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/config.h>
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/string.h>
31 #include <linux/kernel.h>
32 #include <linux/timer.h>
34 #include <linux/interrupt.h>
35 #include <linux/major.h>
36 #include <linux/errno.h>
37 #include <linux/genhd.h>
38 #include <linux/blkpg.h>
39 #include <linux/slab.h>
40 #include <linux/init.h>
41 #include <linux/pci.h>
42 #include <linux/delay.h>
43 #include <linux/ide.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>
51 #include <asm/byteorder.h>
53 #include <asm/uaccess.h>
55 #include <asm/bitops.h>
58 * ide_end_request - complete an IDE I/O
59 * @drive: IDE device for the I/O
61 * @nr_sectors: number of sectors completed
63 * This is our end_request wrapper function. We complete the I/O
64 * update random number input and dequeue the request, which if
65 * it was tagged may be out of order.
68 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
74 spin_lock_irqsave(&ide_lock, flags);
75 rq = HWGROUP(drive)->rq;
77 BUG_ON(!(rq->flags & REQ_STARTED));
80 nr_sectors = rq->hard_cur_sectors;
83 * if failfast is set on a request, override number of sectors and
84 * complete the whole request right now
86 if (blk_noretry_request(rq) && !uptodate)
87 nr_sectors = rq->hard_nr_sectors;
90 * decide whether to reenable DMA -- 3 is a random magic for now,
91 * if we DMA timeout more than 3 times, just stay in PIO
93 if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
95 HWGROUP(drive)->hwif->ide_dma_on(drive);
98 if (!end_that_request_first(rq, uptodate, nr_sectors)) {
99 add_disk_randomness(rq->rq_disk);
100 blkdev_dequeue_request(rq);
101 HWGROUP(drive)->rq = NULL;
102 end_that_request_last(rq);
105 spin_unlock_irqrestore(&ide_lock, flags);
109 EXPORT_SYMBOL(ide_end_request);
112 * ide_complete_pm_request - end the current Power Management request
113 * @drive: target drive
116 * This function cleans up the current PM request and stops the queue
119 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
124 printk("%s: completing PM request, %s\n", drive->name,
125 blk_pm_suspend_request(rq) ? "suspend" : "resume");
127 spin_lock_irqsave(&ide_lock, flags);
128 if (blk_pm_suspend_request(rq)) {
129 blk_stop_queue(drive->queue);
132 blk_start_queue(drive->queue);
134 blkdev_dequeue_request(rq);
135 HWGROUP(drive)->rq = NULL;
136 end_that_request_last(rq);
137 spin_unlock_irqrestore(&ide_lock, flags);
141 * ide_end_drive_cmd - end an explicit drive command
146 * Clean up after success/failure of an explicit drive command.
147 * These get thrown onto the queue so they are synchronized with
148 * real I/O operations on the drive.
150 * In LBA48 mode we have to read the register set twice to get
151 * all the extra information out.
154 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
156 ide_hwif_t *hwif = HWIF(drive);
160 spin_lock_irqsave(&ide_lock, flags);
161 rq = HWGROUP(drive)->rq;
162 spin_unlock_irqrestore(&ide_lock, flags);
164 if (rq->flags & REQ_DRIVE_CMD) {
165 u8 *args = (u8 *) rq->buffer;
167 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
172 args[2] = hwif->INB(IDE_NSECTOR_REG);
174 } else if (rq->flags & REQ_DRIVE_TASK) {
175 u8 *args = (u8 *) rq->buffer;
177 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
182 args[2] = hwif->INB(IDE_NSECTOR_REG);
183 args[3] = hwif->INB(IDE_SECTOR_REG);
184 args[4] = hwif->INB(IDE_LCYL_REG);
185 args[5] = hwif->INB(IDE_HCYL_REG);
186 args[6] = hwif->INB(IDE_SELECT_REG);
188 } else if (rq->flags & REQ_DRIVE_TASKFILE) {
189 ide_task_t *args = (ide_task_t *) rq->special;
191 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
194 if (args->tf_in_flags.b.data) {
195 u16 data = hwif->INW(IDE_DATA_REG);
196 args->tfRegister[IDE_DATA_OFFSET] = (data) & 0xFF;
197 args->hobRegister[IDE_DATA_OFFSET] = (data >> 8) & 0xFF;
199 args->tfRegister[IDE_ERROR_OFFSET] = err;
200 args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
201 args->tfRegister[IDE_SECTOR_OFFSET] = hwif->INB(IDE_SECTOR_REG);
202 args->tfRegister[IDE_LCYL_OFFSET] = hwif->INB(IDE_LCYL_REG);
203 args->tfRegister[IDE_HCYL_OFFSET] = hwif->INB(IDE_HCYL_REG);
204 args->tfRegister[IDE_SELECT_OFFSET] = hwif->INB(IDE_SELECT_REG);
205 args->tfRegister[IDE_STATUS_OFFSET] = stat;
207 if (drive->addressing == 1) {
208 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
209 args->hobRegister[IDE_FEATURE_OFFSET] = hwif->INB(IDE_FEATURE_REG);
210 args->hobRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
211 args->hobRegister[IDE_SECTOR_OFFSET] = hwif->INB(IDE_SECTOR_REG);
212 args->hobRegister[IDE_LCYL_OFFSET] = hwif->INB(IDE_LCYL_REG);
213 args->hobRegister[IDE_HCYL_OFFSET] = hwif->INB(IDE_HCYL_REG);
216 } else if (blk_pm_request(rq)) {
218 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
219 drive->name, rq->pm->pm_step, stat, err);
221 DRIVER(drive)->complete_power_step(drive, rq, stat, err);
222 if (rq->pm->pm_step == ide_pm_state_completed)
223 ide_complete_pm_request(drive, rq);
227 spin_lock_irqsave(&ide_lock, flags);
228 blkdev_dequeue_request(rq);
229 HWGROUP(drive)->rq = NULL;
230 end_that_request_last(rq);
231 spin_unlock_irqrestore(&ide_lock, flags);
234 EXPORT_SYMBOL(ide_end_drive_cmd);
237 * try_to_flush_leftover_data - flush junk
238 * @drive: drive to flush
240 * try_to_flush_leftover_data() is invoked in response to a drive
241 * unexpectedly having its DRQ_STAT bit set. As an alternative to
242 * resetting the drive, this routine tries to clear the condition
243 * by read a sector's worth of data from the drive. Of course,
244 * this may not help if the drive is *waiting* for data from *us*.
246 void try_to_flush_leftover_data (ide_drive_t *drive)
248 int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
250 if (drive->media != ide_disk)
254 u32 wcount = (i > 16) ? 16 : i;
257 HWIF(drive)->ata_input_data(drive, buffer, wcount);
261 EXPORT_SYMBOL(try_to_flush_leftover_data);
264 * FIXME Add an ATAPI error
268 * ide_error - handle an error on the IDE
269 * @drive: drive the error occurred on
270 * @msg: message to report
273 * ide_error() takes action based on the error returned by the drive.
274 * For normal I/O that may well include retries. We deal with
275 * both new-style (taskfile) and old style command handling here.
276 * In the case of taskfile command handling there is work left to
280 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
286 err = ide_dump_status(drive, msg, stat);
287 if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
291 /* retry only "normal" I/O: */
292 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
294 ide_end_drive_cmd(drive, stat, err);
297 if (rq->flags & REQ_DRIVE_TASKFILE) {
299 ide_end_drive_cmd(drive, stat, err);
303 if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
304 /* other bits are useless when BUSY */
305 rq->errors |= ERROR_RESET;
307 if (drive->media != ide_disk)
310 if (stat & ERR_STAT) {
311 /* err has different meaning on cdrom and tape */
312 if (err == ABRT_ERR) {
313 if (drive->select.b.lba &&
314 (hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY))
315 /* some newer drives don't
316 * support WIN_SPECIFY
319 } else if ((err & BAD_CRC) == BAD_CRC) {
321 /* UDMA crc error -- just retry the operation */
322 } else if (err & (BBD_ERR | ECC_ERR)) {
323 /* retries won't help these */
324 rq->errors = ERROR_MAX;
325 } else if (err & TRK0_ERR) {
326 /* help it find track zero */
327 rq->errors |= ERROR_RECAL;
331 if ((stat & DRQ_STAT) && rq_data_dir(rq) != WRITE)
332 try_to_flush_leftover_data(drive);
334 if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT)) {
336 hwif->OUTB(WIN_IDLEIMMEDIATE,IDE_COMMAND_REG);
338 if (rq->errors >= ERROR_MAX) {
339 DRIVER(drive)->end_request(drive, 0, 0);
341 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
343 return ide_do_reset(drive);
345 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
346 drive->special.b.recalibrate = 1;
352 EXPORT_SYMBOL(ide_error);
355 * ide_abort - abort pending IDE operatins
356 * @drive: drive the error occurred on
357 * @msg: message to report
359 * ide_abort kills and cleans up when we are about to do a
360 * host initiated reset on active commands. Longer term we
361 * want handlers to have sensible abort handling themselves
363 * This differs fundamentally from ide_error because in
364 * this case the command is doing just fine when we
368 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
373 if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
377 /* retry only "normal" I/O: */
378 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
380 ide_end_drive_cmd(drive, BUSY_STAT, 0);
383 if (rq->flags & REQ_DRIVE_TASKFILE) {
385 ide_end_drive_cmd(drive, BUSY_STAT, 0);
389 rq->errors |= ERROR_RESET;
390 DRIVER(drive)->end_request(drive, 0, 0);
394 EXPORT_SYMBOL(ide_abort);
397 * ide_cmd - issue a simple drive command
398 * @drive: drive the command is for
400 * @nsect: sector byte
401 * @handler: handler for the command completion
403 * Issue a simple drive command with interrupts.
404 * The drive must be selected beforehand.
407 void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect, ide_handler_t *handler)
409 ide_hwif_t *hwif = HWIF(drive);
411 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
412 SELECT_MASK(drive,0);
413 hwif->OUTB(nsect,IDE_NSECTOR_REG);
414 ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
417 EXPORT_SYMBOL(ide_cmd);
420 * drive_cmd_intr - drive command completion interrupt
421 * @drive: drive the completion interrupt occurred on
423 * drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
424 * We do any necessary daya reading and then wait for the drive to
425 * go non busy. At that point we may read the error data and complete
429 ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
431 struct request *rq = HWGROUP(drive)->rq;
432 ide_hwif_t *hwif = HWIF(drive);
433 u8 *args = (u8 *) rq->buffer;
434 u8 stat = hwif->INB(IDE_STATUS_REG);
438 if ((stat & DRQ_STAT) && args && args[3]) {
439 u8 io_32bit = drive->io_32bit;
441 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
442 drive->io_32bit = io_32bit;
443 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
447 if (!OK_STAT(stat, READY_STAT, BAD_STAT) && DRIVER(drive) != NULL)
448 return DRIVER(drive)->error(drive, "drive_cmd", stat);
449 /* calls ide_end_drive_cmd */
450 ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
454 EXPORT_SYMBOL(drive_cmd_intr);
457 * do_special - issue some special commands
458 * @drive: drive the command is for
460 * do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
461 * commands to a drive. It used to do much more, but has been scaled
465 ide_startstop_t do_special (ide_drive_t *drive)
467 special_t *s = &drive->special;
470 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
474 if (HWIF(drive)->tuneproc != NULL)
475 HWIF(drive)->tuneproc(drive, drive->tune_req);
479 return DRIVER(drive)->special(drive);
482 EXPORT_SYMBOL(do_special);
485 * execute_drive_command - issue special drive command
486 * @drive: the drive to issue th command on
487 * @rq: the request structure holding the command
489 * execute_drive_cmd() issues a special drive command, usually
490 * initiated by ioctl() from the external hdparm program. The
491 * command can be a drive command, drive task or taskfile
492 * operation. Weirdly you can call it with NULL to wait for
493 * all commands to finish. Don't do this as that is due to change
496 ide_startstop_t execute_drive_cmd (ide_drive_t *drive, struct request *rq)
498 ide_hwif_t *hwif = HWIF(drive);
499 if (rq->flags & REQ_DRIVE_TASKFILE) {
500 ide_task_t *args = rq->special;
505 if (args->tf_out_flags.all != 0)
506 return flagged_taskfile(drive, args);
507 return do_rw_taskfile(drive, args);
508 } else if (rq->flags & REQ_DRIVE_TASK) {
509 u8 *args = rq->buffer;
515 printk("%s: DRIVE_TASK_CMD ", drive->name);
516 printk("cmd=0x%02x ", args[0]);
517 printk("fr=0x%02x ", args[1]);
518 printk("ns=0x%02x ", args[2]);
519 printk("sc=0x%02x ", args[3]);
520 printk("lcyl=0x%02x ", args[4]);
521 printk("hcyl=0x%02x ", args[5]);
522 printk("sel=0x%02x\n", args[6]);
524 hwif->OUTB(args[1], IDE_FEATURE_REG);
525 hwif->OUTB(args[3], IDE_SECTOR_REG);
526 hwif->OUTB(args[4], IDE_LCYL_REG);
527 hwif->OUTB(args[5], IDE_HCYL_REG);
528 sel = (args[6] & ~0x10);
529 if (drive->select.b.unit)
531 hwif->OUTB(sel, IDE_SELECT_REG);
532 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
534 } else if (rq->flags & REQ_DRIVE_CMD) {
535 u8 *args = rq->buffer;
540 printk("%s: DRIVE_CMD ", drive->name);
541 printk("cmd=0x%02x ", args[0]);
542 printk("sc=0x%02x ", args[1]);
543 printk("fr=0x%02x ", args[2]);
544 printk("xx=0x%02x\n", args[3]);
546 if (args[0] == WIN_SMART) {
547 hwif->OUTB(0x4f, IDE_LCYL_REG);
548 hwif->OUTB(0xc2, IDE_HCYL_REG);
549 hwif->OUTB(args[2],IDE_FEATURE_REG);
550 hwif->OUTB(args[1],IDE_SECTOR_REG);
551 ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
554 hwif->OUTB(args[2],IDE_FEATURE_REG);
555 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
561 * NULL is actually a valid way of waiting for
562 * all current requests to be flushed from the queue.
565 printk("%s: DRIVE_CMD (null)\n", drive->name);
567 ide_end_drive_cmd(drive,
568 hwif->INB(IDE_STATUS_REG),
569 hwif->INB(IDE_ERROR_REG));
573 EXPORT_SYMBOL(execute_drive_cmd);
576 * start_request - start of I/O and command issuing for IDE
578 * start_request() initiates handling of a new I/O request. It
579 * accepts commands and I/O (read/write) requests. It also does
580 * the final remapping for weird stuff like EZDrive. Once
581 * device mapper can work sector level the EZDrive stuff can go away
583 * FIXME: this function needs a rename
586 ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
588 ide_startstop_t startstop;
591 BUG_ON(!(rq->flags & REQ_STARTED));
594 printk("%s: start_request: current=0x%08lx\n",
595 HWIF(drive)->name, (unsigned long) rq);
598 /* bail early if we've exceeded max_failures */
599 if (drive->max_failures && (drive->failures > drive->max_failures)) {
604 * bail early if we've sent a device to sleep, however how to wake
605 * this needs to be a masked flag. FIXME for proper operations.
607 if (drive->suspend_reset)
611 if (blk_fs_request(rq) &&
612 (drive->media == ide_disk || drive->media == ide_floppy)) {
613 block += drive->sect0;
615 /* Yecch - this will shift the entire interval,
616 possibly killing some innocent following sector */
617 if (block == 0 && drive->remap_0_to_1 == 1)
618 block = 1; /* redirect MBR access to EZ-Drive partn table */
620 if (blk_pm_suspend_request(rq) &&
621 rq->pm->pm_step == ide_pm_state_start_suspend)
622 /* Mark drive blocked when starting the suspend sequence. */
624 else if (blk_pm_resume_request(rq) &&
625 rq->pm->pm_step == ide_pm_state_start_resume) {
627 * The first thing we do on wakeup is to wait for BSY bit to
628 * go away (with a looong timeout) as a drive on this hwif may
629 * just be POSTing itself.
630 * We do that before even selecting as the "other" device on
631 * the bus may be broken enough to walk on our toes at this
636 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
638 rc = ide_wait_not_busy(HWIF(drive), 35000);
640 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
642 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
643 rc = ide_wait_not_busy(HWIF(drive), 10000);
645 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
649 if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
650 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
653 if (!drive->special.all) {
654 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
655 return execute_drive_cmd(drive, rq);
656 else if (rq->flags & REQ_DRIVE_TASKFILE)
657 return execute_drive_cmd(drive, rq);
658 else if (blk_pm_request(rq)) {
660 printk("%s: start_power_step(step: %d)\n",
661 drive->name, rq->pm->pm_step);
663 startstop = DRIVER(drive)->start_power_step(drive, rq);
664 if (startstop == ide_stopped &&
665 rq->pm->pm_step == ide_pm_state_completed)
666 ide_complete_pm_request(drive, rq);
669 return (DRIVER(drive)->do_request(drive, rq, block));
671 return do_special(drive);
673 DRIVER(drive)->end_request(drive, 0, 0);
677 EXPORT_SYMBOL(start_request);
680 * ide_stall_queue - pause an IDE device
681 * @drive: drive to stall
682 * @timeout: time to stall for (jiffies)
684 * ide_stall_queue() can be used by a drive to give excess bandwidth back
685 * to the hwgroup by sleeping for timeout jiffies.
688 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
690 if (timeout > WAIT_WORSTCASE)
691 timeout = WAIT_WORSTCASE;
692 drive->sleep = timeout + jiffies;
695 EXPORT_SYMBOL(ide_stall_queue);
697 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
700 * choose_drive - select a drive to service
701 * @hwgroup: hardware group to select on
703 * choose_drive() selects the next drive which will be serviced.
704 * This is necessary because the IDE layer can't issue commands
705 * to both drives on the same cable, unlike SCSI.
708 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
710 ide_drive_t *drive, *best;
714 drive = hwgroup->drive;
716 if ((!drive->sleep || time_after_eq(jiffies, drive->sleep))
717 && !elv_queue_empty(drive->queue)) {
719 || (drive->sleep && (!best->sleep || 0 < (signed long)(best->sleep - drive->sleep)))
720 || (!best->sleep && 0 < (signed long)(WAKEUP(best) - WAKEUP(drive))))
722 if (!blk_queue_plugged(drive->queue))
726 } while ((drive = drive->next) != hwgroup->drive);
727 if (best && best->nice1 && !best->sleep && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
728 long t = (signed long)(WAKEUP(best) - jiffies);
729 if (t >= WAIT_MIN_SLEEP) {
731 * We *may* have some time to spare, but first let's see if
732 * someone can potentially benefit from our nice mood today..
737 /* FIXME: use time_before */
738 && 0 < (signed long)(WAKEUP(drive) - (jiffies - best->service_time))
739 && 0 < (signed long)((jiffies + t) - WAKEUP(drive)))
741 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
744 } while ((drive = drive->next) != best);
751 * Issue a new request to a drive from hwgroup
752 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
754 * A hwgroup is a serialized group of IDE interfaces. Usually there is
755 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
756 * may have both interfaces in a single hwgroup to "serialize" access.
757 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
758 * together into one hwgroup for serialized access.
760 * Note also that several hwgroups can end up sharing a single IRQ,
761 * possibly along with many other devices. This is especially common in
762 * PCI-based systems with off-board IDE controller cards.
764 * The IDE driver uses the single global ide_lock spinlock to protect
765 * access to the request queues, and to protect the hwgroup->busy flag.
767 * The first thread into the driver for a particular hwgroup sets the
768 * hwgroup->busy flag to indicate that this hwgroup is now active,
769 * and then initiates processing of the top request from the request queue.
771 * Other threads attempting entry notice the busy setting, and will simply
772 * queue their new requests and exit immediately. Note that hwgroup->busy
773 * remains set even when the driver is merely awaiting the next interrupt.
774 * Thus, the meaning is "this hwgroup is busy processing a request".
776 * When processing of a request completes, the completing thread or IRQ-handler
777 * will start the next request from the queue. If no more work remains,
778 * the driver will clear the hwgroup->busy flag and exit.
780 * The ide_lock (spinlock) is used to protect all access to the
781 * hwgroup->busy flag, but is otherwise not needed for most processing in
782 * the driver. This makes the driver much more friendlier to shared IRQs
783 * than previous designs, while remaining 100% (?) SMP safe and capable.
785 /* --BenH: made non-static as ide-pmac.c uses it to kick the hwgroup back
786 * into life on wakeup from machine sleep.
788 void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
793 ide_startstop_t startstop;
795 /* for atari only: POSSIBLY BROKEN HERE(?) */
796 ide_get_lock(ide_intr, hwgroup);
798 /* caller must own ide_lock */
799 BUG_ON(!irqs_disabled());
801 while (!hwgroup->busy) {
803 drive = choose_drive(hwgroup);
805 unsigned long sleep = 0;
807 drive = hwgroup->drive;
809 if (drive->sleep && (!sleep || 0 < (signed long)(sleep - drive->sleep)))
810 sleep = drive->sleep;
811 } while ((drive = drive->next) != hwgroup->drive);
814 * Take a short snooze, and then wake up this hwgroup again.
815 * This gives other hwgroups on the same a chance to
816 * play fairly with us, just in case there are big differences
817 * in relative throughputs.. don't want to hog the cpu too much.
819 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
820 sleep = jiffies + WAIT_MIN_SLEEP;
822 if (timer_pending(&hwgroup->timer))
823 printk(KERN_CRIT "ide_set_handler: timer already active\n");
825 /* so that ide_timer_expiry knows what to do */
826 hwgroup->sleeping = 1;
827 mod_timer(&hwgroup->timer, sleep);
828 /* we purposely leave hwgroup->busy==1
831 /* Ugly, but how can we sleep for the lock
832 * otherwise? perhaps from tq_disk?
840 /* no more work for this hwgroup (for now) */
844 if (hwgroup->hwif->sharing_irq &&
845 hwif != hwgroup->hwif &&
846 hwif->io_ports[IDE_CONTROL_OFFSET]) {
847 /* set nIEN for previous hwif */
848 SELECT_INTERRUPT(drive);
850 hwgroup->hwif = hwif;
851 hwgroup->drive = drive;
853 drive->service_start = jiffies;
855 if (blk_queue_plugged(drive->queue)) {
856 printk(KERN_ERR "ide: huh? queue was plugged!\n");
861 * we know that the queue isn't empty, but this can happen
862 * if the q->prep_rq_fn() decides to kill a request
864 rq = elv_next_request(drive->queue);
871 * Sanity: don't accept a request that isn't a PM request
872 * if we are currently power managed. This is very important as
873 * blk_stop_queue() doesn't prevent the elv_next_request()
874 * above to return us whatever is in the queue. Since we call
875 * ide_do_request() ourselves, we end up taking requests while
876 * the queue is blocked...
878 * We let requests forced at head of queue with ide-preempt
879 * though. I hope that doesn't happen too much, hopefully not
880 * unless the subdriver triggers such a thing in its own PM
883 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
884 /* We clear busy, there should be no pending ATA command at this point. */
892 * Some systems have trouble with IDE IRQs arriving while
893 * the driver is still setting things up. So, here we disable
894 * the IRQ used by this interface while the request is being started.
895 * This may look bad at first, but pretty much the same thing
896 * happens anyway when any interrupt comes in, IDE or otherwise
897 * -- the kernel masks the IRQ while it is being handled.
899 if (hwif->irq != masked_irq)
900 disable_irq_nosync(hwif->irq);
901 spin_unlock(&ide_lock);
903 /* allow other IRQs while we start this request */
904 startstop = start_request(drive, rq);
905 spin_lock_irq(&ide_lock);
906 if (hwif->irq != masked_irq)
907 enable_irq(hwif->irq);
908 if (startstop == ide_stopped)
913 EXPORT_SYMBOL(ide_do_request);
916 * Passes the stuff to ide_do_request
918 void do_ide_request(request_queue_t *q)
920 ide_do_request(q->queuedata, IDE_NO_IRQ);
924 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
925 * retry the current request in pio mode instead of risking tossing it
928 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
930 ide_hwif_t *hwif = HWIF(drive);
932 ide_startstop_t ret = ide_stopped;
935 * end current dma transaction
939 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
940 (void)HWIF(drive)->ide_dma_end(drive);
941 ret = DRIVER(drive)->error(drive, "dma timeout error",
942 hwif->INB(IDE_STATUS_REG));
944 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
945 (void) hwif->ide_dma_timeout(drive);
949 * disable dma for now, but remember that we did so because of
950 * a timeout -- we'll reenable after we finish this next request
951 * (or rather the first chunk of it) in pio.
954 drive->state = DMA_PIO_RETRY;
955 (void) hwif->ide_dma_off_quietly(drive);
958 * un-busy drive etc (hwgroup->busy is cleared on return) and
959 * make sure request is sane
961 rq = HWGROUP(drive)->rq;
962 HWGROUP(drive)->rq = NULL;
965 rq->sector = rq->bio->bi_sector;
966 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
967 rq->hard_cur_sectors = rq->current_nr_sectors;
975 * ide_timer_expiry - handle lack of an IDE interrupt
976 * @data: timer callback magic (hwgroup)
978 * An IDE command has timed out before the expected drive return
979 * occurred. At this point we attempt to clean up the current
980 * mess. If the current handler includes an expiry handler then
981 * we invoke the expiry handler, and providing it is happy the
982 * work is done. If that fails we apply generic recovery rules
983 * invoking the handler and checking the drive DMA status. We
984 * have an excessively incestuous relationship with the DMA
985 * logic that wants cleaning up.
988 void ide_timer_expiry (unsigned long data)
990 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
991 ide_handler_t *handler;
992 ide_expiry_t *expiry;
994 unsigned long wait = -1;
996 spin_lock_irqsave(&ide_lock, flags);
998 if ((handler = hwgroup->handler) == NULL) {
1000 * Either a marginal timeout occurred
1001 * (got the interrupt just as timer expired),
1002 * or we were "sleeping" to give other devices a chance.
1003 * Either way, we don't really want to complain about anything.
1005 if (hwgroup->sleeping) {
1006 hwgroup->sleeping = 0;
1010 ide_drive_t *drive = hwgroup->drive;
1012 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1013 hwgroup->handler = NULL;
1016 ide_startstop_t startstop = ide_stopped;
1017 if (!hwgroup->busy) {
1018 hwgroup->busy = 1; /* paranoia */
1019 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1021 if ((expiry = hwgroup->expiry) != NULL) {
1023 if ((wait = expiry(drive)) > 0) {
1025 hwgroup->timer.expires = jiffies + wait;
1026 add_timer(&hwgroup->timer);
1027 spin_unlock_irqrestore(&ide_lock, flags);
1031 hwgroup->handler = NULL;
1033 * We need to simulate a real interrupt when invoking
1034 * the handler() function, which means we need to
1035 * globally mask the specific IRQ:
1037 spin_unlock(&ide_lock);
1039 #if DISABLE_IRQ_NOSYNC
1040 disable_irq_nosync(hwif->irq);
1042 /* disable_irq_nosync ?? */
1043 disable_irq(hwif->irq);
1044 #endif /* DISABLE_IRQ_NOSYNC */
1046 * as if we were handling an interrupt */
1047 local_irq_disable();
1048 if (hwgroup->poll_timeout != 0) {
1049 startstop = handler(drive);
1050 } else if (drive_is_ready(drive)) {
1051 if (drive->waiting_for_dma)
1052 (void) hwgroup->hwif->ide_dma_lostirq(drive);
1053 (void)ide_ack_intr(hwif);
1054 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1055 startstop = handler(drive);
1057 if (drive->waiting_for_dma) {
1058 startstop = ide_dma_timeout_retry(drive, wait);
1061 DRIVER(drive)->error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1063 drive->service_time = jiffies - drive->service_start;
1064 spin_lock_irq(&ide_lock);
1065 enable_irq(hwif->irq);
1066 if (startstop == ide_stopped)
1070 ide_do_request(hwgroup, IDE_NO_IRQ);
1071 spin_unlock_irqrestore(&ide_lock, flags);
1074 EXPORT_SYMBOL(ide_timer_expiry);
1077 * unexpected_intr - handle an unexpected IDE interrupt
1078 * @irq: interrupt line
1079 * @hwgroup: hwgroup being processed
1081 * There's nothing really useful we can do with an unexpected interrupt,
1082 * other than reading the status register (to clear it), and logging it.
1083 * There should be no way that an irq can happen before we're ready for it,
1084 * so we needn't worry much about losing an "important" interrupt here.
1086 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1087 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1088 * looks "good", we just ignore the interrupt completely.
1090 * This routine assumes __cli() is in effect when called.
1092 * If an unexpected interrupt happens on irq15 while we are handling irq14
1093 * and if the two interfaces are "serialized" (CMD640), then it looks like
1094 * we could screw up by interfering with a new request being set up for
1097 * In reality, this is a non-issue. The new command is not sent unless
1098 * the drive is ready to accept one, in which case we know the drive is
1099 * not trying to interrupt us. And ide_set_handler() is always invoked
1100 * before completing the issuance of any new drive command, so we will not
1101 * be accidentally invoked as a result of any valid command completion
1104 * Note that we must walk the entire hwgroup here. We know which hwif
1105 * is doing the current command, but we don't know which hwif burped
1109 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1112 ide_hwif_t *hwif = hwgroup->hwif;
1115 * handle the unexpected interrupt
1118 if (hwif->irq == irq) {
1119 stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1120 if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1121 /* Try to not flood the console with msgs */
1122 static unsigned long last_msgtime, count;
1124 if (time_after(jiffies, last_msgtime + HZ)) {
1125 last_msgtime = jiffies;
1126 printk(KERN_ERR "%s%s: unexpected interrupt, "
1127 "status=0x%02x, count=%ld\n",
1129 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1133 } while ((hwif = hwif->next) != hwgroup->hwif);
1137 * ide_intr - default IDE interrupt handler
1138 * @irq: interrupt number
1139 * @dev_id: hwif group
1140 * @regs: unused weirdness from the kernel irq layer
1142 * This is the default IRQ handler for the IDE layer. You should
1143 * not need to override it. If you do be aware it is subtle in
1146 * hwgroup->hwif is the interface in the group currently performing
1147 * a command. hwgroup->drive is the drive and hwgroup->handler is
1148 * the IRQ handler to call. As we issue a command the handlers
1149 * step through multiple states, reassigning the handler to the
1150 * next step in the process. Unlike a smart SCSI controller IDE
1151 * expects the main processor to sequence the various transfer
1152 * stages. We also manage a poll timer to catch up with most
1153 * timeout situations. There are still a few where the handlers
1154 * don't ever decide to give up.
1156 * The handler eventually returns ide_stopped to indicate the
1157 * request completed. At this point we issue the next request
1158 * on the hwgroup and the process begins again.
1161 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1163 unsigned long flags;
1164 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1167 ide_handler_t *handler;
1168 ide_startstop_t startstop;
1170 spin_lock_irqsave(&ide_lock, flags);
1171 hwif = hwgroup->hwif;
1173 if (!ide_ack_intr(hwif)) {
1174 spin_unlock_irqrestore(&ide_lock, flags);
1178 if ((handler = hwgroup->handler) == NULL ||
1179 hwgroup->poll_timeout != 0) {
1181 * Not expecting an interrupt from this drive.
1182 * That means this could be:
1183 * (1) an interrupt from another PCI device
1184 * sharing the same PCI INT# as us.
1185 * or (2) a drive just entered sleep or standby mode,
1186 * and is interrupting to let us know.
1187 * or (3) a spurious interrupt of unknown origin.
1189 * For PCI, we cannot tell the difference,
1190 * so in that case we just ignore it and hope it goes away.
1192 * FIXME: unexpected_intr should be hwif-> then we can
1193 * remove all the ifdef PCI crap
1195 #ifdef CONFIG_BLK_DEV_IDEPCI
1196 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1197 #endif /* CONFIG_BLK_DEV_IDEPCI */
1200 * Probably not a shared PCI interrupt,
1201 * so we can safely try to do something about it:
1203 unexpected_intr(irq, hwgroup);
1204 #ifdef CONFIG_BLK_DEV_IDEPCI
1207 * Whack the status register, just in case
1208 * we have a leftover pending IRQ.
1210 (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1211 #endif /* CONFIG_BLK_DEV_IDEPCI */
1213 spin_unlock_irqrestore(&ide_lock, flags);
1216 drive = hwgroup->drive;
1219 * This should NEVER happen, and there isn't much
1220 * we could do about it here.
1222 * [Note - this can occur if the drive is hot unplugged]
1224 spin_unlock_irqrestore(&ide_lock, flags);
1227 if (!drive_is_ready(drive)) {
1229 * This happens regularly when we share a PCI IRQ with
1230 * another device. Unfortunately, it can also happen
1231 * with some buggy drives that trigger the IRQ before
1232 * their status register is up to date. Hopefully we have
1233 * enough advance overhead that the latter isn't a problem.
1235 spin_unlock_irqrestore(&ide_lock, flags);
1238 if (!hwgroup->busy) {
1239 hwgroup->busy = 1; /* paranoia */
1240 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1242 hwgroup->handler = NULL;
1243 del_timer(&hwgroup->timer);
1244 spin_unlock(&ide_lock);
1248 /* service this interrupt, may set handler for next interrupt */
1249 startstop = handler(drive);
1250 spin_lock_irq(&ide_lock);
1253 * Note that handler() may have set things up for another
1254 * interrupt to occur soon, but it cannot happen until
1255 * we exit from this routine, because it will be the
1256 * same irq as is currently being serviced here, and Linux
1257 * won't allow another of the same (on any CPU) until we return.
1259 drive->service_time = jiffies - drive->service_start;
1260 if (startstop == ide_stopped) {
1261 if (hwgroup->handler == NULL) { /* paranoia */
1263 ide_do_request(hwgroup, hwif->irq);
1265 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1266 "on exit\n", drive->name);
1269 spin_unlock_irqrestore(&ide_lock, flags);
1273 EXPORT_SYMBOL(ide_intr);
1276 * ide_init_drive_cmd - initialize a drive command request
1277 * @rq: request object
1279 * Initialize a request before we fill it in and send it down to
1280 * ide_do_drive_cmd. Commands must be set up by this function. Right
1281 * now it doesn't do a lot, but if that changes abusers will have a
1285 void ide_init_drive_cmd (struct request *rq)
1287 memset(rq, 0, sizeof(*rq));
1288 rq->flags = REQ_DRIVE_CMD;
1291 EXPORT_SYMBOL(ide_init_drive_cmd);
1294 * ide_do_drive_cmd - issue IDE special command
1295 * @drive: device to issue command
1296 * @rq: request to issue
1297 * @action: action for processing
1299 * This function issues a special IDE device request
1300 * onto the request queue.
1302 * If action is ide_wait, then the rq is queued at the end of the
1303 * request queue, and the function sleeps until it has been processed.
1304 * This is for use when invoked from an ioctl handler.
1306 * If action is ide_preempt, then the rq is queued at the head of
1307 * the request queue, displacing the currently-being-processed
1308 * request and this function returns immediately without waiting
1309 * for the new rq to be completed. This is VERY DANGEROUS, and is
1310 * intended for careful use by the ATAPI tape/cdrom driver code.
1312 * If action is ide_next, then the rq is queued immediately after
1313 * the currently-being-processed-request (if any), and the function
1314 * returns without waiting for the new rq to be completed. As above,
1315 * This is VERY DANGEROUS, and is intended for careful use by the
1316 * ATAPI tape/cdrom driver code.
1318 * If action is ide_end, then the rq is queued at the end of the
1319 * request queue, and the function returns immediately without waiting
1320 * for the new rq to be completed. This is again intended for careful
1321 * use by the ATAPI tape/cdrom driver code.
1324 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1326 unsigned long flags;
1327 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1328 DECLARE_COMPLETION(wait);
1329 int where = ELEVATOR_INSERT_BACK, err;
1330 int must_wait = (action == ide_wait || action == ide_head_wait);
1332 #ifdef CONFIG_BLK_DEV_PDC4030
1334 * FIXME: there should be a drive or hwif->special
1335 * handler that points here by default, not hacks
1336 * in the ide-io.c code
1338 * FIXME2: That code breaks power management if used with
1339 * this chipset, that really doesn't belong here !
1341 if (HWIF(drive)->chipset == ide_pdc4030 && rq->buffer != NULL)
1342 return -ENOSYS; /* special drive cmds not supported */
1345 rq->rq_status = RQ_ACTIVE;
1347 rq->rq_disk = drive->disk;
1350 * we need to hold an extra reference to request for safe inspection
1355 rq->waiting = &wait;
1358 spin_lock_irqsave(&ide_lock, flags);
1359 if (action == ide_preempt)
1361 if (action == ide_preempt || action == ide_head_wait) {
1362 where = ELEVATOR_INSERT_FRONT;
1363 rq->flags |= REQ_PREEMPT;
1365 __elv_add_request(drive->queue, rq, where, 0);
1366 ide_do_request(hwgroup, IDE_NO_IRQ);
1367 spin_unlock_irqrestore(&ide_lock, flags);
1371 wait_for_completion(&wait);
1376 blk_put_request(rq);
1382 EXPORT_SYMBOL(ide_do_drive_cmd);