xref: /linux/drivers/ata/libata-eh.c (revision 8e59a2a5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  libata-eh.c - libata error handling
4  *
5  *  Copyright 2006 Tejun Heo <htejun@gmail.com>
6  *
7  *  libata documentation is available via 'make {ps|pdf}docs',
8  *  as Documentation/driver-api/libata.rst
9  *
10  *  Hardware documentation available from http://www.t13.org/ and
11  *  http://www.sata-io.org/
12  */
13 
14 #include <linux/kernel.h>
15 #include <linux/blkdev.h>
16 #include <linux/export.h>
17 #include <linux/pci.h>
18 #include <scsi/scsi.h>
19 #include <scsi/scsi_host.h>
20 #include <scsi/scsi_eh.h>
21 #include <scsi/scsi_device.h>
22 #include <scsi/scsi_cmnd.h>
23 #include <scsi/scsi_dbg.h>
24 #include "../scsi/scsi_transport_api.h"
25 
26 #include <linux/libata.h>
27 
28 #include <trace/events/libata.h>
29 #include "libata.h"
30 
31 enum {
32 	/* speed down verdicts */
33 	ATA_EH_SPDN_NCQ_OFF		= (1 << 0),
34 	ATA_EH_SPDN_SPEED_DOWN		= (1 << 1),
35 	ATA_EH_SPDN_FALLBACK_TO_PIO	= (1 << 2),
36 	ATA_EH_SPDN_KEEP_ERRORS		= (1 << 3),
37 
38 	/* error flags */
39 	ATA_EFLAG_IS_IO			= (1 << 0),
40 	ATA_EFLAG_DUBIOUS_XFER		= (1 << 1),
41 	ATA_EFLAG_OLD_ER                = (1 << 31),
42 
43 	/* error categories */
44 	ATA_ECAT_NONE			= 0,
45 	ATA_ECAT_ATA_BUS		= 1,
46 	ATA_ECAT_TOUT_HSM		= 2,
47 	ATA_ECAT_UNK_DEV		= 3,
48 	ATA_ECAT_DUBIOUS_NONE		= 4,
49 	ATA_ECAT_DUBIOUS_ATA_BUS	= 5,
50 	ATA_ECAT_DUBIOUS_TOUT_HSM	= 6,
51 	ATA_ECAT_DUBIOUS_UNK_DEV	= 7,
52 	ATA_ECAT_NR			= 8,
53 
54 	ATA_EH_CMD_DFL_TIMEOUT		=  5000,
55 
56 	/* always put at least this amount of time between resets */
57 	ATA_EH_RESET_COOL_DOWN		=  5000,
58 
59 	/* Waiting in ->prereset can never be reliable.  It's
60 	 * sometimes nice to wait there but it can't be depended upon;
61 	 * otherwise, we wouldn't be resetting.  Just give it enough
62 	 * time for most drives to spin up.
63 	 */
64 	ATA_EH_PRERESET_TIMEOUT		= 10000,
65 	ATA_EH_FASTDRAIN_INTERVAL	=  3000,
66 
67 	ATA_EH_UA_TRIES			= 5,
68 
69 	/* probe speed down parameters, see ata_eh_schedule_probe() */
70 	ATA_EH_PROBE_TRIAL_INTERVAL	= 60000,	/* 1 min */
71 	ATA_EH_PROBE_TRIALS		= 2,
72 };
73 
74 /* The following table determines how we sequence resets.  Each entry
75  * represents timeout for that try.  The first try can be soft or
76  * hardreset.  All others are hardreset if available.  In most cases
77  * the first reset w/ 10sec timeout should succeed.  Following entries
78  * are mostly for error handling, hotplug and those outlier devices that
79  * take an exceptionally long time to recover from reset.
80  */
81 static const unsigned int ata_eh_reset_timeouts[] = {
82 	10000,	/* most drives spin up by 10sec */
83 	10000,	/* > 99% working drives spin up before 20sec */
84 	35000,	/* give > 30 secs of idleness for outlier devices */
85 	 5000,	/* and sweet one last chance */
86 	UINT_MAX, /* > 1 min has elapsed, give up */
87 };
88 
89 static const unsigned int ata_eh_identify_timeouts[] = {
90 	 5000,	/* covers > 99% of successes and not too boring on failures */
91 	10000,  /* combined time till here is enough even for media access */
92 	30000,	/* for true idiots */
93 	UINT_MAX,
94 };
95 
96 static const unsigned int ata_eh_revalidate_timeouts[] = {
97 	15000,	/* Some drives are slow to read log pages when waking-up */
98 	15000,  /* combined time till here is enough even for media access */
99 	UINT_MAX,
100 };
101 
102 static const unsigned int ata_eh_flush_timeouts[] = {
103 	15000,	/* be generous with flush */
104 	15000,  /* ditto */
105 	30000,	/* and even more generous */
106 	UINT_MAX,
107 };
108 
109 static const unsigned int ata_eh_other_timeouts[] = {
110 	 5000,	/* same rationale as identify timeout */
111 	10000,	/* ditto */
112 	/* but no merciful 30sec for other commands, it just isn't worth it */
113 	UINT_MAX,
114 };
115 
116 struct ata_eh_cmd_timeout_ent {
117 	const u8		*commands;
118 	const unsigned int	*timeouts;
119 };
120 
121 /* The following table determines timeouts to use for EH internal
122  * commands.  Each table entry is a command class and matches the
123  * commands the entry applies to and the timeout table to use.
124  *
125  * On the retry after a command timed out, the next timeout value from
126  * the table is used.  If the table doesn't contain further entries,
127  * the last value is used.
128  *
129  * ehc->cmd_timeout_idx keeps track of which timeout to use per
130  * command class, so if SET_FEATURES times out on the first try, the
131  * next try will use the second timeout value only for that class.
132  */
133 #define CMDS(cmds...)	(const u8 []){ cmds, 0 }
134 static const struct ata_eh_cmd_timeout_ent
135 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
136 	{ .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
137 	  .timeouts = ata_eh_identify_timeouts, },
138 	{ .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT),
139 	  .timeouts = ata_eh_revalidate_timeouts, },
140 	{ .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
141 	  .timeouts = ata_eh_other_timeouts, },
142 	{ .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
143 	  .timeouts = ata_eh_other_timeouts, },
144 	{ .commands = CMDS(ATA_CMD_SET_FEATURES),
145 	  .timeouts = ata_eh_other_timeouts, },
146 	{ .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
147 	  .timeouts = ata_eh_other_timeouts, },
148 	{ .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
149 	  .timeouts = ata_eh_flush_timeouts },
150 	{ .commands = CMDS(ATA_CMD_VERIFY),
151 	  .timeouts = ata_eh_reset_timeouts },
152 };
153 #undef CMDS
154 
155 static void __ata_port_freeze(struct ata_port *ap);
156 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
157 			  struct ata_device **r_failed_dev);
158 #ifdef CONFIG_PM
159 static void ata_eh_handle_port_suspend(struct ata_port *ap);
160 static void ata_eh_handle_port_resume(struct ata_port *ap);
161 #else /* CONFIG_PM */
ata_eh_handle_port_suspend(struct ata_port * ap)162 static void ata_eh_handle_port_suspend(struct ata_port *ap)
163 { }
164 
ata_eh_handle_port_resume(struct ata_port * ap)165 static void ata_eh_handle_port_resume(struct ata_port *ap)
166 { }
167 #endif /* CONFIG_PM */
168 
__ata_ehi_pushv_desc(struct ata_eh_info * ehi,const char * fmt,va_list args)169 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
170 				 const char *fmt, va_list args)
171 {
172 	ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
173 				     ATA_EH_DESC_LEN - ehi->desc_len,
174 				     fmt, args);
175 }
176 
177 /**
178  *	__ata_ehi_push_desc - push error description without adding separator
179  *	@ehi: target EHI
180  *	@fmt: printf format string
181  *
182  *	Format string according to @fmt and append it to @ehi->desc.
183  *
184  *	LOCKING:
185  *	spin_lock_irqsave(host lock)
186  */
__ata_ehi_push_desc(struct ata_eh_info * ehi,const char * fmt,...)187 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
188 {
189 	va_list args;
190 
191 	va_start(args, fmt);
192 	__ata_ehi_pushv_desc(ehi, fmt, args);
193 	va_end(args);
194 }
195 EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
196 
197 /**
198  *	ata_ehi_push_desc - push error description with separator
199  *	@ehi: target EHI
200  *	@fmt: printf format string
201  *
202  *	Format string according to @fmt and append it to @ehi->desc.
203  *	If @ehi->desc is not empty, ", " is added in-between.
204  *
205  *	LOCKING:
206  *	spin_lock_irqsave(host lock)
207  */
ata_ehi_push_desc(struct ata_eh_info * ehi,const char * fmt,...)208 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
209 {
210 	va_list args;
211 
212 	if (ehi->desc_len)
213 		__ata_ehi_push_desc(ehi, ", ");
214 
215 	va_start(args, fmt);
216 	__ata_ehi_pushv_desc(ehi, fmt, args);
217 	va_end(args);
218 }
219 EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
220 
221 /**
222  *	ata_ehi_clear_desc - clean error description
223  *	@ehi: target EHI
224  *
225  *	Clear @ehi->desc.
226  *
227  *	LOCKING:
228  *	spin_lock_irqsave(host lock)
229  */
ata_ehi_clear_desc(struct ata_eh_info * ehi)230 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
231 {
232 	ehi->desc[0] = '\0';
233 	ehi->desc_len = 0;
234 }
235 EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
236 
237 /**
238  *	ata_port_desc - append port description
239  *	@ap: target ATA port
240  *	@fmt: printf format string
241  *
242  *	Format string according to @fmt and append it to port
243  *	description.  If port description is not empty, " " is added
244  *	in-between.  This function is to be used while initializing
245  *	ata_host.  The description is printed on host registration.
246  *
247  *	LOCKING:
248  *	None.
249  */
ata_port_desc(struct ata_port * ap,const char * fmt,...)250 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
251 {
252 	va_list args;
253 
254 	WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
255 
256 	if (ap->link.eh_info.desc_len)
257 		__ata_ehi_push_desc(&ap->link.eh_info, " ");
258 
259 	va_start(args, fmt);
260 	__ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
261 	va_end(args);
262 }
263 EXPORT_SYMBOL_GPL(ata_port_desc);
264 
265 #ifdef CONFIG_PCI
266 /**
267  *	ata_port_pbar_desc - append PCI BAR description
268  *	@ap: target ATA port
269  *	@bar: target PCI BAR
270  *	@offset: offset into PCI BAR
271  *	@name: name of the area
272  *
273  *	If @offset is negative, this function formats a string which
274  *	contains the name, address, size and type of the BAR and
275  *	appends it to the port description.  If @offset is zero or
276  *	positive, only name and offsetted address is appended.
277  *
278  *	LOCKING:
279  *	None.
280  */
ata_port_pbar_desc(struct ata_port * ap,int bar,ssize_t offset,const char * name)281 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
282 			const char *name)
283 {
284 	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
285 	char *type = "";
286 	unsigned long long start, len;
287 
288 	if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
289 		type = "m";
290 	else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
291 		type = "i";
292 
293 	start = (unsigned long long)pci_resource_start(pdev, bar);
294 	len = (unsigned long long)pci_resource_len(pdev, bar);
295 
296 	if (offset < 0)
297 		ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
298 	else
299 		ata_port_desc(ap, "%s 0x%llx", name,
300 				start + (unsigned long long)offset);
301 }
302 EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
303 #endif /* CONFIG_PCI */
304 
ata_lookup_timeout_table(u8 cmd)305 static int ata_lookup_timeout_table(u8 cmd)
306 {
307 	int i;
308 
309 	for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
310 		const u8 *cur;
311 
312 		for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
313 			if (*cur == cmd)
314 				return i;
315 	}
316 
317 	return -1;
318 }
319 
320 /**
321  *	ata_internal_cmd_timeout - determine timeout for an internal command
322  *	@dev: target device
323  *	@cmd: internal command to be issued
324  *
325  *	Determine timeout for internal command @cmd for @dev.
326  *
327  *	LOCKING:
328  *	EH context.
329  *
330  *	RETURNS:
331  *	Determined timeout.
332  */
ata_internal_cmd_timeout(struct ata_device * dev,u8 cmd)333 unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
334 {
335 	struct ata_eh_context *ehc = &dev->link->eh_context;
336 	int ent = ata_lookup_timeout_table(cmd);
337 	int idx;
338 
339 	if (ent < 0)
340 		return ATA_EH_CMD_DFL_TIMEOUT;
341 
342 	idx = ehc->cmd_timeout_idx[dev->devno][ent];
343 	return ata_eh_cmd_timeout_table[ent].timeouts[idx];
344 }
345 
346 /**
347  *	ata_internal_cmd_timed_out - notification for internal command timeout
348  *	@dev: target device
349  *	@cmd: internal command which timed out
350  *
351  *	Notify EH that internal command @cmd for @dev timed out.  This
352  *	function should be called only for commands whose timeouts are
353  *	determined using ata_internal_cmd_timeout().
354  *
355  *	LOCKING:
356  *	EH context.
357  */
ata_internal_cmd_timed_out(struct ata_device * dev,u8 cmd)358 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
359 {
360 	struct ata_eh_context *ehc = &dev->link->eh_context;
361 	int ent = ata_lookup_timeout_table(cmd);
362 	int idx;
363 
364 	if (ent < 0)
365 		return;
366 
367 	idx = ehc->cmd_timeout_idx[dev->devno][ent];
368 	if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
369 		ehc->cmd_timeout_idx[dev->devno][ent]++;
370 }
371 
ata_ering_record(struct ata_ering * ering,unsigned int eflags,unsigned int err_mask)372 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
373 			     unsigned int err_mask)
374 {
375 	struct ata_ering_entry *ent;
376 
377 	WARN_ON(!err_mask);
378 
379 	ering->cursor++;
380 	ering->cursor %= ATA_ERING_SIZE;
381 
382 	ent = &ering->ring[ering->cursor];
383 	ent->eflags = eflags;
384 	ent->err_mask = err_mask;
385 	ent->timestamp = get_jiffies_64();
386 }
387 
ata_ering_top(struct ata_ering * ering)388 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
389 {
390 	struct ata_ering_entry *ent = &ering->ring[ering->cursor];
391 
392 	if (ent->err_mask)
393 		return ent;
394 	return NULL;
395 }
396 
ata_ering_map(struct ata_ering * ering,int (* map_fn)(struct ata_ering_entry *,void *),void * arg)397 int ata_ering_map(struct ata_ering *ering,
398 		  int (*map_fn)(struct ata_ering_entry *, void *),
399 		  void *arg)
400 {
401 	int idx, rc = 0;
402 	struct ata_ering_entry *ent;
403 
404 	idx = ering->cursor;
405 	do {
406 		ent = &ering->ring[idx];
407 		if (!ent->err_mask)
408 			break;
409 		rc = map_fn(ent, arg);
410 		if (rc)
411 			break;
412 		idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
413 	} while (idx != ering->cursor);
414 
415 	return rc;
416 }
417 
ata_ering_clear_cb(struct ata_ering_entry * ent,void * void_arg)418 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
419 {
420 	ent->eflags |= ATA_EFLAG_OLD_ER;
421 	return 0;
422 }
423 
ata_ering_clear(struct ata_ering * ering)424 static void ata_ering_clear(struct ata_ering *ering)
425 {
426 	ata_ering_map(ering, ata_ering_clear_cb, NULL);
427 }
428 
ata_eh_dev_action(struct ata_device * dev)429 static unsigned int ata_eh_dev_action(struct ata_device *dev)
430 {
431 	struct ata_eh_context *ehc = &dev->link->eh_context;
432 
433 	return ehc->i.action | ehc->i.dev_action[dev->devno];
434 }
435 
ata_eh_clear_action(struct ata_link * link,struct ata_device * dev,struct ata_eh_info * ehi,unsigned int action)436 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
437 				struct ata_eh_info *ehi, unsigned int action)
438 {
439 	struct ata_device *tdev;
440 
441 	if (!dev) {
442 		ehi->action &= ~action;
443 		ata_for_each_dev(tdev, link, ALL)
444 			ehi->dev_action[tdev->devno] &= ~action;
445 	} else {
446 		/* doesn't make sense for port-wide EH actions */
447 		WARN_ON(!(action & ATA_EH_PERDEV_MASK));
448 
449 		/* break ehi->action into ehi->dev_action */
450 		if (ehi->action & action) {
451 			ata_for_each_dev(tdev, link, ALL)
452 				ehi->dev_action[tdev->devno] |=
453 					ehi->action & action;
454 			ehi->action &= ~action;
455 		}
456 
457 		/* turn off the specified per-dev action */
458 		ehi->dev_action[dev->devno] &= ~action;
459 	}
460 }
461 
462 /**
463  *	ata_eh_acquire - acquire EH ownership
464  *	@ap: ATA port to acquire EH ownership for
465  *
466  *	Acquire EH ownership for @ap.  This is the basic exclusion
467  *	mechanism for ports sharing a host.  Only one port hanging off
468  *	the same host can claim the ownership of EH.
469  *
470  *	LOCKING:
471  *	EH context.
472  */
ata_eh_acquire(struct ata_port * ap)473 void ata_eh_acquire(struct ata_port *ap)
474 {
475 	mutex_lock(&ap->host->eh_mutex);
476 	WARN_ON_ONCE(ap->host->eh_owner);
477 	ap->host->eh_owner = current;
478 }
479 
480 /**
481  *	ata_eh_release - release EH ownership
482  *	@ap: ATA port to release EH ownership for
483  *
484  *	Release EH ownership for @ap if the caller.  The caller must
485  *	have acquired EH ownership using ata_eh_acquire() previously.
486  *
487  *	LOCKING:
488  *	EH context.
489  */
ata_eh_release(struct ata_port * ap)490 void ata_eh_release(struct ata_port *ap)
491 {
492 	WARN_ON_ONCE(ap->host->eh_owner != current);
493 	ap->host->eh_owner = NULL;
494 	mutex_unlock(&ap->host->eh_mutex);
495 }
496 
ata_eh_dev_disable(struct ata_device * dev)497 static void ata_eh_dev_disable(struct ata_device *dev)
498 {
499 	ata_acpi_on_disable(dev);
500 	ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
501 	dev->class++;
502 
503 	/*
504 	 * From now till the next successful probe, ering is used to
505 	 * track probe failures.  Clear accumulated device error info.
506 	 */
507 	ata_ering_clear(&dev->ering);
508 
509 	ata_dev_free_resources(dev);
510 }
511 
ata_eh_unload(struct ata_port * ap)512 static void ata_eh_unload(struct ata_port *ap)
513 {
514 	struct ata_link *link;
515 	struct ata_device *dev;
516 	unsigned long flags;
517 
518 	/*
519 	 * Unless we are restarting, transition all enabled devices to
520 	 * standby power mode.
521 	 */
522 	if (system_state != SYSTEM_RESTART) {
523 		ata_for_each_link(link, ap, PMP_FIRST) {
524 			ata_for_each_dev(dev, link, ENABLED)
525 				ata_dev_power_set_standby(dev);
526 		}
527 	}
528 
529 	/*
530 	 * Restore SControl IPM and SPD for the next driver and
531 	 * disable attached devices.
532 	 */
533 	ata_for_each_link(link, ap, PMP_FIRST) {
534 		sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
535 		ata_for_each_dev(dev, link, ENABLED)
536 			ata_eh_dev_disable(dev);
537 	}
538 
539 	/* freeze and set UNLOADED */
540 	spin_lock_irqsave(ap->lock, flags);
541 
542 	ata_port_freeze(ap);			/* won't be thawed */
543 	ap->pflags &= ~ATA_PFLAG_EH_PENDING;	/* clear pending from freeze */
544 	ap->pflags |= ATA_PFLAG_UNLOADED;
545 
546 	spin_unlock_irqrestore(ap->lock, flags);
547 }
548 
549 /**
550  *	ata_scsi_error - SCSI layer error handler callback
551  *	@host: SCSI host on which error occurred
552  *
553  *	Handles SCSI-layer-thrown error events.
554  *
555  *	LOCKING:
556  *	Inherited from SCSI layer (none, can sleep)
557  *
558  *	RETURNS:
559  *	Zero.
560  */
ata_scsi_error(struct Scsi_Host * host)561 void ata_scsi_error(struct Scsi_Host *host)
562 {
563 	struct ata_port *ap = ata_shost_to_port(host);
564 	unsigned long flags;
565 	LIST_HEAD(eh_work_q);
566 
567 	spin_lock_irqsave(host->host_lock, flags);
568 	list_splice_init(&host->eh_cmd_q, &eh_work_q);
569 	spin_unlock_irqrestore(host->host_lock, flags);
570 
571 	ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
572 
573 	/* If we timed raced normal completion and there is nothing to
574 	   recover nr_timedout == 0 why exactly are we doing error recovery ? */
575 	ata_scsi_port_error_handler(host, ap);
576 
577 	/* finish or retry handled scmd's and clean up */
578 	WARN_ON(!list_empty(&eh_work_q));
579 
580 }
581 
582 /**
583  * ata_scsi_cmd_error_handler - error callback for a list of commands
584  * @host:	scsi host containing the port
585  * @ap:		ATA port within the host
586  * @eh_work_q:	list of commands to process
587  *
588  * process the given list of commands and return those finished to the
589  * ap->eh_done_q.  This function is the first part of the libata error
590  * handler which processes a given list of failed commands.
591  */
ata_scsi_cmd_error_handler(struct Scsi_Host * host,struct ata_port * ap,struct list_head * eh_work_q)592 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
593 				struct list_head *eh_work_q)
594 {
595 	int i;
596 	unsigned long flags;
597 	struct scsi_cmnd *scmd, *tmp;
598 	int nr_timedout = 0;
599 
600 	/* make sure sff pio task is not running */
601 	ata_sff_flush_pio_task(ap);
602 
603 	/* synchronize with host lock and sort out timeouts */
604 
605 	/*
606 	 * For EH, all qcs are finished in one of three ways -
607 	 * normal completion, error completion, and SCSI timeout.
608 	 * Both completions can race against SCSI timeout.  When normal
609 	 * completion wins, the qc never reaches EH.  When error
610 	 * completion wins, the qc has ATA_QCFLAG_EH set.
611 	 *
612 	 * When SCSI timeout wins, things are a bit more complex.
613 	 * Normal or error completion can occur after the timeout but
614 	 * before this point.  In such cases, both types of
615 	 * completions are honored.  A scmd is determined to have
616 	 * timed out iff its associated qc is active and not failed.
617 	 */
618 	spin_lock_irqsave(ap->lock, flags);
619 
620 	/*
621 	 * This must occur under the ap->lock as we don't want
622 	 * a polled recovery to race the real interrupt handler
623 	 *
624 	 * The lost_interrupt handler checks for any completed but
625 	 * non-notified command and completes much like an IRQ handler.
626 	 *
627 	 * We then fall into the error recovery code which will treat
628 	 * this as if normal completion won the race
629 	 */
630 	if (ap->ops->lost_interrupt)
631 		ap->ops->lost_interrupt(ap);
632 
633 	list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
634 		struct ata_queued_cmd *qc;
635 
636 		/*
637 		 * If the scmd was added to EH, via ata_qc_schedule_eh() ->
638 		 * scsi_timeout() -> scsi_eh_scmd_add(), scsi_timeout() will
639 		 * have set DID_TIME_OUT (since libata does not have an abort
640 		 * handler). Thus, to clear DID_TIME_OUT, clear the host byte.
641 		 */
642 		set_host_byte(scmd, DID_OK);
643 
644 		ata_qc_for_each_raw(ap, qc, i) {
645 			if (qc->flags & ATA_QCFLAG_ACTIVE &&
646 			    qc->scsicmd == scmd)
647 				break;
648 		}
649 
650 		if (i < ATA_MAX_QUEUE) {
651 			/* the scmd has an associated qc */
652 			if (!(qc->flags & ATA_QCFLAG_EH)) {
653 				/* which hasn't failed yet, timeout */
654 				set_host_byte(scmd, DID_TIME_OUT);
655 				qc->err_mask |= AC_ERR_TIMEOUT;
656 				qc->flags |= ATA_QCFLAG_EH;
657 				nr_timedout++;
658 			}
659 		} else {
660 			/* Normal completion occurred after
661 			 * SCSI timeout but before this point.
662 			 * Successfully complete it.
663 			 */
664 			scmd->retries = scmd->allowed;
665 			scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
666 		}
667 	}
668 
669 	/*
670 	 * If we have timed out qcs.  They belong to EH from
671 	 * this point but the state of the controller is
672 	 * unknown.  Freeze the port to make sure the IRQ
673 	 * handler doesn't diddle with those qcs.  This must
674 	 * be done atomically w.r.t. setting ATA_QCFLAG_EH.
675 	 */
676 	if (nr_timedout)
677 		__ata_port_freeze(ap);
678 
679 	/* initialize eh_tries */
680 	ap->eh_tries = ATA_EH_MAX_TRIES;
681 
682 	spin_unlock_irqrestore(ap->lock, flags);
683 }
684 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
685 
686 /**
687  * ata_scsi_port_error_handler - recover the port after the commands
688  * @host:	SCSI host containing the port
689  * @ap:		the ATA port
690  *
691  * Handle the recovery of the port @ap after all the commands
692  * have been recovered.
693  */
ata_scsi_port_error_handler(struct Scsi_Host * host,struct ata_port * ap)694 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
695 {
696 	unsigned long flags;
697 	struct ata_link *link;
698 
699 	/* acquire EH ownership */
700 	ata_eh_acquire(ap);
701  repeat:
702 	/* kill fast drain timer */
703 	del_timer_sync(&ap->fastdrain_timer);
704 
705 	/* process port resume request */
706 	ata_eh_handle_port_resume(ap);
707 
708 	/* fetch & clear EH info */
709 	spin_lock_irqsave(ap->lock, flags);
710 
711 	ata_for_each_link(link, ap, HOST_FIRST) {
712 		struct ata_eh_context *ehc = &link->eh_context;
713 		struct ata_device *dev;
714 
715 		memset(&link->eh_context, 0, sizeof(link->eh_context));
716 		link->eh_context.i = link->eh_info;
717 		memset(&link->eh_info, 0, sizeof(link->eh_info));
718 
719 		ata_for_each_dev(dev, link, ENABLED) {
720 			int devno = dev->devno;
721 
722 			ehc->saved_xfer_mode[devno] = dev->xfer_mode;
723 			if (ata_ncq_enabled(dev))
724 				ehc->saved_ncq_enabled |= 1 << devno;
725 
726 			/* If we are resuming, wake up the device */
727 			if (ap->pflags & ATA_PFLAG_RESUMING) {
728 				dev->flags |= ATA_DFLAG_RESUMING;
729 				ehc->i.dev_action[devno] |= ATA_EH_SET_ACTIVE;
730 			}
731 		}
732 	}
733 
734 	ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
735 	ap->pflags &= ~ATA_PFLAG_EH_PENDING;
736 	ap->excl_link = NULL;	/* don't maintain exclusion over EH */
737 
738 	spin_unlock_irqrestore(ap->lock, flags);
739 
740 	/* invoke EH, skip if unloading or suspended */
741 	if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
742 		ap->ops->error_handler(ap);
743 	else {
744 		/* if unloading, commence suicide */
745 		if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
746 		    !(ap->pflags & ATA_PFLAG_UNLOADED))
747 			ata_eh_unload(ap);
748 		ata_eh_finish(ap);
749 	}
750 
751 	/* process port suspend request */
752 	ata_eh_handle_port_suspend(ap);
753 
754 	/*
755 	 * Exception might have happened after ->error_handler recovered the
756 	 * port but before this point.  Repeat EH in such case.
757 	 */
758 	spin_lock_irqsave(ap->lock, flags);
759 
760 	if (ap->pflags & ATA_PFLAG_EH_PENDING) {
761 		if (--ap->eh_tries) {
762 			spin_unlock_irqrestore(ap->lock, flags);
763 			goto repeat;
764 		}
765 		ata_port_err(ap,
766 			     "EH pending after %d tries, giving up\n",
767 			     ATA_EH_MAX_TRIES);
768 		ap->pflags &= ~ATA_PFLAG_EH_PENDING;
769 	}
770 
771 	/* this run is complete, make sure EH info is clear */
772 	ata_for_each_link(link, ap, HOST_FIRST)
773 		memset(&link->eh_info, 0, sizeof(link->eh_info));
774 
775 	/*
776 	 * end eh (clear host_eh_scheduled) while holding ap->lock such that if
777 	 * exception occurs after this point but before EH completion, SCSI
778 	 * midlayer will re-initiate EH.
779 	 */
780 	ap->ops->end_eh(ap);
781 
782 	spin_unlock_irqrestore(ap->lock, flags);
783 	ata_eh_release(ap);
784 
785 	scsi_eh_flush_done_q(&ap->eh_done_q);
786 
787 	/* clean up */
788 	spin_lock_irqsave(ap->lock, flags);
789 
790 	ap->pflags &= ~ATA_PFLAG_RESUMING;
791 
792 	if (ap->pflags & ATA_PFLAG_LOADING)
793 		ap->pflags &= ~ATA_PFLAG_LOADING;
794 	else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
795 		!(ap->flags & ATA_FLAG_SAS_HOST))
796 		schedule_delayed_work(&ap->hotplug_task, 0);
797 
798 	if (ap->pflags & ATA_PFLAG_RECOVERED)
799 		ata_port_info(ap, "EH complete\n");
800 
801 	ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
802 
803 	/* tell wait_eh that we're done */
804 	ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
805 	wake_up_all(&ap->eh_wait_q);
806 
807 	spin_unlock_irqrestore(ap->lock, flags);
808 }
809 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
810 
811 /**
812  *	ata_port_wait_eh - Wait for the currently pending EH to complete
813  *	@ap: Port to wait EH for
814  *
815  *	Wait until the currently pending EH is complete.
816  *
817  *	LOCKING:
818  *	Kernel thread context (may sleep).
819  */
ata_port_wait_eh(struct ata_port * ap)820 void ata_port_wait_eh(struct ata_port *ap)
821 {
822 	unsigned long flags;
823 	DEFINE_WAIT(wait);
824 
825  retry:
826 	spin_lock_irqsave(ap->lock, flags);
827 
828 	while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
829 		prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
830 		spin_unlock_irqrestore(ap->lock, flags);
831 		schedule();
832 		spin_lock_irqsave(ap->lock, flags);
833 	}
834 	finish_wait(&ap->eh_wait_q, &wait);
835 
836 	spin_unlock_irqrestore(ap->lock, flags);
837 
838 	/* make sure SCSI EH is complete */
839 	if (scsi_host_in_recovery(ap->scsi_host)) {
840 		ata_msleep(ap, 10);
841 		goto retry;
842 	}
843 }
844 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
845 
ata_eh_nr_in_flight(struct ata_port * ap)846 static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
847 {
848 	struct ata_queued_cmd *qc;
849 	unsigned int tag;
850 	unsigned int nr = 0;
851 
852 	/* count only non-internal commands */
853 	ata_qc_for_each(ap, qc, tag) {
854 		if (qc)
855 			nr++;
856 	}
857 
858 	return nr;
859 }
860 
ata_eh_fastdrain_timerfn(struct timer_list * t)861 void ata_eh_fastdrain_timerfn(struct timer_list *t)
862 {
863 	struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
864 	unsigned long flags;
865 	unsigned int cnt;
866 
867 	spin_lock_irqsave(ap->lock, flags);
868 
869 	cnt = ata_eh_nr_in_flight(ap);
870 
871 	/* are we done? */
872 	if (!cnt)
873 		goto out_unlock;
874 
875 	if (cnt == ap->fastdrain_cnt) {
876 		struct ata_queued_cmd *qc;
877 		unsigned int tag;
878 
879 		/* No progress during the last interval, tag all
880 		 * in-flight qcs as timed out and freeze the port.
881 		 */
882 		ata_qc_for_each(ap, qc, tag) {
883 			if (qc)
884 				qc->err_mask |= AC_ERR_TIMEOUT;
885 		}
886 
887 		ata_port_freeze(ap);
888 	} else {
889 		/* some qcs have finished, give it another chance */
890 		ap->fastdrain_cnt = cnt;
891 		ap->fastdrain_timer.expires =
892 			ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
893 		add_timer(&ap->fastdrain_timer);
894 	}
895 
896  out_unlock:
897 	spin_unlock_irqrestore(ap->lock, flags);
898 }
899 
900 /**
901  *	ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
902  *	@ap: target ATA port
903  *	@fastdrain: activate fast drain
904  *
905  *	Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
906  *	is non-zero and EH wasn't pending before.  Fast drain ensures
907  *	that EH kicks in in timely manner.
908  *
909  *	LOCKING:
910  *	spin_lock_irqsave(host lock)
911  */
ata_eh_set_pending(struct ata_port * ap,int fastdrain)912 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
913 {
914 	unsigned int cnt;
915 
916 	/* already scheduled? */
917 	if (ap->pflags & ATA_PFLAG_EH_PENDING)
918 		return;
919 
920 	ap->pflags |= ATA_PFLAG_EH_PENDING;
921 
922 	if (!fastdrain)
923 		return;
924 
925 	/* do we have in-flight qcs? */
926 	cnt = ata_eh_nr_in_flight(ap);
927 	if (!cnt)
928 		return;
929 
930 	/* activate fast drain */
931 	ap->fastdrain_cnt = cnt;
932 	ap->fastdrain_timer.expires =
933 		ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
934 	add_timer(&ap->fastdrain_timer);
935 }
936 
937 /**
938  *	ata_qc_schedule_eh - schedule qc for error handling
939  *	@qc: command to schedule error handling for
940  *
941  *	Schedule error handling for @qc.  EH will kick in as soon as
942  *	other commands are drained.
943  *
944  *	LOCKING:
945  *	spin_lock_irqsave(host lock)
946  */
ata_qc_schedule_eh(struct ata_queued_cmd * qc)947 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
948 {
949 	struct ata_port *ap = qc->ap;
950 
951 	qc->flags |= ATA_QCFLAG_EH;
952 	ata_eh_set_pending(ap, 1);
953 
954 	/* The following will fail if timeout has already expired.
955 	 * ata_scsi_error() takes care of such scmds on EH entry.
956 	 * Note that ATA_QCFLAG_EH is unconditionally set after
957 	 * this function completes.
958 	 */
959 	blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
960 }
961 
962 /**
963  * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
964  * @ap: ATA port to schedule EH for
965  *
966  *	LOCKING: inherited from ata_port_schedule_eh
967  *	spin_lock_irqsave(host lock)
968  */
ata_std_sched_eh(struct ata_port * ap)969 void ata_std_sched_eh(struct ata_port *ap)
970 {
971 	if (ap->pflags & ATA_PFLAG_INITIALIZING)
972 		return;
973 
974 	ata_eh_set_pending(ap, 1);
975 	scsi_schedule_eh(ap->scsi_host);
976 
977 	trace_ata_std_sched_eh(ap);
978 }
979 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
980 
981 /**
982  * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
983  * @ap: ATA port to end EH for
984  *
985  * In the libata object model there is a 1:1 mapping of ata_port to
986  * shost, so host fields can be directly manipulated under ap->lock, in
987  * the libsas case we need to hold a lock at the ha->level to coordinate
988  * these events.
989  *
990  *	LOCKING:
991  *	spin_lock_irqsave(host lock)
992  */
ata_std_end_eh(struct ata_port * ap)993 void ata_std_end_eh(struct ata_port *ap)
994 {
995 	struct Scsi_Host *host = ap->scsi_host;
996 
997 	host->host_eh_scheduled = 0;
998 }
999 EXPORT_SYMBOL(ata_std_end_eh);
1000 
1001 
1002 /**
1003  *	ata_port_schedule_eh - schedule error handling without a qc
1004  *	@ap: ATA port to schedule EH for
1005  *
1006  *	Schedule error handling for @ap.  EH will kick in as soon as
1007  *	all commands are drained.
1008  *
1009  *	LOCKING:
1010  *	spin_lock_irqsave(host lock)
1011  */
ata_port_schedule_eh(struct ata_port * ap)1012 void ata_port_schedule_eh(struct ata_port *ap)
1013 {
1014 	/* see: ata_std_sched_eh, unless you know better */
1015 	ap->ops->sched_eh(ap);
1016 }
1017 EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
1018 
ata_do_link_abort(struct ata_port * ap,struct ata_link * link)1019 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1020 {
1021 	struct ata_queued_cmd *qc;
1022 	int tag, nr_aborted = 0;
1023 
1024 	/* we're gonna abort all commands, no need for fast drain */
1025 	ata_eh_set_pending(ap, 0);
1026 
1027 	/* include internal tag in iteration */
1028 	ata_qc_for_each_with_internal(ap, qc, tag) {
1029 		if (qc && (!link || qc->dev->link == link)) {
1030 			qc->flags |= ATA_QCFLAG_EH;
1031 			ata_qc_complete(qc);
1032 			nr_aborted++;
1033 		}
1034 	}
1035 
1036 	if (!nr_aborted)
1037 		ata_port_schedule_eh(ap);
1038 
1039 	return nr_aborted;
1040 }
1041 
1042 /**
1043  *	ata_link_abort - abort all qc's on the link
1044  *	@link: ATA link to abort qc's for
1045  *
1046  *	Abort all active qc's active on @link and schedule EH.
1047  *
1048  *	LOCKING:
1049  *	spin_lock_irqsave(host lock)
1050  *
1051  *	RETURNS:
1052  *	Number of aborted qc's.
1053  */
ata_link_abort(struct ata_link * link)1054 int ata_link_abort(struct ata_link *link)
1055 {
1056 	return ata_do_link_abort(link->ap, link);
1057 }
1058 EXPORT_SYMBOL_GPL(ata_link_abort);
1059 
1060 /**
1061  *	ata_port_abort - abort all qc's on the port
1062  *	@ap: ATA port to abort qc's for
1063  *
1064  *	Abort all active qc's of @ap and schedule EH.
1065  *
1066  *	LOCKING:
1067  *	spin_lock_irqsave(host_set lock)
1068  *
1069  *	RETURNS:
1070  *	Number of aborted qc's.
1071  */
ata_port_abort(struct ata_port * ap)1072 int ata_port_abort(struct ata_port *ap)
1073 {
1074 	return ata_do_link_abort(ap, NULL);
1075 }
1076 EXPORT_SYMBOL_GPL(ata_port_abort);
1077 
1078 /**
1079  *	__ata_port_freeze - freeze port
1080  *	@ap: ATA port to freeze
1081  *
1082  *	This function is called when HSM violation or some other
1083  *	condition disrupts normal operation of the port.  Frozen port
1084  *	is not allowed to perform any operation until the port is
1085  *	thawed, which usually follows a successful reset.
1086  *
1087  *	ap->ops->freeze() callback can be used for freezing the port
1088  *	hardware-wise (e.g. mask interrupt and stop DMA engine).  If a
1089  *	port cannot be frozen hardware-wise, the interrupt handler
1090  *	must ack and clear interrupts unconditionally while the port
1091  *	is frozen.
1092  *
1093  *	LOCKING:
1094  *	spin_lock_irqsave(host lock)
1095  */
__ata_port_freeze(struct ata_port * ap)1096 static void __ata_port_freeze(struct ata_port *ap)
1097 {
1098 	if (ap->ops->freeze)
1099 		ap->ops->freeze(ap);
1100 
1101 	ap->pflags |= ATA_PFLAG_FROZEN;
1102 
1103 	trace_ata_port_freeze(ap);
1104 }
1105 
1106 /**
1107  *	ata_port_freeze - abort & freeze port
1108  *	@ap: ATA port to freeze
1109  *
1110  *	Abort and freeze @ap.  The freeze operation must be called
1111  *	first, because some hardware requires special operations
1112  *	before the taskfile registers are accessible.
1113  *
1114  *	LOCKING:
1115  *	spin_lock_irqsave(host lock)
1116  *
1117  *	RETURNS:
1118  *	Number of aborted commands.
1119  */
ata_port_freeze(struct ata_port * ap)1120 int ata_port_freeze(struct ata_port *ap)
1121 {
1122 	__ata_port_freeze(ap);
1123 
1124 	return ata_port_abort(ap);
1125 }
1126 EXPORT_SYMBOL_GPL(ata_port_freeze);
1127 
1128 /**
1129  *	ata_eh_freeze_port - EH helper to freeze port
1130  *	@ap: ATA port to freeze
1131  *
1132  *	Freeze @ap.
1133  *
1134  *	LOCKING:
1135  *	None.
1136  */
ata_eh_freeze_port(struct ata_port * ap)1137 void ata_eh_freeze_port(struct ata_port *ap)
1138 {
1139 	unsigned long flags;
1140 
1141 	spin_lock_irqsave(ap->lock, flags);
1142 	__ata_port_freeze(ap);
1143 	spin_unlock_irqrestore(ap->lock, flags);
1144 }
1145 EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1146 
1147 /**
1148  *	ata_eh_thaw_port - EH helper to thaw port
1149  *	@ap: ATA port to thaw
1150  *
1151  *	Thaw frozen port @ap.
1152  *
1153  *	LOCKING:
1154  *	None.
1155  */
ata_eh_thaw_port(struct ata_port * ap)1156 void ata_eh_thaw_port(struct ata_port *ap)
1157 {
1158 	unsigned long flags;
1159 
1160 	spin_lock_irqsave(ap->lock, flags);
1161 
1162 	ap->pflags &= ~ATA_PFLAG_FROZEN;
1163 
1164 	if (ap->ops->thaw)
1165 		ap->ops->thaw(ap);
1166 
1167 	spin_unlock_irqrestore(ap->lock, flags);
1168 
1169 	trace_ata_port_thaw(ap);
1170 }
1171 
ata_eh_scsidone(struct scsi_cmnd * scmd)1172 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1173 {
1174 	/* nada */
1175 }
1176 
__ata_eh_qc_complete(struct ata_queued_cmd * qc)1177 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1178 {
1179 	struct ata_port *ap = qc->ap;
1180 	struct scsi_cmnd *scmd = qc->scsicmd;
1181 	unsigned long flags;
1182 
1183 	spin_lock_irqsave(ap->lock, flags);
1184 	qc->scsidone = ata_eh_scsidone;
1185 	__ata_qc_complete(qc);
1186 	WARN_ON(ata_tag_valid(qc->tag));
1187 	spin_unlock_irqrestore(ap->lock, flags);
1188 
1189 	scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1190 }
1191 
1192 /**
1193  *	ata_eh_qc_complete - Complete an active ATA command from EH
1194  *	@qc: Command to complete
1195  *
1196  *	Indicate to the mid and upper layers that an ATA command has
1197  *	completed.  To be used from EH.
1198  */
ata_eh_qc_complete(struct ata_queued_cmd * qc)1199 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1200 {
1201 	struct scsi_cmnd *scmd = qc->scsicmd;
1202 	scmd->retries = scmd->allowed;
1203 	__ata_eh_qc_complete(qc);
1204 }
1205 
1206 /**
1207  *	ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1208  *	@qc: Command to retry
1209  *
1210  *	Indicate to the mid and upper layers that an ATA command
1211  *	should be retried.  To be used from EH.
1212  *
1213  *	SCSI midlayer limits the number of retries to scmd->allowed.
1214  *	scmd->allowed is incremented for commands which get retried
1215  *	due to unrelated failures (qc->err_mask is zero).
1216  */
ata_eh_qc_retry(struct ata_queued_cmd * qc)1217 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1218 {
1219 	struct scsi_cmnd *scmd = qc->scsicmd;
1220 	if (!qc->err_mask)
1221 		scmd->allowed++;
1222 	__ata_eh_qc_complete(qc);
1223 }
1224 
1225 /**
1226  *	ata_dev_disable - disable ATA device
1227  *	@dev: ATA device to disable
1228  *
1229  *	Disable @dev.
1230  *
1231  *	Locking:
1232  *	EH context.
1233  */
ata_dev_disable(struct ata_device * dev)1234 void ata_dev_disable(struct ata_device *dev)
1235 {
1236 	if (!ata_dev_enabled(dev))
1237 		return;
1238 
1239 	ata_dev_warn(dev, "disable device\n");
1240 
1241 	ata_eh_dev_disable(dev);
1242 }
1243 EXPORT_SYMBOL_GPL(ata_dev_disable);
1244 
1245 /**
1246  *	ata_eh_detach_dev - detach ATA device
1247  *	@dev: ATA device to detach
1248  *
1249  *	Detach @dev.
1250  *
1251  *	LOCKING:
1252  *	None.
1253  */
ata_eh_detach_dev(struct ata_device * dev)1254 void ata_eh_detach_dev(struct ata_device *dev)
1255 {
1256 	struct ata_link *link = dev->link;
1257 	struct ata_port *ap = link->ap;
1258 	struct ata_eh_context *ehc = &link->eh_context;
1259 	unsigned long flags;
1260 
1261 	/*
1262 	 * If the device is still enabled, transition it to standby power mode
1263 	 * (i.e. spin down HDDs) and disable it.
1264 	 */
1265 	if (ata_dev_enabled(dev)) {
1266 		ata_dev_power_set_standby(dev);
1267 		ata_eh_dev_disable(dev);
1268 	}
1269 
1270 	spin_lock_irqsave(ap->lock, flags);
1271 
1272 	dev->flags &= ~ATA_DFLAG_DETACH;
1273 
1274 	if (ata_scsi_offline_dev(dev)) {
1275 		dev->flags |= ATA_DFLAG_DETACHED;
1276 		ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1277 	}
1278 
1279 	/* clear per-dev EH info */
1280 	ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1281 	ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1282 	ehc->saved_xfer_mode[dev->devno] = 0;
1283 	ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1284 
1285 	spin_unlock_irqrestore(ap->lock, flags);
1286 }
1287 
1288 /**
1289  *	ata_eh_about_to_do - about to perform eh_action
1290  *	@link: target ATA link
1291  *	@dev: target ATA dev for per-dev action (can be NULL)
1292  *	@action: action about to be performed
1293  *
1294  *	Called just before performing EH actions to clear related bits
1295  *	in @link->eh_info such that eh actions are not unnecessarily
1296  *	repeated.
1297  *
1298  *	LOCKING:
1299  *	None.
1300  */
ata_eh_about_to_do(struct ata_link * link,struct ata_device * dev,unsigned int action)1301 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1302 			unsigned int action)
1303 {
1304 	struct ata_port *ap = link->ap;
1305 	struct ata_eh_info *ehi = &link->eh_info;
1306 	struct ata_eh_context *ehc = &link->eh_context;
1307 	unsigned long flags;
1308 
1309 	trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1310 
1311 	spin_lock_irqsave(ap->lock, flags);
1312 
1313 	ata_eh_clear_action(link, dev, ehi, action);
1314 
1315 	/* About to take EH action, set RECOVERED.  Ignore actions on
1316 	 * slave links as master will do them again.
1317 	 */
1318 	if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1319 		ap->pflags |= ATA_PFLAG_RECOVERED;
1320 
1321 	spin_unlock_irqrestore(ap->lock, flags);
1322 }
1323 
1324 /**
1325  *	ata_eh_done - EH action complete
1326  *	@link: ATA link for which EH actions are complete
1327  *	@dev: target ATA dev for per-dev action (can be NULL)
1328  *	@action: action just completed
1329  *
1330  *	Called right after performing EH actions to clear related bits
1331  *	in @link->eh_context.
1332  *
1333  *	LOCKING:
1334  *	None.
1335  */
ata_eh_done(struct ata_link * link,struct ata_device * dev,unsigned int action)1336 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1337 		 unsigned int action)
1338 {
1339 	struct ata_eh_context *ehc = &link->eh_context;
1340 
1341 	trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1342 
1343 	ata_eh_clear_action(link, dev, &ehc->i, action);
1344 }
1345 
1346 /**
1347  *	ata_err_string - convert err_mask to descriptive string
1348  *	@err_mask: error mask to convert to string
1349  *
1350  *	Convert @err_mask to descriptive string.  Errors are
1351  *	prioritized according to severity and only the most severe
1352  *	error is reported.
1353  *
1354  *	LOCKING:
1355  *	None.
1356  *
1357  *	RETURNS:
1358  *	Descriptive string for @err_mask
1359  */
ata_err_string(unsigned int err_mask)1360 static const char *ata_err_string(unsigned int err_mask)
1361 {
1362 	if (err_mask & AC_ERR_HOST_BUS)
1363 		return "host bus error";
1364 	if (err_mask & AC_ERR_ATA_BUS)
1365 		return "ATA bus error";
1366 	if (err_mask & AC_ERR_TIMEOUT)
1367 		return "timeout";
1368 	if (err_mask & AC_ERR_HSM)
1369 		return "HSM violation";
1370 	if (err_mask & AC_ERR_SYSTEM)
1371 		return "internal error";
1372 	if (err_mask & AC_ERR_MEDIA)
1373 		return "media error";
1374 	if (err_mask & AC_ERR_INVALID)
1375 		return "invalid argument";
1376 	if (err_mask & AC_ERR_DEV)
1377 		return "device error";
1378 	if (err_mask & AC_ERR_NCQ)
1379 		return "NCQ error";
1380 	if (err_mask & AC_ERR_NODEV_HINT)
1381 		return "Polling detection error";
1382 	return "unknown error";
1383 }
1384 
1385 /**
1386  *	atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1387  *	@dev: target ATAPI device
1388  *	@r_sense_key: out parameter for sense_key
1389  *
1390  *	Perform ATAPI TEST_UNIT_READY.
1391  *
1392  *	LOCKING:
1393  *	EH context (may sleep).
1394  *
1395  *	RETURNS:
1396  *	0 on success, AC_ERR_* mask on failure.
1397  */
atapi_eh_tur(struct ata_device * dev,u8 * r_sense_key)1398 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1399 {
1400 	u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1401 	struct ata_taskfile tf;
1402 	unsigned int err_mask;
1403 
1404 	ata_tf_init(dev, &tf);
1405 
1406 	tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1407 	tf.command = ATA_CMD_PACKET;
1408 	tf.protocol = ATAPI_PROT_NODATA;
1409 
1410 	err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1411 	if (err_mask == AC_ERR_DEV)
1412 		*r_sense_key = tf.error >> 4;
1413 	return err_mask;
1414 }
1415 
1416 /**
1417  *	ata_eh_decide_disposition - Disposition a qc based on sense data
1418  *	@qc: qc to examine
1419  *
1420  *	For a regular SCSI command, the SCSI completion callback (scsi_done())
1421  *	will call scsi_complete(), which will call scsi_decide_disposition(),
1422  *	which will call scsi_check_sense(). scsi_complete() finally calls
1423  *	scsi_finish_command(). This is fine for SCSI, since any eventual sense
1424  *	data is usually returned in the completion itself (without invoking SCSI
1425  *	EH). However, for a QC, we always need to fetch the sense data
1426  *	explicitly using SCSI EH.
1427  *
1428  *	A command that is completed via SCSI EH will instead be completed using
1429  *	scsi_eh_flush_done_q(), which will call scsi_finish_command() directly
1430  *	(without ever calling scsi_check_sense()).
1431  *
1432  *	For a command that went through SCSI EH, it is the responsibility of the
1433  *	SCSI EH strategy handler to call scsi_decide_disposition(), see e.g. how
1434  *	scsi_eh_get_sense() calls scsi_decide_disposition() for SCSI LLDDs that
1435  *	do not get the sense data as part of the completion.
1436  *
1437  *	Thus, for QC commands that went via SCSI EH, we need to call
1438  *	scsi_check_sense() ourselves, similar to how scsi_eh_get_sense() calls
1439  *	scsi_decide_disposition(), which calls scsi_check_sense(), in order to
1440  *	set the correct SCSI ML byte (if any).
1441  *
1442  *	LOCKING:
1443  *	EH context.
1444  *
1445  *	RETURNS:
1446  *	SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
1447  */
ata_eh_decide_disposition(struct ata_queued_cmd * qc)1448 enum scsi_disposition ata_eh_decide_disposition(struct ata_queued_cmd *qc)
1449 {
1450 	return scsi_check_sense(qc->scsicmd);
1451 }
1452 
1453 /**
1454  *	ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1455  *	@qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1456  *
1457  *	Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1458  *	SENSE.  This function is an EH helper.
1459  *
1460  *	LOCKING:
1461  *	Kernel thread context (may sleep).
1462  *
1463  *	RETURNS:
1464  *	true if sense data could be fetched, false otherwise.
1465  */
ata_eh_request_sense(struct ata_queued_cmd * qc)1466 static bool ata_eh_request_sense(struct ata_queued_cmd *qc)
1467 {
1468 	struct scsi_cmnd *cmd = qc->scsicmd;
1469 	struct ata_device *dev = qc->dev;
1470 	struct ata_taskfile tf;
1471 	unsigned int err_mask;
1472 
1473 	if (ata_port_is_frozen(qc->ap)) {
1474 		ata_dev_warn(dev, "sense data available but port frozen\n");
1475 		return false;
1476 	}
1477 
1478 	if (!ata_id_sense_reporting_enabled(dev->id)) {
1479 		ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1480 		return false;
1481 	}
1482 
1483 	ata_tf_init(dev, &tf);
1484 	tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1485 	tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1486 	tf.command = ATA_CMD_REQ_SENSE_DATA;
1487 	tf.protocol = ATA_PROT_NODATA;
1488 
1489 	err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1490 	/* Ignore err_mask; ATA_ERR might be set */
1491 	if (tf.status & ATA_SENSE) {
1492 		if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1493 			/* Set sense without also setting scsicmd->result */
1494 			scsi_build_sense_buffer(dev->flags & ATA_DFLAG_D_SENSE,
1495 						cmd->sense_buffer, tf.lbah,
1496 						tf.lbam, tf.lbal);
1497 			qc->flags |= ATA_QCFLAG_SENSE_VALID;
1498 			return true;
1499 		}
1500 	} else {
1501 		ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1502 			     tf.status, err_mask);
1503 	}
1504 
1505 	return false;
1506 }
1507 
1508 /**
1509  *	atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1510  *	@dev: device to perform REQUEST_SENSE to
1511  *	@sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1512  *	@dfl_sense_key: default sense key to use
1513  *
1514  *	Perform ATAPI REQUEST_SENSE after the device reported CHECK
1515  *	SENSE.  This function is EH helper.
1516  *
1517  *	LOCKING:
1518  *	Kernel thread context (may sleep).
1519  *
1520  *	RETURNS:
1521  *	0 on success, AC_ERR_* mask on failure
1522  */
atapi_eh_request_sense(struct ata_device * dev,u8 * sense_buf,u8 dfl_sense_key)1523 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1524 					   u8 *sense_buf, u8 dfl_sense_key)
1525 {
1526 	u8 cdb[ATAPI_CDB_LEN] =
1527 		{ REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1528 	struct ata_port *ap = dev->link->ap;
1529 	struct ata_taskfile tf;
1530 
1531 	memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1532 
1533 	/* initialize sense_buf with the error register,
1534 	 * for the case where they are -not- overwritten
1535 	 */
1536 	sense_buf[0] = 0x70;
1537 	sense_buf[2] = dfl_sense_key;
1538 
1539 	/* some devices time out if garbage left in tf */
1540 	ata_tf_init(dev, &tf);
1541 
1542 	tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1543 	tf.command = ATA_CMD_PACKET;
1544 
1545 	/* is it pointless to prefer PIO for "safety reasons"? */
1546 	if (ap->flags & ATA_FLAG_PIO_DMA) {
1547 		tf.protocol = ATAPI_PROT_DMA;
1548 		tf.feature |= ATAPI_PKT_DMA;
1549 	} else {
1550 		tf.protocol = ATAPI_PROT_PIO;
1551 		tf.lbam = SCSI_SENSE_BUFFERSIZE;
1552 		tf.lbah = 0;
1553 	}
1554 
1555 	return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1556 				 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1557 }
1558 
1559 /**
1560  *	ata_eh_analyze_serror - analyze SError for a failed port
1561  *	@link: ATA link to analyze SError for
1562  *
1563  *	Analyze SError if available and further determine cause of
1564  *	failure.
1565  *
1566  *	LOCKING:
1567  *	None.
1568  */
ata_eh_analyze_serror(struct ata_link * link)1569 static void ata_eh_analyze_serror(struct ata_link *link)
1570 {
1571 	struct ata_eh_context *ehc = &link->eh_context;
1572 	u32 serror = ehc->i.serror;
1573 	unsigned int err_mask = 0, action = 0;
1574 	u32 hotplug_mask;
1575 
1576 	if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1577 		err_mask |= AC_ERR_ATA_BUS;
1578 		action |= ATA_EH_RESET;
1579 	}
1580 	if (serror & SERR_PROTOCOL) {
1581 		err_mask |= AC_ERR_HSM;
1582 		action |= ATA_EH_RESET;
1583 	}
1584 	if (serror & SERR_INTERNAL) {
1585 		err_mask |= AC_ERR_SYSTEM;
1586 		action |= ATA_EH_RESET;
1587 	}
1588 
1589 	/* Determine whether a hotplug event has occurred.  Both
1590 	 * SError.N/X are considered hotplug events for enabled or
1591 	 * host links.  For disabled PMP links, only N bit is
1592 	 * considered as X bit is left at 1 for link plugging.
1593 	 */
1594 	if (link->lpm_policy > ATA_LPM_MAX_POWER)
1595 		hotplug_mask = 0;	/* hotplug doesn't work w/ LPM */
1596 	else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1597 		hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1598 	else
1599 		hotplug_mask = SERR_PHYRDY_CHG;
1600 
1601 	if (serror & hotplug_mask)
1602 		ata_ehi_hotplugged(&ehc->i);
1603 
1604 	ehc->i.err_mask |= err_mask;
1605 	ehc->i.action |= action;
1606 }
1607 
1608 /**
1609  *	ata_eh_analyze_tf - analyze taskfile of a failed qc
1610  *	@qc: qc to analyze
1611  *
1612  *	Analyze taskfile of @qc and further determine cause of
1613  *	failure.  This function also requests ATAPI sense data if
1614  *	available.
1615  *
1616  *	LOCKING:
1617  *	Kernel thread context (may sleep).
1618  *
1619  *	RETURNS:
1620  *	Determined recovery action
1621  */
ata_eh_analyze_tf(struct ata_queued_cmd * qc)1622 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1623 {
1624 	const struct ata_taskfile *tf = &qc->result_tf;
1625 	unsigned int tmp, action = 0;
1626 	u8 stat = tf->status, err = tf->error;
1627 
1628 	if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1629 		qc->err_mask |= AC_ERR_HSM;
1630 		return ATA_EH_RESET;
1631 	}
1632 
1633 	if (stat & (ATA_ERR | ATA_DF)) {
1634 		qc->err_mask |= AC_ERR_DEV;
1635 		/*
1636 		 * Sense data reporting does not work if the
1637 		 * device fault bit is set.
1638 		 */
1639 		if (stat & ATA_DF)
1640 			stat &= ~ATA_SENSE;
1641 	} else {
1642 		return 0;
1643 	}
1644 
1645 	switch (qc->dev->class) {
1646 	case ATA_DEV_ATA:
1647 	case ATA_DEV_ZAC:
1648 		/*
1649 		 * Fetch the sense data explicitly if:
1650 		 * -It was a non-NCQ command that failed, or
1651 		 * -It was a NCQ command that failed, but the sense data
1652 		 *  was not included in the NCQ command error log
1653 		 *  (i.e. NCQ autosense is not supported by the device).
1654 		 */
1655 		if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) &&
1656 		    (stat & ATA_SENSE) && ata_eh_request_sense(qc))
1657 			set_status_byte(qc->scsicmd, SAM_STAT_CHECK_CONDITION);
1658 		if (err & ATA_ICRC)
1659 			qc->err_mask |= AC_ERR_ATA_BUS;
1660 		if (err & (ATA_UNC | ATA_AMNF))
1661 			qc->err_mask |= AC_ERR_MEDIA;
1662 		if (err & ATA_IDNF)
1663 			qc->err_mask |= AC_ERR_INVALID;
1664 		break;
1665 
1666 	case ATA_DEV_ATAPI:
1667 		if (!ata_port_is_frozen(qc->ap)) {
1668 			tmp = atapi_eh_request_sense(qc->dev,
1669 						qc->scsicmd->sense_buffer,
1670 						qc->result_tf.error >> 4);
1671 			if (!tmp)
1672 				qc->flags |= ATA_QCFLAG_SENSE_VALID;
1673 			else
1674 				qc->err_mask |= tmp;
1675 		}
1676 	}
1677 
1678 	if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1679 		enum scsi_disposition ret = ata_eh_decide_disposition(qc);
1680 
1681 		/*
1682 		 * SUCCESS here means that the sense code could be
1683 		 * evaluated and should be passed to the upper layers
1684 		 * for correct evaluation.
1685 		 * FAILED means the sense code could not be interpreted
1686 		 * and the device would need to be reset.
1687 		 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1688 		 * command would need to be retried.
1689 		 */
1690 		if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1691 			qc->flags |= ATA_QCFLAG_RETRY;
1692 			qc->err_mask |= AC_ERR_OTHER;
1693 		} else if (ret != SUCCESS) {
1694 			qc->err_mask |= AC_ERR_HSM;
1695 		}
1696 	}
1697 	if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1698 		action |= ATA_EH_RESET;
1699 
1700 	return action;
1701 }
1702 
ata_eh_categorize_error(unsigned int eflags,unsigned int err_mask,int * xfer_ok)1703 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1704 				   int *xfer_ok)
1705 {
1706 	int base = 0;
1707 
1708 	if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1709 		*xfer_ok = 1;
1710 
1711 	if (!*xfer_ok)
1712 		base = ATA_ECAT_DUBIOUS_NONE;
1713 
1714 	if (err_mask & AC_ERR_ATA_BUS)
1715 		return base + ATA_ECAT_ATA_BUS;
1716 
1717 	if (err_mask & AC_ERR_TIMEOUT)
1718 		return base + ATA_ECAT_TOUT_HSM;
1719 
1720 	if (eflags & ATA_EFLAG_IS_IO) {
1721 		if (err_mask & AC_ERR_HSM)
1722 			return base + ATA_ECAT_TOUT_HSM;
1723 		if ((err_mask &
1724 		     (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1725 			return base + ATA_ECAT_UNK_DEV;
1726 	}
1727 
1728 	return 0;
1729 }
1730 
1731 struct speed_down_verdict_arg {
1732 	u64 since;
1733 	int xfer_ok;
1734 	int nr_errors[ATA_ECAT_NR];
1735 };
1736 
speed_down_verdict_cb(struct ata_ering_entry * ent,void * void_arg)1737 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1738 {
1739 	struct speed_down_verdict_arg *arg = void_arg;
1740 	int cat;
1741 
1742 	if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1743 		return -1;
1744 
1745 	cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1746 				      &arg->xfer_ok);
1747 	arg->nr_errors[cat]++;
1748 
1749 	return 0;
1750 }
1751 
1752 /**
1753  *	ata_eh_speed_down_verdict - Determine speed down verdict
1754  *	@dev: Device of interest
1755  *
1756  *	This function examines error ring of @dev and determines
1757  *	whether NCQ needs to be turned off, transfer speed should be
1758  *	stepped down, or falling back to PIO is necessary.
1759  *
1760  *	ECAT_ATA_BUS	: ATA_BUS error for any command
1761  *
1762  *	ECAT_TOUT_HSM	: TIMEOUT for any command or HSM violation for
1763  *			  IO commands
1764  *
1765  *	ECAT_UNK_DEV	: Unknown DEV error for IO commands
1766  *
1767  *	ECAT_DUBIOUS_*	: Identical to above three but occurred while
1768  *			  data transfer hasn't been verified.
1769  *
1770  *	Verdicts are
1771  *
1772  *	NCQ_OFF		: Turn off NCQ.
1773  *
1774  *	SPEED_DOWN	: Speed down transfer speed but don't fall back
1775  *			  to PIO.
1776  *
1777  *	FALLBACK_TO_PIO	: Fall back to PIO.
1778  *
1779  *	Even if multiple verdicts are returned, only one action is
1780  *	taken per error.  An action triggered by non-DUBIOUS errors
1781  *	clears ering, while one triggered by DUBIOUS_* errors doesn't.
1782  *	This is to expedite speed down decisions right after device is
1783  *	initially configured.
1784  *
1785  *	The following are speed down rules.  #1 and #2 deal with
1786  *	DUBIOUS errors.
1787  *
1788  *	1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1789  *	   occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1790  *
1791  *	2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1792  *	   occurred during last 5 mins, NCQ_OFF.
1793  *
1794  *	3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1795  *	   occurred during last 5 mins, FALLBACK_TO_PIO
1796  *
1797  *	4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1798  *	   during last 10 mins, NCQ_OFF.
1799  *
1800  *	5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1801  *	   UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1802  *
1803  *	LOCKING:
1804  *	Inherited from caller.
1805  *
1806  *	RETURNS:
1807  *	OR of ATA_EH_SPDN_* flags.
1808  */
ata_eh_speed_down_verdict(struct ata_device * dev)1809 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1810 {
1811 	const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1812 	u64 j64 = get_jiffies_64();
1813 	struct speed_down_verdict_arg arg;
1814 	unsigned int verdict = 0;
1815 
1816 	/* scan past 5 mins of error history */
1817 	memset(&arg, 0, sizeof(arg));
1818 	arg.since = j64 - min(j64, j5mins);
1819 	ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1820 
1821 	if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1822 	    arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1823 		verdict |= ATA_EH_SPDN_SPEED_DOWN |
1824 			ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1825 
1826 	if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1827 	    arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1828 		verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1829 
1830 	if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1831 	    arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1832 	    arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1833 		verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1834 
1835 	/* scan past 10 mins of error history */
1836 	memset(&arg, 0, sizeof(arg));
1837 	arg.since = j64 - min(j64, j10mins);
1838 	ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1839 
1840 	if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1841 	    arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1842 		verdict |= ATA_EH_SPDN_NCQ_OFF;
1843 
1844 	if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1845 	    arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1846 	    arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1847 		verdict |= ATA_EH_SPDN_SPEED_DOWN;
1848 
1849 	return verdict;
1850 }
1851 
1852 /**
1853  *	ata_eh_speed_down - record error and speed down if necessary
1854  *	@dev: Failed device
1855  *	@eflags: mask of ATA_EFLAG_* flags
1856  *	@err_mask: err_mask of the error
1857  *
1858  *	Record error and examine error history to determine whether
1859  *	adjusting transmission speed is necessary.  It also sets
1860  *	transmission limits appropriately if such adjustment is
1861  *	necessary.
1862  *
1863  *	LOCKING:
1864  *	Kernel thread context (may sleep).
1865  *
1866  *	RETURNS:
1867  *	Determined recovery action.
1868  */
ata_eh_speed_down(struct ata_device * dev,unsigned int eflags,unsigned int err_mask)1869 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1870 				unsigned int eflags, unsigned int err_mask)
1871 {
1872 	struct ata_link *link = ata_dev_phys_link(dev);
1873 	int xfer_ok = 0;
1874 	unsigned int verdict;
1875 	unsigned int action = 0;
1876 
1877 	/* don't bother if Cat-0 error */
1878 	if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1879 		return 0;
1880 
1881 	/* record error and determine whether speed down is necessary */
1882 	ata_ering_record(&dev->ering, eflags, err_mask);
1883 	verdict = ata_eh_speed_down_verdict(dev);
1884 
1885 	/* turn off NCQ? */
1886 	if ((verdict & ATA_EH_SPDN_NCQ_OFF) && ata_ncq_enabled(dev)) {
1887 		dev->flags |= ATA_DFLAG_NCQ_OFF;
1888 		ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1889 		goto done;
1890 	}
1891 
1892 	/* speed down? */
1893 	if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1894 		/* speed down SATA link speed if possible */
1895 		if (sata_down_spd_limit(link, 0) == 0) {
1896 			action |= ATA_EH_RESET;
1897 			goto done;
1898 		}
1899 
1900 		/* lower transfer mode */
1901 		if (dev->spdn_cnt < 2) {
1902 			static const int dma_dnxfer_sel[] =
1903 				{ ATA_DNXFER_DMA, ATA_DNXFER_40C };
1904 			static const int pio_dnxfer_sel[] =
1905 				{ ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1906 			int sel;
1907 
1908 			if (dev->xfer_shift != ATA_SHIFT_PIO)
1909 				sel = dma_dnxfer_sel[dev->spdn_cnt];
1910 			else
1911 				sel = pio_dnxfer_sel[dev->spdn_cnt];
1912 
1913 			dev->spdn_cnt++;
1914 
1915 			if (ata_down_xfermask_limit(dev, sel) == 0) {
1916 				action |= ATA_EH_RESET;
1917 				goto done;
1918 			}
1919 		}
1920 	}
1921 
1922 	/* Fall back to PIO?  Slowing down to PIO is meaningless for
1923 	 * SATA ATA devices.  Consider it only for PATA and SATAPI.
1924 	 */
1925 	if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1926 	    (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1927 	    (dev->xfer_shift != ATA_SHIFT_PIO)) {
1928 		if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1929 			dev->spdn_cnt = 0;
1930 			action |= ATA_EH_RESET;
1931 			goto done;
1932 		}
1933 	}
1934 
1935 	return 0;
1936  done:
1937 	/* device has been slowed down, blow error history */
1938 	if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1939 		ata_ering_clear(&dev->ering);
1940 	return action;
1941 }
1942 
1943 /**
1944  *	ata_eh_worth_retry - analyze error and decide whether to retry
1945  *	@qc: qc to possibly retry
1946  *
1947  *	Look at the cause of the error and decide if a retry
1948  * 	might be useful or not.  We don't want to retry media errors
1949  *	because the drive itself has probably already taken 10-30 seconds
1950  *	doing its own internal retries before reporting the failure.
1951  */
ata_eh_worth_retry(struct ata_queued_cmd * qc)1952 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1953 {
1954 	if (qc->err_mask & AC_ERR_MEDIA)
1955 		return 0;	/* don't retry media errors */
1956 	if (qc->flags & ATA_QCFLAG_IO)
1957 		return 1;	/* otherwise retry anything from fs stack */
1958 	if (qc->err_mask & AC_ERR_INVALID)
1959 		return 0;	/* don't retry these */
1960 	return qc->err_mask != AC_ERR_DEV;  /* retry if not dev error */
1961 }
1962 
1963 /**
1964  *      ata_eh_quiet - check if we need to be quiet about a command error
1965  *      @qc: qc to check
1966  *
1967  *      Look at the qc flags anbd its scsi command request flags to determine
1968  *      if we need to be quiet about the command failure.
1969  */
ata_eh_quiet(struct ata_queued_cmd * qc)1970 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1971 {
1972 	if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1973 		qc->flags |= ATA_QCFLAG_QUIET;
1974 	return qc->flags & ATA_QCFLAG_QUIET;
1975 }
1976 
ata_eh_get_non_ncq_success_sense(struct ata_link * link)1977 static int ata_eh_get_non_ncq_success_sense(struct ata_link *link)
1978 {
1979 	struct ata_port *ap = link->ap;
1980 	struct ata_queued_cmd *qc;
1981 
1982 	qc = __ata_qc_from_tag(ap, link->active_tag);
1983 	if (!qc)
1984 		return -EIO;
1985 
1986 	if (!(qc->flags & ATA_QCFLAG_EH) ||
1987 	    !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
1988 	    qc->err_mask)
1989 		return -EIO;
1990 
1991 	if (!ata_eh_request_sense(qc))
1992 		return -EIO;
1993 
1994 	/*
1995 	 * No point in checking the return value, since the command has already
1996 	 * completed successfully.
1997 	 */
1998 	ata_eh_decide_disposition(qc);
1999 
2000 	return 0;
2001 }
2002 
ata_eh_get_success_sense(struct ata_link * link)2003 static void ata_eh_get_success_sense(struct ata_link *link)
2004 {
2005 	struct ata_eh_context *ehc = &link->eh_context;
2006 	struct ata_device *dev = link->device;
2007 	struct ata_port *ap = link->ap;
2008 	struct ata_queued_cmd *qc;
2009 	int tag, ret = 0;
2010 
2011 	if (!(ehc->i.dev_action[dev->devno] & ATA_EH_GET_SUCCESS_SENSE))
2012 		return;
2013 
2014 	/* if frozen, we can't do much */
2015 	if (ata_port_is_frozen(ap)) {
2016 		ata_dev_warn(dev,
2017 			"successful sense data available but port frozen\n");
2018 		goto out;
2019 	}
2020 
2021 	/*
2022 	 * If the link has sactive set, then we have outstanding NCQ commands
2023 	 * and have to read the Successful NCQ Commands log to get the sense
2024 	 * data. Otherwise, we are dealing with a non-NCQ command and use
2025 	 * request sense ext command to retrieve the sense data.
2026 	 */
2027 	if (link->sactive)
2028 		ret = ata_eh_get_ncq_success_sense(link);
2029 	else
2030 		ret = ata_eh_get_non_ncq_success_sense(link);
2031 	if (ret)
2032 		goto out;
2033 
2034 	ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
2035 	return;
2036 
2037 out:
2038 	/*
2039 	 * If we failed to get sense data for a successful command that ought to
2040 	 * have sense data, we cannot simply return BLK_STS_OK to user space.
2041 	 * This is because we can't know if the sense data that we couldn't get
2042 	 * was actually "DATA CURRENTLY UNAVAILABLE". Reporting such a command
2043 	 * as success to user space would result in a silent data corruption.
2044 	 * Thus, add a bogus ABORTED_COMMAND sense data to such commands, such
2045 	 * that SCSI will report these commands as BLK_STS_IOERR to user space.
2046 	 */
2047 	ata_qc_for_each_raw(ap, qc, tag) {
2048 		if (!(qc->flags & ATA_QCFLAG_EH) ||
2049 		    !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
2050 		    qc->err_mask ||
2051 		    ata_dev_phys_link(qc->dev) != link)
2052 			continue;
2053 
2054 		/* We managed to get sense for this success command, skip. */
2055 		if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2056 			continue;
2057 
2058 		/* This success command did not have any sense data, skip. */
2059 		if (!(qc->result_tf.status & ATA_SENSE))
2060 			continue;
2061 
2062 		/* This success command had sense data, but we failed to get. */
2063 		ata_scsi_set_sense(dev, qc->scsicmd, ABORTED_COMMAND, 0, 0);
2064 		qc->flags |= ATA_QCFLAG_SENSE_VALID;
2065 	}
2066 	ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
2067 }
2068 
2069 /**
2070  *	ata_eh_link_autopsy - analyze error and determine recovery action
2071  *	@link: host link to perform autopsy on
2072  *
2073  *	Analyze why @link failed and determine which recovery actions
2074  *	are needed.  This function also sets more detailed AC_ERR_*
2075  *	values and fills sense data for ATAPI CHECK SENSE.
2076  *
2077  *	LOCKING:
2078  *	Kernel thread context (may sleep).
2079  */
ata_eh_link_autopsy(struct ata_link * link)2080 static void ata_eh_link_autopsy(struct ata_link *link)
2081 {
2082 	struct ata_port *ap = link->ap;
2083 	struct ata_eh_context *ehc = &link->eh_context;
2084 	struct ata_queued_cmd *qc;
2085 	struct ata_device *dev;
2086 	unsigned int all_err_mask = 0, eflags = 0;
2087 	int tag, nr_failed = 0, nr_quiet = 0;
2088 	u32 serror;
2089 	int rc;
2090 
2091 	if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2092 		return;
2093 
2094 	/* obtain and analyze SError */
2095 	rc = sata_scr_read(link, SCR_ERROR, &serror);
2096 	if (rc == 0) {
2097 		ehc->i.serror |= serror;
2098 		ata_eh_analyze_serror(link);
2099 	} else if (rc != -EOPNOTSUPP) {
2100 		/* SError read failed, force reset and probing */
2101 		ehc->i.probe_mask |= ATA_ALL_DEVICES;
2102 		ehc->i.action |= ATA_EH_RESET;
2103 		ehc->i.err_mask |= AC_ERR_OTHER;
2104 	}
2105 
2106 	/* analyze NCQ failure */
2107 	ata_eh_analyze_ncq_error(link);
2108 
2109 	/*
2110 	 * Check if this was a successful command that simply needs sense data.
2111 	 * Since the sense data is not part of the completion, we need to fetch
2112 	 * it using an additional command. Since this can't be done from irq
2113 	 * context, the sense data for successful commands are fetched by EH.
2114 	 */
2115 	ata_eh_get_success_sense(link);
2116 
2117 	/* any real error trumps AC_ERR_OTHER */
2118 	if (ehc->i.err_mask & ~AC_ERR_OTHER)
2119 		ehc->i.err_mask &= ~AC_ERR_OTHER;
2120 
2121 	all_err_mask |= ehc->i.err_mask;
2122 
2123 	ata_qc_for_each_raw(ap, qc, tag) {
2124 		if (!(qc->flags & ATA_QCFLAG_EH) ||
2125 		    qc->flags & ATA_QCFLAG_RETRY ||
2126 		    qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD ||
2127 		    ata_dev_phys_link(qc->dev) != link)
2128 			continue;
2129 
2130 		/* inherit upper level err_mask */
2131 		qc->err_mask |= ehc->i.err_mask;
2132 
2133 		/* analyze TF */
2134 		ehc->i.action |= ata_eh_analyze_tf(qc);
2135 
2136 		/* DEV errors are probably spurious in case of ATA_BUS error */
2137 		if (qc->err_mask & AC_ERR_ATA_BUS)
2138 			qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2139 					  AC_ERR_INVALID);
2140 
2141 		/* any real error trumps unknown error */
2142 		if (qc->err_mask & ~AC_ERR_OTHER)
2143 			qc->err_mask &= ~AC_ERR_OTHER;
2144 
2145 		/*
2146 		 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2147 		 * layers will determine whether the command is worth retrying
2148 		 * based on the sense data and device class/type. Otherwise,
2149 		 * determine directly if the command is worth retrying using its
2150 		 * error mask and flags.
2151 		 */
2152 		if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2153 			qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2154 		else if (ata_eh_worth_retry(qc))
2155 			qc->flags |= ATA_QCFLAG_RETRY;
2156 
2157 		/* accumulate error info */
2158 		ehc->i.dev = qc->dev;
2159 		all_err_mask |= qc->err_mask;
2160 		if (qc->flags & ATA_QCFLAG_IO)
2161 			eflags |= ATA_EFLAG_IS_IO;
2162 		trace_ata_eh_link_autopsy_qc(qc);
2163 
2164 		/* Count quiet errors */
2165 		if (ata_eh_quiet(qc))
2166 			nr_quiet++;
2167 		nr_failed++;
2168 	}
2169 
2170 	/* If all failed commands requested silence, then be quiet */
2171 	if (nr_quiet == nr_failed)
2172 		ehc->i.flags |= ATA_EHI_QUIET;
2173 
2174 	/* enforce default EH actions */
2175 	if (ata_port_is_frozen(ap) ||
2176 	    all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2177 		ehc->i.action |= ATA_EH_RESET;
2178 	else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2179 		 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2180 		ehc->i.action |= ATA_EH_REVALIDATE;
2181 
2182 	/* If we have offending qcs and the associated failed device,
2183 	 * perform per-dev EH action only on the offending device.
2184 	 */
2185 	if (ehc->i.dev) {
2186 		ehc->i.dev_action[ehc->i.dev->devno] |=
2187 			ehc->i.action & ATA_EH_PERDEV_MASK;
2188 		ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2189 	}
2190 
2191 	/* propagate timeout to host link */
2192 	if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2193 		ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2194 
2195 	/* record error and consider speeding down */
2196 	dev = ehc->i.dev;
2197 	if (!dev && ((ata_link_max_devices(link) == 1 &&
2198 		      ata_dev_enabled(link->device))))
2199 	    dev = link->device;
2200 
2201 	if (dev) {
2202 		if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2203 			eflags |= ATA_EFLAG_DUBIOUS_XFER;
2204 		ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2205 		trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2206 	}
2207 }
2208 
2209 /**
2210  *	ata_eh_autopsy - analyze error and determine recovery action
2211  *	@ap: host port to perform autopsy on
2212  *
2213  *	Analyze all links of @ap and determine why they failed and
2214  *	which recovery actions are needed.
2215  *
2216  *	LOCKING:
2217  *	Kernel thread context (may sleep).
2218  */
ata_eh_autopsy(struct ata_port * ap)2219 void ata_eh_autopsy(struct ata_port *ap)
2220 {
2221 	struct ata_link *link;
2222 
2223 	ata_for_each_link(link, ap, EDGE)
2224 		ata_eh_link_autopsy(link);
2225 
2226 	/* Handle the frigging slave link.  Autopsy is done similarly
2227 	 * but actions and flags are transferred over to the master
2228 	 * link and handled from there.
2229 	 */
2230 	if (ap->slave_link) {
2231 		struct ata_eh_context *mehc = &ap->link.eh_context;
2232 		struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2233 
2234 		/* transfer control flags from master to slave */
2235 		sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2236 
2237 		/* perform autopsy on the slave link */
2238 		ata_eh_link_autopsy(ap->slave_link);
2239 
2240 		/* transfer actions from slave to master and clear slave */
2241 		ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2242 		mehc->i.action		|= sehc->i.action;
2243 		mehc->i.dev_action[1]	|= sehc->i.dev_action[1];
2244 		mehc->i.flags		|= sehc->i.flags;
2245 		ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2246 	}
2247 
2248 	/* Autopsy of fanout ports can affect host link autopsy.
2249 	 * Perform host link autopsy last.
2250 	 */
2251 	if (sata_pmp_attached(ap))
2252 		ata_eh_link_autopsy(&ap->link);
2253 }
2254 
2255 /**
2256  *	ata_get_cmd_name - get name for ATA command
2257  *	@command: ATA command code to get name for
2258  *
2259  *	Return a textual name of the given command or "unknown"
2260  *
2261  *	LOCKING:
2262  *	None
2263  */
ata_get_cmd_name(u8 command)2264 const char *ata_get_cmd_name(u8 command)
2265 {
2266 #ifdef CONFIG_ATA_VERBOSE_ERROR
2267 	static const struct
2268 	{
2269 		u8 command;
2270 		const char *text;
2271 	} cmd_descr[] = {
2272 		{ ATA_CMD_DEV_RESET,		"DEVICE RESET" },
2273 		{ ATA_CMD_CHK_POWER,		"CHECK POWER MODE" },
2274 		{ ATA_CMD_STANDBY,		"STANDBY" },
2275 		{ ATA_CMD_IDLE,			"IDLE" },
2276 		{ ATA_CMD_EDD,			"EXECUTE DEVICE DIAGNOSTIC" },
2277 		{ ATA_CMD_DOWNLOAD_MICRO,	"DOWNLOAD MICROCODE" },
2278 		{ ATA_CMD_DOWNLOAD_MICRO_DMA,	"DOWNLOAD MICROCODE DMA" },
2279 		{ ATA_CMD_NOP,			"NOP" },
2280 		{ ATA_CMD_FLUSH,		"FLUSH CACHE" },
2281 		{ ATA_CMD_FLUSH_EXT,		"FLUSH CACHE EXT" },
2282 		{ ATA_CMD_ID_ATA,		"IDENTIFY DEVICE" },
2283 		{ ATA_CMD_ID_ATAPI,		"IDENTIFY PACKET DEVICE" },
2284 		{ ATA_CMD_SERVICE,		"SERVICE" },
2285 		{ ATA_CMD_READ,			"READ DMA" },
2286 		{ ATA_CMD_READ_EXT,		"READ DMA EXT" },
2287 		{ ATA_CMD_READ_QUEUED,		"READ DMA QUEUED" },
2288 		{ ATA_CMD_READ_STREAM_EXT,	"READ STREAM EXT" },
2289 		{ ATA_CMD_READ_STREAM_DMA_EXT,  "READ STREAM DMA EXT" },
2290 		{ ATA_CMD_WRITE,		"WRITE DMA" },
2291 		{ ATA_CMD_WRITE_EXT,		"WRITE DMA EXT" },
2292 		{ ATA_CMD_WRITE_QUEUED,		"WRITE DMA QUEUED EXT" },
2293 		{ ATA_CMD_WRITE_STREAM_EXT,	"WRITE STREAM EXT" },
2294 		{ ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2295 		{ ATA_CMD_WRITE_FUA_EXT,	"WRITE DMA FUA EXT" },
2296 		{ ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2297 		{ ATA_CMD_FPDMA_READ,		"READ FPDMA QUEUED" },
2298 		{ ATA_CMD_FPDMA_WRITE,		"WRITE FPDMA QUEUED" },
2299 		{ ATA_CMD_NCQ_NON_DATA,		"NCQ NON-DATA" },
2300 		{ ATA_CMD_FPDMA_SEND,		"SEND FPDMA QUEUED" },
2301 		{ ATA_CMD_FPDMA_RECV,		"RECEIVE FPDMA QUEUED" },
2302 		{ ATA_CMD_PIO_READ,		"READ SECTOR(S)" },
2303 		{ ATA_CMD_PIO_READ_EXT,		"READ SECTOR(S) EXT" },
2304 		{ ATA_CMD_PIO_WRITE,		"WRITE SECTOR(S)" },
2305 		{ ATA_CMD_PIO_WRITE_EXT,	"WRITE SECTOR(S) EXT" },
2306 		{ ATA_CMD_READ_MULTI,		"READ MULTIPLE" },
2307 		{ ATA_CMD_READ_MULTI_EXT,	"READ MULTIPLE EXT" },
2308 		{ ATA_CMD_WRITE_MULTI,		"WRITE MULTIPLE" },
2309 		{ ATA_CMD_WRITE_MULTI_EXT,	"WRITE MULTIPLE EXT" },
2310 		{ ATA_CMD_WRITE_MULTI_FUA_EXT,	"WRITE MULTIPLE FUA EXT" },
2311 		{ ATA_CMD_SET_FEATURES,		"SET FEATURES" },
2312 		{ ATA_CMD_SET_MULTI,		"SET MULTIPLE MODE" },
2313 		{ ATA_CMD_VERIFY,		"READ VERIFY SECTOR(S)" },
2314 		{ ATA_CMD_VERIFY_EXT,		"READ VERIFY SECTOR(S) EXT" },
2315 		{ ATA_CMD_WRITE_UNCORR_EXT,	"WRITE UNCORRECTABLE EXT" },
2316 		{ ATA_CMD_STANDBYNOW1,		"STANDBY IMMEDIATE" },
2317 		{ ATA_CMD_IDLEIMMEDIATE,	"IDLE IMMEDIATE" },
2318 		{ ATA_CMD_SLEEP,		"SLEEP" },
2319 		{ ATA_CMD_INIT_DEV_PARAMS,	"INITIALIZE DEVICE PARAMETERS" },
2320 		{ ATA_CMD_READ_NATIVE_MAX,	"READ NATIVE MAX ADDRESS" },
2321 		{ ATA_CMD_READ_NATIVE_MAX_EXT,	"READ NATIVE MAX ADDRESS EXT" },
2322 		{ ATA_CMD_SET_MAX,		"SET MAX ADDRESS" },
2323 		{ ATA_CMD_SET_MAX_EXT,		"SET MAX ADDRESS EXT" },
2324 		{ ATA_CMD_READ_LOG_EXT,		"READ LOG EXT" },
2325 		{ ATA_CMD_WRITE_LOG_EXT,	"WRITE LOG EXT" },
2326 		{ ATA_CMD_READ_LOG_DMA_EXT,	"READ LOG DMA EXT" },
2327 		{ ATA_CMD_WRITE_LOG_DMA_EXT,	"WRITE LOG DMA EXT" },
2328 		{ ATA_CMD_TRUSTED_NONDATA,	"TRUSTED NON-DATA" },
2329 		{ ATA_CMD_TRUSTED_RCV,		"TRUSTED RECEIVE" },
2330 		{ ATA_CMD_TRUSTED_RCV_DMA,	"TRUSTED RECEIVE DMA" },
2331 		{ ATA_CMD_TRUSTED_SND,		"TRUSTED SEND" },
2332 		{ ATA_CMD_TRUSTED_SND_DMA,	"TRUSTED SEND DMA" },
2333 		{ ATA_CMD_PMP_READ,		"READ BUFFER" },
2334 		{ ATA_CMD_PMP_READ_DMA,		"READ BUFFER DMA" },
2335 		{ ATA_CMD_PMP_WRITE,		"WRITE BUFFER" },
2336 		{ ATA_CMD_PMP_WRITE_DMA,	"WRITE BUFFER DMA" },
2337 		{ ATA_CMD_CONF_OVERLAY,		"DEVICE CONFIGURATION OVERLAY" },
2338 		{ ATA_CMD_SEC_SET_PASS,		"SECURITY SET PASSWORD" },
2339 		{ ATA_CMD_SEC_UNLOCK,		"SECURITY UNLOCK" },
2340 		{ ATA_CMD_SEC_ERASE_PREP,	"SECURITY ERASE PREPARE" },
2341 		{ ATA_CMD_SEC_ERASE_UNIT,	"SECURITY ERASE UNIT" },
2342 		{ ATA_CMD_SEC_FREEZE_LOCK,	"SECURITY FREEZE LOCK" },
2343 		{ ATA_CMD_SEC_DISABLE_PASS,	"SECURITY DISABLE PASSWORD" },
2344 		{ ATA_CMD_CONFIG_STREAM,	"CONFIGURE STREAM" },
2345 		{ ATA_CMD_SMART,		"SMART" },
2346 		{ ATA_CMD_MEDIA_LOCK,		"DOOR LOCK" },
2347 		{ ATA_CMD_MEDIA_UNLOCK,		"DOOR UNLOCK" },
2348 		{ ATA_CMD_DSM,			"DATA SET MANAGEMENT" },
2349 		{ ATA_CMD_CHK_MED_CRD_TYP,	"CHECK MEDIA CARD TYPE" },
2350 		{ ATA_CMD_CFA_REQ_EXT_ERR,	"CFA REQUEST EXTENDED ERROR" },
2351 		{ ATA_CMD_CFA_WRITE_NE,		"CFA WRITE SECTORS WITHOUT ERASE" },
2352 		{ ATA_CMD_CFA_TRANS_SECT,	"CFA TRANSLATE SECTOR" },
2353 		{ ATA_CMD_CFA_ERASE,		"CFA ERASE SECTORS" },
2354 		{ ATA_CMD_CFA_WRITE_MULT_NE,	"CFA WRITE MULTIPLE WITHOUT ERASE" },
2355 		{ ATA_CMD_REQ_SENSE_DATA,	"REQUEST SENSE DATA EXT" },
2356 		{ ATA_CMD_SANITIZE_DEVICE,	"SANITIZE DEVICE" },
2357 		{ ATA_CMD_ZAC_MGMT_IN,		"ZAC MANAGEMENT IN" },
2358 		{ ATA_CMD_ZAC_MGMT_OUT,		"ZAC MANAGEMENT OUT" },
2359 		{ ATA_CMD_READ_LONG,		"READ LONG (with retries)" },
2360 		{ ATA_CMD_READ_LONG_ONCE,	"READ LONG (without retries)" },
2361 		{ ATA_CMD_WRITE_LONG,		"WRITE LONG (with retries)" },
2362 		{ ATA_CMD_WRITE_LONG_ONCE,	"WRITE LONG (without retries)" },
2363 		{ ATA_CMD_RESTORE,		"RECALIBRATE" },
2364 		{ 0,				NULL } /* terminate list */
2365 	};
2366 
2367 	unsigned int i;
2368 	for (i = 0; cmd_descr[i].text; i++)
2369 		if (cmd_descr[i].command == command)
2370 			return cmd_descr[i].text;
2371 #endif
2372 
2373 	return "unknown";
2374 }
2375 EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2376 
2377 /**
2378  *	ata_eh_link_report - report error handling to user
2379  *	@link: ATA link EH is going on
2380  *
2381  *	Report EH to user.
2382  *
2383  *	LOCKING:
2384  *	None.
2385  */
ata_eh_link_report(struct ata_link * link)2386 static void ata_eh_link_report(struct ata_link *link)
2387 {
2388 	struct ata_port *ap = link->ap;
2389 	struct ata_eh_context *ehc = &link->eh_context;
2390 	struct ata_queued_cmd *qc;
2391 	const char *frozen, *desc;
2392 	char tries_buf[16] = "";
2393 	int tag, nr_failed = 0;
2394 
2395 	if (ehc->i.flags & ATA_EHI_QUIET)
2396 		return;
2397 
2398 	desc = NULL;
2399 	if (ehc->i.desc[0] != '\0')
2400 		desc = ehc->i.desc;
2401 
2402 	ata_qc_for_each_raw(ap, qc, tag) {
2403 		if (!(qc->flags & ATA_QCFLAG_EH) ||
2404 		    ata_dev_phys_link(qc->dev) != link ||
2405 		    ((qc->flags & ATA_QCFLAG_QUIET) &&
2406 		     qc->err_mask == AC_ERR_DEV))
2407 			continue;
2408 		if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2409 			continue;
2410 
2411 		nr_failed++;
2412 	}
2413 
2414 	if (!nr_failed && !ehc->i.err_mask)
2415 		return;
2416 
2417 	frozen = "";
2418 	if (ata_port_is_frozen(ap))
2419 		frozen = " frozen";
2420 
2421 	if (ap->eh_tries < ATA_EH_MAX_TRIES)
2422 		snprintf(tries_buf, sizeof(tries_buf), " t%d",
2423 			 ap->eh_tries);
2424 
2425 	if (ehc->i.dev) {
2426 		ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2427 			    "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2428 			    ehc->i.err_mask, link->sactive, ehc->i.serror,
2429 			    ehc->i.action, frozen, tries_buf);
2430 		if (desc)
2431 			ata_dev_err(ehc->i.dev, "%s\n", desc);
2432 	} else {
2433 		ata_link_err(link, "exception Emask 0x%x "
2434 			     "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2435 			     ehc->i.err_mask, link->sactive, ehc->i.serror,
2436 			     ehc->i.action, frozen, tries_buf);
2437 		if (desc)
2438 			ata_link_err(link, "%s\n", desc);
2439 	}
2440 
2441 #ifdef CONFIG_ATA_VERBOSE_ERROR
2442 	if (ehc->i.serror)
2443 		ata_link_err(link,
2444 		  "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2445 		  ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2446 		  ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2447 		  ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2448 		  ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2449 		  ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2450 		  ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2451 		  ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2452 		  ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2453 		  ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2454 		  ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2455 		  ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2456 		  ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2457 		  ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2458 		  ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2459 		  ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2460 		  ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2461 		  ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2462 #endif
2463 
2464 	ata_qc_for_each_raw(ap, qc, tag) {
2465 		struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2466 		char data_buf[20] = "";
2467 		char cdb_buf[70] = "";
2468 
2469 		if (!(qc->flags & ATA_QCFLAG_EH) ||
2470 		    ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2471 			continue;
2472 
2473 		if (qc->dma_dir != DMA_NONE) {
2474 			static const char *dma_str[] = {
2475 				[DMA_BIDIRECTIONAL]	= "bidi",
2476 				[DMA_TO_DEVICE]		= "out",
2477 				[DMA_FROM_DEVICE]	= "in",
2478 			};
2479 			const char *prot_str = NULL;
2480 
2481 			switch (qc->tf.protocol) {
2482 			case ATA_PROT_UNKNOWN:
2483 				prot_str = "unknown";
2484 				break;
2485 			case ATA_PROT_NODATA:
2486 				prot_str = "nodata";
2487 				break;
2488 			case ATA_PROT_PIO:
2489 				prot_str = "pio";
2490 				break;
2491 			case ATA_PROT_DMA:
2492 				prot_str = "dma";
2493 				break;
2494 			case ATA_PROT_NCQ:
2495 				prot_str = "ncq dma";
2496 				break;
2497 			case ATA_PROT_NCQ_NODATA:
2498 				prot_str = "ncq nodata";
2499 				break;
2500 			case ATAPI_PROT_NODATA:
2501 				prot_str = "nodata";
2502 				break;
2503 			case ATAPI_PROT_PIO:
2504 				prot_str = "pio";
2505 				break;
2506 			case ATAPI_PROT_DMA:
2507 				prot_str = "dma";
2508 				break;
2509 			}
2510 			snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2511 				 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2512 		}
2513 
2514 		if (ata_is_atapi(qc->tf.protocol)) {
2515 			const u8 *cdb = qc->cdb;
2516 			size_t cdb_len = qc->dev->cdb_len;
2517 
2518 			if (qc->scsicmd) {
2519 				cdb = qc->scsicmd->cmnd;
2520 				cdb_len = qc->scsicmd->cmd_len;
2521 			}
2522 			__scsi_format_command(cdb_buf, sizeof(cdb_buf),
2523 					      cdb, cdb_len);
2524 		} else
2525 			ata_dev_err(qc->dev, "failed command: %s\n",
2526 				    ata_get_cmd_name(cmd->command));
2527 
2528 		ata_dev_err(qc->dev,
2529 			"cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2530 			"tag %d%s\n         %s"
2531 			"res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2532 			"Emask 0x%x (%s)%s\n",
2533 			cmd->command, cmd->feature, cmd->nsect,
2534 			cmd->lbal, cmd->lbam, cmd->lbah,
2535 			cmd->hob_feature, cmd->hob_nsect,
2536 			cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2537 			cmd->device, qc->tag, data_buf, cdb_buf,
2538 			res->status, res->error, res->nsect,
2539 			res->lbal, res->lbam, res->lbah,
2540 			res->hob_feature, res->hob_nsect,
2541 			res->hob_lbal, res->hob_lbam, res->hob_lbah,
2542 			res->device, qc->err_mask, ata_err_string(qc->err_mask),
2543 			qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2544 
2545 #ifdef CONFIG_ATA_VERBOSE_ERROR
2546 		if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2547 				   ATA_SENSE | ATA_ERR)) {
2548 			if (res->status & ATA_BUSY)
2549 				ata_dev_err(qc->dev, "status: { Busy }\n");
2550 			else
2551 				ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2552 				  res->status & ATA_DRDY ? "DRDY " : "",
2553 				  res->status & ATA_DF ? "DF " : "",
2554 				  res->status & ATA_DRQ ? "DRQ " : "",
2555 				  res->status & ATA_SENSE ? "SENSE " : "",
2556 				  res->status & ATA_ERR ? "ERR " : "");
2557 		}
2558 
2559 		if (cmd->command != ATA_CMD_PACKET &&
2560 		    (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2561 				   ATA_ABORTED)))
2562 			ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2563 				    res->error & ATA_ICRC ? "ICRC " : "",
2564 				    res->error & ATA_UNC ? "UNC " : "",
2565 				    res->error & ATA_AMNF ? "AMNF " : "",
2566 				    res->error & ATA_IDNF ? "IDNF " : "",
2567 				    res->error & ATA_ABORTED ? "ABRT " : "");
2568 #endif
2569 	}
2570 }
2571 
2572 /**
2573  *	ata_eh_report - report error handling to user
2574  *	@ap: ATA port to report EH about
2575  *
2576  *	Report EH to user.
2577  *
2578  *	LOCKING:
2579  *	None.
2580  */
ata_eh_report(struct ata_port * ap)2581 void ata_eh_report(struct ata_port *ap)
2582 {
2583 	struct ata_link *link;
2584 
2585 	ata_for_each_link(link, ap, HOST_FIRST)
2586 		ata_eh_link_report(link);
2587 }
2588 
ata_do_reset(struct ata_link * link,ata_reset_fn_t reset,unsigned int * classes,unsigned long deadline,bool clear_classes)2589 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2590 			unsigned int *classes, unsigned long deadline,
2591 			bool clear_classes)
2592 {
2593 	struct ata_device *dev;
2594 
2595 	if (clear_classes)
2596 		ata_for_each_dev(dev, link, ALL)
2597 			classes[dev->devno] = ATA_DEV_UNKNOWN;
2598 
2599 	return reset(link, classes, deadline);
2600 }
2601 
ata_eh_followup_srst_needed(struct ata_link * link,int rc)2602 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2603 {
2604 	if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2605 		return 0;
2606 	if (rc == -EAGAIN)
2607 		return 1;
2608 	if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2609 		return 1;
2610 	return 0;
2611 }
2612 
ata_eh_reset(struct ata_link * link,int classify,ata_prereset_fn_t prereset,ata_reset_fn_t softreset,ata_reset_fn_t hardreset,ata_postreset_fn_t postreset)2613 int ata_eh_reset(struct ata_link *link, int classify,
2614 		 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2615 		 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2616 {
2617 	struct ata_port *ap = link->ap;
2618 	struct ata_link *slave = ap->slave_link;
2619 	struct ata_eh_context *ehc = &link->eh_context;
2620 	struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2621 	unsigned int *classes = ehc->classes;
2622 	unsigned int lflags = link->flags;
2623 	int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2624 	int max_tries = 0, try = 0;
2625 	struct ata_link *failed_link;
2626 	struct ata_device *dev;
2627 	unsigned long deadline, now;
2628 	ata_reset_fn_t reset;
2629 	unsigned long flags;
2630 	u32 sstatus;
2631 	int nr_unknown, rc;
2632 
2633 	/*
2634 	 * Prepare to reset
2635 	 */
2636 	while (ata_eh_reset_timeouts[max_tries] != UINT_MAX)
2637 		max_tries++;
2638 	if (link->flags & ATA_LFLAG_RST_ONCE)
2639 		max_tries = 1;
2640 	if (link->flags & ATA_LFLAG_NO_HRST)
2641 		hardreset = NULL;
2642 	if (link->flags & ATA_LFLAG_NO_SRST)
2643 		softreset = NULL;
2644 
2645 	/* make sure each reset attempt is at least COOL_DOWN apart */
2646 	if (ehc->i.flags & ATA_EHI_DID_RESET) {
2647 		now = jiffies;
2648 		WARN_ON(time_after(ehc->last_reset, now));
2649 		deadline = ata_deadline(ehc->last_reset,
2650 					ATA_EH_RESET_COOL_DOWN);
2651 		if (time_before(now, deadline))
2652 			schedule_timeout_uninterruptible(deadline - now);
2653 	}
2654 
2655 	spin_lock_irqsave(ap->lock, flags);
2656 	ap->pflags |= ATA_PFLAG_RESETTING;
2657 	spin_unlock_irqrestore(ap->lock, flags);
2658 
2659 	ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2660 
2661 	ata_for_each_dev(dev, link, ALL) {
2662 		/* If we issue an SRST then an ATA drive (not ATAPI)
2663 		 * may change configuration and be in PIO0 timing. If
2664 		 * we do a hard reset (or are coming from power on)
2665 		 * this is true for ATA or ATAPI. Until we've set a
2666 		 * suitable controller mode we should not touch the
2667 		 * bus as we may be talking too fast.
2668 		 */
2669 		dev->pio_mode = XFER_PIO_0;
2670 		dev->dma_mode = 0xff;
2671 
2672 		/* If the controller has a pio mode setup function
2673 		 * then use it to set the chipset to rights. Don't
2674 		 * touch the DMA setup as that will be dealt with when
2675 		 * configuring devices.
2676 		 */
2677 		if (ap->ops->set_piomode)
2678 			ap->ops->set_piomode(ap, dev);
2679 	}
2680 
2681 	/* prefer hardreset */
2682 	reset = NULL;
2683 	ehc->i.action &= ~ATA_EH_RESET;
2684 	if (hardreset) {
2685 		reset = hardreset;
2686 		ehc->i.action |= ATA_EH_HARDRESET;
2687 	} else if (softreset) {
2688 		reset = softreset;
2689 		ehc->i.action |= ATA_EH_SOFTRESET;
2690 	}
2691 
2692 	if (prereset) {
2693 		unsigned long deadline = ata_deadline(jiffies,
2694 						      ATA_EH_PRERESET_TIMEOUT);
2695 
2696 		if (slave) {
2697 			sehc->i.action &= ~ATA_EH_RESET;
2698 			sehc->i.action |= ehc->i.action;
2699 		}
2700 
2701 		rc = prereset(link, deadline);
2702 
2703 		/* If present, do prereset on slave link too.  Reset
2704 		 * is skipped iff both master and slave links report
2705 		 * -ENOENT or clear ATA_EH_RESET.
2706 		 */
2707 		if (slave && (rc == 0 || rc == -ENOENT)) {
2708 			int tmp;
2709 
2710 			tmp = prereset(slave, deadline);
2711 			if (tmp != -ENOENT)
2712 				rc = tmp;
2713 
2714 			ehc->i.action |= sehc->i.action;
2715 		}
2716 
2717 		if (rc) {
2718 			if (rc == -ENOENT) {
2719 				ata_link_dbg(link, "port disabled--ignoring\n");
2720 				ehc->i.action &= ~ATA_EH_RESET;
2721 
2722 				ata_for_each_dev(dev, link, ALL)
2723 					classes[dev->devno] = ATA_DEV_NONE;
2724 
2725 				rc = 0;
2726 			} else
2727 				ata_link_err(link,
2728 					     "prereset failed (errno=%d)\n",
2729 					     rc);
2730 			goto out;
2731 		}
2732 
2733 		/* prereset() might have cleared ATA_EH_RESET.  If so,
2734 		 * bang classes, thaw and return.
2735 		 */
2736 		if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2737 			ata_for_each_dev(dev, link, ALL)
2738 				classes[dev->devno] = ATA_DEV_NONE;
2739 			if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2740 				ata_eh_thaw_port(ap);
2741 			rc = 0;
2742 			goto out;
2743 		}
2744 	}
2745 
2746  retry:
2747 	/*
2748 	 * Perform reset
2749 	 */
2750 	if (ata_is_host_link(link))
2751 		ata_eh_freeze_port(ap);
2752 
2753 	deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2754 
2755 	if (reset) {
2756 		if (verbose)
2757 			ata_link_info(link, "%s resetting link\n",
2758 				      reset == softreset ? "soft" : "hard");
2759 
2760 		/* mark that this EH session started with reset */
2761 		ehc->last_reset = jiffies;
2762 		if (reset == hardreset) {
2763 			ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2764 			trace_ata_link_hardreset_begin(link, classes, deadline);
2765 		} else {
2766 			ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2767 			trace_ata_link_softreset_begin(link, classes, deadline);
2768 		}
2769 
2770 		rc = ata_do_reset(link, reset, classes, deadline, true);
2771 		if (reset == hardreset)
2772 			trace_ata_link_hardreset_end(link, classes, rc);
2773 		else
2774 			trace_ata_link_softreset_end(link, classes, rc);
2775 		if (rc && rc != -EAGAIN) {
2776 			failed_link = link;
2777 			goto fail;
2778 		}
2779 
2780 		/* hardreset slave link if existent */
2781 		if (slave && reset == hardreset) {
2782 			int tmp;
2783 
2784 			if (verbose)
2785 				ata_link_info(slave, "hard resetting link\n");
2786 
2787 			ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2788 			trace_ata_slave_hardreset_begin(slave, classes,
2789 							deadline);
2790 			tmp = ata_do_reset(slave, reset, classes, deadline,
2791 					   false);
2792 			trace_ata_slave_hardreset_end(slave, classes, tmp);
2793 			switch (tmp) {
2794 			case -EAGAIN:
2795 				rc = -EAGAIN;
2796 				break;
2797 			case 0:
2798 				break;
2799 			default:
2800 				failed_link = slave;
2801 				rc = tmp;
2802 				goto fail;
2803 			}
2804 		}
2805 
2806 		/* perform follow-up SRST if necessary */
2807 		if (reset == hardreset &&
2808 		    ata_eh_followup_srst_needed(link, rc)) {
2809 			reset = softreset;
2810 
2811 			if (!reset) {
2812 				ata_link_err(link,
2813 	     "follow-up softreset required but no softreset available\n");
2814 				failed_link = link;
2815 				rc = -EINVAL;
2816 				goto fail;
2817 			}
2818 
2819 			ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2820 			trace_ata_link_softreset_begin(link, classes, deadline);
2821 			rc = ata_do_reset(link, reset, classes, deadline, true);
2822 			trace_ata_link_softreset_end(link, classes, rc);
2823 			if (rc) {
2824 				failed_link = link;
2825 				goto fail;
2826 			}
2827 		}
2828 	} else {
2829 		if (verbose)
2830 			ata_link_info(link,
2831 	"no reset method available, skipping reset\n");
2832 		if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2833 			lflags |= ATA_LFLAG_ASSUME_ATA;
2834 	}
2835 
2836 	/*
2837 	 * Post-reset processing
2838 	 */
2839 	ata_for_each_dev(dev, link, ALL) {
2840 		/* After the reset, the device state is PIO 0 and the
2841 		 * controller state is undefined.  Reset also wakes up
2842 		 * drives from sleeping mode.
2843 		 */
2844 		dev->pio_mode = XFER_PIO_0;
2845 		dev->flags &= ~ATA_DFLAG_SLEEPING;
2846 
2847 		if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2848 			continue;
2849 
2850 		/* apply class override */
2851 		if (lflags & ATA_LFLAG_ASSUME_ATA)
2852 			classes[dev->devno] = ATA_DEV_ATA;
2853 		else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2854 			classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2855 	}
2856 
2857 	/* record current link speed */
2858 	if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2859 		link->sata_spd = (sstatus >> 4) & 0xf;
2860 	if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2861 		slave->sata_spd = (sstatus >> 4) & 0xf;
2862 
2863 	/* thaw the port */
2864 	if (ata_is_host_link(link))
2865 		ata_eh_thaw_port(ap);
2866 
2867 	/* postreset() should clear hardware SError.  Although SError
2868 	 * is cleared during link resume, clearing SError here is
2869 	 * necessary as some PHYs raise hotplug events after SRST.
2870 	 * This introduces race condition where hotplug occurs between
2871 	 * reset and here.  This race is mediated by cross checking
2872 	 * link onlineness and classification result later.
2873 	 */
2874 	if (postreset) {
2875 		postreset(link, classes);
2876 		trace_ata_link_postreset(link, classes, rc);
2877 		if (slave) {
2878 			postreset(slave, classes);
2879 			trace_ata_slave_postreset(slave, classes, rc);
2880 		}
2881 	}
2882 
2883 	/* clear cached SError */
2884 	spin_lock_irqsave(link->ap->lock, flags);
2885 	link->eh_info.serror = 0;
2886 	if (slave)
2887 		slave->eh_info.serror = 0;
2888 	spin_unlock_irqrestore(link->ap->lock, flags);
2889 
2890 	/*
2891 	 * Make sure onlineness and classification result correspond.
2892 	 * Hotplug could have happened during reset and some
2893 	 * controllers fail to wait while a drive is spinning up after
2894 	 * being hotplugged causing misdetection.  By cross checking
2895 	 * link on/offlineness and classification result, those
2896 	 * conditions can be reliably detected and retried.
2897 	 */
2898 	nr_unknown = 0;
2899 	ata_for_each_dev(dev, link, ALL) {
2900 		if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2901 			if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2902 				ata_dev_dbg(dev, "link online but device misclassified\n");
2903 				classes[dev->devno] = ATA_DEV_NONE;
2904 				nr_unknown++;
2905 			}
2906 		} else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2907 			if (ata_class_enabled(classes[dev->devno]))
2908 				ata_dev_dbg(dev,
2909 					    "link offline, clearing class %d to NONE\n",
2910 					    classes[dev->devno]);
2911 			classes[dev->devno] = ATA_DEV_NONE;
2912 		} else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2913 			ata_dev_dbg(dev,
2914 				    "link status unknown, clearing UNKNOWN to NONE\n");
2915 			classes[dev->devno] = ATA_DEV_NONE;
2916 		}
2917 	}
2918 
2919 	if (classify && nr_unknown) {
2920 		if (try < max_tries) {
2921 			ata_link_warn(link,
2922 				      "link online but %d devices misclassified, retrying\n",
2923 				      nr_unknown);
2924 			failed_link = link;
2925 			rc = -EAGAIN;
2926 			goto fail;
2927 		}
2928 		ata_link_warn(link,
2929 			      "link online but %d devices misclassified, "
2930 			      "device detection might fail\n", nr_unknown);
2931 	}
2932 
2933 	/* reset successful, schedule revalidation */
2934 	ata_eh_done(link, NULL, ATA_EH_RESET);
2935 	if (slave)
2936 		ata_eh_done(slave, NULL, ATA_EH_RESET);
2937 	ehc->last_reset = jiffies;		/* update to completion time */
2938 	ehc->i.action |= ATA_EH_REVALIDATE;
2939 	link->lpm_policy = ATA_LPM_UNKNOWN;	/* reset LPM state */
2940 
2941 	rc = 0;
2942  out:
2943 	/* clear hotplug flag */
2944 	ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2945 	if (slave)
2946 		sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2947 
2948 	spin_lock_irqsave(ap->lock, flags);
2949 	ap->pflags &= ~ATA_PFLAG_RESETTING;
2950 	spin_unlock_irqrestore(ap->lock, flags);
2951 
2952 	return rc;
2953 
2954  fail:
2955 	/* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2956 	if (!ata_is_host_link(link) &&
2957 	    sata_scr_read(link, SCR_STATUS, &sstatus))
2958 		rc = -ERESTART;
2959 
2960 	if (try >= max_tries) {
2961 		/*
2962 		 * Thaw host port even if reset failed, so that the port
2963 		 * can be retried on the next phy event.  This risks
2964 		 * repeated EH runs but seems to be a better tradeoff than
2965 		 * shutting down a port after a botched hotplug attempt.
2966 		 */
2967 		if (ata_is_host_link(link))
2968 			ata_eh_thaw_port(ap);
2969 		ata_link_warn(link, "%s failed\n",
2970 			      reset == hardreset ? "hardreset" : "softreset");
2971 		goto out;
2972 	}
2973 
2974 	now = jiffies;
2975 	if (time_before(now, deadline)) {
2976 		unsigned long delta = deadline - now;
2977 
2978 		ata_link_warn(failed_link,
2979 			"reset failed (errno=%d), retrying in %u secs\n",
2980 			rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2981 
2982 		ata_eh_release(ap);
2983 		while (delta)
2984 			delta = schedule_timeout_uninterruptible(delta);
2985 		ata_eh_acquire(ap);
2986 	}
2987 
2988 	/*
2989 	 * While disks spinup behind PMP, some controllers fail sending SRST.
2990 	 * They need to be reset - as well as the PMP - before retrying.
2991 	 */
2992 	if (rc == -ERESTART) {
2993 		if (ata_is_host_link(link))
2994 			ata_eh_thaw_port(ap);
2995 		goto out;
2996 	}
2997 
2998 	if (try == max_tries - 1) {
2999 		sata_down_spd_limit(link, 0);
3000 		if (slave)
3001 			sata_down_spd_limit(slave, 0);
3002 	} else if (rc == -EPIPE)
3003 		sata_down_spd_limit(failed_link, 0);
3004 
3005 	if (hardreset)
3006 		reset = hardreset;
3007 	goto retry;
3008 }
3009 
ata_eh_pull_park_action(struct ata_port * ap)3010 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3011 {
3012 	struct ata_link *link;
3013 	struct ata_device *dev;
3014 	unsigned long flags;
3015 
3016 	/*
3017 	 * This function can be thought of as an extended version of
3018 	 * ata_eh_about_to_do() specially crafted to accommodate the
3019 	 * requirements of ATA_EH_PARK handling. Since the EH thread
3020 	 * does not leave the do {} while () loop in ata_eh_recover as
3021 	 * long as the timeout for a park request to *one* device on
3022 	 * the port has not expired, and since we still want to pick
3023 	 * up park requests to other devices on the same port or
3024 	 * timeout updates for the same device, we have to pull
3025 	 * ATA_EH_PARK actions from eh_info into eh_context.i
3026 	 * ourselves at the beginning of each pass over the loop.
3027 	 *
3028 	 * Additionally, all write accesses to &ap->park_req_pending
3029 	 * through reinit_completion() (see below) or complete_all()
3030 	 * (see ata_scsi_park_store()) are protected by the host lock.
3031 	 * As a result we have that park_req_pending.done is zero on
3032 	 * exit from this function, i.e. when ATA_EH_PARK actions for
3033 	 * *all* devices on port ap have been pulled into the
3034 	 * respective eh_context structs. If, and only if,
3035 	 * park_req_pending.done is non-zero by the time we reach
3036 	 * wait_for_completion_timeout(), another ATA_EH_PARK action
3037 	 * has been scheduled for at least one of the devices on port
3038 	 * ap and we have to cycle over the do {} while () loop in
3039 	 * ata_eh_recover() again.
3040 	 */
3041 
3042 	spin_lock_irqsave(ap->lock, flags);
3043 	reinit_completion(&ap->park_req_pending);
3044 	ata_for_each_link(link, ap, EDGE) {
3045 		ata_for_each_dev(dev, link, ALL) {
3046 			struct ata_eh_info *ehi = &link->eh_info;
3047 
3048 			link->eh_context.i.dev_action[dev->devno] |=
3049 				ehi->dev_action[dev->devno] & ATA_EH_PARK;
3050 			ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3051 		}
3052 	}
3053 	spin_unlock_irqrestore(ap->lock, flags);
3054 }
3055 
ata_eh_park_issue_cmd(struct ata_device * dev,int park)3056 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3057 {
3058 	struct ata_eh_context *ehc = &dev->link->eh_context;
3059 	struct ata_taskfile tf;
3060 	unsigned int err_mask;
3061 
3062 	ata_tf_init(dev, &tf);
3063 	if (park) {
3064 		ehc->unloaded_mask |= 1 << dev->devno;
3065 		tf.command = ATA_CMD_IDLEIMMEDIATE;
3066 		tf.feature = 0x44;
3067 		tf.lbal = 0x4c;
3068 		tf.lbam = 0x4e;
3069 		tf.lbah = 0x55;
3070 	} else {
3071 		ehc->unloaded_mask &= ~(1 << dev->devno);
3072 		tf.command = ATA_CMD_CHK_POWER;
3073 	}
3074 
3075 	tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3076 	tf.protocol = ATA_PROT_NODATA;
3077 	err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3078 	if (park && (err_mask || tf.lbal != 0xc4)) {
3079 		ata_dev_err(dev, "head unload failed!\n");
3080 		ehc->unloaded_mask &= ~(1 << dev->devno);
3081 	}
3082 }
3083 
ata_eh_revalidate_and_attach(struct ata_link * link,struct ata_device ** r_failed_dev)3084 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3085 					struct ata_device **r_failed_dev)
3086 {
3087 	struct ata_port *ap = link->ap;
3088 	struct ata_eh_context *ehc = &link->eh_context;
3089 	struct ata_device *dev;
3090 	unsigned int new_mask = 0;
3091 	unsigned long flags;
3092 	int rc = 0;
3093 
3094 	/* For PATA drive side cable detection to work, IDENTIFY must
3095 	 * be done backwards such that PDIAG- is released by the slave
3096 	 * device before the master device is identified.
3097 	 */
3098 	ata_for_each_dev(dev, link, ALL_REVERSE) {
3099 		unsigned int action = ata_eh_dev_action(dev);
3100 		unsigned int readid_flags = 0;
3101 
3102 		if (ehc->i.flags & ATA_EHI_DID_RESET)
3103 			readid_flags |= ATA_READID_POSTRESET;
3104 
3105 		if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3106 			WARN_ON(dev->class == ATA_DEV_PMP);
3107 
3108 			/*
3109 			 * The link may be in a deep sleep, wake it up.
3110 			 *
3111 			 * If the link is in deep sleep, ata_phys_link_offline()
3112 			 * will return true, causing the revalidation to fail,
3113 			 * which leads to a (potentially) needless hard reset.
3114 			 *
3115 			 * ata_eh_recover() will later restore the link policy
3116 			 * to ap->target_lpm_policy after revalidation is done.
3117 			 */
3118 			if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3119 				rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER,
3120 						    r_failed_dev);
3121 				if (rc)
3122 					goto err;
3123 			}
3124 
3125 			if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3126 				rc = -EIO;
3127 				goto err;
3128 			}
3129 
3130 			ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3131 			rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3132 						readid_flags);
3133 			if (rc)
3134 				goto err;
3135 
3136 			ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3137 
3138 			/* Configuration may have changed, reconfigure
3139 			 * transfer mode.
3140 			 */
3141 			ehc->i.flags |= ATA_EHI_SETMODE;
3142 
3143 			/* schedule the scsi_rescan_device() here */
3144 			schedule_delayed_work(&ap->scsi_rescan_task, 0);
3145 		} else if (dev->class == ATA_DEV_UNKNOWN &&
3146 			   ehc->tries[dev->devno] &&
3147 			   ata_class_enabled(ehc->classes[dev->devno])) {
3148 			/* Temporarily set dev->class, it will be
3149 			 * permanently set once all configurations are
3150 			 * complete.  This is necessary because new
3151 			 * device configuration is done in two
3152 			 * separate loops.
3153 			 */
3154 			dev->class = ehc->classes[dev->devno];
3155 
3156 			if (dev->class == ATA_DEV_PMP)
3157 				rc = sata_pmp_attach(dev);
3158 			else
3159 				rc = ata_dev_read_id(dev, &dev->class,
3160 						     readid_flags, dev->id);
3161 
3162 			/* read_id might have changed class, store and reset */
3163 			ehc->classes[dev->devno] = dev->class;
3164 			dev->class = ATA_DEV_UNKNOWN;
3165 
3166 			switch (rc) {
3167 			case 0:
3168 				/* clear error info accumulated during probe */
3169 				ata_ering_clear(&dev->ering);
3170 				new_mask |= 1 << dev->devno;
3171 				break;
3172 			case -ENOENT:
3173 				/* IDENTIFY was issued to non-existent
3174 				 * device.  No need to reset.  Just
3175 				 * thaw and ignore the device.
3176 				 */
3177 				ata_eh_thaw_port(ap);
3178 				break;
3179 			default:
3180 				goto err;
3181 			}
3182 		}
3183 	}
3184 
3185 	/* PDIAG- should have been released, ask cable type if post-reset */
3186 	if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3187 		if (ap->ops->cable_detect)
3188 			ap->cbl = ap->ops->cable_detect(ap);
3189 		ata_force_cbl(ap);
3190 	}
3191 
3192 	/* Configure new devices forward such that user doesn't see
3193 	 * device detection messages backwards.
3194 	 */
3195 	ata_for_each_dev(dev, link, ALL) {
3196 		if (!(new_mask & (1 << dev->devno)))
3197 			continue;
3198 
3199 		dev->class = ehc->classes[dev->devno];
3200 
3201 		if (dev->class == ATA_DEV_PMP)
3202 			continue;
3203 
3204 		ehc->i.flags |= ATA_EHI_PRINTINFO;
3205 		rc = ata_dev_configure(dev);
3206 		ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3207 		if (rc) {
3208 			dev->class = ATA_DEV_UNKNOWN;
3209 			goto err;
3210 		}
3211 
3212 		spin_lock_irqsave(ap->lock, flags);
3213 		ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3214 		spin_unlock_irqrestore(ap->lock, flags);
3215 
3216 		/* new device discovered, configure xfermode */
3217 		ehc->i.flags |= ATA_EHI_SETMODE;
3218 	}
3219 
3220 	return 0;
3221 
3222  err:
3223 	dev->flags &= ~ATA_DFLAG_RESUMING;
3224 	*r_failed_dev = dev;
3225 	return rc;
3226 }
3227 
3228 /**
3229  *	ata_set_mode - Program timings and issue SET FEATURES - XFER
3230  *	@link: link on which timings will be programmed
3231  *	@r_failed_dev: out parameter for failed device
3232  *
3233  *	Set ATA device disk transfer mode (PIO3, UDMA6, etc.).  If
3234  *	ata_set_mode() fails, pointer to the failing device is
3235  *	returned in @r_failed_dev.
3236  *
3237  *	LOCKING:
3238  *	PCI/etc. bus probe sem.
3239  *
3240  *	RETURNS:
3241  *	0 on success, negative errno otherwise
3242  */
ata_set_mode(struct ata_link * link,struct ata_device ** r_failed_dev)3243 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3244 {
3245 	struct ata_port *ap = link->ap;
3246 	struct ata_device *dev;
3247 	int rc;
3248 
3249 	/* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3250 	ata_for_each_dev(dev, link, ENABLED) {
3251 		if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3252 			struct ata_ering_entry *ent;
3253 
3254 			ent = ata_ering_top(&dev->ering);
3255 			if (ent)
3256 				ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3257 		}
3258 	}
3259 
3260 	/* has private set_mode? */
3261 	if (ap->ops->set_mode)
3262 		rc = ap->ops->set_mode(link, r_failed_dev);
3263 	else
3264 		rc = ata_do_set_mode(link, r_failed_dev);
3265 
3266 	/* if transfer mode has changed, set DUBIOUS_XFER on device */
3267 	ata_for_each_dev(dev, link, ENABLED) {
3268 		struct ata_eh_context *ehc = &link->eh_context;
3269 		u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3270 		u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3271 
3272 		if (dev->xfer_mode != saved_xfer_mode ||
3273 		    ata_ncq_enabled(dev) != saved_ncq)
3274 			dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3275 	}
3276 
3277 	return rc;
3278 }
3279 
3280 /**
3281  *	atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3282  *	@dev: ATAPI device to clear UA for
3283  *
3284  *	Resets and other operations can make an ATAPI device raise
3285  *	UNIT ATTENTION which causes the next operation to fail.  This
3286  *	function clears UA.
3287  *
3288  *	LOCKING:
3289  *	EH context (may sleep).
3290  *
3291  *	RETURNS:
3292  *	0 on success, -errno on failure.
3293  */
atapi_eh_clear_ua(struct ata_device * dev)3294 static int atapi_eh_clear_ua(struct ata_device *dev)
3295 {
3296 	int i;
3297 
3298 	for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3299 		u8 *sense_buffer = dev->sector_buf;
3300 		u8 sense_key = 0;
3301 		unsigned int err_mask;
3302 
3303 		err_mask = atapi_eh_tur(dev, &sense_key);
3304 		if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3305 			ata_dev_warn(dev,
3306 				     "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3307 				     err_mask);
3308 			return -EIO;
3309 		}
3310 
3311 		if (!err_mask || sense_key != UNIT_ATTENTION)
3312 			return 0;
3313 
3314 		err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3315 		if (err_mask) {
3316 			ata_dev_warn(dev, "failed to clear "
3317 				"UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3318 			return -EIO;
3319 		}
3320 	}
3321 
3322 	ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3323 		     ATA_EH_UA_TRIES);
3324 
3325 	return 0;
3326 }
3327 
3328 /**
3329  *	ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3330  *	@dev: ATA device which may need FLUSH retry
3331  *
3332  *	If @dev failed FLUSH, it needs to be reported upper layer
3333  *	immediately as it means that @dev failed to remap and already
3334  *	lost at least a sector and further FLUSH retrials won't make
3335  *	any difference to the lost sector.  However, if FLUSH failed
3336  *	for other reasons, for example transmission error, FLUSH needs
3337  *	to be retried.
3338  *
3339  *	This function determines whether FLUSH failure retry is
3340  *	necessary and performs it if so.
3341  *
3342  *	RETURNS:
3343  *	0 if EH can continue, -errno if EH needs to be repeated.
3344  */
ata_eh_maybe_retry_flush(struct ata_device * dev)3345 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3346 {
3347 	struct ata_link *link = dev->link;
3348 	struct ata_port *ap = link->ap;
3349 	struct ata_queued_cmd *qc;
3350 	struct ata_taskfile tf;
3351 	unsigned int err_mask;
3352 	int rc = 0;
3353 
3354 	/* did flush fail for this device? */
3355 	if (!ata_tag_valid(link->active_tag))
3356 		return 0;
3357 
3358 	qc = __ata_qc_from_tag(ap, link->active_tag);
3359 	if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3360 			       qc->tf.command != ATA_CMD_FLUSH))
3361 		return 0;
3362 
3363 	/* if the device failed it, it should be reported to upper layers */
3364 	if (qc->err_mask & AC_ERR_DEV)
3365 		return 0;
3366 
3367 	/* flush failed for some other reason, give it another shot */
3368 	ata_tf_init(dev, &tf);
3369 
3370 	tf.command = qc->tf.command;
3371 	tf.flags |= ATA_TFLAG_DEVICE;
3372 	tf.protocol = ATA_PROT_NODATA;
3373 
3374 	ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3375 		       tf.command, qc->err_mask);
3376 
3377 	err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3378 	if (!err_mask) {
3379 		/*
3380 		 * FLUSH is complete but there's no way to
3381 		 * successfully complete a failed command from EH.
3382 		 * Making sure retry is allowed at least once and
3383 		 * retrying it should do the trick - whatever was in
3384 		 * the cache is already on the platter and this won't
3385 		 * cause infinite loop.
3386 		 */
3387 		qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3388 	} else {
3389 		ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3390 			       err_mask);
3391 		rc = -EIO;
3392 
3393 		/* if device failed it, report it to upper layers */
3394 		if (err_mask & AC_ERR_DEV) {
3395 			qc->err_mask |= AC_ERR_DEV;
3396 			qc->result_tf = tf;
3397 			if (!ata_port_is_frozen(ap))
3398 				rc = 0;
3399 		}
3400 	}
3401 	return rc;
3402 }
3403 
3404 /**
3405  *	ata_eh_set_lpm - configure SATA interface power management
3406  *	@link: link to configure power management
3407  *	@policy: the link power management policy
3408  *	@r_failed_dev: out parameter for failed device
3409  *
3410  *	Enable SATA Interface power management.  This will enable
3411  *	Device Interface Power Management (DIPM) for min_power and
3412  *	medium_power_with_dipm policies, and then call driver specific
3413  *	callbacks for enabling Host Initiated Power management.
3414  *
3415  *	LOCKING:
3416  *	EH context.
3417  *
3418  *	RETURNS:
3419  *	0 on success, -errno on failure.
3420  */
ata_eh_set_lpm(struct ata_link * link,enum ata_lpm_policy policy,struct ata_device ** r_failed_dev)3421 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3422 			  struct ata_device **r_failed_dev)
3423 {
3424 	struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3425 	struct ata_eh_context *ehc = &link->eh_context;
3426 	struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3427 	enum ata_lpm_policy old_policy = link->lpm_policy;
3428 	bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3429 	unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3430 	unsigned int err_mask;
3431 	int rc;
3432 
3433 	/* if the link or host doesn't do LPM, noop */
3434 	if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3435 	    (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3436 		return 0;
3437 
3438 	/*
3439 	 * DIPM is enabled only for MIN_POWER as some devices
3440 	 * misbehave when the host NACKs transition to SLUMBER.  Order
3441 	 * device and link configurations such that the host always
3442 	 * allows DIPM requests.
3443 	 */
3444 	ata_for_each_dev(dev, link, ENABLED) {
3445 		bool hipm = ata_id_has_hipm(dev->id);
3446 		bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3447 
3448 		/* find the first enabled and LPM enabled devices */
3449 		if (!link_dev)
3450 			link_dev = dev;
3451 
3452 		if (!lpm_dev && (hipm || dipm))
3453 			lpm_dev = dev;
3454 
3455 		hints &= ~ATA_LPM_EMPTY;
3456 		if (!hipm)
3457 			hints &= ~ATA_LPM_HIPM;
3458 
3459 		/* disable DIPM before changing link config */
3460 		if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3461 			err_mask = ata_dev_set_feature(dev,
3462 					SETFEATURES_SATA_DISABLE, SATA_DIPM);
3463 			if (err_mask && err_mask != AC_ERR_DEV) {
3464 				ata_dev_warn(dev,
3465 					     "failed to disable DIPM, Emask 0x%x\n",
3466 					     err_mask);
3467 				rc = -EIO;
3468 				goto fail;
3469 			}
3470 		}
3471 	}
3472 
3473 	if (ap) {
3474 		rc = ap->ops->set_lpm(link, policy, hints);
3475 		if (!rc && ap->slave_link)
3476 			rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3477 	} else
3478 		rc = sata_pmp_set_lpm(link, policy, hints);
3479 
3480 	/*
3481 	 * Attribute link config failure to the first (LPM) enabled
3482 	 * device on the link.
3483 	 */
3484 	if (rc) {
3485 		if (rc == -EOPNOTSUPP) {
3486 			link->flags |= ATA_LFLAG_NO_LPM;
3487 			return 0;
3488 		}
3489 		dev = lpm_dev ? lpm_dev : link_dev;
3490 		goto fail;
3491 	}
3492 
3493 	/*
3494 	 * Low level driver acked the transition.  Issue DIPM command
3495 	 * with the new policy set.
3496 	 */
3497 	link->lpm_policy = policy;
3498 	if (ap && ap->slave_link)
3499 		ap->slave_link->lpm_policy = policy;
3500 
3501 	/* host config updated, enable DIPM if transitioning to MIN_POWER */
3502 	ata_for_each_dev(dev, link, ENABLED) {
3503 		if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3504 		    ata_id_has_dipm(dev->id)) {
3505 			err_mask = ata_dev_set_feature(dev,
3506 					SETFEATURES_SATA_ENABLE, SATA_DIPM);
3507 			if (err_mask && err_mask != AC_ERR_DEV) {
3508 				ata_dev_warn(dev,
3509 					"failed to enable DIPM, Emask 0x%x\n",
3510 					err_mask);
3511 				rc = -EIO;
3512 				goto fail;
3513 			}
3514 		}
3515 	}
3516 
3517 	link->last_lpm_change = jiffies;
3518 	link->flags |= ATA_LFLAG_CHANGED;
3519 
3520 	return 0;
3521 
3522 fail:
3523 	/* restore the old policy */
3524 	link->lpm_policy = old_policy;
3525 	if (ap && ap->slave_link)
3526 		ap->slave_link->lpm_policy = old_policy;
3527 
3528 	/* if no device or only one more chance is left, disable LPM */
3529 	if (!dev || ehc->tries[dev->devno] <= 2) {
3530 		ata_link_warn(link, "disabling LPM on the link\n");
3531 		link->flags |= ATA_LFLAG_NO_LPM;
3532 	}
3533 	if (r_failed_dev)
3534 		*r_failed_dev = dev;
3535 	return rc;
3536 }
3537 
ata_link_nr_enabled(struct ata_link * link)3538 int ata_link_nr_enabled(struct ata_link *link)
3539 {
3540 	struct ata_device *dev;
3541 	int cnt = 0;
3542 
3543 	ata_for_each_dev(dev, link, ENABLED)
3544 		cnt++;
3545 	return cnt;
3546 }
3547 
ata_link_nr_vacant(struct ata_link * link)3548 static int ata_link_nr_vacant(struct ata_link *link)
3549 {
3550 	struct ata_device *dev;
3551 	int cnt = 0;
3552 
3553 	ata_for_each_dev(dev, link, ALL)
3554 		if (dev->class == ATA_DEV_UNKNOWN)
3555 			cnt++;
3556 	return cnt;
3557 }
3558 
ata_eh_skip_recovery(struct ata_link * link)3559 static int ata_eh_skip_recovery(struct ata_link *link)
3560 {
3561 	struct ata_port *ap = link->ap;
3562 	struct ata_eh_context *ehc = &link->eh_context;
3563 	struct ata_device *dev;
3564 
3565 	/* skip disabled links */
3566 	if (link->flags & ATA_LFLAG_DISABLED)
3567 		return 1;
3568 
3569 	/* skip if explicitly requested */
3570 	if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3571 		return 1;
3572 
3573 	/* thaw frozen port and recover failed devices */
3574 	if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3575 		return 0;
3576 
3577 	/* reset at least once if reset is requested */
3578 	if ((ehc->i.action & ATA_EH_RESET) &&
3579 	    !(ehc->i.flags & ATA_EHI_DID_RESET))
3580 		return 0;
3581 
3582 	/* skip if class codes for all vacant slots are ATA_DEV_NONE */
3583 	ata_for_each_dev(dev, link, ALL) {
3584 		if (dev->class == ATA_DEV_UNKNOWN &&
3585 		    ehc->classes[dev->devno] != ATA_DEV_NONE)
3586 			return 0;
3587 	}
3588 
3589 	return 1;
3590 }
3591 
ata_count_probe_trials_cb(struct ata_ering_entry * ent,void * void_arg)3592 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3593 {
3594 	u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3595 	u64 now = get_jiffies_64();
3596 	int *trials = void_arg;
3597 
3598 	if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3599 	    (ent->timestamp < now - min(now, interval)))
3600 		return -1;
3601 
3602 	(*trials)++;
3603 	return 0;
3604 }
3605 
ata_eh_schedule_probe(struct ata_device * dev)3606 static int ata_eh_schedule_probe(struct ata_device *dev)
3607 {
3608 	struct ata_eh_context *ehc = &dev->link->eh_context;
3609 	struct ata_link *link = ata_dev_phys_link(dev);
3610 	int trials = 0;
3611 
3612 	if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3613 	    (ehc->did_probe_mask & (1 << dev->devno)))
3614 		return 0;
3615 
3616 	ata_eh_detach_dev(dev);
3617 	ata_dev_init(dev);
3618 	ehc->did_probe_mask |= (1 << dev->devno);
3619 	ehc->i.action |= ATA_EH_RESET;
3620 	ehc->saved_xfer_mode[dev->devno] = 0;
3621 	ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3622 
3623 	/* the link maybe in a deep sleep, wake it up */
3624 	if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3625 		if (ata_is_host_link(link))
3626 			link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3627 					       ATA_LPM_EMPTY);
3628 		else
3629 			sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3630 					 ATA_LPM_EMPTY);
3631 	}
3632 
3633 	/* Record and count probe trials on the ering.  The specific
3634 	 * error mask used is irrelevant.  Because a successful device
3635 	 * detection clears the ering, this count accumulates only if
3636 	 * there are consecutive failed probes.
3637 	 *
3638 	 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3639 	 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3640 	 * forced to 1.5Gbps.
3641 	 *
3642 	 * This is to work around cases where failed link speed
3643 	 * negotiation results in device misdetection leading to
3644 	 * infinite DEVXCHG or PHRDY CHG events.
3645 	 */
3646 	ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3647 	ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3648 
3649 	if (trials > ATA_EH_PROBE_TRIALS)
3650 		sata_down_spd_limit(link, 1);
3651 
3652 	return 1;
3653 }
3654 
ata_eh_handle_dev_fail(struct ata_device * dev,int err)3655 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3656 {
3657 	struct ata_eh_context *ehc = &dev->link->eh_context;
3658 
3659 	/* -EAGAIN from EH routine indicates retry without prejudice.
3660 	 * The requester is responsible for ensuring forward progress.
3661 	 */
3662 	if (err != -EAGAIN)
3663 		ehc->tries[dev->devno]--;
3664 
3665 	switch (err) {
3666 	case -ENODEV:
3667 		/* device missing or wrong IDENTIFY data, schedule probing */
3668 		ehc->i.probe_mask |= (1 << dev->devno);
3669 		fallthrough;
3670 	case -EINVAL:
3671 		/* give it just one more chance */
3672 		ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3673 		fallthrough;
3674 	case -EIO:
3675 		if (ehc->tries[dev->devno] == 1) {
3676 			/* This is the last chance, better to slow
3677 			 * down than lose it.
3678 			 */
3679 			sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3680 			if (dev->pio_mode > XFER_PIO_0)
3681 				ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3682 		}
3683 	}
3684 
3685 	if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3686 		/* disable device if it has used up all its chances */
3687 		ata_dev_disable(dev);
3688 
3689 		/* detach if offline */
3690 		if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3691 			ata_eh_detach_dev(dev);
3692 
3693 		/* schedule probe if necessary */
3694 		if (ata_eh_schedule_probe(dev)) {
3695 			ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3696 			memset(ehc->cmd_timeout_idx[dev->devno], 0,
3697 			       sizeof(ehc->cmd_timeout_idx[dev->devno]));
3698 		}
3699 
3700 		return 1;
3701 	} else {
3702 		ehc->i.action |= ATA_EH_RESET;
3703 		return 0;
3704 	}
3705 }
3706 
3707 /**
3708  *	ata_eh_recover - recover host port after error
3709  *	@ap: host port to recover
3710  *	@prereset: prereset method (can be NULL)
3711  *	@softreset: softreset method (can be NULL)
3712  *	@hardreset: hardreset method (can be NULL)
3713  *	@postreset: postreset method (can be NULL)
3714  *	@r_failed_link: out parameter for failed link
3715  *
3716  *	This is the alpha and omega, eum and yang, heart and soul of
3717  *	libata exception handling.  On entry, actions required to
3718  *	recover each link and hotplug requests are recorded in the
3719  *	link's eh_context.  This function executes all the operations
3720  *	with appropriate retrials and fallbacks to resurrect failed
3721  *	devices, detach goners and greet newcomers.
3722  *
3723  *	LOCKING:
3724  *	Kernel thread context (may sleep).
3725  *
3726  *	RETURNS:
3727  *	0 on success, -errno on failure.
3728  */
ata_eh_recover(struct ata_port * ap,ata_prereset_fn_t prereset,ata_reset_fn_t softreset,ata_reset_fn_t hardreset,ata_postreset_fn_t postreset,struct ata_link ** r_failed_link)3729 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3730 		   ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3731 		   ata_postreset_fn_t postreset,
3732 		   struct ata_link **r_failed_link)
3733 {
3734 	struct ata_link *link;
3735 	struct ata_device *dev;
3736 	int rc, nr_fails;
3737 	unsigned long flags, deadline;
3738 
3739 	/* prep for recovery */
3740 	ata_for_each_link(link, ap, EDGE) {
3741 		struct ata_eh_context *ehc = &link->eh_context;
3742 
3743 		/* re-enable link? */
3744 		if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3745 			ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3746 			spin_lock_irqsave(ap->lock, flags);
3747 			link->flags &= ~ATA_LFLAG_DISABLED;
3748 			spin_unlock_irqrestore(ap->lock, flags);
3749 			ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3750 		}
3751 
3752 		ata_for_each_dev(dev, link, ALL) {
3753 			if (link->flags & ATA_LFLAG_NO_RETRY)
3754 				ehc->tries[dev->devno] = 1;
3755 			else
3756 				ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3757 
3758 			/* collect port action mask recorded in dev actions */
3759 			ehc->i.action |= ehc->i.dev_action[dev->devno] &
3760 					 ~ATA_EH_PERDEV_MASK;
3761 			ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3762 
3763 			/* process hotplug request */
3764 			if (dev->flags & ATA_DFLAG_DETACH)
3765 				ata_eh_detach_dev(dev);
3766 
3767 			/* schedule probe if necessary */
3768 			if (!ata_dev_enabled(dev))
3769 				ata_eh_schedule_probe(dev);
3770 		}
3771 	}
3772 
3773  retry:
3774 	rc = 0;
3775 
3776 	/* if UNLOADING, finish immediately */
3777 	if (ap->pflags & ATA_PFLAG_UNLOADING)
3778 		goto out;
3779 
3780 	/* prep for EH */
3781 	ata_for_each_link(link, ap, EDGE) {
3782 		struct ata_eh_context *ehc = &link->eh_context;
3783 
3784 		/* skip EH if possible. */
3785 		if (ata_eh_skip_recovery(link))
3786 			ehc->i.action = 0;
3787 
3788 		ata_for_each_dev(dev, link, ALL)
3789 			ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3790 	}
3791 
3792 	/* reset */
3793 	ata_for_each_link(link, ap, EDGE) {
3794 		struct ata_eh_context *ehc = &link->eh_context;
3795 
3796 		if (!(ehc->i.action & ATA_EH_RESET))
3797 			continue;
3798 
3799 		rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3800 				  prereset, softreset, hardreset, postreset);
3801 		if (rc) {
3802 			ata_link_err(link, "reset failed, giving up\n");
3803 			goto out;
3804 		}
3805 	}
3806 
3807 	do {
3808 		unsigned long now;
3809 
3810 		/*
3811 		 * clears ATA_EH_PARK in eh_info and resets
3812 		 * ap->park_req_pending
3813 		 */
3814 		ata_eh_pull_park_action(ap);
3815 
3816 		deadline = jiffies;
3817 		ata_for_each_link(link, ap, EDGE) {
3818 			ata_for_each_dev(dev, link, ALL) {
3819 				struct ata_eh_context *ehc = &link->eh_context;
3820 				unsigned long tmp;
3821 
3822 				if (dev->class != ATA_DEV_ATA &&
3823 				    dev->class != ATA_DEV_ZAC)
3824 					continue;
3825 				if (!(ehc->i.dev_action[dev->devno] &
3826 				      ATA_EH_PARK))
3827 					continue;
3828 				tmp = dev->unpark_deadline;
3829 				if (time_before(deadline, tmp))
3830 					deadline = tmp;
3831 				else if (time_before_eq(tmp, jiffies))
3832 					continue;
3833 				if (ehc->unloaded_mask & (1 << dev->devno))
3834 					continue;
3835 
3836 				ata_eh_park_issue_cmd(dev, 1);
3837 			}
3838 		}
3839 
3840 		now = jiffies;
3841 		if (time_before_eq(deadline, now))
3842 			break;
3843 
3844 		ata_eh_release(ap);
3845 		deadline = wait_for_completion_timeout(&ap->park_req_pending,
3846 						       deadline - now);
3847 		ata_eh_acquire(ap);
3848 	} while (deadline);
3849 	ata_for_each_link(link, ap, EDGE) {
3850 		ata_for_each_dev(dev, link, ALL) {
3851 			if (!(link->eh_context.unloaded_mask &
3852 			      (1 << dev->devno)))
3853 				continue;
3854 
3855 			ata_eh_park_issue_cmd(dev, 0);
3856 			ata_eh_done(link, dev, ATA_EH_PARK);
3857 		}
3858 	}
3859 
3860 	/* the rest */
3861 	nr_fails = 0;
3862 	ata_for_each_link(link, ap, PMP_FIRST) {
3863 		struct ata_eh_context *ehc = &link->eh_context;
3864 
3865 		if (sata_pmp_attached(ap) && ata_is_host_link(link))
3866 			goto config_lpm;
3867 
3868 		/* revalidate existing devices and attach new ones */
3869 		rc = ata_eh_revalidate_and_attach(link, &dev);
3870 		if (rc)
3871 			goto rest_fail;
3872 
3873 		/* if PMP got attached, return, pmp EH will take care of it */
3874 		if (link->device->class == ATA_DEV_PMP) {
3875 			ehc->i.action = 0;
3876 			return 0;
3877 		}
3878 
3879 		/* configure transfer mode if necessary */
3880 		if (ehc->i.flags & ATA_EHI_SETMODE) {
3881 			rc = ata_set_mode(link, &dev);
3882 			if (rc)
3883 				goto rest_fail;
3884 			ehc->i.flags &= ~ATA_EHI_SETMODE;
3885 		}
3886 
3887 		/* If reset has been issued, clear UA to avoid
3888 		 * disrupting the current users of the device.
3889 		 */
3890 		if (ehc->i.flags & ATA_EHI_DID_RESET) {
3891 			ata_for_each_dev(dev, link, ALL) {
3892 				if (dev->class != ATA_DEV_ATAPI)
3893 					continue;
3894 				rc = atapi_eh_clear_ua(dev);
3895 				if (rc)
3896 					goto rest_fail;
3897 				if (zpodd_dev_enabled(dev))
3898 					zpodd_post_poweron(dev);
3899 			}
3900 		}
3901 
3902 		/*
3903 		 * Make sure to transition devices to the active power mode
3904 		 * if needed (e.g. if we were scheduled on system resume).
3905 		 */
3906 		ata_for_each_dev(dev, link, ENABLED) {
3907 			if (ehc->i.dev_action[dev->devno] & ATA_EH_SET_ACTIVE) {
3908 				ata_dev_power_set_active(dev);
3909 				ata_eh_done(link, dev, ATA_EH_SET_ACTIVE);
3910 			}
3911 		}
3912 
3913 		/* retry flush if necessary */
3914 		ata_for_each_dev(dev, link, ALL) {
3915 			if (dev->class != ATA_DEV_ATA &&
3916 			    dev->class != ATA_DEV_ZAC)
3917 				continue;
3918 			rc = ata_eh_maybe_retry_flush(dev);
3919 			if (rc)
3920 				goto rest_fail;
3921 		}
3922 
3923 	config_lpm:
3924 		/* configure link power saving */
3925 		if (link->lpm_policy != ap->target_lpm_policy) {
3926 			rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3927 			if (rc)
3928 				goto rest_fail;
3929 		}
3930 
3931 		/* this link is okay now */
3932 		ehc->i.flags = 0;
3933 		continue;
3934 
3935 	rest_fail:
3936 		nr_fails++;
3937 		if (dev)
3938 			ata_eh_handle_dev_fail(dev, rc);
3939 
3940 		if (ata_port_is_frozen(ap)) {
3941 			/* PMP reset requires working host port.
3942 			 * Can't retry if it's frozen.
3943 			 */
3944 			if (sata_pmp_attached(ap))
3945 				goto out;
3946 			break;
3947 		}
3948 	}
3949 
3950 	if (nr_fails)
3951 		goto retry;
3952 
3953  out:
3954 	if (rc && r_failed_link)
3955 		*r_failed_link = link;
3956 
3957 	return rc;
3958 }
3959 
3960 /**
3961  *	ata_eh_finish - finish up EH
3962  *	@ap: host port to finish EH for
3963  *
3964  *	Recovery is complete.  Clean up EH states and retry or finish
3965  *	failed qcs.
3966  *
3967  *	LOCKING:
3968  *	None.
3969  */
ata_eh_finish(struct ata_port * ap)3970 void ata_eh_finish(struct ata_port *ap)
3971 {
3972 	struct ata_queued_cmd *qc;
3973 	int tag;
3974 
3975 	/* retry or finish qcs */
3976 	ata_qc_for_each_raw(ap, qc, tag) {
3977 		if (!(qc->flags & ATA_QCFLAG_EH))
3978 			continue;
3979 
3980 		if (qc->err_mask) {
3981 			/* FIXME: Once EH migration is complete,
3982 			 * generate sense data in this function,
3983 			 * considering both err_mask and tf.
3984 			 */
3985 			if (qc->flags & ATA_QCFLAG_RETRY) {
3986 				/*
3987 				 * Since qc->err_mask is set, ata_eh_qc_retry()
3988 				 * will not increment scmd->allowed, so upper
3989 				 * layer will only retry the command if it has
3990 				 * not already been retried too many times.
3991 				 */
3992 				ata_eh_qc_retry(qc);
3993 			} else {
3994 				ata_eh_qc_complete(qc);
3995 			}
3996 		} else {
3997 			if (qc->flags & ATA_QCFLAG_SENSE_VALID ||
3998 			    qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) {
3999 				ata_eh_qc_complete(qc);
4000 			} else {
4001 				/* feed zero TF to sense generation */
4002 				memset(&qc->result_tf, 0, sizeof(qc->result_tf));
4003 				/*
4004 				 * Since qc->err_mask is not set,
4005 				 * ata_eh_qc_retry() will increment
4006 				 * scmd->allowed, so upper layer is guaranteed
4007 				 * to retry the command.
4008 				 */
4009 				ata_eh_qc_retry(qc);
4010 			}
4011 		}
4012 	}
4013 
4014 	/* make sure nr_active_links is zero after EH */
4015 	WARN_ON(ap->nr_active_links);
4016 	ap->nr_active_links = 0;
4017 }
4018 
4019 /**
4020  *	ata_do_eh - do standard error handling
4021  *	@ap: host port to handle error for
4022  *
4023  *	@prereset: prereset method (can be NULL)
4024  *	@softreset: softreset method (can be NULL)
4025  *	@hardreset: hardreset method (can be NULL)
4026  *	@postreset: postreset method (can be NULL)
4027  *
4028  *	Perform standard error handling sequence.
4029  *
4030  *	LOCKING:
4031  *	Kernel thread context (may sleep).
4032  */
ata_do_eh(struct ata_port * ap,ata_prereset_fn_t prereset,ata_reset_fn_t softreset,ata_reset_fn_t hardreset,ata_postreset_fn_t postreset)4033 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4034 	       ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4035 	       ata_postreset_fn_t postreset)
4036 {
4037 	struct ata_device *dev;
4038 	int rc;
4039 
4040 	ata_eh_autopsy(ap);
4041 	ata_eh_report(ap);
4042 
4043 	rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4044 			    NULL);
4045 	if (rc) {
4046 		ata_for_each_dev(dev, &ap->link, ALL)
4047 			ata_dev_disable(dev);
4048 	}
4049 
4050 	ata_eh_finish(ap);
4051 }
4052 
4053 /**
4054  *	ata_std_error_handler - standard error handler
4055  *	@ap: host port to handle error for
4056  *
4057  *	Standard error handler
4058  *
4059  *	LOCKING:
4060  *	Kernel thread context (may sleep).
4061  */
ata_std_error_handler(struct ata_port * ap)4062 void ata_std_error_handler(struct ata_port *ap)
4063 {
4064 	struct ata_port_operations *ops = ap->ops;
4065 	ata_reset_fn_t hardreset = ops->hardreset;
4066 
4067 	/* ignore built-in hardreset if SCR access is not available */
4068 	if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4069 		hardreset = NULL;
4070 
4071 	ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4072 }
4073 EXPORT_SYMBOL_GPL(ata_std_error_handler);
4074 
4075 #ifdef CONFIG_PM
4076 /**
4077  *	ata_eh_handle_port_suspend - perform port suspend operation
4078  *	@ap: port to suspend
4079  *
4080  *	Suspend @ap.
4081  *
4082  *	LOCKING:
4083  *	Kernel thread context (may sleep).
4084  */
ata_eh_handle_port_suspend(struct ata_port * ap)4085 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4086 {
4087 	unsigned long flags;
4088 	int rc = 0;
4089 	struct ata_device *dev;
4090 	struct ata_link *link;
4091 
4092 	/* are we suspending? */
4093 	spin_lock_irqsave(ap->lock, flags);
4094 	if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4095 	    ap->pm_mesg.event & PM_EVENT_RESUME) {
4096 		spin_unlock_irqrestore(ap->lock, flags);
4097 		return;
4098 	}
4099 	spin_unlock_irqrestore(ap->lock, flags);
4100 
4101 	WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4102 
4103 	/*
4104 	 * We will reach this point for all of the PM events:
4105 	 * PM_EVENT_SUSPEND (if runtime pm, PM_EVENT_AUTO will also be set)
4106 	 * PM_EVENT_FREEZE, and PM_EVENT_HIBERNATE.
4107 	 *
4108 	 * We do not want to perform disk spin down for PM_EVENT_FREEZE.
4109 	 * (Spin down will be performed by the subsequent PM_EVENT_HIBERNATE.)
4110 	 */
4111 	if (!(ap->pm_mesg.event & PM_EVENT_FREEZE)) {
4112 		/* Set all devices attached to the port in standby mode */
4113 		ata_for_each_link(link, ap, HOST_FIRST) {
4114 			ata_for_each_dev(dev, link, ENABLED)
4115 				ata_dev_power_set_standby(dev);
4116 		}
4117 	}
4118 
4119 	/*
4120 	 * If we have a ZPODD attached, check its zero
4121 	 * power ready status before the port is frozen.
4122 	 * Only needed for runtime suspend.
4123 	 */
4124 	if (PMSG_IS_AUTO(ap->pm_mesg)) {
4125 		ata_for_each_dev(dev, &ap->link, ENABLED) {
4126 			if (zpodd_dev_enabled(dev))
4127 				zpodd_on_suspend(dev);
4128 		}
4129 	}
4130 
4131 	/* suspend */
4132 	ata_eh_freeze_port(ap);
4133 
4134 	if (ap->ops->port_suspend)
4135 		rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4136 
4137 	ata_acpi_set_state(ap, ap->pm_mesg);
4138 
4139 	/* update the flags */
4140 	spin_lock_irqsave(ap->lock, flags);
4141 
4142 	ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4143 	if (rc == 0)
4144 		ap->pflags |= ATA_PFLAG_SUSPENDED;
4145 	else if (ata_port_is_frozen(ap))
4146 		ata_port_schedule_eh(ap);
4147 
4148 	spin_unlock_irqrestore(ap->lock, flags);
4149 
4150 	return;
4151 }
4152 
4153 /**
4154  *	ata_eh_handle_port_resume - perform port resume operation
4155  *	@ap: port to resume
4156  *
4157  *	Resume @ap.
4158  *
4159  *	LOCKING:
4160  *	Kernel thread context (may sleep).
4161  */
ata_eh_handle_port_resume(struct ata_port * ap)4162 static void ata_eh_handle_port_resume(struct ata_port *ap)
4163 {
4164 	struct ata_link *link;
4165 	struct ata_device *dev;
4166 	unsigned long flags;
4167 
4168 	/* are we resuming? */
4169 	spin_lock_irqsave(ap->lock, flags);
4170 	if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4171 	    !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4172 		spin_unlock_irqrestore(ap->lock, flags);
4173 		return;
4174 	}
4175 	spin_unlock_irqrestore(ap->lock, flags);
4176 
4177 	WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4178 
4179 	/*
4180 	 * Error timestamps are in jiffies which doesn't run while
4181 	 * suspended and PHY events during resume isn't too uncommon.
4182 	 * When the two are combined, it can lead to unnecessary speed
4183 	 * downs if the machine is suspended and resumed repeatedly.
4184 	 * Clear error history.
4185 	 */
4186 	ata_for_each_link(link, ap, HOST_FIRST)
4187 		ata_for_each_dev(dev, link, ALL)
4188 			ata_ering_clear(&dev->ering);
4189 
4190 	ata_acpi_set_state(ap, ap->pm_mesg);
4191 
4192 	if (ap->ops->port_resume)
4193 		ap->ops->port_resume(ap);
4194 
4195 	/* tell ACPI that we're resuming */
4196 	ata_acpi_on_resume(ap);
4197 
4198 	/* update the flags */
4199 	spin_lock_irqsave(ap->lock, flags);
4200 	ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4201 	ap->pflags |= ATA_PFLAG_RESUMING;
4202 	spin_unlock_irqrestore(ap->lock, flags);
4203 }
4204 #endif /* CONFIG_PM */
4205