Tejun Heo955e57d2007-08-06 18:36:23 +09001210 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001211
Tejun Heoa568d1d2008-10-21 20:37:21 +09001212 /* About to take EH action, set RECOVERED. Ignore actions on
1213 * slave links as master will do them again.
1214 */
1215 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
Tejun Heo1cdaf532006-07-03 16:07:26 +09001216 ap->pflags |= ATA_PFLAG_RECOVERED;
1217
Jeff Garzikba6a1302006-06-22 23:46:10 -04001218 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001219}
1220
1221/**
Tejun Heo47005f22006-06-19 18:27:23 +09001222 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001223* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001224 * @dev: target ATA dev for per-dev action (can be NULL)
1225 * @action: action just completed
1226 *
1227 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001228 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001229 *
1230 * LOCKING:
1231 * None.
1232 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001233void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1234 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001235{
Tejun Heo955e57d2007-08-06 18:36:23 +09001236 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001237
Tejun Heo955e57d2007-08-06 18:36:23 +09001238 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001239}
1240
1241/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001242 * ata_err_string - convert err_mask to descriptive string
1243 * @err_mask: error mask to convert to string
1244 *
1245 * Convert @err_mask to descriptive string. Errors are
1246 * prioritized according to severity and only the most severe
1247 * error is reported.
1248 *
1249 * LOCKING:
1250 * None.
1251 *
1252 * RETURNS:
1253 * Descriptive string for @err_mask
1254 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001255static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001256{
1257 if (err_mask & AC_ERR_HOST_BUS)
1258 return "host bus error";
1259 if (err_mask & AC_ERR_ATA_BUS)
1260 return "ATA bus error";
1261 if (err_mask & AC_ERR_TIMEOUT)
1262 return "timeout";
1263 if (err_mask & AC_ERR_HSM)
1264 return "HSM violation";
1265 if (err_mask & AC_ERR_SYSTEM)
1266 return "internal error";
1267 if (err_mask & AC_ERR_MEDIA)
1268 return "media error";
1269 if (err_mask & AC_ERR_INVALID)
1270 return "invalid argument";
1271 if (err_mask & AC_ERR_DEV)
1272 return "device error";
1273 return "unknown error";
1274}
1275
1276/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001277 * ata_read_log_page - read a specific log page
1278 * @dev: target device
1279 * @page: page to read
1280 * @buf: buffer to store read page
1281 * @sectors: number of sectors to read
1282 *
1283 * Read log page using READ_LOG_EXT command.
1284 *
1285 * LOCKING:
1286 * Kernel thread context (may sleep).
1287 *
1288 * RETURNS:
1289 * 0 on success, AC_ERR_* mask otherwise.
1290 */
1291static unsigned int ata_read_log_page(struct ata_device *dev,
1292 u8 page, void *buf, unsigned int sectors)
1293{
1294 struct ata_taskfile tf;
1295 unsigned int err_mask;
1296
1297 DPRINTK("read log page - page %d\n", page);
1298
1299 ata_tf_init(dev, &tf);
1300 tf.command = ATA_CMD_READ_LOG_EXT;
1301 tf.lbal = page;
1302 tf.nsect = sectors;
1303 tf.hob_nsect = sectors >> 8;
1304 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1305 tf.protocol = ATA_PROT_PIO;
1306
1307 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001308 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001309
1310 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1311 return err_mask;
1312}
1313
1314/**
1315 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1316 * @dev: Device to read log page 10h from
1317 * @tag: Resulting tag of the failed command
1318 * @tf: Resulting taskfile registers of the failed command
1319 *
1320 * Read log page 10h to obtain NCQ error details and clear error
1321 * condition.
1322 *
1323 * LOCKING:
1324 * Kernel thread context (may sleep).
1325 *
1326 * RETURNS:
1327 * 0 on success, -errno otherwise.
1328 */
1329static int ata_eh_read_log_10h(struct ata_device *dev,
1330 int *tag, struct ata_taskfile *tf)
1331{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001332 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001333 unsigned int err_mask;
1334 u8 csum;
1335 int i;
1336
1337 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1338 if (err_mask)
1339 return -EIO;
1340
1341 csum = 0;
1342 for (i = 0; i < ATA_SECT_SIZE; i++)
1343 csum += buf[i];
1344 if (csum)
1345 ata_dev_printk(dev, KERN_WARNING,
1346 "invalid checksum 0x%x on log page 10h\n", csum);
1347
1348 if (buf[0] & 0x80)
1349 return -ENOENT;
1350
1351 *tag = buf[0] & 0x1f;
1352
1353 tf->command = buf[2];
1354 tf->feature = buf[3];
1355 tf->lbal = buf[4];
1356 tf->lbam = buf[5];
1357 tf->lbah = buf[6];
1358 tf->device = buf[7];
1359 tf->hob_lbal = buf[8];
1360 tf->hob_lbam = buf[9];
1361 tf->hob_lbah = buf[10];
1362 tf->nsect = buf[12];
1363 tf->hob_nsect = buf[13];
1364
1365 return 0;
1366}
1367
1368/**
Tejun Heo11fc33d2008-08-30 14:20:01 +02001369 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1370 * @dev: target ATAPI device
1371 * @r_sense_key: out parameter for sense_key
1372 *
1373 * Perform ATAPI TEST_UNIT_READY.
1374 *
1375 * LOCKING:
1376 * EH context (may sleep).
1377 *
1378 * RETURNS:
1379 * 0 on success, AC_ERR_* mask on failure.
1380 */
1381static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1382{
1383 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1384 struct ata_taskfile tf;
1385 unsigned int err_mask;
1386
1387 ata_tf_init(dev, &tf);
1388
1389 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1390 tf.command = ATA_CMD_PACKET;
1391 tf.protocol = ATAPI_PROT_NODATA;
1392
1393 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1394 if (err_mask == AC_ERR_DEV)
1395 *r_sense_key = tf.feature >> 4;
1396 return err_mask;
1397}
1398
1399/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001400 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1401 * @dev: device to perform REQUEST_SENSE to
1402 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
Tejun Heo3eabddb2008-06-10 18:28:05 +09001403 * @dfl_sense_key: default sense key to use
Tejun Heo022bdb02006-05-15 20:58:22 +09001404 *
1405 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1406 * SENSE. This function is EH helper.
1407 *
1408 * LOCKING:
1409 * Kernel thread context (may sleep).
1410 *
1411 * RETURNS:
1412 * 0 on success, AC_ERR_* mask on failure
1413 */
Tejun Heo3eabddb2008-06-10 18:28:05 +09001414static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1415 u8 *sense_buf, u8 dfl_sense_key)
Tejun Heo022bdb02006-05-15 20:58:22 +09001416{
Tejun Heo3eabddb2008-06-10 18:28:05 +09001417 u8 cdb[ATAPI_CDB_LEN] =
1418 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001419 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001420 struct ata_taskfile tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001421
1422 DPRINTK("ATAPI request sense\n");
1423
Tejun Heo022bdb02006-05-15 20:58:22 +09001424 /* FIXME: is this needed? */
1425 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1426
Albert Lee56287762007-04-02 11:30:46 +08001427 /* initialize sense_buf with the error register,
1428 * for the case where they are -not- overwritten
1429 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001430 sense_buf[0] = 0x70;
Tejun Heo3eabddb2008-06-10 18:28:05 +09001431 sense_buf[2] = dfl_sense_key;
Albert Lee56287762007-04-02 11:30:46 +08001432
Jeff Garzika617c092007-05-21 20:14:23 -04001433 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001434 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001435
Tejun Heo022bdb02006-05-15 20:58:22 +09001436 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1437 tf.command = ATA_CMD_PACKET;
1438
1439 /* is it pointless to prefer PIO for "safety reasons"? */
1440 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001441 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001442 tf.feature |= ATAPI_PKT_DMA;
1443 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001444 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001445 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1446 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001447 }
1448
1449 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001450 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001451}
1452
1453/**
1454 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001455 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001456 *
1457 * Analyze SError if available and further determine cause of
1458 * failure.
1459 *
1460 * LOCKING:
1461 * None.
1462 */
Tejun Heo02607312007-08-06 18:36:23 +09001463static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001464{
Tejun Heo02607312007-08-06 18:36:23 +09001465 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001466 u32 serror = ehc->i.serror;
1467 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001468 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001469
Tejun Heoe0614db2008-05-19 01:15:09 +09001470 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001471 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001472 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001473 }
1474 if (serror & SERR_PROTOCOL) {
1475 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001476 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001477 }
1478 if (serror & SERR_INTERNAL) {
1479 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001480 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001481 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001482
1483 /* Determine whether a hotplug event has occurred. Both
1484 * SError.N/X are considered hotplug events for enabled or
1485 * host links. For disabled PMP links, only N bit is
1486 * considered as X bit is left at 1 for link plugging.
1487 */
1488 hotplug_mask = 0;
1489
1490 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1491 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1492 else
1493 hotplug_mask = SERR_PHYRDY_CHG;
1494
1495 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001496 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001497
1498 ehc->i.err_mask |= err_mask;
1499 ehc->i.action |= action;
1500}
1501
1502/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001503 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001504 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001505 *
1506 * Read log page 10h, determine the offending qc and acquire
1507 * error status TF. For NCQ device errors, all LLDDs have to do
1508 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1509 * care of the rest.
1510 *
1511 * LOCKING:
1512 * Kernel thread context (may sleep).
1513 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001514void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001515{
Tejun Heo02607312007-08-06 18:36:23 +09001516 struct ata_port *ap = link->ap;
1517 struct ata_eh_context *ehc = &link->eh_context;
1518 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001519 struct ata_queued_cmd *qc;
1520 struct ata_taskfile tf;
1521 int tag, rc;
1522
1523 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001524 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001525 return;
1526
1527 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001528 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001529 return;
1530
1531 /* has LLDD analyzed already? */
1532 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1533 qc = __ata_qc_from_tag(ap, tag);
1534
1535 if (!(qc->flags & ATA_QCFLAG_FAILED))
1536 continue;
1537
1538 if (qc->err_mask)
1539 return;
1540 }
1541
1542 /* okay, this error is ours */
1543 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1544 if (rc) {
Tejun Heo02607312007-08-06 18:36:23 +09001545 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001546 "(errno=%d)\n", rc);
1547 return;
1548 }
1549
Tejun Heo02607312007-08-06 18:36:23 +09001550 if (!(link->sactive & (1 << tag))) {
1551 ata_link_printk(link, KERN_ERR, "log page 10h reported "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001552 "inactive tag %d\n", tag);
1553 return;
1554 }
1555
1556 /* we've got the perpetrator, condemn it */
1557 qc = __ata_qc_from_tag(ap, tag);
1558 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001559 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001560 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001561 ehc->i.err_mask &= ~AC_ERR_DEV;
1562}
1563
1564/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001565 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1566 * @qc: qc to analyze
1567 * @tf: Taskfile registers to analyze
1568 *
1569 * Analyze taskfile of @qc and further determine cause of
1570 * failure. This function also requests ATAPI sense data if
1571 * avaliable.
1572 *
1573 * LOCKING:
1574 * Kernel thread context (may sleep).
1575 *
1576 * RETURNS:
1577 * Determined recovery action
1578 */
1579static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1580 const struct ata_taskfile *tf)
1581{
1582 unsigned int tmp, action = 0;
1583 u8 stat = tf->command, err = tf->feature;
1584
1585 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1586 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001587 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001588 }
1589
Tejun Heoa51d6442007-03-20 15:24:11 +09001590 if (stat & (ATA_ERR | ATA_DF))
1591 qc->err_mask |= AC_ERR_DEV;
1592 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001593 return 0;
1594
1595 switch (qc->dev->class) {
1596 case ATA_DEV_ATA:
1597 if (err & ATA_ICRC)
1598 qc->err_mask |= AC_ERR_ATA_BUS;
1599 if (err & ATA_UNC)
1600 qc->err_mask |= AC_ERR_MEDIA;
1601 if (err & ATA_IDNF)
1602 qc->err_mask |= AC_ERR_INVALID;
1603 break;
1604
1605 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001606 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Tejun Heo3eabddb2008-06-10 18:28:05 +09001607 tmp = atapi_eh_request_sense(qc->dev,
1608 qc->scsicmd->sense_buffer,
1609 qc->result_tf.feature >> 4);
Tejun Heoa569a302006-11-21 10:40:51 +09001610 if (!tmp) {
1611 /* ATA_QCFLAG_SENSE_VALID is used to
1612 * tell atapi_qc_complete() that sense
1613 * data is already valid.
1614 *
1615 * TODO: interpret sense data and set
1616 * appropriate err_mask.
1617 */
1618 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1619 } else
1620 qc->err_mask |= tmp;
1621 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001622 }
1623
1624 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001625 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001626
1627 return action;
1628}
1629
Tejun Heo76326ac2007-11-27 19:28:59 +09001630static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1631 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001632{
Tejun Heo76326ac2007-11-27 19:28:59 +09001633 int base = 0;
1634
1635 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1636 *xfer_ok = 1;
1637
1638 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001639 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001640
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001641 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001642 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001643
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001644 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001645 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001646
Tejun Heo3884f7b2007-11-27 19:28:56 +09001647 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001648 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001649 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001650 if ((err_mask &
1651 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001652 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001653 }
1654
1655 return 0;
1656}
1657
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001658struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001659 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001660 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001661 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001662};
1663
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001664static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001665{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001666 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001667 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001668
1669 if (ent->timestamp < arg->since)
1670 return -1;
1671
Tejun Heo76326ac2007-11-27 19:28:59 +09001672 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1673 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001674 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001675
Tejun Heo022bdb02006-05-15 20:58:22 +09001676 return 0;
1677}
1678
1679/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001680 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001681 * @dev: Device of interest
1682 *
1683 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001684 * whether NCQ needs to be turned off, transfer speed should be
1685 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001686 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001687 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001688 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001689 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1690 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001691 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001692 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001693 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001694 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1695 * data transfer hasn't been verified.
1696 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001697 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001698 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001699 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001700 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001701 * SPEED_DOWN : Speed down transfer speed but don't fall back
1702 * to PIO.
1703 *
1704 * FALLBACK_TO_PIO : Fall back to PIO.
1705 *
1706 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001707 * taken per error. An action triggered by non-DUBIOUS errors
1708 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1709 * This is to expedite speed down decisions right after device is
1710 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001711 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001712 * The followings are speed down rules. #1 and #2 deal with
1713 * DUBIOUS errors.
1714 *
1715 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1716 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1717 *
1718 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1719 * occurred during last 5 mins, NCQ_OFF.
1720 *
1721 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Tejun Heo3884f7b2007-11-27 19:28:56 +09001722 * ocurred during last 5 mins, FALLBACK_TO_PIO
1723 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001724 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001725 * during last 10 mins, NCQ_OFF.
1726 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001727 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001728 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001729 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001730 * LOCKING:
1731 * Inherited from caller.
1732 *
1733 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001734 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001735 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001736static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001737{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001738 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1739 u64 j64 = get_jiffies_64();
1740 struct speed_down_verdict_arg arg;
1741 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001742
Tejun Heo3884f7b2007-11-27 19:28:56 +09001743 /* scan past 5 mins of error history */
1744 memset(&arg, 0, sizeof(arg));
1745 arg.since = j64 - min(j64, j5mins);
1746 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1747
Tejun Heo76326ac2007-11-27 19:28:59 +09001748 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1749 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1750 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1751 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1752
1753 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1754 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1755 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1756
Tejun Heo3884f7b2007-11-27 19:28:56 +09001757 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1758 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b2007-11-27 19:28:57 +09001759 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001760 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1761
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001762 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001763 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001764 arg.since = j64 - min(j64, j10mins);
1765 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001766
Tejun Heo3884f7b2007-11-27 19:28:56 +09001767 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1768 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001769 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001770
1771 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1772 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b2007-11-27 19:28:57 +09001773 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001774 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001775
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001776 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001777}
1778
1779/**
1780 * ata_eh_speed_down - record error and speed down if necessary
1781 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001782 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001783 * @err_mask: err_mask of the error
1784 *
1785 * Record error and examine error history to determine whether
1786 * adjusting transmission speed is necessary. It also sets
1787 * transmission limits appropriately if such adjustment is
1788 * necessary.
1789 *
1790 * LOCKING:
1791 * Kernel thread context (may sleep).
1792 *
1793 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001794 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001795 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001796static unsigned int ata_eh_speed_down(struct ata_device *dev,
1797 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001798{
Tejun Heob1c72912008-07-31 17:02:43 +09001799 struct ata_link *link = ata_dev_phys_link(dev);
Tejun Heo76326ac2007-11-27 19:28:59 +09001800 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001801 unsigned int verdict;
1802 unsigned int action = 0;
1803
1804 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001805 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001806 return 0;
1807
1808 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001809 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001810 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001811
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001812 /* turn off NCQ? */
1813 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1814 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1815 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1816 dev->flags |= ATA_DFLAG_NCQ_OFF;
1817 ata_dev_printk(dev, KERN_WARNING,
1818 "NCQ disabled due to excessive errors\n");
1819 goto done;
1820 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001821
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001822 /* speed down? */
1823 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1824 /* speed down SATA link speed if possible */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001825 if (sata_down_spd_limit(link) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001826 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001827 goto done;
1828 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001829
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001830 /* lower transfer mode */
1831 if (dev->spdn_cnt < 2) {
1832 static const int dma_dnxfer_sel[] =
1833 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1834 static const int pio_dnxfer_sel[] =
1835 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1836 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001837
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001838 if (dev->xfer_shift != ATA_SHIFT_PIO)
1839 sel = dma_dnxfer_sel[dev->spdn_cnt];
1840 else
1841 sel = pio_dnxfer_sel[dev->spdn_cnt];
1842
1843 dev->spdn_cnt++;
1844
1845 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001846 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001847 goto done;
1848 }
1849 }
1850 }
1851
1852 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b2007-11-27 19:28:57 +09001853 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001854 */
1855 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b2007-11-27 19:28:57 +09001856 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001857 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1858 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1859 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09001860 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001861 goto done;
1862 }
1863 }
1864
Tejun Heo022bdb02006-05-15 20:58:22 +09001865 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001866 done:
1867 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09001868 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1869 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001870 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001871}
1872
1873/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001874 * ata_eh_link_autopsy - analyze error and determine recovery action
1875 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09001876 *
Tejun Heo02607312007-08-06 18:36:23 +09001877 * Analyze why @link failed and determine which recovery actions
1878 * are needed. This function also sets more detailed AC_ERR_*
1879 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09001880 *
1881 * LOCKING:
1882 * Kernel thread context (may sleep).
1883 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001884static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001885{
Tejun Heo02607312007-08-06 18:36:23 +09001886 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09001887 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09001888 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001889 unsigned int all_err_mask = 0, eflags = 0;
1890 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09001891 u32 serror;
1892 int rc;
1893
1894 DPRINTK("ENTER\n");
1895
Tejun Heo1cdaf532006-07-03 16:07:26 +09001896 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1897 return;
1898
Tejun Heo022bdb02006-05-15 20:58:22 +09001899 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09001900 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09001901 if (rc == 0) {
1902 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09001903 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09001904 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09001905 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09001906 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09001907 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09001908 ehc->i.err_mask |= AC_ERR_OTHER;
1909 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001910
Tejun Heoe8ee8452006-05-15 21:03:46 +09001911 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09001912 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001913
Tejun Heo022bdb02006-05-15 20:58:22 +09001914 /* any real error trumps AC_ERR_OTHER */
1915 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1916 ehc->i.err_mask &= ~AC_ERR_OTHER;
1917
1918 all_err_mask |= ehc->i.err_mask;
1919
1920 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1921 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1922
Tejun Heob1c72912008-07-31 17:02:43 +09001923 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1924 ata_dev_phys_link(qc->dev) != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001925 continue;
1926
1927 /* inherit upper level err_mask */
1928 qc->err_mask |= ehc->i.err_mask;
1929
Tejun Heo022bdb02006-05-15 20:58:22 +09001930 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001931 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001932
1933 /* DEV errors are probably spurious in case of ATA_BUS error */
1934 if (qc->err_mask & AC_ERR_ATA_BUS)
1935 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1936 AC_ERR_INVALID);
1937
1938 /* any real error trumps unknown error */
1939 if (qc->err_mask & ~AC_ERR_OTHER)
1940 qc->err_mask &= ~AC_ERR_OTHER;
1941
1942 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09001943 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09001944 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09001945
Tejun Heo03faab72008-03-27 19:14:24 +09001946 /* determine whether the command is worth retrying */
1947 if (!(qc->err_mask & AC_ERR_INVALID) &&
1948 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1949 qc->flags |= ATA_QCFLAG_RETRY;
1950
Tejun Heo022bdb02006-05-15 20:58:22 +09001951 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001952 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09001953 all_err_mask |= qc->err_mask;
1954 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001955 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09001956 }
1957
Tejun Heoa20f33f2006-05-16 12:58:24 +09001958 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09001959 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09001960 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09001961 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001962 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1963 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09001964 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09001965
Tejun Heodfcc1732007-10-31 10:17:05 +09001966 /* If we have offending qcs and the associated failed device,
1967 * perform per-dev EH action only on the offending device.
1968 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09001969 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09001970 ehc->i.dev_action[ehc->i.dev->devno] |=
1971 ehc->i.action & ATA_EH_PERDEV_MASK;
1972 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09001973 }
1974
Tejun Heo2695e362008-01-10 13:41:23 +09001975 /* propagate timeout to host link */
1976 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1977 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1978
1979 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09001980 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09001981 if (!dev && ((ata_link_max_devices(link) == 1 &&
1982 ata_dev_enabled(link->device))))
1983 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09001984
Tejun Heo76326ac2007-11-27 19:28:59 +09001985 if (dev) {
1986 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1987 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001988 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09001989 }
Tejun Heodfcc1732007-10-31 10:17:05 +09001990
Tejun Heo022bdb02006-05-15 20:58:22 +09001991 DPRINTK("EXIT\n");
1992}
1993
1994/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001995 * ata_eh_autopsy - analyze error and determine recovery action
1996 * @ap: host port to perform autopsy on
1997 *
1998 * Analyze all links of @ap and determine why they failed and
1999 * which recovery actions are needed.
2000 *
2001 * LOCKING:
2002 * Kernel thread context (may sleep).
2003 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002004void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002005{
2006 struct ata_link *link;
2007
Tejun Heo2695e362008-01-10 13:41:23 +09002008 ata_port_for_each_link(link, ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002009 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09002010
Tejun Heob1c72912008-07-31 17:02:43 +09002011 /* Handle the frigging slave link. Autopsy is done similarly
2012 * but actions and flags are transferred over to the master
2013 * link and handled from there.
2014 */
2015 if (ap->slave_link) {
2016 struct ata_eh_context *mehc = &ap->link.eh_context;
2017 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2018
Tejun Heo848e4c62008-10-21 14:26:39 +09002019 /* transfer control flags from master to slave */
2020 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2021
2022 /* perform autopsy on the slave link */
Tejun Heob1c72912008-07-31 17:02:43 +09002023 ata_eh_link_autopsy(ap->slave_link);
2024
Tejun Heo848e4c62008-10-21 14:26:39 +09002025 /* transfer actions from slave to master and clear slave */
Tejun Heob1c72912008-07-31 17:02:43 +09002026 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2027 mehc->i.action |= sehc->i.action;
2028 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2029 mehc->i.flags |= sehc->i.flags;
2030 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2031 }
2032
Tejun Heo2695e362008-01-10 13:41:23 +09002033 /* Autopsy of fanout ports can affect host link autopsy.
2034 * Perform host link autopsy last.
2035 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002036 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09002037 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09002038}
2039
2040/**
2041 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09002042 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09002043 *
2044 * Report EH to user.
2045 *
2046 * LOCKING:
2047 * None.
2048 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002049static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002050{
Tejun Heo02607312007-08-06 18:36:23 +09002051 struct ata_port *ap = link->ap;
2052 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002053 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09002054 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09002055 int tag, nr_failed = 0;
2056
Tejun Heo94ff3d52007-10-09 14:57:56 +09002057 if (ehc->i.flags & ATA_EHI_QUIET)
2058 return;
2059
Tejun Heo022bdb02006-05-15 20:58:22 +09002060 desc = NULL;
2061 if (ehc->i.desc[0] != '\0')
2062 desc = ehc->i.desc;
2063
2064 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2065 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2066
Tejun Heob1c72912008-07-31 17:02:43 +09002067 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2068 ata_dev_phys_link(qc->dev) != link ||
Tejun Heoe027bd32007-10-26 16:19:26 +09002069 ((qc->flags & ATA_QCFLAG_QUIET) &&
2070 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002071 continue;
2072 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2073 continue;
2074
2075 nr_failed++;
2076 }
2077
2078 if (!nr_failed && !ehc->i.err_mask)
2079 return;
2080
2081 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09002082 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09002083 frozen = " frozen";
2084
Tejun Heoa1e10f72007-08-18 13:28:49 +09002085 memset(tries_buf, 0, sizeof(tries_buf));
2086 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2087 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2088 ap->eh_tries);
2089
Tejun Heo022bdb02006-05-15 20:58:22 +09002090 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09002091 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002092 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2093 ehc->i.err_mask, link->sactive, ehc->i.serror,
2094 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002095 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09002096 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002097 } else {
Tejun Heo02607312007-08-06 18:36:23 +09002098 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002099 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2100 ehc->i.err_mask, link->sactive, ehc->i.serror,
2101 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002102 if (desc)
Tejun Heo02607312007-08-06 18:36:23 +09002103 ata_link_printk(link, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002104 }
2105
Robert Hancock1333e192007-10-02 11:22:02 -04002106 if (ehc->i.serror)
Tejun Heoda0e21d2008-07-31 16:08:38 +09002107 ata_link_printk(link, KERN_ERR,
Robert Hancock1333e192007-10-02 11:22:02 -04002108 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2109 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2110 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2111 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2112 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2113 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2114 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2115 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2116 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2117 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2118 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2119 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2120 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2121 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2122 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2123 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2124 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002125 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002126
Tejun Heo022bdb02006-05-15 20:58:22 +09002127 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2128 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09002129 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09002130 const u8 *cdb = qc->cdb;
2131 char data_buf[20] = "";
2132 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002133
Tejun Heo02607312007-08-06 18:36:23 +09002134 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
Tejun Heob1c72912008-07-31 17:02:43 +09002135 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002136 continue;
2137
Tejun Heoabb6a882007-11-28 23:16:09 +09002138 if (qc->dma_dir != DMA_NONE) {
2139 static const char *dma_str[] = {
2140 [DMA_BIDIRECTIONAL] = "bidi",
2141 [DMA_TO_DEVICE] = "out",
2142 [DMA_FROM_DEVICE] = "in",
2143 };
2144 static const char *prot_str[] = {
2145 [ATA_PROT_PIO] = "pio",
2146 [ATA_PROT_DMA] = "dma",
2147 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05002148 [ATAPI_PROT_PIO] = "pio",
2149 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09002150 };
2151
2152 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2153 prot_str[qc->tf.protocol], qc->nbytes,
2154 dma_str[qc->dma_dir]);
2155 }
2156
Jeff Garzike39eec12007-12-01 18:05:39 -05002157 if (ata_is_atapi(qc->tf.protocol))
Tejun Heoabb6a882007-11-28 23:16:09 +09002158 snprintf(cdb_buf, sizeof(cdb_buf),
2159 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2160 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2161 cdb[0], cdb[1], cdb[2], cdb[3],
2162 cdb[4], cdb[5], cdb[6], cdb[7],
2163 cdb[8], cdb[9], cdb[10], cdb[11],
2164 cdb[12], cdb[13], cdb[14], cdb[15]);
2165
Tejun Heo8a937582006-11-14 22:36:12 +09002166 ata_dev_printk(qc->dev, KERN_ERR,
2167 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002168 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002169 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002170 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002171 cmd->command, cmd->feature, cmd->nsect,
2172 cmd->lbal, cmd->lbam, cmd->lbah,
2173 cmd->hob_feature, cmd->hob_nsect,
2174 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002175 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002176 res->command, res->feature, res->nsect,
2177 res->lbal, res->lbam, res->lbah,
2178 res->hob_feature, res->hob_nsect,
2179 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002180 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2181 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002182
2183 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002184 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002185 if (res->command & ATA_BUSY)
2186 ata_dev_printk(qc->dev, KERN_ERR,
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002187 "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002188 else
2189 ata_dev_printk(qc->dev, KERN_ERR,
2190 "status: { %s%s%s%s}\n",
2191 res->command & ATA_DRDY ? "DRDY " : "",
2192 res->command & ATA_DF ? "DF " : "",
2193 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002194 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002195 }
2196
2197 if (cmd->command != ATA_CMD_PACKET &&
2198 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2199 ATA_ABORTED)))
2200 ata_dev_printk(qc->dev, KERN_ERR,
2201 "error: { %s%s%s%s}\n",
2202 res->feature & ATA_ICRC ? "ICRC " : "",
2203 res->feature & ATA_UNC ? "UNC " : "",
2204 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002205 res->feature & ATA_ABORTED ? "ABRT " : "");
Tejun Heo022bdb02006-05-15 20:58:22 +09002206 }
2207}
2208
Tejun Heo9b1e2652007-08-06 18:36:24 +09002209/**
2210 * ata_eh_report - report error handling to user
2211 * @ap: ATA port to report EH about
2212 *
2213 * Report EH to user.
2214 *
2215 * LOCKING:
2216 * None.
2217 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002218void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002219{
2220 struct ata_link *link;
2221
2222 __ata_port_for_each_link(link, ap)
2223 ata_eh_link_report(link);
2224}
2225
Tejun Heocc0680a2007-08-06 18:36:23 +09002226static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heob1c72912008-07-31 17:02:43 +09002227 unsigned int *classes, unsigned long deadline,
2228 bool clear_classes)
Tejun Heod87fa382006-05-31 18:28:24 +09002229{
Tejun Heof58229f2007-08-06 18:36:23 +09002230 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002231
Tejun Heob1c72912008-07-31 17:02:43 +09002232 if (clear_classes)
2233 ata_link_for_each_dev(dev, link)
2234 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002235
Tejun Heof0465192008-05-19 01:15:08 +09002236 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002237}
2238
Tejun Heoae791c02007-09-23 13:14:12 +09002239static int ata_eh_followup_srst_needed(struct ata_link *link,
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002240 int rc, const unsigned int *classes)
Tejun Heo664faf02006-05-31 18:27:50 +09002241{
Tejun Heo45db2f62008-04-08 01:46:56 +09002242 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002243 return 0;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002244 if (rc == -EAGAIN)
2245 return 1;
Tejun Heo071f44b2008-04-07 22:47:22 +09002246 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002247 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002248 return 0;
2249}
2250
Tejun Heofb7fd612007-09-23 13:14:12 +09002251int ata_eh_reset(struct ata_link *link, int classify,
2252 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2253 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002254{
Tejun Heoafaa5c32007-10-09 15:06:10 +09002255 struct ata_port *ap = link->ap;
Tejun Heob1c72912008-07-31 17:02:43 +09002256 struct ata_link *slave = ap->slave_link;
Tejun Heo936fd732007-08-06 18:36:23 +09002257 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heob1c72912008-07-31 17:02:43 +09002258 struct ata_eh_context *sehc = &slave->eh_context;
Tejun Heo664faf02006-05-31 18:27:50 +09002259 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002260 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002261 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heod8af0eb2008-05-20 02:17:53 +09002262 int max_tries = 0, try = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002263 struct ata_link *failed_link;
Tejun Heof58229f2007-08-06 18:36:23 +09002264 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002265 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002266 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002267 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002268 u32 sstatus;
Tejun Heob1c72912008-07-31 17:02:43 +09002269 int nr_unknown, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002270
Tejun Heo932648b2008-05-19 01:15:06 +09002271 /*
2272 * Prepare to reset
2273 */
Tejun Heod8af0eb2008-05-20 02:17:53 +09002274 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2275 max_tries++;
Tejun Heo05944bd2008-08-13 20:19:09 +09002276 if (link->flags & ATA_LFLAG_NO_HRST)
2277 hardreset = NULL;
2278 if (link->flags & ATA_LFLAG_NO_SRST)
2279 softreset = NULL;
Tejun Heod8af0eb2008-05-20 02:17:53 +09002280
Tejun Heo19b72322008-11-04 17:08:40 +09002281 /* make sure each reset attemp is at least COOL_DOWN apart */
2282 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2283 now = jiffies;
2284 WARN_ON(time_after(ehc->last_reset, now));
2285 deadline = ata_deadline(ehc->last_reset,
2286 ATA_EH_RESET_COOL_DOWN);
2287 if (time_before(now, deadline))
2288 schedule_timeout_uninterruptible(deadline - now);
2289 }
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002290
Tejun Heoafaa5c32007-10-09 15:06:10 +09002291 spin_lock_irqsave(ap->lock, flags);
2292 ap->pflags |= ATA_PFLAG_RESETTING;
2293 spin_unlock_irqrestore(ap->lock, flags);
2294
Tejun Heocf480622008-01-24 00:05:14 +09002295 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002296
Tejun Heocdeab112007-10-29 16:41:09 +09002297 ata_link_for_each_dev(dev, link) {
2298 /* If we issue an SRST then an ATA drive (not ATAPI)
2299 * may change configuration and be in PIO0 timing. If
2300 * we do a hard reset (or are coming from power on)
2301 * this is true for ATA or ATAPI. Until we've set a
2302 * suitable controller mode we should not touch the
2303 * bus as we may be talking too fast.
2304 */
2305 dev->pio_mode = XFER_PIO_0;
2306
2307 /* If the controller has a pio mode setup function
2308 * then use it to set the chipset to rights. Don't
2309 * touch the DMA setup as that will be dealt with when
2310 * configuring devices.
2311 */
2312 if (ap->ops->set_piomode)
2313 ap->ops->set_piomode(ap, dev);
2314 }
2315
Tejun Heocf480622008-01-24 00:05:14 +09002316 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002317 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002318 ehc->i.action &= ~ATA_EH_RESET;
2319 if (hardreset) {
2320 reset = hardreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002321 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002322 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002323 reset = softreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002324 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heocf480622008-01-24 00:05:14 +09002325 }
Tejun Heof5914a42006-05-31 18:27:48 +09002326
2327 if (prereset) {
Tejun Heob1c72912008-07-31 17:02:43 +09002328 unsigned long deadline = ata_deadline(jiffies,
2329 ATA_EH_PRERESET_TIMEOUT);
2330
2331 if (slave) {
2332 sehc->i.action &= ~ATA_EH_RESET;
2333 sehc->i.action |= ehc->i.action;
2334 }
2335
2336 rc = prereset(link, deadline);
2337
2338 /* If present, do prereset on slave link too. Reset
2339 * is skipped iff both master and slave links report
2340 * -ENOENT or clear ATA_EH_RESET.
2341 */
2342 if (slave && (rc == 0 || rc == -ENOENT)) {
2343 int tmp;
2344
2345 tmp = prereset(slave, deadline);
2346 if (tmp != -ENOENT)
2347 rc = tmp;
2348
2349 ehc->i.action |= sehc->i.action;
2350 }
2351
Tejun Heof5914a42006-05-31 18:27:48 +09002352 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002353 if (rc == -ENOENT) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002354 ata_link_printk(link, KERN_DEBUG,
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002355 "port disabled. ignoring.\n");
Tejun Heocf480622008-01-24 00:05:14 +09002356 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002357
Tejun Heo936fd732007-08-06 18:36:23 +09002358 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002359 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002360
2361 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002362 } else
Tejun Heocc0680a2007-08-06 18:36:23 +09002363 ata_link_printk(link, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09002364 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002365 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002366 }
Tejun Heof5914a42006-05-31 18:27:48 +09002367
Tejun Heo932648b2008-05-19 01:15:06 +09002368 /* prereset() might have cleared ATA_EH_RESET. If so,
2369 * bang classes and return.
2370 */
2371 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2372 ata_link_for_each_dev(dev, link)
2373 classes[dev->devno] = ATA_DEV_NONE;
2374 rc = 0;
2375 goto out;
2376 }
Tejun Heof5914a42006-05-31 18:27:48 +09002377 }
2378
Tejun Heo022bdb02006-05-15 20:58:22 +09002379 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002380 /*
2381 * Perform reset
2382 */
Tejun Heodc98c322008-05-19 01:15:07 +09002383 if (ata_is_host_link(link))
2384 ata_eh_freeze_port(ap);
2385
Tejun Heo341c2c92008-05-20 02:17:51 +09002386 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
Tejun Heo31daabd2007-02-02 16:50:52 +09002387
Tejun Heo932648b2008-05-19 01:15:06 +09002388 if (reset) {
2389 if (verbose)
2390 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2391 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002392
Tejun Heo932648b2008-05-19 01:15:06 +09002393 /* mark that this EH session started with reset */
Tejun Heo19b72322008-11-04 17:08:40 +09002394 ehc->last_reset = jiffies;
Tejun Heo932648b2008-05-19 01:15:06 +09002395 if (reset == hardreset)
2396 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2397 else
2398 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002399
Tejun Heob1c72912008-07-31 17:02:43 +09002400 rc = ata_do_reset(link, reset, classes, deadline, true);
2401 if (rc && rc != -EAGAIN) {
2402 failed_link = link;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002403 goto fail;
Tejun Heob1c72912008-07-31 17:02:43 +09002404 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002405
Tejun Heob1c72912008-07-31 17:02:43 +09002406 /* hardreset slave link if existent */
2407 if (slave && reset == hardreset) {
2408 int tmp;
2409
2410 if (verbose)
2411 ata_link_printk(slave, KERN_INFO,
2412 "hard resetting link\n");
2413
2414 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2415 tmp = ata_do_reset(slave, reset, classes, deadline,
2416 false);
2417 switch (tmp) {
2418 case -EAGAIN:
2419 rc = -EAGAIN;
2420 case 0:
2421 break;
2422 default:
2423 failed_link = slave;
2424 rc = tmp;
2425 goto fail;
2426 }
2427 }
2428
2429 /* perform follow-up SRST if necessary */
Tejun Heo932648b2008-05-19 01:15:06 +09002430 if (reset == hardreset &&
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002431 ata_eh_followup_srst_needed(link, rc, classes)) {
Tejun Heo932648b2008-05-19 01:15:06 +09002432 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002433
Tejun Heo932648b2008-05-19 01:15:06 +09002434 if (!reset) {
2435 ata_link_printk(link, KERN_ERR,
2436 "follow-up softreset required "
2437 "but no softreset avaliable\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002438 failed_link = link;
Tejun Heo932648b2008-05-19 01:15:06 +09002439 rc = -EINVAL;
2440 goto fail;
2441 }
2442
2443 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002444 rc = ata_do_reset(link, reset, classes, deadline, true);
Tejun Heo664faf02006-05-31 18:27:50 +09002445 }
Tejun Heo932648b2008-05-19 01:15:06 +09002446 } else {
2447 if (verbose)
2448 ata_link_printk(link, KERN_INFO, "no reset method "
2449 "available, skipping reset\n");
2450 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2451 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002452 }
2453
Tejun Heo932648b2008-05-19 01:15:06 +09002454 /*
2455 * Post-reset processing
2456 */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002457 ata_link_for_each_dev(dev, link) {
2458 /* After the reset, the device state is PIO 0 and the
2459 * controller state is undefined. Reset also wakes up
2460 * drives from sleeping mode.
2461 */
2462 dev->pio_mode = XFER_PIO_0;
2463 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002464
Tejun Heo816ab892008-10-22 00:31:34 +09002465 if (!ata_phys_link_offline(ata_dev_phys_link(dev))) {
2466 /* apply class override */
2467 if (lflags & ATA_LFLAG_ASSUME_ATA)
2468 classes[dev->devno] = ATA_DEV_ATA;
2469 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2470 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2471 } else
2472 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002473 }
2474
Tejun Heo416dc9e2007-10-31 10:17:03 +09002475 /* record current link speed */
2476 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2477 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heob1c72912008-07-31 17:02:43 +09002478 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2479 slave->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002480
Tejun Heodc98c322008-05-19 01:15:07 +09002481 /* thaw the port */
2482 if (ata_is_host_link(link))
2483 ata_eh_thaw_port(ap);
2484
Tejun Heof0465192008-05-19 01:15:08 +09002485 /* postreset() should clear hardware SError. Although SError
2486 * is cleared during link resume, clearing SError here is
2487 * necessary as some PHYs raise hotplug events after SRST.
2488 * This introduces race condition where hotplug occurs between
2489 * reset and here. This race is mediated by cross checking
2490 * link onlineness and classification result later.
2491 */
Tejun Heob1c72912008-07-31 17:02:43 +09002492 if (postreset) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002493 postreset(link, classes);
Tejun Heob1c72912008-07-31 17:02:43 +09002494 if (slave)
2495 postreset(slave, classes);
2496 }
Tejun Heo20952b62006-05-31 18:27:23 +09002497
Tejun Heof0465192008-05-19 01:15:08 +09002498 /* clear cached SError */
2499 spin_lock_irqsave(link->ap->lock, flags);
2500 link->eh_info.serror = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002501 if (slave)
2502 slave->eh_info.serror = 0;
Tejun Heof0465192008-05-19 01:15:08 +09002503 spin_unlock_irqrestore(link->ap->lock, flags);
2504
2505 /* Make sure onlineness and classification result correspond.
2506 * Hotplug could have happened during reset and some
2507 * controllers fail to wait while a drive is spinning up after
2508 * being hotplugged causing misdetection. By cross checking
2509 * link onlineness and classification result, those conditions
2510 * can be reliably detected and retried.
2511 */
Tejun Heob1c72912008-07-31 17:02:43 +09002512 nr_unknown = 0;
Tejun Heof0465192008-05-19 01:15:08 +09002513 ata_link_for_each_dev(dev, link) {
2514 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
Tejun Heob1c72912008-07-31 17:02:43 +09002515 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
Tejun Heof0465192008-05-19 01:15:08 +09002516 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002517 if (ata_phys_link_online(ata_dev_phys_link(dev)))
2518 nr_unknown++;
2519 }
Tejun Heof0465192008-05-19 01:15:08 +09002520 }
2521
Tejun Heob1c72912008-07-31 17:02:43 +09002522 if (classify && nr_unknown) {
Tejun Heof0465192008-05-19 01:15:08 +09002523 if (try < max_tries) {
2524 ata_link_printk(link, KERN_WARNING, "link online but "
2525 "device misclassified, retrying\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002526 failed_link = link;
Tejun Heof0465192008-05-19 01:15:08 +09002527 rc = -EAGAIN;
2528 goto fail;
2529 }
2530 ata_link_printk(link, KERN_WARNING,
2531 "link online but device misclassified, "
2532 "device detection might fail\n");
2533 }
2534
Tejun Heo416dc9e2007-10-31 10:17:03 +09002535 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002536 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002537 if (slave)
2538 ata_eh_done(slave, NULL, ATA_EH_RESET);
Tejun Heo19b72322008-11-04 17:08:40 +09002539 ehc->last_reset = jiffies; /* update to completion time */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002540 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heoae791c02007-09-23 13:14:12 +09002541
Tejun Heo416dc9e2007-10-31 10:17:03 +09002542 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002543 out:
2544 /* clear hotplug flag */
2545 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heob1c72912008-07-31 17:02:43 +09002546 if (slave)
2547 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002548
2549 spin_lock_irqsave(ap->lock, flags);
2550 ap->pflags &= ~ATA_PFLAG_RESETTING;
2551 spin_unlock_irqrestore(ap->lock, flags);
2552
Tejun Heo022bdb02006-05-15 20:58:22 +09002553 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002554
2555 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002556 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2557 if (!ata_is_host_link(link) &&
2558 sata_scr_read(link, SCR_STATUS, &sstatus))
2559 rc = -ERESTART;
2560
Tejun Heo416dc9e2007-10-31 10:17:03 +09002561 if (rc == -ERESTART || try >= max_tries)
2562 goto out;
2563
2564 now = jiffies;
2565 if (time_before(now, deadline)) {
2566 unsigned long delta = deadline - now;
2567
Tejun Heob1c72912008-07-31 17:02:43 +09002568 ata_link_printk(failed_link, KERN_WARNING,
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002569 "reset failed (errno=%d), retrying in %u secs\n",
2570 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
Tejun Heo416dc9e2007-10-31 10:17:03 +09002571
2572 while (delta)
2573 delta = schedule_timeout_uninterruptible(delta);
2574 }
2575
Tejun Heob1c72912008-07-31 17:02:43 +09002576 if (try == max_tries - 1) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002577 sata_down_spd_limit(link);
Tejun Heob1c72912008-07-31 17:02:43 +09002578 if (slave)
2579 sata_down_spd_limit(slave);
2580 } else if (rc == -EPIPE)
2581 sata_down_spd_limit(failed_link);
2582
Tejun Heo416dc9e2007-10-31 10:17:03 +09002583 if (hardreset)
2584 reset = hardreset;
2585 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002586}
2587
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002588static inline void ata_eh_pull_park_action(struct ata_port *ap)
2589{
2590 struct ata_link *link;
2591 struct ata_device *dev;
2592 unsigned long flags;
2593
2594 /*
2595 * This function can be thought of as an extended version of
2596 * ata_eh_about_to_do() specially crafted to accommodate the
2597 * requirements of ATA_EH_PARK handling. Since the EH thread
2598 * does not leave the do {} while () loop in ata_eh_recover as
2599 * long as the timeout for a park request to *one* device on
2600 * the port has not expired, and since we still want to pick
2601 * up park requests to other devices on the same port or
2602 * timeout updates for the same device, we have to pull
2603 * ATA_EH_PARK actions from eh_info into eh_context.i
2604 * ourselves at the beginning of each pass over the loop.
2605 *
2606 * Additionally, all write accesses to &ap->park_req_pending
2607 * through INIT_COMPLETION() (see below) or complete_all()
2608 * (see ata_scsi_park_store()) are protected by the host lock.
2609 * As a result we have that park_req_pending.done is zero on
2610 * exit from this function, i.e. when ATA_EH_PARK actions for
2611 * *all* devices on port ap have been pulled into the
2612 * respective eh_context structs. If, and only if,
2613 * park_req_pending.done is non-zero by the time we reach
2614 * wait_for_completion_timeout(), another ATA_EH_PARK action
2615 * has been scheduled for at least one of the devices on port
2616 * ap and we have to cycle over the do {} while () loop in
2617 * ata_eh_recover() again.
2618 */
2619
2620 spin_lock_irqsave(ap->lock, flags);
2621 INIT_COMPLETION(ap->park_req_pending);
2622 ata_port_for_each_link(link, ap) {
2623 ata_link_for_each_dev(dev, link) {
2624 struct ata_eh_info *ehi = &link->eh_info;
2625
2626 link->eh_context.i.dev_action[dev->devno] |=
2627 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2628 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2629 }
2630 }
2631 spin_unlock_irqrestore(ap->lock, flags);
2632}
2633
2634static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2635{
2636 struct ata_eh_context *ehc = &dev->link->eh_context;
2637 struct ata_taskfile tf;
2638 unsigned int err_mask;
2639
2640 ata_tf_init(dev, &tf);
2641 if (park) {
2642 ehc->unloaded_mask |= 1 << dev->devno;
2643 tf.command = ATA_CMD_IDLEIMMEDIATE;
2644 tf.feature = 0x44;
2645 tf.lbal = 0x4c;
2646 tf.lbam = 0x4e;
2647 tf.lbah = 0x55;
2648 } else {
2649 ehc->unloaded_mask &= ~(1 << dev->devno);
2650 tf.command = ATA_CMD_CHK_POWER;
2651 }
2652
2653 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2654 tf.protocol |= ATA_PROT_NODATA;
2655 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2656 if (park && (err_mask || tf.lbal != 0xc4)) {
2657 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2658 ehc->unloaded_mask &= ~(1 << dev->devno);
2659 }
2660}
2661
Tejun Heo02607312007-08-06 18:36:23 +09002662static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09002663 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09002664{
Tejun Heo02607312007-08-06 18:36:23 +09002665 struct ata_port *ap = link->ap;
2666 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002667 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002668 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002669 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09002670 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002671
2672 DPRINTK("ENTER\n");
2673
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002674 /* For PATA drive side cable detection to work, IDENTIFY must
2675 * be done backwards such that PDIAG- is released by the slave
2676 * device before the master device is identified.
2677 */
Tejun Heo02607312007-08-06 18:36:23 +09002678 ata_link_for_each_dev_reverse(dev, link) {
Tejun Heof58229f2007-08-06 18:36:23 +09002679 unsigned int action = ata_eh_dev_action(dev);
2680 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09002681
Tejun Heobff04642006-11-10 18:08:10 +09002682 if (ehc->i.flags & ATA_EHI_DID_RESET)
2683 readid_flags |= ATA_READID_POSTRESET;
2684
Tejun Heo9666f402007-05-04 21:27:47 +02002685 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a32007-09-23 13:19:54 +09002686 WARN_ON(dev->class == ATA_DEV_PMP);
2687
Tejun Heob1c72912008-07-31 17:02:43 +09002688 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002689 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002690 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002691 }
2692
Tejun Heo02607312007-08-06 18:36:23 +09002693 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09002694 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2695 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09002696 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002697 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002698
Tejun Heo02607312007-08-06 18:36:23 +09002699 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09002700
Tejun Heobaa1e782006-11-01 18:39:27 +09002701 /* Configuration may have changed, reconfigure
2702 * transfer mode.
2703 */
2704 ehc->i.flags |= ATA_EHI_SETMODE;
2705
zhao, forrest3057ac32006-06-12 12:01:34 +08002706 /* schedule the scsi_rescan_device() here */
2707 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09002708 } else if (dev->class == ATA_DEV_UNKNOWN &&
2709 ehc->tries[dev->devno] &&
2710 ata_class_enabled(ehc->classes[dev->devno])) {
2711 dev->class = ehc->classes[dev->devno];
2712
Tejun Heo633273a32007-09-23 13:19:54 +09002713 if (dev->class == ATA_DEV_PMP)
2714 rc = sata_pmp_attach(dev);
2715 else
2716 rc = ata_dev_read_id(dev, &dev->class,
2717 readid_flags, dev->id);
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002718 switch (rc) {
2719 case 0:
Tejun Heof58229f2007-08-06 18:36:23 +09002720 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002721 break;
2722 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002723 /* IDENTIFY was issued to non-existent
2724 * device. No need to reset. Just
2725 * thaw and kill the device.
2726 */
2727 ata_eh_thaw_port(ap);
2728 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo084fe632006-05-31 18:28:03 +09002729 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002730 default:
2731 dev->class = ATA_DEV_UNKNOWN;
2732 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002733 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002734 }
2735 }
2736
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002737 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09002738 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2739 if (ap->ops->cable_detect)
2740 ap->cbl = ap->ops->cable_detect(ap);
2741 ata_force_cbl(ap);
2742 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002743
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002744 /* Configure new devices forward such that user doesn't see
2745 * device detection messages backwards.
2746 */
Tejun Heo02607312007-08-06 18:36:23 +09002747 ata_link_for_each_dev(dev, link) {
Tejun Heo633273a32007-09-23 13:19:54 +09002748 if (!(new_mask & (1 << dev->devno)) ||
2749 dev->class == ATA_DEV_PMP)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002750 continue;
2751
2752 ehc->i.flags |= ATA_EHI_PRINTINFO;
2753 rc = ata_dev_configure(dev);
2754 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2755 if (rc)
2756 goto err;
2757
2758 spin_lock_irqsave(ap->lock, flags);
2759 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2760 spin_unlock_irqrestore(ap->lock, flags);
2761
2762 /* new device discovered, configure xfermode */
2763 ehc->i.flags |= ATA_EHI_SETMODE;
2764 }
2765
2766 return 0;
2767
2768 err:
2769 *r_failed_dev = dev;
2770 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002771 return rc;
2772}
2773
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002774/**
2775 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2776 * @link: link on which timings will be programmed
2777 * @r_failed_dev: out paramter for failed device
2778 *
2779 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2780 * ata_set_mode() fails, pointer to the failing device is
2781 * returned in @r_failed_dev.
2782 *
2783 * LOCKING:
2784 * PCI/etc. bus probe sem.
2785 *
2786 * RETURNS:
2787 * 0 on success, negative errno otherwise
2788 */
2789int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2790{
2791 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09002792 struct ata_device *dev;
2793 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002794
Tejun Heo76326ac2007-11-27 19:28:59 +09002795 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2796 ata_link_for_each_dev(dev, link) {
Tejun Heo4a9c7b32008-10-27 19:59:23 +09002797 if (!ata_dev_enabled(dev))
2798 continue;
2799
Tejun Heo76326ac2007-11-27 19:28:59 +09002800 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2801 struct ata_ering_entry *ent;
2802
2803 ent = ata_ering_top(&dev->ering);
2804 if (ent)
2805 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2806 }
2807 }
2808
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002809 /* has private set_mode? */
2810 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09002811 rc = ap->ops->set_mode(link, r_failed_dev);
2812 else
2813 rc = ata_do_set_mode(link, r_failed_dev);
2814
2815 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2816 ata_link_for_each_dev(dev, link) {
2817 struct ata_eh_context *ehc = &link->eh_context;
2818 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2819 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2820
Tejun Heo4a9c7b32008-10-27 19:59:23 +09002821 if (!ata_dev_enabled(dev))
2822 continue;
2823
Tejun Heo00115e02007-11-27 19:28:58 +09002824 if (dev->xfer_mode != saved_xfer_mode ||
2825 ata_ncq_enabled(dev) != saved_ncq)
2826 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2827 }
2828
2829 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09002830}
2831
Tejun Heo11fc33d2008-08-30 14:20:01 +02002832/**
2833 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2834 * @dev: ATAPI device to clear UA for
2835 *
2836 * Resets and other operations can make an ATAPI device raise
2837 * UNIT ATTENTION which causes the next operation to fail. This
2838 * function clears UA.
2839 *
2840 * LOCKING:
2841 * EH context (may sleep).
2842 *
2843 * RETURNS:
2844 * 0 on success, -errno on failure.
2845 */
2846static int atapi_eh_clear_ua(struct ata_device *dev)
2847{
2848 int i;
2849
2850 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
2851 u8 sense_buffer[SCSI_SENSE_BUFFERSIZE];
2852 u8 sense_key = 0;
2853 unsigned int err_mask;
2854
2855 err_mask = atapi_eh_tur(dev, &sense_key);
2856 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
2857 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
2858 "failed (err_mask=0x%x)\n", err_mask);
2859 return -EIO;
2860 }
2861
2862 if (!err_mask || sense_key != UNIT_ATTENTION)
2863 return 0;
2864
2865 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
2866 if (err_mask) {
2867 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
2868 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
2869 return -EIO;
2870 }
2871 }
2872
2873 ata_dev_printk(dev, KERN_WARNING,
2874 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
2875
2876 return 0;
2877}
2878
Tejun Heo02607312007-08-06 18:36:23 +09002879static int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002880{
Tejun Heof58229f2007-08-06 18:36:23 +09002881 struct ata_device *dev;
2882 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002883
Tejun Heo02607312007-08-06 18:36:23 +09002884 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002885 if (ata_dev_enabled(dev))
Tejun Heo022bdb02006-05-15 20:58:22 +09002886 cnt++;
2887 return cnt;
2888}
2889
Tejun Heo02607312007-08-06 18:36:23 +09002890static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002891{
Tejun Heof58229f2007-08-06 18:36:23 +09002892 struct ata_device *dev;
2893 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002894
Tejun Heo02607312007-08-06 18:36:23 +09002895 ata_link_for_each_dev(dev, link)
Tejun Heof58229f2007-08-06 18:36:23 +09002896 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09002897 cnt++;
2898 return cnt;
2899}
2900
Tejun Heo02607312007-08-06 18:36:23 +09002901static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09002902{
Tejun Heo672b2d62008-01-24 00:05:14 +09002903 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09002904 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09002905 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09002906
Tejun Heof9df58c2007-09-23 13:14:13 +09002907 /* skip disabled links */
2908 if (link->flags & ATA_LFLAG_DISABLED)
2909 return 1;
2910
Tejun Heo672b2d62008-01-24 00:05:14 +09002911 /* thaw frozen port and recover failed devices */
2912 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2913 return 0;
2914
2915 /* reset at least once if reset is requested */
2916 if ((ehc->i.action & ATA_EH_RESET) &&
2917 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09002918 return 0;
2919
2920 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo02607312007-08-06 18:36:23 +09002921 ata_link_for_each_dev(dev, link) {
Tejun Heo084fe632006-05-31 18:28:03 +09002922 if (dev->class == ATA_DEV_UNKNOWN &&
2923 ehc->classes[dev->devno] != ATA_DEV_NONE)
2924 return 0;
2925 }
2926
2927 return 1;
2928}
2929
Tejun Heo02c05a22007-11-27 19:28:54 +09002930static int ata_eh_schedule_probe(struct ata_device *dev)
2931{
2932 struct ata_eh_context *ehc = &dev->link->eh_context;
2933
2934 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2935 (ehc->did_probe_mask & (1 << dev->devno)))
2936 return 0;
2937
2938 ata_eh_detach_dev(dev);
2939 ata_dev_init(dev);
2940 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09002941 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09002942 ehc->saved_xfer_mode[dev->devno] = 0;
2943 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09002944
2945 return 1;
2946}
2947
Tejun Heo9b1e2652007-08-06 18:36:24 +09002948static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09002949{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09002950 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09002951
2952 ehc->tries[dev->devno]--;
2953
2954 switch (err) {
2955 case -ENODEV:
2956 /* device missing or wrong IDENTIFY data, schedule probing */
2957 ehc->i.probe_mask |= (1 << dev->devno);
2958 case -EINVAL:
2959 /* give it just one more chance */
2960 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2961 case -EIO:
Tejun Heo4fb46152007-10-29 16:45:05 +09002962 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
Tejun Heofee7ca72007-07-01 19:05:58 +09002963 /* This is the last chance, better to slow
2964 * down than lose it.
2965 */
Tejun Heob1c72912008-07-31 17:02:43 +09002966 sata_down_spd_limit(ata_dev_phys_link(dev));
Tejun Heofee7ca72007-07-01 19:05:58 +09002967 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2968 }
2969 }
2970
2971 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2972 /* disable device if it has used up all its chances */
2973 ata_dev_disable(dev);
2974
2975 /* detach if offline */
Tejun Heob1c72912008-07-31 17:02:43 +09002976 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
Tejun Heofee7ca72007-07-01 19:05:58 +09002977 ata_eh_detach_dev(dev);
2978
Tejun Heo02c05a22007-11-27 19:28:54 +09002979 /* schedule probe if necessary */
Tejun Heo87fbc5a2008-05-20 02:17:54 +09002980 if (ata_eh_schedule_probe(dev)) {
Tejun Heofee7ca72007-07-01 19:05:58 +09002981 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo87fbc5a2008-05-20 02:17:54 +09002982 memset(ehc->cmd_timeout_idx[dev->devno], 0,
2983 sizeof(ehc->cmd_timeout_idx[dev->devno]));
2984 }
Tejun Heo9b1e2652007-08-06 18:36:24 +09002985
2986 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09002987 } else {
Tejun Heocf480622008-01-24 00:05:14 +09002988 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09002989 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09002990 }
2991}
2992
Tejun Heo022bdb02006-05-15 20:58:22 +09002993/**
2994 * ata_eh_recover - recover host port after error
2995 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09002996 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09002997 * @softreset: softreset method (can be NULL)
2998 * @hardreset: hardreset method (can be NULL)
2999 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003000 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09003001 *
3002 * This is the alpha and omega, eum and yang, heart and soul of
3003 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09003004 * recover each link and hotplug requests are recorded in the
3005 * link's eh_context. This function executes all the operations
3006 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09003007 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09003008 *
3009 * LOCKING:
3010 * Kernel thread context (may sleep).
3011 *
3012 * RETURNS:
3013 * 0 on success, -errno on failure.
3014 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003015int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3016 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3017 ata_postreset_fn_t postreset,
3018 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003019{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003020 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003021 struct ata_device *dev;
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003022 int nr_failed_devs;
Tejun Heodc98c322008-05-19 01:15:07 +09003023 int rc;
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003024 unsigned long flags, deadline;
Tejun Heo022bdb02006-05-15 20:58:22 +09003025
3026 DPRINTK("ENTER\n");
3027
3028 /* prep for recovery */
Tejun Heo9b1e2652007-08-06 18:36:24 +09003029 ata_port_for_each_link(link, ap) {
3030 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09003031
Tejun Heof9df58c2007-09-23 13:14:13 +09003032 /* re-enable link? */
3033 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3034 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3035 spin_lock_irqsave(ap->lock, flags);
3036 link->flags &= ~ATA_LFLAG_DISABLED;
3037 spin_unlock_irqrestore(ap->lock, flags);
3038 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3039 }
3040
Tejun Heo9b1e2652007-08-06 18:36:24 +09003041 ata_link_for_each_dev(dev, link) {
Tejun Heofd995f72007-09-23 13:14:12 +09003042 if (link->flags & ATA_LFLAG_NO_RETRY)
3043 ehc->tries[dev->devno] = 1;
3044 else
3045 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09003046
Tejun Heo9b1e2652007-08-06 18:36:24 +09003047 /* collect port action mask recorded in dev actions */
3048 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3049 ~ATA_EH_PERDEV_MASK;
3050 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09003051
Tejun Heo9b1e2652007-08-06 18:36:24 +09003052 /* process hotplug request */
3053 if (dev->flags & ATA_DFLAG_DETACH)
3054 ata_eh_detach_dev(dev);
3055
Tejun Heo02c05a22007-11-27 19:28:54 +09003056 /* schedule probe if necessary */
3057 if (!ata_dev_enabled(dev))
3058 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09003059 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003060 }
3061
3062 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09003063 rc = 0;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003064 nr_failed_devs = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003065
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003066 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09003067 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003068 goto out;
3069
Tejun Heo9b1e2652007-08-06 18:36:24 +09003070 /* prep for EH */
3071 ata_port_for_each_link(link, ap) {
3072 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003073
Tejun Heo9b1e2652007-08-06 18:36:24 +09003074 /* skip EH if possible. */
3075 if (ata_eh_skip_recovery(link))
3076 ehc->i.action = 0;
3077
Tejun Heo9b1e2652007-08-06 18:36:24 +09003078 ata_link_for_each_dev(dev, link)
3079 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3080 }
Tejun Heo084fe632006-05-31 18:28:03 +09003081
Tejun Heo022bdb02006-05-15 20:58:22 +09003082 /* reset */
Tejun Heodc98c322008-05-19 01:15:07 +09003083 ata_port_for_each_link(link, ap) {
3084 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003085
Tejun Heodc98c322008-05-19 01:15:07 +09003086 if (!(ehc->i.action & ATA_EH_RESET))
3087 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003088
Tejun Heodc98c322008-05-19 01:15:07 +09003089 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3090 prereset, softreset, hardreset, postreset);
3091 if (rc) {
3092 ata_link_printk(link, KERN_ERR,
3093 "reset failed, giving up\n");
3094 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09003095 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003096 }
3097
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003098 do {
3099 unsigned long now;
3100
3101 /*
3102 * clears ATA_EH_PARK in eh_info and resets
3103 * ap->park_req_pending
3104 */
3105 ata_eh_pull_park_action(ap);
3106
3107 deadline = jiffies;
3108 ata_port_for_each_link(link, ap) {
3109 ata_link_for_each_dev(dev, link) {
3110 struct ata_eh_context *ehc = &link->eh_context;
3111 unsigned long tmp;
3112
3113 if (dev->class != ATA_DEV_ATA)
3114 continue;
3115 if (!(ehc->i.dev_action[dev->devno] &
3116 ATA_EH_PARK))
3117 continue;
3118 tmp = dev->unpark_deadline;
3119 if (time_before(deadline, tmp))
3120 deadline = tmp;
3121 else if (time_before_eq(tmp, jiffies))
3122 continue;
3123 if (ehc->unloaded_mask & (1 << dev->devno))
3124 continue;
3125
3126 ata_eh_park_issue_cmd(dev, 1);
3127 }
3128 }
3129
3130 now = jiffies;
3131 if (time_before_eq(deadline, now))
3132 break;
3133
3134 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3135 deadline - now);
3136 } while (deadline);
3137 ata_port_for_each_link(link, ap) {
3138 ata_link_for_each_dev(dev, link) {
3139 if (!(link->eh_context.unloaded_mask &
3140 (1 << dev->devno)))
3141 continue;
3142
3143 ata_eh_park_issue_cmd(dev, 0);
3144 ata_eh_done(link, dev, ATA_EH_PARK);
3145 }
3146 }
3147
Tejun Heo9b1e2652007-08-06 18:36:24 +09003148 /* the rest */
3149 ata_port_for_each_link(link, ap) {
3150 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003151
Tejun Heo9b1e2652007-08-06 18:36:24 +09003152 /* revalidate existing devices and attach new ones */
3153 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09003154 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09003155 goto dev_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003156
Tejun Heo633273a32007-09-23 13:19:54 +09003157 /* if PMP got attached, return, pmp EH will take care of it */
3158 if (link->device->class == ATA_DEV_PMP) {
3159 ehc->i.action = 0;
3160 return 0;
3161 }
3162
Tejun Heo9b1e2652007-08-06 18:36:24 +09003163 /* configure transfer mode if necessary */
3164 if (ehc->i.flags & ATA_EHI_SETMODE) {
3165 rc = ata_set_mode(link, &dev);
3166 if (rc)
3167 goto dev_fail;
3168 ehc->i.flags &= ~ATA_EHI_SETMODE;
3169 }
3170
Tejun Heo11fc33d2008-08-30 14:20:01 +02003171 /* If reset has been issued, clear UA to avoid
3172 * disrupting the current users of the device.
3173 */
3174 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3175 ata_link_for_each_dev(dev, link) {
3176 if (dev->class != ATA_DEV_ATAPI)
3177 continue;
3178 rc = atapi_eh_clear_ua(dev);
3179 if (rc)
3180 goto dev_fail;
3181 }
3182 }
3183
3184 /* configure link power saving */
Tejun Heo3ec25eb2008-03-27 18:37:14 +09003185 if (ehc->i.action & ATA_EH_LPM)
Kristen Carlson Accardica773292007-10-25 00:58:59 -04003186 ata_link_for_each_dev(dev, link)
3187 ata_dev_enable_pm(dev, ap->pm_policy);
3188
Tejun Heo9b1e2652007-08-06 18:36:24 +09003189 /* this link is okay now */
3190 ehc->i.flags = 0;
3191 continue;
3192
Jeff Garzik2dcb4072007-10-19 06:42:56 -04003193dev_fail:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003194 nr_failed_devs++;
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003195 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo9b1e2652007-08-06 18:36:24 +09003196
Tejun Heob06ce3e2007-10-09 15:06:48 +09003197 if (ap->pflags & ATA_PFLAG_FROZEN) {
3198 /* PMP reset requires working host port.
3199 * Can't retry if it's frozen.
3200 */
Tejun Heo071f44b2008-04-07 22:47:22 +09003201 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09003202 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003203 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09003204 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003205 }
3206
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003207 if (nr_failed_devs)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003208 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003209
Tejun Heo022bdb02006-05-15 20:58:22 +09003210 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003211 if (rc && r_failed_link)
3212 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003213
3214 DPRINTK("EXIT, rc=%d\n", rc);
3215 return rc;
3216}
3217
3218/**
3219 * ata_eh_finish - finish up EH
3220 * @ap: host port to finish EH for
3221 *
3222 * Recovery is complete. Clean up EH states and retry or finish
3223 * failed qcs.
3224 *
3225 * LOCKING:
3226 * None.
3227 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003228void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09003229{
3230 int tag;
3231
3232 /* retry or finish qcs */
3233 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3234 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3235
3236 if (!(qc->flags & ATA_QCFLAG_FAILED))
3237 continue;
3238
3239 if (qc->err_mask) {
3240 /* FIXME: Once EH migration is complete,
3241 * generate sense data in this function,
3242 * considering both err_mask and tf.
3243 */
Tejun Heo03faab72008-03-27 19:14:24 +09003244 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09003245 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09003246 else
3247 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003248 } else {
3249 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3250 ata_eh_qc_complete(qc);
3251 } else {
3252 /* feed zero TF to sense generation */
3253 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3254 ata_eh_qc_retry(qc);
3255 }
3256 }
3257 }
Tejun Heoda917d62007-09-23 13:14:12 +09003258
3259 /* make sure nr_active_links is zero after EH */
3260 WARN_ON(ap->nr_active_links);
3261 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003262}
3263
3264/**
3265 * ata_do_eh - do standard error handling
3266 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09003267 *
Tejun Heof5914a42006-05-31 18:27:48 +09003268 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003269 * @softreset: softreset method (can be NULL)
3270 * @hardreset: hardreset method (can be NULL)
3271 * @postreset: postreset method (can be NULL)
3272 *
3273 * Perform standard error handling sequence.
3274 *
3275 * LOCKING:
3276 * Kernel thread context (may sleep).
3277 */
Tejun Heof5914a42006-05-31 18:27:48 +09003278void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3279 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3280 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09003281{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003282 struct ata_device *dev;
3283 int rc;
3284
3285 ata_eh_autopsy(ap);
3286 ata_eh_report(ap);
3287
3288 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3289 NULL);
3290 if (rc) {
3291 ata_link_for_each_dev(dev, &ap->link)
3292 ata_dev_disable(dev);
3293 }
3294
Tejun Heo022bdb02006-05-15 20:58:22 +09003295 ata_eh_finish(ap);
3296}
Tejun Heo500530f2006-07-03 16:07:27 +09003297
Tejun Heoa1efdab2008-03-25 12:22:50 +09003298/**
3299 * ata_std_error_handler - standard error handler
3300 * @ap: host port to handle error for
3301 *
3302 * Standard error handler
3303 *
3304 * LOCKING:
3305 * Kernel thread context (may sleep).
3306 */
3307void ata_std_error_handler(struct ata_port *ap)
3308{
3309 struct ata_port_operations *ops = ap->ops;
3310 ata_reset_fn_t hardreset = ops->hardreset;
3311
Tejun Heo57c9efd2008-04-07 22:47:19 +09003312 /* ignore built-in hardreset if SCR access is not available */
3313 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09003314 hardreset = NULL;
3315
3316 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3317}
3318
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003319#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09003320/**
3321 * ata_eh_handle_port_suspend - perform port suspend operation
3322 * @ap: port to suspend
3323 *
3324 * Suspend @ap.
3325 *
3326 * LOCKING:
3327 * Kernel thread context (may sleep).
3328 */
3329static void ata_eh_handle_port_suspend(struct ata_port *ap)
3330{
3331 unsigned long flags;
3332 int rc = 0;
3333
3334 /* are we suspending? */
3335 spin_lock_irqsave(ap->lock, flags);
3336 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3337 ap->pm_mesg.event == PM_EVENT_ON) {
3338 spin_unlock_irqrestore(ap->lock, flags);
3339 return;
3340 }
3341 spin_unlock_irqrestore(ap->lock, flags);
3342
3343 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3344
Tejun Heo64578a32007-05-15 03:28:16 +09003345 /* tell ACPI we're suspending */
3346 rc = ata_acpi_on_suspend(ap);
3347 if (rc)
3348 goto out;
3349
Tejun Heo500530f2006-07-03 16:07:27 +09003350 /* suspend */
3351 ata_eh_freeze_port(ap);
3352
3353 if (ap->ops->port_suspend)
3354 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3355
Shaohua Libd3adca2007-11-02 09:32:38 +08003356 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a32007-05-15 03:28:16 +09003357 out:
Tejun Heo500530f2006-07-03 16:07:27 +09003358 /* report result */
3359 spin_lock_irqsave(ap->lock, flags);
3360
3361 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3362 if (rc == 0)
3363 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a32007-05-15 03:28:16 +09003364 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09003365 ata_port_schedule_eh(ap);
3366
3367 if (ap->pm_result) {
3368 *ap->pm_result = rc;
3369 ap->pm_result = NULL;
3370 }
3371
3372 spin_unlock_irqrestore(ap->lock, flags);
3373
3374 return;
3375}
3376
3377/**
3378 * ata_eh_handle_port_resume - perform port resume operation
3379 * @ap: port to resume
3380 *
3381 * Resume @ap.
3382 *
Tejun Heo500530f2006-07-03 16:07:27 +09003383 * LOCKING:
3384 * Kernel thread context (may sleep).
3385 */
3386static void ata_eh_handle_port_resume(struct ata_port *ap)
3387{
Tejun Heo500530f2006-07-03 16:07:27 +09003388 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02003389 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09003390
3391 /* are we resuming? */
3392 spin_lock_irqsave(ap->lock, flags);
3393 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3394 ap->pm_mesg.event != PM_EVENT_ON) {
3395 spin_unlock_irqrestore(ap->lock, flags);
3396 return;
3397 }
3398 spin_unlock_irqrestore(ap->lock, flags);
3399
Tejun Heo9666f402007-05-04 21:27:47 +02003400 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09003401
Shaohua Libd3adca2007-11-02 09:32:38 +08003402 ata_acpi_set_state(ap, PMSG_ON);
3403
Tejun Heo500530f2006-07-03 16:07:27 +09003404 if (ap->ops->port_resume)
3405 rc = ap->ops->port_resume(ap);
3406
Tejun Heo67465442007-05-15 03:28:16 +09003407 /* tell ACPI that we're resuming */
3408 ata_acpi_on_resume(ap);
3409
Tejun Heo9666f402007-05-04 21:27:47 +02003410 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09003411 spin_lock_irqsave(ap->lock, flags);
3412 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3413 if (ap->pm_result) {
3414 *ap->pm_result = rc;
3415 ap->pm_result = NULL;
3416 }
3417 spin_unlock_irqrestore(ap->lock, flags);
3418}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003419#endif /* CONFIG_PM */