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);
123 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq)
125 struct request_pm_state *pm = rq->data;
128 printk(KERN_INFO "%s: complete_power_step(step: %d)\n",
129 drive->name, pm->pm_step);
131 if (drive->media != ide_disk)
134 switch (pm->pm_step) {
135 case IDE_PM_FLUSH_CACHE: /* Suspend step 1 (flush cache) */
136 if (pm->pm_state == PM_EVENT_FREEZE)
137 pm->pm_step = IDE_PM_COMPLETED;
139 pm->pm_step = IDE_PM_STANDBY;
141 case IDE_PM_STANDBY: /* Suspend step 2 (standby) */
142 pm->pm_step = IDE_PM_COMPLETED;
144 case IDE_PM_RESTORE_PIO: /* Resume step 1 (restore PIO) */
145 pm->pm_step = IDE_PM_IDLE;
147 case IDE_PM_IDLE: /* Resume step 2 (idle)*/
148 pm->pm_step = IDE_PM_RESTORE_DMA;
153 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
155 struct request_pm_state *pm = rq->data;
156 ide_task_t *args = rq->special;
158 memset(args, 0, sizeof(*args));
160 switch (pm->pm_step) {
161 case IDE_PM_FLUSH_CACHE: /* Suspend step 1 (flush cache) */
162 if (drive->media != ide_disk)
164 /* Not supported? Switch to next step now. */
165 if (ata_id_flush_enabled(drive->id) == 0 ||
166 (drive->dev_flags & IDE_DFLAG_WCACHE) == 0) {
167 ide_complete_power_step(drive, rq);
170 if (ata_id_flush_ext_enabled(drive->id))
171 args->tf.command = ATA_CMD_FLUSH_EXT;
173 args->tf.command = ATA_CMD_FLUSH;
175 case IDE_PM_STANDBY: /* Suspend step 2 (standby) */
176 args->tf.command = ATA_CMD_STANDBYNOW1;
178 case IDE_PM_RESTORE_PIO: /* Resume step 1 (restore PIO) */
179 ide_set_max_pio(drive);
181 * skip IDE_PM_IDLE for ATAPI devices
183 if (drive->media != ide_disk)
184 pm->pm_step = IDE_PM_RESTORE_DMA;
186 ide_complete_power_step(drive, rq);
188 case IDE_PM_IDLE: /* Resume step 2 (idle) */
189 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
191 case IDE_PM_RESTORE_DMA: /* Resume step 3 (restore DMA) */
193 * Right now, all we do is call ide_set_dma(drive),
194 * we could be smarter and check for current xfer_speed
195 * in struct drive etc...
197 if (drive->hwif->dma_ops == NULL)
200 * TODO: respect IDE_DFLAG_USING_DMA
206 pm->pm_step = IDE_PM_COMPLETED;
210 args->tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
211 args->data_phase = TASKFILE_NO_DATA;
212 return do_rw_taskfile(drive, args);
216 * ide_end_dequeued_request - complete an IDE I/O
217 * @drive: IDE device for the I/O
219 * @nr_sectors: number of sectors completed
221 * Complete an I/O that is no longer on the request queue. This
222 * typically occurs when we pull the request and issue a REQUEST_SENSE.
223 * We must still finish the old request but we must not tamper with the
224 * queue in the meantime.
226 * NOTE: This path does not handle barrier, but barrier is not supported
230 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
231 int uptodate, int nr_sectors)
233 BUG_ON(!blk_rq_started(rq));
235 return __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
237 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
241 * ide_complete_pm_request - end the current Power Management request
242 * @drive: target drive
245 * This function cleans up the current PM request and stops the queue
248 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
253 printk("%s: completing PM request, %s\n", drive->name,
254 blk_pm_suspend_request(rq) ? "suspend" : "resume");
256 spin_lock_irqsave(&ide_lock, flags);
257 if (blk_pm_suspend_request(rq)) {
258 blk_stop_queue(drive->queue);
260 drive->dev_flags &= ~IDE_DFLAG_BLOCKED;
261 blk_start_queue(drive->queue);
263 spin_unlock_irqrestore(&ide_lock, flags);
265 drive->hwif->hwgroup->rq = NULL;
267 if (blk_end_request(rq, 0, 0))
272 * ide_end_drive_cmd - end an explicit drive command
277 * Clean up after success/failure of an explicit drive command.
278 * These get thrown onto the queue so they are synchronized with
279 * real I/O operations on the drive.
281 * In LBA48 mode we have to read the register set twice to get
282 * all the extra information out.
285 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
287 ide_hwgroup_t *hwgroup = drive->hwif->hwgroup;
288 struct request *rq = hwgroup->rq;
290 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
291 ide_task_t *task = (ide_task_t *)rq->special;
294 rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
297 struct ide_taskfile *tf = &task->tf;
302 drive->hwif->tp_ops->tf_read(drive, task);
304 if (task->tf_flags & IDE_TFLAG_DYN)
307 } else if (blk_pm_request(rq)) {
308 struct request_pm_state *pm = rq->data;
310 ide_complete_power_step(drive, rq);
311 if (pm->pm_step == IDE_PM_COMPLETED)
312 ide_complete_pm_request(drive, rq);
320 if (unlikely(blk_end_request(rq, (rq->errors ? -EIO : 0),
324 EXPORT_SYMBOL(ide_end_drive_cmd);
326 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
331 drv = *(ide_driver_t **)rq->rq_disk->private_data;
332 drv->end_request(drive, 0, 0);
334 ide_end_request(drive, 0, 0);
337 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
339 ide_hwif_t *hwif = drive->hwif;
341 if ((stat & ATA_BUSY) ||
342 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
343 /* other bits are useless when BUSY */
344 rq->errors |= ERROR_RESET;
345 } else if (stat & ATA_ERR) {
346 /* err has different meaning on cdrom and tape */
347 if (err == ATA_ABORTED) {
348 if ((drive->dev_flags & IDE_DFLAG_LBA) &&
349 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
350 hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
352 } else if ((err & BAD_CRC) == BAD_CRC) {
353 /* UDMA crc error, just retry the operation */
355 } else if (err & (ATA_BBK | ATA_UNC)) {
356 /* retries won't help these */
357 rq->errors = ERROR_MAX;
358 } else if (err & ATA_TRK0NF) {
359 /* help it find track zero */
360 rq->errors |= ERROR_RECAL;
364 if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
365 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
366 int nsect = drive->mult_count ? drive->mult_count : 1;
368 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
371 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
372 ide_kill_rq(drive, rq);
376 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
377 rq->errors |= ERROR_RESET;
379 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
381 return ide_do_reset(drive);
384 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
385 drive->special.b.recalibrate = 1;
392 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
394 ide_hwif_t *hwif = drive->hwif;
396 if ((stat & ATA_BUSY) ||
397 ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
398 /* other bits are useless when BUSY */
399 rq->errors |= ERROR_RESET;
401 /* add decoding error stuff */
404 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
406 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
408 if (rq->errors >= ERROR_MAX) {
409 ide_kill_rq(drive, rq);
411 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
413 return ide_do_reset(drive);
422 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
424 if (drive->media == ide_disk)
425 return ide_ata_error(drive, rq, stat, err);
426 return ide_atapi_error(drive, rq, stat, err);
429 EXPORT_SYMBOL_GPL(__ide_error);
432 * ide_error - handle an error on the IDE
433 * @drive: drive the error occurred on
434 * @msg: message to report
437 * ide_error() takes action based on the error returned by the drive.
438 * For normal I/O that may well include retries. We deal with
439 * both new-style (taskfile) and old style command handling here.
440 * In the case of taskfile command handling there is work left to
444 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
449 err = ide_dump_status(drive, msg, stat);
451 if ((rq = HWGROUP(drive)->rq) == NULL)
454 /* retry only "normal" I/O: */
455 if (!blk_fs_request(rq)) {
457 ide_end_drive_cmd(drive, stat, err);
464 drv = *(ide_driver_t **)rq->rq_disk->private_data;
465 return drv->error(drive, rq, stat, err);
467 return __ide_error(drive, rq, stat, err);
470 EXPORT_SYMBOL_GPL(ide_error);
472 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
474 tf->nsect = drive->sect;
475 tf->lbal = drive->sect;
476 tf->lbam = drive->cyl;
477 tf->lbah = drive->cyl >> 8;
478 tf->device = (drive->head - 1) | drive->select;
479 tf->command = ATA_CMD_INIT_DEV_PARAMS;
482 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
484 tf->nsect = drive->sect;
485 tf->command = ATA_CMD_RESTORE;
488 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
490 tf->nsect = drive->mult_req;
491 tf->command = ATA_CMD_SET_MULTI;
494 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
496 special_t *s = &drive->special;
499 memset(&args, 0, sizeof(ide_task_t));
500 args.data_phase = TASKFILE_NO_DATA;
502 if (s->b.set_geometry) {
503 s->b.set_geometry = 0;
504 ide_tf_set_specify_cmd(drive, &args.tf);
505 } else if (s->b.recalibrate) {
506 s->b.recalibrate = 0;
507 ide_tf_set_restore_cmd(drive, &args.tf);
508 } else if (s->b.set_multmode) {
509 s->b.set_multmode = 0;
510 ide_tf_set_setmult_cmd(drive, &args.tf);
512 int special = s->all;
514 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
518 args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
519 IDE_TFLAG_CUSTOM_HANDLER;
521 do_rw_taskfile(drive, &args);
527 * do_special - issue some special commands
528 * @drive: drive the command is for
530 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
531 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
533 * It used to do much more, but has been scaled back.
536 static ide_startstop_t do_special (ide_drive_t *drive)
538 special_t *s = &drive->special;
541 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
543 if (drive->media == ide_disk)
544 return ide_disk_special(drive);
551 void ide_map_sg(ide_drive_t *drive, struct request *rq)
553 ide_hwif_t *hwif = drive->hwif;
554 struct scatterlist *sg = hwif->sg_table;
556 if (hwif->sg_mapped) /* needed by ide-scsi */
559 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
560 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
562 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
567 EXPORT_SYMBOL_GPL(ide_map_sg);
569 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
571 ide_hwif_t *hwif = drive->hwif;
573 hwif->nsect = hwif->nleft = rq->nr_sectors;
578 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
581 * execute_drive_command - issue special drive command
582 * @drive: the drive to issue the command on
583 * @rq: the request structure holding the command
585 * execute_drive_cmd() issues a special drive command, usually
586 * initiated by ioctl() from the external hdparm program. The
587 * command can be a drive command, drive task or taskfile
588 * operation. Weirdly you can call it with NULL to wait for
589 * all commands to finish. Don't do this as that is due to change
592 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
595 ide_hwif_t *hwif = HWIF(drive);
596 ide_task_t *task = rq->special;
599 hwif->data_phase = task->data_phase;
601 switch (hwif->data_phase) {
602 case TASKFILE_MULTI_OUT:
604 case TASKFILE_MULTI_IN:
606 ide_init_sg_cmd(drive, rq);
607 ide_map_sg(drive, rq);
612 return do_rw_taskfile(drive, task);
616 * NULL is actually a valid way of waiting for
617 * all current requests to be flushed from the queue.
620 printk("%s: DRIVE_CMD (null)\n", drive->name);
622 ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
623 ide_read_error(drive));
628 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
631 struct request_queue *q = drive->queue;
635 if (!(setting->flags & DS_SYNC))
636 return setting->set(drive, arg);
638 rq = blk_get_request(q, READ, __GFP_WAIT);
639 rq->cmd_type = REQ_TYPE_SPECIAL;
641 rq->cmd[0] = REQ_DEVSET_EXEC;
642 *(int *)&rq->cmd[1] = arg;
643 rq->special = setting->set;
645 if (blk_execute_rq(q, NULL, rq, 0))
651 EXPORT_SYMBOL_GPL(ide_devset_execute);
653 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
657 if (cmd == REQ_PARK_HEADS || cmd == REQ_UNPARK_HEADS) {
659 struct ide_taskfile *tf = &task.tf;
661 memset(&task, 0, sizeof(task));
662 if (cmd == REQ_PARK_HEADS) {
663 drive->sleep = *(unsigned long *)rq->special;
664 drive->dev_flags |= IDE_DFLAG_SLEEPING;
665 tf->command = ATA_CMD_IDLEIMMEDIATE;
670 task.tf_flags |= IDE_TFLAG_CUSTOM_HANDLER;
671 } else /* cmd == REQ_UNPARK_HEADS */
672 tf->command = ATA_CMD_CHK_POWER;
674 task.tf_flags |= IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
676 drive->hwif->data_phase = task.data_phase = TASKFILE_NO_DATA;
677 return do_rw_taskfile(drive, &task);
681 case REQ_DEVSET_EXEC:
683 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
685 err = setfunc(drive, *(int *)&rq->cmd[1]);
690 ide_end_request(drive, err, 0);
693 case REQ_DRIVE_RESET:
694 return ide_do_reset(drive);
696 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
697 ide_end_request(drive, 0, 0);
702 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
704 struct request_pm_state *pm = rq->data;
706 if (blk_pm_suspend_request(rq) &&
707 pm->pm_step == IDE_PM_START_SUSPEND)
708 /* Mark drive blocked when starting the suspend sequence. */
709 drive->dev_flags |= IDE_DFLAG_BLOCKED;
710 else if (blk_pm_resume_request(rq) &&
711 pm->pm_step == IDE_PM_START_RESUME) {
713 * The first thing we do on wakeup is to wait for BSY bit to
714 * go away (with a looong timeout) as a drive on this hwif may
715 * just be POSTing itself.
716 * We do that before even selecting as the "other" device on
717 * the bus may be broken enough to walk on our toes at this
720 ide_hwif_t *hwif = drive->hwif;
723 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
725 rc = ide_wait_not_busy(hwif, 35000);
727 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
729 hwif->tp_ops->set_irq(hwif, 1);
730 rc = ide_wait_not_busy(hwif, 100000);
732 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
737 * start_request - start of I/O and command issuing for IDE
739 * start_request() initiates handling of a new I/O request. It
740 * accepts commands and I/O (read/write) requests.
742 * FIXME: this function needs a rename
745 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
747 ide_startstop_t startstop;
749 BUG_ON(!blk_rq_started(rq));
752 printk("%s: start_request: current=0x%08lx\n",
753 HWIF(drive)->name, (unsigned long) rq);
756 /* bail early if we've exceeded max_failures */
757 if (drive->max_failures && (drive->failures > drive->max_failures)) {
758 rq->cmd_flags |= REQ_FAILED;
762 if (blk_pm_request(rq))
763 ide_check_pm_state(drive, rq);
766 if (ide_wait_stat(&startstop, drive, drive->ready_stat,
767 ATA_BUSY | ATA_DRQ, WAIT_READY)) {
768 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
771 if (!drive->special.all) {
775 * We reset the drive so we need to issue a SETFEATURES.
776 * Do it _after_ do_special() restored device parameters.
778 if (drive->current_speed == 0xff)
779 ide_config_drive_speed(drive, drive->desired_speed);
781 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
782 return execute_drive_cmd(drive, rq);
783 else if (blk_pm_request(rq)) {
784 struct request_pm_state *pm = rq->data;
786 printk("%s: start_power_step(step: %d)\n",
787 drive->name, pm->pm_step);
789 startstop = ide_start_power_step(drive, rq);
790 if (startstop == ide_stopped &&
791 pm->pm_step == IDE_PM_COMPLETED)
792 ide_complete_pm_request(drive, rq);
794 } else if (!rq->rq_disk && blk_special_request(rq))
796 * TODO: Once all ULDs have been modified to
797 * check for specific op codes rather than
798 * blindly accepting any special request, the
799 * check for ->rq_disk above may be replaced
800 * by a more suitable mechanism or even
803 return ide_special_rq(drive, rq);
805 drv = *(ide_driver_t **)rq->rq_disk->private_data;
807 return drv->do_request(drive, rq, rq->sector);
809 return do_special(drive);
811 ide_kill_rq(drive, rq);
816 * ide_stall_queue - pause an IDE device
817 * @drive: drive to stall
818 * @timeout: time to stall for (jiffies)
820 * ide_stall_queue() can be used by a drive to give excess bandwidth back
821 * to the hwgroup by sleeping for timeout jiffies.
824 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
826 if (timeout > WAIT_WORSTCASE)
827 timeout = WAIT_WORSTCASE;
828 drive->sleep = timeout + jiffies;
829 drive->dev_flags |= IDE_DFLAG_SLEEPING;
832 EXPORT_SYMBOL(ide_stall_queue);
834 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
837 * choose_drive - select a drive to service
838 * @hwgroup: hardware group to select on
840 * choose_drive() selects the next drive which will be serviced.
841 * This is necessary because the IDE layer can't issue commands
842 * to both drives on the same cable, unlike SCSI.
845 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
847 ide_drive_t *drive, *best;
851 drive = hwgroup->drive;
854 * drive is doing pre-flush, ordered write, post-flush sequence. even
855 * though that is 3 requests, it must be seen as a single transaction.
856 * we must not preempt this drive until that is complete
858 if (blk_queue_flushing(drive->queue)) {
860 * small race where queue could get replugged during
861 * the 3-request flush cycle, just yank the plug since
862 * we want it to finish asap
864 blk_remove_plug(drive->queue);
869 u8 dev_s = !!(drive->dev_flags & IDE_DFLAG_SLEEPING);
870 u8 best_s = (best && !!(best->dev_flags & IDE_DFLAG_SLEEPING));
872 if ((dev_s == 0 || time_after_eq(jiffies, drive->sleep)) &&
873 !elv_queue_empty(drive->queue)) {
875 (dev_s && (best_s == 0 || time_before(drive->sleep, best->sleep))) ||
876 (best_s == 0 && time_before(WAKEUP(drive), WAKEUP(best)))) {
877 if (!blk_queue_plugged(drive->queue))
881 } while ((drive = drive->next) != hwgroup->drive);
883 if (best && (best->dev_flags & IDE_DFLAG_NICE1) &&
884 (best->dev_flags & IDE_DFLAG_SLEEPING) == 0 &&
885 best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
886 long t = (signed long)(WAKEUP(best) - jiffies);
887 if (t >= WAIT_MIN_SLEEP) {
889 * We *may* have some time to spare, but first let's see if
890 * someone can potentially benefit from our nice mood today..
894 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) == 0
895 && time_before(jiffies - best->service_time, WAKEUP(drive))
896 && time_before(WAKEUP(drive), jiffies + t))
898 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
901 } while ((drive = drive->next) != best);
908 * Issue a new request to a drive from hwgroup
909 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
911 * A hwgroup is a serialized group of IDE interfaces. Usually there is
912 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
913 * may have both interfaces in a single hwgroup to "serialize" access.
914 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
915 * together into one hwgroup for serialized access.
917 * Note also that several hwgroups can end up sharing a single IRQ,
918 * possibly along with many other devices. This is especially common in
919 * PCI-based systems with off-board IDE controller cards.
921 * The IDE driver uses the single global ide_lock spinlock to protect
922 * access to the request queues, and to protect the hwgroup->busy flag.
924 * The first thread into the driver for a particular hwgroup sets the
925 * hwgroup->busy flag to indicate that this hwgroup is now active,
926 * and then initiates processing of the top request from the request queue.
928 * Other threads attempting entry notice the busy setting, and will simply
929 * queue their new requests and exit immediately. Note that hwgroup->busy
930 * remains set even when the driver is merely awaiting the next interrupt.
931 * Thus, the meaning is "this hwgroup is busy processing a request".
933 * When processing of a request completes, the completing thread or IRQ-handler
934 * will start the next request from the queue. If no more work remains,
935 * the driver will clear the hwgroup->busy flag and exit.
937 * The ide_lock (spinlock) is used to protect all access to the
938 * hwgroup->busy flag, but is otherwise not needed for most processing in
939 * the driver. This makes the driver much more friendlier to shared IRQs
940 * than previous designs, while remaining 100% (?) SMP safe and capable.
942 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
947 ide_startstop_t startstop;
950 /* caller must own ide_lock */
951 BUG_ON(!irqs_disabled());
953 while (!hwgroup->busy) {
956 ide_get_lock(ide_intr, hwgroup);
957 drive = choose_drive(hwgroup);
960 unsigned long sleep = 0; /* shut up, gcc */
962 drive = hwgroup->drive;
964 if ((drive->dev_flags & IDE_DFLAG_SLEEPING) &&
966 time_before(drive->sleep, sleep))) {
968 sleep = drive->sleep;
970 } while ((drive = drive->next) != hwgroup->drive);
973 * Take a short snooze, and then wake up this hwgroup again.
974 * This gives other hwgroups on the same a chance to
975 * play fairly with us, just in case there are big differences
976 * in relative throughputs.. don't want to hog the cpu too much.
978 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
979 sleep = jiffies + WAIT_MIN_SLEEP;
981 if (timer_pending(&hwgroup->timer))
982 printk(KERN_CRIT "ide_set_handler: timer already active\n");
984 /* so that ide_timer_expiry knows what to do */
985 hwgroup->sleeping = 1;
986 hwgroup->req_gen_timer = hwgroup->req_gen;
987 mod_timer(&hwgroup->timer, sleep);
988 /* we purposely leave hwgroup->busy==1
991 /* Ugly, but how can we sleep for the lock
992 * otherwise? perhaps from tq_disk?
1000 /* no more work for this hwgroup (for now) */
1005 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1007 * set nIEN for previous hwif, drives in the
1008 * quirk_list may not like intr setups/cleanups
1010 if (drive->quirk_list != 1)
1011 hwif->tp_ops->set_irq(hwif, 0);
1013 hwgroup->hwif = hwif;
1014 hwgroup->drive = drive;
1015 drive->dev_flags &= ~(IDE_DFLAG_SLEEPING | IDE_DFLAG_PARKED);
1016 drive->service_start = jiffies;
1018 if (blk_queue_plugged(drive->queue)) {
1019 printk(KERN_ERR "ide: huh? queue was plugged!\n");
1024 * we know that the queue isn't empty, but this can happen
1025 * if the q->prep_rq_fn() decides to kill a request
1027 rq = elv_next_request(drive->queue);
1034 * Sanity: don't accept a request that isn't a PM request
1035 * if we are currently power managed. This is very important as
1036 * blk_stop_queue() doesn't prevent the elv_next_request()
1037 * above to return us whatever is in the queue. Since we call
1038 * ide_do_request() ourselves, we end up taking requests while
1039 * the queue is blocked...
1041 * We let requests forced at head of queue with ide-preempt
1042 * though. I hope that doesn't happen too much, hopefully not
1043 * unless the subdriver triggers such a thing in its own PM
1046 * We count how many times we loop here to make sure we service
1047 * all drives in the hwgroup without looping for ever
1049 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
1050 blk_pm_request(rq) == 0 &&
1051 (rq->cmd_flags & REQ_PREEMPT) == 0) {
1052 drive = drive->next ? drive->next : hwgroup->drive;
1053 if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1055 /* We clear busy, there should be no pending ATA command at this point. */
1063 * Some systems have trouble with IDE IRQs arriving while
1064 * the driver is still setting things up. So, here we disable
1065 * the IRQ used by this interface while the request is being started.
1066 * This may look bad at first, but pretty much the same thing
1067 * happens anyway when any interrupt comes in, IDE or otherwise
1068 * -- the kernel masks the IRQ while it is being handled.
1070 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1071 disable_irq_nosync(hwif->irq);
1072 spin_unlock(&ide_lock);
1073 local_irq_enable_in_hardirq();
1074 /* allow other IRQs while we start this request */
1075 startstop = start_request(drive, rq);
1076 spin_lock_irq(&ide_lock);
1077 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1078 enable_irq(hwif->irq);
1079 if (startstop == ide_stopped)
1085 * Passes the stuff to ide_do_request
1087 void do_ide_request(struct request_queue *q)
1089 ide_drive_t *drive = q->queuedata;
1091 ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1095 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1096 * retry the current request in pio mode instead of risking tossing it
1099 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1101 ide_hwif_t *hwif = HWIF(drive);
1103 ide_startstop_t ret = ide_stopped;
1106 * end current dma transaction
1110 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1111 (void)hwif->dma_ops->dma_end(drive);
1112 ret = ide_error(drive, "dma timeout error",
1113 hwif->tp_ops->read_status(hwif));
1115 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1116 hwif->dma_ops->dma_timeout(drive);
1120 * disable dma for now, but remember that we did so because of
1121 * a timeout -- we'll reenable after we finish this next request
1122 * (or rather the first chunk of it) in pio.
1124 drive->dev_flags |= IDE_DFLAG_DMA_PIO_RETRY;
1126 ide_dma_off_quietly(drive);
1129 * un-busy drive etc (hwgroup->busy is cleared on return) and
1130 * make sure request is sane
1132 rq = HWGROUP(drive)->rq;
1137 HWGROUP(drive)->rq = NULL;
1144 rq->sector = rq->bio->bi_sector;
1145 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1146 rq->hard_cur_sectors = rq->current_nr_sectors;
1147 rq->buffer = bio_data(rq->bio);
1153 * ide_timer_expiry - handle lack of an IDE interrupt
1154 * @data: timer callback magic (hwgroup)
1156 * An IDE command has timed out before the expected drive return
1157 * occurred. At this point we attempt to clean up the current
1158 * mess. If the current handler includes an expiry handler then
1159 * we invoke the expiry handler, and providing it is happy the
1160 * work is done. If that fails we apply generic recovery rules
1161 * invoking the handler and checking the drive DMA status. We
1162 * have an excessively incestuous relationship with the DMA
1163 * logic that wants cleaning up.
1166 void ide_timer_expiry (unsigned long data)
1168 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1169 ide_handler_t *handler;
1170 ide_expiry_t *expiry;
1171 unsigned long flags;
1172 unsigned long wait = -1;
1174 spin_lock_irqsave(&ide_lock, flags);
1176 if (((handler = hwgroup->handler) == NULL) ||
1177 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1179 * Either a marginal timeout occurred
1180 * (got the interrupt just as timer expired),
1181 * or we were "sleeping" to give other devices a chance.
1182 * Either way, we don't really want to complain about anything.
1184 if (hwgroup->sleeping) {
1185 hwgroup->sleeping = 0;
1189 ide_drive_t *drive = hwgroup->drive;
1191 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1192 hwgroup->handler = NULL;
1195 ide_startstop_t startstop = ide_stopped;
1196 if (!hwgroup->busy) {
1197 hwgroup->busy = 1; /* paranoia */
1198 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1200 if ((expiry = hwgroup->expiry) != NULL) {
1202 if ((wait = expiry(drive)) > 0) {
1204 hwgroup->timer.expires = jiffies + wait;
1205 hwgroup->req_gen_timer = hwgroup->req_gen;
1206 add_timer(&hwgroup->timer);
1207 spin_unlock_irqrestore(&ide_lock, flags);
1211 hwgroup->handler = NULL;
1213 * We need to simulate a real interrupt when invoking
1214 * the handler() function, which means we need to
1215 * globally mask the specific IRQ:
1217 spin_unlock(&ide_lock);
1219 /* disable_irq_nosync ?? */
1220 disable_irq(hwif->irq);
1222 * as if we were handling an interrupt */
1223 local_irq_disable();
1224 if (hwgroup->polling) {
1225 startstop = handler(drive);
1226 } else if (drive_is_ready(drive)) {
1227 if (drive->waiting_for_dma)
1228 hwif->dma_ops->dma_lost_irq(drive);
1229 (void)ide_ack_intr(hwif);
1230 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1231 startstop = handler(drive);
1233 if (drive->waiting_for_dma) {
1234 startstop = ide_dma_timeout_retry(drive, wait);
1237 ide_error(drive, "irq timeout",
1238 hwif->tp_ops->read_status(hwif));
1240 drive->service_time = jiffies - drive->service_start;
1241 spin_lock_irq(&ide_lock);
1242 enable_irq(hwif->irq);
1243 if (startstop == ide_stopped)
1247 ide_do_request(hwgroup, IDE_NO_IRQ);
1248 spin_unlock_irqrestore(&ide_lock, flags);
1252 * unexpected_intr - handle an unexpected IDE interrupt
1253 * @irq: interrupt line
1254 * @hwgroup: hwgroup being processed
1256 * There's nothing really useful we can do with an unexpected interrupt,
1257 * other than reading the status register (to clear it), and logging it.
1258 * There should be no way that an irq can happen before we're ready for it,
1259 * so we needn't worry much about losing an "important" interrupt here.
1261 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1262 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1263 * looks "good", we just ignore the interrupt completely.
1265 * This routine assumes __cli() is in effect when called.
1267 * If an unexpected interrupt happens on irq15 while we are handling irq14
1268 * and if the two interfaces are "serialized" (CMD640), then it looks like
1269 * we could screw up by interfering with a new request being set up for
1272 * In reality, this is a non-issue. The new command is not sent unless
1273 * the drive is ready to accept one, in which case we know the drive is
1274 * not trying to interrupt us. And ide_set_handler() is always invoked
1275 * before completing the issuance of any new drive command, so we will not
1276 * be accidentally invoked as a result of any valid command completion
1279 * Note that we must walk the entire hwgroup here. We know which hwif
1280 * is doing the current command, but we don't know which hwif burped
1284 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1287 ide_hwif_t *hwif = hwgroup->hwif;
1290 * handle the unexpected interrupt
1293 if (hwif->irq == irq) {
1294 stat = hwif->tp_ops->read_status(hwif);
1296 if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1297 /* Try to not flood the console with msgs */
1298 static unsigned long last_msgtime, count;
1300 if (time_after(jiffies, last_msgtime + HZ)) {
1301 last_msgtime = jiffies;
1302 printk(KERN_ERR "%s%s: unexpected interrupt, "
1303 "status=0x%02x, count=%ld\n",
1305 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1309 } while ((hwif = hwif->next) != hwgroup->hwif);
1313 * ide_intr - default IDE interrupt handler
1314 * @irq: interrupt number
1315 * @dev_id: hwif group
1316 * @regs: unused weirdness from the kernel irq layer
1318 * This is the default IRQ handler for the IDE layer. You should
1319 * not need to override it. If you do be aware it is subtle in
1322 * hwgroup->hwif is the interface in the group currently performing
1323 * a command. hwgroup->drive is the drive and hwgroup->handler is
1324 * the IRQ handler to call. As we issue a command the handlers
1325 * step through multiple states, reassigning the handler to the
1326 * next step in the process. Unlike a smart SCSI controller IDE
1327 * expects the main processor to sequence the various transfer
1328 * stages. We also manage a poll timer to catch up with most
1329 * timeout situations. There are still a few where the handlers
1330 * don't ever decide to give up.
1332 * The handler eventually returns ide_stopped to indicate the
1333 * request completed. At this point we issue the next request
1334 * on the hwgroup and the process begins again.
1337 irqreturn_t ide_intr (int irq, void *dev_id)
1339 unsigned long flags;
1340 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1343 ide_handler_t *handler;
1344 ide_startstop_t startstop;
1345 irqreturn_t irq_ret = IRQ_NONE;
1347 spin_lock_irqsave(&ide_lock, flags);
1348 hwif = hwgroup->hwif;
1350 if (!ide_ack_intr(hwif))
1353 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1355 * Not expecting an interrupt from this drive.
1356 * That means this could be:
1357 * (1) an interrupt from another PCI device
1358 * sharing the same PCI INT# as us.
1359 * or (2) a drive just entered sleep or standby mode,
1360 * and is interrupting to let us know.
1361 * or (3) a spurious interrupt of unknown origin.
1363 * For PCI, we cannot tell the difference,
1364 * so in that case we just ignore it and hope it goes away.
1366 * FIXME: unexpected_intr should be hwif-> then we can
1367 * remove all the ifdef PCI crap
1369 #ifdef CONFIG_BLK_DEV_IDEPCI
1370 if (hwif->chipset != ide_pci)
1371 #endif /* CONFIG_BLK_DEV_IDEPCI */
1374 * Probably not a shared PCI interrupt,
1375 * so we can safely try to do something about it:
1377 unexpected_intr(irq, hwgroup);
1378 #ifdef CONFIG_BLK_DEV_IDEPCI
1381 * Whack the status register, just in case
1382 * we have a leftover pending IRQ.
1384 (void)hwif->tp_ops->read_status(hwif);
1385 #endif /* CONFIG_BLK_DEV_IDEPCI */
1390 drive = hwgroup->drive;
1393 * This should NEVER happen, and there isn't much
1394 * we could do about it here.
1396 * [Note - this can occur if the drive is hot unplugged]
1401 if (!drive_is_ready(drive))
1403 * This happens regularly when we share a PCI IRQ with
1404 * another device. Unfortunately, it can also happen
1405 * with some buggy drives that trigger the IRQ before
1406 * their status register is up to date. Hopefully we have
1407 * enough advance overhead that the latter isn't a problem.
1411 if (!hwgroup->busy) {
1412 hwgroup->busy = 1; /* paranoia */
1413 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1415 hwgroup->handler = NULL;
1417 del_timer(&hwgroup->timer);
1418 spin_unlock(&ide_lock);
1420 if (hwif->port_ops && hwif->port_ops->clear_irq)
1421 hwif->port_ops->clear_irq(drive);
1423 if (drive->dev_flags & IDE_DFLAG_UNMASK)
1424 local_irq_enable_in_hardirq();
1426 /* service this interrupt, may set handler for next interrupt */
1427 startstop = handler(drive);
1429 spin_lock_irq(&ide_lock);
1431 * Note that handler() may have set things up for another
1432 * interrupt to occur soon, but it cannot happen until
1433 * we exit from this routine, because it will be the
1434 * same irq as is currently being serviced here, and Linux
1435 * won't allow another of the same (on any CPU) until we return.
1437 drive->service_time = jiffies - drive->service_start;
1438 if (startstop == ide_stopped) {
1439 if (hwgroup->handler == NULL) { /* paranoia */
1441 ide_do_request(hwgroup, hwif->irq);
1443 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1444 "on exit\n", drive->name);
1448 irq_ret = IRQ_HANDLED;
1450 spin_unlock_irqrestore(&ide_lock, flags);
1455 * ide_do_drive_cmd - issue IDE special command
1456 * @drive: device to issue command
1457 * @rq: request to issue
1459 * This function issues a special IDE device request
1460 * onto the request queue.
1462 * the rq is queued at the head of the request queue, displacing
1463 * the currently-being-processed request and this function
1464 * returns immediately without waiting for the new rq to be
1465 * completed. This is VERY DANGEROUS, and is intended for
1466 * careful use by the ATAPI tape/cdrom driver code.
1469 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1471 ide_hwgroup_t *hwgroup = drive->hwif->hwgroup;
1472 unsigned long flags;
1476 spin_lock_irqsave(&ide_lock, flags);
1477 __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 0);
1478 blk_start_queueing(drive->queue);
1479 spin_unlock_irqrestore(&ide_lock, flags);
1482 EXPORT_SYMBOL(ide_do_drive_cmd);
1484 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1486 ide_hwif_t *hwif = drive->hwif;
1489 memset(&task, 0, sizeof(task));
1490 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1491 IDE_TFLAG_OUT_FEATURE | tf_flags;
1492 task.tf.feature = dma; /* Use PIO/DMA */
1493 task.tf.lbam = bcount & 0xff;
1494 task.tf.lbah = (bcount >> 8) & 0xff;
1496 ide_tf_dump(drive->name, &task.tf);
1497 hwif->tp_ops->set_irq(hwif, 1);
1498 SELECT_MASK(drive, 0);
1499 hwif->tp_ops->tf_load(drive, &task);
1502 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1504 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1506 ide_hwif_t *hwif = drive->hwif;
1511 hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1513 hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1517 EXPORT_SYMBOL_GPL(ide_pad_transfer);