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->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
86 if (!__blk_end_request(rq, error, nr_bytes)) {
88 HWGROUP(drive)->rq = NULL;
96 * ide_end_request - complete an IDE I/O
97 * @drive: IDE device for the I/O
99 * @nr_sectors: number of sectors completed
101 * This is our end_request wrapper function. We complete the I/O
102 * update random number input and dequeue the request, which if
103 * it was tagged may be out of order.
106 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
108 unsigned int nr_bytes = nr_sectors << 9;
114 * room for locking improvements here, the calls below don't
115 * need the queue lock held at all
117 spin_lock_irqsave(&ide_lock, flags);
118 rq = HWGROUP(drive)->rq;
121 if (blk_pc_request(rq))
122 nr_bytes = rq->data_len;
124 nr_bytes = rq->hard_cur_sectors << 9;
127 ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
129 spin_unlock_irqrestore(&ide_lock, flags);
132 EXPORT_SYMBOL(ide_end_request);
135 * Power Management state machine. This one is rather trivial for now,
136 * we should probably add more, like switching back to PIO on suspend
137 * to help some BIOSes, re-do the door locking on resume, etc...
141 ide_pm_flush_cache = ide_pm_state_start_suspend,
144 idedisk_pm_restore_pio = ide_pm_state_start_resume,
149 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
151 struct request_pm_state *pm = rq->data;
153 if (drive->media != ide_disk)
156 switch (pm->pm_step) {
157 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) complete */
158 if (pm->pm_state == PM_EVENT_FREEZE)
159 pm->pm_step = ide_pm_state_completed;
161 pm->pm_step = idedisk_pm_standby;
163 case idedisk_pm_standby: /* Suspend step 2 (standby) complete */
164 pm->pm_step = ide_pm_state_completed;
166 case idedisk_pm_restore_pio: /* Resume step 1 complete */
167 pm->pm_step = idedisk_pm_idle;
169 case idedisk_pm_idle: /* Resume step 2 (idle) complete */
170 pm->pm_step = ide_pm_restore_dma;
175 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
177 struct request_pm_state *pm = rq->data;
178 ide_task_t *args = rq->special;
180 memset(args, 0, sizeof(*args));
182 switch (pm->pm_step) {
183 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) */
184 if (drive->media != ide_disk)
186 /* Not supported? Switch to next step now. */
187 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
188 ide_complete_power_step(drive, rq, 0, 0);
191 if (ide_id_has_flush_cache_ext(drive->id))
192 args->tf.command = ATA_CMD_FLUSH_EXT;
194 args->tf.command = ATA_CMD_FLUSH;
197 case idedisk_pm_standby: /* Suspend step 2 (standby) */
198 args->tf.command = ATA_CMD_STANDBYNOW1;
201 case idedisk_pm_restore_pio: /* Resume step 1 (restore PIO) */
202 ide_set_max_pio(drive);
204 * skip idedisk_pm_idle for ATAPI devices
206 if (drive->media != ide_disk)
207 pm->pm_step = ide_pm_restore_dma;
209 ide_complete_power_step(drive, rq, 0, 0);
212 case idedisk_pm_idle: /* Resume step 2 (idle) */
213 args->tf.command = ATA_CMD_IDLEIMMEDIATE;
216 case ide_pm_restore_dma: /* Resume step 3 (restore DMA) */
218 * Right now, all we do is call ide_set_dma(drive),
219 * we could be smarter and check for current xfer_speed
220 * in struct drive etc...
222 if (drive->hwif->dma_ops == NULL)
225 * TODO: respect ->using_dma setting
230 pm->pm_step = ide_pm_state_completed;
234 args->tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
235 args->data_phase = TASKFILE_NO_DATA;
236 return do_rw_taskfile(drive, args);
240 * ide_end_dequeued_request - complete an IDE I/O
241 * @drive: IDE device for the I/O
243 * @nr_sectors: number of sectors completed
245 * Complete an I/O that is no longer on the request queue. This
246 * typically occurs when we pull the request and issue a REQUEST_SENSE.
247 * We must still finish the old request but we must not tamper with the
248 * queue in the meantime.
250 * NOTE: This path does not handle barrier, but barrier is not supported
254 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
255 int uptodate, int nr_sectors)
260 spin_lock_irqsave(&ide_lock, flags);
261 BUG_ON(!blk_rq_started(rq));
262 ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
263 spin_unlock_irqrestore(&ide_lock, flags);
267 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
271 * ide_complete_pm_request - end the current Power Management request
272 * @drive: target drive
275 * This function cleans up the current PM request and stops the queue
278 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
283 printk("%s: completing PM request, %s\n", drive->name,
284 blk_pm_suspend_request(rq) ? "suspend" : "resume");
286 spin_lock_irqsave(&ide_lock, flags);
287 if (blk_pm_suspend_request(rq)) {
288 blk_stop_queue(drive->queue);
291 blk_start_queue(drive->queue);
293 HWGROUP(drive)->rq = NULL;
294 if (__blk_end_request(rq, 0, 0))
296 spin_unlock_irqrestore(&ide_lock, flags);
300 * ide_end_drive_cmd - end an explicit drive command
305 * Clean up after success/failure of an explicit drive command.
306 * These get thrown onto the queue so they are synchronized with
307 * real I/O operations on the drive.
309 * In LBA48 mode we have to read the register set twice to get
310 * all the extra information out.
313 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
318 spin_lock_irqsave(&ide_lock, flags);
319 rq = HWGROUP(drive)->rq;
320 spin_unlock_irqrestore(&ide_lock, flags);
322 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
323 ide_task_t *task = (ide_task_t *)rq->special;
326 rq->errors = !OK_STAT(stat, ATA_DRDY, BAD_STAT);
329 struct ide_taskfile *tf = &task->tf;
334 drive->hwif->tp_ops->tf_read(drive, task);
336 if (task->tf_flags & IDE_TFLAG_DYN)
339 } else if (blk_pm_request(rq)) {
340 struct request_pm_state *pm = rq->data;
342 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
343 drive->name, rq->pm->pm_step, stat, err);
345 ide_complete_power_step(drive, rq, stat, err);
346 if (pm->pm_step == ide_pm_state_completed)
347 ide_complete_pm_request(drive, rq);
351 spin_lock_irqsave(&ide_lock, flags);
352 HWGROUP(drive)->rq = NULL;
354 if (unlikely(__blk_end_request(rq, (rq->errors ? -EIO : 0),
357 spin_unlock_irqrestore(&ide_lock, flags);
360 EXPORT_SYMBOL(ide_end_drive_cmd);
362 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
367 drv = *(ide_driver_t **)rq->rq_disk->private_data;
368 drv->end_request(drive, 0, 0);
370 ide_end_request(drive, 0, 0);
373 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
375 ide_hwif_t *hwif = drive->hwif;
377 if ((stat & ATA_BUSY) || ((stat & ATA_DF) && !drive->nowerr)) {
378 /* other bits are useless when BUSY */
379 rq->errors |= ERROR_RESET;
380 } else if (stat & ATA_ERR) {
381 /* err has different meaning on cdrom and tape */
382 if (err == ATA_ABORTED) {
383 if (drive->select.b.lba &&
384 /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
385 hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
387 } else if ((err & BAD_CRC) == BAD_CRC) {
388 /* UDMA crc error, just retry the operation */
390 } else if (err & (ATA_BBK | ATA_UNC)) {
391 /* retries won't help these */
392 rq->errors = ERROR_MAX;
393 } else if (err & ATA_TRK0NF) {
394 /* help it find track zero */
395 rq->errors |= ERROR_RECAL;
399 if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
400 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
401 int nsect = drive->mult_count ? drive->mult_count : 1;
403 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
406 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
407 ide_kill_rq(drive, rq);
411 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
412 rq->errors |= ERROR_RESET;
414 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
416 return ide_do_reset(drive);
419 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
420 drive->special.b.recalibrate = 1;
427 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
429 ide_hwif_t *hwif = drive->hwif;
431 if ((stat & ATA_BUSY) || ((stat & ATA_DF) && !drive->nowerr)) {
432 /* other bits are useless when BUSY */
433 rq->errors |= ERROR_RESET;
435 /* add decoding error stuff */
438 if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
440 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
442 if (rq->errors >= ERROR_MAX) {
443 ide_kill_rq(drive, rq);
445 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
447 return ide_do_reset(drive);
456 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
458 if (drive->media == ide_disk)
459 return ide_ata_error(drive, rq, stat, err);
460 return ide_atapi_error(drive, rq, stat, err);
463 EXPORT_SYMBOL_GPL(__ide_error);
466 * ide_error - handle an error on the IDE
467 * @drive: drive the error occurred on
468 * @msg: message to report
471 * ide_error() takes action based on the error returned by the drive.
472 * For normal I/O that may well include retries. We deal with
473 * both new-style (taskfile) and old style command handling here.
474 * In the case of taskfile command handling there is work left to
478 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
483 err = ide_dump_status(drive, msg, stat);
485 if ((rq = HWGROUP(drive)->rq) == NULL)
488 /* retry only "normal" I/O: */
489 if (!blk_fs_request(rq)) {
491 ide_end_drive_cmd(drive, stat, err);
498 drv = *(ide_driver_t **)rq->rq_disk->private_data;
499 return drv->error(drive, rq, stat, err);
501 return __ide_error(drive, rq, stat, err);
504 EXPORT_SYMBOL_GPL(ide_error);
506 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
508 tf->nsect = drive->sect;
509 tf->lbal = drive->sect;
510 tf->lbam = drive->cyl;
511 tf->lbah = drive->cyl >> 8;
512 tf->device = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
513 tf->command = ATA_CMD_INIT_DEV_PARAMS;
516 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
518 tf->nsect = drive->sect;
519 tf->command = ATA_CMD_RESTORE;
522 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
524 tf->nsect = drive->mult_req;
525 tf->command = ATA_CMD_SET_MULTI;
528 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
530 special_t *s = &drive->special;
533 memset(&args, 0, sizeof(ide_task_t));
534 args.data_phase = TASKFILE_NO_DATA;
536 if (s->b.set_geometry) {
537 s->b.set_geometry = 0;
538 ide_tf_set_specify_cmd(drive, &args.tf);
539 } else if (s->b.recalibrate) {
540 s->b.recalibrate = 0;
541 ide_tf_set_restore_cmd(drive, &args.tf);
542 } else if (s->b.set_multmode) {
543 s->b.set_multmode = 0;
544 ide_tf_set_setmult_cmd(drive, &args.tf);
546 int special = s->all;
548 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
552 args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
553 IDE_TFLAG_CUSTOM_HANDLER;
555 do_rw_taskfile(drive, &args);
561 * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
563 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
572 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
575 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
578 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
585 * do_special - issue some special commands
586 * @drive: drive the command is for
588 * do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
589 * ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
591 * It used to do much more, but has been scaled back.
594 static ide_startstop_t do_special (ide_drive_t *drive)
596 special_t *s = &drive->special;
599 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
602 ide_hwif_t *hwif = drive->hwif;
603 const struct ide_port_ops *port_ops = hwif->port_ops;
604 u8 req_pio = drive->tune_req;
608 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
610 * take ide_lock for drive->[no_]unmask/[no_]io_32bit
612 if (req_pio == 8 || req_pio == 9) {
615 spin_lock_irqsave(&ide_lock, flags);
616 port_ops->set_pio_mode(drive, req_pio);
617 spin_unlock_irqrestore(&ide_lock, flags);
619 port_ops->set_pio_mode(drive, req_pio);
621 int keep_dma = drive->using_dma;
623 ide_set_pio(drive, req_pio);
625 if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
633 if (drive->media == ide_disk)
634 return ide_disk_special(drive);
642 void ide_map_sg(ide_drive_t *drive, struct request *rq)
644 ide_hwif_t *hwif = drive->hwif;
645 struct scatterlist *sg = hwif->sg_table;
647 if (hwif->sg_mapped) /* needed by ide-scsi */
650 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
651 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
653 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
658 EXPORT_SYMBOL_GPL(ide_map_sg);
660 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
662 ide_hwif_t *hwif = drive->hwif;
664 hwif->nsect = hwif->nleft = rq->nr_sectors;
669 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
672 * execute_drive_command - issue special drive command
673 * @drive: the drive to issue the command on
674 * @rq: the request structure holding the command
676 * execute_drive_cmd() issues a special drive command, usually
677 * initiated by ioctl() from the external hdparm program. The
678 * command can be a drive command, drive task or taskfile
679 * operation. Weirdly you can call it with NULL to wait for
680 * all commands to finish. Don't do this as that is due to change
683 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
686 ide_hwif_t *hwif = HWIF(drive);
687 ide_task_t *task = rq->special;
690 hwif->data_phase = task->data_phase;
692 switch (hwif->data_phase) {
693 case TASKFILE_MULTI_OUT:
695 case TASKFILE_MULTI_IN:
697 ide_init_sg_cmd(drive, rq);
698 ide_map_sg(drive, rq);
703 return do_rw_taskfile(drive, task);
707 * NULL is actually a valid way of waiting for
708 * all current requests to be flushed from the queue.
711 printk("%s: DRIVE_CMD (null)\n", drive->name);
713 ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
714 ide_read_error(drive));
719 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
721 switch (rq->cmd[0]) {
722 case REQ_DRIVE_RESET:
723 return ide_do_reset(drive);
725 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
726 ide_end_request(drive, 0, 0);
731 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
733 struct request_pm_state *pm = rq->data;
735 if (blk_pm_suspend_request(rq) &&
736 pm->pm_step == ide_pm_state_start_suspend)
737 /* Mark drive blocked when starting the suspend sequence. */
739 else if (blk_pm_resume_request(rq) &&
740 pm->pm_step == ide_pm_state_start_resume) {
742 * The first thing we do on wakeup is to wait for BSY bit to
743 * go away (with a looong timeout) as a drive on this hwif may
744 * just be POSTing itself.
745 * We do that before even selecting as the "other" device on
746 * the bus may be broken enough to walk on our toes at this
749 ide_hwif_t *hwif = drive->hwif;
752 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
754 rc = ide_wait_not_busy(hwif, 35000);
756 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
758 hwif->tp_ops->set_irq(hwif, 1);
759 rc = ide_wait_not_busy(hwif, 100000);
761 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
766 * start_request - start of I/O and command issuing for IDE
768 * start_request() initiates handling of a new I/O request. It
769 * accepts commands and I/O (read/write) requests.
771 * FIXME: this function needs a rename
774 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
776 ide_startstop_t startstop;
778 BUG_ON(!blk_rq_started(rq));
781 printk("%s: start_request: current=0x%08lx\n",
782 HWIF(drive)->name, (unsigned long) rq);
785 /* bail early if we've exceeded max_failures */
786 if (drive->max_failures && (drive->failures > drive->max_failures)) {
787 rq->cmd_flags |= REQ_FAILED;
791 if (blk_pm_request(rq))
792 ide_check_pm_state(drive, rq);
795 if (ide_wait_stat(&startstop, drive, drive->ready_stat,
796 ATA_BUSY | ATA_DRQ, WAIT_READY)) {
797 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
800 if (!drive->special.all) {
804 * We reset the drive so we need to issue a SETFEATURES.
805 * Do it _after_ do_special() restored device parameters.
807 if (drive->current_speed == 0xff)
808 ide_config_drive_speed(drive, drive->desired_speed);
810 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
811 return execute_drive_cmd(drive, rq);
812 else if (blk_pm_request(rq)) {
813 struct request_pm_state *pm = rq->data;
815 printk("%s: start_power_step(step: %d)\n",
816 drive->name, rq->pm->pm_step);
818 startstop = ide_start_power_step(drive, rq);
819 if (startstop == ide_stopped &&
820 pm->pm_step == ide_pm_state_completed)
821 ide_complete_pm_request(drive, rq);
823 } else if (!rq->rq_disk && blk_special_request(rq))
825 * TODO: Once all ULDs have been modified to
826 * check for specific op codes rather than
827 * blindly accepting any special request, the
828 * check for ->rq_disk above may be replaced
829 * by a more suitable mechanism or even
832 return ide_special_rq(drive, rq);
834 drv = *(ide_driver_t **)rq->rq_disk->private_data;
836 return drv->do_request(drive, rq, rq->sector);
838 return do_special(drive);
840 ide_kill_rq(drive, rq);
845 * ide_stall_queue - pause an IDE device
846 * @drive: drive to stall
847 * @timeout: time to stall for (jiffies)
849 * ide_stall_queue() can be used by a drive to give excess bandwidth back
850 * to the hwgroup by sleeping for timeout jiffies.
853 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
855 if (timeout > WAIT_WORSTCASE)
856 timeout = WAIT_WORSTCASE;
857 drive->sleep = timeout + jiffies;
861 EXPORT_SYMBOL(ide_stall_queue);
863 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
866 * choose_drive - select a drive to service
867 * @hwgroup: hardware group to select on
869 * choose_drive() selects the next drive which will be serviced.
870 * This is necessary because the IDE layer can't issue commands
871 * to both drives on the same cable, unlike SCSI.
874 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
876 ide_drive_t *drive, *best;
880 drive = hwgroup->drive;
883 * drive is doing pre-flush, ordered write, post-flush sequence. even
884 * though that is 3 requests, it must be seen as a single transaction.
885 * we must not preempt this drive until that is complete
887 if (blk_queue_flushing(drive->queue)) {
889 * small race where queue could get replugged during
890 * the 3-request flush cycle, just yank the plug since
891 * we want it to finish asap
893 blk_remove_plug(drive->queue);
898 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
899 && !elv_queue_empty(drive->queue)) {
901 || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
902 || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
904 if (!blk_queue_plugged(drive->queue))
908 } while ((drive = drive->next) != hwgroup->drive);
909 if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
910 long t = (signed long)(WAKEUP(best) - jiffies);
911 if (t >= WAIT_MIN_SLEEP) {
913 * We *may* have some time to spare, but first let's see if
914 * someone can potentially benefit from our nice mood today..
919 && time_before(jiffies - best->service_time, WAKEUP(drive))
920 && time_before(WAKEUP(drive), jiffies + t))
922 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
925 } while ((drive = drive->next) != best);
932 * Issue a new request to a drive from hwgroup
933 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
935 * A hwgroup is a serialized group of IDE interfaces. Usually there is
936 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
937 * may have both interfaces in a single hwgroup to "serialize" access.
938 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
939 * together into one hwgroup for serialized access.
941 * Note also that several hwgroups can end up sharing a single IRQ,
942 * possibly along with many other devices. This is especially common in
943 * PCI-based systems with off-board IDE controller cards.
945 * The IDE driver uses the single global ide_lock spinlock to protect
946 * access to the request queues, and to protect the hwgroup->busy flag.
948 * The first thread into the driver for a particular hwgroup sets the
949 * hwgroup->busy flag to indicate that this hwgroup is now active,
950 * and then initiates processing of the top request from the request queue.
952 * Other threads attempting entry notice the busy setting, and will simply
953 * queue their new requests and exit immediately. Note that hwgroup->busy
954 * remains set even when the driver is merely awaiting the next interrupt.
955 * Thus, the meaning is "this hwgroup is busy processing a request".
957 * When processing of a request completes, the completing thread or IRQ-handler
958 * will start the next request from the queue. If no more work remains,
959 * the driver will clear the hwgroup->busy flag and exit.
961 * The ide_lock (spinlock) is used to protect all access to the
962 * hwgroup->busy flag, but is otherwise not needed for most processing in
963 * the driver. This makes the driver much more friendlier to shared IRQs
964 * than previous designs, while remaining 100% (?) SMP safe and capable.
966 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
971 ide_startstop_t startstop;
974 /* for atari only: POSSIBLY BROKEN HERE(?) */
975 ide_get_lock(ide_intr, hwgroup);
977 /* caller must own ide_lock */
978 BUG_ON(!irqs_disabled());
980 while (!hwgroup->busy) {
982 drive = choose_drive(hwgroup);
985 unsigned long sleep = 0; /* shut up, gcc */
987 drive = hwgroup->drive;
989 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
991 sleep = drive->sleep;
993 } while ((drive = drive->next) != hwgroup->drive);
996 * Take a short snooze, and then wake up this hwgroup again.
997 * This gives other hwgroups on the same a chance to
998 * play fairly with us, just in case there are big differences
999 * in relative throughputs.. don't want to hog the cpu too much.
1001 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1002 sleep = jiffies + WAIT_MIN_SLEEP;
1004 if (timer_pending(&hwgroup->timer))
1005 printk(KERN_CRIT "ide_set_handler: timer already active\n");
1007 /* so that ide_timer_expiry knows what to do */
1008 hwgroup->sleeping = 1;
1009 hwgroup->req_gen_timer = hwgroup->req_gen;
1010 mod_timer(&hwgroup->timer, sleep);
1011 /* we purposely leave hwgroup->busy==1
1014 /* Ugly, but how can we sleep for the lock
1015 * otherwise? perhaps from tq_disk?
1018 /* for atari only */
1023 /* no more work for this hwgroup (for now) */
1028 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1030 * set nIEN for previous hwif, drives in the
1031 * quirk_list may not like intr setups/cleanups
1033 if (drive->quirk_list != 1)
1034 hwif->tp_ops->set_irq(hwif, 0);
1036 hwgroup->hwif = hwif;
1037 hwgroup->drive = drive;
1038 drive->sleeping = 0;
1039 drive->service_start = jiffies;
1041 if (blk_queue_plugged(drive->queue)) {
1042 printk(KERN_ERR "ide: huh? queue was plugged!\n");
1047 * we know that the queue isn't empty, but this can happen
1048 * if the q->prep_rq_fn() decides to kill a request
1050 rq = elv_next_request(drive->queue);
1057 * Sanity: don't accept a request that isn't a PM request
1058 * if we are currently power managed. This is very important as
1059 * blk_stop_queue() doesn't prevent the elv_next_request()
1060 * above to return us whatever is in the queue. Since we call
1061 * ide_do_request() ourselves, we end up taking requests while
1062 * the queue is blocked...
1064 * We let requests forced at head of queue with ide-preempt
1065 * though. I hope that doesn't happen too much, hopefully not
1066 * unless the subdriver triggers such a thing in its own PM
1069 * We count how many times we loop here to make sure we service
1070 * all drives in the hwgroup without looping for ever
1072 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1073 drive = drive->next ? drive->next : hwgroup->drive;
1074 if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1076 /* We clear busy, there should be no pending ATA command at this point. */
1084 * Some systems have trouble with IDE IRQs arriving while
1085 * the driver is still setting things up. So, here we disable
1086 * the IRQ used by this interface while the request is being started.
1087 * This may look bad at first, but pretty much the same thing
1088 * happens anyway when any interrupt comes in, IDE or otherwise
1089 * -- the kernel masks the IRQ while it is being handled.
1091 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1092 disable_irq_nosync(hwif->irq);
1093 spin_unlock(&ide_lock);
1094 local_irq_enable_in_hardirq();
1095 /* allow other IRQs while we start this request */
1096 startstop = start_request(drive, rq);
1097 spin_lock_irq(&ide_lock);
1098 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1099 enable_irq(hwif->irq);
1100 if (startstop == ide_stopped)
1106 * Passes the stuff to ide_do_request
1108 void do_ide_request(struct request_queue *q)
1110 ide_drive_t *drive = q->queuedata;
1112 ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1116 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1117 * retry the current request in pio mode instead of risking tossing it
1120 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1122 ide_hwif_t *hwif = HWIF(drive);
1124 ide_startstop_t ret = ide_stopped;
1127 * end current dma transaction
1131 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1132 (void)hwif->dma_ops->dma_end(drive);
1133 ret = ide_error(drive, "dma timeout error",
1134 hwif->tp_ops->read_status(hwif));
1136 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1137 hwif->dma_ops->dma_timeout(drive);
1141 * disable dma for now, but remember that we did so because of
1142 * a timeout -- we'll reenable after we finish this next request
1143 * (or rather the first chunk of it) in pio.
1146 drive->state = DMA_PIO_RETRY;
1147 ide_dma_off_quietly(drive);
1150 * un-busy drive etc (hwgroup->busy is cleared on return) and
1151 * make sure request is sane
1153 rq = HWGROUP(drive)->rq;
1158 HWGROUP(drive)->rq = NULL;
1165 rq->sector = rq->bio->bi_sector;
1166 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1167 rq->hard_cur_sectors = rq->current_nr_sectors;
1168 rq->buffer = bio_data(rq->bio);
1174 * ide_timer_expiry - handle lack of an IDE interrupt
1175 * @data: timer callback magic (hwgroup)
1177 * An IDE command has timed out before the expected drive return
1178 * occurred. At this point we attempt to clean up the current
1179 * mess. If the current handler includes an expiry handler then
1180 * we invoke the expiry handler, and providing it is happy the
1181 * work is done. If that fails we apply generic recovery rules
1182 * invoking the handler and checking the drive DMA status. We
1183 * have an excessively incestuous relationship with the DMA
1184 * logic that wants cleaning up.
1187 void ide_timer_expiry (unsigned long data)
1189 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1190 ide_handler_t *handler;
1191 ide_expiry_t *expiry;
1192 unsigned long flags;
1193 unsigned long wait = -1;
1195 spin_lock_irqsave(&ide_lock, flags);
1197 if (((handler = hwgroup->handler) == NULL) ||
1198 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1200 * Either a marginal timeout occurred
1201 * (got the interrupt just as timer expired),
1202 * or we were "sleeping" to give other devices a chance.
1203 * Either way, we don't really want to complain about anything.
1205 if (hwgroup->sleeping) {
1206 hwgroup->sleeping = 0;
1210 ide_drive_t *drive = hwgroup->drive;
1212 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1213 hwgroup->handler = NULL;
1216 ide_startstop_t startstop = ide_stopped;
1217 if (!hwgroup->busy) {
1218 hwgroup->busy = 1; /* paranoia */
1219 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1221 if ((expiry = hwgroup->expiry) != NULL) {
1223 if ((wait = expiry(drive)) > 0) {
1225 hwgroup->timer.expires = jiffies + wait;
1226 hwgroup->req_gen_timer = hwgroup->req_gen;
1227 add_timer(&hwgroup->timer);
1228 spin_unlock_irqrestore(&ide_lock, flags);
1232 hwgroup->handler = NULL;
1234 * We need to simulate a real interrupt when invoking
1235 * the handler() function, which means we need to
1236 * globally mask the specific IRQ:
1238 spin_unlock(&ide_lock);
1240 /* disable_irq_nosync ?? */
1241 disable_irq(hwif->irq);
1243 * as if we were handling an interrupt */
1244 local_irq_disable();
1245 if (hwgroup->polling) {
1246 startstop = handler(drive);
1247 } else if (drive_is_ready(drive)) {
1248 if (drive->waiting_for_dma)
1249 hwif->dma_ops->dma_lost_irq(drive);
1250 (void)ide_ack_intr(hwif);
1251 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1252 startstop = handler(drive);
1254 if (drive->waiting_for_dma) {
1255 startstop = ide_dma_timeout_retry(drive, wait);
1258 ide_error(drive, "irq timeout",
1259 hwif->tp_ops->read_status(hwif));
1261 drive->service_time = jiffies - drive->service_start;
1262 spin_lock_irq(&ide_lock);
1263 enable_irq(hwif->irq);
1264 if (startstop == ide_stopped)
1268 ide_do_request(hwgroup, IDE_NO_IRQ);
1269 spin_unlock_irqrestore(&ide_lock, flags);
1273 * unexpected_intr - handle an unexpected IDE interrupt
1274 * @irq: interrupt line
1275 * @hwgroup: hwgroup being processed
1277 * There's nothing really useful we can do with an unexpected interrupt,
1278 * other than reading the status register (to clear it), and logging it.
1279 * There should be no way that an irq can happen before we're ready for it,
1280 * so we needn't worry much about losing an "important" interrupt here.
1282 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1283 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1284 * looks "good", we just ignore the interrupt completely.
1286 * This routine assumes __cli() is in effect when called.
1288 * If an unexpected interrupt happens on irq15 while we are handling irq14
1289 * and if the two interfaces are "serialized" (CMD640), then it looks like
1290 * we could screw up by interfering with a new request being set up for
1293 * In reality, this is a non-issue. The new command is not sent unless
1294 * the drive is ready to accept one, in which case we know the drive is
1295 * not trying to interrupt us. And ide_set_handler() is always invoked
1296 * before completing the issuance of any new drive command, so we will not
1297 * be accidentally invoked as a result of any valid command completion
1300 * Note that we must walk the entire hwgroup here. We know which hwif
1301 * is doing the current command, but we don't know which hwif burped
1305 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1308 ide_hwif_t *hwif = hwgroup->hwif;
1311 * handle the unexpected interrupt
1314 if (hwif->irq == irq) {
1315 stat = hwif->tp_ops->read_status(hwif);
1317 if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1318 /* Try to not flood the console with msgs */
1319 static unsigned long last_msgtime, count;
1321 if (time_after(jiffies, last_msgtime + HZ)) {
1322 last_msgtime = jiffies;
1323 printk(KERN_ERR "%s%s: unexpected interrupt, "
1324 "status=0x%02x, count=%ld\n",
1326 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1330 } while ((hwif = hwif->next) != hwgroup->hwif);
1334 * ide_intr - default IDE interrupt handler
1335 * @irq: interrupt number
1336 * @dev_id: hwif group
1337 * @regs: unused weirdness from the kernel irq layer
1339 * This is the default IRQ handler for the IDE layer. You should
1340 * not need to override it. If you do be aware it is subtle in
1343 * hwgroup->hwif is the interface in the group currently performing
1344 * a command. hwgroup->drive is the drive and hwgroup->handler is
1345 * the IRQ handler to call. As we issue a command the handlers
1346 * step through multiple states, reassigning the handler to the
1347 * next step in the process. Unlike a smart SCSI controller IDE
1348 * expects the main processor to sequence the various transfer
1349 * stages. We also manage a poll timer to catch up with most
1350 * timeout situations. There are still a few where the handlers
1351 * don't ever decide to give up.
1353 * The handler eventually returns ide_stopped to indicate the
1354 * request completed. At this point we issue the next request
1355 * on the hwgroup and the process begins again.
1358 irqreturn_t ide_intr (int irq, void *dev_id)
1360 unsigned long flags;
1361 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1364 ide_handler_t *handler;
1365 ide_startstop_t startstop;
1367 spin_lock_irqsave(&ide_lock, flags);
1368 hwif = hwgroup->hwif;
1370 if (!ide_ack_intr(hwif)) {
1371 spin_unlock_irqrestore(&ide_lock, flags);
1375 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1377 * Not expecting an interrupt from this drive.
1378 * That means this could be:
1379 * (1) an interrupt from another PCI device
1380 * sharing the same PCI INT# as us.
1381 * or (2) a drive just entered sleep or standby mode,
1382 * and is interrupting to let us know.
1383 * or (3) a spurious interrupt of unknown origin.
1385 * For PCI, we cannot tell the difference,
1386 * so in that case we just ignore it and hope it goes away.
1388 * FIXME: unexpected_intr should be hwif-> then we can
1389 * remove all the ifdef PCI crap
1391 #ifdef CONFIG_BLK_DEV_IDEPCI
1392 if (hwif->chipset != ide_pci)
1393 #endif /* CONFIG_BLK_DEV_IDEPCI */
1396 * Probably not a shared PCI interrupt,
1397 * so we can safely try to do something about it:
1399 unexpected_intr(irq, hwgroup);
1400 #ifdef CONFIG_BLK_DEV_IDEPCI
1403 * Whack the status register, just in case
1404 * we have a leftover pending IRQ.
1406 (void)hwif->tp_ops->read_status(hwif);
1407 #endif /* CONFIG_BLK_DEV_IDEPCI */
1409 spin_unlock_irqrestore(&ide_lock, flags);
1412 drive = hwgroup->drive;
1415 * This should NEVER happen, and there isn't much
1416 * we could do about it here.
1418 * [Note - this can occur if the drive is hot unplugged]
1420 spin_unlock_irqrestore(&ide_lock, flags);
1423 if (!drive_is_ready(drive)) {
1425 * This happens regularly when we share a PCI IRQ with
1426 * another device. Unfortunately, it can also happen
1427 * with some buggy drives that trigger the IRQ before
1428 * their status register is up to date. Hopefully we have
1429 * enough advance overhead that the latter isn't a problem.
1431 spin_unlock_irqrestore(&ide_lock, flags);
1434 if (!hwgroup->busy) {
1435 hwgroup->busy = 1; /* paranoia */
1436 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1438 hwgroup->handler = NULL;
1440 del_timer(&hwgroup->timer);
1441 spin_unlock(&ide_lock);
1443 /* Some controllers might set DMA INTR no matter DMA or PIO;
1444 * bmdma status might need to be cleared even for
1445 * PIO interrupts to prevent spurious/lost irq.
1447 if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1448 /* ide_dma_end() needs bmdma status for error checking.
1449 * So, skip clearing bmdma status here and leave it
1450 * to ide_dma_end() if this is dma interrupt.
1452 hwif->ide_dma_clear_irq(drive);
1455 local_irq_enable_in_hardirq();
1456 /* service this interrupt, may set handler for next interrupt */
1457 startstop = handler(drive);
1458 spin_lock_irq(&ide_lock);
1461 * Note that handler() may have set things up for another
1462 * interrupt to occur soon, but it cannot happen until
1463 * we exit from this routine, because it will be the
1464 * same irq as is currently being serviced here, and Linux
1465 * won't allow another of the same (on any CPU) until we return.
1467 drive->service_time = jiffies - drive->service_start;
1468 if (startstop == ide_stopped) {
1469 if (hwgroup->handler == NULL) { /* paranoia */
1471 ide_do_request(hwgroup, hwif->irq);
1473 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1474 "on exit\n", drive->name);
1477 spin_unlock_irqrestore(&ide_lock, flags);
1482 * ide_do_drive_cmd - issue IDE special command
1483 * @drive: device to issue command
1484 * @rq: request to issue
1486 * This function issues a special IDE device request
1487 * onto the request queue.
1489 * the rq is queued at the head of the request queue, displacing
1490 * the currently-being-processed request and this function
1491 * returns immediately without waiting for the new rq to be
1492 * completed. This is VERY DANGEROUS, and is intended for
1493 * careful use by the ATAPI tape/cdrom driver code.
1496 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1498 unsigned long flags;
1499 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1501 spin_lock_irqsave(&ide_lock, flags);
1503 __elv_add_request(drive->queue, rq, ELEVATOR_INSERT_FRONT, 1);
1504 __generic_unplug_device(drive->queue);
1505 spin_unlock_irqrestore(&ide_lock, flags);
1508 EXPORT_SYMBOL(ide_do_drive_cmd);
1510 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1512 ide_hwif_t *hwif = drive->hwif;
1515 memset(&task, 0, sizeof(task));
1516 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1517 IDE_TFLAG_OUT_FEATURE | tf_flags;
1518 task.tf.feature = dma; /* Use PIO/DMA */
1519 task.tf.lbam = bcount & 0xff;
1520 task.tf.lbah = (bcount >> 8) & 0xff;
1522 ide_tf_dump(drive->name, &task.tf);
1523 hwif->tp_ops->set_irq(hwif, 1);
1524 SELECT_MASK(drive, 0);
1525 hwif->tp_ops->tf_load(drive, &task);
1528 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1530 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1532 ide_hwif_t *hwif = drive->hwif;
1537 hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1539 hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1543 EXPORT_SYMBOL_GPL(ide_pad_transfer);