]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/ide/ide-io.c
ide-tape: fix handling of non-special requests in ->end_request method
[mirror_ubuntu-bionic-kernel.git] / drivers / ide / ide-io.c
1 /*
2 * IDE I/O functions
3 *
4 * Basic PIO and command management functionality.
5 *
6 * This code was split off from ide.c. See ide.c for history and original
7 * copyrights.
8 *
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
12 * later version.
13 *
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.
18 *
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.
24 */
25
26
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>
32 #include <linux/mm.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/completion.h>
44 #include <linux/reboot.h>
45 #include <linux/cdrom.h>
46 #include <linux/seq_file.h>
47 #include <linux/device.h>
48 #include <linux/kmod.h>
49 #include <linux/scatterlist.h>
50 #include <linux/bitops.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/irq.h>
54 #include <asm/uaccess.h>
55 #include <asm/io.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58 int uptodate, unsigned int nr_bytes, int dequeue)
59 {
60 int ret = 1;
61
62 /*
63 * if failfast is set on a request, override number of sectors and
64 * complete the whole request right now
65 */
66 if (blk_noretry_request(rq) && end_io_error(uptodate))
67 nr_bytes = rq->hard_nr_sectors << 9;
68
69 if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
70 rq->errors = -EIO;
71
72 /*
73 * decide whether to reenable DMA -- 3 is a random magic for now,
74 * if we DMA timeout more than 3 times, just stay in PIO
75 */
76 if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
77 drive->state = 0;
78 ide_dma_on(drive);
79 }
80
81 if (!end_that_request_chunk(rq, uptodate, nr_bytes)) {
82 add_disk_randomness(rq->rq_disk);
83 if (dequeue) {
84 if (!list_empty(&rq->queuelist))
85 blkdev_dequeue_request(rq);
86 HWGROUP(drive)->rq = NULL;
87 }
88 end_that_request_last(rq, uptodate);
89 ret = 0;
90 }
91
92 return ret;
93 }
94
95 /**
96 * ide_end_request - complete an IDE I/O
97 * @drive: IDE device for the I/O
98 * @uptodate:
99 * @nr_sectors: number of sectors completed
100 *
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.
104 */
105
106 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
107 {
108 unsigned int nr_bytes = nr_sectors << 9;
109 struct request *rq;
110 unsigned long flags;
111 int ret = 1;
112
113 /*
114 * room for locking improvements here, the calls below don't
115 * need the queue lock held at all
116 */
117 spin_lock_irqsave(&ide_lock, flags);
118 rq = HWGROUP(drive)->rq;
119
120 if (!nr_bytes) {
121 if (blk_pc_request(rq))
122 nr_bytes = rq->data_len;
123 else
124 nr_bytes = rq->hard_cur_sectors << 9;
125 }
126
127 ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
128
129 spin_unlock_irqrestore(&ide_lock, flags);
130 return ret;
131 }
132 EXPORT_SYMBOL(ide_end_request);
133
134 /*
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...
138 */
139
140 enum {
141 ide_pm_flush_cache = ide_pm_state_start_suspend,
142 idedisk_pm_standby,
143
144 idedisk_pm_restore_pio = ide_pm_state_start_resume,
145 idedisk_pm_idle,
146 ide_pm_restore_dma,
147 };
148
149 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
150 {
151 struct request_pm_state *pm = rq->data;
152
153 if (drive->media != ide_disk)
154 return;
155
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;
160 else
161 pm->pm_step = idedisk_pm_standby;
162 break;
163 case idedisk_pm_standby: /* Suspend step 2 (standby) complete */
164 pm->pm_step = ide_pm_state_completed;
165 break;
166 case idedisk_pm_restore_pio: /* Resume step 1 complete */
167 pm->pm_step = idedisk_pm_idle;
168 break;
169 case idedisk_pm_idle: /* Resume step 2 (idle) complete */
170 pm->pm_step = ide_pm_restore_dma;
171 break;
172 }
173 }
174
175 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
176 {
177 struct request_pm_state *pm = rq->data;
178 ide_task_t *args = rq->special;
179
180 memset(args, 0, sizeof(*args));
181
182 switch (pm->pm_step) {
183 case ide_pm_flush_cache: /* Suspend step 1 (flush cache) */
184 if (drive->media != ide_disk)
185 break;
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);
189 return ide_stopped;
190 }
191 if (ide_id_has_flush_cache_ext(drive->id))
192 args->tf.command = WIN_FLUSH_CACHE_EXT;
193 else
194 args->tf.command = WIN_FLUSH_CACHE;
195 goto out_do_tf;
196
197 case idedisk_pm_standby: /* Suspend step 2 (standby) */
198 args->tf.command = WIN_STANDBYNOW1;
199 goto out_do_tf;
200
201 case idedisk_pm_restore_pio: /* Resume step 1 (restore PIO) */
202 ide_set_max_pio(drive);
203 /*
204 * skip idedisk_pm_idle for ATAPI devices
205 */
206 if (drive->media != ide_disk)
207 pm->pm_step = ide_pm_restore_dma;
208 else
209 ide_complete_power_step(drive, rq, 0, 0);
210 return ide_stopped;
211
212 case idedisk_pm_idle: /* Resume step 2 (idle) */
213 args->tf.command = WIN_IDLEIMMEDIATE;
214 goto out_do_tf;
215
216 case ide_pm_restore_dma: /* Resume step 3 (restore DMA) */
217 /*
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...
221 */
222 if (drive->hwif->dma_host_set == NULL)
223 break;
224 /*
225 * TODO: respect ->using_dma setting
226 */
227 ide_set_dma(drive);
228 break;
229 }
230 pm->pm_step = ide_pm_state_completed;
231 return ide_stopped;
232
233 out_do_tf:
234 args->tf_flags = IDE_TFLAG_OUT_TF | IDE_TFLAG_OUT_DEVICE;
235 args->data_phase = TASKFILE_NO_DATA;
236 return do_rw_taskfile(drive, args);
237 }
238
239 /**
240 * ide_end_dequeued_request - complete an IDE I/O
241 * @drive: IDE device for the I/O
242 * @uptodate:
243 * @nr_sectors: number of sectors completed
244 *
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.
249 *
250 * NOTE: This path does not handle barrier, but barrier is not supported
251 * on ide-cd anyway.
252 */
253
254 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
255 int uptodate, int nr_sectors)
256 {
257 unsigned long flags;
258 int ret;
259
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);
264
265 return ret;
266 }
267 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
268
269
270 /**
271 * ide_complete_pm_request - end the current Power Management request
272 * @drive: target drive
273 * @rq: request
274 *
275 * This function cleans up the current PM request and stops the queue
276 * if necessary.
277 */
278 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
279 {
280 unsigned long flags;
281
282 #ifdef DEBUG_PM
283 printk("%s: completing PM request, %s\n", drive->name,
284 blk_pm_suspend_request(rq) ? "suspend" : "resume");
285 #endif
286 spin_lock_irqsave(&ide_lock, flags);
287 if (blk_pm_suspend_request(rq)) {
288 blk_stop_queue(drive->queue);
289 } else {
290 drive->blocked = 0;
291 blk_start_queue(drive->queue);
292 }
293 blkdev_dequeue_request(rq);
294 HWGROUP(drive)->rq = NULL;
295 end_that_request_last(rq, 1);
296 spin_unlock_irqrestore(&ide_lock, flags);
297 }
298
299 void ide_tf_read(ide_drive_t *drive, ide_task_t *task)
300 {
301 ide_hwif_t *hwif = drive->hwif;
302 struct ide_taskfile *tf = &task->tf;
303
304 if (task->tf_flags & IDE_TFLAG_IN_DATA) {
305 u16 data = hwif->INW(IDE_DATA_REG);
306
307 tf->data = data & 0xff;
308 tf->hob_data = (data >> 8) & 0xff;
309 }
310
311 /* be sure we're looking at the low order bits */
312 hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
313
314 if (task->tf_flags & IDE_TFLAG_IN_NSECT)
315 tf->nsect = hwif->INB(IDE_NSECTOR_REG);
316 if (task->tf_flags & IDE_TFLAG_IN_LBAL)
317 tf->lbal = hwif->INB(IDE_SECTOR_REG);
318 if (task->tf_flags & IDE_TFLAG_IN_LBAM)
319 tf->lbam = hwif->INB(IDE_LCYL_REG);
320 if (task->tf_flags & IDE_TFLAG_IN_LBAH)
321 tf->lbah = hwif->INB(IDE_HCYL_REG);
322 if (task->tf_flags & IDE_TFLAG_IN_DEVICE)
323 tf->device = hwif->INB(IDE_SELECT_REG);
324
325 if (task->tf_flags & IDE_TFLAG_LBA48) {
326 hwif->OUTB(drive->ctl | 0x80, IDE_CONTROL_REG);
327
328 if (task->tf_flags & IDE_TFLAG_IN_HOB_FEATURE)
329 tf->hob_feature = hwif->INB(IDE_FEATURE_REG);
330 if (task->tf_flags & IDE_TFLAG_IN_HOB_NSECT)
331 tf->hob_nsect = hwif->INB(IDE_NSECTOR_REG);
332 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAL)
333 tf->hob_lbal = hwif->INB(IDE_SECTOR_REG);
334 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAM)
335 tf->hob_lbam = hwif->INB(IDE_LCYL_REG);
336 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAH)
337 tf->hob_lbah = hwif->INB(IDE_HCYL_REG);
338 }
339 }
340
341 /**
342 * ide_end_drive_cmd - end an explicit drive command
343 * @drive: command
344 * @stat: status bits
345 * @err: error bits
346 *
347 * Clean up after success/failure of an explicit drive command.
348 * These get thrown onto the queue so they are synchronized with
349 * real I/O operations on the drive.
350 *
351 * In LBA48 mode we have to read the register set twice to get
352 * all the extra information out.
353 */
354
355 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
356 {
357 ide_hwif_t *hwif = HWIF(drive);
358 unsigned long flags;
359 struct request *rq;
360
361 spin_lock_irqsave(&ide_lock, flags);
362 rq = HWGROUP(drive)->rq;
363 spin_unlock_irqrestore(&ide_lock, flags);
364
365 if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
366 u8 *args = (u8 *) rq->buffer;
367 if (rq->errors == 0)
368 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
369
370 if (args) {
371 args[0] = stat;
372 args[1] = err;
373 /* be sure we're looking at the low order bits */
374 hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
375 args[2] = hwif->INB(IDE_NSECTOR_REG);
376 }
377 } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
378 ide_task_t *args = (ide_task_t *) rq->special;
379 if (rq->errors == 0)
380 rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
381
382 if (args) {
383 struct ide_taskfile *tf = &args->tf;
384
385 tf->error = err;
386 tf->status = stat;
387
388 args->tf_flags |= (IDE_TFLAG_IN_TF|IDE_TFLAG_IN_DEVICE);
389 if (args->tf_flags & IDE_TFLAG_LBA48)
390 args->tf_flags |= IDE_TFLAG_IN_HOB;
391
392 ide_tf_read(drive, args);
393 }
394 } else if (blk_pm_request(rq)) {
395 struct request_pm_state *pm = rq->data;
396 #ifdef DEBUG_PM
397 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
398 drive->name, rq->pm->pm_step, stat, err);
399 #endif
400 ide_complete_power_step(drive, rq, stat, err);
401 if (pm->pm_step == ide_pm_state_completed)
402 ide_complete_pm_request(drive, rq);
403 return;
404 }
405
406 spin_lock_irqsave(&ide_lock, flags);
407 blkdev_dequeue_request(rq);
408 HWGROUP(drive)->rq = NULL;
409 rq->errors = err;
410 end_that_request_last(rq, !rq->errors);
411 spin_unlock_irqrestore(&ide_lock, flags);
412 }
413
414 EXPORT_SYMBOL(ide_end_drive_cmd);
415
416 /**
417 * try_to_flush_leftover_data - flush junk
418 * @drive: drive to flush
419 *
420 * try_to_flush_leftover_data() is invoked in response to a drive
421 * unexpectedly having its DRQ_STAT bit set. As an alternative to
422 * resetting the drive, this routine tries to clear the condition
423 * by read a sector's worth of data from the drive. Of course,
424 * this may not help if the drive is *waiting* for data from *us*.
425 */
426 static void try_to_flush_leftover_data (ide_drive_t *drive)
427 {
428 int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
429
430 if (drive->media != ide_disk)
431 return;
432 while (i > 0) {
433 u32 buffer[16];
434 u32 wcount = (i > 16) ? 16 : i;
435
436 i -= wcount;
437 HWIF(drive)->ata_input_data(drive, buffer, wcount);
438 }
439 }
440
441 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
442 {
443 if (rq->rq_disk) {
444 ide_driver_t *drv;
445
446 drv = *(ide_driver_t **)rq->rq_disk->private_data;
447 drv->end_request(drive, 0, 0);
448 } else
449 ide_end_request(drive, 0, 0);
450 }
451
452 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
453 {
454 ide_hwif_t *hwif = drive->hwif;
455
456 if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
457 /* other bits are useless when BUSY */
458 rq->errors |= ERROR_RESET;
459 } else if (stat & ERR_STAT) {
460 /* err has different meaning on cdrom and tape */
461 if (err == ABRT_ERR) {
462 if (drive->select.b.lba &&
463 /* some newer drives don't support WIN_SPECIFY */
464 hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
465 return ide_stopped;
466 } else if ((err & BAD_CRC) == BAD_CRC) {
467 /* UDMA crc error, just retry the operation */
468 drive->crc_count++;
469 } else if (err & (BBD_ERR | ECC_ERR)) {
470 /* retries won't help these */
471 rq->errors = ERROR_MAX;
472 } else if (err & TRK0_ERR) {
473 /* help it find track zero */
474 rq->errors |= ERROR_RECAL;
475 }
476 }
477
478 if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ &&
479 (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0)
480 try_to_flush_leftover_data(drive);
481
482 if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
483 ide_kill_rq(drive, rq);
484 return ide_stopped;
485 }
486
487 if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
488 rq->errors |= ERROR_RESET;
489
490 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
491 ++rq->errors;
492 return ide_do_reset(drive);
493 }
494
495 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
496 drive->special.b.recalibrate = 1;
497
498 ++rq->errors;
499
500 return ide_stopped;
501 }
502
503 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
504 {
505 ide_hwif_t *hwif = drive->hwif;
506
507 if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
508 /* other bits are useless when BUSY */
509 rq->errors |= ERROR_RESET;
510 } else {
511 /* add decoding error stuff */
512 }
513
514 if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
515 /* force an abort */
516 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
517
518 if (rq->errors >= ERROR_MAX) {
519 ide_kill_rq(drive, rq);
520 } else {
521 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
522 ++rq->errors;
523 return ide_do_reset(drive);
524 }
525 ++rq->errors;
526 }
527
528 return ide_stopped;
529 }
530
531 ide_startstop_t
532 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
533 {
534 if (drive->media == ide_disk)
535 return ide_ata_error(drive, rq, stat, err);
536 return ide_atapi_error(drive, rq, stat, err);
537 }
538
539 EXPORT_SYMBOL_GPL(__ide_error);
540
541 /**
542 * ide_error - handle an error on the IDE
543 * @drive: drive the error occurred on
544 * @msg: message to report
545 * @stat: status bits
546 *
547 * ide_error() takes action based on the error returned by the drive.
548 * For normal I/O that may well include retries. We deal with
549 * both new-style (taskfile) and old style command handling here.
550 * In the case of taskfile command handling there is work left to
551 * do
552 */
553
554 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
555 {
556 struct request *rq;
557 u8 err;
558
559 err = ide_dump_status(drive, msg, stat);
560
561 if ((rq = HWGROUP(drive)->rq) == NULL)
562 return ide_stopped;
563
564 /* retry only "normal" I/O: */
565 if (!blk_fs_request(rq)) {
566 rq->errors = 1;
567 ide_end_drive_cmd(drive, stat, err);
568 return ide_stopped;
569 }
570
571 if (rq->rq_disk) {
572 ide_driver_t *drv;
573
574 drv = *(ide_driver_t **)rq->rq_disk->private_data;
575 return drv->error(drive, rq, stat, err);
576 } else
577 return __ide_error(drive, rq, stat, err);
578 }
579
580 EXPORT_SYMBOL_GPL(ide_error);
581
582 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
583 {
584 if (drive->media != ide_disk)
585 rq->errors |= ERROR_RESET;
586
587 ide_kill_rq(drive, rq);
588
589 return ide_stopped;
590 }
591
592 EXPORT_SYMBOL_GPL(__ide_abort);
593
594 /**
595 * ide_abort - abort pending IDE operations
596 * @drive: drive the error occurred on
597 * @msg: message to report
598 *
599 * ide_abort kills and cleans up when we are about to do a
600 * host initiated reset on active commands. Longer term we
601 * want handlers to have sensible abort handling themselves
602 *
603 * This differs fundamentally from ide_error because in
604 * this case the command is doing just fine when we
605 * blow it away.
606 */
607
608 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
609 {
610 struct request *rq;
611
612 if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
613 return ide_stopped;
614
615 /* retry only "normal" I/O: */
616 if (!blk_fs_request(rq)) {
617 rq->errors = 1;
618 ide_end_drive_cmd(drive, BUSY_STAT, 0);
619 return ide_stopped;
620 }
621
622 if (rq->rq_disk) {
623 ide_driver_t *drv;
624
625 drv = *(ide_driver_t **)rq->rq_disk->private_data;
626 return drv->abort(drive, rq);
627 } else
628 return __ide_abort(drive, rq);
629 }
630
631 /**
632 * drive_cmd_intr - drive command completion interrupt
633 * @drive: drive the completion interrupt occurred on
634 *
635 * drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
636 * We do any necessary data reading and then wait for the drive to
637 * go non busy. At that point we may read the error data and complete
638 * the request
639 */
640
641 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
642 {
643 struct request *rq = HWGROUP(drive)->rq;
644 ide_hwif_t *hwif = HWIF(drive);
645 u8 *args = (u8 *) rq->buffer;
646 u8 stat = hwif->INB(IDE_STATUS_REG);
647 int retries = 10;
648
649 local_irq_enable_in_hardirq();
650 if (rq->cmd_type == REQ_TYPE_ATA_CMD &&
651 (stat & DRQ_STAT) && args && args[3]) {
652 u8 io_32bit = drive->io_32bit;
653 drive->io_32bit = 0;
654 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
655 drive->io_32bit = io_32bit;
656 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
657 udelay(100);
658 }
659
660 if (!OK_STAT(stat, READY_STAT, BAD_STAT))
661 return ide_error(drive, "drive_cmd", stat);
662 /* calls ide_end_drive_cmd */
663 ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
664 return ide_stopped;
665 }
666
667 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
668 {
669 tf->nsect = drive->sect;
670 tf->lbal = drive->sect;
671 tf->lbam = drive->cyl;
672 tf->lbah = drive->cyl >> 8;
673 tf->device = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
674 tf->command = WIN_SPECIFY;
675 }
676
677 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
678 {
679 tf->nsect = drive->sect;
680 tf->command = WIN_RESTORE;
681 }
682
683 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
684 {
685 tf->nsect = drive->mult_req;
686 tf->command = WIN_SETMULT;
687 }
688
689 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
690 {
691 special_t *s = &drive->special;
692 ide_task_t args;
693
694 memset(&args, 0, sizeof(ide_task_t));
695 args.data_phase = TASKFILE_NO_DATA;
696
697 if (s->b.set_geometry) {
698 s->b.set_geometry = 0;
699 ide_tf_set_specify_cmd(drive, &args.tf);
700 } else if (s->b.recalibrate) {
701 s->b.recalibrate = 0;
702 ide_tf_set_restore_cmd(drive, &args.tf);
703 } else if (s->b.set_multmode) {
704 s->b.set_multmode = 0;
705 if (drive->mult_req > drive->id->max_multsect)
706 drive->mult_req = drive->id->max_multsect;
707 ide_tf_set_setmult_cmd(drive, &args.tf);
708 } else if (s->all) {
709 int special = s->all;
710 s->all = 0;
711 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
712 return ide_stopped;
713 }
714
715 args.tf_flags = IDE_TFLAG_OUT_TF | IDE_TFLAG_OUT_DEVICE |
716 IDE_TFLAG_CUSTOM_HANDLER;
717
718 do_rw_taskfile(drive, &args);
719
720 return ide_started;
721 }
722
723 /*
724 * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
725 */
726 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
727 {
728 switch (req_pio) {
729 case 202:
730 case 201:
731 case 200:
732 case 102:
733 case 101:
734 case 100:
735 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
736 case 9:
737 case 8:
738 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
739 case 7:
740 case 6:
741 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
742 default:
743 return 0;
744 }
745 }
746
747 /**
748 * do_special - issue some special commands
749 * @drive: drive the command is for
750 *
751 * do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
752 * commands to a drive. It used to do much more, but has been scaled
753 * back.
754 */
755
756 static ide_startstop_t do_special (ide_drive_t *drive)
757 {
758 special_t *s = &drive->special;
759
760 #ifdef DEBUG
761 printk("%s: do_special: 0x%02x\n", drive->name, s->all);
762 #endif
763 if (s->b.set_tune) {
764 ide_hwif_t *hwif = drive->hwif;
765 u8 req_pio = drive->tune_req;
766
767 s->b.set_tune = 0;
768
769 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
770
771 if (hwif->set_pio_mode == NULL)
772 return ide_stopped;
773
774 /*
775 * take ide_lock for drive->[no_]unmask/[no_]io_32bit
776 */
777 if (req_pio == 8 || req_pio == 9) {
778 unsigned long flags;
779
780 spin_lock_irqsave(&ide_lock, flags);
781 hwif->set_pio_mode(drive, req_pio);
782 spin_unlock_irqrestore(&ide_lock, flags);
783 } else
784 hwif->set_pio_mode(drive, req_pio);
785 } else {
786 int keep_dma = drive->using_dma;
787
788 ide_set_pio(drive, req_pio);
789
790 if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
791 if (keep_dma)
792 ide_dma_on(drive);
793 }
794 }
795
796 return ide_stopped;
797 } else {
798 if (drive->media == ide_disk)
799 return ide_disk_special(drive);
800
801 s->all = 0;
802 drive->mult_req = 0;
803 return ide_stopped;
804 }
805 }
806
807 void ide_map_sg(ide_drive_t *drive, struct request *rq)
808 {
809 ide_hwif_t *hwif = drive->hwif;
810 struct scatterlist *sg = hwif->sg_table;
811
812 if (hwif->sg_mapped) /* needed by ide-scsi */
813 return;
814
815 if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
816 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
817 } else {
818 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
819 hwif->sg_nents = 1;
820 }
821 }
822
823 EXPORT_SYMBOL_GPL(ide_map_sg);
824
825 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
826 {
827 ide_hwif_t *hwif = drive->hwif;
828
829 hwif->nsect = hwif->nleft = rq->nr_sectors;
830 hwif->cursg_ofs = 0;
831 hwif->cursg = NULL;
832 }
833
834 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
835
836 /**
837 * execute_drive_command - issue special drive command
838 * @drive: the drive to issue the command on
839 * @rq: the request structure holding the command
840 *
841 * execute_drive_cmd() issues a special drive command, usually
842 * initiated by ioctl() from the external hdparm program. The
843 * command can be a drive command, drive task or taskfile
844 * operation. Weirdly you can call it with NULL to wait for
845 * all commands to finish. Don't do this as that is due to change
846 */
847
848 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
849 struct request *rq)
850 {
851 ide_hwif_t *hwif = HWIF(drive);
852 u8 *args = rq->buffer;
853 ide_task_t ltask;
854 struct ide_taskfile *tf = &ltask.tf;
855
856 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
857 ide_task_t *task = rq->special;
858
859 if (task == NULL)
860 goto done;
861
862 hwif->data_phase = task->data_phase;
863
864 switch (hwif->data_phase) {
865 case TASKFILE_MULTI_OUT:
866 case TASKFILE_OUT:
867 case TASKFILE_MULTI_IN:
868 case TASKFILE_IN:
869 ide_init_sg_cmd(drive, rq);
870 ide_map_sg(drive, rq);
871 default:
872 break;
873 }
874
875 return do_rw_taskfile(drive, task);
876 }
877
878 if (args == NULL)
879 goto done;
880
881 memset(&ltask, 0, sizeof(ltask));
882 if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
883 #ifdef DEBUG
884 printk("%s: DRIVE_CMD\n", drive->name);
885 #endif
886 tf->feature = args[2];
887 if (args[0] == WIN_SMART) {
888 tf->nsect = args[3];
889 tf->lbal = args[1];
890 tf->lbam = 0x4f;
891 tf->lbah = 0xc2;
892 ltask.tf_flags = IDE_TFLAG_OUT_TF;
893 } else {
894 tf->nsect = args[1];
895 ltask.tf_flags = IDE_TFLAG_OUT_FEATURE |
896 IDE_TFLAG_OUT_NSECT;
897 }
898 }
899 tf->command = args[0];
900 ide_tf_load(drive, &ltask);
901 ide_execute_command(drive, args[0], &drive_cmd_intr, WAIT_WORSTCASE, NULL);
902 return ide_started;
903
904 done:
905 /*
906 * NULL is actually a valid way of waiting for
907 * all current requests to be flushed from the queue.
908 */
909 #ifdef DEBUG
910 printk("%s: DRIVE_CMD (null)\n", drive->name);
911 #endif
912 ide_end_drive_cmd(drive,
913 hwif->INB(IDE_STATUS_REG),
914 hwif->INB(IDE_ERROR_REG));
915 return ide_stopped;
916 }
917
918 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
919 {
920 struct request_pm_state *pm = rq->data;
921
922 if (blk_pm_suspend_request(rq) &&
923 pm->pm_step == ide_pm_state_start_suspend)
924 /* Mark drive blocked when starting the suspend sequence. */
925 drive->blocked = 1;
926 else if (blk_pm_resume_request(rq) &&
927 pm->pm_step == ide_pm_state_start_resume) {
928 /*
929 * The first thing we do on wakeup is to wait for BSY bit to
930 * go away (with a looong timeout) as a drive on this hwif may
931 * just be POSTing itself.
932 * We do that before even selecting as the "other" device on
933 * the bus may be broken enough to walk on our toes at this
934 * point.
935 */
936 int rc;
937 #ifdef DEBUG_PM
938 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
939 #endif
940 rc = ide_wait_not_busy(HWIF(drive), 35000);
941 if (rc)
942 printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
943 SELECT_DRIVE(drive);
944 ide_set_irq(drive, 1);
945 rc = ide_wait_not_busy(HWIF(drive), 100000);
946 if (rc)
947 printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
948 }
949 }
950
951 /**
952 * start_request - start of I/O and command issuing for IDE
953 *
954 * start_request() initiates handling of a new I/O request. It
955 * accepts commands and I/O (read/write) requests. It also does
956 * the final remapping for weird stuff like EZDrive. Once
957 * device mapper can work sector level the EZDrive stuff can go away
958 *
959 * FIXME: this function needs a rename
960 */
961
962 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
963 {
964 ide_startstop_t startstop;
965 sector_t block;
966
967 BUG_ON(!blk_rq_started(rq));
968
969 #ifdef DEBUG
970 printk("%s: start_request: current=0x%08lx\n",
971 HWIF(drive)->name, (unsigned long) rq);
972 #endif
973
974 /* bail early if we've exceeded max_failures */
975 if (drive->max_failures && (drive->failures > drive->max_failures)) {
976 rq->cmd_flags |= REQ_FAILED;
977 goto kill_rq;
978 }
979
980 block = rq->sector;
981 if (blk_fs_request(rq) &&
982 (drive->media == ide_disk || drive->media == ide_floppy)) {
983 block += drive->sect0;
984 }
985 /* Yecch - this will shift the entire interval,
986 possibly killing some innocent following sector */
987 if (block == 0 && drive->remap_0_to_1 == 1)
988 block = 1; /* redirect MBR access to EZ-Drive partn table */
989
990 if (blk_pm_request(rq))
991 ide_check_pm_state(drive, rq);
992
993 SELECT_DRIVE(drive);
994 if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
995 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
996 return startstop;
997 }
998 if (!drive->special.all) {
999 ide_driver_t *drv;
1000
1001 /*
1002 * We reset the drive so we need to issue a SETFEATURES.
1003 * Do it _after_ do_special() restored device parameters.
1004 */
1005 if (drive->current_speed == 0xff)
1006 ide_config_drive_speed(drive, drive->desired_speed);
1007
1008 if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1009 rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1010 return execute_drive_cmd(drive, rq);
1011 else if (blk_pm_request(rq)) {
1012 struct request_pm_state *pm = rq->data;
1013 #ifdef DEBUG_PM
1014 printk("%s: start_power_step(step: %d)\n",
1015 drive->name, rq->pm->pm_step);
1016 #endif
1017 startstop = ide_start_power_step(drive, rq);
1018 if (startstop == ide_stopped &&
1019 pm->pm_step == ide_pm_state_completed)
1020 ide_complete_pm_request(drive, rq);
1021 return startstop;
1022 }
1023
1024 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1025 return drv->do_request(drive, rq, block);
1026 }
1027 return do_special(drive);
1028 kill_rq:
1029 ide_kill_rq(drive, rq);
1030 return ide_stopped;
1031 }
1032
1033 /**
1034 * ide_stall_queue - pause an IDE device
1035 * @drive: drive to stall
1036 * @timeout: time to stall for (jiffies)
1037 *
1038 * ide_stall_queue() can be used by a drive to give excess bandwidth back
1039 * to the hwgroup by sleeping for timeout jiffies.
1040 */
1041
1042 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1043 {
1044 if (timeout > WAIT_WORSTCASE)
1045 timeout = WAIT_WORSTCASE;
1046 drive->sleep = timeout + jiffies;
1047 drive->sleeping = 1;
1048 }
1049
1050 EXPORT_SYMBOL(ide_stall_queue);
1051
1052 #define WAKEUP(drive) ((drive)->service_start + 2 * (drive)->service_time)
1053
1054 /**
1055 * choose_drive - select a drive to service
1056 * @hwgroup: hardware group to select on
1057 *
1058 * choose_drive() selects the next drive which will be serviced.
1059 * This is necessary because the IDE layer can't issue commands
1060 * to both drives on the same cable, unlike SCSI.
1061 */
1062
1063 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1064 {
1065 ide_drive_t *drive, *best;
1066
1067 repeat:
1068 best = NULL;
1069 drive = hwgroup->drive;
1070
1071 /*
1072 * drive is doing pre-flush, ordered write, post-flush sequence. even
1073 * though that is 3 requests, it must be seen as a single transaction.
1074 * we must not preempt this drive until that is complete
1075 */
1076 if (blk_queue_flushing(drive->queue)) {
1077 /*
1078 * small race where queue could get replugged during
1079 * the 3-request flush cycle, just yank the plug since
1080 * we want it to finish asap
1081 */
1082 blk_remove_plug(drive->queue);
1083 return drive;
1084 }
1085
1086 do {
1087 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1088 && !elv_queue_empty(drive->queue)) {
1089 if (!best
1090 || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1091 || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1092 {
1093 if (!blk_queue_plugged(drive->queue))
1094 best = drive;
1095 }
1096 }
1097 } while ((drive = drive->next) != hwgroup->drive);
1098 if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1099 long t = (signed long)(WAKEUP(best) - jiffies);
1100 if (t >= WAIT_MIN_SLEEP) {
1101 /*
1102 * We *may* have some time to spare, but first let's see if
1103 * someone can potentially benefit from our nice mood today..
1104 */
1105 drive = best->next;
1106 do {
1107 if (!drive->sleeping
1108 && time_before(jiffies - best->service_time, WAKEUP(drive))
1109 && time_before(WAKEUP(drive), jiffies + t))
1110 {
1111 ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1112 goto repeat;
1113 }
1114 } while ((drive = drive->next) != best);
1115 }
1116 }
1117 return best;
1118 }
1119
1120 /*
1121 * Issue a new request to a drive from hwgroup
1122 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1123 *
1124 * A hwgroup is a serialized group of IDE interfaces. Usually there is
1125 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1126 * may have both interfaces in a single hwgroup to "serialize" access.
1127 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1128 * together into one hwgroup for serialized access.
1129 *
1130 * Note also that several hwgroups can end up sharing a single IRQ,
1131 * possibly along with many other devices. This is especially common in
1132 * PCI-based systems with off-board IDE controller cards.
1133 *
1134 * The IDE driver uses the single global ide_lock spinlock to protect
1135 * access to the request queues, and to protect the hwgroup->busy flag.
1136 *
1137 * The first thread into the driver for a particular hwgroup sets the
1138 * hwgroup->busy flag to indicate that this hwgroup is now active,
1139 * and then initiates processing of the top request from the request queue.
1140 *
1141 * Other threads attempting entry notice the busy setting, and will simply
1142 * queue their new requests and exit immediately. Note that hwgroup->busy
1143 * remains set even when the driver is merely awaiting the next interrupt.
1144 * Thus, the meaning is "this hwgroup is busy processing a request".
1145 *
1146 * When processing of a request completes, the completing thread or IRQ-handler
1147 * will start the next request from the queue. If no more work remains,
1148 * the driver will clear the hwgroup->busy flag and exit.
1149 *
1150 * The ide_lock (spinlock) is used to protect all access to the
1151 * hwgroup->busy flag, but is otherwise not needed for most processing in
1152 * the driver. This makes the driver much more friendlier to shared IRQs
1153 * than previous designs, while remaining 100% (?) SMP safe and capable.
1154 */
1155 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1156 {
1157 ide_drive_t *drive;
1158 ide_hwif_t *hwif;
1159 struct request *rq;
1160 ide_startstop_t startstop;
1161 int loops = 0;
1162
1163 /* for atari only: POSSIBLY BROKEN HERE(?) */
1164 ide_get_lock(ide_intr, hwgroup);
1165
1166 /* caller must own ide_lock */
1167 BUG_ON(!irqs_disabled());
1168
1169 while (!hwgroup->busy) {
1170 hwgroup->busy = 1;
1171 drive = choose_drive(hwgroup);
1172 if (drive == NULL) {
1173 int sleeping = 0;
1174 unsigned long sleep = 0; /* shut up, gcc */
1175 hwgroup->rq = NULL;
1176 drive = hwgroup->drive;
1177 do {
1178 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1179 sleeping = 1;
1180 sleep = drive->sleep;
1181 }
1182 } while ((drive = drive->next) != hwgroup->drive);
1183 if (sleeping) {
1184 /*
1185 * Take a short snooze, and then wake up this hwgroup again.
1186 * This gives other hwgroups on the same a chance to
1187 * play fairly with us, just in case there are big differences
1188 * in relative throughputs.. don't want to hog the cpu too much.
1189 */
1190 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1191 sleep = jiffies + WAIT_MIN_SLEEP;
1192 #if 1
1193 if (timer_pending(&hwgroup->timer))
1194 printk(KERN_CRIT "ide_set_handler: timer already active\n");
1195 #endif
1196 /* so that ide_timer_expiry knows what to do */
1197 hwgroup->sleeping = 1;
1198 hwgroup->req_gen_timer = hwgroup->req_gen;
1199 mod_timer(&hwgroup->timer, sleep);
1200 /* we purposely leave hwgroup->busy==1
1201 * while sleeping */
1202 } else {
1203 /* Ugly, but how can we sleep for the lock
1204 * otherwise? perhaps from tq_disk?
1205 */
1206
1207 /* for atari only */
1208 ide_release_lock();
1209 hwgroup->busy = 0;
1210 }
1211
1212 /* no more work for this hwgroup (for now) */
1213 return;
1214 }
1215 again:
1216 hwif = HWIF(drive);
1217 if (hwgroup->hwif->sharing_irq && hwif != hwgroup->hwif) {
1218 /*
1219 * set nIEN for previous hwif, drives in the
1220 * quirk_list may not like intr setups/cleanups
1221 */
1222 if (drive->quirk_list != 1)
1223 ide_set_irq(drive, 0);
1224 }
1225 hwgroup->hwif = hwif;
1226 hwgroup->drive = drive;
1227 drive->sleeping = 0;
1228 drive->service_start = jiffies;
1229
1230 if (blk_queue_plugged(drive->queue)) {
1231 printk(KERN_ERR "ide: huh? queue was plugged!\n");
1232 break;
1233 }
1234
1235 /*
1236 * we know that the queue isn't empty, but this can happen
1237 * if the q->prep_rq_fn() decides to kill a request
1238 */
1239 rq = elv_next_request(drive->queue);
1240 if (!rq) {
1241 hwgroup->busy = 0;
1242 break;
1243 }
1244
1245 /*
1246 * Sanity: don't accept a request that isn't a PM request
1247 * if we are currently power managed. This is very important as
1248 * blk_stop_queue() doesn't prevent the elv_next_request()
1249 * above to return us whatever is in the queue. Since we call
1250 * ide_do_request() ourselves, we end up taking requests while
1251 * the queue is blocked...
1252 *
1253 * We let requests forced at head of queue with ide-preempt
1254 * though. I hope that doesn't happen too much, hopefully not
1255 * unless the subdriver triggers such a thing in its own PM
1256 * state machine.
1257 *
1258 * We count how many times we loop here to make sure we service
1259 * all drives in the hwgroup without looping for ever
1260 */
1261 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1262 drive = drive->next ? drive->next : hwgroup->drive;
1263 if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1264 goto again;
1265 /* We clear busy, there should be no pending ATA command at this point. */
1266 hwgroup->busy = 0;
1267 break;
1268 }
1269
1270 hwgroup->rq = rq;
1271
1272 /*
1273 * Some systems have trouble with IDE IRQs arriving while
1274 * the driver is still setting things up. So, here we disable
1275 * the IRQ used by this interface while the request is being started.
1276 * This may look bad at first, but pretty much the same thing
1277 * happens anyway when any interrupt comes in, IDE or otherwise
1278 * -- the kernel masks the IRQ while it is being handled.
1279 */
1280 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1281 disable_irq_nosync(hwif->irq);
1282 spin_unlock(&ide_lock);
1283 local_irq_enable_in_hardirq();
1284 /* allow other IRQs while we start this request */
1285 startstop = start_request(drive, rq);
1286 spin_lock_irq(&ide_lock);
1287 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1288 enable_irq(hwif->irq);
1289 if (startstop == ide_stopped)
1290 hwgroup->busy = 0;
1291 }
1292 }
1293
1294 /*
1295 * Passes the stuff to ide_do_request
1296 */
1297 void do_ide_request(struct request_queue *q)
1298 {
1299 ide_drive_t *drive = q->queuedata;
1300
1301 ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1302 }
1303
1304 /*
1305 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1306 * retry the current request in pio mode instead of risking tossing it
1307 * all away
1308 */
1309 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1310 {
1311 ide_hwif_t *hwif = HWIF(drive);
1312 struct request *rq;
1313 ide_startstop_t ret = ide_stopped;
1314
1315 /*
1316 * end current dma transaction
1317 */
1318
1319 if (error < 0) {
1320 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1321 (void)HWIF(drive)->ide_dma_end(drive);
1322 ret = ide_error(drive, "dma timeout error",
1323 hwif->INB(IDE_STATUS_REG));
1324 } else {
1325 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1326 hwif->dma_timeout(drive);
1327 }
1328
1329 /*
1330 * disable dma for now, but remember that we did so because of
1331 * a timeout -- we'll reenable after we finish this next request
1332 * (or rather the first chunk of it) in pio.
1333 */
1334 drive->retry_pio++;
1335 drive->state = DMA_PIO_RETRY;
1336 ide_dma_off_quietly(drive);
1337
1338 /*
1339 * un-busy drive etc (hwgroup->busy is cleared on return) and
1340 * make sure request is sane
1341 */
1342 rq = HWGROUP(drive)->rq;
1343
1344 if (!rq)
1345 goto out;
1346
1347 HWGROUP(drive)->rq = NULL;
1348
1349 rq->errors = 0;
1350
1351 if (!rq->bio)
1352 goto out;
1353
1354 rq->sector = rq->bio->bi_sector;
1355 rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1356 rq->hard_cur_sectors = rq->current_nr_sectors;
1357 rq->buffer = bio_data(rq->bio);
1358 out:
1359 return ret;
1360 }
1361
1362 /**
1363 * ide_timer_expiry - handle lack of an IDE interrupt
1364 * @data: timer callback magic (hwgroup)
1365 *
1366 * An IDE command has timed out before the expected drive return
1367 * occurred. At this point we attempt to clean up the current
1368 * mess. If the current handler includes an expiry handler then
1369 * we invoke the expiry handler, and providing it is happy the
1370 * work is done. If that fails we apply generic recovery rules
1371 * invoking the handler and checking the drive DMA status. We
1372 * have an excessively incestuous relationship with the DMA
1373 * logic that wants cleaning up.
1374 */
1375
1376 void ide_timer_expiry (unsigned long data)
1377 {
1378 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *) data;
1379 ide_handler_t *handler;
1380 ide_expiry_t *expiry;
1381 unsigned long flags;
1382 unsigned long wait = -1;
1383
1384 spin_lock_irqsave(&ide_lock, flags);
1385
1386 if (((handler = hwgroup->handler) == NULL) ||
1387 (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1388 /*
1389 * Either a marginal timeout occurred
1390 * (got the interrupt just as timer expired),
1391 * or we were "sleeping" to give other devices a chance.
1392 * Either way, we don't really want to complain about anything.
1393 */
1394 if (hwgroup->sleeping) {
1395 hwgroup->sleeping = 0;
1396 hwgroup->busy = 0;
1397 }
1398 } else {
1399 ide_drive_t *drive = hwgroup->drive;
1400 if (!drive) {
1401 printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1402 hwgroup->handler = NULL;
1403 } else {
1404 ide_hwif_t *hwif;
1405 ide_startstop_t startstop = ide_stopped;
1406 if (!hwgroup->busy) {
1407 hwgroup->busy = 1; /* paranoia */
1408 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1409 }
1410 if ((expiry = hwgroup->expiry) != NULL) {
1411 /* continue */
1412 if ((wait = expiry(drive)) > 0) {
1413 /* reset timer */
1414 hwgroup->timer.expires = jiffies + wait;
1415 hwgroup->req_gen_timer = hwgroup->req_gen;
1416 add_timer(&hwgroup->timer);
1417 spin_unlock_irqrestore(&ide_lock, flags);
1418 return;
1419 }
1420 }
1421 hwgroup->handler = NULL;
1422 /*
1423 * We need to simulate a real interrupt when invoking
1424 * the handler() function, which means we need to
1425 * globally mask the specific IRQ:
1426 */
1427 spin_unlock(&ide_lock);
1428 hwif = HWIF(drive);
1429 /* disable_irq_nosync ?? */
1430 disable_irq(hwif->irq);
1431 /* local CPU only,
1432 * as if we were handling an interrupt */
1433 local_irq_disable();
1434 if (hwgroup->polling) {
1435 startstop = handler(drive);
1436 } else if (drive_is_ready(drive)) {
1437 if (drive->waiting_for_dma)
1438 hwgroup->hwif->dma_lost_irq(drive);
1439 (void)ide_ack_intr(hwif);
1440 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1441 startstop = handler(drive);
1442 } else {
1443 if (drive->waiting_for_dma) {
1444 startstop = ide_dma_timeout_retry(drive, wait);
1445 } else
1446 startstop =
1447 ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1448 }
1449 drive->service_time = jiffies - drive->service_start;
1450 spin_lock_irq(&ide_lock);
1451 enable_irq(hwif->irq);
1452 if (startstop == ide_stopped)
1453 hwgroup->busy = 0;
1454 }
1455 }
1456 ide_do_request(hwgroup, IDE_NO_IRQ);
1457 spin_unlock_irqrestore(&ide_lock, flags);
1458 }
1459
1460 /**
1461 * unexpected_intr - handle an unexpected IDE interrupt
1462 * @irq: interrupt line
1463 * @hwgroup: hwgroup being processed
1464 *
1465 * There's nothing really useful we can do with an unexpected interrupt,
1466 * other than reading the status register (to clear it), and logging it.
1467 * There should be no way that an irq can happen before we're ready for it,
1468 * so we needn't worry much about losing an "important" interrupt here.
1469 *
1470 * On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1471 * the drive enters "idle", "standby", or "sleep" mode, so if the status
1472 * looks "good", we just ignore the interrupt completely.
1473 *
1474 * This routine assumes __cli() is in effect when called.
1475 *
1476 * If an unexpected interrupt happens on irq15 while we are handling irq14
1477 * and if the two interfaces are "serialized" (CMD640), then it looks like
1478 * we could screw up by interfering with a new request being set up for
1479 * irq15.
1480 *
1481 * In reality, this is a non-issue. The new command is not sent unless
1482 * the drive is ready to accept one, in which case we know the drive is
1483 * not trying to interrupt us. And ide_set_handler() is always invoked
1484 * before completing the issuance of any new drive command, so we will not
1485 * be accidentally invoked as a result of any valid command completion
1486 * interrupt.
1487 *
1488 * Note that we must walk the entire hwgroup here. We know which hwif
1489 * is doing the current command, but we don't know which hwif burped
1490 * mysteriously.
1491 */
1492
1493 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1494 {
1495 u8 stat;
1496 ide_hwif_t *hwif = hwgroup->hwif;
1497
1498 /*
1499 * handle the unexpected interrupt
1500 */
1501 do {
1502 if (hwif->irq == irq) {
1503 stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1504 if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1505 /* Try to not flood the console with msgs */
1506 static unsigned long last_msgtime, count;
1507 ++count;
1508 if (time_after(jiffies, last_msgtime + HZ)) {
1509 last_msgtime = jiffies;
1510 printk(KERN_ERR "%s%s: unexpected interrupt, "
1511 "status=0x%02x, count=%ld\n",
1512 hwif->name,
1513 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1514 }
1515 }
1516 }
1517 } while ((hwif = hwif->next) != hwgroup->hwif);
1518 }
1519
1520 /**
1521 * ide_intr - default IDE interrupt handler
1522 * @irq: interrupt number
1523 * @dev_id: hwif group
1524 * @regs: unused weirdness from the kernel irq layer
1525 *
1526 * This is the default IRQ handler for the IDE layer. You should
1527 * not need to override it. If you do be aware it is subtle in
1528 * places
1529 *
1530 * hwgroup->hwif is the interface in the group currently performing
1531 * a command. hwgroup->drive is the drive and hwgroup->handler is
1532 * the IRQ handler to call. As we issue a command the handlers
1533 * step through multiple states, reassigning the handler to the
1534 * next step in the process. Unlike a smart SCSI controller IDE
1535 * expects the main processor to sequence the various transfer
1536 * stages. We also manage a poll timer to catch up with most
1537 * timeout situations. There are still a few where the handlers
1538 * don't ever decide to give up.
1539 *
1540 * The handler eventually returns ide_stopped to indicate the
1541 * request completed. At this point we issue the next request
1542 * on the hwgroup and the process begins again.
1543 */
1544
1545 irqreturn_t ide_intr (int irq, void *dev_id)
1546 {
1547 unsigned long flags;
1548 ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1549 ide_hwif_t *hwif;
1550 ide_drive_t *drive;
1551 ide_handler_t *handler;
1552 ide_startstop_t startstop;
1553
1554 spin_lock_irqsave(&ide_lock, flags);
1555 hwif = hwgroup->hwif;
1556
1557 if (!ide_ack_intr(hwif)) {
1558 spin_unlock_irqrestore(&ide_lock, flags);
1559 return IRQ_NONE;
1560 }
1561
1562 if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1563 /*
1564 * Not expecting an interrupt from this drive.
1565 * That means this could be:
1566 * (1) an interrupt from another PCI device
1567 * sharing the same PCI INT# as us.
1568 * or (2) a drive just entered sleep or standby mode,
1569 * and is interrupting to let us know.
1570 * or (3) a spurious interrupt of unknown origin.
1571 *
1572 * For PCI, we cannot tell the difference,
1573 * so in that case we just ignore it and hope it goes away.
1574 *
1575 * FIXME: unexpected_intr should be hwif-> then we can
1576 * remove all the ifdef PCI crap
1577 */
1578 #ifdef CONFIG_BLK_DEV_IDEPCI
1579 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1580 #endif /* CONFIG_BLK_DEV_IDEPCI */
1581 {
1582 /*
1583 * Probably not a shared PCI interrupt,
1584 * so we can safely try to do something about it:
1585 */
1586 unexpected_intr(irq, hwgroup);
1587 #ifdef CONFIG_BLK_DEV_IDEPCI
1588 } else {
1589 /*
1590 * Whack the status register, just in case
1591 * we have a leftover pending IRQ.
1592 */
1593 (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1594 #endif /* CONFIG_BLK_DEV_IDEPCI */
1595 }
1596 spin_unlock_irqrestore(&ide_lock, flags);
1597 return IRQ_NONE;
1598 }
1599 drive = hwgroup->drive;
1600 if (!drive) {
1601 /*
1602 * This should NEVER happen, and there isn't much
1603 * we could do about it here.
1604 *
1605 * [Note - this can occur if the drive is hot unplugged]
1606 */
1607 spin_unlock_irqrestore(&ide_lock, flags);
1608 return IRQ_HANDLED;
1609 }
1610 if (!drive_is_ready(drive)) {
1611 /*
1612 * This happens regularly when we share a PCI IRQ with
1613 * another device. Unfortunately, it can also happen
1614 * with some buggy drives that trigger the IRQ before
1615 * their status register is up to date. Hopefully we have
1616 * enough advance overhead that the latter isn't a problem.
1617 */
1618 spin_unlock_irqrestore(&ide_lock, flags);
1619 return IRQ_NONE;
1620 }
1621 if (!hwgroup->busy) {
1622 hwgroup->busy = 1; /* paranoia */
1623 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1624 }
1625 hwgroup->handler = NULL;
1626 hwgroup->req_gen++;
1627 del_timer(&hwgroup->timer);
1628 spin_unlock(&ide_lock);
1629
1630 /* Some controllers might set DMA INTR no matter DMA or PIO;
1631 * bmdma status might need to be cleared even for
1632 * PIO interrupts to prevent spurious/lost irq.
1633 */
1634 if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1635 /* ide_dma_end() needs bmdma status for error checking.
1636 * So, skip clearing bmdma status here and leave it
1637 * to ide_dma_end() if this is dma interrupt.
1638 */
1639 hwif->ide_dma_clear_irq(drive);
1640
1641 if (drive->unmask)
1642 local_irq_enable_in_hardirq();
1643 /* service this interrupt, may set handler for next interrupt */
1644 startstop = handler(drive);
1645 spin_lock_irq(&ide_lock);
1646
1647 /*
1648 * Note that handler() may have set things up for another
1649 * interrupt to occur soon, but it cannot happen until
1650 * we exit from this routine, because it will be the
1651 * same irq as is currently being serviced here, and Linux
1652 * won't allow another of the same (on any CPU) until we return.
1653 */
1654 drive->service_time = jiffies - drive->service_start;
1655 if (startstop == ide_stopped) {
1656 if (hwgroup->handler == NULL) { /* paranoia */
1657 hwgroup->busy = 0;
1658 ide_do_request(hwgroup, hwif->irq);
1659 } else {
1660 printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1661 "on exit\n", drive->name);
1662 }
1663 }
1664 spin_unlock_irqrestore(&ide_lock, flags);
1665 return IRQ_HANDLED;
1666 }
1667
1668 /**
1669 * ide_init_drive_cmd - initialize a drive command request
1670 * @rq: request object
1671 *
1672 * Initialize a request before we fill it in and send it down to
1673 * ide_do_drive_cmd. Commands must be set up by this function. Right
1674 * now it doesn't do a lot, but if that changes abusers will have a
1675 * nasty surprise.
1676 */
1677
1678 void ide_init_drive_cmd (struct request *rq)
1679 {
1680 memset(rq, 0, sizeof(*rq));
1681 rq->cmd_type = REQ_TYPE_ATA_CMD;
1682 rq->ref_count = 1;
1683 }
1684
1685 EXPORT_SYMBOL(ide_init_drive_cmd);
1686
1687 /**
1688 * ide_do_drive_cmd - issue IDE special command
1689 * @drive: device to issue command
1690 * @rq: request to issue
1691 * @action: action for processing
1692 *
1693 * This function issues a special IDE device request
1694 * onto the request queue.
1695 *
1696 * If action is ide_wait, then the rq is queued at the end of the
1697 * request queue, and the function sleeps until it has been processed.
1698 * This is for use when invoked from an ioctl handler.
1699 *
1700 * If action is ide_preempt, then the rq is queued at the head of
1701 * the request queue, displacing the currently-being-processed
1702 * request and this function returns immediately without waiting
1703 * for the new rq to be completed. This is VERY DANGEROUS, and is
1704 * intended for careful use by the ATAPI tape/cdrom driver code.
1705 *
1706 * If action is ide_end, then the rq is queued at the end of the
1707 * request queue, and the function returns immediately without waiting
1708 * for the new rq to be completed. This is again intended for careful
1709 * use by the ATAPI tape/cdrom driver code.
1710 */
1711
1712 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1713 {
1714 unsigned long flags;
1715 ide_hwgroup_t *hwgroup = HWGROUP(drive);
1716 DECLARE_COMPLETION_ONSTACK(wait);
1717 int where = ELEVATOR_INSERT_BACK, err;
1718 int must_wait = (action == ide_wait || action == ide_head_wait);
1719
1720 rq->errors = 0;
1721
1722 /*
1723 * we need to hold an extra reference to request for safe inspection
1724 * after completion
1725 */
1726 if (must_wait) {
1727 rq->ref_count++;
1728 rq->end_io_data = &wait;
1729 rq->end_io = blk_end_sync_rq;
1730 }
1731
1732 spin_lock_irqsave(&ide_lock, flags);
1733 if (action == ide_preempt)
1734 hwgroup->rq = NULL;
1735 if (action == ide_preempt || action == ide_head_wait) {
1736 where = ELEVATOR_INSERT_FRONT;
1737 rq->cmd_flags |= REQ_PREEMPT;
1738 }
1739 __elv_add_request(drive->queue, rq, where, 0);
1740 ide_do_request(hwgroup, IDE_NO_IRQ);
1741 spin_unlock_irqrestore(&ide_lock, flags);
1742
1743 err = 0;
1744 if (must_wait) {
1745 wait_for_completion(&wait);
1746 if (rq->errors)
1747 err = -EIO;
1748
1749 blk_put_request(rq);
1750 }
1751
1752 return err;
1753 }
1754
1755 EXPORT_SYMBOL(ide_do_drive_cmd);
1756
1757 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1758 {
1759 ide_task_t task;
1760
1761 memset(&task, 0, sizeof(task));
1762 task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1763 IDE_TFLAG_OUT_FEATURE | tf_flags;
1764 task.tf.feature = dma; /* Use PIO/DMA */
1765 task.tf.lbam = bcount & 0xff;
1766 task.tf.lbah = (bcount >> 8) & 0xff;
1767
1768 ide_tf_load(drive, &task);
1769 }
1770
1771 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);