2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_eh.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_cmnd.h>
41 #include "../scsi/scsi_transport_api.h"
43 #include <linux/libata.h>
47 static void __ata_port_freeze(struct ata_port *ap);
48 static void ata_eh_finish(struct ata_port *ap);
49 static void ata_eh_handle_port_suspend(struct ata_port *ap);
50 static void ata_eh_handle_port_resume(struct ata_port *ap);
52 static void ata_ering_record(struct ata_ering *ering, int is_io,
53 unsigned int err_mask)
55 struct ata_ering_entry *ent;
60 ering->cursor %= ATA_ERING_SIZE;
62 ent = &ering->ring[ering->cursor];
64 ent->err_mask = err_mask;
65 ent->timestamp = get_jiffies_64();
68 static struct ata_ering_entry * ata_ering_top(struct ata_ering *ering)
70 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
76 static int ata_ering_map(struct ata_ering *ering,
77 int (*map_fn)(struct ata_ering_entry *, void *),
81 struct ata_ering_entry *ent;
85 ent = &ering->ring[idx];
88 rc = map_fn(ent, arg);
91 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
92 } while (idx != ering->cursor);
97 static unsigned int ata_eh_dev_action(struct ata_device *dev)
99 struct ata_eh_context *ehc = &dev->ap->eh_context;
101 return ehc->i.action | ehc->i.dev_action[dev->devno];
104 static void ata_eh_clear_action(struct ata_device *dev,
105 struct ata_eh_info *ehi, unsigned int action)
110 ehi->action &= ~action;
111 for (i = 0; i < ATA_MAX_DEVICES; i++)
112 ehi->dev_action[i] &= ~action;
114 /* doesn't make sense for port-wide EH actions */
115 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
117 /* break ehi->action into ehi->dev_action */
118 if (ehi->action & action) {
119 for (i = 0; i < ATA_MAX_DEVICES; i++)
120 ehi->dev_action[i] |= ehi->action & action;
121 ehi->action &= ~action;
124 /* turn off the specified per-dev action */
125 ehi->dev_action[dev->devno] &= ~action;
130 * ata_scsi_timed_out - SCSI layer time out callback
131 * @cmd: timed out SCSI command
133 * Handles SCSI layer timeout. We race with normal completion of
134 * the qc for @cmd. If the qc is already gone, we lose and let
135 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
136 * timed out and EH should be invoked. Prevent ata_qc_complete()
137 * from finishing it by setting EH_SCHEDULED and return
140 * TODO: kill this function once old EH is gone.
143 * Called from timer context
146 * EH_HANDLED or EH_NOT_HANDLED
148 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
150 struct Scsi_Host *host = cmd->device->host;
151 struct ata_port *ap = ata_shost_to_port(host);
153 struct ata_queued_cmd *qc;
154 enum scsi_eh_timer_return ret;
158 if (ap->ops->error_handler) {
159 ret = EH_NOT_HANDLED;
164 spin_lock_irqsave(ap->lock, flags);
165 qc = ata_qc_from_tag(ap, ap->active_tag);
167 WARN_ON(qc->scsicmd != cmd);
168 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
169 qc->err_mask |= AC_ERR_TIMEOUT;
170 ret = EH_NOT_HANDLED;
172 spin_unlock_irqrestore(ap->lock, flags);
175 DPRINTK("EXIT, ret=%d\n", ret);
180 * ata_scsi_error - SCSI layer error handler callback
181 * @host: SCSI host on which error occurred
183 * Handles SCSI-layer-thrown error events.
186 * Inherited from SCSI layer (none, can sleep)
191 void ata_scsi_error(struct Scsi_Host *host)
193 struct ata_port *ap = ata_shost_to_port(host);
194 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
199 /* synchronize with port task */
200 ata_port_flush_task(ap);
202 /* synchronize with host lock and sort out timeouts */
204 /* For new EH, all qcs are finished in one of three ways -
205 * normal completion, error completion, and SCSI timeout.
206 * Both cmpletions can race against SCSI timeout. When normal
207 * completion wins, the qc never reaches EH. When error
208 * completion wins, the qc has ATA_QCFLAG_FAILED set.
210 * When SCSI timeout wins, things are a bit more complex.
211 * Normal or error completion can occur after the timeout but
212 * before this point. In such cases, both types of
213 * completions are honored. A scmd is determined to have
214 * timed out iff its associated qc is active and not failed.
216 if (ap->ops->error_handler) {
217 struct scsi_cmnd *scmd, *tmp;
220 spin_lock_irqsave(ap->lock, flags);
222 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
223 struct ata_queued_cmd *qc;
225 for (i = 0; i < ATA_MAX_QUEUE; i++) {
226 qc = __ata_qc_from_tag(ap, i);
227 if (qc->flags & ATA_QCFLAG_ACTIVE &&
232 if (i < ATA_MAX_QUEUE) {
233 /* the scmd has an associated qc */
234 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
235 /* which hasn't failed yet, timeout */
236 qc->err_mask |= AC_ERR_TIMEOUT;
237 qc->flags |= ATA_QCFLAG_FAILED;
241 /* Normal completion occurred after
242 * SCSI timeout but before this point.
243 * Successfully complete it.
245 scmd->retries = scmd->allowed;
246 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
250 /* If we have timed out qcs. They belong to EH from
251 * this point but the state of the controller is
252 * unknown. Freeze the port to make sure the IRQ
253 * handler doesn't diddle with those qcs. This must
254 * be done atomically w.r.t. setting QCFLAG_FAILED.
257 __ata_port_freeze(ap);
259 spin_unlock_irqrestore(ap->lock, flags);
261 spin_unlock_wait(ap->lock);
264 /* invoke error handler */
265 if (ap->ops->error_handler) {
266 /* process port resume request */
267 ata_eh_handle_port_resume(ap);
269 /* fetch & clear EH info */
270 spin_lock_irqsave(ap->lock, flags);
272 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
273 ap->eh_context.i = ap->eh_info;
274 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
276 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
277 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
279 spin_unlock_irqrestore(ap->lock, flags);
281 /* invoke EH, skip if unloading or suspended */
282 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
283 ap->ops->error_handler(ap);
287 /* process port suspend request */
288 ata_eh_handle_port_suspend(ap);
290 /* Exception might have happend after ->error_handler
291 * recovered the port but before this point. Repeat
294 spin_lock_irqsave(ap->lock, flags);
296 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
298 ata_port_printk(ap, KERN_INFO,
299 "EH pending after completion, "
300 "repeating EH (cnt=%d)\n", repeat_cnt);
301 spin_unlock_irqrestore(ap->lock, flags);
304 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
305 "tries, giving up\n", ATA_EH_MAX_REPEAT);
308 /* this run is complete, make sure EH info is clear */
309 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
311 /* Clear host_eh_scheduled while holding ap->lock such
312 * that if exception occurs after this point but
313 * before EH completion, SCSI midlayer will
316 host->host_eh_scheduled = 0;
318 spin_unlock_irqrestore(ap->lock, flags);
320 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
321 ap->ops->eng_timeout(ap);
324 /* finish or retry handled scmd's and clean up */
325 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
327 scsi_eh_flush_done_q(&ap->eh_done_q);
330 spin_lock_irqsave(ap->lock, flags);
332 if (ap->pflags & ATA_PFLAG_LOADING)
333 ap->pflags &= ~ATA_PFLAG_LOADING;
334 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
335 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
337 if (ap->pflags & ATA_PFLAG_RECOVERED)
338 ata_port_printk(ap, KERN_INFO, "EH complete\n");
340 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
342 /* tell wait_eh that we're done */
343 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
344 wake_up_all(&ap->eh_wait_q);
346 spin_unlock_irqrestore(ap->lock, flags);
352 * ata_port_wait_eh - Wait for the currently pending EH to complete
353 * @ap: Port to wait EH for
355 * Wait until the currently pending EH is complete.
358 * Kernel thread context (may sleep).
360 void ata_port_wait_eh(struct ata_port *ap)
366 spin_lock_irqsave(ap->lock, flags);
368 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
369 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
370 spin_unlock_irqrestore(ap->lock, flags);
372 spin_lock_irqsave(ap->lock, flags);
374 finish_wait(&ap->eh_wait_q, &wait);
376 spin_unlock_irqrestore(ap->lock, flags);
378 /* make sure SCSI EH is complete */
379 if (scsi_host_in_recovery(ap->scsi_host)) {
386 * ata_qc_timeout - Handle timeout of queued command
387 * @qc: Command that timed out
389 * Some part of the kernel (currently, only the SCSI layer)
390 * has noticed that the active command on port @ap has not
391 * completed after a specified length of time. Handle this
392 * condition by disabling DMA (if necessary) and completing
393 * transactions, with error if necessary.
395 * This also handles the case of the "lost interrupt", where
396 * for some reason (possibly hardware bug, possibly driver bug)
397 * an interrupt was not delivered to the driver, even though the
398 * transaction completed successfully.
400 * TODO: kill this function once old EH is gone.
403 * Inherited from SCSI layer (none, can sleep)
405 static void ata_qc_timeout(struct ata_queued_cmd *qc)
407 struct ata_port *ap = qc->ap;
408 u8 host_stat = 0, drv_stat;
413 ap->hsm_task_state = HSM_ST_IDLE;
415 spin_lock_irqsave(ap->lock, flags);
417 switch (qc->tf.protocol) {
420 case ATA_PROT_ATAPI_DMA:
421 host_stat = ap->ops->bmdma_status(ap);
423 /* before we do anything else, clear DMA-Start bit */
424 ap->ops->bmdma_stop(qc);
430 drv_stat = ata_chk_status(ap);
432 /* ack bmdma irq events */
433 ap->ops->irq_clear(ap);
435 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
436 "stat 0x%x host_stat 0x%x\n",
437 qc->tf.command, drv_stat, host_stat);
439 /* complete taskfile transaction */
440 qc->err_mask |= AC_ERR_TIMEOUT;
444 spin_unlock_irqrestore(ap->lock, flags);
446 ata_eh_qc_complete(qc);
452 * ata_eng_timeout - Handle timeout of queued command
453 * @ap: Port on which timed-out command is active
455 * Some part of the kernel (currently, only the SCSI layer)
456 * has noticed that the active command on port @ap has not
457 * completed after a specified length of time. Handle this
458 * condition by disabling DMA (if necessary) and completing
459 * transactions, with error if necessary.
461 * This also handles the case of the "lost interrupt", where
462 * for some reason (possibly hardware bug, possibly driver bug)
463 * an interrupt was not delivered to the driver, even though the
464 * transaction completed successfully.
466 * TODO: kill this function once old EH is gone.
469 * Inherited from SCSI layer (none, can sleep)
471 void ata_eng_timeout(struct ata_port *ap)
475 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
481 * ata_qc_schedule_eh - schedule qc for error handling
482 * @qc: command to schedule error handling for
484 * Schedule error handling for @qc. EH will kick in as soon as
485 * other commands are drained.
488 * spin_lock_irqsave(host lock)
490 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
492 struct ata_port *ap = qc->ap;
494 WARN_ON(!ap->ops->error_handler);
496 qc->flags |= ATA_QCFLAG_FAILED;
497 qc->ap->pflags |= ATA_PFLAG_EH_PENDING;
499 /* The following will fail if timeout has already expired.
500 * ata_scsi_error() takes care of such scmds on EH entry.
501 * Note that ATA_QCFLAG_FAILED is unconditionally set after
502 * this function completes.
504 scsi_req_abort_cmd(qc->scsicmd);
508 * ata_port_schedule_eh - schedule error handling without a qc
509 * @ap: ATA port to schedule EH for
511 * Schedule error handling for @ap. EH will kick in as soon as
512 * all commands are drained.
515 * spin_lock_irqsave(host lock)
517 void ata_port_schedule_eh(struct ata_port *ap)
519 WARN_ON(!ap->ops->error_handler);
521 ap->pflags |= ATA_PFLAG_EH_PENDING;
522 scsi_schedule_eh(ap->scsi_host);
524 DPRINTK("port EH scheduled\n");
528 * ata_port_abort - abort all qc's on the port
529 * @ap: ATA port to abort qc's for
531 * Abort all active qc's of @ap and schedule EH.
534 * spin_lock_irqsave(host lock)
537 * Number of aborted qc's.
539 int ata_port_abort(struct ata_port *ap)
541 int tag, nr_aborted = 0;
543 WARN_ON(!ap->ops->error_handler);
545 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
546 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
549 qc->flags |= ATA_QCFLAG_FAILED;
556 ata_port_schedule_eh(ap);
562 * __ata_port_freeze - freeze port
563 * @ap: ATA port to freeze
565 * This function is called when HSM violation or some other
566 * condition disrupts normal operation of the port. Frozen port
567 * is not allowed to perform any operation until the port is
568 * thawed, which usually follows a successful reset.
570 * ap->ops->freeze() callback can be used for freezing the port
571 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
572 * port cannot be frozen hardware-wise, the interrupt handler
573 * must ack and clear interrupts unconditionally while the port
577 * spin_lock_irqsave(host lock)
579 static void __ata_port_freeze(struct ata_port *ap)
581 WARN_ON(!ap->ops->error_handler);
586 ap->pflags |= ATA_PFLAG_FROZEN;
588 DPRINTK("ata%u port frozen\n", ap->id);
592 * ata_port_freeze - abort & freeze port
593 * @ap: ATA port to freeze
595 * Abort and freeze @ap.
598 * spin_lock_irqsave(host lock)
601 * Number of aborted commands.
603 int ata_port_freeze(struct ata_port *ap)
607 WARN_ON(!ap->ops->error_handler);
609 nr_aborted = ata_port_abort(ap);
610 __ata_port_freeze(ap);
616 * ata_eh_freeze_port - EH helper to freeze port
617 * @ap: ATA port to freeze
624 void ata_eh_freeze_port(struct ata_port *ap)
628 if (!ap->ops->error_handler)
631 spin_lock_irqsave(ap->lock, flags);
632 __ata_port_freeze(ap);
633 spin_unlock_irqrestore(ap->lock, flags);
637 * ata_port_thaw_port - EH helper to thaw port
638 * @ap: ATA port to thaw
640 * Thaw frozen port @ap.
645 void ata_eh_thaw_port(struct ata_port *ap)
649 if (!ap->ops->error_handler)
652 spin_lock_irqsave(ap->lock, flags);
654 ap->pflags &= ~ATA_PFLAG_FROZEN;
659 spin_unlock_irqrestore(ap->lock, flags);
661 DPRINTK("ata%u port thawed\n", ap->id);
664 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
669 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
671 struct ata_port *ap = qc->ap;
672 struct scsi_cmnd *scmd = qc->scsicmd;
675 spin_lock_irqsave(ap->lock, flags);
676 qc->scsidone = ata_eh_scsidone;
677 __ata_qc_complete(qc);
678 WARN_ON(ata_tag_valid(qc->tag));
679 spin_unlock_irqrestore(ap->lock, flags);
681 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
685 * ata_eh_qc_complete - Complete an active ATA command from EH
686 * @qc: Command to complete
688 * Indicate to the mid and upper layers that an ATA command has
689 * completed. To be used from EH.
691 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
693 struct scsi_cmnd *scmd = qc->scsicmd;
694 scmd->retries = scmd->allowed;
695 __ata_eh_qc_complete(qc);
699 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
700 * @qc: Command to retry
702 * Indicate to the mid and upper layers that an ATA command
703 * should be retried. To be used from EH.
705 * SCSI midlayer limits the number of retries to scmd->allowed.
706 * scmd->retries is decremented for commands which get retried
707 * due to unrelated failures (qc->err_mask is zero).
709 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
711 struct scsi_cmnd *scmd = qc->scsicmd;
712 if (!qc->err_mask && scmd->retries)
714 __ata_eh_qc_complete(qc);
718 * ata_eh_detach_dev - detach ATA device
719 * @dev: ATA device to detach
726 static void ata_eh_detach_dev(struct ata_device *dev)
728 struct ata_port *ap = dev->ap;
731 ata_dev_disable(dev);
733 spin_lock_irqsave(ap->lock, flags);
735 dev->flags &= ~ATA_DFLAG_DETACH;
737 if (ata_scsi_offline_dev(dev)) {
738 dev->flags |= ATA_DFLAG_DETACHED;
739 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
742 /* clear per-dev EH actions */
743 ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK);
744 ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK);
746 spin_unlock_irqrestore(ap->lock, flags);
750 * ata_eh_about_to_do - about to perform eh_action
751 * @ap: target ATA port
752 * @dev: target ATA dev for per-dev action (can be NULL)
753 * @action: action about to be performed
755 * Called just before performing EH actions to clear related bits
756 * in @ap->eh_info such that eh actions are not unnecessarily
762 static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
766 struct ata_eh_info *ehi = &ap->eh_info;
767 struct ata_eh_context *ehc = &ap->eh_context;
769 spin_lock_irqsave(ap->lock, flags);
771 /* Reset is represented by combination of actions and EHI
772 * flags. Suck in all related bits before clearing eh_info to
773 * avoid losing requested action.
775 if (action & ATA_EH_RESET_MASK) {
776 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
777 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
779 /* make sure all reset actions are cleared & clear EHI flags */
780 action |= ATA_EH_RESET_MASK;
781 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
784 ata_eh_clear_action(dev, ehi, action);
786 if (!(ehc->i.flags & ATA_EHI_QUIET))
787 ap->pflags |= ATA_PFLAG_RECOVERED;
789 spin_unlock_irqrestore(ap->lock, flags);
793 * ata_eh_done - EH action complete
794 * @ap: target ATA port
795 * @dev: target ATA dev for per-dev action (can be NULL)
796 * @action: action just completed
798 * Called right after performing EH actions to clear related bits
799 * in @ap->eh_context.
804 static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
807 /* if reset is complete, clear all reset actions & reset modifier */
808 if (action & ATA_EH_RESET_MASK) {
809 action |= ATA_EH_RESET_MASK;
810 ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
813 ata_eh_clear_action(dev, &ap->eh_context.i, action);
817 * ata_err_string - convert err_mask to descriptive string
818 * @err_mask: error mask to convert to string
820 * Convert @err_mask to descriptive string. Errors are
821 * prioritized according to severity and only the most severe
828 * Descriptive string for @err_mask
830 static const char * ata_err_string(unsigned int err_mask)
832 if (err_mask & AC_ERR_HOST_BUS)
833 return "host bus error";
834 if (err_mask & AC_ERR_ATA_BUS)
835 return "ATA bus error";
836 if (err_mask & AC_ERR_TIMEOUT)
838 if (err_mask & AC_ERR_HSM)
839 return "HSM violation";
840 if (err_mask & AC_ERR_SYSTEM)
841 return "internal error";
842 if (err_mask & AC_ERR_MEDIA)
843 return "media error";
844 if (err_mask & AC_ERR_INVALID)
845 return "invalid argument";
846 if (err_mask & AC_ERR_DEV)
847 return "device error";
848 return "unknown error";
852 * ata_read_log_page - read a specific log page
853 * @dev: target device
854 * @page: page to read
855 * @buf: buffer to store read page
856 * @sectors: number of sectors to read
858 * Read log page using READ_LOG_EXT command.
861 * Kernel thread context (may sleep).
864 * 0 on success, AC_ERR_* mask otherwise.
866 static unsigned int ata_read_log_page(struct ata_device *dev,
867 u8 page, void *buf, unsigned int sectors)
869 struct ata_taskfile tf;
870 unsigned int err_mask;
872 DPRINTK("read log page - page %d\n", page);
874 ata_tf_init(dev, &tf);
875 tf.command = ATA_CMD_READ_LOG_EXT;
878 tf.hob_nsect = sectors >> 8;
879 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
880 tf.protocol = ATA_PROT_PIO;
882 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
883 buf, sectors * ATA_SECT_SIZE);
885 DPRINTK("EXIT, err_mask=%x\n", err_mask);
890 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
891 * @dev: Device to read log page 10h from
892 * @tag: Resulting tag of the failed command
893 * @tf: Resulting taskfile registers of the failed command
895 * Read log page 10h to obtain NCQ error details and clear error
899 * Kernel thread context (may sleep).
902 * 0 on success, -errno otherwise.
904 static int ata_eh_read_log_10h(struct ata_device *dev,
905 int *tag, struct ata_taskfile *tf)
907 u8 *buf = dev->ap->sector_buf;
908 unsigned int err_mask;
912 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
917 for (i = 0; i < ATA_SECT_SIZE; i++)
920 ata_dev_printk(dev, KERN_WARNING,
921 "invalid checksum 0x%x on log page 10h\n", csum);
926 *tag = buf[0] & 0x1f;
928 tf->command = buf[2];
929 tf->feature = buf[3];
934 tf->hob_lbal = buf[8];
935 tf->hob_lbam = buf[9];
936 tf->hob_lbah = buf[10];
938 tf->hob_nsect = buf[13];
944 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
945 * @dev: device to perform REQUEST_SENSE to
946 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
948 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
949 * SENSE. This function is EH helper.
952 * Kernel thread context (may sleep).
955 * 0 on success, AC_ERR_* mask on failure
957 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
958 unsigned char *sense_buf)
960 struct ata_port *ap = dev->ap;
961 struct ata_taskfile tf;
962 u8 cdb[ATAPI_CDB_LEN];
964 DPRINTK("ATAPI request sense\n");
966 ata_tf_init(dev, &tf);
968 /* FIXME: is this needed? */
969 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
971 /* XXX: why tf_read here? */
972 ap->ops->tf_read(ap, &tf);
974 /* fill these in, for the case where they are -not- overwritten */
976 sense_buf[2] = tf.feature >> 4;
978 memset(cdb, 0, ATAPI_CDB_LEN);
979 cdb[0] = REQUEST_SENSE;
980 cdb[4] = SCSI_SENSE_BUFFERSIZE;
982 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
983 tf.command = ATA_CMD_PACKET;
985 /* is it pointless to prefer PIO for "safety reasons"? */
986 if (ap->flags & ATA_FLAG_PIO_DMA) {
987 tf.protocol = ATA_PROT_ATAPI_DMA;
988 tf.feature |= ATAPI_PKT_DMA;
990 tf.protocol = ATA_PROT_ATAPI;
991 tf.lbam = (8 * 1024) & 0xff;
992 tf.lbah = (8 * 1024) >> 8;
995 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
996 sense_buf, SCSI_SENSE_BUFFERSIZE);
1000 * ata_eh_analyze_serror - analyze SError for a failed port
1001 * @ap: ATA port to analyze SError for
1003 * Analyze SError if available and further determine cause of
1009 static void ata_eh_analyze_serror(struct ata_port *ap)
1011 struct ata_eh_context *ehc = &ap->eh_context;
1012 u32 serror = ehc->i.serror;
1013 unsigned int err_mask = 0, action = 0;
1015 if (serror & SERR_PERSISTENT) {
1016 err_mask |= AC_ERR_ATA_BUS;
1017 action |= ATA_EH_HARDRESET;
1020 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1021 err_mask |= AC_ERR_ATA_BUS;
1022 action |= ATA_EH_SOFTRESET;
1024 if (serror & SERR_PROTOCOL) {
1025 err_mask |= AC_ERR_HSM;
1026 action |= ATA_EH_SOFTRESET;
1028 if (serror & SERR_INTERNAL) {
1029 err_mask |= AC_ERR_SYSTEM;
1030 action |= ATA_EH_SOFTRESET;
1032 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1033 ata_ehi_hotplugged(&ehc->i);
1035 ehc->i.err_mask |= err_mask;
1036 ehc->i.action |= action;
1040 * ata_eh_analyze_ncq_error - analyze NCQ error
1041 * @ap: ATA port to analyze NCQ error for
1043 * Read log page 10h, determine the offending qc and acquire
1044 * error status TF. For NCQ device errors, all LLDDs have to do
1045 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1049 * Kernel thread context (may sleep).
1051 static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1053 struct ata_eh_context *ehc = &ap->eh_context;
1054 struct ata_device *dev = ap->device;
1055 struct ata_queued_cmd *qc;
1056 struct ata_taskfile tf;
1059 /* if frozen, we can't do much */
1060 if (ap->pflags & ATA_PFLAG_FROZEN)
1063 /* is it NCQ device error? */
1064 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1067 /* has LLDD analyzed already? */
1068 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1069 qc = __ata_qc_from_tag(ap, tag);
1071 if (!(qc->flags & ATA_QCFLAG_FAILED))
1078 /* okay, this error is ours */
1079 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1081 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1082 "(errno=%d)\n", rc);
1086 if (!(ap->sactive & (1 << tag))) {
1087 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1088 "inactive tag %d\n", tag);
1092 /* we've got the perpetrator, condemn it */
1093 qc = __ata_qc_from_tag(ap, tag);
1094 memcpy(&qc->result_tf, &tf, sizeof(tf));
1095 qc->err_mask |= AC_ERR_DEV;
1096 ehc->i.err_mask &= ~AC_ERR_DEV;
1100 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1101 * @qc: qc to analyze
1102 * @tf: Taskfile registers to analyze
1104 * Analyze taskfile of @qc and further determine cause of
1105 * failure. This function also requests ATAPI sense data if
1109 * Kernel thread context (may sleep).
1112 * Determined recovery action
1114 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1115 const struct ata_taskfile *tf)
1117 unsigned int tmp, action = 0;
1118 u8 stat = tf->command, err = tf->feature;
1120 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1121 qc->err_mask |= AC_ERR_HSM;
1122 return ATA_EH_SOFTRESET;
1125 if (!(qc->err_mask & AC_ERR_DEV))
1128 switch (qc->dev->class) {
1131 qc->err_mask |= AC_ERR_ATA_BUS;
1133 qc->err_mask |= AC_ERR_MEDIA;
1135 qc->err_mask |= AC_ERR_INVALID;
1139 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1140 tmp = atapi_eh_request_sense(qc->dev,
1141 qc->scsicmd->sense_buffer);
1143 /* ATA_QCFLAG_SENSE_VALID is used to
1144 * tell atapi_qc_complete() that sense
1145 * data is already valid.
1147 * TODO: interpret sense data and set
1148 * appropriate err_mask.
1150 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1152 qc->err_mask |= tmp;
1156 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1157 action |= ATA_EH_SOFTRESET;
1162 static int ata_eh_categorize_ering_entry(struct ata_ering_entry *ent)
1164 if (ent->err_mask & (AC_ERR_ATA_BUS | AC_ERR_TIMEOUT))
1168 if (ent->err_mask & AC_ERR_HSM)
1170 if ((ent->err_mask &
1171 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1178 struct speed_down_needed_arg {
1183 static int speed_down_needed_cb(struct ata_ering_entry *ent, void *void_arg)
1185 struct speed_down_needed_arg *arg = void_arg;
1187 if (ent->timestamp < arg->since)
1190 arg->nr_errors[ata_eh_categorize_ering_entry(ent)]++;
1195 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
1196 * @dev: Device of interest
1198 * This function examines error ring of @dev and determines
1199 * whether speed down is necessary. Speed down is necessary if
1200 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
1201 * errors during last 15 minutes.
1203 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
1204 * violation for known supported commands.
1206 * Cat-2 errors are unclassified DEV error for known supported
1210 * Inherited from caller.
1213 * 1 if speed down is necessary, 0 otherwise
1215 static int ata_eh_speed_down_needed(struct ata_device *dev)
1217 const u64 interval = 15LLU * 60 * HZ;
1218 static const int err_limits[3] = { -1, 3, 10 };
1219 struct speed_down_needed_arg arg;
1220 struct ata_ering_entry *ent;
1224 ent = ata_ering_top(&dev->ering);
1228 err_cat = ata_eh_categorize_ering_entry(ent);
1232 memset(&arg, 0, sizeof(arg));
1234 j64 = get_jiffies_64();
1235 if (j64 >= interval)
1236 arg.since = j64 - interval;
1240 ata_ering_map(&dev->ering, speed_down_needed_cb, &arg);
1242 return arg.nr_errors[err_cat] > err_limits[err_cat];
1246 * ata_eh_speed_down - record error and speed down if necessary
1247 * @dev: Failed device
1248 * @is_io: Did the device fail during normal IO?
1249 * @err_mask: err_mask of the error
1251 * Record error and examine error history to determine whether
1252 * adjusting transmission speed is necessary. It also sets
1253 * transmission limits appropriately if such adjustment is
1257 * Kernel thread context (may sleep).
1260 * 0 on success, -errno otherwise
1262 static int ata_eh_speed_down(struct ata_device *dev, int is_io,
1263 unsigned int err_mask)
1268 /* record error and determine whether speed down is necessary */
1269 ata_ering_record(&dev->ering, is_io, err_mask);
1271 if (!ata_eh_speed_down_needed(dev))
1274 /* speed down SATA link speed if possible */
1275 if (sata_down_spd_limit(dev->ap) == 0)
1276 return ATA_EH_HARDRESET;
1278 /* lower transfer mode */
1279 if (ata_down_xfermask_limit(dev, 0) == 0)
1280 return ATA_EH_SOFTRESET;
1282 ata_dev_printk(dev, KERN_ERR,
1283 "speed down requested but no transfer mode left\n");
1288 * ata_eh_autopsy - analyze error and determine recovery action
1289 * @ap: ATA port to perform autopsy on
1291 * Analyze why @ap failed and determine which recovery action is
1292 * needed. This function also sets more detailed AC_ERR_* values
1293 * and fills sense data for ATAPI CHECK SENSE.
1296 * Kernel thread context (may sleep).
1298 static void ata_eh_autopsy(struct ata_port *ap)
1300 struct ata_eh_context *ehc = &ap->eh_context;
1301 unsigned int all_err_mask = 0;
1308 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1311 /* obtain and analyze SError */
1312 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1314 ehc->i.serror |= serror;
1315 ata_eh_analyze_serror(ap);
1316 } else if (rc != -EOPNOTSUPP)
1317 ehc->i.action |= ATA_EH_HARDRESET;
1319 /* analyze NCQ failure */
1320 ata_eh_analyze_ncq_error(ap);
1322 /* any real error trumps AC_ERR_OTHER */
1323 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1324 ehc->i.err_mask &= ~AC_ERR_OTHER;
1326 all_err_mask |= ehc->i.err_mask;
1328 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1329 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1331 if (!(qc->flags & ATA_QCFLAG_FAILED))
1334 /* inherit upper level err_mask */
1335 qc->err_mask |= ehc->i.err_mask;
1338 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1340 /* DEV errors are probably spurious in case of ATA_BUS error */
1341 if (qc->err_mask & AC_ERR_ATA_BUS)
1342 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1345 /* any real error trumps unknown error */
1346 if (qc->err_mask & ~AC_ERR_OTHER)
1347 qc->err_mask &= ~AC_ERR_OTHER;
1349 /* SENSE_VALID trumps dev/unknown error and revalidation */
1350 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1351 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1352 ehc->i.action &= ~ATA_EH_REVALIDATE;
1355 /* accumulate error info */
1356 ehc->i.dev = qc->dev;
1357 all_err_mask |= qc->err_mask;
1358 if (qc->flags & ATA_QCFLAG_IO)
1362 /* enforce default EH actions */
1363 if (ap->pflags & ATA_PFLAG_FROZEN ||
1364 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1365 ehc->i.action |= ATA_EH_SOFTRESET;
1366 else if (all_err_mask)
1367 ehc->i.action |= ATA_EH_REVALIDATE;
1369 /* if we have offending qcs and the associated failed device */
1372 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1375 /* perform per-dev EH action only on the offending device */
1376 ehc->i.dev_action[ehc->i.dev->devno] |=
1377 ehc->i.action & ATA_EH_PERDEV_MASK;
1378 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1385 * ata_eh_report - report error handling to user
1386 * @ap: ATA port EH is going on
1388 * Report EH to user.
1393 static void ata_eh_report(struct ata_port *ap)
1395 struct ata_eh_context *ehc = &ap->eh_context;
1396 const char *frozen, *desc;
1397 int tag, nr_failed = 0;
1400 if (ehc->i.desc[0] != '\0')
1403 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1404 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1406 if (!(qc->flags & ATA_QCFLAG_FAILED))
1408 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1414 if (!nr_failed && !ehc->i.err_mask)
1418 if (ap->pflags & ATA_PFLAG_FROZEN)
1422 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1423 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1424 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1425 ehc->i.action, frozen);
1427 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1429 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1430 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1431 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1432 ehc->i.action, frozen);
1434 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1437 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1438 static const char *dma_str[] = {
1439 [DMA_BIDIRECTIONAL] = "bidi",
1440 [DMA_TO_DEVICE] = "out",
1441 [DMA_FROM_DEVICE] = "in",
1444 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1445 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
1446 unsigned int nbytes;
1448 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1451 nbytes = qc->nbytes;
1453 nbytes = qc->nsect << 9;
1455 ata_dev_printk(qc->dev, KERN_ERR,
1456 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1457 "tag %d cdb 0x%x data %u %s\n "
1458 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1459 "Emask 0x%x (%s)\n",
1460 cmd->command, cmd->feature, cmd->nsect,
1461 cmd->lbal, cmd->lbam, cmd->lbah,
1462 cmd->hob_feature, cmd->hob_nsect,
1463 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
1464 cmd->device, qc->tag, qc->cdb[0], nbytes,
1465 dma_str[qc->dma_dir],
1466 res->command, res->feature, res->nsect,
1467 res->lbal, res->lbam, res->lbah,
1468 res->hob_feature, res->hob_nsect,
1469 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1470 res->device, qc->err_mask, ata_err_string(qc->err_mask));
1474 static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1475 unsigned int *classes)
1479 for (i = 0; i < ATA_MAX_DEVICES; i++)
1480 classes[i] = ATA_DEV_UNKNOWN;
1482 rc = reset(ap, classes);
1486 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1487 * is complete and convert all ATA_DEV_UNKNOWN to
1490 for (i = 0; i < ATA_MAX_DEVICES; i++)
1491 if (classes[i] != ATA_DEV_UNKNOWN)
1494 if (i < ATA_MAX_DEVICES)
1495 for (i = 0; i < ATA_MAX_DEVICES; i++)
1496 if (classes[i] == ATA_DEV_UNKNOWN)
1497 classes[i] = ATA_DEV_NONE;
1502 static int ata_eh_followup_srst_needed(int rc, int classify,
1503 const unsigned int *classes)
1509 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1514 static int ata_eh_reset(struct ata_port *ap, int classify,
1515 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
1516 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1518 struct ata_eh_context *ehc = &ap->eh_context;
1519 unsigned int *classes = ehc->classes;
1520 int tries = ATA_EH_RESET_TRIES;
1521 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
1522 unsigned int action;
1523 ata_reset_fn_t reset;
1524 int i, did_followup_srst, rc;
1526 /* about to reset */
1527 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1529 /* Determine which reset to use and record in ehc->i.action.
1530 * prereset() may examine and modify it.
1532 action = ehc->i.action;
1533 ehc->i.action &= ~ATA_EH_RESET_MASK;
1534 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
1535 !(action & ATA_EH_HARDRESET))))
1536 ehc->i.action |= ATA_EH_SOFTRESET;
1538 ehc->i.action |= ATA_EH_HARDRESET;
1543 if (rc == -ENOENT) {
1544 ata_port_printk(ap, KERN_DEBUG, "port disabled. ignoring.\n");
1545 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
1547 ata_port_printk(ap, KERN_ERR,
1548 "prereset failed (errno=%d)\n", rc);
1553 /* prereset() might have modified ehc->i.action */
1554 if (ehc->i.action & ATA_EH_HARDRESET)
1556 else if (ehc->i.action & ATA_EH_SOFTRESET)
1559 /* prereset told us not to reset, bang classes and return */
1560 for (i = 0; i < ATA_MAX_DEVICES; i++)
1561 classes[i] = ATA_DEV_NONE;
1565 /* did prereset() screw up? if so, fix up to avoid oopsing */
1567 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1568 "invalid reset type\n");
1576 /* shut up during boot probing */
1578 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1579 reset == softreset ? "soft" : "hard");
1581 /* mark that this EH session started with reset */
1582 ehc->i.flags |= ATA_EHI_DID_RESET;
1584 rc = ata_do_reset(ap, reset, classes);
1586 did_followup_srst = 0;
1587 if (reset == hardreset &&
1588 ata_eh_followup_srst_needed(rc, classify, classes)) {
1589 /* okay, let's do follow-up softreset */
1590 did_followup_srst = 1;
1594 ata_port_printk(ap, KERN_ERR,
1595 "follow-up softreset required "
1596 "but no softreset avaliable\n");
1600 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
1601 rc = ata_do_reset(ap, reset, classes);
1603 if (rc == 0 && classify &&
1604 classes[0] == ATA_DEV_UNKNOWN) {
1605 ata_port_printk(ap, KERN_ERR,
1606 "classification failed\n");
1611 if (rc && --tries) {
1614 if (reset == softreset) {
1615 if (did_followup_srst)
1616 type = "follow-up soft";
1622 ata_port_printk(ap, KERN_WARNING,
1623 "%sreset failed, retrying in 5 secs\n", type);
1626 if (reset == hardreset)
1627 sata_down_spd_limit(ap);
1634 /* After the reset, the device state is PIO 0 and the
1635 * controller state is undefined. Record the mode.
1637 for (i = 0; i < ATA_MAX_DEVICES; i++)
1638 ap->device[i].pio_mode = XFER_PIO_0;
1641 postreset(ap, classes);
1643 /* reset successful, schedule revalidation */
1644 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1645 ehc->i.action |= ATA_EH_REVALIDATE;
1651 static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1652 struct ata_device **r_failed_dev)
1654 struct ata_eh_context *ehc = &ap->eh_context;
1655 struct ata_device *dev;
1656 unsigned long flags;
1661 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1662 unsigned int action, readid_flags = 0;
1664 dev = &ap->device[i];
1665 action = ata_eh_dev_action(dev);
1667 if (ehc->i.flags & ATA_EHI_DID_RESET)
1668 readid_flags |= ATA_READID_POSTRESET;
1670 if (action & ATA_EH_REVALIDATE && ata_dev_ready(dev)) {
1671 if (ata_port_offline(ap)) {
1676 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
1677 rc = ata_dev_revalidate(dev, readid_flags);
1681 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1683 /* Configuration may have changed, reconfigure
1686 ehc->i.flags |= ATA_EHI_SETMODE;
1688 /* schedule the scsi_rescan_device() here */
1689 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
1690 } else if (dev->class == ATA_DEV_UNKNOWN &&
1691 ehc->tries[dev->devno] &&
1692 ata_class_enabled(ehc->classes[dev->devno])) {
1693 dev->class = ehc->classes[dev->devno];
1695 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1698 ehc->i.flags |= ATA_EHI_PRINTINFO;
1699 rc = ata_dev_configure(dev);
1700 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
1701 } else if (rc == -ENOENT) {
1702 /* IDENTIFY was issued to non-existent
1703 * device. No need to reset. Just
1704 * thaw and kill the device.
1706 ata_eh_thaw_port(ap);
1707 dev->class = ATA_DEV_UNKNOWN;
1712 dev->class = ATA_DEV_UNKNOWN;
1716 if (ata_dev_enabled(dev)) {
1717 spin_lock_irqsave(ap->lock, flags);
1718 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1719 spin_unlock_irqrestore(ap->lock, flags);
1721 /* new device discovered, configure xfermode */
1722 ehc->i.flags |= ATA_EHI_SETMODE;
1728 *r_failed_dev = dev;
1735 * ata_eh_suspend - handle suspend EH action
1736 * @ap: target host port
1737 * @r_failed_dev: result parameter to indicate failing device
1739 * Handle suspend EH action. Disk devices are spinned down and
1740 * other types of devices are just marked suspended. Once
1741 * suspended, no EH action to the device is allowed until it is
1745 * Kernel thread context (may sleep).
1748 * 0 on success, -errno otherwise
1750 static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev)
1752 struct ata_device *dev;
1757 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1758 unsigned long flags;
1759 unsigned int action, err_mask;
1761 dev = &ap->device[i];
1762 action = ata_eh_dev_action(dev);
1764 if (!ata_dev_enabled(dev) || !(action & ATA_EH_SUSPEND))
1767 WARN_ON(dev->flags & ATA_DFLAG_SUSPENDED);
1769 ata_eh_about_to_do(ap, dev, ATA_EH_SUSPEND);
1771 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1773 rc = ata_flush_cache(dev);
1778 err_mask = ata_do_simple_cmd(dev, ATA_CMD_STANDBYNOW1);
1780 ata_dev_printk(dev, KERN_ERR, "failed to "
1781 "spin down (err_mask=0x%x)\n",
1788 spin_lock_irqsave(ap->lock, flags);
1789 dev->flags |= ATA_DFLAG_SUSPENDED;
1790 spin_unlock_irqrestore(ap->lock, flags);
1792 ata_eh_done(ap, dev, ATA_EH_SUSPEND);
1796 *r_failed_dev = dev;
1803 * ata_eh_prep_resume - prep for resume EH action
1804 * @ap: target host port
1806 * Clear SUSPENDED in preparation for scheduled resume actions.
1807 * This allows other parts of EH to access the devices being
1811 * Kernel thread context (may sleep).
1813 static void ata_eh_prep_resume(struct ata_port *ap)
1815 struct ata_device *dev;
1816 unsigned long flags;
1821 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1822 unsigned int action;
1824 dev = &ap->device[i];
1825 action = ata_eh_dev_action(dev);
1827 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1830 spin_lock_irqsave(ap->lock, flags);
1831 dev->flags &= ~ATA_DFLAG_SUSPENDED;
1832 spin_unlock_irqrestore(ap->lock, flags);
1839 * ata_eh_resume - handle resume EH action
1840 * @ap: target host port
1841 * @r_failed_dev: result parameter to indicate failing device
1843 * Handle resume EH action. Target devices are already reset and
1844 * revalidated. Spinning up is the only operation left.
1847 * Kernel thread context (may sleep).
1850 * 0 on success, -errno otherwise
1852 static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev)
1854 struct ata_device *dev;
1859 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1860 unsigned int action, err_mask;
1862 dev = &ap->device[i];
1863 action = ata_eh_dev_action(dev);
1865 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1868 ata_eh_about_to_do(ap, dev, ATA_EH_RESUME);
1870 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1871 err_mask = ata_do_simple_cmd(dev,
1872 ATA_CMD_IDLEIMMEDIATE);
1874 ata_dev_printk(dev, KERN_ERR, "failed to "
1875 "spin up (err_mask=0x%x)\n",
1882 ata_eh_done(ap, dev, ATA_EH_RESUME);
1886 *r_failed_dev = dev;
1892 static int ata_port_nr_enabled(struct ata_port *ap)
1896 for (i = 0; i < ATA_MAX_DEVICES; i++)
1897 if (ata_dev_enabled(&ap->device[i]))
1902 static int ata_port_nr_vacant(struct ata_port *ap)
1906 for (i = 0; i < ATA_MAX_DEVICES; i++)
1907 if (ap->device[i].class == ATA_DEV_UNKNOWN)
1912 static int ata_eh_skip_recovery(struct ata_port *ap)
1914 struct ata_eh_context *ehc = &ap->eh_context;
1917 /* skip if all possible devices are suspended */
1918 for (i = 0; i < ata_port_max_devices(ap); i++) {
1919 struct ata_device *dev = &ap->device[i];
1921 if (!(dev->flags & ATA_DFLAG_SUSPENDED))
1925 if (i == ata_port_max_devices(ap))
1928 /* thaw frozen port, resume link and recover failed devices */
1929 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
1930 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
1933 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1934 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1935 struct ata_device *dev = &ap->device[i];
1937 if (dev->class == ATA_DEV_UNKNOWN &&
1938 ehc->classes[dev->devno] != ATA_DEV_NONE)
1946 * ata_eh_recover - recover host port after error
1947 * @ap: host port to recover
1948 * @prereset: prereset method (can be NULL)
1949 * @softreset: softreset method (can be NULL)
1950 * @hardreset: hardreset method (can be NULL)
1951 * @postreset: postreset method (can be NULL)
1953 * This is the alpha and omega, eum and yang, heart and soul of
1954 * libata exception handling. On entry, actions required to
1955 * recover the port and hotplug requests are recorded in
1956 * eh_context. This function executes all the operations with
1957 * appropriate retrials and fallbacks to resurrect failed
1958 * devices, detach goners and greet newcomers.
1961 * Kernel thread context (may sleep).
1964 * 0 on success, -errno on failure.
1966 static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
1967 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
1968 ata_postreset_fn_t postreset)
1970 struct ata_eh_context *ehc = &ap->eh_context;
1971 struct ata_device *dev;
1972 int down_xfermask, i, rc;
1976 /* prep for recovery */
1977 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1978 dev = &ap->device[i];
1980 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
1982 /* collect port action mask recorded in dev actions */
1983 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
1984 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
1986 /* process hotplug request */
1987 if (dev->flags & ATA_DFLAG_DETACH)
1988 ata_eh_detach_dev(dev);
1990 if (!ata_dev_enabled(dev) &&
1991 ((ehc->i.probe_mask & (1 << dev->devno)) &&
1992 !(ehc->did_probe_mask & (1 << dev->devno)))) {
1993 ata_eh_detach_dev(dev);
1995 ehc->did_probe_mask |= (1 << dev->devno);
1996 ehc->i.action |= ATA_EH_SOFTRESET;
2004 /* if UNLOADING, finish immediately */
2005 if (ap->pflags & ATA_PFLAG_UNLOADING)
2008 /* prep for resume */
2009 ata_eh_prep_resume(ap);
2011 /* skip EH if possible. */
2012 if (ata_eh_skip_recovery(ap))
2015 for (i = 0; i < ATA_MAX_DEVICES; i++)
2016 ehc->classes[i] = ATA_DEV_UNKNOWN;
2019 if (ehc->i.action & ATA_EH_RESET_MASK) {
2020 ata_eh_freeze_port(ap);
2022 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
2023 softreset, hardreset, postreset);
2025 ata_port_printk(ap, KERN_ERR,
2026 "reset failed, giving up\n");
2030 ata_eh_thaw_port(ap);
2033 /* revalidate existing devices and attach new ones */
2034 rc = ata_eh_revalidate_and_attach(ap, &dev);
2038 /* resume devices */
2039 rc = ata_eh_resume(ap, &dev);
2043 /* configure transfer mode if necessary */
2044 if (ehc->i.flags & ATA_EHI_SETMODE) {
2045 rc = ata_set_mode(ap, &dev);
2050 ehc->i.flags &= ~ATA_EHI_SETMODE;
2053 /* suspend devices */
2054 rc = ata_eh_suspend(ap, &dev);
2063 /* device missing, schedule probing */
2064 ehc->i.probe_mask |= (1 << dev->devno);
2066 ehc->tries[dev->devno] = 0;
2069 sata_down_spd_limit(ap);
2071 ehc->tries[dev->devno]--;
2072 if (down_xfermask &&
2073 ata_down_xfermask_limit(dev, ehc->tries[dev->devno] == 1))
2074 ehc->tries[dev->devno] = 0;
2077 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2078 /* disable device if it has used up all its chances */
2079 ata_dev_disable(dev);
2081 /* detach if offline */
2082 if (ata_port_offline(ap))
2083 ata_eh_detach_dev(dev);
2085 /* probe if requested */
2086 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2087 !(ehc->did_probe_mask & (1 << dev->devno))) {
2088 ata_eh_detach_dev(dev);
2091 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2092 ehc->did_probe_mask |= (1 << dev->devno);
2093 ehc->i.action |= ATA_EH_SOFTRESET;
2096 /* soft didn't work? be haaaaard */
2097 if (ehc->i.flags & ATA_EHI_DID_RESET)
2098 ehc->i.action |= ATA_EH_HARDRESET;
2100 ehc->i.action |= ATA_EH_SOFTRESET;
2103 if (ata_port_nr_enabled(ap)) {
2104 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2105 "devices, retrying in 5 secs\n");
2108 /* no device left, repeat fast */
2116 for (i = 0; i < ATA_MAX_DEVICES; i++)
2117 ata_dev_disable(&ap->device[i]);
2120 DPRINTK("EXIT, rc=%d\n", rc);
2125 * ata_eh_finish - finish up EH
2126 * @ap: host port to finish EH for
2128 * Recovery is complete. Clean up EH states and retry or finish
2134 static void ata_eh_finish(struct ata_port *ap)
2138 /* retry or finish qcs */
2139 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2140 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2142 if (!(qc->flags & ATA_QCFLAG_FAILED))
2146 /* FIXME: Once EH migration is complete,
2147 * generate sense data in this function,
2148 * considering both err_mask and tf.
2150 if (qc->err_mask & AC_ERR_INVALID)
2151 ata_eh_qc_complete(qc);
2153 ata_eh_qc_retry(qc);
2155 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2156 ata_eh_qc_complete(qc);
2158 /* feed zero TF to sense generation */
2159 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2160 ata_eh_qc_retry(qc);
2167 * ata_do_eh - do standard error handling
2168 * @ap: host port to handle error for
2169 * @prereset: prereset method (can be NULL)
2170 * @softreset: softreset method (can be NULL)
2171 * @hardreset: hardreset method (can be NULL)
2172 * @postreset: postreset method (can be NULL)
2174 * Perform standard error handling sequence.
2177 * Kernel thread context (may sleep).
2179 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2180 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2181 ata_postreset_fn_t postreset)
2185 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
2190 * ata_eh_handle_port_suspend - perform port suspend operation
2191 * @ap: port to suspend
2196 * Kernel thread context (may sleep).
2198 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2200 unsigned long flags;
2203 /* are we suspending? */
2204 spin_lock_irqsave(ap->lock, flags);
2205 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2206 ap->pm_mesg.event == PM_EVENT_ON) {
2207 spin_unlock_irqrestore(ap->lock, flags);
2210 spin_unlock_irqrestore(ap->lock, flags);
2212 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2215 ata_eh_freeze_port(ap);
2217 if (ap->ops->port_suspend)
2218 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2221 spin_lock_irqsave(ap->lock, flags);
2223 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2225 ap->pflags |= ATA_PFLAG_SUSPENDED;
2227 ata_port_schedule_eh(ap);
2229 if (ap->pm_result) {
2230 *ap->pm_result = rc;
2231 ap->pm_result = NULL;
2234 spin_unlock_irqrestore(ap->lock, flags);
2240 * ata_eh_handle_port_resume - perform port resume operation
2241 * @ap: port to resume
2245 * This function also waits upto one second until all devices
2246 * hanging off this port requests resume EH action. This is to
2247 * prevent invoking EH and thus reset multiple times on resume.
2249 * On DPM resume, where some of devices might not be resumed
2250 * together, this may delay port resume upto one second, but such
2251 * DPM resumes are rare and 1 sec delay isn't too bad.
2254 * Kernel thread context (may sleep).
2256 static void ata_eh_handle_port_resume(struct ata_port *ap)
2258 unsigned long timeout;
2259 unsigned long flags;
2262 /* are we resuming? */
2263 spin_lock_irqsave(ap->lock, flags);
2264 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2265 ap->pm_mesg.event != PM_EVENT_ON) {
2266 spin_unlock_irqrestore(ap->lock, flags);
2269 spin_unlock_irqrestore(ap->lock, flags);
2272 if (!(ap->pflags & ATA_PFLAG_SUSPENDED))
2275 if (ap->ops->port_resume)
2276 rc = ap->ops->port_resume(ap);
2278 /* give devices time to request EH */
2279 timeout = jiffies + HZ; /* 1s max */
2281 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2282 struct ata_device *dev = &ap->device[i];
2283 unsigned int action = ata_eh_dev_action(dev);
2285 if ((dev->flags & ATA_DFLAG_SUSPENDED) &&
2286 !(action & ATA_EH_RESUME))
2290 if (i == ATA_MAX_DEVICES || time_after(jiffies, timeout))
2296 spin_lock_irqsave(ap->lock, flags);
2297 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2298 if (ap->pm_result) {
2299 *ap->pm_result = rc;
2300 ap->pm_result = NULL;
2302 spin_unlock_irqrestore(ap->lock, flags);