2 * libata-eh.c - libata error handling
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_work(ata_aux_wq, &ap->hotplug_task);
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 tmp = atapi_eh_request_sense(qc->dev,
1140 qc->scsicmd->sense_buffer);
1142 /* ATA_QCFLAG_SENSE_VALID is used to tell
1143 * atapi_qc_complete() that sense data is
1146 * TODO: interpret sense data and set
1147 * appropriate err_mask.
1149 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1151 qc->err_mask |= tmp;
1154 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1155 action |= ATA_EH_SOFTRESET;
1160 static int ata_eh_categorize_ering_entry(struct ata_ering_entry *ent)
1162 if (ent->err_mask & (AC_ERR_ATA_BUS | AC_ERR_TIMEOUT))
1166 if (ent->err_mask & AC_ERR_HSM)
1168 if ((ent->err_mask &
1169 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1176 struct speed_down_needed_arg {
1181 static int speed_down_needed_cb(struct ata_ering_entry *ent, void *void_arg)
1183 struct speed_down_needed_arg *arg = void_arg;
1185 if (ent->timestamp < arg->since)
1188 arg->nr_errors[ata_eh_categorize_ering_entry(ent)]++;
1193 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
1194 * @dev: Device of interest
1196 * This function examines error ring of @dev and determines
1197 * whether speed down is necessary. Speed down is necessary if
1198 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
1199 * errors during last 15 minutes.
1201 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
1202 * violation for known supported commands.
1204 * Cat-2 errors are unclassified DEV error for known supported
1208 * Inherited from caller.
1211 * 1 if speed down is necessary, 0 otherwise
1213 static int ata_eh_speed_down_needed(struct ata_device *dev)
1215 const u64 interval = 15LLU * 60 * HZ;
1216 static const int err_limits[3] = { -1, 3, 10 };
1217 struct speed_down_needed_arg arg;
1218 struct ata_ering_entry *ent;
1222 ent = ata_ering_top(&dev->ering);
1226 err_cat = ata_eh_categorize_ering_entry(ent);
1230 memset(&arg, 0, sizeof(arg));
1232 j64 = get_jiffies_64();
1233 if (j64 >= interval)
1234 arg.since = j64 - interval;
1238 ata_ering_map(&dev->ering, speed_down_needed_cb, &arg);
1240 return arg.nr_errors[err_cat] > err_limits[err_cat];
1244 * ata_eh_speed_down - record error and speed down if necessary
1245 * @dev: Failed device
1246 * @is_io: Did the device fail during normal IO?
1247 * @err_mask: err_mask of the error
1249 * Record error and examine error history to determine whether
1250 * adjusting transmission speed is necessary. It also sets
1251 * transmission limits appropriately if such adjustment is
1255 * Kernel thread context (may sleep).
1258 * 0 on success, -errno otherwise
1260 static int ata_eh_speed_down(struct ata_device *dev, int is_io,
1261 unsigned int err_mask)
1266 /* record error and determine whether speed down is necessary */
1267 ata_ering_record(&dev->ering, is_io, err_mask);
1269 if (!ata_eh_speed_down_needed(dev))
1272 /* speed down SATA link speed if possible */
1273 if (sata_down_spd_limit(dev->ap) == 0)
1274 return ATA_EH_HARDRESET;
1276 /* lower transfer mode */
1277 if (ata_down_xfermask_limit(dev, 0) == 0)
1278 return ATA_EH_SOFTRESET;
1280 ata_dev_printk(dev, KERN_ERR,
1281 "speed down requested but no transfer mode left\n");
1286 * ata_eh_autopsy - analyze error and determine recovery action
1287 * @ap: ATA port to perform autopsy on
1289 * Analyze why @ap failed and determine which recovery action is
1290 * needed. This function also sets more detailed AC_ERR_* values
1291 * and fills sense data for ATAPI CHECK SENSE.
1294 * Kernel thread context (may sleep).
1296 static void ata_eh_autopsy(struct ata_port *ap)
1298 struct ata_eh_context *ehc = &ap->eh_context;
1299 unsigned int all_err_mask = 0;
1306 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1309 /* obtain and analyze SError */
1310 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1312 ehc->i.serror |= serror;
1313 ata_eh_analyze_serror(ap);
1314 } else if (rc != -EOPNOTSUPP)
1315 ehc->i.action |= ATA_EH_HARDRESET;
1317 /* analyze NCQ failure */
1318 ata_eh_analyze_ncq_error(ap);
1320 /* any real error trumps AC_ERR_OTHER */
1321 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1322 ehc->i.err_mask &= ~AC_ERR_OTHER;
1324 all_err_mask |= ehc->i.err_mask;
1326 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1327 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1329 if (!(qc->flags & ATA_QCFLAG_FAILED))
1332 /* inherit upper level err_mask */
1333 qc->err_mask |= ehc->i.err_mask;
1336 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1338 /* DEV errors are probably spurious in case of ATA_BUS error */
1339 if (qc->err_mask & AC_ERR_ATA_BUS)
1340 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1343 /* any real error trumps unknown error */
1344 if (qc->err_mask & ~AC_ERR_OTHER)
1345 qc->err_mask &= ~AC_ERR_OTHER;
1347 /* SENSE_VALID trumps dev/unknown error and revalidation */
1348 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1349 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1350 ehc->i.action &= ~ATA_EH_REVALIDATE;
1353 /* accumulate error info */
1354 ehc->i.dev = qc->dev;
1355 all_err_mask |= qc->err_mask;
1356 if (qc->flags & ATA_QCFLAG_IO)
1360 /* enforce default EH actions */
1361 if (ap->pflags & ATA_PFLAG_FROZEN ||
1362 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1363 ehc->i.action |= ATA_EH_SOFTRESET;
1364 else if (all_err_mask)
1365 ehc->i.action |= ATA_EH_REVALIDATE;
1367 /* if we have offending qcs and the associated failed device */
1370 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1373 /* perform per-dev EH action only on the offending device */
1374 ehc->i.dev_action[ehc->i.dev->devno] |=
1375 ehc->i.action & ATA_EH_PERDEV_MASK;
1376 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1383 * ata_eh_report - report error handling to user
1384 * @ap: ATA port EH is going on
1386 * Report EH to user.
1391 static void ata_eh_report(struct ata_port *ap)
1393 struct ata_eh_context *ehc = &ap->eh_context;
1394 const char *frozen, *desc;
1395 int tag, nr_failed = 0;
1398 if (ehc->i.desc[0] != '\0')
1401 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1402 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1404 if (!(qc->flags & ATA_QCFLAG_FAILED))
1406 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1412 if (!nr_failed && !ehc->i.err_mask)
1416 if (ap->pflags & ATA_PFLAG_FROZEN)
1420 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1421 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1422 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1423 ehc->i.action, frozen);
1425 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1427 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1428 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1429 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1430 ehc->i.action, frozen);
1432 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1435 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1436 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1438 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1441 ata_dev_printk(qc->dev, KERN_ERR, "tag %d cmd 0x%x "
1442 "Emask 0x%x stat 0x%x err 0x%x (%s)\n",
1443 qc->tag, qc->tf.command, qc->err_mask,
1444 qc->result_tf.command, qc->result_tf.feature,
1445 ata_err_string(qc->err_mask));
1449 static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1450 unsigned int *classes)
1454 for (i = 0; i < ATA_MAX_DEVICES; i++)
1455 classes[i] = ATA_DEV_UNKNOWN;
1457 rc = reset(ap, classes);
1461 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1462 * is complete and convert all ATA_DEV_UNKNOWN to
1465 for (i = 0; i < ATA_MAX_DEVICES; i++)
1466 if (classes[i] != ATA_DEV_UNKNOWN)
1469 if (i < ATA_MAX_DEVICES)
1470 for (i = 0; i < ATA_MAX_DEVICES; i++)
1471 if (classes[i] == ATA_DEV_UNKNOWN)
1472 classes[i] = ATA_DEV_NONE;
1477 static int ata_eh_followup_srst_needed(int rc, int classify,
1478 const unsigned int *classes)
1484 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1489 static int ata_eh_reset(struct ata_port *ap, int classify,
1490 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
1491 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1493 struct ata_eh_context *ehc = &ap->eh_context;
1494 unsigned int *classes = ehc->classes;
1495 int tries = ATA_EH_RESET_TRIES;
1496 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
1497 unsigned int action;
1498 ata_reset_fn_t reset;
1499 int i, did_followup_srst, rc;
1501 /* about to reset */
1502 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1504 /* Determine which reset to use and record in ehc->i.action.
1505 * prereset() may examine and modify it.
1507 action = ehc->i.action;
1508 ehc->i.action &= ~ATA_EH_RESET_MASK;
1509 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
1510 !(action & ATA_EH_HARDRESET))))
1511 ehc->i.action |= ATA_EH_SOFTRESET;
1513 ehc->i.action |= ATA_EH_HARDRESET;
1518 if (rc == -ENOENT) {
1519 ata_port_printk(ap, KERN_DEBUG, "port disabled. ignoring.\n");
1520 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
1522 ata_port_printk(ap, KERN_ERR,
1523 "prereset failed (errno=%d)\n", rc);
1528 /* prereset() might have modified ehc->i.action */
1529 if (ehc->i.action & ATA_EH_HARDRESET)
1531 else if (ehc->i.action & ATA_EH_SOFTRESET)
1534 /* prereset told us not to reset, bang classes and return */
1535 for (i = 0; i < ATA_MAX_DEVICES; i++)
1536 classes[i] = ATA_DEV_NONE;
1540 /* did prereset() screw up? if so, fix up to avoid oopsing */
1542 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1543 "invalid reset type\n");
1551 /* shut up during boot probing */
1553 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1554 reset == softreset ? "soft" : "hard");
1556 /* mark that this EH session started with reset */
1557 ehc->i.flags |= ATA_EHI_DID_RESET;
1559 rc = ata_do_reset(ap, reset, classes);
1561 did_followup_srst = 0;
1562 if (reset == hardreset &&
1563 ata_eh_followup_srst_needed(rc, classify, classes)) {
1564 /* okay, let's do follow-up softreset */
1565 did_followup_srst = 1;
1569 ata_port_printk(ap, KERN_ERR,
1570 "follow-up softreset required "
1571 "but no softreset avaliable\n");
1575 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
1576 rc = ata_do_reset(ap, reset, classes);
1578 if (rc == 0 && classify &&
1579 classes[0] == ATA_DEV_UNKNOWN) {
1580 ata_port_printk(ap, KERN_ERR,
1581 "classification failed\n");
1586 if (rc && --tries) {
1589 if (reset == softreset) {
1590 if (did_followup_srst)
1591 type = "follow-up soft";
1597 ata_port_printk(ap, KERN_WARNING,
1598 "%sreset failed, retrying in 5 secs\n", type);
1601 if (reset == hardreset)
1602 sata_down_spd_limit(ap);
1609 /* After the reset, the device state is PIO 0 and the
1610 * controller state is undefined. Record the mode.
1612 for (i = 0; i < ATA_MAX_DEVICES; i++)
1613 ap->device[i].pio_mode = XFER_PIO_0;
1616 postreset(ap, classes);
1618 /* reset successful, schedule revalidation */
1619 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1620 ehc->i.action |= ATA_EH_REVALIDATE;
1626 static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1627 struct ata_device **r_failed_dev)
1629 struct ata_eh_context *ehc = &ap->eh_context;
1630 struct ata_device *dev;
1631 unsigned long flags;
1636 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1637 unsigned int action, readid_flags = 0;
1639 dev = &ap->device[i];
1640 action = ata_eh_dev_action(dev);
1642 if (ehc->i.flags & ATA_EHI_DID_RESET)
1643 readid_flags |= ATA_READID_POSTRESET;
1645 if (action & ATA_EH_REVALIDATE && ata_dev_ready(dev)) {
1646 if (ata_port_offline(ap)) {
1651 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
1652 rc = ata_dev_revalidate(dev, readid_flags);
1656 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1658 /* Configuration may have changed, reconfigure
1661 ehc->i.flags |= ATA_EHI_SETMODE;
1663 /* schedule the scsi_rescan_device() here */
1664 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
1665 } else if (dev->class == ATA_DEV_UNKNOWN &&
1666 ehc->tries[dev->devno] &&
1667 ata_class_enabled(ehc->classes[dev->devno])) {
1668 dev->class = ehc->classes[dev->devno];
1670 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1673 ehc->i.flags |= ATA_EHI_PRINTINFO;
1674 rc = ata_dev_configure(dev);
1675 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
1679 dev->class = ATA_DEV_UNKNOWN;
1683 spin_lock_irqsave(ap->lock, flags);
1684 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1685 spin_unlock_irqrestore(ap->lock, flags);
1687 /* new device discovered, configure transfer mode */
1688 ehc->i.flags |= ATA_EHI_SETMODE;
1693 *r_failed_dev = dev;
1700 * ata_eh_suspend - handle suspend EH action
1701 * @ap: target host port
1702 * @r_failed_dev: result parameter to indicate failing device
1704 * Handle suspend EH action. Disk devices are spinned down and
1705 * other types of devices are just marked suspended. Once
1706 * suspended, no EH action to the device is allowed until it is
1710 * Kernel thread context (may sleep).
1713 * 0 on success, -errno otherwise
1715 static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev)
1717 struct ata_device *dev;
1722 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1723 unsigned long flags;
1724 unsigned int action, err_mask;
1726 dev = &ap->device[i];
1727 action = ata_eh_dev_action(dev);
1729 if (!ata_dev_enabled(dev) || !(action & ATA_EH_SUSPEND))
1732 WARN_ON(dev->flags & ATA_DFLAG_SUSPENDED);
1734 ata_eh_about_to_do(ap, dev, ATA_EH_SUSPEND);
1736 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1738 rc = ata_flush_cache(dev);
1743 err_mask = ata_do_simple_cmd(dev, ATA_CMD_STANDBYNOW1);
1745 ata_dev_printk(dev, KERN_ERR, "failed to "
1746 "spin down (err_mask=0x%x)\n",
1753 spin_lock_irqsave(ap->lock, flags);
1754 dev->flags |= ATA_DFLAG_SUSPENDED;
1755 spin_unlock_irqrestore(ap->lock, flags);
1757 ata_eh_done(ap, dev, ATA_EH_SUSPEND);
1761 *r_failed_dev = dev;
1768 * ata_eh_prep_resume - prep for resume EH action
1769 * @ap: target host port
1771 * Clear SUSPENDED in preparation for scheduled resume actions.
1772 * This allows other parts of EH to access the devices being
1776 * Kernel thread context (may sleep).
1778 static void ata_eh_prep_resume(struct ata_port *ap)
1780 struct ata_device *dev;
1781 unsigned long flags;
1786 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1787 unsigned int action;
1789 dev = &ap->device[i];
1790 action = ata_eh_dev_action(dev);
1792 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1795 spin_lock_irqsave(ap->lock, flags);
1796 dev->flags &= ~ATA_DFLAG_SUSPENDED;
1797 spin_unlock_irqrestore(ap->lock, flags);
1804 * ata_eh_resume - handle resume EH action
1805 * @ap: target host port
1806 * @r_failed_dev: result parameter to indicate failing device
1808 * Handle resume EH action. Target devices are already reset and
1809 * revalidated. Spinning up is the only operation left.
1812 * Kernel thread context (may sleep).
1815 * 0 on success, -errno otherwise
1817 static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev)
1819 struct ata_device *dev;
1824 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1825 unsigned int action, err_mask;
1827 dev = &ap->device[i];
1828 action = ata_eh_dev_action(dev);
1830 if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME))
1833 ata_eh_about_to_do(ap, dev, ATA_EH_RESUME);
1835 if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) {
1836 err_mask = ata_do_simple_cmd(dev,
1837 ATA_CMD_IDLEIMMEDIATE);
1839 ata_dev_printk(dev, KERN_ERR, "failed to "
1840 "spin up (err_mask=0x%x)\n",
1847 ata_eh_done(ap, dev, ATA_EH_RESUME);
1851 *r_failed_dev = dev;
1857 static int ata_port_nr_enabled(struct ata_port *ap)
1861 for (i = 0; i < ATA_MAX_DEVICES; i++)
1862 if (ata_dev_enabled(&ap->device[i]))
1867 static int ata_port_nr_vacant(struct ata_port *ap)
1871 for (i = 0; i < ATA_MAX_DEVICES; i++)
1872 if (ap->device[i].class == ATA_DEV_UNKNOWN)
1877 static int ata_eh_skip_recovery(struct ata_port *ap)
1879 struct ata_eh_context *ehc = &ap->eh_context;
1882 /* skip if all possible devices are suspended */
1883 for (i = 0; i < ata_port_max_devices(ap); i++) {
1884 struct ata_device *dev = &ap->device[i];
1886 if (!(dev->flags & ATA_DFLAG_SUSPENDED))
1890 if (i == ata_port_max_devices(ap))
1893 /* thaw frozen port, resume link and recover failed devices */
1894 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
1895 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
1898 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1899 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1900 struct ata_device *dev = &ap->device[i];
1902 if (dev->class == ATA_DEV_UNKNOWN &&
1903 ehc->classes[dev->devno] != ATA_DEV_NONE)
1911 * ata_eh_recover - recover host port after error
1912 * @ap: host port to recover
1913 * @prereset: prereset method (can be NULL)
1914 * @softreset: softreset method (can be NULL)
1915 * @hardreset: hardreset method (can be NULL)
1916 * @postreset: postreset method (can be NULL)
1918 * This is the alpha and omega, eum and yang, heart and soul of
1919 * libata exception handling. On entry, actions required to
1920 * recover the port and hotplug requests are recorded in
1921 * eh_context. This function executes all the operations with
1922 * appropriate retrials and fallbacks to resurrect failed
1923 * devices, detach goners and greet newcomers.
1926 * Kernel thread context (may sleep).
1929 * 0 on success, -errno on failure.
1931 static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
1932 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
1933 ata_postreset_fn_t postreset)
1935 struct ata_eh_context *ehc = &ap->eh_context;
1936 struct ata_device *dev;
1937 int down_xfermask, i, rc;
1941 /* prep for recovery */
1942 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1943 dev = &ap->device[i];
1945 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
1947 /* process hotplug request */
1948 if (dev->flags & ATA_DFLAG_DETACH)
1949 ata_eh_detach_dev(dev);
1951 if (!ata_dev_enabled(dev) &&
1952 ((ehc->i.probe_mask & (1 << dev->devno)) &&
1953 !(ehc->did_probe_mask & (1 << dev->devno)))) {
1954 ata_eh_detach_dev(dev);
1956 ehc->did_probe_mask |= (1 << dev->devno);
1957 ehc->i.action |= ATA_EH_SOFTRESET;
1965 /* if UNLOADING, finish immediately */
1966 if (ap->pflags & ATA_PFLAG_UNLOADING)
1969 /* prep for resume */
1970 ata_eh_prep_resume(ap);
1972 /* skip EH if possible. */
1973 if (ata_eh_skip_recovery(ap))
1976 for (i = 0; i < ATA_MAX_DEVICES; i++)
1977 ehc->classes[i] = ATA_DEV_UNKNOWN;
1980 if (ehc->i.action & ATA_EH_RESET_MASK) {
1981 ata_eh_freeze_port(ap);
1983 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
1984 softreset, hardreset, postreset);
1986 ata_port_printk(ap, KERN_ERR,
1987 "reset failed, giving up\n");
1991 ata_eh_thaw_port(ap);
1994 /* revalidate existing devices and attach new ones */
1995 rc = ata_eh_revalidate_and_attach(ap, &dev);
1999 /* resume devices */
2000 rc = ata_eh_resume(ap, &dev);
2004 /* configure transfer mode if necessary */
2005 if (ehc->i.flags & ATA_EHI_SETMODE) {
2006 rc = ata_set_mode(ap, &dev);
2011 ehc->i.flags &= ~ATA_EHI_SETMODE;
2014 /* suspend devices */
2015 rc = ata_eh_suspend(ap, &dev);
2024 /* device missing, schedule probing */
2025 ehc->i.probe_mask |= (1 << dev->devno);
2027 ehc->tries[dev->devno] = 0;
2030 sata_down_spd_limit(ap);
2032 ehc->tries[dev->devno]--;
2033 if (down_xfermask &&
2034 ata_down_xfermask_limit(dev, ehc->tries[dev->devno] == 1))
2035 ehc->tries[dev->devno] = 0;
2038 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2039 /* disable device if it has used up all its chances */
2040 ata_dev_disable(dev);
2042 /* detach if offline */
2043 if (ata_port_offline(ap))
2044 ata_eh_detach_dev(dev);
2046 /* probe if requested */
2047 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2048 !(ehc->did_probe_mask & (1 << dev->devno))) {
2049 ata_eh_detach_dev(dev);
2052 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2053 ehc->did_probe_mask |= (1 << dev->devno);
2054 ehc->i.action |= ATA_EH_SOFTRESET;
2057 /* soft didn't work? be haaaaard */
2058 if (ehc->i.flags & ATA_EHI_DID_RESET)
2059 ehc->i.action |= ATA_EH_HARDRESET;
2061 ehc->i.action |= ATA_EH_SOFTRESET;
2064 if (ata_port_nr_enabled(ap)) {
2065 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2066 "devices, retrying in 5 secs\n");
2069 /* no device left, repeat fast */
2077 for (i = 0; i < ATA_MAX_DEVICES; i++)
2078 ata_dev_disable(&ap->device[i]);
2081 DPRINTK("EXIT, rc=%d\n", rc);
2086 * ata_eh_finish - finish up EH
2087 * @ap: host port to finish EH for
2089 * Recovery is complete. Clean up EH states and retry or finish
2095 static void ata_eh_finish(struct ata_port *ap)
2099 /* retry or finish qcs */
2100 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2101 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2103 if (!(qc->flags & ATA_QCFLAG_FAILED))
2107 /* FIXME: Once EH migration is complete,
2108 * generate sense data in this function,
2109 * considering both err_mask and tf.
2111 if (qc->err_mask & AC_ERR_INVALID)
2112 ata_eh_qc_complete(qc);
2114 ata_eh_qc_retry(qc);
2116 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2117 ata_eh_qc_complete(qc);
2119 /* feed zero TF to sense generation */
2120 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2121 ata_eh_qc_retry(qc);
2128 * ata_do_eh - do standard error handling
2129 * @ap: host port to handle error for
2130 * @prereset: prereset method (can be NULL)
2131 * @softreset: softreset method (can be NULL)
2132 * @hardreset: hardreset method (can be NULL)
2133 * @postreset: postreset method (can be NULL)
2135 * Perform standard error handling sequence.
2138 * Kernel thread context (may sleep).
2140 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2141 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2142 ata_postreset_fn_t postreset)
2146 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
2151 * ata_eh_handle_port_suspend - perform port suspend operation
2152 * @ap: port to suspend
2157 * Kernel thread context (may sleep).
2159 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2161 unsigned long flags;
2164 /* are we suspending? */
2165 spin_lock_irqsave(ap->lock, flags);
2166 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2167 ap->pm_mesg.event == PM_EVENT_ON) {
2168 spin_unlock_irqrestore(ap->lock, flags);
2171 spin_unlock_irqrestore(ap->lock, flags);
2173 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2176 ata_eh_freeze_port(ap);
2178 if (ap->ops->port_suspend)
2179 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2182 spin_lock_irqsave(ap->lock, flags);
2184 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2186 ap->pflags |= ATA_PFLAG_SUSPENDED;
2188 ata_port_schedule_eh(ap);
2190 if (ap->pm_result) {
2191 *ap->pm_result = rc;
2192 ap->pm_result = NULL;
2195 spin_unlock_irqrestore(ap->lock, flags);
2201 * ata_eh_handle_port_resume - perform port resume operation
2202 * @ap: port to resume
2206 * This function also waits upto one second until all devices
2207 * hanging off this port requests resume EH action. This is to
2208 * prevent invoking EH and thus reset multiple times on resume.
2210 * On DPM resume, where some of devices might not be resumed
2211 * together, this may delay port resume upto one second, but such
2212 * DPM resumes are rare and 1 sec delay isn't too bad.
2215 * Kernel thread context (may sleep).
2217 static void ata_eh_handle_port_resume(struct ata_port *ap)
2219 unsigned long timeout;
2220 unsigned long flags;
2223 /* are we resuming? */
2224 spin_lock_irqsave(ap->lock, flags);
2225 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2226 ap->pm_mesg.event != PM_EVENT_ON) {
2227 spin_unlock_irqrestore(ap->lock, flags);
2230 spin_unlock_irqrestore(ap->lock, flags);
2233 if (!(ap->pflags & ATA_PFLAG_SUSPENDED))
2236 if (ap->ops->port_resume)
2237 rc = ap->ops->port_resume(ap);
2239 /* give devices time to request EH */
2240 timeout = jiffies + HZ; /* 1s max */
2242 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2243 struct ata_device *dev = &ap->device[i];
2244 unsigned int action = ata_eh_dev_action(dev);
2246 if ((dev->flags & ATA_DFLAG_SUSPENDED) &&
2247 !(action & ATA_EH_RESUME))
2251 if (i == ATA_MAX_DEVICES || time_after(jiffies, timeout))
2257 spin_lock_irqsave(ap->lock, flags);
2258 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2259 if (ap->pm_result) {
2260 *ap->pm_result = rc;
2261 ap->pm_result = NULL;
2263 spin_unlock_irqrestore(ap->lock, flags);