4 * Basic PIO and command management functionality.
6 * This code was split off from ide.c. See ide.c for history and original
9 * This program is free software; you can redistribute it and/or modify it
10 * under the terms of the GNU General Public License as published by the
11 * Free Software Foundation; either version 2, or (at your option) any
14 * This program is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * General Public License for more details.
19 * For the avoidance of doubt the "preferred form" of this code is one which
20 * is in an open non patent encumbered format. Where cryptographic key signing
21 * forms part of the process of creating an executable the information
22 * including keys needed to generate an equivalently functional executable
23 * are deemed to be part of the source code.
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/hdreg.h>
44 #include <linux/completion.h>
45 #include <linux/reboot.h>
46 #include <linux/cdrom.h>
47 #include <linux/seq_file.h>
48 #include <linux/device.h>
49 #include <linux/kmod.h>
50 #include <linux/scatterlist.h>
51 #include <linux/bitops.h>
53 #include <asm/byteorder.h>
55 #include <asm/uaccess.h>
58 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
59 int uptodate, unsigned int nr_bytes, int dequeue)
65 error = uptodate ? uptodate : -EIO;
68 * if failfast is set on a request, override number of sectors and
69 * complete the whole request right now
71 if (blk_noretry_request(rq) && error)
72 nr_bytes = rq->hard_nr_sectors << 9;
74 if (!blk_fs_request(rq) && error && !rq->errors)
78 * decide whether to reenable DMA -- 3 is a random magic for now,
79 * if we DMA timeout more than 3 times, just stay in PIO
81 if ((drive->dev_flags & IDE_DFLAG_DMA_PIO_RETRY) &&
82 drive->retry_pio <= 3) {
83 drive->dev_flags &= ~IDE_DFLAG_DMA_PIO_RETRY;
87 if (!blk_end_request(rq, error, nr_bytes))
90 if (ret == 0 && dequeue)
91 drive->hwif->hwgroup->rq = NULL;
97 * ide_end_request - complete an IDE I/O
98 * @drive: IDE device for the I/O
100 * @nr_sectors: number of sectors completed
102 * This is our end_request wrapper function. We complete the I/O
103 * update random number input and dequeue the request, which if
104 * it was tagged may be out of order.
107 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
109 unsigned int nr_bytes = nr_sectors << 9;
110 struct request *rq = drive->hwif->hwgroup->rq;
113 if (blk_pc_request(rq))
114 nr_bytes = rq->data_len;
116 nr_bytes = rq->hard_cur_sectors << 9;
119 return __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
121 EXPORT_SYMBOL(ide_end_request);
124 * ide_end_dequeued_request - complete an IDE I/O
125 * @drive: IDE device for the I/O
127 * @nr_sectors: number of sectors completed
129 * Complete an I/O that is no longer on the request queue. This
130 * typically occurs when we pull the request and issue a REQUEST_SENSE.
131 * We must still finish the old request but we must not tamper with the
132 * queue in the meantime.
134 * NOTE: This path does not handle barrier, but barrier is not supported
138 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
139 int uptodate, int nr_sectors)
141 BUG_ON(!blk_rq_started(rq));
143 return __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
145 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
148 * ide_end_drive_cmd - end an explicit drive command
153 * Clean up after success/failure of an explicit drive command.
154 * These get thrown onto the queue so they are synchronized with
155 * real I/O operations on the drive.
157 * In LBA48 mode we have to read the register set twice to get
158 * all the extra information out.
161 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
163 ide_hwgroup_t *hwgroup = drive->hwif->hwgroup;
164 struct request *rq = hwgroup->rq;
166 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
167 ide_task_t *task = (ide_task_t *)rq->special;
170 struct ide_taskfile *tf = &task->tf;
175 drive->hwif->tp_ops->tf_read(drive, task);
177 if (task->tf_flags & IDE_TFLAG_DYN)
180 } else if (blk_pm_request(rq)) {
181 struct request_pm_state *pm = rq->data;
183 ide_complete_power_step(drive, rq);
184 if (pm->pm_step == IDE_PM_COMPLETED)
185 ide_complete_pm_request(drive, rq);
193 if (unlikely(blk_end_request(rq, (rq->errors ? -EIO : 0),
197 EXPORT_SYMBOL(ide_end_drive_cmd);
199 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
204 drv = *(ide_driver_t **)rq->rq_disk->private_data;
205 drv->end_request(drive, 0, 0);
207 ide_end_request(drive, 0, 0);
210 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
212 ide_hwif_t *hwif = drive->hwif;
214 if ((stat & ATA_BUSY) ||
215 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
216 /* other bits are useless when BUSY */
217 rq->errors |= ERROR_RESET;
218 } else if (stat & ATA_ERR) {
219 /* err has different meaning on cdrom and tape */
220 if (err == ATA_ABORTED) {
221 if ((drive->dev_flags & IDE_DFLAG_LBA) &&
222 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
223 hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
225 } else if ((err & BAD_CRC) == BAD_CRC) {
226 /* UDMA crc error, just retry the operation */
228 } else if (err & (ATA_BBK | ATA_UNC)) {
229 /* retries won't help these */
230 rq->errors = ERROR_MAX;
231 } else if (err & ATA_TRK0NF) {
232 /* help it find track zero */
233 rq->errors |= ERROR_RECAL;
237 if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
238 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
239 int nsect = drive->mult_count ? drive->mult_count : 1;
241 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
244 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
245 ide_kill_rq(drive, rq);
249 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
250 rq->errors |= ERROR_RESET;
252 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
254 return ide_do_reset(drive);
257 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
258 drive->special.b.recalibrate = 1;
265 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
267 ide_hwif_t *hwif = drive->hwif;
269 if ((stat & ATA_BUSY) ||
270 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
271 /* other bits are useless when BUSY */
272 rq->errors |= ERROR_RESET;
274 /* add decoding error stuff */
277 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
279 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
281 if (rq->errors >= ERROR_MAX) {
282 ide_kill_rq(drive, rq);
284 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
286 return ide_do_reset(drive);
295 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
297 if (drive->media == ide_disk)
298 return ide_ata_error(drive, rq, stat, err);
299 return ide_atapi_error(drive, rq, stat, err);
302 EXPORT_SYMBOL_GPL(__ide_error);
305 * ide_error - handle an error on the IDE
306 * @drive: drive the error occurred on
307 * @msg: message to report
310 * ide_error() takes action based on the error returned by the drive.
311 * For normal I/O that may well include retries. We deal with
312 * both new-style (taskfile) and old style command handling here.
313 * In the case of taskfile command handling there is work left to
317 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
322 err = ide_dump_status(drive, msg, stat);
324 if ((rq = HWGROUP(drive)->rq) == NULL)
327 /* retry only "normal" I/O: */
328 if (!blk_fs_request(rq)) {
330 ide_end_drive_cmd(drive, stat, err);
337 drv = *(ide_driver_t **)rq->rq_disk->private_data;
338 return drv->error(drive, rq, stat, err);
340 return __ide_error(drive, rq, stat, err);
343 EXPORT_SYMBOL_GPL(ide_error);
345 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
347 tf->nsect = drive->sect;
348 tf->lbal = drive->sect;
349 tf->lbam = drive->cyl;
350 tf->lbah = drive->cyl >> 8;
351 tf->device = (drive->head - 1) | drive->select;
352 tf->command = ATA_CMD_INIT_DEV_PARAMS;
355 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
357 tf->nsect = drive->sect;
358 tf->command = ATA_CMD_RESTORE;
361 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
363 tf->nsect = drive->mult_req;
364 tf->command = ATA_CMD_SET_MULTI;
367 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
369 special_t *s = &drive->special;
372 memset(&args, 0, sizeof(ide_task_t));
373 args.data_phase = TASKFILE_NO_DATA;
375 if (s->b.set_geometry) {
376 s->b.set_geometry = 0;
377 ide_tf_set_specify_cmd(drive, &args.tf);
378 } else if (s->b.recalibrate) {
379 s->b.recalibrate = 0;
380 ide_tf_set_restore_cmd(drive, &args.tf);
381 } else if (s->b.set_multmode) {
382 s->b.set_multmode = 0;
383 ide_tf_set_setmult_cmd(drive, &args.tf);
385 int special = s->all;
387 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
391 args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
392 IDE_TFLAG_CUSTOM_HANDLER;
394 do_rw_taskfile(drive, &args);
400 * do_special - issue some special commands
401 * @drive: drive the command is for
403 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
404 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
406 * It used to do much more, but has been scaled back.
409 static ide_startstop_t do_special (ide_drive_t *drive)
411 special_t *s = &drive->special;
414 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
416 if (drive->media == ide_disk)
417 return ide_disk_special(drive);
424 void ide_map_sg(ide_drive_t *drive, struct request *rq)
426 ide_hwif_t *hwif = drive->hwif;
427 struct scatterlist *sg = hwif->sg_table;
429 if (hwif->sg_mapped) /* needed by ide-scsi */
432 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
433 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
435 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
440 EXPORT_SYMBOL_GPL(ide_map_sg);
442 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
444 ide_hwif_t *hwif = drive->hwif;
446 hwif->nsect = hwif->nleft = rq->nr_sectors;
451 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
454 * execute_drive_command - issue special drive command
455 * @drive: the drive to issue the command on
456 * @rq: the request structure holding the command
458 * execute_drive_cmd() issues a special drive command, usually
459 * initiated by ioctl() from the external hdparm program. The
460 * command can be a drive command, drive task or taskfile
461 * operation. Weirdly you can call it with NULL to wait for
462 * all commands to finish. Don't do this as that is due to change
465 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
468 ide_hwif_t *hwif = HWIF(drive);
469 ide_task_t *task = rq->special;
472 hwif->data_phase = task->data_phase;
474 switch (hwif->data_phase) {
475 case TASKFILE_MULTI_OUT:
477 case TASKFILE_MULTI_IN:
479 ide_init_sg_cmd(drive, rq);
480 ide_map_sg(drive, rq);
485 return do_rw_taskfile(drive, task);
489 * NULL is actually a valid way of waiting for
490 * all current requests to be flushed from the queue.
493 printk("%s: DRIVE_CMD (null)\n", drive->name);
495 ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
496 ide_read_error(drive));
501 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
504 struct request_queue *q = drive->queue;
508 if (!(setting->flags & DS_SYNC))
509 return setting->set(drive, arg);
511 rq = blk_get_request(q, READ, __GFP_WAIT);
512 rq->cmd_type = REQ_TYPE_SPECIAL;
514 rq->cmd[0] = REQ_DEVSET_EXEC;
515 *(int *)&rq->cmd[1] = arg;
516 rq->special = setting->set;
518 if (blk_execute_rq(q, NULL, rq, 0))
524 EXPORT_SYMBOL_GPL(ide_devset_execute);
526 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
530 if (cmd == REQ_PARK_HEADS || cmd == REQ_UNPARK_HEADS) {
532 struct ide_taskfile *tf = &task.tf;
534 memset(&task, 0, sizeof(task));
535 if (cmd == REQ_PARK_HEADS) {
536 drive->sleep = *(unsigned long *)rq->special;
537 drive->dev_flags |= IDE_DFLAG_SLEEPING;
538 tf->command = ATA_CMD_IDLEIMMEDIATE;
543 task.tf_flags |= IDE_TFLAG_CUSTOM_HANDLER;
544 } else /* cmd == REQ_UNPARK_HEADS */
545 tf->command = ATA_CMD_CHK_POWER;
547 task.tf_flags |= IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
549 drive->hwif->data_phase = task.data_phase = TASKFILE_NO_DATA;
550 return do_rw_taskfile(drive, &task);
554 case REQ_DEVSET_EXEC:
556 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
558 err = setfunc(drive, *(int *)&rq->cmd[1]);
563 ide_end_request(drive, err, 0);
566 case REQ_DRIVE_RESET:
567 return ide_do_reset(drive);
569 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
570 ide_end_request(drive, 0, 0);
576 * start_request - start of I/O and command issuing for IDE
578 * start_request() initiates handling of a new I/O request. It
579 * accepts commands and I/O (read/write) requests.
581 * FIXME: this function needs a rename
584 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
586 ide_startstop_t startstop;
588 BUG_ON(!blk_rq_started(rq));
591 printk("%s: start_request: current=0x%08lx\n",
592 HWIF(drive)->name, (unsigned long) rq);
595 /* bail early if we've exceeded max_failures */
596 if (drive->max_failures && (drive->failures > drive->max_failures)) {
597 rq->cmd_flags |= REQ_FAILED;
601 if (blk_pm_request(rq))
602 ide_check_pm_state(drive, rq);
605 if (ide_wait_stat(&startstop, drive, drive->ready_stat,
606 ATA_BUSY | ATA_DRQ, WAIT_READY)) {
607 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
610 if (!drive->special.all) {
614 * We reset the drive so we need to issue a SETFEATURES.
615 * Do it _after_ do_special() restored device parameters.
617 if (drive->current_speed == 0xff)
618 ide_config_drive_speed(drive, drive->desired_speed);
620 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
621 return execute_drive_cmd(drive, rq);
622 else if (blk_pm_request(rq)) {
623 struct request_pm_state *pm = rq->data;
625 printk("%s: start_power_step(step: %d)\n",
626 drive->name, pm->pm_step);
628 startstop = ide_start_power_step(drive, rq);
629 if (startstop == ide_stopped &&
630 pm->pm_step == IDE_PM_COMPLETED)
631 ide_complete_pm_request(drive, rq);
633 } else if (!rq->rq_disk && blk_special_request(rq))
635 * TODO: Once all ULDs have been modified to
636 * check for specific op codes rather than
637 * blindly accepting any special request, the
638 * check for ->rq_disk above may be replaced
639 * by a more suitable mechanism or even
642 return ide_special_rq(drive, rq);
644 drv = *(ide_driver_t **)rq->rq_disk->private_data;
646 return drv->do_request(drive, rq, rq->sector);
648 return do_special(drive);
650 ide_kill_rq(drive, rq);
655 * ide_stall_queue - pause an IDE device
656 * @drive: drive to stall
657 * @timeout: time to stall for (jiffies)
659 * ide_stall_queue() can be used by a drive to give excess bandwidth back
660 * to the hwgroup by sleeping for timeout jiffies.
663 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
665 if (timeout > WAIT_WORSTCASE)
666 timeout = WAIT_WORSTCASE;
667 drive->sleep = timeout + jiffies;
668 drive->dev_flags |= IDE_DFLAG_SLEEPING;
671 EXPORT_SYMBOL(ide_stall_queue);
673 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
676 * choose_drive - select a drive to service
677 * @hwgroup: hardware group to select on
679 * choose_drive() selects the next drive which will be serviced.
680 * This is necessary because the IDE layer can't issue commands
681 * to both drives on the same cable, unlike SCSI.
684 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
686 ide_drive_t *drive, *best;
690 drive = hwgroup->drive;
693 * drive is doing pre-flush, ordered write, post-flush sequence. even
694 * though that is 3 requests, it must be seen as a single transaction.
695 * we must not preempt this drive until that is complete
697 if (blk_queue_flushing(drive->queue)) {
699 * small race where queue could get replugged during
700 * the 3-request flush cycle, just yank the plug since
701 * we want it to finish asap
703 blk_remove_plug(drive->queue);
708 u8 dev_s = !!(drive->dev_flags & IDE_DFLAG_SLEEPING);
709 u8 best_s = (best && !!(best->dev_flags & IDE_DFLAG_SLEEPING));
711 if ((dev_s == 0 || time_after_eq(jiffies, drive->sleep)) &&
712 !elv_queue_empty(drive->queue)) {
714 (dev_s && (best_s == 0 || time_before(drive->sleep, best->sleep))) ||
715 (best_s == 0 && time_before(WAKEUP(drive), WAKEUP(best)))) {
716 if (!blk_queue_plugged(drive->queue))
720 } while ((drive = drive->next) != hwgroup->drive);
722 if (best && (best->dev_flags & IDE_DFLAG_NICE1) &&
723 (best->dev_flags & IDE_DFLAG_SLEEPING) == 0 &&
724 best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
725 long t = (signed long)(WAKEUP(best) - jiffies);
726 if (t >= WAIT_MIN_SLEEP) {
728 * We *may* have some time to spare, but first let's see if
729 * someone can potentially benefit from our nice mood today..
733 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) == 0
734 && time_before(jiffies - best->service_time, WAKEUP(drive))
735 && time_before(WAKEUP(drive), jiffies + t))
737 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
740 } while ((drive = drive->next) != best);
747 * Issue a new request to a drive from hwgroup
748 * Caller must have already done spin_lock_irqsave(&hwgroup->lock, ..);
750 * A hwgroup is a serialized group of IDE interfaces. Usually there is
751 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
752 * may have both interfaces in a single hwgroup to "serialize" access.
753 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
754 * together into one hwgroup for serialized access.
756 * Note also that several hwgroups can end up sharing a single IRQ,
757 * possibly along with many other devices. This is especially common in
758 * PCI-based systems with off-board IDE controller cards.
760 * The IDE driver uses a per-hwgroup spinlock to protect
761 * access to the request queues, and to protect the hwgroup->busy flag.
763 * The first thread into the driver for a particular hwgroup sets the
764 * hwgroup->busy flag to indicate that this hwgroup is now active,
765 * and then initiates processing of the top request from the request queue.
767 * Other threads attempting entry notice the busy setting, and will simply
768 * queue their new requests and exit immediately. Note that hwgroup->busy
769 * remains set even when the driver is merely awaiting the next interrupt.
770 * Thus, the meaning is "this hwgroup is busy processing a request".
772 * When processing of a request completes, the completing thread or IRQ-handler
773 * will start the next request from the queue. If no more work remains,
774 * the driver will clear the hwgroup->busy flag and exit.
776 * The per-hwgroup spinlock is used to protect all access to the
777 * hwgroup->busy flag, but is otherwise not needed for most processing in
778 * the driver. This makes the driver much more friendlier to shared IRQs
779 * than previous designs, while remaining 100% (?) SMP safe and capable.
781 void do_ide_request(struct request_queue *q)
783 ide_drive_t *orig_drive = q->queuedata;
784 ide_hwgroup_t *hwgroup = orig_drive->hwif->hwgroup;
788 ide_startstop_t startstop;
790 /* caller must own hwgroup->lock */
791 BUG_ON(!irqs_disabled());
793 while (!hwgroup->busy) {
796 ide_get_lock(ide_intr, hwgroup);
797 drive = choose_drive(hwgroup);
800 unsigned long sleep = 0; /* shut up, gcc */
802 drive = hwgroup->drive;
804 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) &&
806 time_before(drive->sleep, sleep))) {
808 sleep = drive->sleep;
810 } while ((drive = drive->next) != hwgroup->drive);
813 * Take a short snooze, and then wake up this hwgroup again.
814 * This gives other hwgroups on the same a chance to
815 * play fairly with us, just in case there are big differences
816 * in relative throughputs.. don't want to hog the cpu too much.
818 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
819 sleep = jiffies + WAIT_MIN_SLEEP;
821 if (timer_pending(&hwgroup->timer))
822 printk(KERN_CRIT "ide_set_handler: timer already active\n");
824 /* so that ide_timer_expiry knows what to do */
825 hwgroup->sleeping = 1;
826 hwgroup->req_gen_timer = hwgroup->req_gen;
827 mod_timer(&hwgroup->timer, sleep);
828 /* we purposely leave hwgroup->busy==1
831 /* Ugly, but how can we sleep for the lock
832 * otherwise? perhaps from tq_disk?
840 /* no more work for this hwgroup (for now) */
844 if (drive != orig_drive)
849 if (hwif != hwgroup->hwif) {
851 * set nIEN for previous hwif, drives in the
852 * quirk_list may not like intr setups/cleanups
854 if (drive->quirk_list == 0)
855 hwif->tp_ops->set_irq(hwif, 0);
857 hwgroup->hwif = hwif;
858 hwgroup->drive = drive;
859 drive->dev_flags &= ~(IDE_DFLAG_SLEEPING | IDE_DFLAG_PARKED);
860 drive->service_start = jiffies;
863 * we know that the queue isn't empty, but this can happen
864 * if the q->prep_rq_fn() decides to kill a request
866 rq = elv_next_request(drive->queue);
873 * Sanity: don't accept a request that isn't a PM request
874 * if we are currently power managed. This is very important as
875 * blk_stop_queue() doesn't prevent the elv_next_request()
876 * above to return us whatever is in the queue. Since we call
877 * ide_do_request() ourselves, we end up taking requests while
878 * the queue is blocked...
880 * We let requests forced at head of queue with ide-preempt
881 * though. I hope that doesn't happen too much, hopefully not
882 * unless the subdriver triggers such a thing in its own PM
885 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
886 blk_pm_request(rq) == 0 &&
887 (rq->cmd_flags & REQ_PREEMPT) == 0) {
888 /* We clear busy, there should be no pending ATA command at this point. */
895 spin_unlock_irq(&hwgroup->lock);
896 startstop = start_request(drive, rq);
897 spin_lock_irq(&hwgroup->lock);
899 if (startstop == ide_stopped) {
901 if (!elv_queue_empty(orig_drive->queue))
902 blk_plug_device(orig_drive->queue);
908 if (!elv_queue_empty(orig_drive->queue))
909 blk_plug_device(orig_drive->queue);
913 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
914 * retry the current request in pio mode instead of risking tossing it
917 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
919 ide_hwif_t *hwif = HWIF(drive);
921 ide_startstop_t ret = ide_stopped;
924 * end current dma transaction
928 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
929 (void)hwif->dma_ops->dma_end(drive);
930 ret = ide_error(drive, "dma timeout error",
931 hwif->tp_ops->read_status(hwif));
933 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
934 hwif->dma_ops->dma_timeout(drive);
938 * disable dma for now, but remember that we did so because of
939 * a timeout -- we'll reenable after we finish this next request
940 * (or rather the first chunk of it) in pio.
942 drive->dev_flags |= IDE_DFLAG_DMA_PIO_RETRY;
944 ide_dma_off_quietly(drive);
947 * un-busy drive etc (hwgroup->busy is cleared on return) and
948 * make sure request is sane
950 rq = HWGROUP(drive)->rq;
955 HWGROUP(drive)->rq = NULL;
962 rq->sector = rq->bio->bi_sector;
963 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
964 rq->hard_cur_sectors = rq->current_nr_sectors;
965 rq->buffer = bio_data(rq->bio);
971 * ide_timer_expiry - handle lack of an IDE interrupt
972 * @data: timer callback magic (hwgroup)
974 * An IDE command has timed out before the expected drive return
975 * occurred. At this point we attempt to clean up the current
976 * mess. If the current handler includes an expiry handler then
977 * we invoke the expiry handler, and providing it is happy the
978 * work is done. If that fails we apply generic recovery rules
979 * invoking the handler and checking the drive DMA status. We
980 * have an excessively incestuous relationship with the DMA
981 * logic that wants cleaning up.
984 void ide_timer_expiry (unsigned long data)
986 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
987 ide_handler_t *handler;
988 ide_expiry_t *expiry;
990 unsigned long wait = -1;
992 spin_lock_irqsave(&hwgroup->lock, flags);
994 if (((handler = hwgroup->handler) == NULL) ||
995 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
997 * Either a marginal timeout occurred
998 * (got the interrupt just as timer expired),
999 * or we were "sleeping" to give other devices a chance.
1000 * Either way, we don't really want to complain about anything.
1002 if (hwgroup->sleeping) {
1003 hwgroup->sleeping = 0;
1007 ide_drive_t *drive = hwgroup->drive;
1009 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1010 hwgroup->handler = NULL;
1013 ide_startstop_t startstop = ide_stopped;
1015 if ((expiry = hwgroup->expiry) != NULL) {
1017 if ((wait = expiry(drive)) > 0) {
1019 hwgroup->timer.expires = jiffies + wait;
1020 hwgroup->req_gen_timer = hwgroup->req_gen;
1021 add_timer(&hwgroup->timer);
1022 spin_unlock_irqrestore(&hwgroup->lock, flags);
1026 hwgroup->handler = NULL;
1028 * We need to simulate a real interrupt when invoking
1029 * the handler() function, which means we need to
1030 * globally mask the specific IRQ:
1032 spin_unlock(&hwgroup->lock);
1034 /* disable_irq_nosync ?? */
1035 disable_irq(hwif->irq);
1037 * as if we were handling an interrupt */
1038 local_irq_disable();
1039 if (hwgroup->polling) {
1040 startstop = handler(drive);
1041 } else if (drive_is_ready(drive)) {
1042 if (drive->waiting_for_dma)
1043 hwif->dma_ops->dma_lost_irq(drive);
1044 (void)ide_ack_intr(hwif);
1045 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1046 startstop = handler(drive);
1048 if (drive->waiting_for_dma) {
1049 startstop = ide_dma_timeout_retry(drive, wait);
1052 ide_error(drive, "irq timeout",
1053 hwif->tp_ops->read_status(hwif));
1055 drive->service_time = jiffies - drive->service_start;
1056 spin_lock_irq(&hwgroup->lock);
1057 enable_irq(hwif->irq);
1058 if (startstop == ide_stopped) {
1060 if (!elv_queue_empty(drive->queue))
1061 blk_plug_device(drive->queue);
1065 spin_unlock_irqrestore(&hwgroup->lock, flags);
1069 * unexpected_intr - handle an unexpected IDE interrupt
1070 * @irq: interrupt line
1071 * @hwgroup: hwgroup being processed
1073 * There's nothing really useful we can do with an unexpected interrupt,
1074 * other than reading the status register (to clear it), and logging it.
1075 * There should be no way that an irq can happen before we're ready for it,
1076 * so we needn't worry much about losing an "important" interrupt here.
1078 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1079 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1080 * looks "good", we just ignore the interrupt completely.
1082 * This routine assumes __cli() is in effect when called.
1084 * If an unexpected interrupt happens on irq15 while we are handling irq14
1085 * and if the two interfaces are "serialized" (CMD640), then it looks like
1086 * we could screw up by interfering with a new request being set up for
1089 * In reality, this is a non-issue. The new command is not sent unless
1090 * the drive is ready to accept one, in which case we know the drive is
1091 * not trying to interrupt us. And ide_set_handler() is always invoked
1092 * before completing the issuance of any new drive command, so we will not
1093 * be accidentally invoked as a result of any valid command completion
1096 * Note that we must walk the entire hwgroup here. We know which hwif
1097 * is doing the current command, but we don't know which hwif burped
1101 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1104 ide_hwif_t *hwif = hwgroup->hwif;
1107 * handle the unexpected interrupt
1110 if (hwif->irq == irq) {
1111 stat = hwif->tp_ops->read_status(hwif);
1113 if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1114 /* Try to not flood the console with msgs */
1115 static unsigned long last_msgtime, count;
1117 if (time_after(jiffies, last_msgtime + HZ)) {
1118 last_msgtime = jiffies;
1119 printk(KERN_ERR "%s%s: unexpected interrupt, "
1120 "status=0x%02x, count=%ld\n",
1122 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1126 } while ((hwif = hwif->next) != hwgroup->hwif);
1130 * ide_intr - default IDE interrupt handler
1131 * @irq: interrupt number
1132 * @dev_id: hwif group
1133 * @regs: unused weirdness from the kernel irq layer
1135 * This is the default IRQ handler for the IDE layer. You should
1136 * not need to override it. If you do be aware it is subtle in
1139 * hwgroup->hwif is the interface in the group currently performing
1140 * a command. hwgroup->drive is the drive and hwgroup->handler is
1141 * the IRQ handler to call. As we issue a command the handlers
1142 * step through multiple states, reassigning the handler to the
1143 * next step in the process. Unlike a smart SCSI controller IDE
1144 * expects the main processor to sequence the various transfer
1145 * stages. We also manage a poll timer to catch up with most
1146 * timeout situations. There are still a few where the handlers
1147 * don't ever decide to give up.
1149 * The handler eventually returns ide_stopped to indicate the
1150 * request completed. At this point we issue the next request
1151 * on the hwgroup and the process begins again.
1154 irqreturn_t ide_intr (int irq, void *dev_id)
1156 unsigned long flags;
1157 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1158 ide_hwif_t *hwif = hwgroup->hwif;
1160 ide_handler_t *handler;
1161 ide_startstop_t startstop;
1162 irqreturn_t irq_ret = IRQ_NONE;
1164 spin_lock_irqsave(&hwgroup->lock, flags);
1166 if (!ide_ack_intr(hwif))
1169 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1171 * Not expecting an interrupt from this drive.
1172 * That means this could be:
1173 * (1) an interrupt from another PCI device
1174 * sharing the same PCI INT# as us.
1175 * or (2) a drive just entered sleep or standby mode,
1176 * and is interrupting to let us know.
1177 * or (3) a spurious interrupt of unknown origin.
1179 * For PCI, we cannot tell the difference,
1180 * so in that case we just ignore it and hope it goes away.
1182 * FIXME: unexpected_intr should be hwif-> then we can
1183 * remove all the ifdef PCI crap
1185 #ifdef CONFIG_BLK_DEV_IDEPCI
1186 if (hwif->chipset != ide_pci)
1187 #endif /* CONFIG_BLK_DEV_IDEPCI */
1190 * Probably not a shared PCI interrupt,
1191 * so we can safely try to do something about it:
1193 unexpected_intr(irq, hwgroup);
1194 #ifdef CONFIG_BLK_DEV_IDEPCI
1197 * Whack the status register, just in case
1198 * we have a leftover pending IRQ.
1200 (void)hwif->tp_ops->read_status(hwif);
1201 #endif /* CONFIG_BLK_DEV_IDEPCI */
1206 drive = hwgroup->drive;
1209 * This should NEVER happen, and there isn't much
1210 * we could do about it here.
1212 * [Note - this can occur if the drive is hot unplugged]
1217 if (!drive_is_ready(drive))
1219 * This happens regularly when we share a PCI IRQ with
1220 * another device. Unfortunately, it can also happen
1221 * with some buggy drives that trigger the IRQ before
1222 * their status register is up to date. Hopefully we have
1223 * enough advance overhead that the latter isn't a problem.
1227 hwgroup->handler = NULL;
1229 del_timer(&hwgroup->timer);
1230 spin_unlock(&hwgroup->lock);
1232 if (hwif->port_ops && hwif->port_ops->clear_irq)
1233 hwif->port_ops->clear_irq(drive);
1235 if (drive->dev_flags & IDE_DFLAG_UNMASK)
1236 local_irq_enable_in_hardirq();
1238 /* service this interrupt, may set handler for next interrupt */
1239 startstop = handler(drive);
1241 spin_lock_irq(&hwgroup->lock);
1243 * Note that handler() may have set things up for another
1244 * interrupt to occur soon, but it cannot happen until
1245 * we exit from this routine, because it will be the
1246 * same irq as is currently being serviced here, and Linux
1247 * won't allow another of the same (on any CPU) until we return.
1249 drive->service_time = jiffies - drive->service_start;
1250 if (startstop == ide_stopped) {
1251 if (hwgroup->handler == NULL) { /* paranoia */
1253 if (!elv_queue_empty(drive->queue))
1254 blk_plug_device(drive->queue);
1256 printk(KERN_ERR "%s: %s: huh? expected NULL handler "
1257 "on exit\n", __func__, drive->name);
1260 irq_ret = IRQ_HANDLED;
1262 spin_unlock_irqrestore(&hwgroup->lock, flags);
1267 * ide_do_drive_cmd - issue IDE special command
1268 * @drive: device to issue command
1269 * @rq: request to issue
1271 * This function issues a special IDE device request
1272 * onto the request queue.
1274 * the rq is queued at the head of the request queue, displacing
1275 * the currently-being-processed request and this function
1276 * returns immediately without waiting for the new rq to be
1277 * completed. This is VERY DANGEROUS, and is intended for
1278 * careful use by the ATAPI tape/cdrom driver code.
1281 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1283 ide_hwgroup_t *hwgroup = drive->hwif->hwgroup;
1284 struct request_queue *q = drive->queue;
1285 unsigned long flags;
1289 spin_lock_irqsave(q->queue_lock, flags);
1290 __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0);
1291 blk_start_queueing(q);
1292 spin_unlock_irqrestore(q->queue_lock, flags);
1294 EXPORT_SYMBOL(ide_do_drive_cmd);
1296 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1298 ide_hwif_t *hwif = drive->hwif;
1301 memset(&task, 0, sizeof(task));
1302 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1303 IDE_TFLAG_OUT_FEATURE | tf_flags;
1304 task.tf.feature = dma; /* Use PIO/DMA */
1305 task.tf.lbam = bcount & 0xff;
1306 task.tf.lbah = (bcount >> 8) & 0xff;
1308 ide_tf_dump(drive->name, &task.tf);
1309 hwif->tp_ops->set_irq(hwif, 1);
1310 SELECT_MASK(drive, 0);
1311 hwif->tp_ops->tf_load(drive, &task);
1314 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1316 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1318 ide_hwif_t *hwif = drive->hwif;
1323 hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1325 hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1329 EXPORT_SYMBOL_GPL(ide_pad_transfer);