blob: f77a67303f8b3d0d724e4fccd3510a269a72b0c8 [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>
Robert Hancock65211482009-07-14 20:43:39 -060043#include <scsi/scsi_dbg.h>
Jeff Garzikc6fd2802006-08-10 07:31:37 -040044#include "../scsi/scsi_transport_api.h"
Tejun Heoece1d632006-04-02 18:51:53 +090045
46#include <linux/libata.h>
47
48#include "libata.h"
49
Tejun Heo7d47e8d2007-02-02 16:22:31 +090050enum {
Tejun Heo3884f7b2007-11-27 19:28:56 +090051 /* speed down verdicts */
Tejun Heo7d47e8d2007-02-02 16:22:31 +090052 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
Tejun Heo76326ac2007-11-27 19:28:59 +090055 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
Tejun Heo3884f7b2007-11-27 19:28:56 +090056
57 /* error flags */
58 ATA_EFLAG_IS_IO = (1 << 0),
Tejun Heo76326ac2007-11-27 19:28:59 +090059 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
Tejun Heo3884f7b2007-11-27 19:28:56 +090060
61 /* error categories */
62 ATA_ECAT_NONE = 0,
63 ATA_ECAT_ATA_BUS = 1,
64 ATA_ECAT_TOUT_HSM = 2,
65 ATA_ECAT_UNK_DEV = 3,
Tejun Heo75f9caf2008-01-03 01:21:14 +090066 ATA_ECAT_DUBIOUS_NONE = 4,
67 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
68 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
69 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
70 ATA_ECAT_NR = 8,
Tejun Heo7d47e8d2007-02-02 16:22:31 +090071
Tejun Heo87fbc5a2008-05-20 02:17:54 +090072 ATA_EH_CMD_DFL_TIMEOUT = 5000,
73
Tejun Heo0a2c0f52008-05-20 02:17:52 +090074 /* always put at least this amount of time between resets */
75 ATA_EH_RESET_COOL_DOWN = 5000,
76
Tejun Heo341c2c92008-05-20 02:17:51 +090077 /* Waiting in ->prereset can never be reliable. It's
78 * sometimes nice to wait there but it can't be depended upon;
79 * otherwise, we wouldn't be resetting. Just give it enough
80 * time for most drives to spin up.
81 */
82 ATA_EH_PRERESET_TIMEOUT = 10000,
83 ATA_EH_FASTDRAIN_INTERVAL = 3000,
Tejun Heo11fc33d2008-08-30 14:20:01 +020084
85 ATA_EH_UA_TRIES = 5,
Tejun Heoc2c7a892009-01-29 20:31:34 +090086
87 /* probe speed down parameters, see ata_eh_schedule_probe() */
88 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
89 ATA_EH_PROBE_TRIALS = 2,
Tejun Heo31daabd2007-02-02 16:50:52 +090090};
91
92/* The following table determines how we sequence resets. Each entry
93 * represents timeout for that try. The first try can be soft or
94 * hardreset. All others are hardreset if available. In most cases
95 * the first reset w/ 10sec timeout should succeed. Following entries
96 * are mostly for error handling, hotplug and retarded devices.
97 */
98static const unsigned long ata_eh_reset_timeouts[] = {
Tejun Heo341c2c92008-05-20 02:17:51 +090099 10000, /* most drives spin up by 10sec */
100 10000, /* > 99% working drives spin up before 20sec */
101 35000, /* give > 30 secs of idleness for retarded devices */
102 5000, /* and sweet one last chance */
Tejun Heod8af0eb2008-05-20 02:17:53 +0900103 ULONG_MAX, /* > 1 min has elapsed, give up */
Tejun Heo31daabd2007-02-02 16:50:52 +0900104};
105
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900106static const unsigned long ata_eh_identify_timeouts[] = {
107 5000, /* covers > 99% of successes and not too boring on failures */
108 10000, /* combined time till here is enough even for media access */
109 30000, /* for true idiots */
110 ULONG_MAX,
111};
112
Tejun Heo6013efd2009-11-19 15:36:45 +0900113static const unsigned long ata_eh_flush_timeouts[] = {
114 15000, /* be generous with flush */
115 15000, /* ditto */
116 30000, /* and even more generous */
117 ULONG_MAX,
118};
119
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900120static const unsigned long ata_eh_other_timeouts[] = {
121 5000, /* same rationale as identify timeout */
122 10000, /* ditto */
123 /* but no merciful 30sec for other commands, it just isn't worth it */
124 ULONG_MAX,
125};
126
127struct ata_eh_cmd_timeout_ent {
128 const u8 *commands;
129 const unsigned long *timeouts;
130};
131
132/* The following table determines timeouts to use for EH internal
133 * commands. Each table entry is a command class and matches the
134 * commands the entry applies to and the timeout table to use.
135 *
136 * On the retry after a command timed out, the next timeout value from
137 * the table is used. If the table doesn't contain further entries,
138 * the last value is used.
139 *
140 * ehc->cmd_timeout_idx keeps track of which timeout to use per
141 * command class, so if SET_FEATURES times out on the first try, the
142 * next try will use the second timeout value only for that class.
143 */
144#define CMDS(cmds...) (const u8 []){ cmds, 0 }
145static const struct ata_eh_cmd_timeout_ent
146ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
147 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
148 .timeouts = ata_eh_identify_timeouts, },
149 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
150 .timeouts = ata_eh_other_timeouts, },
151 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
152 .timeouts = ata_eh_other_timeouts, },
153 { .commands = CMDS(ATA_CMD_SET_FEATURES),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
156 .timeouts = ata_eh_other_timeouts, },
Tejun Heo6013efd2009-11-19 15:36:45 +0900157 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
158 .timeouts = ata_eh_flush_timeouts },
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900159};
160#undef CMDS
161
Tejun Heoad9e2762006-05-15 20:58:12 +0900162static void __ata_port_freeze(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900163#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +0900164static void ata_eh_handle_port_suspend(struct ata_port *ap);
165static void ata_eh_handle_port_resume(struct ata_port *ap);
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900166#else /* CONFIG_PM */
167static void ata_eh_handle_port_suspend(struct ata_port *ap)
168{ }
169
170static void ata_eh_handle_port_resume(struct ata_port *ap)
171{ }
Tejun Heo6ffa01d2007-03-02 17:32:47 +0900172#endif /* CONFIG_PM */
Tejun Heoad9e2762006-05-15 20:58:12 +0900173
Tejun Heob64bbc32007-07-16 14:29:39 +0900174static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
175 va_list args)
176{
177 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
178 ATA_EH_DESC_LEN - ehi->desc_len,
179 fmt, args);
180}
181
182/**
183 * __ata_ehi_push_desc - push error description without adding separator
184 * @ehi: target EHI
185 * @fmt: printf format string
186 *
187 * Format string according to @fmt and append it to @ehi->desc.
188 *
189 * LOCKING:
190 * spin_lock_irqsave(host lock)
191 */
192void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
193{
194 va_list args;
195
196 va_start(args, fmt);
197 __ata_ehi_pushv_desc(ehi, fmt, args);
198 va_end(args);
199}
200
201/**
202 * ata_ehi_push_desc - push error description with separator
203 * @ehi: target EHI
204 * @fmt: printf format string
205 *
206 * Format string according to @fmt and append it to @ehi->desc.
207 * If @ehi->desc is not empty, ", " is added in-between.
208 *
209 * LOCKING:
210 * spin_lock_irqsave(host lock)
211 */
212void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
213{
214 va_list args;
215
216 if (ehi->desc_len)
217 __ata_ehi_push_desc(ehi, ", ");
218
219 va_start(args, fmt);
220 __ata_ehi_pushv_desc(ehi, fmt, args);
221 va_end(args);
222}
223
224/**
225 * ata_ehi_clear_desc - clean error description
226 * @ehi: target EHI
227 *
228 * Clear @ehi->desc.
229 *
230 * LOCKING:
231 * spin_lock_irqsave(host lock)
232 */
233void ata_ehi_clear_desc(struct ata_eh_info *ehi)
234{
235 ehi->desc[0] = '\0';
236 ehi->desc_len = 0;
237}
238
Tejun Heocbcdd872007-08-18 13:14:55 +0900239/**
240 * ata_port_desc - append port description
241 * @ap: target ATA port
242 * @fmt: printf format string
243 *
244 * Format string according to @fmt and append it to port
245 * description. If port description is not empty, " " is added
246 * in-between. This function is to be used while initializing
247 * ata_host. The description is printed on host registration.
248 *
249 * LOCKING:
250 * None.
251 */
252void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
253{
254 va_list args;
255
256 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
257
258 if (ap->link.eh_info.desc_len)
259 __ata_ehi_push_desc(&ap->link.eh_info, " ");
260
261 va_start(args, fmt);
262 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
263 va_end(args);
264}
265
266#ifdef CONFIG_PCI
267
268/**
269 * ata_port_pbar_desc - append PCI BAR description
270 * @ap: target ATA port
271 * @bar: target PCI BAR
272 * @offset: offset into PCI BAR
273 * @name: name of the area
274 *
275 * If @offset is negative, this function formats a string which
276 * contains the name, address, size and type of the BAR and
277 * appends it to the port description. If @offset is zero or
278 * positive, only name and offsetted address is appended.
279 *
280 * LOCKING:
281 * None.
282 */
283void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
284 const char *name)
285{
286 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
287 char *type = "";
288 unsigned long long start, len;
289
290 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
291 type = "m";
292 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
293 type = "i";
294
295 start = (unsigned long long)pci_resource_start(pdev, bar);
296 len = (unsigned long long)pci_resource_len(pdev, bar);
297
298 if (offset < 0)
299 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
300 else
Andrew Mortone6a73ab2007-12-13 16:01:37 -0800301 ata_port_desc(ap, "%s 0x%llx", name,
302 start + (unsigned long long)offset);
Tejun Heocbcdd872007-08-18 13:14:55 +0900303}
304
305#endif /* CONFIG_PCI */
306
Tejun Heo87fbc5a2008-05-20 02:17:54 +0900307static int ata_lookup_timeout_table(u8 cmd)
308{
309 int i;
310
311 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
312 const u8 *cur;
313
314 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
315 if (*cur == cmd)
316 return i;
317 }
318
319 return -1;
320}
321
322/**
323 * ata_internal_cmd_timeout - determine timeout for an internal command
324 * @dev: target device
325 * @cmd: internal command to be issued
326 *
327 * Determine timeout for internal command @cmd for @dev.
328 *
329 * LOCKING:
330 * EH context.
331 *
332 * RETURNS:
333 * Determined timeout.
334 */
335unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
336{
337 struct ata_eh_context *ehc = &dev->link->eh_context;
338 int ent = ata_lookup_timeout_table(cmd);
339 int idx;
340
341 if (ent < 0)
342 return ATA_EH_CMD_DFL_TIMEOUT;
343
344 idx = ehc->cmd_timeout_idx[dev->devno][ent];
345 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
346}
347
348/**
349 * ata_internal_cmd_timed_out - notification for internal command timeout
350 * @dev: target device
351 * @cmd: internal command which timed out
352 *
353 * Notify EH that internal command @cmd for @dev timed out. This
354 * function should be called only for commands whose timeouts are
355 * determined using ata_internal_cmd_timeout().
356 *
357 * LOCKING:
358 * EH context.
359 */
360void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
361{
362 struct ata_eh_context *ehc = &dev->link->eh_context;
363 int ent = ata_lookup_timeout_table(cmd);
364 int idx;
365
366 if (ent < 0)
367 return;
368
369 idx = ehc->cmd_timeout_idx[dev->devno][ent];
370 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
371 ehc->cmd_timeout_idx[dev->devno][ent]++;
372}
373
Tejun Heo3884f7b2007-11-27 19:28:56 +0900374static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
Tejun Heo0c247c52006-05-15 20:58:19 +0900375 unsigned int err_mask)
376{
377 struct ata_ering_entry *ent;
378
379 WARN_ON(!err_mask);
380
381 ering->cursor++;
382 ering->cursor %= ATA_ERING_SIZE;
383
384 ent = &ering->ring[ering->cursor];
Tejun Heo3884f7b2007-11-27 19:28:56 +0900385 ent->eflags = eflags;
Tejun Heo0c247c52006-05-15 20:58:19 +0900386 ent->err_mask = err_mask;
387 ent->timestamp = get_jiffies_64();
388}
389
Tejun Heo76326ac2007-11-27 19:28:59 +0900390static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
391{
392 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
393
394 if (ent->err_mask)
395 return ent;
396 return NULL;
397}
398
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900399static void ata_ering_clear(struct ata_ering *ering)
Tejun Heo0c247c52006-05-15 20:58:19 +0900400{
Tejun Heo7d47e8d2007-02-02 16:22:31 +0900401 memset(ering, 0, sizeof(*ering));
Tejun Heo0c247c52006-05-15 20:58:19 +0900402}
403
404static int ata_ering_map(struct ata_ering *ering,
405 int (*map_fn)(struct ata_ering_entry *, void *),
406 void *arg)
407{
408 int idx, rc = 0;
409 struct ata_ering_entry *ent;
410
411 idx = ering->cursor;
412 do {
413 ent = &ering->ring[idx];
414 if (!ent->err_mask)
415 break;
416 rc = map_fn(ent, arg);
417 if (rc)
418 break;
419 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
420 } while (idx != ering->cursor);
421
422 return rc;
423}
424
Tejun Heo64f65ca2006-06-24 20:30:18 +0900425static unsigned int ata_eh_dev_action(struct ata_device *dev)
426{
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900427 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heo64f65ca2006-06-24 20:30:18 +0900428
429 return ehc->i.action | ehc->i.dev_action[dev->devno];
430}
431
Tejun Heof58229f2007-08-06 18:36:23 +0900432static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
Tejun Heoaf181c22006-06-24 20:30:18 +0900433 struct ata_eh_info *ehi, unsigned int action)
434{
Tejun Heof58229f2007-08-06 18:36:23 +0900435 struct ata_device *tdev;
Tejun Heoaf181c22006-06-24 20:30:18 +0900436
437 if (!dev) {
438 ehi->action &= ~action;
Tejun Heo1eca4362008-11-03 20:03:17 +0900439 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900440 ehi->dev_action[tdev->devno] &= ~action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900441 } else {
442 /* doesn't make sense for port-wide EH actions */
443 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
444
445 /* break ehi->action into ehi->dev_action */
446 if (ehi->action & action) {
Tejun Heo1eca4362008-11-03 20:03:17 +0900447 ata_for_each_dev(tdev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +0900448 ehi->dev_action[tdev->devno] |=
449 ehi->action & action;
Tejun Heoaf181c22006-06-24 20:30:18 +0900450 ehi->action &= ~action;
451 }
452
453 /* turn off the specified per-dev action */
454 ehi->dev_action[dev->devno] &= ~action;
455 }
456}
457
Tejun Heoece1d632006-04-02 18:51:53 +0900458/**
459 * ata_scsi_timed_out - SCSI layer time out callback
460 * @cmd: timed out SCSI command
461 *
462 * Handles SCSI layer timeout. We race with normal completion of
463 * the qc for @cmd. If the qc is already gone, we lose and let
464 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
465 * timed out and EH should be invoked. Prevent ata_qc_complete()
466 * from finishing it by setting EH_SCHEDULED and return
467 * EH_NOT_HANDLED.
468 *
Tejun Heoad9e2762006-05-15 20:58:12 +0900469 * TODO: kill this function once old EH is gone.
470 *
Tejun Heoece1d632006-04-02 18:51:53 +0900471 * LOCKING:
472 * Called from timer context
473 *
474 * RETURNS:
475 * EH_HANDLED or EH_NOT_HANDLED
476 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700477enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
Tejun Heoece1d632006-04-02 18:51:53 +0900478{
479 struct Scsi_Host *host = cmd->device->host;
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400480 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoece1d632006-04-02 18:51:53 +0900481 unsigned long flags;
482 struct ata_queued_cmd *qc;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700483 enum blk_eh_timer_return ret;
Tejun Heoece1d632006-04-02 18:51:53 +0900484
485 DPRINTK("ENTER\n");
486
Tejun Heoad9e2762006-05-15 20:58:12 +0900487 if (ap->ops->error_handler) {
Jens Axboe242f9dc2008-09-14 05:55:09 -0700488 ret = BLK_EH_NOT_HANDLED;
Tejun Heoad9e2762006-05-15 20:58:12 +0900489 goto out;
490 }
491
Jens Axboe242f9dc2008-09-14 05:55:09 -0700492 ret = BLK_EH_HANDLED;
Jeff Garzikba6a1302006-06-22 23:46:10 -0400493 spin_lock_irqsave(ap->lock, flags);
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900494 qc = ata_qc_from_tag(ap, ap->link.active_tag);
Tejun Heoece1d632006-04-02 18:51:53 +0900495 if (qc) {
496 WARN_ON(qc->scsicmd != cmd);
497 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
498 qc->err_mask |= AC_ERR_TIMEOUT;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700499 ret = BLK_EH_NOT_HANDLED;
Tejun Heoece1d632006-04-02 18:51:53 +0900500 }
Jeff Garzikba6a1302006-06-22 23:46:10 -0400501 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +0900502
Tejun Heoad9e2762006-05-15 20:58:12 +0900503 out:
Tejun Heoece1d632006-04-02 18:51:53 +0900504 DPRINTK("EXIT, ret=%d\n", ret);
505 return ret;
506}
507
Tejun Heoece180d2008-11-03 20:04:37 +0900508static void ata_eh_unload(struct ata_port *ap)
509{
510 struct ata_link *link;
511 struct ata_device *dev;
512 unsigned long flags;
513
514 /* Restore SControl IPM and SPD for the next driver and
515 * disable attached devices.
516 */
517 ata_for_each_link(link, ap, PMP_FIRST) {
518 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
519 ata_for_each_dev(dev, link, ALL)
520 ata_dev_disable(dev);
521 }
522
523 /* freeze and set UNLOADED */
524 spin_lock_irqsave(ap->lock, flags);
525
526 ata_port_freeze(ap); /* won't be thawed */
527 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
528 ap->pflags |= ATA_PFLAG_UNLOADED;
529
530 spin_unlock_irqrestore(ap->lock, flags);
531}
532
Tejun Heoece1d632006-04-02 18:51:53 +0900533/**
534 * ata_scsi_error - SCSI layer error handler callback
535 * @host: SCSI host on which error occurred
536 *
537 * Handles SCSI-layer-thrown error events.
538 *
539 * LOCKING:
540 * Inherited from SCSI layer (none, can sleep)
541 *
542 * RETURNS:
543 * Zero.
544 */
Jeff Garzik381544b2006-04-11 13:04:39 -0400545void ata_scsi_error(struct Scsi_Host *host)
Tejun Heoece1d632006-04-02 18:51:53 +0900546{
Jeff Garzik35bb94b2006-04-11 13:12:34 -0400547 struct ata_port *ap = ata_shost_to_port(host);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900548 int i;
Tejun Heoad9e2762006-05-15 20:58:12 +0900549 unsigned long flags;
Tejun Heoece1d632006-04-02 18:51:53 +0900550
551 DPRINTK("ENTER\n");
552
Tejun Heoc4291372010-05-10 21:41:38 +0200553 /* make sure sff pio task is not running */
554 ata_sff_flush_pio_task(ap);
Tejun Heoece1d632006-04-02 18:51:53 +0900555
Jeff Garzikcca39742006-08-24 03:19:22 -0400556 /* synchronize with host lock and sort out timeouts */
Tejun Heoece1d632006-04-02 18:51:53 +0900557
Tejun Heoad9e2762006-05-15 20:58:12 +0900558 /* For new EH, all qcs are finished in one of three ways -
559 * normal completion, error completion, and SCSI timeout.
Alan Coxc96f1732009-03-24 10:23:46 +0000560 * Both completions can race against SCSI timeout. When normal
Tejun Heoad9e2762006-05-15 20:58:12 +0900561 * completion wins, the qc never reaches EH. When error
562 * completion wins, the qc has ATA_QCFLAG_FAILED set.
563 *
564 * When SCSI timeout wins, things are a bit more complex.
565 * Normal or error completion can occur after the timeout but
566 * before this point. In such cases, both types of
567 * completions are honored. A scmd is determined to have
568 * timed out iff its associated qc is active and not failed.
569 */
570 if (ap->ops->error_handler) {
571 struct scsi_cmnd *scmd, *tmp;
572 int nr_timedout = 0;
Tejun Heoece1d632006-04-02 18:51:53 +0900573
Tejun Heoe30349d2006-07-03 03:02:15 +0900574 spin_lock_irqsave(ap->lock, flags);
Alan Coxc96f1732009-03-24 10:23:46 +0000575
576 /* This must occur under the ap->lock as we don't want
577 a polled recovery to race the real interrupt handler
578
579 The lost_interrupt handler checks for any completed but
580 non-notified command and completes much like an IRQ handler.
581
582 We then fall into the error recovery code which will treat
583 this as if normal completion won the race */
Tejun Heoad9e2762006-05-15 20:58:12 +0900584
Alan Coxc96f1732009-03-24 10:23:46 +0000585 if (ap->ops->lost_interrupt)
586 ap->ops->lost_interrupt(ap);
587
Tejun Heoad9e2762006-05-15 20:58:12 +0900588 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
589 struct ata_queued_cmd *qc;
590
591 for (i = 0; i < ATA_MAX_QUEUE; i++) {
592 qc = __ata_qc_from_tag(ap, i);
593 if (qc->flags & ATA_QCFLAG_ACTIVE &&
594 qc->scsicmd == scmd)
595 break;
596 }
597
598 if (i < ATA_MAX_QUEUE) {
599 /* the scmd has an associated qc */
600 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
601 /* which hasn't failed yet, timeout */
602 qc->err_mask |= AC_ERR_TIMEOUT;
603 qc->flags |= ATA_QCFLAG_FAILED;
604 nr_timedout++;
605 }
606 } else {
607 /* Normal completion occurred after
608 * SCSI timeout but before this point.
609 * Successfully complete it.
610 */
611 scmd->retries = scmd->allowed;
612 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
613 }
614 }
615
616 /* If we have timed out qcs. They belong to EH from
617 * this point but the state of the controller is
618 * unknown. Freeze the port to make sure the IRQ
619 * handler doesn't diddle with those qcs. This must
620 * be done atomically w.r.t. setting QCFLAG_FAILED.
621 */
622 if (nr_timedout)
623 __ata_port_freeze(ap);
624
Tejun Heoe30349d2006-07-03 03:02:15 +0900625 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoa1e10f72007-08-18 13:28:49 +0900626
627 /* initialize eh_tries */
628 ap->eh_tries = ATA_EH_MAX_TRIES;
Tejun Heoad9e2762006-05-15 20:58:12 +0900629 } else
Tejun Heoe30349d2006-07-03 03:02:15 +0900630 spin_unlock_wait(ap->lock);
Alan Coxc96f1732009-03-24 10:23:46 +0000631
632 /* If we timed raced normal completion and there is nothing to
633 recover nr_timedout == 0 why exactly are we doing error recovery ? */
Tejun Heoad9e2762006-05-15 20:58:12 +0900634
635 repeat:
636 /* invoke error handler */
637 if (ap->ops->error_handler) {
Tejun Heocf1b86c2007-08-06 18:36:23 +0900638 struct ata_link *link;
639
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900640 /* kill fast drain timer */
641 del_timer_sync(&ap->fastdrain_timer);
642
Tejun Heo500530f2006-07-03 16:07:27 +0900643 /* process port resume request */
644 ata_eh_handle_port_resume(ap);
645
Tejun Heof3e81b12006-05-15 20:58:21 +0900646 /* fetch & clear EH info */
Tejun Heoe30349d2006-07-03 03:02:15 +0900647 spin_lock_irqsave(ap->lock, flags);
Tejun Heof3e81b12006-05-15 20:58:21 +0900648
Tejun Heo1eca4362008-11-03 20:03:17 +0900649 ata_for_each_link(link, ap, HOST_FIRST) {
Tejun Heo00115e02007-11-27 19:28:58 +0900650 struct ata_eh_context *ehc = &link->eh_context;
651 struct ata_device *dev;
652
Tejun Heocf1b86c2007-08-06 18:36:23 +0900653 memset(&link->eh_context, 0, sizeof(link->eh_context));
654 link->eh_context.i = link->eh_info;
655 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heo00115e02007-11-27 19:28:58 +0900656
Tejun Heo1eca4362008-11-03 20:03:17 +0900657 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +0900658 int devno = dev->devno;
659
660 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
661 if (ata_ncq_enabled(dev))
662 ehc->saved_ncq_enabled |= 1 << devno;
663 }
Tejun Heocf1b86c2007-08-06 18:36:23 +0900664 }
Tejun Heof3e81b12006-05-15 20:58:21 +0900665
Tejun Heob51e9e52006-06-29 01:29:30 +0900666 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
667 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoda917d62007-09-23 13:14:12 +0900668 ap->excl_link = NULL; /* don't maintain exclusion over EH */
Tejun Heof3e81b12006-05-15 20:58:21 +0900669
Tejun Heoe30349d2006-07-03 03:02:15 +0900670 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900671
Tejun Heo500530f2006-07-03 16:07:27 +0900672 /* invoke EH, skip if unloading or suspended */
673 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
Tejun Heo720ba122006-05-31 18:28:13 +0900674 ap->ops->error_handler(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900675 else {
676 /* if unloading, commence suicide */
677 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
678 !(ap->pflags & ATA_PFLAG_UNLOADED))
679 ata_eh_unload(ap);
Tejun Heo720ba122006-05-31 18:28:13 +0900680 ata_eh_finish(ap);
Tejun Heoece180d2008-11-03 20:04:37 +0900681 }
Tejun Heoad9e2762006-05-15 20:58:12 +0900682
Tejun Heo500530f2006-07-03 16:07:27 +0900683 /* process port suspend request */
684 ata_eh_handle_port_suspend(ap);
685
Tejun Heoad9e2762006-05-15 20:58:12 +0900686 /* Exception might have happend after ->error_handler
687 * recovered the port but before this point. Repeat
688 * EH in such case.
689 */
Tejun Heoe30349d2006-07-03 03:02:15 +0900690 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900691
Tejun Heob51e9e52006-06-29 01:29:30 +0900692 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
Tejun Heoa1e10f72007-08-18 13:28:49 +0900693 if (--ap->eh_tries) {
Tejun Heoe30349d2006-07-03 03:02:15 +0900694 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900695 goto repeat;
696 }
697 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
Tejun Heoa1e10f72007-08-18 13:28:49 +0900698 "tries, giving up\n", ATA_EH_MAX_TRIES);
Tejun Heo914616a2007-06-25 21:47:11 +0900699 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heoad9e2762006-05-15 20:58:12 +0900700 }
701
Tejun Heof3e81b12006-05-15 20:58:21 +0900702 /* this run is complete, make sure EH info is clear */
Tejun Heo1eca4362008-11-03 20:03:17 +0900703 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heocf1b86c2007-08-06 18:36:23 +0900704 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heof3e81b12006-05-15 20:58:21 +0900705
Tejun Heoe30349d2006-07-03 03:02:15 +0900706 /* Clear host_eh_scheduled while holding ap->lock such
Tejun Heoad9e2762006-05-15 20:58:12 +0900707 * that if exception occurs after this point but
708 * before EH completion, SCSI midlayer will
709 * re-initiate EH.
710 */
711 host->host_eh_scheduled = 0;
712
Tejun Heoe30349d2006-07-03 03:02:15 +0900713 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900714 } else {
Tejun Heo9af5c9c2007-08-06 18:36:22 +0900715 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
Tejun Heoad9e2762006-05-15 20:58:12 +0900716 ap->ops->eng_timeout(ap);
717 }
718
719 /* finish or retry handled scmd's and clean up */
Tejun Heoece1d632006-04-02 18:51:53 +0900720 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
721
722 scsi_eh_flush_done_q(&ap->eh_done_q);
723
Tejun Heoad9e2762006-05-15 20:58:12 +0900724 /* clean up */
Tejun Heoe30349d2006-07-03 03:02:15 +0900725 spin_lock_irqsave(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900726
Tejun Heo1cdaf532006-07-03 16:07:26 +0900727 if (ap->pflags & ATA_PFLAG_LOADING)
Tejun Heob51e9e52006-06-29 01:29:30 +0900728 ap->pflags &= ~ATA_PFLAG_LOADING;
Tejun Heo1cdaf532006-07-03 16:07:26 +0900729 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
David Howells52bad642006-11-22 14:54:01 +0000730 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
Tejun Heo1cdaf532006-07-03 16:07:26 +0900731
732 if (ap->pflags & ATA_PFLAG_RECOVERED)
733 ata_port_printk(ap, KERN_INFO, "EH complete\n");
Tejun Heo580b21022006-05-31 18:28:05 +0900734
Tejun Heob51e9e52006-06-29 01:29:30 +0900735 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
Tejun Heoad9e2762006-05-15 20:58:12 +0900736
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900737 /* tell wait_eh that we're done */
Tejun Heob51e9e52006-06-29 01:29:30 +0900738 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900739 wake_up_all(&ap->eh_wait_q);
740
Tejun Heoe30349d2006-07-03 03:02:15 +0900741 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoad9e2762006-05-15 20:58:12 +0900742
Tejun Heoece1d632006-04-02 18:51:53 +0900743 DPRINTK("EXIT\n");
Tejun Heoece1d632006-04-02 18:51:53 +0900744}
745
746/**
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900747 * ata_port_wait_eh - Wait for the currently pending EH to complete
748 * @ap: Port to wait EH for
749 *
750 * Wait until the currently pending EH is complete.
751 *
752 * LOCKING:
753 * Kernel thread context (may sleep).
754 */
755void ata_port_wait_eh(struct ata_port *ap)
756{
757 unsigned long flags;
758 DEFINE_WAIT(wait);
759
760 retry:
Jeff Garzikba6a1302006-06-22 23:46:10 -0400761 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900762
Tejun Heob51e9e52006-06-29 01:29:30 +0900763 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900764 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
Jeff Garzikba6a1302006-06-22 23:46:10 -0400765 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900766 schedule();
Jeff Garzikba6a1302006-06-22 23:46:10 -0400767 spin_lock_irqsave(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900768 }
Tejun Heo0a1b6222006-06-11 11:01:38 +0900769 finish_wait(&ap->eh_wait_q, &wait);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900770
Jeff Garzikba6a1302006-06-22 23:46:10 -0400771 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900772
773 /* make sure SCSI EH is complete */
Jeff Garzikcca39742006-08-24 03:19:22 -0400774 if (scsi_host_in_recovery(ap->scsi_host)) {
Tejun Heoc6cf9e92006-05-31 18:27:27 +0900775 msleep(10);
776 goto retry;
777 }
778}
779
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900780static int ata_eh_nr_in_flight(struct ata_port *ap)
781{
782 unsigned int tag;
783 int nr = 0;
784
785 /* count only non-internal commands */
786 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
787 if (ata_qc_from_tag(ap, tag))
788 nr++;
789
790 return nr;
791}
792
793void ata_eh_fastdrain_timerfn(unsigned long arg)
794{
795 struct ata_port *ap = (void *)arg;
796 unsigned long flags;
797 int cnt;
798
799 spin_lock_irqsave(ap->lock, flags);
800
801 cnt = ata_eh_nr_in_flight(ap);
802
803 /* are we done? */
804 if (!cnt)
805 goto out_unlock;
806
807 if (cnt == ap->fastdrain_cnt) {
808 unsigned int tag;
809
810 /* No progress during the last interval, tag all
811 * in-flight qcs as timed out and freeze the port.
812 */
813 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
814 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
815 if (qc)
816 qc->err_mask |= AC_ERR_TIMEOUT;
817 }
818
819 ata_port_freeze(ap);
820 } else {
821 /* some qcs have finished, give it another chance */
822 ap->fastdrain_cnt = cnt;
823 ap->fastdrain_timer.expires =
Tejun Heo341c2c92008-05-20 02:17:51 +0900824 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900825 add_timer(&ap->fastdrain_timer);
826 }
827
828 out_unlock:
829 spin_unlock_irqrestore(ap->lock, flags);
830}
831
832/**
833 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
834 * @ap: target ATA port
835 * @fastdrain: activate fast drain
836 *
837 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
838 * is non-zero and EH wasn't pending before. Fast drain ensures
839 * that EH kicks in in timely manner.
840 *
841 * LOCKING:
842 * spin_lock_irqsave(host lock)
843 */
844static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
845{
846 int cnt;
847
848 /* already scheduled? */
849 if (ap->pflags & ATA_PFLAG_EH_PENDING)
850 return;
851
852 ap->pflags |= ATA_PFLAG_EH_PENDING;
853
854 if (!fastdrain)
855 return;
856
857 /* do we have in-flight qcs? */
858 cnt = ata_eh_nr_in_flight(ap);
859 if (!cnt)
860 return;
861
862 /* activate fast drain */
863 ap->fastdrain_cnt = cnt;
Tejun Heo341c2c92008-05-20 02:17:51 +0900864 ap->fastdrain_timer.expires =
865 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900866 add_timer(&ap->fastdrain_timer);
867}
868
Tejun Heof686bcb2006-05-15 20:58:05 +0900869/**
870 * ata_qc_schedule_eh - schedule qc for error handling
871 * @qc: command to schedule error handling for
872 *
873 * Schedule error handling for @qc. EH will kick in as soon as
874 * other commands are drained.
875 *
876 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400877 * spin_lock_irqsave(host lock)
Tejun Heof686bcb2006-05-15 20:58:05 +0900878 */
879void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
880{
881 struct ata_port *ap = qc->ap;
Tejun Heofa41efd2010-04-15 08:57:37 +0900882 struct request_queue *q = qc->scsicmd->device->request_queue;
883 unsigned long flags;
Tejun Heof686bcb2006-05-15 20:58:05 +0900884
885 WARN_ON(!ap->ops->error_handler);
886
887 qc->flags |= ATA_QCFLAG_FAILED;
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900888 ata_eh_set_pending(ap, 1);
Tejun Heof686bcb2006-05-15 20:58:05 +0900889
890 /* The following will fail if timeout has already expired.
891 * ata_scsi_error() takes care of such scmds on EH entry.
892 * Note that ATA_QCFLAG_FAILED is unconditionally set after
893 * this function completes.
894 */
Tejun Heofa41efd2010-04-15 08:57:37 +0900895 spin_lock_irqsave(q->queue_lock, flags);
Jens Axboe242f9dc2008-09-14 05:55:09 -0700896 blk_abort_request(qc->scsicmd->request);
Tejun Heofa41efd2010-04-15 08:57:37 +0900897 spin_unlock_irqrestore(q->queue_lock, flags);
Tejun Heof686bcb2006-05-15 20:58:05 +0900898}
899
Tejun Heo7b70fc02006-05-15 20:58:07 +0900900/**
901 * ata_port_schedule_eh - schedule error handling without a qc
902 * @ap: ATA port to schedule EH for
903 *
904 * Schedule error handling for @ap. EH will kick in as soon as
905 * all commands are drained.
906 *
907 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400908 * spin_lock_irqsave(host lock)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900909 */
910void ata_port_schedule_eh(struct ata_port *ap)
911{
912 WARN_ON(!ap->ops->error_handler);
913
Tejun Heof4d6d002007-05-01 11:50:15 +0200914 if (ap->pflags & ATA_PFLAG_INITIALIZING)
915 return;
916
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900917 ata_eh_set_pending(ap, 1);
Jeff Garzikcca39742006-08-24 03:19:22 -0400918 scsi_schedule_eh(ap->scsi_host);
Tejun Heo7b70fc02006-05-15 20:58:07 +0900919
920 DPRINTK("port EH scheduled\n");
921}
922
Tejun Heodbd82612007-08-06 18:36:23 +0900923static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
Tejun Heo7b70fc02006-05-15 20:58:07 +0900924{
925 int tag, nr_aborted = 0;
926
927 WARN_ON(!ap->ops->error_handler);
928
Tejun Heo5ddf24c2007-07-16 14:29:41 +0900929 /* we're gonna abort all commands, no need for fast drain */
930 ata_eh_set_pending(ap, 0);
931
Tejun Heo7b70fc02006-05-15 20:58:07 +0900932 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
933 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
934
Tejun Heodbd82612007-08-06 18:36:23 +0900935 if (qc && (!link || qc->dev->link == link)) {
Tejun Heo7b70fc02006-05-15 20:58:07 +0900936 qc->flags |= ATA_QCFLAG_FAILED;
937 ata_qc_complete(qc);
938 nr_aborted++;
939 }
940 }
941
942 if (!nr_aborted)
943 ata_port_schedule_eh(ap);
944
945 return nr_aborted;
946}
947
Tejun Heoe3180492006-05-15 20:58:09 +0900948/**
Tejun Heodbd82612007-08-06 18:36:23 +0900949 * ata_link_abort - abort all qc's on the link
950 * @link: ATA link to abort qc's for
951 *
952 * Abort all active qc's active on @link and schedule EH.
953 *
954 * LOCKING:
955 * spin_lock_irqsave(host lock)
956 *
957 * RETURNS:
958 * Number of aborted qc's.
959 */
960int ata_link_abort(struct ata_link *link)
961{
962 return ata_do_link_abort(link->ap, link);
963}
964
965/**
966 * ata_port_abort - abort all qc's on the port
967 * @ap: ATA port to abort qc's for
968 *
969 * Abort all active qc's of @ap and schedule EH.
970 *
971 * LOCKING:
972 * spin_lock_irqsave(host_set lock)
973 *
974 * RETURNS:
975 * Number of aborted qc's.
976 */
977int ata_port_abort(struct ata_port *ap)
978{
979 return ata_do_link_abort(ap, NULL);
980}
981
982/**
Tejun Heoe3180492006-05-15 20:58:09 +0900983 * __ata_port_freeze - freeze port
984 * @ap: ATA port to freeze
985 *
986 * This function is called when HSM violation or some other
987 * condition disrupts normal operation of the port. Frozen port
988 * is not allowed to perform any operation until the port is
989 * thawed, which usually follows a successful reset.
990 *
991 * ap->ops->freeze() callback can be used for freezing the port
992 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
993 * port cannot be frozen hardware-wise, the interrupt handler
994 * must ack and clear interrupts unconditionally while the port
995 * is frozen.
996 *
997 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -0400998 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +0900999 */
1000static void __ata_port_freeze(struct ata_port *ap)
1001{
1002 WARN_ON(!ap->ops->error_handler);
1003
1004 if (ap->ops->freeze)
1005 ap->ops->freeze(ap);
1006
Tejun Heob51e9e52006-06-29 01:29:30 +09001007 ap->pflags |= ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001008
Tejun Heo44877b42007-02-21 01:06:51 +09001009 DPRINTK("ata%u port frozen\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001010}
1011
1012/**
1013 * ata_port_freeze - abort & freeze port
1014 * @ap: ATA port to freeze
1015 *
Jeff Garzik54c38442009-04-07 19:13:15 -04001016 * Abort and freeze @ap. The freeze operation must be called
1017 * first, because some hardware requires special operations
1018 * before the taskfile registers are accessible.
Tejun Heoe3180492006-05-15 20:58:09 +09001019 *
1020 * LOCKING:
Jeff Garzikcca39742006-08-24 03:19:22 -04001021 * spin_lock_irqsave(host lock)
Tejun Heoe3180492006-05-15 20:58:09 +09001022 *
1023 * RETURNS:
1024 * Number of aborted commands.
1025 */
1026int ata_port_freeze(struct ata_port *ap)
1027{
1028 int nr_aborted;
1029
1030 WARN_ON(!ap->ops->error_handler);
1031
Tejun Heoe3180492006-05-15 20:58:09 +09001032 __ata_port_freeze(ap);
Jeff Garzik54c38442009-04-07 19:13:15 -04001033 nr_aborted = ata_port_abort(ap);
Tejun Heoe3180492006-05-15 20:58:09 +09001034
1035 return nr_aborted;
1036}
1037
1038/**
Tejun Heo7d77b242007-09-23 13:14:13 +09001039 * sata_async_notification - SATA async notification handler
1040 * @ap: ATA port where async notification is received
1041 *
1042 * Handler to be called when async notification via SDB FIS is
1043 * received. This function schedules EH if necessary.
1044 *
1045 * LOCKING:
1046 * spin_lock_irqsave(host lock)
1047 *
1048 * RETURNS:
1049 * 1 if EH is scheduled, 0 otherwise.
1050 */
1051int sata_async_notification(struct ata_port *ap)
1052{
1053 u32 sntf;
1054 int rc;
1055
1056 if (!(ap->flags & ATA_FLAG_AN))
1057 return 0;
1058
1059 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1060 if (rc == 0)
1061 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1062
Tejun Heo071f44b2008-04-07 22:47:22 +09001063 if (!sata_pmp_attached(ap) || rc) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001064 /* PMP is not attached or SNTF is not available */
Tejun Heo071f44b2008-04-07 22:47:22 +09001065 if (!sata_pmp_attached(ap)) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001066 /* PMP is not attached. Check whether ATAPI
1067 * AN is configured. If so, notify media
1068 * change.
1069 */
1070 struct ata_device *dev = ap->link.device;
1071
1072 if ((dev->class == ATA_DEV_ATAPI) &&
1073 (dev->flags & ATA_DFLAG_AN))
1074 ata_scsi_media_change_notify(dev);
1075 return 0;
1076 } else {
1077 /* PMP is attached but SNTF is not available.
1078 * ATAPI async media change notification is
1079 * not used. The PMP must be reporting PHY
1080 * status change, schedule EH.
1081 */
1082 ata_port_schedule_eh(ap);
1083 return 1;
1084 }
1085 } else {
1086 /* PMP is attached and SNTF is available */
1087 struct ata_link *link;
1088
1089 /* check and notify ATAPI AN */
Tejun Heo1eca4362008-11-03 20:03:17 +09001090 ata_for_each_link(link, ap, EDGE) {
Tejun Heo7d77b242007-09-23 13:14:13 +09001091 if (!(sntf & (1 << link->pmp)))
1092 continue;
1093
1094 if ((link->device->class == ATA_DEV_ATAPI) &&
1095 (link->device->flags & ATA_DFLAG_AN))
1096 ata_scsi_media_change_notify(link->device);
1097 }
1098
1099 /* If PMP is reporting that PHY status of some
1100 * downstream ports has changed, schedule EH.
1101 */
1102 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1103 ata_port_schedule_eh(ap);
1104 return 1;
1105 }
1106
1107 return 0;
1108 }
1109}
1110
1111/**
Tejun Heoe3180492006-05-15 20:58:09 +09001112 * ata_eh_freeze_port - EH helper to freeze port
1113 * @ap: ATA port to freeze
1114 *
1115 * Freeze @ap.
1116 *
1117 * LOCKING:
1118 * None.
1119 */
1120void ata_eh_freeze_port(struct ata_port *ap)
1121{
1122 unsigned long flags;
1123
1124 if (!ap->ops->error_handler)
1125 return;
1126
Jeff Garzikba6a1302006-06-22 23:46:10 -04001127 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001128 __ata_port_freeze(ap);
Jeff Garzikba6a1302006-06-22 23:46:10 -04001129 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001130}
1131
1132/**
1133 * ata_port_thaw_port - EH helper to thaw port
1134 * @ap: ATA port to thaw
1135 *
1136 * Thaw frozen port @ap.
1137 *
1138 * LOCKING:
1139 * None.
1140 */
1141void ata_eh_thaw_port(struct ata_port *ap)
1142{
1143 unsigned long flags;
1144
1145 if (!ap->ops->error_handler)
1146 return;
1147
Jeff Garzikba6a1302006-06-22 23:46:10 -04001148 spin_lock_irqsave(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001149
Tejun Heob51e9e52006-06-29 01:29:30 +09001150 ap->pflags &= ~ATA_PFLAG_FROZEN;
Tejun Heoe3180492006-05-15 20:58:09 +09001151
1152 if (ap->ops->thaw)
1153 ap->ops->thaw(ap);
1154
Jeff Garzikba6a1302006-06-22 23:46:10 -04001155 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoe3180492006-05-15 20:58:09 +09001156
Tejun Heo44877b42007-02-21 01:06:51 +09001157 DPRINTK("ata%u port thawed\n", ap->print_id);
Tejun Heoe3180492006-05-15 20:58:09 +09001158}
1159
Tejun Heoece1d632006-04-02 18:51:53 +09001160static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1161{
1162 /* nada */
1163}
1164
1165static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1166{
1167 struct ata_port *ap = qc->ap;
1168 struct scsi_cmnd *scmd = qc->scsicmd;
1169 unsigned long flags;
1170
Jeff Garzikba6a1302006-06-22 23:46:10 -04001171 spin_lock_irqsave(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001172 qc->scsidone = ata_eh_scsidone;
1173 __ata_qc_complete(qc);
1174 WARN_ON(ata_tag_valid(qc->tag));
Jeff Garzikba6a1302006-06-22 23:46:10 -04001175 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heoece1d632006-04-02 18:51:53 +09001176
1177 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1178}
1179
1180/**
1181 * ata_eh_qc_complete - Complete an active ATA command from EH
1182 * @qc: Command to complete
1183 *
1184 * Indicate to the mid and upper layers that an ATA command has
1185 * completed. To be used from EH.
1186 */
1187void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1188{
1189 struct scsi_cmnd *scmd = qc->scsicmd;
1190 scmd->retries = scmd->allowed;
1191 __ata_eh_qc_complete(qc);
1192}
1193
1194/**
1195 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1196 * @qc: Command to retry
1197 *
1198 * Indicate to the mid and upper layers that an ATA command
1199 * should be retried. To be used from EH.
1200 *
1201 * SCSI midlayer limits the number of retries to scmd->allowed.
1202 * scmd->retries is decremented for commands which get retried
1203 * due to unrelated failures (qc->err_mask is zero).
1204 */
1205void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1206{
1207 struct scsi_cmnd *scmd = qc->scsicmd;
1208 if (!qc->err_mask && scmd->retries)
1209 scmd->retries--;
1210 __ata_eh_qc_complete(qc);
1211}
Tejun Heo022bdb02006-05-15 20:58:22 +09001212
1213/**
Tejun Heo678afac62009-01-29 20:31:30 +09001214 * ata_dev_disable - disable ATA device
1215 * @dev: ATA device to disable
1216 *
1217 * Disable @dev.
1218 *
1219 * Locking:
1220 * EH context.
1221 */
1222void ata_dev_disable(struct ata_device *dev)
1223{
1224 if (!ata_dev_enabled(dev))
1225 return;
1226
1227 if (ata_msg_drv(dev->link->ap))
1228 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1229 ata_acpi_on_disable(dev);
1230 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1231 dev->class++;
Tejun Heo99cf6102009-01-29 20:31:32 +09001232
1233 /* From now till the next successful probe, ering is used to
1234 * track probe failures. Clear accumulated device error info.
1235 */
1236 ata_ering_clear(&dev->ering);
Tejun Heo678afac62009-01-29 20:31:30 +09001237}
1238
1239/**
Tejun Heo0ea035a2006-05-31 18:28:01 +09001240 * ata_eh_detach_dev - detach ATA device
1241 * @dev: ATA device to detach
1242 *
1243 * Detach @dev.
1244 *
1245 * LOCKING:
1246 * None.
1247 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001248void ata_eh_detach_dev(struct ata_device *dev)
Tejun Heo0ea035a2006-05-31 18:28:01 +09001249{
Tejun Heof58229f2007-08-06 18:36:23 +09001250 struct ata_link *link = dev->link;
1251 struct ata_port *ap = link->ap;
Tejun Heo90484eb2008-10-26 15:43:03 +09001252 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001253 unsigned long flags;
1254
1255 ata_dev_disable(dev);
1256
Jeff Garzikba6a1302006-06-22 23:46:10 -04001257 spin_lock_irqsave(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001258
1259 dev->flags &= ~ATA_DFLAG_DETACH;
1260
1261 if (ata_scsi_offline_dev(dev)) {
1262 dev->flags |= ATA_DFLAG_DETACHED;
Tejun Heob51e9e52006-06-29 01:29:30 +09001263 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
Tejun Heo0ea035a2006-05-31 18:28:01 +09001264 }
1265
Tejun Heo90484eb2008-10-26 15:43:03 +09001266 /* clear per-dev EH info */
Tejun Heof58229f2007-08-06 18:36:23 +09001267 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1268 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
Tejun Heo90484eb2008-10-26 15:43:03 +09001269 ehc->saved_xfer_mode[dev->devno] = 0;
1270 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heobeb07c12006-06-24 20:30:19 +09001271
Jeff Garzikba6a1302006-06-22 23:46:10 -04001272 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo0ea035a2006-05-31 18:28:01 +09001273}
1274
1275/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001276 * ata_eh_about_to_do - about to perform eh_action
Tejun Heo955e57d2007-08-06 18:36:23 +09001277 * @link: target ATA link
Tejun Heo47005f22006-06-19 18:27:23 +09001278 * @dev: target ATA dev for per-dev action (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09001279 * @action: action about to be performed
1280 *
1281 * Called just before performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001282 * in @link->eh_info such that eh actions are not unnecessarily
1283 * repeated.
Tejun Heo022bdb02006-05-15 20:58:22 +09001284 *
1285 * LOCKING:
1286 * None.
1287 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001288void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1289 unsigned int action)
Tejun Heo022bdb02006-05-15 20:58:22 +09001290{
Tejun Heo955e57d2007-08-06 18:36:23 +09001291 struct ata_port *ap = link->ap;
1292 struct ata_eh_info *ehi = &link->eh_info;
1293 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001294 unsigned long flags;
1295
Jeff Garzikba6a1302006-06-22 23:46:10 -04001296 spin_lock_irqsave(ap->lock, flags);
Tejun Heo1cdaf532006-07-03 16:07:26 +09001297
Tejun Heo955e57d2007-08-06 18:36:23 +09001298 ata_eh_clear_action(link, dev, ehi, action);
Tejun Heo13abf502006-07-10 23:18:46 +09001299
Tejun Heoa568d1d2008-10-21 20:37:21 +09001300 /* About to take EH action, set RECOVERED. Ignore actions on
1301 * slave links as master will do them again.
1302 */
1303 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
Tejun Heo1cdaf532006-07-03 16:07:26 +09001304 ap->pflags |= ATA_PFLAG_RECOVERED;
1305
Jeff Garzikba6a1302006-06-22 23:46:10 -04001306 spin_unlock_irqrestore(ap->lock, flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09001307}
1308
1309/**
Tejun Heo47005f22006-06-19 18:27:23 +09001310 * ata_eh_done - EH action complete
Tejun Heo955e57d2007-08-06 18:36:23 +09001311* @ap: target ATA port
Tejun Heo47005f22006-06-19 18:27:23 +09001312 * @dev: target ATA dev for per-dev action (can be NULL)
1313 * @action: action just completed
1314 *
1315 * Called right after performing EH actions to clear related bits
Tejun Heo955e57d2007-08-06 18:36:23 +09001316 * in @link->eh_context.
Tejun Heo47005f22006-06-19 18:27:23 +09001317 *
1318 * LOCKING:
1319 * None.
1320 */
Tejun Heofb7fd612007-09-23 13:14:12 +09001321void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1322 unsigned int action)
Tejun Heo47005f22006-06-19 18:27:23 +09001323{
Tejun Heo955e57d2007-08-06 18:36:23 +09001324 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001325
Tejun Heo955e57d2007-08-06 18:36:23 +09001326 ata_eh_clear_action(link, dev, &ehc->i, action);
Tejun Heo47005f22006-06-19 18:27:23 +09001327}
1328
1329/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001330 * ata_err_string - convert err_mask to descriptive string
1331 * @err_mask: error mask to convert to string
1332 *
1333 * Convert @err_mask to descriptive string. Errors are
1334 * prioritized according to severity and only the most severe
1335 * error is reported.
1336 *
1337 * LOCKING:
1338 * None.
1339 *
1340 * RETURNS:
1341 * Descriptive string for @err_mask
1342 */
Jeff Garzik2dcb4072007-10-19 06:42:56 -04001343static const char *ata_err_string(unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001344{
1345 if (err_mask & AC_ERR_HOST_BUS)
1346 return "host bus error";
1347 if (err_mask & AC_ERR_ATA_BUS)
1348 return "ATA bus error";
1349 if (err_mask & AC_ERR_TIMEOUT)
1350 return "timeout";
1351 if (err_mask & AC_ERR_HSM)
1352 return "HSM violation";
1353 if (err_mask & AC_ERR_SYSTEM)
1354 return "internal error";
1355 if (err_mask & AC_ERR_MEDIA)
1356 return "media error";
1357 if (err_mask & AC_ERR_INVALID)
1358 return "invalid argument";
1359 if (err_mask & AC_ERR_DEV)
1360 return "device error";
1361 return "unknown error";
1362}
1363
1364/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001365 * ata_read_log_page - read a specific log page
1366 * @dev: target device
1367 * @page: page to read
1368 * @buf: buffer to store read page
1369 * @sectors: number of sectors to read
1370 *
1371 * Read log page using READ_LOG_EXT command.
1372 *
1373 * LOCKING:
1374 * Kernel thread context (may sleep).
1375 *
1376 * RETURNS:
1377 * 0 on success, AC_ERR_* mask otherwise.
1378 */
1379static unsigned int ata_read_log_page(struct ata_device *dev,
1380 u8 page, void *buf, unsigned int sectors)
1381{
1382 struct ata_taskfile tf;
1383 unsigned int err_mask;
1384
1385 DPRINTK("read log page - page %d\n", page);
1386
1387 ata_tf_init(dev, &tf);
1388 tf.command = ATA_CMD_READ_LOG_EXT;
1389 tf.lbal = page;
1390 tf.nsect = sectors;
1391 tf.hob_nsect = sectors >> 8;
1392 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1393 tf.protocol = ATA_PROT_PIO;
1394
1395 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001396 buf, sectors * ATA_SECT_SIZE, 0);
Tejun Heoe8ee8452006-05-15 21:03:46 +09001397
1398 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1399 return err_mask;
1400}
1401
1402/**
1403 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1404 * @dev: Device to read log page 10h from
1405 * @tag: Resulting tag of the failed command
1406 * @tf: Resulting taskfile registers of the failed command
1407 *
1408 * Read log page 10h to obtain NCQ error details and clear error
1409 * condition.
1410 *
1411 * LOCKING:
1412 * Kernel thread context (may sleep).
1413 *
1414 * RETURNS:
1415 * 0 on success, -errno otherwise.
1416 */
1417static int ata_eh_read_log_10h(struct ata_device *dev,
1418 int *tag, struct ata_taskfile *tf)
1419{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001420 u8 *buf = dev->link->ap->sector_buf;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001421 unsigned int err_mask;
1422 u8 csum;
1423 int i;
1424
1425 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1426 if (err_mask)
1427 return -EIO;
1428
1429 csum = 0;
1430 for (i = 0; i < ATA_SECT_SIZE; i++)
1431 csum += buf[i];
1432 if (csum)
1433 ata_dev_printk(dev, KERN_WARNING,
1434 "invalid checksum 0x%x on log page 10h\n", csum);
1435
1436 if (buf[0] & 0x80)
1437 return -ENOENT;
1438
1439 *tag = buf[0] & 0x1f;
1440
1441 tf->command = buf[2];
1442 tf->feature = buf[3];
1443 tf->lbal = buf[4];
1444 tf->lbam = buf[5];
1445 tf->lbah = buf[6];
1446 tf->device = buf[7];
1447 tf->hob_lbal = buf[8];
1448 tf->hob_lbam = buf[9];
1449 tf->hob_lbah = buf[10];
1450 tf->nsect = buf[12];
1451 tf->hob_nsect = buf[13];
1452
1453 return 0;
1454}
1455
1456/**
Tejun Heo11fc33d2008-08-30 14:20:01 +02001457 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1458 * @dev: target ATAPI device
1459 * @r_sense_key: out parameter for sense_key
1460 *
1461 * Perform ATAPI TEST_UNIT_READY.
1462 *
1463 * LOCKING:
1464 * EH context (may sleep).
1465 *
1466 * RETURNS:
1467 * 0 on success, AC_ERR_* mask on failure.
1468 */
1469static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1470{
1471 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1472 struct ata_taskfile tf;
1473 unsigned int err_mask;
1474
1475 ata_tf_init(dev, &tf);
1476
1477 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1478 tf.command = ATA_CMD_PACKET;
1479 tf.protocol = ATAPI_PROT_NODATA;
1480
1481 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1482 if (err_mask == AC_ERR_DEV)
1483 *r_sense_key = tf.feature >> 4;
1484 return err_mask;
1485}
1486
1487/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001488 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1489 * @dev: device to perform REQUEST_SENSE to
1490 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
Tejun Heo3eabddb2008-06-10 18:28:05 +09001491 * @dfl_sense_key: default sense key to use
Tejun Heo022bdb02006-05-15 20:58:22 +09001492 *
1493 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1494 * SENSE. This function is EH helper.
1495 *
1496 * LOCKING:
1497 * Kernel thread context (may sleep).
1498 *
1499 * RETURNS:
1500 * 0 on success, AC_ERR_* mask on failure
1501 */
Tejun Heo3eabddb2008-06-10 18:28:05 +09001502static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1503 u8 *sense_buf, u8 dfl_sense_key)
Tejun Heo022bdb02006-05-15 20:58:22 +09001504{
Tejun Heo3eabddb2008-06-10 18:28:05 +09001505 u8 cdb[ATAPI_CDB_LEN] =
1506 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
Tejun Heo9af5c9c2007-08-06 18:36:22 +09001507 struct ata_port *ap = dev->link->ap;
Tejun Heo022bdb02006-05-15 20:58:22 +09001508 struct ata_taskfile tf;
Tejun Heo022bdb02006-05-15 20:58:22 +09001509
1510 DPRINTK("ATAPI request sense\n");
1511
Tejun Heo022bdb02006-05-15 20:58:22 +09001512 /* FIXME: is this needed? */
1513 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1514
Albert Lee56287762007-04-02 11:30:46 +08001515 /* initialize sense_buf with the error register,
1516 * for the case where they are -not- overwritten
1517 */
Tejun Heo022bdb02006-05-15 20:58:22 +09001518 sense_buf[0] = 0x70;
Tejun Heo3eabddb2008-06-10 18:28:05 +09001519 sense_buf[2] = dfl_sense_key;
Albert Lee56287762007-04-02 11:30:46 +08001520
Jeff Garzika617c092007-05-21 20:14:23 -04001521 /* some devices time out if garbage left in tf */
Albert Lee56287762007-04-02 11:30:46 +08001522 ata_tf_init(dev, &tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09001523
Tejun Heo022bdb02006-05-15 20:58:22 +09001524 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1525 tf.command = ATA_CMD_PACKET;
1526
1527 /* is it pointless to prefer PIO for "safety reasons"? */
1528 if (ap->flags & ATA_FLAG_PIO_DMA) {
Tejun Heo0dc36882007-12-18 16:34:43 -05001529 tf.protocol = ATAPI_PROT_DMA;
Tejun Heo022bdb02006-05-15 20:58:22 +09001530 tf.feature |= ATAPI_PKT_DMA;
1531 } else {
Tejun Heo0dc36882007-12-18 16:34:43 -05001532 tf.protocol = ATAPI_PROT_PIO;
Tejun Heof2dfc1a2007-12-12 12:12:46 +09001533 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1534 tf.lbah = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001535 }
1536
1537 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
Tejun Heo2b789102007-10-09 15:05:44 +09001538 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
Tejun Heo022bdb02006-05-15 20:58:22 +09001539}
1540
1541/**
1542 * ata_eh_analyze_serror - analyze SError for a failed port
Tejun Heo02607312007-08-06 18:36:23 +09001543 * @link: ATA link to analyze SError for
Tejun Heo022bdb02006-05-15 20:58:22 +09001544 *
1545 * Analyze SError if available and further determine cause of
1546 * failure.
1547 *
1548 * LOCKING:
1549 * None.
1550 */
Tejun Heo02607312007-08-06 18:36:23 +09001551static void ata_eh_analyze_serror(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001552{
Tejun Heo02607312007-08-06 18:36:23 +09001553 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09001554 u32 serror = ehc->i.serror;
1555 unsigned int err_mask = 0, action = 0;
Tejun Heof9df58c2007-09-23 13:14:13 +09001556 u32 hotplug_mask;
Tejun Heo022bdb02006-05-15 20:58:22 +09001557
Tejun Heoe0614db2008-05-19 01:15:09 +09001558 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
Tejun Heo022bdb02006-05-15 20:58:22 +09001559 err_mask |= AC_ERR_ATA_BUS;
Tejun Heocf480622008-01-24 00:05:14 +09001560 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001561 }
1562 if (serror & SERR_PROTOCOL) {
1563 err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001564 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001565 }
1566 if (serror & SERR_INTERNAL) {
1567 err_mask |= AC_ERR_SYSTEM;
Tejun Heocf480622008-01-24 00:05:14 +09001568 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001569 }
Tejun Heof9df58c2007-09-23 13:14:13 +09001570
1571 /* Determine whether a hotplug event has occurred. Both
1572 * SError.N/X are considered hotplug events for enabled or
1573 * host links. For disabled PMP links, only N bit is
1574 * considered as X bit is left at 1 for link plugging.
1575 */
1576 hotplug_mask = 0;
1577
1578 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1579 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1580 else
1581 hotplug_mask = SERR_PHYRDY_CHG;
1582
1583 if (serror & hotplug_mask)
Tejun Heo084fe632006-05-31 18:28:03 +09001584 ata_ehi_hotplugged(&ehc->i);
Tejun Heo022bdb02006-05-15 20:58:22 +09001585
1586 ehc->i.err_mask |= err_mask;
1587 ehc->i.action |= action;
1588}
1589
1590/**
Tejun Heoe8ee8452006-05-15 21:03:46 +09001591 * ata_eh_analyze_ncq_error - analyze NCQ error
Tejun Heo02607312007-08-06 18:36:23 +09001592 * @link: ATA link to analyze NCQ error for
Tejun Heoe8ee8452006-05-15 21:03:46 +09001593 *
1594 * Read log page 10h, determine the offending qc and acquire
1595 * error status TF. For NCQ device errors, all LLDDs have to do
1596 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1597 * care of the rest.
1598 *
1599 * LOCKING:
1600 * Kernel thread context (may sleep).
1601 */
Mark Lord10acf3b2008-05-02 02:14:53 -04001602void ata_eh_analyze_ncq_error(struct ata_link *link)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001603{
Tejun Heo02607312007-08-06 18:36:23 +09001604 struct ata_port *ap = link->ap;
1605 struct ata_eh_context *ehc = &link->eh_context;
1606 struct ata_device *dev = link->device;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001607 struct ata_queued_cmd *qc;
1608 struct ata_taskfile tf;
1609 int tag, rc;
1610
1611 /* if frozen, we can't do much */
Tejun Heob51e9e52006-06-29 01:29:30 +09001612 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heoe8ee8452006-05-15 21:03:46 +09001613 return;
1614
1615 /* is it NCQ device error? */
Tejun Heo02607312007-08-06 18:36:23 +09001616 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
Tejun Heoe8ee8452006-05-15 21:03:46 +09001617 return;
1618
1619 /* has LLDD analyzed already? */
1620 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1621 qc = __ata_qc_from_tag(ap, tag);
1622
1623 if (!(qc->flags & ATA_QCFLAG_FAILED))
1624 continue;
1625
1626 if (qc->err_mask)
1627 return;
1628 }
1629
1630 /* okay, this error is ours */
Jeff Garzika09bf4c2010-04-22 21:59:13 -04001631 memset(&tf, 0, sizeof(tf));
Tejun Heoe8ee8452006-05-15 21:03:46 +09001632 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1633 if (rc) {
Tejun Heo02607312007-08-06 18:36:23 +09001634 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001635 "(errno=%d)\n", rc);
1636 return;
1637 }
1638
Tejun Heo02607312007-08-06 18:36:23 +09001639 if (!(link->sactive & (1 << tag))) {
1640 ata_link_printk(link, KERN_ERR, "log page 10h reported "
Tejun Heoe8ee8452006-05-15 21:03:46 +09001641 "inactive tag %d\n", tag);
1642 return;
1643 }
1644
1645 /* we've got the perpetrator, condemn it */
1646 qc = __ata_qc_from_tag(ap, tag);
1647 memcpy(&qc->result_tf, &tf, sizeof(tf));
Mark Lorda6116c92008-04-23 22:36:25 -04001648 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
Tejun Heo5335b722007-07-16 14:29:40 +09001649 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
Tejun Heoe8ee8452006-05-15 21:03:46 +09001650 ehc->i.err_mask &= ~AC_ERR_DEV;
1651}
1652
1653/**
Tejun Heo022bdb02006-05-15 20:58:22 +09001654 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1655 * @qc: qc to analyze
1656 * @tf: Taskfile registers to analyze
1657 *
1658 * Analyze taskfile of @qc and further determine cause of
1659 * failure. This function also requests ATAPI sense data if
1660 * avaliable.
1661 *
1662 * LOCKING:
1663 * Kernel thread context (may sleep).
1664 *
1665 * RETURNS:
1666 * Determined recovery action
1667 */
1668static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1669 const struct ata_taskfile *tf)
1670{
1671 unsigned int tmp, action = 0;
1672 u8 stat = tf->command, err = tf->feature;
1673
1674 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1675 qc->err_mask |= AC_ERR_HSM;
Tejun Heocf480622008-01-24 00:05:14 +09001676 return ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001677 }
1678
Tejun Heoa51d6442007-03-20 15:24:11 +09001679 if (stat & (ATA_ERR | ATA_DF))
1680 qc->err_mask |= AC_ERR_DEV;
1681 else
Tejun Heo022bdb02006-05-15 20:58:22 +09001682 return 0;
1683
1684 switch (qc->dev->class) {
1685 case ATA_DEV_ATA:
1686 if (err & ATA_ICRC)
1687 qc->err_mask |= AC_ERR_ATA_BUS;
1688 if (err & ATA_UNC)
1689 qc->err_mask |= AC_ERR_MEDIA;
1690 if (err & ATA_IDNF)
1691 qc->err_mask |= AC_ERR_INVALID;
1692 break;
1693
1694 case ATA_DEV_ATAPI:
Tejun Heoa569a302006-11-21 10:40:51 +09001695 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
Tejun Heo3eabddb2008-06-10 18:28:05 +09001696 tmp = atapi_eh_request_sense(qc->dev,
1697 qc->scsicmd->sense_buffer,
1698 qc->result_tf.feature >> 4);
Tejun Heoa569a302006-11-21 10:40:51 +09001699 if (!tmp) {
1700 /* ATA_QCFLAG_SENSE_VALID is used to
1701 * tell atapi_qc_complete() that sense
1702 * data is already valid.
1703 *
1704 * TODO: interpret sense data and set
1705 * appropriate err_mask.
1706 */
1707 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1708 } else
1709 qc->err_mask |= tmp;
1710 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001711 }
1712
1713 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
Tejun Heocf480622008-01-24 00:05:14 +09001714 action |= ATA_EH_RESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09001715
1716 return action;
1717}
1718
Tejun Heo76326ac2007-11-27 19:28:59 +09001719static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1720 int *xfer_ok)
Tejun Heo022bdb02006-05-15 20:58:22 +09001721{
Tejun Heo76326ac2007-11-27 19:28:59 +09001722 int base = 0;
1723
1724 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1725 *xfer_ok = 1;
1726
1727 if (!*xfer_ok)
Tejun Heo75f9caf2008-01-03 01:21:14 +09001728 base = ATA_ECAT_DUBIOUS_NONE;
Tejun Heo76326ac2007-11-27 19:28:59 +09001729
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001730 if (err_mask & AC_ERR_ATA_BUS)
Tejun Heo76326ac2007-11-27 19:28:59 +09001731 return base + ATA_ECAT_ATA_BUS;
Tejun Heo022bdb02006-05-15 20:58:22 +09001732
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001733 if (err_mask & AC_ERR_TIMEOUT)
Tejun Heo76326ac2007-11-27 19:28:59 +09001734 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001735
Tejun Heo3884f7b2007-11-27 19:28:56 +09001736 if (eflags & ATA_EFLAG_IS_IO) {
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001737 if (err_mask & AC_ERR_HSM)
Tejun Heo76326ac2007-11-27 19:28:59 +09001738 return base + ATA_ECAT_TOUT_HSM;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001739 if ((err_mask &
1740 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
Tejun Heo76326ac2007-11-27 19:28:59 +09001741 return base + ATA_ECAT_UNK_DEV;
Tejun Heo022bdb02006-05-15 20:58:22 +09001742 }
1743
1744 return 0;
1745}
1746
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001747struct speed_down_verdict_arg {
Tejun Heo022bdb02006-05-15 20:58:22 +09001748 u64 since;
Tejun Heo76326ac2007-11-27 19:28:59 +09001749 int xfer_ok;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001750 int nr_errors[ATA_ECAT_NR];
Tejun Heo022bdb02006-05-15 20:58:22 +09001751};
1752
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001753static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
Tejun Heo022bdb02006-05-15 20:58:22 +09001754{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001755 struct speed_down_verdict_arg *arg = void_arg;
Tejun Heo76326ac2007-11-27 19:28:59 +09001756 int cat;
Tejun Heo022bdb02006-05-15 20:58:22 +09001757
1758 if (ent->timestamp < arg->since)
1759 return -1;
1760
Tejun Heo76326ac2007-11-27 19:28:59 +09001761 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1762 &arg->xfer_ok);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001763 arg->nr_errors[cat]++;
Tejun Heo76326ac2007-11-27 19:28:59 +09001764
Tejun Heo022bdb02006-05-15 20:58:22 +09001765 return 0;
1766}
1767
1768/**
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001769 * ata_eh_speed_down_verdict - Determine speed down verdict
Tejun Heo022bdb02006-05-15 20:58:22 +09001770 * @dev: Device of interest
1771 *
1772 * This function examines error ring of @dev and determines
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001773 * whether NCQ needs to be turned off, transfer speed should be
1774 * stepped down, or falling back to PIO is necessary.
Tejun Heo022bdb02006-05-15 20:58:22 +09001775 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001776 * ECAT_ATA_BUS : ATA_BUS error for any command
Tejun Heo022bdb02006-05-15 20:58:22 +09001777 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001778 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1779 * IO commands
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001780 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001781 * ECAT_UNK_DEV : Unknown DEV error for IO commands
Tejun Heo022bdb02006-05-15 20:58:22 +09001782 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001783 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1784 * data transfer hasn't been verified.
1785 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001786 * Verdicts are
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001787 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001788 * NCQ_OFF : Turn off NCQ.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001789 *
Tejun Heo3884f7b2007-11-27 19:28:56 +09001790 * SPEED_DOWN : Speed down transfer speed but don't fall back
1791 * to PIO.
1792 *
1793 * FALLBACK_TO_PIO : Fall back to PIO.
1794 *
1795 * Even if multiple verdicts are returned, only one action is
Tejun Heo76326ac2007-11-27 19:28:59 +09001796 * taken per error. An action triggered by non-DUBIOUS errors
1797 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1798 * This is to expedite speed down decisions right after device is
1799 * initially configured.
Tejun Heo3884f7b2007-11-27 19:28:56 +09001800 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001801 * The followings are speed down rules. #1 and #2 deal with
1802 * DUBIOUS errors.
1803 *
1804 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1805 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1806 *
1807 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1808 * occurred during last 5 mins, NCQ_OFF.
1809 *
1810 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
Tejun Heo3884f7b2007-11-27 19:28:56 +09001811 * ocurred during last 5 mins, FALLBACK_TO_PIO
1812 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001813 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
Tejun Heo3884f7b2007-11-27 19:28:56 +09001814 * during last 10 mins, NCQ_OFF.
1815 *
Tejun Heo76326ac2007-11-27 19:28:59 +09001816 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
Tejun Heo3884f7b2007-11-27 19:28:56 +09001817 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001818 *
Tejun Heo022bdb02006-05-15 20:58:22 +09001819 * LOCKING:
1820 * Inherited from caller.
1821 *
1822 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001823 * OR of ATA_EH_SPDN_* flags.
Tejun Heo022bdb02006-05-15 20:58:22 +09001824 */
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001825static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09001826{
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001827 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1828 u64 j64 = get_jiffies_64();
1829 struct speed_down_verdict_arg arg;
1830 unsigned int verdict = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09001831
Tejun Heo3884f7b2007-11-27 19:28:56 +09001832 /* scan past 5 mins of error history */
1833 memset(&arg, 0, sizeof(arg));
1834 arg.since = j64 - min(j64, j5mins);
1835 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1836
Tejun Heo76326ac2007-11-27 19:28:59 +09001837 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1838 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1839 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1840 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1841
1842 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1843 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1844 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1845
Tejun Heo3884f7b2007-11-27 19:28:56 +09001846 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1847 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
Tejun Heo663f99b82007-11-27 19:28:57 +09001848 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo3884f7b2007-11-27 19:28:56 +09001849 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1850
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001851 /* scan past 10 mins of error history */
Tejun Heo022bdb02006-05-15 20:58:22 +09001852 memset(&arg, 0, sizeof(arg));
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001853 arg.since = j64 - min(j64, j10mins);
1854 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
Tejun Heo022bdb02006-05-15 20:58:22 +09001855
Tejun Heo3884f7b2007-11-27 19:28:56 +09001856 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1857 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001858 verdict |= ATA_EH_SPDN_NCQ_OFF;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001859
1860 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1861 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
Tejun Heo663f99b82007-11-27 19:28:57 +09001862 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001863 verdict |= ATA_EH_SPDN_SPEED_DOWN;
Tejun Heo022bdb02006-05-15 20:58:22 +09001864
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001865 return verdict;
Tejun Heo022bdb02006-05-15 20:58:22 +09001866}
1867
1868/**
1869 * ata_eh_speed_down - record error and speed down if necessary
1870 * @dev: Failed device
Tejun Heo3884f7b2007-11-27 19:28:56 +09001871 * @eflags: mask of ATA_EFLAG_* flags
Tejun Heo022bdb02006-05-15 20:58:22 +09001872 * @err_mask: err_mask of the error
1873 *
1874 * Record error and examine error history to determine whether
1875 * adjusting transmission speed is necessary. It also sets
1876 * transmission limits appropriately if such adjustment is
1877 * necessary.
1878 *
1879 * LOCKING:
1880 * Kernel thread context (may sleep).
1881 *
1882 * RETURNS:
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001883 * Determined recovery action.
Tejun Heo022bdb02006-05-15 20:58:22 +09001884 */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001885static unsigned int ata_eh_speed_down(struct ata_device *dev,
1886 unsigned int eflags, unsigned int err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09001887{
Tejun Heob1c72912008-07-31 17:02:43 +09001888 struct ata_link *link = ata_dev_phys_link(dev);
Tejun Heo76326ac2007-11-27 19:28:59 +09001889 int xfer_ok = 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001890 unsigned int verdict;
1891 unsigned int action = 0;
1892
1893 /* don't bother if Cat-0 error */
Tejun Heo76326ac2007-11-27 19:28:59 +09001894 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
Tejun Heo022bdb02006-05-15 20:58:22 +09001895 return 0;
1896
1897 /* record error and determine whether speed down is necessary */
Tejun Heo3884f7b2007-11-27 19:28:56 +09001898 ata_ering_record(&dev->ering, eflags, err_mask);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001899 verdict = ata_eh_speed_down_verdict(dev);
Tejun Heo022bdb02006-05-15 20:58:22 +09001900
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001901 /* turn off NCQ? */
1902 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1903 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1904 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1905 dev->flags |= ATA_DFLAG_NCQ_OFF;
1906 ata_dev_printk(dev, KERN_WARNING,
1907 "NCQ disabled due to excessive errors\n");
1908 goto done;
1909 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001910
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001911 /* speed down? */
1912 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1913 /* speed down SATA link speed if possible */
Tejun Heoa07d4992009-01-29 20:31:33 +09001914 if (sata_down_spd_limit(link, 0) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001915 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001916 goto done;
1917 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001918
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001919 /* lower transfer mode */
1920 if (dev->spdn_cnt < 2) {
1921 static const int dma_dnxfer_sel[] =
1922 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1923 static const int pio_dnxfer_sel[] =
1924 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1925 int sel;
Tejun Heo022bdb02006-05-15 20:58:22 +09001926
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001927 if (dev->xfer_shift != ATA_SHIFT_PIO)
1928 sel = dma_dnxfer_sel[dev->spdn_cnt];
1929 else
1930 sel = pio_dnxfer_sel[dev->spdn_cnt];
1931
1932 dev->spdn_cnt++;
1933
1934 if (ata_down_xfermask_limit(dev, sel) == 0) {
Tejun Heocf480622008-01-24 00:05:14 +09001935 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001936 goto done;
1937 }
1938 }
1939 }
1940
1941 /* Fall back to PIO? Slowing down to PIO is meaningless for
Tejun Heo663f99b82007-11-27 19:28:57 +09001942 * SATA ATA devices. Consider it only for PATA and SATAPI.
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001943 */
1944 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
Tejun Heo663f99b82007-11-27 19:28:57 +09001945 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001946 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1947 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1948 dev->spdn_cnt = 0;
Tejun Heocf480622008-01-24 00:05:14 +09001949 action |= ATA_EH_RESET;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001950 goto done;
1951 }
1952 }
1953
Tejun Heo022bdb02006-05-15 20:58:22 +09001954 return 0;
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001955 done:
1956 /* device has been slowed down, blow error history */
Tejun Heo76326ac2007-11-27 19:28:59 +09001957 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1958 ata_ering_clear(&dev->ering);
Tejun Heo7d47e8d2007-02-02 16:22:31 +09001959 return action;
Tejun Heo022bdb02006-05-15 20:58:22 +09001960}
1961
1962/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09001963 * ata_eh_link_autopsy - analyze error and determine recovery action
1964 * @link: host link to perform autopsy on
Tejun Heo022bdb02006-05-15 20:58:22 +09001965 *
Tejun Heo02607312007-08-06 18:36:23 +09001966 * Analyze why @link failed and determine which recovery actions
1967 * are needed. This function also sets more detailed AC_ERR_*
1968 * values and fills sense data for ATAPI CHECK SENSE.
Tejun Heo022bdb02006-05-15 20:58:22 +09001969 *
1970 * LOCKING:
1971 * Kernel thread context (may sleep).
1972 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09001973static void ata_eh_link_autopsy(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09001974{
Tejun Heo02607312007-08-06 18:36:23 +09001975 struct ata_port *ap = link->ap;
Tejun Heo936fd732007-08-06 18:36:23 +09001976 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heodfcc1732007-10-31 10:17:05 +09001977 struct ata_device *dev;
Tejun Heo3884f7b2007-11-27 19:28:56 +09001978 unsigned int all_err_mask = 0, eflags = 0;
1979 int tag;
Tejun Heo022bdb02006-05-15 20:58:22 +09001980 u32 serror;
1981 int rc;
1982
1983 DPRINTK("ENTER\n");
1984
Tejun Heo1cdaf532006-07-03 16:07:26 +09001985 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1986 return;
1987
Tejun Heo022bdb02006-05-15 20:58:22 +09001988 /* obtain and analyze SError */
Tejun Heo936fd732007-08-06 18:36:23 +09001989 rc = sata_scr_read(link, SCR_ERROR, &serror);
Tejun Heo022bdb02006-05-15 20:58:22 +09001990 if (rc == 0) {
1991 ehc->i.serror |= serror;
Tejun Heo02607312007-08-06 18:36:23 +09001992 ata_eh_analyze_serror(link);
Tejun Heo4e57c512007-07-16 14:29:41 +09001993 } else if (rc != -EOPNOTSUPP) {
Tejun Heocf480622008-01-24 00:05:14 +09001994 /* SError read failed, force reset and probing */
Tejun Heob558edd2008-01-24 00:05:14 +09001995 ehc->i.probe_mask |= ATA_ALL_DEVICES;
Tejun Heocf480622008-01-24 00:05:14 +09001996 ehc->i.action |= ATA_EH_RESET;
Tejun Heo4e57c512007-07-16 14:29:41 +09001997 ehc->i.err_mask |= AC_ERR_OTHER;
1998 }
Tejun Heo022bdb02006-05-15 20:58:22 +09001999
Tejun Heoe8ee8452006-05-15 21:03:46 +09002000 /* analyze NCQ failure */
Tejun Heo02607312007-08-06 18:36:23 +09002001 ata_eh_analyze_ncq_error(link);
Tejun Heoe8ee8452006-05-15 21:03:46 +09002002
Tejun Heo022bdb02006-05-15 20:58:22 +09002003 /* any real error trumps AC_ERR_OTHER */
2004 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2005 ehc->i.err_mask &= ~AC_ERR_OTHER;
2006
2007 all_err_mask |= ehc->i.err_mask;
2008
2009 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2010 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2011
Tejun Heob1c72912008-07-31 17:02:43 +09002012 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2013 ata_dev_phys_link(qc->dev) != link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002014 continue;
2015
2016 /* inherit upper level err_mask */
2017 qc->err_mask |= ehc->i.err_mask;
2018
Tejun Heo022bdb02006-05-15 20:58:22 +09002019 /* analyze TF */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002020 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002021
2022 /* DEV errors are probably spurious in case of ATA_BUS error */
2023 if (qc->err_mask & AC_ERR_ATA_BUS)
2024 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2025 AC_ERR_INVALID);
2026
2027 /* any real error trumps unknown error */
2028 if (qc->err_mask & ~AC_ERR_OTHER)
2029 qc->err_mask &= ~AC_ERR_OTHER;
2030
2031 /* SENSE_VALID trumps dev/unknown error and revalidation */
Tejun Heof90f0822007-10-26 16:12:41 +09002032 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
Tejun Heo022bdb02006-05-15 20:58:22 +09002033 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
Tejun Heo022bdb02006-05-15 20:58:22 +09002034
Tejun Heo03faab72008-03-27 19:14:24 +09002035 /* determine whether the command is worth retrying */
Tejun Heo534ead72010-01-14 16:18:09 +09002036 if (qc->flags & ATA_QCFLAG_IO ||
2037 (!(qc->err_mask & AC_ERR_INVALID) &&
2038 qc->err_mask != AC_ERR_DEV))
Tejun Heo03faab72008-03-27 19:14:24 +09002039 qc->flags |= ATA_QCFLAG_RETRY;
2040
Tejun Heo022bdb02006-05-15 20:58:22 +09002041 /* accumulate error info */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002042 ehc->i.dev = qc->dev;
Tejun Heo022bdb02006-05-15 20:58:22 +09002043 all_err_mask |= qc->err_mask;
2044 if (qc->flags & ATA_QCFLAG_IO)
Tejun Heo3884f7b2007-11-27 19:28:56 +09002045 eflags |= ATA_EFLAG_IS_IO;
Tejun Heo022bdb02006-05-15 20:58:22 +09002046 }
2047
Tejun Heoa20f33f2006-05-16 12:58:24 +09002048 /* enforce default EH actions */
Tejun Heob51e9e52006-06-29 01:29:30 +09002049 if (ap->pflags & ATA_PFLAG_FROZEN ||
Tejun Heoa20f33f2006-05-16 12:58:24 +09002050 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
Tejun Heocf480622008-01-24 00:05:14 +09002051 ehc->i.action |= ATA_EH_RESET;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002052 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2053 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
Tejun Heo4528e4d2006-07-08 20:17:26 +09002054 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heo022bdb02006-05-15 20:58:22 +09002055
Tejun Heodfcc1732007-10-31 10:17:05 +09002056 /* If we have offending qcs and the associated failed device,
2057 * perform per-dev EH action only on the offending device.
2058 */
Tejun Heo4528e4d2006-07-08 20:17:26 +09002059 if (ehc->i.dev) {
Tejun Heo4528e4d2006-07-08 20:17:26 +09002060 ehc->i.dev_action[ehc->i.dev->devno] |=
2061 ehc->i.action & ATA_EH_PERDEV_MASK;
2062 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
Tejun Heo47005f22006-06-19 18:27:23 +09002063 }
2064
Tejun Heo2695e362008-01-10 13:41:23 +09002065 /* propagate timeout to host link */
2066 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2067 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2068
2069 /* record error and consider speeding down */
Tejun Heodfcc1732007-10-31 10:17:05 +09002070 dev = ehc->i.dev;
Tejun Heo2695e362008-01-10 13:41:23 +09002071 if (!dev && ((ata_link_max_devices(link) == 1 &&
2072 ata_dev_enabled(link->device))))
2073 dev = link->device;
Tejun Heodfcc1732007-10-31 10:17:05 +09002074
Tejun Heo76326ac2007-11-27 19:28:59 +09002075 if (dev) {
2076 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2077 eflags |= ATA_EFLAG_DUBIOUS_XFER;
Tejun Heo3884f7b2007-11-27 19:28:56 +09002078 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
Tejun Heo76326ac2007-11-27 19:28:59 +09002079 }
Tejun Heodfcc1732007-10-31 10:17:05 +09002080
Tejun Heo022bdb02006-05-15 20:58:22 +09002081 DPRINTK("EXIT\n");
2082}
2083
2084/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002085 * ata_eh_autopsy - analyze error and determine recovery action
2086 * @ap: host port to perform autopsy on
2087 *
2088 * Analyze all links of @ap and determine why they failed and
2089 * which recovery actions are needed.
2090 *
2091 * LOCKING:
2092 * Kernel thread context (may sleep).
2093 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002094void ata_eh_autopsy(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002095{
2096 struct ata_link *link;
2097
Tejun Heo1eca4362008-11-03 20:03:17 +09002098 ata_for_each_link(link, ap, EDGE)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002099 ata_eh_link_autopsy(link);
Tejun Heo2695e362008-01-10 13:41:23 +09002100
Tejun Heob1c72912008-07-31 17:02:43 +09002101 /* Handle the frigging slave link. Autopsy is done similarly
2102 * but actions and flags are transferred over to the master
2103 * link and handled from there.
2104 */
2105 if (ap->slave_link) {
2106 struct ata_eh_context *mehc = &ap->link.eh_context;
2107 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2108
Tejun Heo848e4c62008-10-21 14:26:39 +09002109 /* transfer control flags from master to slave */
2110 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2111
2112 /* perform autopsy on the slave link */
Tejun Heob1c72912008-07-31 17:02:43 +09002113 ata_eh_link_autopsy(ap->slave_link);
2114
Tejun Heo848e4c62008-10-21 14:26:39 +09002115 /* transfer actions from slave to master and clear slave */
Tejun Heob1c72912008-07-31 17:02:43 +09002116 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2117 mehc->i.action |= sehc->i.action;
2118 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2119 mehc->i.flags |= sehc->i.flags;
2120 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2121 }
2122
Tejun Heo2695e362008-01-10 13:41:23 +09002123 /* Autopsy of fanout ports can affect host link autopsy.
2124 * Perform host link autopsy last.
2125 */
Tejun Heo071f44b2008-04-07 22:47:22 +09002126 if (sata_pmp_attached(ap))
Tejun Heo2695e362008-01-10 13:41:23 +09002127 ata_eh_link_autopsy(&ap->link);
Tejun Heo9b1e2652007-08-06 18:36:24 +09002128}
2129
2130/**
Robert Hancock65211482009-07-14 20:43:39 -06002131 * ata_get_cmd_descript - get description for ATA command
2132 * @command: ATA command code to get description for
2133 *
2134 * Return a textual description of the given command, or NULL if the
2135 * command is not known.
2136 *
2137 * LOCKING:
2138 * None
2139 */
2140const char *ata_get_cmd_descript(u8 command)
2141{
2142#ifdef CONFIG_ATA_VERBOSE_ERROR
2143 static const struct
2144 {
2145 u8 command;
2146 const char *text;
2147 } cmd_descr[] = {
2148 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2149 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2150 { ATA_CMD_STANDBY, "STANDBY" },
2151 { ATA_CMD_IDLE, "IDLE" },
2152 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2153 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2154 { ATA_CMD_NOP, "NOP" },
2155 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2156 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2157 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2158 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2159 { ATA_CMD_SERVICE, "SERVICE" },
2160 { ATA_CMD_READ, "READ DMA" },
2161 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2162 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2163 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2164 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2165 { ATA_CMD_WRITE, "WRITE DMA" },
2166 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2167 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2168 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2169 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2170 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2171 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2172 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2173 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2174 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2175 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2176 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2177 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2178 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2179 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2180 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2181 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2182 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2183 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2184 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2185 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2186 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2187 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2188 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2189 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2190 { ATA_CMD_SLEEP, "SLEEP" },
2191 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2192 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2193 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2194 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2195 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2196 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2197 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2198 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2199 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2200 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2201 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2202 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2203 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2204 { ATA_CMD_PMP_READ, "READ BUFFER" },
2205 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2206 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2207 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2208 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2209 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2210 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2211 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2212 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2213 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2214 { ATA_CMD_SMART, "SMART" },
2215 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2216 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2217 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2218 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2219 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2220 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2221 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2222 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2223 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2224 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2225 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2226 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2227 { ATA_CMD_RESTORE, "RECALIBRATE" },
2228 { 0, NULL } /* terminate list */
2229 };
2230
2231 unsigned int i;
2232 for (i = 0; cmd_descr[i].text; i++)
2233 if (cmd_descr[i].command == command)
2234 return cmd_descr[i].text;
2235#endif
2236
2237 return NULL;
2238}
2239
2240/**
Tejun Heo9b1e2652007-08-06 18:36:24 +09002241 * ata_eh_link_report - report error handling to user
Tejun Heo02607312007-08-06 18:36:23 +09002242 * @link: ATA link EH is going on
Tejun Heo022bdb02006-05-15 20:58:22 +09002243 *
2244 * Report EH to user.
2245 *
2246 * LOCKING:
2247 * None.
2248 */
Tejun Heo9b1e2652007-08-06 18:36:24 +09002249static void ata_eh_link_report(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09002250{
Tejun Heo02607312007-08-06 18:36:23 +09002251 struct ata_port *ap = link->ap;
2252 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002253 const char *frozen, *desc;
Tejun Heoa1e10f72007-08-18 13:28:49 +09002254 char tries_buf[6];
Tejun Heo022bdb02006-05-15 20:58:22 +09002255 int tag, nr_failed = 0;
2256
Tejun Heo94ff3d52007-10-09 14:57:56 +09002257 if (ehc->i.flags & ATA_EHI_QUIET)
2258 return;
2259
Tejun Heo022bdb02006-05-15 20:58:22 +09002260 desc = NULL;
2261 if (ehc->i.desc[0] != '\0')
2262 desc = ehc->i.desc;
2263
2264 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2265 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2266
Tejun Heob1c72912008-07-31 17:02:43 +09002267 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2268 ata_dev_phys_link(qc->dev) != link ||
Tejun Heoe027bd32007-10-26 16:19:26 +09002269 ((qc->flags & ATA_QCFLAG_QUIET) &&
2270 qc->err_mask == AC_ERR_DEV))
Tejun Heo022bdb02006-05-15 20:58:22 +09002271 continue;
2272 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2273 continue;
2274
2275 nr_failed++;
2276 }
2277
2278 if (!nr_failed && !ehc->i.err_mask)
2279 return;
2280
2281 frozen = "";
Tejun Heob51e9e52006-06-29 01:29:30 +09002282 if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo022bdb02006-05-15 20:58:22 +09002283 frozen = " frozen";
2284
Tejun Heoa1e10f72007-08-18 13:28:49 +09002285 memset(tries_buf, 0, sizeof(tries_buf));
2286 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2287 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2288 ap->eh_tries);
2289
Tejun Heo022bdb02006-05-15 20:58:22 +09002290 if (ehc->i.dev) {
Tejun Heoe8ee8452006-05-15 21:03:46 +09002291 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002292 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2293 ehc->i.err_mask, link->sactive, ehc->i.serror,
2294 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002295 if (desc)
Tejun Heob64bbc32007-07-16 14:29:39 +09002296 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002297 } else {
Tejun Heo02607312007-08-06 18:36:23 +09002298 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
Tejun Heoa1e10f72007-08-18 13:28:49 +09002299 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2300 ehc->i.err_mask, link->sactive, ehc->i.serror,
2301 ehc->i.action, frozen, tries_buf);
Tejun Heo022bdb02006-05-15 20:58:22 +09002302 if (desc)
Tejun Heo02607312007-08-06 18:36:23 +09002303 ata_link_printk(link, KERN_ERR, "%s\n", desc);
Tejun Heo022bdb02006-05-15 20:58:22 +09002304 }
2305
Robert Hancock65211482009-07-14 20:43:39 -06002306#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002307 if (ehc->i.serror)
Tejun Heoda0e21d2008-07-31 16:08:38 +09002308 ata_link_printk(link, KERN_ERR,
Robert Hancock1333e192007-10-02 11:22:02 -04002309 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2310 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2311 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2312 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2313 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2314 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2315 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2316 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2317 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2318 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2319 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2320 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2321 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2322 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2323 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2324 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2325 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002326 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002327#endif
Robert Hancock1333e192007-10-02 11:22:02 -04002328
Tejun Heo022bdb02006-05-15 20:58:22 +09002329 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2330 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
Tejun Heo8a937582006-11-14 22:36:12 +09002331 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
Tejun Heoabb6a882007-11-28 23:16:09 +09002332 const u8 *cdb = qc->cdb;
2333 char data_buf[20] = "";
2334 char cdb_buf[70] = "";
Tejun Heo022bdb02006-05-15 20:58:22 +09002335
Tejun Heo02607312007-08-06 18:36:23 +09002336 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
Tejun Heob1c72912008-07-31 17:02:43 +09002337 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
Tejun Heo022bdb02006-05-15 20:58:22 +09002338 continue;
2339
Tejun Heoabb6a882007-11-28 23:16:09 +09002340 if (qc->dma_dir != DMA_NONE) {
2341 static const char *dma_str[] = {
2342 [DMA_BIDIRECTIONAL] = "bidi",
2343 [DMA_TO_DEVICE] = "out",
2344 [DMA_FROM_DEVICE] = "in",
2345 };
2346 static const char *prot_str[] = {
2347 [ATA_PROT_PIO] = "pio",
2348 [ATA_PROT_DMA] = "dma",
2349 [ATA_PROT_NCQ] = "ncq",
Tejun Heo0dc36882007-12-18 16:34:43 -05002350 [ATAPI_PROT_PIO] = "pio",
2351 [ATAPI_PROT_DMA] = "dma",
Tejun Heoabb6a882007-11-28 23:16:09 +09002352 };
2353
2354 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2355 prot_str[qc->tf.protocol], qc->nbytes,
2356 dma_str[qc->dma_dir]);
2357 }
2358
Robert Hancock65211482009-07-14 20:43:39 -06002359 if (ata_is_atapi(qc->tf.protocol)) {
2360 if (qc->scsicmd)
2361 scsi_print_command(qc->scsicmd);
2362 else
2363 snprintf(cdb_buf, sizeof(cdb_buf),
Tejun Heoabb6a882007-11-28 23:16:09 +09002364 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2365 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2366 cdb[0], cdb[1], cdb[2], cdb[3],
2367 cdb[4], cdb[5], cdb[6], cdb[7],
2368 cdb[8], cdb[9], cdb[10], cdb[11],
2369 cdb[12], cdb[13], cdb[14], cdb[15]);
Robert Hancock65211482009-07-14 20:43:39 -06002370 } else {
2371 const char *descr = ata_get_cmd_descript(cmd->command);
2372 if (descr)
2373 ata_dev_printk(qc->dev, KERN_ERR,
2374 "failed command: %s\n", descr);
2375 }
Tejun Heoabb6a882007-11-28 23:16:09 +09002376
Tejun Heo8a937582006-11-14 22:36:12 +09002377 ata_dev_printk(qc->dev, KERN_ERR,
2378 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heoabb6a882007-11-28 23:16:09 +09002379 "tag %d%s\n %s"
Tejun Heo8a937582006-11-14 22:36:12 +09002380 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
Tejun Heo5335b722007-07-16 14:29:40 +09002381 "Emask 0x%x (%s)%s\n",
Tejun Heo8a937582006-11-14 22:36:12 +09002382 cmd->command, cmd->feature, cmd->nsect,
2383 cmd->lbal, cmd->lbam, cmd->lbah,
2384 cmd->hob_feature, cmd->hob_nsect,
2385 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
Tejun Heoabb6a882007-11-28 23:16:09 +09002386 cmd->device, qc->tag, data_buf, cdb_buf,
Tejun Heo8a937582006-11-14 22:36:12 +09002387 res->command, res->feature, res->nsect,
2388 res->lbal, res->lbam, res->lbah,
2389 res->hob_feature, res->hob_nsect,
2390 res->hob_lbal, res->hob_lbam, res->hob_lbah,
Tejun Heo5335b722007-07-16 14:29:40 +09002391 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2392 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002393
Robert Hancock65211482009-07-14 20:43:39 -06002394#ifdef CONFIG_ATA_VERBOSE_ERROR
Robert Hancock1333e192007-10-02 11:22:02 -04002395 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002396 ATA_ERR)) {
Robert Hancock1333e192007-10-02 11:22:02 -04002397 if (res->command & ATA_BUSY)
2398 ata_dev_printk(qc->dev, KERN_ERR,
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002399 "status: { Busy }\n");
Robert Hancock1333e192007-10-02 11:22:02 -04002400 else
2401 ata_dev_printk(qc->dev, KERN_ERR,
2402 "status: { %s%s%s%s}\n",
2403 res->command & ATA_DRDY ? "DRDY " : "",
2404 res->command & ATA_DF ? "DF " : "",
2405 res->command & ATA_DRQ ? "DRQ " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002406 res->command & ATA_ERR ? "ERR " : "");
Robert Hancock1333e192007-10-02 11:22:02 -04002407 }
2408
2409 if (cmd->command != ATA_CMD_PACKET &&
2410 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2411 ATA_ABORTED)))
2412 ata_dev_printk(qc->dev, KERN_ERR,
2413 "error: { %s%s%s%s}\n",
2414 res->feature & ATA_ICRC ? "ICRC " : "",
2415 res->feature & ATA_UNC ? "UNC " : "",
2416 res->feature & ATA_IDNF ? "IDNF " : "",
Jeff Garzik2dcb4072007-10-19 06:42:56 -04002417 res->feature & ATA_ABORTED ? "ABRT " : "");
Robert Hancock65211482009-07-14 20:43:39 -06002418#endif
Tejun Heo022bdb02006-05-15 20:58:22 +09002419 }
2420}
2421
Tejun Heo9b1e2652007-08-06 18:36:24 +09002422/**
2423 * ata_eh_report - report error handling to user
2424 * @ap: ATA port to report EH about
2425 *
2426 * Report EH to user.
2427 *
2428 * LOCKING:
2429 * None.
2430 */
Tejun Heofb7fd612007-09-23 13:14:12 +09002431void ata_eh_report(struct ata_port *ap)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002432{
2433 struct ata_link *link;
2434
Tejun Heo1eca4362008-11-03 20:03:17 +09002435 ata_for_each_link(link, ap, HOST_FIRST)
Tejun Heo9b1e2652007-08-06 18:36:24 +09002436 ata_eh_link_report(link);
2437}
2438
Tejun Heocc0680a2007-08-06 18:36:23 +09002439static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
Tejun Heob1c72912008-07-31 17:02:43 +09002440 unsigned int *classes, unsigned long deadline,
2441 bool clear_classes)
Tejun Heod87fa382006-05-31 18:28:24 +09002442{
Tejun Heof58229f2007-08-06 18:36:23 +09002443 struct ata_device *dev;
Tejun Heod87fa382006-05-31 18:28:24 +09002444
Tejun Heob1c72912008-07-31 17:02:43 +09002445 if (clear_classes)
Tejun Heo1eca4362008-11-03 20:03:17 +09002446 ata_for_each_dev(dev, link, ALL)
Tejun Heob1c72912008-07-31 17:02:43 +09002447 classes[dev->devno] = ATA_DEV_UNKNOWN;
Tejun Heod87fa382006-05-31 18:28:24 +09002448
Tejun Heof0465192008-05-19 01:15:08 +09002449 return reset(link, classes, deadline);
Tejun Heod87fa382006-05-31 18:28:24 +09002450}
2451
Tejun Heoae791c02007-09-23 13:14:12 +09002452static int ata_eh_followup_srst_needed(struct ata_link *link,
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002453 int rc, const unsigned int *classes)
Tejun Heo664faf02006-05-31 18:27:50 +09002454{
Tejun Heo45db2f62008-04-08 01:46:56 +09002455 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
Tejun Heoae791c02007-09-23 13:14:12 +09002456 return 0;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002457 if (rc == -EAGAIN)
2458 return 1;
Tejun Heo071f44b2008-04-07 22:47:22 +09002459 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
Tejun Heo3495de72007-09-23 13:19:53 +09002460 return 1;
Tejun Heo664faf02006-05-31 18:27:50 +09002461 return 0;
2462}
2463
Tejun Heofb7fd612007-09-23 13:14:12 +09002464int ata_eh_reset(struct ata_link *link, int classify,
2465 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2466 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09002467{
Tejun Heoafaa5c32007-10-09 15:06:10 +09002468 struct ata_port *ap = link->ap;
Tejun Heob1c72912008-07-31 17:02:43 +09002469 struct ata_link *slave = ap->slave_link;
Tejun Heo936fd732007-08-06 18:36:23 +09002470 struct ata_eh_context *ehc = &link->eh_context;
Bartlomiej Zolnierkiewicz705d2012009-07-26 16:21:01 +02002471 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
Tejun Heo664faf02006-05-31 18:27:50 +09002472 unsigned int *classes = ehc->classes;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002473 unsigned int lflags = link->flags;
Tejun Heo1cdaf532006-07-03 16:07:26 +09002474 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
Tejun Heod8af0eb2008-05-20 02:17:53 +09002475 int max_tries = 0, try = 0;
Tejun Heob1c72912008-07-31 17:02:43 +09002476 struct ata_link *failed_link;
Tejun Heof58229f2007-08-06 18:36:23 +09002477 struct ata_device *dev;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002478 unsigned long deadline, now;
Tejun Heo022bdb02006-05-15 20:58:22 +09002479 ata_reset_fn_t reset;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002480 unsigned long flags;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002481 u32 sstatus;
Tejun Heob1c72912008-07-31 17:02:43 +09002482 int nr_unknown, rc;
Tejun Heo022bdb02006-05-15 20:58:22 +09002483
Tejun Heo932648b2008-05-19 01:15:06 +09002484 /*
2485 * Prepare to reset
2486 */
Tejun Heod8af0eb2008-05-20 02:17:53 +09002487 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2488 max_tries++;
Tejun Heo05944bd2008-08-13 20:19:09 +09002489 if (link->flags & ATA_LFLAG_NO_HRST)
2490 hardreset = NULL;
2491 if (link->flags & ATA_LFLAG_NO_SRST)
2492 softreset = NULL;
Tejun Heod8af0eb2008-05-20 02:17:53 +09002493
Tejun Heo19b72322008-11-04 17:08:40 +09002494 /* make sure each reset attemp is at least COOL_DOWN apart */
2495 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2496 now = jiffies;
2497 WARN_ON(time_after(ehc->last_reset, now));
2498 deadline = ata_deadline(ehc->last_reset,
2499 ATA_EH_RESET_COOL_DOWN);
2500 if (time_before(now, deadline))
2501 schedule_timeout_uninterruptible(deadline - now);
2502 }
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002503
Tejun Heoafaa5c32007-10-09 15:06:10 +09002504 spin_lock_irqsave(ap->lock, flags);
2505 ap->pflags |= ATA_PFLAG_RESETTING;
2506 spin_unlock_irqrestore(ap->lock, flags);
2507
Tejun Heocf480622008-01-24 00:05:14 +09002508 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heo13abf502006-07-10 23:18:46 +09002509
Tejun Heo1eca4362008-11-03 20:03:17 +09002510 ata_for_each_dev(dev, link, ALL) {
Tejun Heocdeab112007-10-29 16:41:09 +09002511 /* If we issue an SRST then an ATA drive (not ATAPI)
2512 * may change configuration and be in PIO0 timing. If
2513 * we do a hard reset (or are coming from power on)
2514 * this is true for ATA or ATAPI. Until we've set a
2515 * suitable controller mode we should not touch the
2516 * bus as we may be talking too fast.
2517 */
2518 dev->pio_mode = XFER_PIO_0;
2519
2520 /* If the controller has a pio mode setup function
2521 * then use it to set the chipset to rights. Don't
2522 * touch the DMA setup as that will be dealt with when
2523 * configuring devices.
2524 */
2525 if (ap->ops->set_piomode)
2526 ap->ops->set_piomode(ap, dev);
2527 }
2528
Tejun Heocf480622008-01-24 00:05:14 +09002529 /* prefer hardreset */
Tejun Heo932648b2008-05-19 01:15:06 +09002530 reset = NULL;
Tejun Heocf480622008-01-24 00:05:14 +09002531 ehc->i.action &= ~ATA_EH_RESET;
2532 if (hardreset) {
2533 reset = hardreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002534 ehc->i.action |= ATA_EH_HARDRESET;
Tejun Heo4f7faa32008-01-30 18:18:26 +09002535 } else if (softreset) {
Tejun Heocf480622008-01-24 00:05:14 +09002536 reset = softreset;
Tejun Heoa6740502008-07-31 16:07:04 +09002537 ehc->i.action |= ATA_EH_SOFTRESET;
Tejun Heocf480622008-01-24 00:05:14 +09002538 }
Tejun Heof5914a42006-05-31 18:27:48 +09002539
2540 if (prereset) {
Tejun Heob1c72912008-07-31 17:02:43 +09002541 unsigned long deadline = ata_deadline(jiffies,
2542 ATA_EH_PRERESET_TIMEOUT);
2543
2544 if (slave) {
2545 sehc->i.action &= ~ATA_EH_RESET;
2546 sehc->i.action |= ehc->i.action;
2547 }
2548
2549 rc = prereset(link, deadline);
2550
2551 /* If present, do prereset on slave link too. Reset
2552 * is skipped iff both master and slave links report
2553 * -ENOENT or clear ATA_EH_RESET.
2554 */
2555 if (slave && (rc == 0 || rc == -ENOENT)) {
2556 int tmp;
2557
2558 tmp = prereset(slave, deadline);
2559 if (tmp != -ENOENT)
2560 rc = tmp;
2561
2562 ehc->i.action |= sehc->i.action;
2563 }
2564
Tejun Heof5914a42006-05-31 18:27:48 +09002565 if (rc) {
Alan Coxc9619222006-09-26 17:53:38 +01002566 if (rc == -ENOENT) {
Tejun Heocc0680a2007-08-06 18:36:23 +09002567 ata_link_printk(link, KERN_DEBUG,
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002568 "port disabled. ignoring.\n");
Tejun Heocf480622008-01-24 00:05:14 +09002569 ehc->i.action &= ~ATA_EH_RESET;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002570
Tejun Heo1eca4362008-11-03 20:03:17 +09002571 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09002572 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heo4aa9ab62007-03-12 17:24:08 +09002573
2574 rc = 0;
Alan Coxc9619222006-09-26 17:53:38 +01002575 } else
Tejun Heocc0680a2007-08-06 18:36:23 +09002576 ata_link_printk(link, KERN_ERR,
Tejun Heof5914a42006-05-31 18:27:48 +09002577 "prereset failed (errno=%d)\n", rc);
Tejun Heofccb6ea2007-07-16 14:29:41 +09002578 goto out;
Tejun Heof5914a42006-05-31 18:27:48 +09002579 }
Tejun Heof5914a42006-05-31 18:27:48 +09002580
Tejun Heo932648b2008-05-19 01:15:06 +09002581 /* prereset() might have cleared ATA_EH_RESET. If so,
Tejun Heod6515e62009-03-04 15:59:30 +09002582 * bang classes, thaw and return.
Tejun Heo932648b2008-05-19 01:15:06 +09002583 */
2584 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
Tejun Heo1eca4362008-11-03 20:03:17 +09002585 ata_for_each_dev(dev, link, ALL)
Tejun Heo932648b2008-05-19 01:15:06 +09002586 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heod6515e62009-03-04 15:59:30 +09002587 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2588 ata_is_host_link(link))
2589 ata_eh_thaw_port(ap);
Tejun Heo932648b2008-05-19 01:15:06 +09002590 rc = 0;
2591 goto out;
2592 }
Tejun Heof5914a42006-05-31 18:27:48 +09002593 }
2594
Tejun Heo022bdb02006-05-15 20:58:22 +09002595 retry:
Tejun Heo932648b2008-05-19 01:15:06 +09002596 /*
2597 * Perform reset
2598 */
Tejun Heodc98c322008-05-19 01:15:07 +09002599 if (ata_is_host_link(link))
2600 ata_eh_freeze_port(ap);
2601
Tejun Heo341c2c92008-05-20 02:17:51 +09002602 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
Tejun Heo31daabd2007-02-02 16:50:52 +09002603
Tejun Heo932648b2008-05-19 01:15:06 +09002604 if (reset) {
2605 if (verbose)
2606 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2607 reset == softreset ? "soft" : "hard");
Tejun Heo022bdb02006-05-15 20:58:22 +09002608
Tejun Heo932648b2008-05-19 01:15:06 +09002609 /* mark that this EH session started with reset */
Tejun Heo19b72322008-11-04 17:08:40 +09002610 ehc->last_reset = jiffies;
Tejun Heo932648b2008-05-19 01:15:06 +09002611 if (reset == hardreset)
2612 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2613 else
2614 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
Tejun Heo022bdb02006-05-15 20:58:22 +09002615
Tejun Heob1c72912008-07-31 17:02:43 +09002616 rc = ata_do_reset(link, reset, classes, deadline, true);
2617 if (rc && rc != -EAGAIN) {
2618 failed_link = link;
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002619 goto fail;
Tejun Heob1c72912008-07-31 17:02:43 +09002620 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002621
Tejun Heob1c72912008-07-31 17:02:43 +09002622 /* hardreset slave link if existent */
2623 if (slave && reset == hardreset) {
2624 int tmp;
2625
2626 if (verbose)
2627 ata_link_printk(slave, KERN_INFO,
2628 "hard resetting link\n");
2629
2630 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2631 tmp = ata_do_reset(slave, reset, classes, deadline,
2632 false);
2633 switch (tmp) {
2634 case -EAGAIN:
2635 rc = -EAGAIN;
2636 case 0:
2637 break;
2638 default:
2639 failed_link = slave;
2640 rc = tmp;
2641 goto fail;
2642 }
2643 }
2644
2645 /* perform follow-up SRST if necessary */
Tejun Heo932648b2008-05-19 01:15:06 +09002646 if (reset == hardreset &&
Tejun Heo5dbfc9c2008-07-31 16:08:02 +09002647 ata_eh_followup_srst_needed(link, rc, classes)) {
Tejun Heo932648b2008-05-19 01:15:06 +09002648 reset = softreset;
Tejun Heo664faf02006-05-31 18:27:50 +09002649
Tejun Heo932648b2008-05-19 01:15:06 +09002650 if (!reset) {
2651 ata_link_printk(link, KERN_ERR,
2652 "follow-up softreset required "
2653 "but no softreset avaliable\n");
Tejun Heob1c72912008-07-31 17:02:43 +09002654 failed_link = link;
Tejun Heo932648b2008-05-19 01:15:06 +09002655 rc = -EINVAL;
2656 goto fail;
2657 }
2658
2659 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002660 rc = ata_do_reset(link, reset, classes, deadline, true);
Tejun Heofe2c4d02009-07-08 12:16:37 +09002661 if (rc) {
2662 failed_link = link;
2663 goto fail;
2664 }
Tejun Heo664faf02006-05-31 18:27:50 +09002665 }
Tejun Heo932648b2008-05-19 01:15:06 +09002666 } else {
2667 if (verbose)
2668 ata_link_printk(link, KERN_INFO, "no reset method "
2669 "available, skipping reset\n");
2670 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2671 lflags |= ATA_LFLAG_ASSUME_ATA;
Tejun Heo664faf02006-05-31 18:27:50 +09002672 }
2673
Tejun Heo932648b2008-05-19 01:15:06 +09002674 /*
2675 * Post-reset processing
2676 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002677 ata_for_each_dev(dev, link, ALL) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002678 /* After the reset, the device state is PIO 0 and the
2679 * controller state is undefined. Reset also wakes up
2680 * drives from sleeping mode.
2681 */
2682 dev->pio_mode = XFER_PIO_0;
2683 dev->flags &= ~ATA_DFLAG_SLEEPING;
Tejun Heo664faf02006-05-31 18:27:50 +09002684
Tejun Heo3b761d32009-10-06 17:08:40 +09002685 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2686 continue;
2687
2688 /* apply class override */
2689 if (lflags & ATA_LFLAG_ASSUME_ATA)
2690 classes[dev->devno] = ATA_DEV_ATA;
2691 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2692 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
Tejun Heo022bdb02006-05-15 20:58:22 +09002693 }
2694
Tejun Heo416dc9e2007-10-31 10:17:03 +09002695 /* record current link speed */
2696 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2697 link->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heob1c72912008-07-31 17:02:43 +09002698 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2699 slave->sata_spd = (sstatus >> 4) & 0xf;
Tejun Heo008a7892007-07-16 14:29:40 +09002700
Tejun Heodc98c322008-05-19 01:15:07 +09002701 /* thaw the port */
2702 if (ata_is_host_link(link))
2703 ata_eh_thaw_port(ap);
2704
Tejun Heof0465192008-05-19 01:15:08 +09002705 /* postreset() should clear hardware SError. Although SError
2706 * is cleared during link resume, clearing SError here is
2707 * necessary as some PHYs raise hotplug events after SRST.
2708 * This introduces race condition where hotplug occurs between
2709 * reset and here. This race is mediated by cross checking
2710 * link onlineness and classification result later.
2711 */
Tejun Heob1c72912008-07-31 17:02:43 +09002712 if (postreset) {
Tejun Heo416dc9e2007-10-31 10:17:03 +09002713 postreset(link, classes);
Tejun Heob1c72912008-07-31 17:02:43 +09002714 if (slave)
2715 postreset(slave, classes);
2716 }
Tejun Heo20952b62006-05-31 18:27:23 +09002717
Tejun Heo1e641062009-07-17 11:58:33 +09002718 /*
2719 * Some controllers can't be frozen very well and may set
2720 * spuruious error conditions during reset. Clear accumulated
2721 * error information. As reset is the final recovery action,
2722 * nothing is lost by doing this.
2723 */
Tejun Heof0465192008-05-19 01:15:08 +09002724 spin_lock_irqsave(link->ap->lock, flags);
Tejun Heo1e641062009-07-17 11:58:33 +09002725 memset(&link->eh_info, 0, sizeof(link->eh_info));
Tejun Heob1c72912008-07-31 17:02:43 +09002726 if (slave)
Tejun Heo1e641062009-07-17 11:58:33 +09002727 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2728 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
Tejun Heof0465192008-05-19 01:15:08 +09002729 spin_unlock_irqrestore(link->ap->lock, flags);
2730
Tejun Heo3b761d32009-10-06 17:08:40 +09002731 /*
2732 * Make sure onlineness and classification result correspond.
Tejun Heof0465192008-05-19 01:15:08 +09002733 * Hotplug could have happened during reset and some
2734 * controllers fail to wait while a drive is spinning up after
2735 * being hotplugged causing misdetection. By cross checking
Tejun Heo3b761d32009-10-06 17:08:40 +09002736 * link on/offlineness and classification result, those
2737 * conditions can be reliably detected and retried.
Tejun Heof0465192008-05-19 01:15:08 +09002738 */
Tejun Heob1c72912008-07-31 17:02:43 +09002739 nr_unknown = 0;
Tejun Heo1eca4362008-11-03 20:03:17 +09002740 ata_for_each_dev(dev, link, ALL) {
Tejun Heo3b761d32009-10-06 17:08:40 +09002741 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2742 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2743 ata_dev_printk(dev, KERN_DEBUG, "link online "
2744 "but device misclassifed\n");
2745 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002746 nr_unknown++;
Tejun Heo3b761d32009-10-06 17:08:40 +09002747 }
2748 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2749 if (ata_class_enabled(classes[dev->devno]))
2750 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2751 "clearing class %d to NONE\n",
2752 classes[dev->devno]);
2753 classes[dev->devno] = ATA_DEV_NONE;
2754 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2755 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2756 "clearing UNKNOWN to NONE\n");
2757 classes[dev->devno] = ATA_DEV_NONE;
Tejun Heob1c72912008-07-31 17:02:43 +09002758 }
Tejun Heof0465192008-05-19 01:15:08 +09002759 }
2760
Tejun Heob1c72912008-07-31 17:02:43 +09002761 if (classify && nr_unknown) {
Tejun Heof0465192008-05-19 01:15:08 +09002762 if (try < max_tries) {
2763 ata_link_printk(link, KERN_WARNING, "link online but "
Tejun Heo3b761d32009-10-06 17:08:40 +09002764 "%d devices misclassified, retrying\n",
2765 nr_unknown);
Tejun Heob1c72912008-07-31 17:02:43 +09002766 failed_link = link;
Tejun Heof0465192008-05-19 01:15:08 +09002767 rc = -EAGAIN;
2768 goto fail;
2769 }
2770 ata_link_printk(link, KERN_WARNING,
Tejun Heo3b761d32009-10-06 17:08:40 +09002771 "link online but %d devices misclassified, "
2772 "device detection might fail\n", nr_unknown);
Tejun Heof0465192008-05-19 01:15:08 +09002773 }
2774
Tejun Heo416dc9e2007-10-31 10:17:03 +09002775 /* reset successful, schedule revalidation */
Tejun Heocf480622008-01-24 00:05:14 +09002776 ata_eh_done(link, NULL, ATA_EH_RESET);
Tejun Heob1c72912008-07-31 17:02:43 +09002777 if (slave)
2778 ata_eh_done(slave, NULL, ATA_EH_RESET);
Tejun Heo19b72322008-11-04 17:08:40 +09002779 ehc->last_reset = jiffies; /* update to completion time */
Tejun Heo416dc9e2007-10-31 10:17:03 +09002780 ehc->i.action |= ATA_EH_REVALIDATE;
Tejun Heoae791c02007-09-23 13:14:12 +09002781
Tejun Heo416dc9e2007-10-31 10:17:03 +09002782 rc = 0;
Tejun Heofccb6ea2007-07-16 14:29:41 +09002783 out:
2784 /* clear hotplug flag */
2785 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heob1c72912008-07-31 17:02:43 +09002786 if (slave)
2787 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
Tejun Heoafaa5c32007-10-09 15:06:10 +09002788
2789 spin_lock_irqsave(ap->lock, flags);
2790 ap->pflags &= ~ATA_PFLAG_RESETTING;
2791 spin_unlock_irqrestore(ap->lock, flags);
2792
Tejun Heo022bdb02006-05-15 20:58:22 +09002793 return rc;
Tejun Heo416dc9e2007-10-31 10:17:03 +09002794
2795 fail:
Tejun Heo5958e302008-04-07 22:47:20 +09002796 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2797 if (!ata_is_host_link(link) &&
2798 sata_scr_read(link, SCR_STATUS, &sstatus))
2799 rc = -ERESTART;
2800
Tejun Heo416dc9e2007-10-31 10:17:03 +09002801 if (rc == -ERESTART || try >= max_tries)
2802 goto out;
2803
2804 now = jiffies;
2805 if (time_before(now, deadline)) {
2806 unsigned long delta = deadline - now;
2807
Tejun Heob1c72912008-07-31 17:02:43 +09002808 ata_link_printk(failed_link, KERN_WARNING,
Tejun Heo0a2c0f52008-05-20 02:17:52 +09002809 "reset failed (errno=%d), retrying in %u secs\n",
2810 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
Tejun Heo416dc9e2007-10-31 10:17:03 +09002811
2812 while (delta)
2813 delta = schedule_timeout_uninterruptible(delta);
2814 }
2815
Tejun Heob1c72912008-07-31 17:02:43 +09002816 if (try == max_tries - 1) {
Tejun Heoa07d4992009-01-29 20:31:33 +09002817 sata_down_spd_limit(link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002818 if (slave)
Tejun Heoa07d4992009-01-29 20:31:33 +09002819 sata_down_spd_limit(slave, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002820 } else if (rc == -EPIPE)
Tejun Heoa07d4992009-01-29 20:31:33 +09002821 sata_down_spd_limit(failed_link, 0);
Tejun Heob1c72912008-07-31 17:02:43 +09002822
Tejun Heo416dc9e2007-10-31 10:17:03 +09002823 if (hardreset)
2824 reset = hardreset;
2825 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09002826}
2827
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002828static inline void ata_eh_pull_park_action(struct ata_port *ap)
2829{
2830 struct ata_link *link;
2831 struct ata_device *dev;
2832 unsigned long flags;
2833
2834 /*
2835 * This function can be thought of as an extended version of
2836 * ata_eh_about_to_do() specially crafted to accommodate the
2837 * requirements of ATA_EH_PARK handling. Since the EH thread
2838 * does not leave the do {} while () loop in ata_eh_recover as
2839 * long as the timeout for a park request to *one* device on
2840 * the port has not expired, and since we still want to pick
2841 * up park requests to other devices on the same port or
2842 * timeout updates for the same device, we have to pull
2843 * ATA_EH_PARK actions from eh_info into eh_context.i
2844 * ourselves at the beginning of each pass over the loop.
2845 *
2846 * Additionally, all write accesses to &ap->park_req_pending
2847 * through INIT_COMPLETION() (see below) or complete_all()
2848 * (see ata_scsi_park_store()) are protected by the host lock.
2849 * As a result we have that park_req_pending.done is zero on
2850 * exit from this function, i.e. when ATA_EH_PARK actions for
2851 * *all* devices on port ap have been pulled into the
2852 * respective eh_context structs. If, and only if,
2853 * park_req_pending.done is non-zero by the time we reach
2854 * wait_for_completion_timeout(), another ATA_EH_PARK action
2855 * has been scheduled for at least one of the devices on port
2856 * ap and we have to cycle over the do {} while () loop in
2857 * ata_eh_recover() again.
2858 */
2859
2860 spin_lock_irqsave(ap->lock, flags);
2861 INIT_COMPLETION(ap->park_req_pending);
Tejun Heo1eca4362008-11-03 20:03:17 +09002862 ata_for_each_link(link, ap, EDGE) {
2863 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02002864 struct ata_eh_info *ehi = &link->eh_info;
2865
2866 link->eh_context.i.dev_action[dev->devno] |=
2867 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2868 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2869 }
2870 }
2871 spin_unlock_irqrestore(ap->lock, flags);
2872}
2873
2874static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2875{
2876 struct ata_eh_context *ehc = &dev->link->eh_context;
2877 struct ata_taskfile tf;
2878 unsigned int err_mask;
2879
2880 ata_tf_init(dev, &tf);
2881 if (park) {
2882 ehc->unloaded_mask |= 1 << dev->devno;
2883 tf.command = ATA_CMD_IDLEIMMEDIATE;
2884 tf.feature = 0x44;
2885 tf.lbal = 0x4c;
2886 tf.lbam = 0x4e;
2887 tf.lbah = 0x55;
2888 } else {
2889 ehc->unloaded_mask &= ~(1 << dev->devno);
2890 tf.command = ATA_CMD_CHK_POWER;
2891 }
2892
2893 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2894 tf.protocol |= ATA_PROT_NODATA;
2895 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2896 if (park && (err_mask || tf.lbal != 0xc4)) {
2897 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2898 ehc->unloaded_mask &= ~(1 << dev->devno);
2899 }
2900}
2901
Tejun Heo02607312007-08-06 18:36:23 +09002902static int ata_eh_revalidate_and_attach(struct ata_link *link,
Tejun Heo084fe632006-05-31 18:28:03 +09002903 struct ata_device **r_failed_dev)
Tejun Heo022bdb02006-05-15 20:58:22 +09002904{
Tejun Heo02607312007-08-06 18:36:23 +09002905 struct ata_port *ap = link->ap;
2906 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09002907 struct ata_device *dev;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002908 unsigned int new_mask = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09002909 unsigned long flags;
Tejun Heof58229f2007-08-06 18:36:23 +09002910 int rc = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09002911
2912 DPRINTK("ENTER\n");
2913
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002914 /* For PATA drive side cable detection to work, IDENTIFY must
2915 * be done backwards such that PDIAG- is released by the slave
2916 * device before the master device is identified.
2917 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002918 ata_for_each_dev(dev, link, ALL_REVERSE) {
Tejun Heof58229f2007-08-06 18:36:23 +09002919 unsigned int action = ata_eh_dev_action(dev);
2920 unsigned int readid_flags = 0;
Tejun Heo47005f22006-06-19 18:27:23 +09002921
Tejun Heobff04642006-11-10 18:08:10 +09002922 if (ehc->i.flags & ATA_EHI_DID_RESET)
2923 readid_flags |= ATA_READID_POSTRESET;
2924
Tejun Heo9666f402007-05-04 21:27:47 +02002925 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
Tejun Heo633273a2007-09-23 13:19:54 +09002926 WARN_ON(dev->class == ATA_DEV_PMP);
2927
Tejun Heob1c72912008-07-31 17:02:43 +09002928 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
Tejun Heo022bdb02006-05-15 20:58:22 +09002929 rc = -EIO;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002930 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002931 }
2932
Tejun Heo02607312007-08-06 18:36:23 +09002933 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
Tejun Heo422c9da2007-09-23 13:14:12 +09002934 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2935 readid_flags);
Tejun Heo022bdb02006-05-15 20:58:22 +09002936 if (rc)
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002937 goto err;
Tejun Heo022bdb02006-05-15 20:58:22 +09002938
Tejun Heo02607312007-08-06 18:36:23 +09002939 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
Tejun Heo47005f22006-06-19 18:27:23 +09002940
Tejun Heobaa1e782006-11-01 18:39:27 +09002941 /* Configuration may have changed, reconfigure
2942 * transfer mode.
2943 */
2944 ehc->i.flags |= ATA_EHI_SETMODE;
2945
zhao, forrest3057ac32006-06-12 12:01:34 +08002946 /* schedule the scsi_rescan_device() here */
2947 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
Tejun Heo084fe632006-05-31 18:28:03 +09002948 } else if (dev->class == ATA_DEV_UNKNOWN &&
2949 ehc->tries[dev->devno] &&
2950 ata_class_enabled(ehc->classes[dev->devno])) {
Tejun Heo842faa62009-05-10 01:06:54 +09002951 /* Temporarily set dev->class, it will be
2952 * permanently set once all configurations are
2953 * complete. This is necessary because new
2954 * device configuration is done in two
2955 * separate loops.
2956 */
Tejun Heo084fe632006-05-31 18:28:03 +09002957 dev->class = ehc->classes[dev->devno];
2958
Tejun Heo633273a2007-09-23 13:19:54 +09002959 if (dev->class == ATA_DEV_PMP)
2960 rc = sata_pmp_attach(dev);
2961 else
2962 rc = ata_dev_read_id(dev, &dev->class,
2963 readid_flags, dev->id);
Tejun Heo842faa62009-05-10 01:06:54 +09002964
2965 /* read_id might have changed class, store and reset */
2966 ehc->classes[dev->devno] = dev->class;
2967 dev->class = ATA_DEV_UNKNOWN;
2968
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002969 switch (rc) {
2970 case 0:
Tejun Heo99cf6102009-01-29 20:31:32 +09002971 /* clear error info accumulated during probe */
2972 ata_ering_clear(&dev->ering);
Tejun Heof58229f2007-08-06 18:36:23 +09002973 new_mask |= 1 << dev->devno;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002974 break;
2975 case -ENOENT:
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002976 /* IDENTIFY was issued to non-existent
2977 * device. No need to reset. Just
Tejun Heo842faa62009-05-10 01:06:54 +09002978 * thaw and ignore the device.
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002979 */
2980 ata_eh_thaw_port(ap);
Tejun Heo084fe632006-05-31 18:28:03 +09002981 break;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002982 default:
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002983 goto err;
Tejun Heo55a8e2c2006-11-10 18:08:10 +09002984 }
Tejun Heo022bdb02006-05-15 20:58:22 +09002985 }
2986 }
2987
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002988 /* PDIAG- should have been released, ask cable type if post-reset */
Tejun Heo33267322008-02-13 09:15:09 +09002989 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2990 if (ap->ops->cable_detect)
2991 ap->cbl = ap->ops->cable_detect(ap);
2992 ata_force_cbl(ap);
2993 }
Tejun Heoc1c4e8d2007-04-23 02:05:53 +09002994
Tejun Heo8c3c52a2007-03-22 22:24:19 +09002995 /* Configure new devices forward such that user doesn't see
2996 * device detection messages backwards.
2997 */
Tejun Heo1eca4362008-11-03 20:03:17 +09002998 ata_for_each_dev(dev, link, ALL) {
Tejun Heo4f7c2872009-10-15 23:37:32 +09002999 if (!(new_mask & (1 << dev->devno)))
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003000 continue;
3001
Tejun Heo842faa62009-05-10 01:06:54 +09003002 dev->class = ehc->classes[dev->devno];
3003
Tejun Heo4f7c2872009-10-15 23:37:32 +09003004 if (dev->class == ATA_DEV_PMP)
3005 continue;
3006
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003007 ehc->i.flags |= ATA_EHI_PRINTINFO;
3008 rc = ata_dev_configure(dev);
3009 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
Tejun Heo842faa62009-05-10 01:06:54 +09003010 if (rc) {
3011 dev->class = ATA_DEV_UNKNOWN;
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003012 goto err;
Tejun Heo842faa62009-05-10 01:06:54 +09003013 }
Tejun Heo8c3c52a2007-03-22 22:24:19 +09003014
3015 spin_lock_irqsave(ap->lock, flags);
3016 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3017 spin_unlock_irqrestore(ap->lock, flags);
3018
3019 /* new device discovered, configure xfermode */
3020 ehc->i.flags |= ATA_EHI_SETMODE;
3021 }
3022
3023 return 0;
3024
3025 err:
3026 *r_failed_dev = dev;
3027 DPRINTK("EXIT rc=%d\n", rc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003028 return rc;
3029}
3030
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003031/**
3032 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3033 * @link: link on which timings will be programmed
Martin Olsson98a17082009-04-22 18:21:29 +02003034 * @r_failed_dev: out parameter for failed device
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003035 *
3036 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3037 * ata_set_mode() fails, pointer to the failing device is
3038 * returned in @r_failed_dev.
3039 *
3040 * LOCKING:
3041 * PCI/etc. bus probe sem.
3042 *
3043 * RETURNS:
3044 * 0 on success, negative errno otherwise
3045 */
3046int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3047{
3048 struct ata_port *ap = link->ap;
Tejun Heo00115e02007-11-27 19:28:58 +09003049 struct ata_device *dev;
3050 int rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003051
Tejun Heo76326ac2007-11-27 19:28:59 +09003052 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
Tejun Heo1eca4362008-11-03 20:03:17 +09003053 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo76326ac2007-11-27 19:28:59 +09003054 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3055 struct ata_ering_entry *ent;
3056
3057 ent = ata_ering_top(&dev->ering);
3058 if (ent)
3059 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3060 }
3061 }
3062
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003063 /* has private set_mode? */
3064 if (ap->ops->set_mode)
Tejun Heo00115e02007-11-27 19:28:58 +09003065 rc = ap->ops->set_mode(link, r_failed_dev);
3066 else
3067 rc = ata_do_set_mode(link, r_failed_dev);
3068
3069 /* if transfer mode has changed, set DUBIOUS_XFER on device */
Tejun Heo1eca4362008-11-03 20:03:17 +09003070 ata_for_each_dev(dev, link, ENABLED) {
Tejun Heo00115e02007-11-27 19:28:58 +09003071 struct ata_eh_context *ehc = &link->eh_context;
3072 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3073 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3074
3075 if (dev->xfer_mode != saved_xfer_mode ||
3076 ata_ncq_enabled(dev) != saved_ncq)
3077 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3078 }
3079
3080 return rc;
Tejun Heo6f1d1e32007-11-27 19:28:55 +09003081}
3082
Tejun Heo11fc33d2008-08-30 14:20:01 +02003083/**
3084 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3085 * @dev: ATAPI device to clear UA for
3086 *
3087 * Resets and other operations can make an ATAPI device raise
3088 * UNIT ATTENTION which causes the next operation to fail. This
3089 * function clears UA.
3090 *
3091 * LOCKING:
3092 * EH context (may sleep).
3093 *
3094 * RETURNS:
3095 * 0 on success, -errno on failure.
3096 */
3097static int atapi_eh_clear_ua(struct ata_device *dev)
3098{
3099 int i;
3100
3101 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
Tejun Heob5357082009-03-02 18:55:16 +09003102 u8 *sense_buffer = dev->link->ap->sector_buf;
Tejun Heo11fc33d2008-08-30 14:20:01 +02003103 u8 sense_key = 0;
3104 unsigned int err_mask;
3105
3106 err_mask = atapi_eh_tur(dev, &sense_key);
3107 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3108 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3109 "failed (err_mask=0x%x)\n", err_mask);
3110 return -EIO;
3111 }
3112
3113 if (!err_mask || sense_key != UNIT_ATTENTION)
3114 return 0;
3115
3116 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3117 if (err_mask) {
3118 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3119 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3120 return -EIO;
3121 }
3122 }
3123
3124 ata_dev_printk(dev, KERN_WARNING,
3125 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3126
3127 return 0;
3128}
3129
Tejun Heo6013efd2009-11-19 15:36:45 +09003130/**
3131 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3132 * @dev: ATA device which may need FLUSH retry
3133 *
3134 * If @dev failed FLUSH, it needs to be reported upper layer
3135 * immediately as it means that @dev failed to remap and already
3136 * lost at least a sector and further FLUSH retrials won't make
3137 * any difference to the lost sector. However, if FLUSH failed
3138 * for other reasons, for example transmission error, FLUSH needs
3139 * to be retried.
3140 *
3141 * This function determines whether FLUSH failure retry is
3142 * necessary and performs it if so.
3143 *
3144 * RETURNS:
3145 * 0 if EH can continue, -errno if EH needs to be repeated.
3146 */
3147static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3148{
3149 struct ata_link *link = dev->link;
3150 struct ata_port *ap = link->ap;
3151 struct ata_queued_cmd *qc;
3152 struct ata_taskfile tf;
3153 unsigned int err_mask;
3154 int rc = 0;
3155
3156 /* did flush fail for this device? */
3157 if (!ata_tag_valid(link->active_tag))
3158 return 0;
3159
3160 qc = __ata_qc_from_tag(ap, link->active_tag);
3161 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3162 qc->tf.command != ATA_CMD_FLUSH))
3163 return 0;
3164
3165 /* if the device failed it, it should be reported to upper layers */
3166 if (qc->err_mask & AC_ERR_DEV)
3167 return 0;
3168
3169 /* flush failed for some other reason, give it another shot */
3170 ata_tf_init(dev, &tf);
3171
3172 tf.command = qc->tf.command;
3173 tf.flags |= ATA_TFLAG_DEVICE;
3174 tf.protocol = ATA_PROT_NODATA;
3175
3176 ata_dev_printk(dev, KERN_WARNING, "retrying FLUSH 0x%x Emask 0x%x\n",
3177 tf.command, qc->err_mask);
3178
3179 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3180 if (!err_mask) {
3181 /*
3182 * FLUSH is complete but there's no way to
3183 * successfully complete a failed command from EH.
3184 * Making sure retry is allowed at least once and
3185 * retrying it should do the trick - whatever was in
3186 * the cache is already on the platter and this won't
3187 * cause infinite loop.
3188 */
3189 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3190 } else {
3191 ata_dev_printk(dev, KERN_WARNING, "FLUSH failed Emask 0x%x\n",
3192 err_mask);
3193 rc = -EIO;
3194
3195 /* if device failed it, report it to upper layers */
3196 if (err_mask & AC_ERR_DEV) {
3197 qc->err_mask |= AC_ERR_DEV;
3198 qc->result_tf = tf;
3199 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3200 rc = 0;
3201 }
3202 }
3203 return rc;
3204}
3205
Tejun Heo02607312007-08-06 18:36:23 +09003206static int ata_link_nr_enabled(struct ata_link *link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003207{
Tejun Heof58229f2007-08-06 18:36:23 +09003208 struct ata_device *dev;
3209 int cnt = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003210
Tejun Heo1eca4362008-11-03 20:03:17 +09003211 ata_for_each_dev(dev, link, ENABLED)
3212 cnt++;
Tejun Heo022bdb02006-05-15 20:58:22 +09003213 return cnt;
3214}
3215
Tejun Heo02607312007-08-06 18:36:23 +09003216static int ata_link_nr_vacant(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003217{
Tejun Heof58229f2007-08-06 18:36:23 +09003218 struct ata_device *dev;
3219 int cnt = 0;
Tejun Heo084fe632006-05-31 18:28:03 +09003220
Tejun Heo1eca4362008-11-03 20:03:17 +09003221 ata_for_each_dev(dev, link, ALL)
Tejun Heof58229f2007-08-06 18:36:23 +09003222 if (dev->class == ATA_DEV_UNKNOWN)
Tejun Heo084fe632006-05-31 18:28:03 +09003223 cnt++;
3224 return cnt;
3225}
3226
Tejun Heo02607312007-08-06 18:36:23 +09003227static int ata_eh_skip_recovery(struct ata_link *link)
Tejun Heo084fe632006-05-31 18:28:03 +09003228{
Tejun Heo672b2d62008-01-24 00:05:14 +09003229 struct ata_port *ap = link->ap;
Tejun Heo02607312007-08-06 18:36:23 +09003230 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heof58229f2007-08-06 18:36:23 +09003231 struct ata_device *dev;
Tejun Heo084fe632006-05-31 18:28:03 +09003232
Tejun Heof9df58c2007-09-23 13:14:13 +09003233 /* skip disabled links */
3234 if (link->flags & ATA_LFLAG_DISABLED)
3235 return 1;
3236
Tejun Heo672b2d62008-01-24 00:05:14 +09003237 /* thaw frozen port and recover failed devices */
3238 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3239 return 0;
3240
3241 /* reset at least once if reset is requested */
3242 if ((ehc->i.action & ATA_EH_RESET) &&
3243 !(ehc->i.flags & ATA_EHI_DID_RESET))
Tejun Heo084fe632006-05-31 18:28:03 +09003244 return 0;
3245
3246 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
Tejun Heo1eca4362008-11-03 20:03:17 +09003247 ata_for_each_dev(dev, link, ALL) {
Tejun Heo084fe632006-05-31 18:28:03 +09003248 if (dev->class == ATA_DEV_UNKNOWN &&
3249 ehc->classes[dev->devno] != ATA_DEV_NONE)
3250 return 0;
3251 }
3252
3253 return 1;
3254}
3255
Tejun Heoc2c7a892009-01-29 20:31:34 +09003256static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3257{
3258 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3259 u64 now = get_jiffies_64();
3260 int *trials = void_arg;
3261
3262 if (ent->timestamp < now - min(now, interval))
3263 return -1;
3264
3265 (*trials)++;
3266 return 0;
3267}
3268
Tejun Heo02c05a22007-11-27 19:28:54 +09003269static int ata_eh_schedule_probe(struct ata_device *dev)
3270{
3271 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heoc2c7a892009-01-29 20:31:34 +09003272 struct ata_link *link = ata_dev_phys_link(dev);
3273 int trials = 0;
Tejun Heo02c05a22007-11-27 19:28:54 +09003274
3275 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3276 (ehc->did_probe_mask & (1 << dev->devno)))
3277 return 0;
3278
3279 ata_eh_detach_dev(dev);
3280 ata_dev_init(dev);
3281 ehc->did_probe_mask |= (1 << dev->devno);
Tejun Heocf480622008-01-24 00:05:14 +09003282 ehc->i.action |= ATA_EH_RESET;
Tejun Heo00115e02007-11-27 19:28:58 +09003283 ehc->saved_xfer_mode[dev->devno] = 0;
3284 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
Tejun Heo02c05a22007-11-27 19:28:54 +09003285
Tejun Heoc2c7a892009-01-29 20:31:34 +09003286 /* Record and count probe trials on the ering. The specific
3287 * error mask used is irrelevant. Because a successful device
3288 * detection clears the ering, this count accumulates only if
3289 * there are consecutive failed probes.
3290 *
3291 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3292 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3293 * forced to 1.5Gbps.
3294 *
3295 * This is to work around cases where failed link speed
3296 * negotiation results in device misdetection leading to
3297 * infinite DEVXCHG or PHRDY CHG events.
3298 */
3299 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3300 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3301
3302 if (trials > ATA_EH_PROBE_TRIALS)
3303 sata_down_spd_limit(link, 1);
3304
Tejun Heo02c05a22007-11-27 19:28:54 +09003305 return 1;
3306}
3307
Tejun Heo9b1e2652007-08-06 18:36:24 +09003308static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
Tejun Heofee7ca72007-07-01 19:05:58 +09003309{
Tejun Heo9af5c9c2007-08-06 18:36:22 +09003310 struct ata_eh_context *ehc = &dev->link->eh_context;
Tejun Heofee7ca72007-07-01 19:05:58 +09003311
Tejun Heocf9a5902009-01-29 20:31:35 +09003312 /* -EAGAIN from EH routine indicates retry without prejudice.
3313 * The requester is responsible for ensuring forward progress.
3314 */
3315 if (err != -EAGAIN)
3316 ehc->tries[dev->devno]--;
Tejun Heofee7ca72007-07-01 19:05:58 +09003317
3318 switch (err) {
3319 case -ENODEV:
3320 /* device missing or wrong IDENTIFY data, schedule probing */
3321 ehc->i.probe_mask |= (1 << dev->devno);
3322 case -EINVAL:
3323 /* give it just one more chance */
3324 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3325 case -EIO:
Tejun Heod89293a2009-01-29 20:31:29 +09003326 if (ehc->tries[dev->devno] == 1) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003327 /* This is the last chance, better to slow
3328 * down than lose it.
3329 */
Tejun Heoa07d4992009-01-29 20:31:33 +09003330 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
Tejun Heod89293a2009-01-29 20:31:29 +09003331 if (dev->pio_mode > XFER_PIO_0)
3332 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
Tejun Heofee7ca72007-07-01 19:05:58 +09003333 }
3334 }
3335
3336 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3337 /* disable device if it has used up all its chances */
3338 ata_dev_disable(dev);
3339
3340 /* detach if offline */
Tejun Heob1c72912008-07-31 17:02:43 +09003341 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
Tejun Heofee7ca72007-07-01 19:05:58 +09003342 ata_eh_detach_dev(dev);
3343
Tejun Heo02c05a22007-11-27 19:28:54 +09003344 /* schedule probe if necessary */
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003345 if (ata_eh_schedule_probe(dev)) {
Tejun Heofee7ca72007-07-01 19:05:58 +09003346 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo87fbc5a2008-05-20 02:17:54 +09003347 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3348 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3349 }
Tejun Heo9b1e2652007-08-06 18:36:24 +09003350
3351 return 1;
Tejun Heofee7ca72007-07-01 19:05:58 +09003352 } else {
Tejun Heocf480622008-01-24 00:05:14 +09003353 ehc->i.action |= ATA_EH_RESET;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003354 return 0;
Tejun Heofee7ca72007-07-01 19:05:58 +09003355 }
3356}
3357
Tejun Heo022bdb02006-05-15 20:58:22 +09003358/**
3359 * ata_eh_recover - recover host port after error
3360 * @ap: host port to recover
Tejun Heof5914a42006-05-31 18:27:48 +09003361 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003362 * @softreset: softreset method (can be NULL)
3363 * @hardreset: hardreset method (can be NULL)
3364 * @postreset: postreset method (can be NULL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003365 * @r_failed_link: out parameter for failed link
Tejun Heo022bdb02006-05-15 20:58:22 +09003366 *
3367 * This is the alpha and omega, eum and yang, heart and soul of
3368 * libata exception handling. On entry, actions required to
Tejun Heo9b1e2652007-08-06 18:36:24 +09003369 * recover each link and hotplug requests are recorded in the
3370 * link's eh_context. This function executes all the operations
3371 * with appropriate retrials and fallbacks to resurrect failed
Tejun Heo084fe632006-05-31 18:28:03 +09003372 * devices, detach goners and greet newcomers.
Tejun Heo022bdb02006-05-15 20:58:22 +09003373 *
3374 * LOCKING:
3375 * Kernel thread context (may sleep).
3376 *
3377 * RETURNS:
3378 * 0 on success, -errno on failure.
3379 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003380int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3381 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3382 ata_postreset_fn_t postreset,
3383 struct ata_link **r_failed_link)
Tejun Heo022bdb02006-05-15 20:58:22 +09003384{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003385 struct ata_link *link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003386 struct ata_device *dev;
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003387 int nr_failed_devs;
Tejun Heodc98c322008-05-19 01:15:07 +09003388 int rc;
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003389 unsigned long flags, deadline;
Tejun Heo022bdb02006-05-15 20:58:22 +09003390
3391 DPRINTK("ENTER\n");
3392
3393 /* prep for recovery */
Tejun Heo1eca4362008-11-03 20:03:17 +09003394 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003395 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo084fe632006-05-31 18:28:03 +09003396
Tejun Heof9df58c2007-09-23 13:14:13 +09003397 /* re-enable link? */
3398 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3399 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3400 spin_lock_irqsave(ap->lock, flags);
3401 link->flags &= ~ATA_LFLAG_DISABLED;
3402 spin_unlock_irqrestore(ap->lock, flags);
3403 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3404 }
3405
Tejun Heo1eca4362008-11-03 20:03:17 +09003406 ata_for_each_dev(dev, link, ALL) {
Tejun Heofd995f72007-09-23 13:14:12 +09003407 if (link->flags & ATA_LFLAG_NO_RETRY)
3408 ehc->tries[dev->devno] = 1;
3409 else
3410 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
Tejun Heo79a55b72007-01-18 17:22:18 +09003411
Tejun Heo9b1e2652007-08-06 18:36:24 +09003412 /* collect port action mask recorded in dev actions */
3413 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3414 ~ATA_EH_PERDEV_MASK;
3415 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
Tejun Heo084fe632006-05-31 18:28:03 +09003416
Tejun Heo9b1e2652007-08-06 18:36:24 +09003417 /* process hotplug request */
3418 if (dev->flags & ATA_DFLAG_DETACH)
3419 ata_eh_detach_dev(dev);
3420
Tejun Heo02c05a22007-11-27 19:28:54 +09003421 /* schedule probe if necessary */
3422 if (!ata_dev_enabled(dev))
3423 ata_eh_schedule_probe(dev);
Tejun Heo084fe632006-05-31 18:28:03 +09003424 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003425 }
3426
3427 retry:
Tejun Heo022bdb02006-05-15 20:58:22 +09003428 rc = 0;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003429 nr_failed_devs = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003430
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003431 /* if UNLOADING, finish immediately */
Tejun Heob51e9e52006-06-29 01:29:30 +09003432 if (ap->pflags & ATA_PFLAG_UNLOADING)
Tejun Heoaeb2ecd2006-06-12 14:11:43 +09003433 goto out;
3434
Tejun Heo9b1e2652007-08-06 18:36:24 +09003435 /* prep for EH */
Tejun Heo1eca4362008-11-03 20:03:17 +09003436 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003437 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003438
Tejun Heo9b1e2652007-08-06 18:36:24 +09003439 /* skip EH if possible. */
3440 if (ata_eh_skip_recovery(link))
3441 ehc->i.action = 0;
3442
Tejun Heo1eca4362008-11-03 20:03:17 +09003443 ata_for_each_dev(dev, link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003444 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3445 }
Tejun Heo084fe632006-05-31 18:28:03 +09003446
Tejun Heo022bdb02006-05-15 20:58:22 +09003447 /* reset */
Tejun Heo1eca4362008-11-03 20:03:17 +09003448 ata_for_each_link(link, ap, EDGE) {
Tejun Heodc98c322008-05-19 01:15:07 +09003449 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003450
Tejun Heodc98c322008-05-19 01:15:07 +09003451 if (!(ehc->i.action & ATA_EH_RESET))
3452 continue;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003453
Tejun Heodc98c322008-05-19 01:15:07 +09003454 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3455 prereset, softreset, hardreset, postreset);
3456 if (rc) {
3457 ata_link_printk(link, KERN_ERR,
3458 "reset failed, giving up\n");
3459 goto out;
Tejun Heo022bdb02006-05-15 20:58:22 +09003460 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003461 }
3462
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003463 do {
3464 unsigned long now;
3465
3466 /*
3467 * clears ATA_EH_PARK in eh_info and resets
3468 * ap->park_req_pending
3469 */
3470 ata_eh_pull_park_action(ap);
3471
3472 deadline = jiffies;
Tejun Heo1eca4362008-11-03 20:03:17 +09003473 ata_for_each_link(link, ap, EDGE) {
3474 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003475 struct ata_eh_context *ehc = &link->eh_context;
3476 unsigned long tmp;
3477
3478 if (dev->class != ATA_DEV_ATA)
3479 continue;
3480 if (!(ehc->i.dev_action[dev->devno] &
3481 ATA_EH_PARK))
3482 continue;
3483 tmp = dev->unpark_deadline;
3484 if (time_before(deadline, tmp))
3485 deadline = tmp;
3486 else if (time_before_eq(tmp, jiffies))
3487 continue;
3488 if (ehc->unloaded_mask & (1 << dev->devno))
3489 continue;
3490
3491 ata_eh_park_issue_cmd(dev, 1);
3492 }
3493 }
3494
3495 now = jiffies;
3496 if (time_before_eq(deadline, now))
3497 break;
3498
3499 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3500 deadline - now);
3501 } while (deadline);
Tejun Heo1eca4362008-11-03 20:03:17 +09003502 ata_for_each_link(link, ap, EDGE) {
3503 ata_for_each_dev(dev, link, ALL) {
Elias Oltmanns45fabbb2008-09-21 11:54:08 +02003504 if (!(link->eh_context.unloaded_mask &
3505 (1 << dev->devno)))
3506 continue;
3507
3508 ata_eh_park_issue_cmd(dev, 0);
3509 ata_eh_done(link, dev, ATA_EH_PARK);
3510 }
3511 }
3512
Tejun Heo9b1e2652007-08-06 18:36:24 +09003513 /* the rest */
Tejun Heo1eca4362008-11-03 20:03:17 +09003514 ata_for_each_link(link, ap, EDGE) {
Tejun Heo9b1e2652007-08-06 18:36:24 +09003515 struct ata_eh_context *ehc = &link->eh_context;
Tejun Heo022bdb02006-05-15 20:58:22 +09003516
Tejun Heo9b1e2652007-08-06 18:36:24 +09003517 /* revalidate existing devices and attach new ones */
3518 rc = ata_eh_revalidate_and_attach(link, &dev);
Tejun Heo4ae72a12007-02-02 16:22:30 +09003519 if (rc)
Tejun Heo022bdb02006-05-15 20:58:22 +09003520 goto dev_fail;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003521
Tejun Heo633273a2007-09-23 13:19:54 +09003522 /* if PMP got attached, return, pmp EH will take care of it */
3523 if (link->device->class == ATA_DEV_PMP) {
3524 ehc->i.action = 0;
3525 return 0;
3526 }
3527
Tejun Heo9b1e2652007-08-06 18:36:24 +09003528 /* configure transfer mode if necessary */
3529 if (ehc->i.flags & ATA_EHI_SETMODE) {
3530 rc = ata_set_mode(link, &dev);
3531 if (rc)
3532 goto dev_fail;
3533 ehc->i.flags &= ~ATA_EHI_SETMODE;
3534 }
3535
Tejun Heo11fc33d2008-08-30 14:20:01 +02003536 /* If reset has been issued, clear UA to avoid
3537 * disrupting the current users of the device.
3538 */
3539 if (ehc->i.flags & ATA_EHI_DID_RESET) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003540 ata_for_each_dev(dev, link, ALL) {
Tejun Heo11fc33d2008-08-30 14:20:01 +02003541 if (dev->class != ATA_DEV_ATAPI)
3542 continue;
3543 rc = atapi_eh_clear_ua(dev);
3544 if (rc)
3545 goto dev_fail;
3546 }
3547 }
3548
Tejun Heo6013efd2009-11-19 15:36:45 +09003549 /* retry flush if necessary */
3550 ata_for_each_dev(dev, link, ALL) {
3551 if (dev->class != ATA_DEV_ATA)
3552 continue;
3553 rc = ata_eh_maybe_retry_flush(dev);
3554 if (rc)
3555 goto dev_fail;
3556 }
3557
Tejun Heo11fc33d2008-08-30 14:20:01 +02003558 /* configure link power saving */
Tejun Heo3ec25eb2008-03-27 18:37:14 +09003559 if (ehc->i.action & ATA_EH_LPM)
Tejun Heo1eca4362008-11-03 20:03:17 +09003560 ata_for_each_dev(dev, link, ALL)
Kristen Carlson Accardica773292007-10-25 00:58:59 -04003561 ata_dev_enable_pm(dev, ap->pm_policy);
3562
Tejun Heo9b1e2652007-08-06 18:36:24 +09003563 /* this link is okay now */
3564 ehc->i.flags = 0;
3565 continue;
3566
Jeff Garzik2dcb4072007-10-19 06:42:56 -04003567dev_fail:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003568 nr_failed_devs++;
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003569 ata_eh_handle_dev_fail(dev, rc);
Tejun Heo9b1e2652007-08-06 18:36:24 +09003570
Tejun Heob06ce3e2007-10-09 15:06:48 +09003571 if (ap->pflags & ATA_PFLAG_FROZEN) {
3572 /* PMP reset requires working host port.
3573 * Can't retry if it's frozen.
3574 */
Tejun Heo071f44b2008-04-07 22:47:22 +09003575 if (sata_pmp_attached(ap))
Tejun Heob06ce3e2007-10-09 15:06:48 +09003576 goto out;
Tejun Heo9b1e2652007-08-06 18:36:24 +09003577 break;
Tejun Heob06ce3e2007-10-09 15:06:48 +09003578 }
Tejun Heo022bdb02006-05-15 20:58:22 +09003579 }
3580
Tejun Heo0a2c0f52008-05-20 02:17:52 +09003581 if (nr_failed_devs)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003582 goto retry;
Tejun Heo022bdb02006-05-15 20:58:22 +09003583
Tejun Heo022bdb02006-05-15 20:58:22 +09003584 out:
Tejun Heo9b1e2652007-08-06 18:36:24 +09003585 if (rc && r_failed_link)
3586 *r_failed_link = link;
Tejun Heo022bdb02006-05-15 20:58:22 +09003587
3588 DPRINTK("EXIT, rc=%d\n", rc);
3589 return rc;
3590}
3591
3592/**
3593 * ata_eh_finish - finish up EH
3594 * @ap: host port to finish EH for
3595 *
3596 * Recovery is complete. Clean up EH states and retry or finish
3597 * failed qcs.
3598 *
3599 * LOCKING:
3600 * None.
3601 */
Tejun Heofb7fd612007-09-23 13:14:12 +09003602void ata_eh_finish(struct ata_port *ap)
Tejun Heo022bdb02006-05-15 20:58:22 +09003603{
3604 int tag;
3605
3606 /* retry or finish qcs */
3607 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3608 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3609
3610 if (!(qc->flags & ATA_QCFLAG_FAILED))
3611 continue;
3612
3613 if (qc->err_mask) {
3614 /* FIXME: Once EH migration is complete,
3615 * generate sense data in this function,
3616 * considering both err_mask and tf.
3617 */
Tejun Heo03faab72008-03-27 19:14:24 +09003618 if (qc->flags & ATA_QCFLAG_RETRY)
Tejun Heo022bdb02006-05-15 20:58:22 +09003619 ata_eh_qc_retry(qc);
Tejun Heo03faab72008-03-27 19:14:24 +09003620 else
3621 ata_eh_qc_complete(qc);
Tejun Heo022bdb02006-05-15 20:58:22 +09003622 } else {
3623 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3624 ata_eh_qc_complete(qc);
3625 } else {
3626 /* feed zero TF to sense generation */
3627 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3628 ata_eh_qc_retry(qc);
3629 }
3630 }
3631 }
Tejun Heoda917d62007-09-23 13:14:12 +09003632
3633 /* make sure nr_active_links is zero after EH */
3634 WARN_ON(ap->nr_active_links);
3635 ap->nr_active_links = 0;
Tejun Heo022bdb02006-05-15 20:58:22 +09003636}
3637
3638/**
3639 * ata_do_eh - do standard error handling
3640 * @ap: host port to handle error for
Tejun Heoa1efdab2008-03-25 12:22:50 +09003641 *
Tejun Heof5914a42006-05-31 18:27:48 +09003642 * @prereset: prereset method (can be NULL)
Tejun Heo022bdb02006-05-15 20:58:22 +09003643 * @softreset: softreset method (can be NULL)
3644 * @hardreset: hardreset method (can be NULL)
3645 * @postreset: postreset method (can be NULL)
3646 *
3647 * Perform standard error handling sequence.
3648 *
3649 * LOCKING:
3650 * Kernel thread context (may sleep).
3651 */
Tejun Heof5914a42006-05-31 18:27:48 +09003652void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3653 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3654 ata_postreset_fn_t postreset)
Tejun Heo022bdb02006-05-15 20:58:22 +09003655{
Tejun Heo9b1e2652007-08-06 18:36:24 +09003656 struct ata_device *dev;
3657 int rc;
3658
3659 ata_eh_autopsy(ap);
3660 ata_eh_report(ap);
3661
3662 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3663 NULL);
3664 if (rc) {
Tejun Heo1eca4362008-11-03 20:03:17 +09003665 ata_for_each_dev(dev, &ap->link, ALL)
Tejun Heo9b1e2652007-08-06 18:36:24 +09003666 ata_dev_disable(dev);
3667 }
3668
Tejun Heo022bdb02006-05-15 20:58:22 +09003669 ata_eh_finish(ap);
3670}
Tejun Heo500530f2006-07-03 16:07:27 +09003671
Tejun Heoa1efdab2008-03-25 12:22:50 +09003672/**
3673 * ata_std_error_handler - standard error handler
3674 * @ap: host port to handle error for
3675 *
3676 * Standard error handler
3677 *
3678 * LOCKING:
3679 * Kernel thread context (may sleep).
3680 */
3681void ata_std_error_handler(struct ata_port *ap)
3682{
3683 struct ata_port_operations *ops = ap->ops;
3684 ata_reset_fn_t hardreset = ops->hardreset;
3685
Tejun Heo57c9efd2008-04-07 22:47:19 +09003686 /* ignore built-in hardreset if SCR access is not available */
Tejun Heofe06e5f2010-05-10 21:41:39 +02003687 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
Tejun Heoa1efdab2008-03-25 12:22:50 +09003688 hardreset = NULL;
3689
3690 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3691}
3692
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003693#ifdef CONFIG_PM
Tejun Heo500530f2006-07-03 16:07:27 +09003694/**
3695 * ata_eh_handle_port_suspend - perform port suspend operation
3696 * @ap: port to suspend
3697 *
3698 * Suspend @ap.
3699 *
3700 * LOCKING:
3701 * Kernel thread context (may sleep).
3702 */
3703static void ata_eh_handle_port_suspend(struct ata_port *ap)
3704{
3705 unsigned long flags;
3706 int rc = 0;
3707
3708 /* are we suspending? */
3709 spin_lock_irqsave(ap->lock, flags);
3710 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3711 ap->pm_mesg.event == PM_EVENT_ON) {
3712 spin_unlock_irqrestore(ap->lock, flags);
3713 return;
3714 }
3715 spin_unlock_irqrestore(ap->lock, flags);
3716
3717 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3718
Tejun Heo64578a3de2007-05-15 03:28:16 +09003719 /* tell ACPI we're suspending */
3720 rc = ata_acpi_on_suspend(ap);
3721 if (rc)
3722 goto out;
3723
Tejun Heo500530f2006-07-03 16:07:27 +09003724 /* suspend */
3725 ata_eh_freeze_port(ap);
3726
3727 if (ap->ops->port_suspend)
3728 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3729
Shaohua Libd3adca2007-11-02 09:32:38 +08003730 ata_acpi_set_state(ap, PMSG_SUSPEND);
Tejun Heo64578a3de2007-05-15 03:28:16 +09003731 out:
Tejun Heo500530f2006-07-03 16:07:27 +09003732 /* report result */
3733 spin_lock_irqsave(ap->lock, flags);
3734
3735 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3736 if (rc == 0)
3737 ap->pflags |= ATA_PFLAG_SUSPENDED;
Tejun Heo64578a3de2007-05-15 03:28:16 +09003738 else if (ap->pflags & ATA_PFLAG_FROZEN)
Tejun Heo500530f2006-07-03 16:07:27 +09003739 ata_port_schedule_eh(ap);
3740
3741 if (ap->pm_result) {
3742 *ap->pm_result = rc;
3743 ap->pm_result = NULL;
3744 }
3745
3746 spin_unlock_irqrestore(ap->lock, flags);
3747
3748 return;
3749}
3750
3751/**
3752 * ata_eh_handle_port_resume - perform port resume operation
3753 * @ap: port to resume
3754 *
3755 * Resume @ap.
3756 *
Tejun Heo500530f2006-07-03 16:07:27 +09003757 * LOCKING:
3758 * Kernel thread context (may sleep).
3759 */
3760static void ata_eh_handle_port_resume(struct ata_port *ap)
3761{
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09003762 struct ata_link *link;
3763 struct ata_device *dev;
Tejun Heo500530f2006-07-03 16:07:27 +09003764 unsigned long flags;
Tejun Heo9666f402007-05-04 21:27:47 +02003765 int rc = 0;
Tejun Heo500530f2006-07-03 16:07:27 +09003766
3767 /* are we resuming? */
3768 spin_lock_irqsave(ap->lock, flags);
3769 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3770 ap->pm_mesg.event != PM_EVENT_ON) {
3771 spin_unlock_irqrestore(ap->lock, flags);
3772 return;
3773 }
3774 spin_unlock_irqrestore(ap->lock, flags);
3775
Tejun Heo9666f402007-05-04 21:27:47 +02003776 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
Tejun Heo500530f2006-07-03 16:07:27 +09003777
Tejun Heo6f9c1ea2009-04-23 09:55:28 +09003778 /*
3779 * Error timestamps are in jiffies which doesn't run while
3780 * suspended and PHY events during resume isn't too uncommon.
3781 * When the two are combined, it can lead to unnecessary speed
3782 * downs if the machine is suspended and resumed repeatedly.
3783 * Clear error history.
3784 */
3785 ata_for_each_link(link, ap, HOST_FIRST)
3786 ata_for_each_dev(dev, link, ALL)
3787 ata_ering_clear(&dev->ering);
3788
Shaohua Libd3adca2007-11-02 09:32:38 +08003789 ata_acpi_set_state(ap, PMSG_ON);
3790
Tejun Heo500530f2006-07-03 16:07:27 +09003791 if (ap->ops->port_resume)
3792 rc = ap->ops->port_resume(ap);
3793
Tejun Heo67465442007-05-15 03:28:16 +09003794 /* tell ACPI that we're resuming */
3795 ata_acpi_on_resume(ap);
3796
Tejun Heo9666f402007-05-04 21:27:47 +02003797 /* report result */
Tejun Heo500530f2006-07-03 16:07:27 +09003798 spin_lock_irqsave(ap->lock, flags);
3799 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3800 if (ap->pm_result) {
3801 *ap->pm_result = rc;
3802 ap->pm_result = NULL;
3803 }
3804 spin_unlock_irqrestore(ap->lock, flags);
3805}
Tejun Heo6ffa01d2007-03-02 17:32:47 +09003806#endif /* CONFIG_PM */