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)) {
89 HWGROUP(drive)->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;
115 * room for locking improvements here, the calls below don't
116 * need the queue lock held at all
118 spin_lock_irqsave(&ide_lock, flags);
119 rq = HWGROUP(drive)->rq;
122 if (blk_pc_request(rq))
123 nr_bytes = rq->data_len;
125 nr_bytes = rq->hard_cur_sectors << 9;
128 ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
130 spin_unlock_irqrestore(&ide_lock, flags);
133 EXPORT_SYMBOL(ide_end_request);
135 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
137 struct request_pm_state *pm = rq->data;
139 if (drive->media != ide_disk)
142 switch (pm->pm_step) {
143 case IDE_PM_FLUSH_CACHE: /* Suspend step 1 (flush cache) */
144 if (pm->pm_state == PM_EVENT_FREEZE)
145 pm->pm_step = IDE_PM_COMPLETED;
147 pm->pm_step = IDE_PM_STANDBY;
149 case IDE_PM_STANDBY: /* Suspend step 2 (standby) */
150 pm->pm_step = IDE_PM_COMPLETED;
152 case IDE_PM_RESTORE_PIO: /* Resume step 1 (restore PIO) */
153 pm->pm_step = IDE_PM_IDLE;
155 case IDE_PM_IDLE: /* Resume step 2 (idle)*/
156 pm->pm_step = IDE_PM_RESTORE_DMA;
161 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
163 struct request_pm_state *pm = rq->data;
164 ide_task_t *args = rq->special;
166 memset(args, 0, sizeof(*args));
168 switch (pm->pm_step) {
169 case IDE_PM_FLUSH_CACHE: /* Suspend step 1 (flush cache) */
170 if (drive->media != ide_disk)
172 /* Not supported? Switch to next step now. */
173 if (ata_id_flush_enabled(drive->id) == 0 ||
174 (drive->dev_flags & IDE_DFLAG_WCACHE) == 0) {
175 ide_complete_power_step(drive, rq, 0, 0);
178 if (ata_id_flush_ext_enabled(drive->id))
179 args->tf.command = ATA_CMD_FLUSH_EXT;
181 args->tf.command = ATA_CMD_FLUSH;
183 case IDE_PM_STANDBY: /* Suspend step 2 (standby) */
184 args->tf.command = ATA_CMD_STANDBYNOW1;
186 case IDE_PM_RESTORE_PIO: /* Resume step 1 (restore PIO) */
187 ide_set_max_pio(drive);
189 * skip IDE_PM_IDLE for ATAPI devices
191 if (drive->media != ide_disk)
192 pm->pm_step = IDE_PM_RESTORE_DMA;
194 ide_complete_power_step(drive, rq, 0, 0);
196 case IDE_PM_IDLE: /* Resume step 2 (idle) */
197 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
199 case IDE_PM_RESTORE_DMA: /* Resume step 3 (restore DMA) */
201 * Right now, all we do is call ide_set_dma(drive),
202 * we could be smarter and check for current xfer_speed
203 * in struct drive etc...
205 if (drive->hwif->dma_ops == NULL)
207 if (drive->dev_flags & IDE_DFLAG_USING_DMA)
212 pm->pm_step = IDE_PM_COMPLETED;
216 args->tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
217 args->data_phase = TASKFILE_NO_DATA;
218 return do_rw_taskfile(drive, args);
222 * ide_end_dequeued_request - complete an IDE I/O
223 * @drive: IDE device for the I/O
225 * @nr_sectors: number of sectors completed
227 * Complete an I/O that is no longer on the request queue. This
228 * typically occurs when we pull the request and issue a REQUEST_SENSE.
229 * We must still finish the old request but we must not tamper with the
230 * queue in the meantime.
232 * NOTE: This path does not handle barrier, but barrier is not supported
236 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
237 int uptodate, int nr_sectors)
242 spin_lock_irqsave(&ide_lock, flags);
243 BUG_ON(!blk_rq_started(rq));
244 ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
245 spin_unlock_irqrestore(&ide_lock, flags);
249 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
253 * ide_complete_pm_request - end the current Power Management request
254 * @drive: target drive
257 * This function cleans up the current PM request and stops the queue
260 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
265 printk("%s: completing PM request, %s\n", drive->name,
266 blk_pm_suspend_request(rq) ? "suspend" : "resume");
268 spin_lock_irqsave(&ide_lock, flags);
269 if (blk_pm_suspend_request(rq)) {
270 blk_stop_queue(drive->queue);
272 drive->dev_flags &= ~IDE_DFLAG_BLOCKED;
273 blk_start_queue(drive->queue);
275 HWGROUP(drive)->rq = NULL;
276 if (__blk_end_request(rq, 0, 0))
278 spin_unlock_irqrestore(&ide_lock, flags);
282 * ide_end_drive_cmd - end an explicit drive command
287 * Clean up after success/failure of an explicit drive command.
288 * These get thrown onto the queue so they are synchronized with
289 * real I/O operations on the drive.
291 * In LBA48 mode we have to read the register set twice to get
292 * all the extra information out.
295 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
300 spin_lock_irqsave(&ide_lock, flags);
301 rq = HWGROUP(drive)->rq;
302 spin_unlock_irqrestore(&ide_lock, flags);
304 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
305 ide_task_t *task = (ide_task_t *)rq->special;
308 rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
311 struct ide_taskfile *tf = &task->tf;
316 drive->hwif->tp_ops->tf_read(drive, task);
318 if (task->tf_flags & IDE_TFLAG_DYN)
321 } else if (blk_pm_request(rq)) {
322 struct request_pm_state *pm = rq->data;
324 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
325 drive->name, rq->pm->pm_step, stat, err);
327 ide_complete_power_step(drive, rq, stat, err);
328 if (pm->pm_step == IDE_PM_COMPLETED)
329 ide_complete_pm_request(drive, rq);
333 spin_lock_irqsave(&ide_lock, flags);
334 HWGROUP(drive)->rq = NULL;
336 if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
339 spin_unlock_irqrestore(&ide_lock, flags);
342 EXPORT_SYMBOL(ide_end_drive_cmd);
344 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
349 drv = *(ide_driver_t **)rq->rq_disk->private_data;
350 drv->end_request(drive, 0, 0);
352 ide_end_request(drive, 0, 0);
355 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
357 ide_hwif_t *hwif = drive->hwif;
359 if ((stat & ATA_BUSY) ||
360 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
361 /* other bits are useless when BUSY */
362 rq->errors |= ERROR_RESET;
363 } else if (stat & ATA_ERR) {
364 /* err has different meaning on cdrom and tape */
365 if (err == ATA_ABORTED) {
366 if ((drive->dev_flags & IDE_DFLAG_LBA) &&
367 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
368 hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
370 } else if ((err & BAD_CRC) == BAD_CRC) {
371 /* UDMA crc error, just retry the operation */
373 } else if (err & (ATA_BBK | ATA_UNC)) {
374 /* retries won't help these */
375 rq->errors = ERROR_MAX;
376 } else if (err & ATA_TRK0NF) {
377 /* help it find track zero */
378 rq->errors |= ERROR_RECAL;
382 if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
383 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
384 int nsect = drive->mult_count ? drive->mult_count : 1;
386 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
389 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
390 ide_kill_rq(drive, rq);
394 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
395 rq->errors |= ERROR_RESET;
397 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
399 return ide_do_reset(drive);
402 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
403 drive->special.b.recalibrate = 1;
410 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
412 ide_hwif_t *hwif = drive->hwif;
414 if ((stat & ATA_BUSY) ||
415 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
416 /* other bits are useless when BUSY */
417 rq->errors |= ERROR_RESET;
419 /* add decoding error stuff */
422 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
424 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
426 if (rq->errors >= ERROR_MAX) {
427 ide_kill_rq(drive, rq);
429 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
431 return ide_do_reset(drive);
440 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
442 if (drive->media == ide_disk)
443 return ide_ata_error(drive, rq, stat, err);
444 return ide_atapi_error(drive, rq, stat, err);
447 EXPORT_SYMBOL_GPL(__ide_error);
450 * ide_error - handle an error on the IDE
451 * @drive: drive the error occurred on
452 * @msg: message to report
455 * ide_error() takes action based on the error returned by the drive.
456 * For normal I/O that may well include retries. We deal with
457 * both new-style (taskfile) and old style command handling here.
458 * In the case of taskfile command handling there is work left to
462 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
467 err = ide_dump_status(drive, msg, stat);
469 if ((rq = HWGROUP(drive)->rq) == NULL)
472 /* retry only "normal" I/O: */
473 if (!blk_fs_request(rq)) {
475 ide_end_drive_cmd(drive, stat, err);
482 drv = *(ide_driver_t **)rq->rq_disk->private_data;
483 return drv->error(drive, rq, stat, err);
485 return __ide_error(drive, rq, stat, err);
488 EXPORT_SYMBOL_GPL(ide_error);
490 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
492 tf->nsect = drive->sect;
493 tf->lbal = drive->sect;
494 tf->lbam = drive->cyl;
495 tf->lbah = drive->cyl >> 8;
496 tf->device = (drive->head - 1) | drive->select;
497 tf->command = ATA_CMD_INIT_DEV_PARAMS;
500 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
502 tf->nsect = drive->sect;
503 tf->command = ATA_CMD_RESTORE;
506 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
508 tf->nsect = drive->mult_req;
509 tf->command = ATA_CMD_SET_MULTI;
512 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
514 special_t *s = &drive->special;
517 memset(&args, 0, sizeof(ide_task_t));
518 args.data_phase = TASKFILE_NO_DATA;
520 if (s->b.set_geometry) {
521 s->b.set_geometry = 0;
522 ide_tf_set_specify_cmd(drive, &args.tf);
523 } else if (s->b.recalibrate) {
524 s->b.recalibrate = 0;
525 ide_tf_set_restore_cmd(drive, &args.tf);
526 } else if (s->b.set_multmode) {
527 s->b.set_multmode = 0;
528 ide_tf_set_setmult_cmd(drive, &args.tf);
530 int special = s->all;
532 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
536 args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
537 IDE_TFLAG_CUSTOM_HANDLER;
539 do_rw_taskfile(drive, &args);
545 * do_special - issue some special commands
546 * @drive: drive the command is for
548 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
549 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
551 * It used to do much more, but has been scaled back.
554 static ide_startstop_t do_special (ide_drive_t *drive)
556 special_t *s = &drive->special;
559 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
561 if (drive->media == ide_disk)
562 return ide_disk_special(drive);
569 void ide_map_sg(ide_drive_t *drive, struct request *rq)
571 ide_hwif_t *hwif = drive->hwif;
572 struct scatterlist *sg = hwif->sg_table;
574 if (hwif->sg_mapped) /* needed by ide-scsi */
577 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
578 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
580 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
585 EXPORT_SYMBOL_GPL(ide_map_sg);
587 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
589 ide_hwif_t *hwif = drive->hwif;
591 hwif->nsect = hwif->nleft = rq->nr_sectors;
596 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
599 * execute_drive_command - issue special drive command
600 * @drive: the drive to issue the command on
601 * @rq: the request structure holding the command
603 * execute_drive_cmd() issues a special drive command, usually
604 * initiated by ioctl() from the external hdparm program. The
605 * command can be a drive command, drive task or taskfile
606 * operation. Weirdly you can call it with NULL to wait for
607 * all commands to finish. Don't do this as that is due to change
610 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
613 ide_hwif_t *hwif = HWIF(drive);
614 ide_task_t *task = rq->special;
617 hwif->data_phase = task->data_phase;
619 switch (hwif->data_phase) {
620 case TASKFILE_MULTI_OUT:
622 case TASKFILE_MULTI_IN:
624 ide_init_sg_cmd(drive, rq);
625 ide_map_sg(drive, rq);
630 return do_rw_taskfile(drive, task);
634 * NULL is actually a valid way of waiting for
635 * all current requests to be flushed from the queue.
638 printk("%s: DRIVE_CMD (null)\n", drive->name);
640 ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
641 ide_read_error(drive));
646 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
649 struct request_queue *q = drive->queue;
653 if (!(setting->flags & DS_SYNC))
654 return setting->set(drive, arg);
656 rq = blk_get_request(q, READ, __GFP_WAIT);
657 rq->cmd_type = REQ_TYPE_SPECIAL;
659 rq->cmd[0] = REQ_DEVSET_EXEC;
660 *(int *)&rq->cmd[1] = arg;
661 rq->special = setting->set;
663 if (blk_execute_rq(q, NULL, rq, 0))
669 EXPORT_SYMBOL_GPL(ide_devset_execute);
671 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
675 if (cmd == REQ_PARK_HEADS || cmd == REQ_UNPARK_HEADS) {
677 struct ide_taskfile *tf = &task.tf;
679 memset(&task, 0, sizeof(task));
680 if (cmd == REQ_PARK_HEADS) {
681 drive->sleep = *(unsigned long *)rq->special;
682 drive->dev_flags |= IDE_DFLAG_SLEEPING;
683 tf->command = ATA_CMD_IDLEIMMEDIATE;
688 task.tf_flags |= IDE_TFLAG_CUSTOM_HANDLER;
689 } else /* cmd == REQ_UNPARK_HEADS */
690 tf->command = ATA_CMD_CHK_POWER;
692 task.tf_flags |= IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
694 drive->hwif->data_phase = task.data_phase = TASKFILE_NO_DATA;
695 return do_rw_taskfile(drive, &task);
699 case REQ_DEVSET_EXEC:
701 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
703 err = setfunc(drive, *(int *)&rq->cmd[1]);
708 ide_end_request(drive, err, 0);
711 case REQ_DRIVE_RESET:
712 return ide_do_reset(drive);
714 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
715 ide_end_request(drive, 0, 0);
720 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
722 struct request_pm_state *pm = rq->data;
724 if (blk_pm_suspend_request(rq) &&
725 pm->pm_step == IDE_PM_START_SUSPEND)
726 /* Mark drive blocked when starting the suspend sequence. */
727 drive->dev_flags |= IDE_DFLAG_BLOCKED;
728 else if (blk_pm_resume_request(rq) &&
729 pm->pm_step == IDE_PM_START_RESUME) {
731 * The first thing we do on wakeup is to wait for BSY bit to
732 * go away (with a looong timeout) as a drive on this hwif may
733 * just be POSTing itself.
734 * We do that before even selecting as the "other" device on
735 * the bus may be broken enough to walk on our toes at this
738 ide_hwif_t *hwif = drive->hwif;
741 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
743 rc = ide_wait_not_busy(hwif, 35000);
745 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
747 hwif->tp_ops->set_irq(hwif, 1);
748 rc = ide_wait_not_busy(hwif, 100000);
750 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
755 * start_request - start of I/O and command issuing for IDE
757 * start_request() initiates handling of a new I/O request. It
758 * accepts commands and I/O (read/write) requests.
760 * FIXME: this function needs a rename
763 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
765 ide_startstop_t startstop;
767 BUG_ON(!blk_rq_started(rq));
770 printk("%s: start_request: current=0x%08lx\n",
771 HWIF(drive)->name, (unsigned long) rq);
774 /* bail early if we've exceeded max_failures */
775 if (drive->max_failures && (drive->failures > drive->max_failures)) {
776 rq->cmd_flags |= REQ_FAILED;
780 if (blk_pm_request(rq))
781 ide_check_pm_state(drive, rq);
784 if (ide_wait_stat(&startstop, drive, drive->ready_stat,
785 ATA_BUSY | ATA_DRQ, WAIT_READY)) {
786 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
789 if (!drive->special.all) {
793 * We reset the drive so we need to issue a SETFEATURES.
794 * Do it _after_ do_special() restored device parameters.
796 if (drive->current_speed == 0xff)
797 ide_config_drive_speed(drive, drive->desired_speed);
799 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
800 return execute_drive_cmd(drive, rq);
801 else if (blk_pm_request(rq)) {
802 struct request_pm_state *pm = rq->data;
804 printk("%s: start_power_step(step: %d)\n",
805 drive->name, rq->pm->pm_step);
807 startstop = ide_start_power_step(drive, rq);
808 if (startstop == ide_stopped &&
809 pm->pm_step == IDE_PM_COMPLETED)
810 ide_complete_pm_request(drive, rq);
812 } else if (!rq->rq_disk && blk_special_request(rq))
814 * TODO: Once all ULDs have been modified to
815 * check for specific op codes rather than
816 * blindly accepting any special request, the
817 * check for ->rq_disk above may be replaced
818 * by a more suitable mechanism or even
821 return ide_special_rq(drive, rq);
823 drv = *(ide_driver_t **)rq->rq_disk->private_data;
825 return drv->do_request(drive, rq, rq->sector);
827 return do_special(drive);
829 ide_kill_rq(drive, rq);
834 * ide_stall_queue - pause an IDE device
835 * @drive: drive to stall
836 * @timeout: time to stall for (jiffies)
838 * ide_stall_queue() can be used by a drive to give excess bandwidth back
839 * to the hwgroup by sleeping for timeout jiffies.
842 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
844 if (timeout > WAIT_WORSTCASE)
845 timeout = WAIT_WORSTCASE;
846 drive->sleep = timeout + jiffies;
847 drive->dev_flags |= IDE_DFLAG_SLEEPING;
850 EXPORT_SYMBOL(ide_stall_queue);
852 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
855 * choose_drive - select a drive to service
856 * @hwgroup: hardware group to select on
858 * choose_drive() selects the next drive which will be serviced.
859 * This is necessary because the IDE layer can't issue commands
860 * to both drives on the same cable, unlike SCSI.
863 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
865 ide_drive_t *drive, *best;
869 drive = hwgroup->drive;
872 * drive is doing pre-flush, ordered write, post-flush sequence. even
873 * though that is 3 requests, it must be seen as a single transaction.
874 * we must not preempt this drive until that is complete
876 if (blk_queue_flushing(drive->queue)) {
878 * small race where queue could get replugged during
879 * the 3-request flush cycle, just yank the plug since
880 * we want it to finish asap
882 blk_remove_plug(drive->queue);
887 u8 dev_s = !!(drive->dev_flags & IDE_DFLAG_SLEEPING);
888 u8 best_s = (best && !!(best->dev_flags & IDE_DFLAG_SLEEPING));
890 if ((dev_s == 0 || time_after_eq(jiffies, drive->sleep)) &&
891 !elv_queue_empty(drive->queue)) {
893 (dev_s && (best_s == 0 || time_before(drive->sleep, best->sleep))) ||
894 (best_s == 0 && time_before(WAKEUP(drive), WAKEUP(best)))) {
895 if (!blk_queue_plugged(drive->queue))
899 } while ((drive = drive->next) != hwgroup->drive);
901 if (best && (best->dev_flags & IDE_DFLAG_NICE1) &&
902 (best->dev_flags & IDE_DFLAG_SLEEPING) == 0 &&
903 best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
904 long t = (signed long)(WAKEUP(best) - jiffies);
905 if (t >= WAIT_MIN_SLEEP) {
907 * We *may* have some time to spare, but first let's see if
908 * someone can potentially benefit from our nice mood today..
912 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) == 0
913 && time_before(jiffies - best->service_time, WAKEUP(drive))
914 && time_before(WAKEUP(drive), jiffies + t))
916 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
919 } while ((drive = drive->next) != best);
926 * Issue a new request to a drive from hwgroup
927 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
929 * A hwgroup is a serialized group of IDE interfaces. Usually there is
930 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
931 * may have both interfaces in a single hwgroup to "serialize" access.
932 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
933 * together into one hwgroup for serialized access.
935 * Note also that several hwgroups can end up sharing a single IRQ,
936 * possibly along with many other devices. This is especially common in
937 * PCI-based systems with off-board IDE controller cards.
939 * The IDE driver uses the single global ide_lock spinlock to protect
940 * access to the request queues, and to protect the hwgroup->busy flag.
942 * The first thread into the driver for a particular hwgroup sets the
943 * hwgroup->busy flag to indicate that this hwgroup is now active,
944 * and then initiates processing of the top request from the request queue.
946 * Other threads attempting entry notice the busy setting, and will simply
947 * queue their new requests and exit immediately. Note that hwgroup->busy
948 * remains set even when the driver is merely awaiting the next interrupt.
949 * Thus, the meaning is "this hwgroup is busy processing a request".
951 * When processing of a request completes, the completing thread or IRQ-handler
952 * will start the next request from the queue. If no more work remains,
953 * the driver will clear the hwgroup->busy flag and exit.
955 * The ide_lock (spinlock) is used to protect all access to the
956 * hwgroup->busy flag, but is otherwise not needed for most processing in
957 * the driver. This makes the driver much more friendlier to shared IRQs
958 * than previous designs, while remaining 100% (?) SMP safe and capable.
960 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
965 ide_startstop_t startstop;
968 /* caller must own ide_lock */
969 BUG_ON(!irqs_disabled());
971 while (!hwgroup->busy) {
974 ide_get_lock(ide_intr, hwgroup);
975 drive = choose_drive(hwgroup);
978 unsigned long sleep = 0; /* shut up, gcc */
980 drive = hwgroup->drive;
982 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) &&
984 time_before(drive->sleep, sleep))) {
986 sleep = drive->sleep;
988 } while ((drive = drive->next) != hwgroup->drive);
991 * Take a short snooze, and then wake up this hwgroup again.
992 * This gives other hwgroups on the same a chance to
993 * play fairly with us, just in case there are big differences
994 * in relative throughputs.. don't want to hog the cpu too much.
996 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
997 sleep = jiffies + WAIT_MIN_SLEEP;
999 if (timer_pending(&hwgroup->timer))
1000 printk(KERN_CRIT "ide_set_handler: timer already active\n");
1002 /* so that ide_timer_expiry knows what to do */
1003 hwgroup->sleeping = 1;
1004 hwgroup->req_gen_timer = hwgroup->req_gen;
1005 mod_timer(&hwgroup->timer, sleep);
1006 /* we purposely leave hwgroup->busy==1
1009 /* Ugly, but how can we sleep for the lock
1010 * otherwise? perhaps from tq_disk?
1013 /* for atari only */
1018 /* no more work for this hwgroup (for now) */
1023 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1025 * set nIEN for previous hwif, drives in the
1026 * quirk_list may not like intr setups/cleanups
1028 if (drive->quirk_list != 1)
1029 hwif->tp_ops->set_irq(hwif, 0);
1031 hwgroup->hwif = hwif;
1032 hwgroup->drive = drive;
1033 drive->dev_flags &= ~(IDE_DFLAG_SLEEPING | IDE_DFLAG_PARKED);
1034 drive->service_start = jiffies;
1036 if (blk_queue_plugged(drive->queue)) {
1037 printk(KERN_ERR "ide: huh? queue was plugged!\n");
1042 * we know that the queue isn't empty, but this can happen
1043 * if the q->prep_rq_fn() decides to kill a request
1045 rq = elv_next_request(drive->queue);
1052 * Sanity: don't accept a request that isn't a PM request
1053 * if we are currently power managed. This is very important as
1054 * blk_stop_queue() doesn't prevent the elv_next_request()
1055 * above to return us whatever is in the queue. Since we call
1056 * ide_do_request() ourselves, we end up taking requests while
1057 * the queue is blocked...
1059 * We let requests forced at head of queue with ide-preempt
1060 * though. I hope that doesn't happen too much, hopefully not
1061 * unless the subdriver triggers such a thing in its own PM
1064 * We count how many times we loop here to make sure we service
1065 * all drives in the hwgroup without looping for ever
1067 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
1068 blk_pm_request(rq) == 0 &&
1069 (rq->cmd_flags & REQ_PREEMPT) == 0) {
1070 drive = drive->next ? drive->next : hwgroup->drive;
1071 if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1073 /* We clear busy, there should be no pending ATA command at this point. */
1081 * Some systems have trouble with IDE IRQs arriving while
1082 * the driver is still setting things up. So, here we disable
1083 * the IRQ used by this interface while the request is being started.
1084 * This may look bad at first, but pretty much the same thing
1085 * happens anyway when any interrupt comes in, IDE or otherwise
1086 * -- the kernel masks the IRQ while it is being handled.
1088 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1089 disable_irq_nosync(hwif->irq);
1090 spin_unlock(&ide_lock);
1091 local_irq_enable_in_hardirq();
1092 /* allow other IRQs while we start this request */
1093 startstop = start_request(drive, rq);
1094 spin_lock_irq(&ide_lock);
1095 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1096 enable_irq(hwif->irq);
1097 if (startstop == ide_stopped)
1103 * Passes the stuff to ide_do_request
1105 void do_ide_request(struct request_queue *q)
1107 ide_drive_t *drive = q->queuedata;
1109 ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1113 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1114 * retry the current request in pio mode instead of risking tossing it
1117 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1119 ide_hwif_t *hwif = HWIF(drive);
1121 ide_startstop_t ret = ide_stopped;
1124 * end current dma transaction
1128 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1129 (void)hwif->dma_ops->dma_end(drive);
1130 ret = ide_error(drive, "dma timeout error",
1131 hwif->tp_ops->read_status(hwif));
1133 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1134 hwif->dma_ops->dma_timeout(drive);
1138 * disable dma for now, but remember that we did so because of
1139 * a timeout -- we'll reenable after we finish this next request
1140 * (or rather the first chunk of it) in pio.
1142 drive->dev_flags |= IDE_DFLAG_DMA_PIO_RETRY;
1144 ide_dma_off_quietly(drive);
1147 * un-busy drive etc (hwgroup->busy is cleared on return) and
1148 * make sure request is sane
1150 rq = HWGROUP(drive)->rq;
1155 HWGROUP(drive)->rq = NULL;
1162 rq->sector = rq->bio->bi_sector;
1163 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1164 rq->hard_cur_sectors = rq->current_nr_sectors;
1165 rq->buffer = bio_data(rq->bio);
1171 * ide_timer_expiry - handle lack of an IDE interrupt
1172 * @data: timer callback magic (hwgroup)
1174 * An IDE command has timed out before the expected drive return
1175 * occurred. At this point we attempt to clean up the current
1176 * mess. If the current handler includes an expiry handler then
1177 * we invoke the expiry handler, and providing it is happy the
1178 * work is done. If that fails we apply generic recovery rules
1179 * invoking the handler and checking the drive DMA status. We
1180 * have an excessively incestuous relationship with the DMA
1181 * logic that wants cleaning up.
1184 void ide_timer_expiry (unsigned long data)
1186 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1187 ide_handler_t *handler;
1188 ide_expiry_t *expiry;
1189 unsigned long flags;
1190 unsigned long wait = -1;
1192 spin_lock_irqsave(&ide_lock, flags);
1194 if (((handler = hwgroup->handler) == NULL) ||
1195 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1197 * Either a marginal timeout occurred
1198 * (got the interrupt just as timer expired),
1199 * or we were "sleeping" to give other devices a chance.
1200 * Either way, we don't really want to complain about anything.
1202 if (hwgroup->sleeping) {
1203 hwgroup->sleeping = 0;
1207 ide_drive_t *drive = hwgroup->drive;
1209 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1210 hwgroup->handler = NULL;
1213 ide_startstop_t startstop = ide_stopped;
1214 if (!hwgroup->busy) {
1215 hwgroup->busy = 1; /* paranoia */
1216 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1218 if ((expiry = hwgroup->expiry) != NULL) {
1220 if ((wait = expiry(drive)) > 0) {
1222 hwgroup->timer.expires = jiffies + wait;
1223 hwgroup->req_gen_timer = hwgroup->req_gen;
1224 add_timer(&hwgroup->timer);
1225 spin_unlock_irqrestore(&ide_lock, flags);
1229 hwgroup->handler = NULL;
1231 * We need to simulate a real interrupt when invoking
1232 * the handler() function, which means we need to
1233 * globally mask the specific IRQ:
1235 spin_unlock(&ide_lock);
1237 /* disable_irq_nosync ?? */
1238 disable_irq(hwif->irq);
1240 * as if we were handling an interrupt */
1241 local_irq_disable();
1242 if (hwgroup->polling) {
1243 startstop = handler(drive);
1244 } else if (drive_is_ready(drive)) {
1245 if (drive->waiting_for_dma)
1246 hwif->dma_ops->dma_lost_irq(drive);
1247 (void)ide_ack_intr(hwif);
1248 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1249 startstop = handler(drive);
1251 if (drive->waiting_for_dma) {
1252 startstop = ide_dma_timeout_retry(drive, wait);
1255 ide_error(drive, "irq timeout",
1256 hwif->tp_ops->read_status(hwif));
1258 drive->service_time = jiffies - drive->service_start;
1259 spin_lock_irq(&ide_lock);
1260 enable_irq(hwif->irq);
1261 if (startstop == ide_stopped)
1265 ide_do_request(hwgroup, IDE_NO_IRQ);
1266 spin_unlock_irqrestore(&ide_lock, flags);
1270 * unexpected_intr - handle an unexpected IDE interrupt
1271 * @irq: interrupt line
1272 * @hwgroup: hwgroup being processed
1274 * There's nothing really useful we can do with an unexpected interrupt,
1275 * other than reading the status register (to clear it), and logging it.
1276 * There should be no way that an irq can happen before we're ready for it,
1277 * so we needn't worry much about losing an "important" interrupt here.
1279 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1280 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1281 * looks "good", we just ignore the interrupt completely.
1283 * This routine assumes __cli() is in effect when called.
1285 * If an unexpected interrupt happens on irq15 while we are handling irq14
1286 * and if the two interfaces are "serialized" (CMD640), then it looks like
1287 * we could screw up by interfering with a new request being set up for
1290 * In reality, this is a non-issue. The new command is not sent unless
1291 * the drive is ready to accept one, in which case we know the drive is
1292 * not trying to interrupt us. And ide_set_handler() is always invoked
1293 * before completing the issuance of any new drive command, so we will not
1294 * be accidentally invoked as a result of any valid command completion
1297 * Note that we must walk the entire hwgroup here. We know which hwif
1298 * is doing the current command, but we don't know which hwif burped
1302 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1305 ide_hwif_t *hwif = hwgroup->hwif;
1308 * handle the unexpected interrupt
1311 if (hwif->irq == irq) {
1312 stat = hwif->tp_ops->read_status(hwif);
1314 if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1315 /* Try to not flood the console with msgs */
1316 static unsigned long last_msgtime, count;
1318 if (time_after(jiffies, last_msgtime + HZ)) {
1319 last_msgtime = jiffies;
1320 printk(KERN_ERR "%s%s: unexpected interrupt, "
1321 "status=0x%02x, count=%ld\n",
1323 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1327 } while ((hwif = hwif->next) != hwgroup->hwif);
1331 * ide_intr - default IDE interrupt handler
1332 * @irq: interrupt number
1333 * @dev_id: hwif group
1334 * @regs: unused weirdness from the kernel irq layer
1336 * This is the default IRQ handler for the IDE layer. You should
1337 * not need to override it. If you do be aware it is subtle in
1340 * hwgroup->hwif is the interface in the group currently performing
1341 * a command. hwgroup->drive is the drive and hwgroup->handler is
1342 * the IRQ handler to call. As we issue a command the handlers
1343 * step through multiple states, reassigning the handler to the
1344 * next step in the process. Unlike a smart SCSI controller IDE
1345 * expects the main processor to sequence the various transfer
1346 * stages. We also manage a poll timer to catch up with most
1347 * timeout situations. There are still a few where the handlers
1348 * don't ever decide to give up.
1350 * The handler eventually returns ide_stopped to indicate the
1351 * request completed. At this point we issue the next request
1352 * on the hwgroup and the process begins again.
1355 irqreturn_t ide_intr (int irq, void *dev_id)
1357 unsigned long flags;
1358 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1361 ide_handler_t *handler;
1362 ide_startstop_t startstop;
1364 spin_lock_irqsave(&ide_lock, flags);
1365 hwif = hwgroup->hwif;
1367 if (!ide_ack_intr(hwif)) {
1368 spin_unlock_irqrestore(&ide_lock, flags);
1372 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1374 * Not expecting an interrupt from this drive.
1375 * That means this could be:
1376 * (1) an interrupt from another PCI device
1377 * sharing the same PCI INT# as us.
1378 * or (2) a drive just entered sleep or standby mode,
1379 * and is interrupting to let us know.
1380 * or (3) a spurious interrupt of unknown origin.
1382 * For PCI, we cannot tell the difference,
1383 * so in that case we just ignore it and hope it goes away.
1385 * FIXME: unexpected_intr should be hwif-> then we can
1386 * remove all the ifdef PCI crap
1388 #ifdef CONFIG_BLK_DEV_IDEPCI
1389 if (hwif->chipset != ide_pci)
1390 #endif /* CONFIG_BLK_DEV_IDEPCI */
1393 * Probably not a shared PCI interrupt,
1394 * so we can safely try to do something about it:
1396 unexpected_intr(irq, hwgroup);
1397 #ifdef CONFIG_BLK_DEV_IDEPCI
1400 * Whack the status register, just in case
1401 * we have a leftover pending IRQ.
1403 (void)hwif->tp_ops->read_status(hwif);
1404 #endif /* CONFIG_BLK_DEV_IDEPCI */
1406 spin_unlock_irqrestore(&ide_lock, flags);
1409 drive = hwgroup->drive;
1412 * This should NEVER happen, and there isn't much
1413 * we could do about it here.
1415 * [Note - this can occur if the drive is hot unplugged]
1417 spin_unlock_irqrestore(&ide_lock, flags);
1420 if (!drive_is_ready(drive)) {
1422 * This happens regularly when we share a PCI IRQ with
1423 * another device. Unfortunately, it can also happen
1424 * with some buggy drives that trigger the IRQ before
1425 * their status register is up to date. Hopefully we have
1426 * enough advance overhead that the latter isn't a problem.
1428 spin_unlock_irqrestore(&ide_lock, flags);
1431 if (!hwgroup->busy) {
1432 hwgroup->busy = 1; /* paranoia */
1433 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1435 hwgroup->handler = NULL;
1437 del_timer(&hwgroup->timer);
1438 spin_unlock(&ide_lock);
1440 if (hwif->port_ops && hwif->port_ops->clear_irq)
1441 hwif->port_ops->clear_irq(drive);
1443 if (drive->dev_flags & IDE_DFLAG_UNMASK)
1444 local_irq_enable_in_hardirq();
1446 /* service this interrupt, may set handler for next interrupt */
1447 startstop = handler(drive);
1449 spin_lock_irq(&ide_lock);
1451 * Note that handler() may have set things up for another
1452 * interrupt to occur soon, but it cannot happen until
1453 * we exit from this routine, because it will be the
1454 * same irq as is currently being serviced here, and Linux
1455 * won't allow another of the same (on any CPU) until we return.
1457 drive->service_time = jiffies - drive->service_start;
1458 if (startstop == ide_stopped) {
1459 if (hwgroup->handler == NULL) { /* paranoia */
1461 ide_do_request(hwgroup, hwif->irq);
1463 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1464 "on exit\n", drive->name);
1467 spin_unlock_irqrestore(&ide_lock, flags);
1472 * ide_do_drive_cmd - issue IDE special command
1473 * @drive: device to issue command
1474 * @rq: request to issue
1476 * This function issues a special IDE device request
1477 * onto the request queue.
1479 * the rq is queued at the head of the request queue, displacing
1480 * the currently-being-processed request and this function
1481 * returns immediately without waiting for the new rq to be
1482 * completed. This is VERY DANGEROUS, and is intended for
1483 * careful use by the ATAPI tape/cdrom driver code.
1486 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1488 unsigned long flags;
1489 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1491 spin_lock_irqsave(&ide_lock, flags);
1493 __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 0);
1494 blk_start_queueing(drive->queue);
1495 spin_unlock_irqrestore(&ide_lock, flags);
1498 EXPORT_SYMBOL(ide_do_drive_cmd);
1500 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1502 ide_hwif_t *hwif = drive->hwif;
1505 memset(&task, 0, sizeof(task));
1506 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1507 IDE_TFLAG_OUT_FEATURE | tf_flags;
1508 task.tf.feature = dma; /* Use PIO/DMA */
1509 task.tf.lbam = bcount & 0xff;
1510 task.tf.lbah = (bcount >> 8) & 0xff;
1512 ide_tf_dump(drive->name, &task.tf);
1513 hwif->tp_ops->set_irq(hwif, 1);
1514 SELECT_MASK(drive, 0);
1515 hwif->tp_ops->tf_load(drive, &task);
1518 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1520 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1522 ide_hwif_t *hwif = drive->hwif;
1527 hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1529 hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1533 EXPORT_SYMBOL_GPL(ide_pad_transfer);