Merge rsync://rsync.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / 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
51 #include <asm/byteorder.h>
52 #include <asm/irq.h>
53 #include <asm/uaccess.h>
54 #include <asm/io.h>
55 #include <asm/bitops.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, int nr_sectors)
59 {
60         int ret = 1;
61
62         BUG_ON(!(rq->flags & REQ_STARTED));
63
64         /*
65          * if failfast is set on a request, override number of sectors and
66          * complete the whole request right now
67          */
68         if (blk_noretry_request(rq) && end_io_error(uptodate))
69                 nr_sectors = rq->hard_nr_sectors;
70
71         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
72                 rq->errors = -EIO;
73
74         /*
75          * decide whether to reenable DMA -- 3 is a random magic for now,
76          * if we DMA timeout more than 3 times, just stay in PIO
77          */
78         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
79                 drive->state = 0;
80                 HWGROUP(drive)->hwif->ide_dma_on(drive);
81         }
82
83         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
84                 add_disk_randomness(rq->rq_disk);
85                 blkdev_dequeue_request(rq);
86                 HWGROUP(drive)->rq = NULL;
87                 end_that_request_last(rq, uptodate);
88                 ret = 0;
89         }
90
91         return ret;
92 }
93
94 /**
95  *      ide_end_request         -       complete an IDE I/O
96  *      @drive: IDE device for the I/O
97  *      @uptodate:
98  *      @nr_sectors: number of sectors completed
99  *
100  *      This is our end_request wrapper function. We complete the I/O
101  *      update random number input and dequeue the request, which if
102  *      it was tagged may be out of order.
103  */
104
105 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
106 {
107         struct request *rq;
108         unsigned long flags;
109         int ret = 1;
110
111         /*
112          * room for locking improvements here, the calls below don't
113          * need the queue lock held at all
114          */
115         spin_lock_irqsave(&ide_lock, flags);
116         rq = HWGROUP(drive)->rq;
117
118         if (!nr_sectors)
119                 nr_sectors = rq->hard_cur_sectors;
120
121         ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
122
123         spin_unlock_irqrestore(&ide_lock, flags);
124         return ret;
125 }
126 EXPORT_SYMBOL(ide_end_request);
127
128 /*
129  * Power Management state machine. This one is rather trivial for now,
130  * we should probably add more, like switching back to PIO on suspend
131  * to help some BIOSes, re-do the door locking on resume, etc...
132  */
133
134 enum {
135         ide_pm_flush_cache      = ide_pm_state_start_suspend,
136         idedisk_pm_standby,
137
138         idedisk_pm_idle         = ide_pm_state_start_resume,
139         ide_pm_restore_dma,
140 };
141
142 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
143 {
144         struct request_pm_state *pm = rq->end_io_data;
145
146         if (drive->media != ide_disk)
147                 return;
148
149         switch (pm->pm_step) {
150         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
151                 if (pm->pm_state == PM_EVENT_FREEZE)
152                         pm->pm_step = ide_pm_state_completed;
153                 else
154                         pm->pm_step = idedisk_pm_standby;
155                 break;
156         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
157                 pm->pm_step = ide_pm_state_completed;
158                 break;
159         case idedisk_pm_idle:           /* Resume step 1 (idle) complete */
160                 pm->pm_step = ide_pm_restore_dma;
161                 break;
162         }
163 }
164
165 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
166 {
167         struct request_pm_state *pm = rq->end_io_data;
168         ide_task_t *args = rq->special;
169
170         memset(args, 0, sizeof(*args));
171
172         if (drive->media != ide_disk) {
173                 /* skip idedisk_pm_idle for ATAPI devices */
174                 if (pm->pm_step == idedisk_pm_idle)
175                         pm->pm_step = ide_pm_restore_dma;
176         }
177
178         switch (pm->pm_step) {
179         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
180                 if (drive->media != ide_disk)
181                         break;
182                 /* Not supported? Switch to next step now. */
183                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
184                         ide_complete_power_step(drive, rq, 0, 0);
185                         return ide_stopped;
186                 }
187                 if (ide_id_has_flush_cache_ext(drive->id))
188                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
189                 else
190                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
191                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
192                 args->handler      = &task_no_data_intr;
193                 return do_rw_taskfile(drive, args);
194
195         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
196                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
197                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
198                 args->handler      = &task_no_data_intr;
199                 return do_rw_taskfile(drive, args);
200
201         case idedisk_pm_idle:           /* Resume step 1 (idle) */
202                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
203                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
204                 args->handler = task_no_data_intr;
205                 return do_rw_taskfile(drive, args);
206
207         case ide_pm_restore_dma:        /* Resume step 2 (restore DMA) */
208                 /*
209                  * Right now, all we do is call hwif->ide_dma_check(drive),
210                  * we could be smarter and check for current xfer_speed
211                  * in struct drive etc...
212                  */
213                 if ((drive->id->capability & 1) == 0)
214                         break;
215                 if (drive->hwif->ide_dma_check == NULL)
216                         break;
217                 drive->hwif->ide_dma_check(drive);
218                 break;
219         }
220         pm->pm_step = ide_pm_state_completed;
221         return ide_stopped;
222 }
223
224 /**
225  *      ide_end_dequeued_request        -       complete an IDE I/O
226  *      @drive: IDE device for the I/O
227  *      @uptodate:
228  *      @nr_sectors: number of sectors completed
229  *
230  *      Complete an I/O that is no longer on the request queue. This
231  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
232  *      We must still finish the old request but we must not tamper with the
233  *      queue in the meantime.
234  *
235  *      NOTE: This path does not handle barrier, but barrier is not supported
236  *      on ide-cd anyway.
237  */
238
239 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
240                              int uptodate, int nr_sectors)
241 {
242         unsigned long flags;
243         int ret = 1;
244
245         spin_lock_irqsave(&ide_lock, flags);
246
247         BUG_ON(!(rq->flags & REQ_STARTED));
248
249         /*
250          * if failfast is set on a request, override number of sectors and
251          * complete the whole request right now
252          */
253         if (blk_noretry_request(rq) && end_io_error(uptodate))
254                 nr_sectors = rq->hard_nr_sectors;
255
256         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
257                 rq->errors = -EIO;
258
259         /*
260          * decide whether to reenable DMA -- 3 is a random magic for now,
261          * if we DMA timeout more than 3 times, just stay in PIO
262          */
263         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
264                 drive->state = 0;
265                 HWGROUP(drive)->hwif->ide_dma_on(drive);
266         }
267
268         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
269                 add_disk_randomness(rq->rq_disk);
270                 if (blk_rq_tagged(rq))
271                         blk_queue_end_tag(drive->queue, rq);
272                 end_that_request_last(rq, uptodate);
273                 ret = 0;
274         }
275         spin_unlock_irqrestore(&ide_lock, flags);
276         return ret;
277 }
278 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
279
280
281 /**
282  *      ide_complete_pm_request - end the current Power Management request
283  *      @drive: target drive
284  *      @rq: request
285  *
286  *      This function cleans up the current PM request and stops the queue
287  *      if necessary.
288  */
289 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
290 {
291         unsigned long flags;
292
293 #ifdef DEBUG_PM
294         printk("%s: completing PM request, %s\n", drive->name,
295                blk_pm_suspend_request(rq) ? "suspend" : "resume");
296 #endif
297         spin_lock_irqsave(&ide_lock, flags);
298         if (blk_pm_suspend_request(rq)) {
299                 blk_stop_queue(drive->queue);
300         } else {
301                 drive->blocked = 0;
302                 blk_start_queue(drive->queue);
303         }
304         blkdev_dequeue_request(rq);
305         HWGROUP(drive)->rq = NULL;
306         end_that_request_last(rq, 1);
307         spin_unlock_irqrestore(&ide_lock, flags);
308 }
309
310 /*
311  * FIXME: probably move this somewhere else, name is bad too :)
312  */
313 u64 ide_get_error_location(ide_drive_t *drive, char *args)
314 {
315         u32 high, low;
316         u8 hcyl, lcyl, sect;
317         u64 sector;
318
319         high = 0;
320         hcyl = args[5];
321         lcyl = args[4];
322         sect = args[3];
323
324         if (ide_id_has_flush_cache_ext(drive->id)) {
325                 low = (hcyl << 16) | (lcyl << 8) | sect;
326                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
327                 high = ide_read_24(drive);
328         } else {
329                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
330                 if (cur & 0x40) {
331                         high = cur & 0xf;
332                         low = (hcyl << 16) | (lcyl << 8) | sect;
333                 } else {
334                         low = hcyl * drive->head * drive->sect;
335                         low += lcyl * drive->sect;
336                         low += sect - 1;
337                 }
338         }
339
340         sector = ((u64) high << 24) | low;
341         return sector;
342 }
343 EXPORT_SYMBOL(ide_get_error_location);
344
345 /**
346  *      ide_end_drive_cmd       -       end an explicit drive command
347  *      @drive: command 
348  *      @stat: status bits
349  *      @err: error bits
350  *
351  *      Clean up after success/failure of an explicit drive command.
352  *      These get thrown onto the queue so they are synchronized with
353  *      real I/O operations on the drive.
354  *
355  *      In LBA48 mode we have to read the register set twice to get
356  *      all the extra information out.
357  */
358  
359 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
360 {
361         ide_hwif_t *hwif = HWIF(drive);
362         unsigned long flags;
363         struct request *rq;
364
365         spin_lock_irqsave(&ide_lock, flags);
366         rq = HWGROUP(drive)->rq;
367         spin_unlock_irqrestore(&ide_lock, flags);
368
369         if (rq->flags & REQ_DRIVE_CMD) {
370                 u8 *args = (u8 *) rq->buffer;
371                 if (rq->errors == 0)
372                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
373
374                 if (args) {
375                         args[0] = stat;
376                         args[1] = err;
377                         args[2] = hwif->INB(IDE_NSECTOR_REG);
378                 }
379         } else if (rq->flags & REQ_DRIVE_TASK) {
380                 u8 *args = (u8 *) rq->buffer;
381                 if (rq->errors == 0)
382                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
383
384                 if (args) {
385                         args[0] = stat;
386                         args[1] = err;
387                         args[2] = hwif->INB(IDE_NSECTOR_REG);
388                         args[3] = hwif->INB(IDE_SECTOR_REG);
389                         args[4] = hwif->INB(IDE_LCYL_REG);
390                         args[5] = hwif->INB(IDE_HCYL_REG);
391                         args[6] = hwif->INB(IDE_SELECT_REG);
392                 }
393         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
394                 ide_task_t *args = (ide_task_t *) rq->special;
395                 if (rq->errors == 0)
396                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
397                         
398                 if (args) {
399                         if (args->tf_in_flags.b.data) {
400                                 u16 data                                = hwif->INW(IDE_DATA_REG);
401                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
402                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
403                         }
404                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
405                         /* be sure we're looking at the low order bits */
406                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
407                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
408                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
409                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
410                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
411                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
412                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
413
414                         if (drive->addressing == 1) {
415                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
416                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
417                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
418                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
419                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
420                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
421                         }
422                 }
423         } else if (blk_pm_request(rq)) {
424                 struct request_pm_state *pm = rq->end_io_data;
425 #ifdef DEBUG_PM
426                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
427                         drive->name, rq->pm->pm_step, stat, err);
428 #endif
429                 ide_complete_power_step(drive, rq, stat, err);
430                 if (pm->pm_step == ide_pm_state_completed)
431                         ide_complete_pm_request(drive, rq);
432                 return;
433         }
434
435         spin_lock_irqsave(&ide_lock, flags);
436         blkdev_dequeue_request(rq);
437         HWGROUP(drive)->rq = NULL;
438         rq->errors = err;
439         end_that_request_last(rq, !rq->errors);
440         spin_unlock_irqrestore(&ide_lock, flags);
441 }
442
443 EXPORT_SYMBOL(ide_end_drive_cmd);
444
445 /**
446  *      try_to_flush_leftover_data      -       flush junk
447  *      @drive: drive to flush
448  *
449  *      try_to_flush_leftover_data() is invoked in response to a drive
450  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
451  *      resetting the drive, this routine tries to clear the condition
452  *      by read a sector's worth of data from the drive.  Of course,
453  *      this may not help if the drive is *waiting* for data from *us*.
454  */
455 static void try_to_flush_leftover_data (ide_drive_t *drive)
456 {
457         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
458
459         if (drive->media != ide_disk)
460                 return;
461         while (i > 0) {
462                 u32 buffer[16];
463                 u32 wcount = (i > 16) ? 16 : i;
464
465                 i -= wcount;
466                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
467         }
468 }
469
470 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
471 {
472         if (rq->rq_disk) {
473                 ide_driver_t *drv;
474
475                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
476                 drv->end_request(drive, 0, 0);
477         } else
478                 ide_end_request(drive, 0, 0);
479 }
480
481 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
482 {
483         ide_hwif_t *hwif = drive->hwif;
484
485         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
486                 /* other bits are useless when BUSY */
487                 rq->errors |= ERROR_RESET;
488         } else if (stat & ERR_STAT) {
489                 /* err has different meaning on cdrom and tape */
490                 if (err == ABRT_ERR) {
491                         if (drive->select.b.lba &&
492                             /* some newer drives don't support WIN_SPECIFY */
493                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
494                                 return ide_stopped;
495                 } else if ((err & BAD_CRC) == BAD_CRC) {
496                         /* UDMA crc error, just retry the operation */
497                         drive->crc_count++;
498                 } else if (err & (BBD_ERR | ECC_ERR)) {
499                         /* retries won't help these */
500                         rq->errors = ERROR_MAX;
501                 } else if (err & TRK0_ERR) {
502                         /* help it find track zero */
503                         rq->errors |= ERROR_RECAL;
504                 }
505         }
506
507         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ && hwif->err_stops_fifo == 0)
508                 try_to_flush_leftover_data(drive);
509
510         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
511                 /* force an abort */
512                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
513
514         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
515                 ide_kill_rq(drive, rq);
516         else {
517                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
518                         ++rq->errors;
519                         return ide_do_reset(drive);
520                 }
521                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
522                         drive->special.b.recalibrate = 1;
523                 ++rq->errors;
524         }
525         return ide_stopped;
526 }
527
528 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
529 {
530         ide_hwif_t *hwif = drive->hwif;
531
532         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
533                 /* other bits are useless when BUSY */
534                 rq->errors |= ERROR_RESET;
535         } else {
536                 /* add decoding error stuff */
537         }
538
539         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
540                 /* force an abort */
541                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
542
543         if (rq->errors >= ERROR_MAX) {
544                 ide_kill_rq(drive, rq);
545         } else {
546                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
547                         ++rq->errors;
548                         return ide_do_reset(drive);
549                 }
550                 ++rq->errors;
551         }
552
553         return ide_stopped;
554 }
555
556 ide_startstop_t
557 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
558 {
559         if (drive->media == ide_disk)
560                 return ide_ata_error(drive, rq, stat, err);
561         return ide_atapi_error(drive, rq, stat, err);
562 }
563
564 EXPORT_SYMBOL_GPL(__ide_error);
565
566 /**
567  *      ide_error       -       handle an error on the IDE
568  *      @drive: drive the error occurred on
569  *      @msg: message to report
570  *      @stat: status bits
571  *
572  *      ide_error() takes action based on the error returned by the drive.
573  *      For normal I/O that may well include retries. We deal with
574  *      both new-style (taskfile) and old style command handling here.
575  *      In the case of taskfile command handling there is work left to
576  *      do
577  */
578  
579 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
580 {
581         struct request *rq;
582         u8 err;
583
584         err = ide_dump_status(drive, msg, stat);
585
586         if ((rq = HWGROUP(drive)->rq) == NULL)
587                 return ide_stopped;
588
589         /* retry only "normal" I/O: */
590         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
591                 rq->errors = 1;
592                 ide_end_drive_cmd(drive, stat, err);
593                 return ide_stopped;
594         }
595
596         if (rq->rq_disk) {
597                 ide_driver_t *drv;
598
599                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
600                 return drv->error(drive, rq, stat, err);
601         } else
602                 return __ide_error(drive, rq, stat, err);
603 }
604
605 EXPORT_SYMBOL_GPL(ide_error);
606
607 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
608 {
609         if (drive->media != ide_disk)
610                 rq->errors |= ERROR_RESET;
611
612         ide_kill_rq(drive, rq);
613
614         return ide_stopped;
615 }
616
617 EXPORT_SYMBOL_GPL(__ide_abort);
618
619 /**
620  *      ide_abort       -       abort pending IDE operations
621  *      @drive: drive the error occurred on
622  *      @msg: message to report
623  *
624  *      ide_abort kills and cleans up when we are about to do a 
625  *      host initiated reset on active commands. Longer term we
626  *      want handlers to have sensible abort handling themselves
627  *
628  *      This differs fundamentally from ide_error because in 
629  *      this case the command is doing just fine when we
630  *      blow it away.
631  */
632  
633 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
634 {
635         struct request *rq;
636
637         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
638                 return ide_stopped;
639
640         /* retry only "normal" I/O: */
641         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
642                 rq->errors = 1;
643                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
644                 return ide_stopped;
645         }
646
647         if (rq->rq_disk) {
648                 ide_driver_t *drv;
649
650                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
651                 return drv->abort(drive, rq);
652         } else
653                 return __ide_abort(drive, rq);
654 }
655
656 /**
657  *      ide_cmd         -       issue a simple drive command
658  *      @drive: drive the command is for
659  *      @cmd: command byte
660  *      @nsect: sector byte
661  *      @handler: handler for the command completion
662  *
663  *      Issue a simple drive command with interrupts.
664  *      The drive must be selected beforehand.
665  */
666
667 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
668                 ide_handler_t *handler)
669 {
670         ide_hwif_t *hwif = HWIF(drive);
671         if (IDE_CONTROL_REG)
672                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
673         SELECT_MASK(drive,0);
674         hwif->OUTB(nsect,IDE_NSECTOR_REG);
675         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
676 }
677
678 /**
679  *      drive_cmd_intr          -       drive command completion interrupt
680  *      @drive: drive the completion interrupt occurred on
681  *
682  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
683  *      We do any necessary data reading and then wait for the drive to
684  *      go non busy. At that point we may read the error data and complete
685  *      the request
686  */
687  
688 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
689 {
690         struct request *rq = HWGROUP(drive)->rq;
691         ide_hwif_t *hwif = HWIF(drive);
692         u8 *args = (u8 *) rq->buffer;
693         u8 stat = hwif->INB(IDE_STATUS_REG);
694         int retries = 10;
695
696         local_irq_enable_in_hardirq();
697         if ((stat & DRQ_STAT) && args && args[3]) {
698                 u8 io_32bit = drive->io_32bit;
699                 drive->io_32bit = 0;
700                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
701                 drive->io_32bit = io_32bit;
702                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
703                         udelay(100);
704         }
705
706         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
707                 return ide_error(drive, "drive_cmd", stat);
708                 /* calls ide_end_drive_cmd */
709         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
710         return ide_stopped;
711 }
712
713 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
714 {
715         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
716         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
717         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
718         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
719         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
720         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
721
722         task->handler = &set_geometry_intr;
723 }
724
725 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
726 {
727         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
728         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
729
730         task->handler = &recal_intr;
731 }
732
733 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
734 {
735         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
736         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
737
738         task->handler = &set_multmode_intr;
739 }
740
741 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
742 {
743         special_t *s = &drive->special;
744         ide_task_t args;
745
746         memset(&args, 0, sizeof(ide_task_t));
747         args.command_type = IDE_DRIVE_TASK_NO_DATA;
748
749         if (s->b.set_geometry) {
750                 s->b.set_geometry = 0;
751                 ide_init_specify_cmd(drive, &args);
752         } else if (s->b.recalibrate) {
753                 s->b.recalibrate = 0;
754                 ide_init_restore_cmd(drive, &args);
755         } else if (s->b.set_multmode) {
756                 s->b.set_multmode = 0;
757                 if (drive->mult_req > drive->id->max_multsect)
758                         drive->mult_req = drive->id->max_multsect;
759                 ide_init_setmult_cmd(drive, &args);
760         } else if (s->all) {
761                 int special = s->all;
762                 s->all = 0;
763                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
764                 return ide_stopped;
765         }
766
767         do_rw_taskfile(drive, &args);
768
769         return ide_started;
770 }
771
772 /**
773  *      do_special              -       issue some special commands
774  *      @drive: drive the command is for
775  *
776  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
777  *      commands to a drive.  It used to do much more, but has been scaled
778  *      back.
779  */
780
781 static ide_startstop_t do_special (ide_drive_t *drive)
782 {
783         special_t *s = &drive->special;
784
785 #ifdef DEBUG
786         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
787 #endif
788         if (s->b.set_tune) {
789                 s->b.set_tune = 0;
790                 if (HWIF(drive)->tuneproc != NULL)
791                         HWIF(drive)->tuneproc(drive, drive->tune_req);
792                 return ide_stopped;
793         } else {
794                 if (drive->media == ide_disk)
795                         return ide_disk_special(drive);
796
797                 s->all = 0;
798                 drive->mult_req = 0;
799                 return ide_stopped;
800         }
801 }
802
803 void ide_map_sg(ide_drive_t *drive, struct request *rq)
804 {
805         ide_hwif_t *hwif = drive->hwif;
806         struct scatterlist *sg = hwif->sg_table;
807
808         if (hwif->sg_mapped)    /* needed by ide-scsi */
809                 return;
810
811         if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
812                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
813         } else {
814                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
815                 hwif->sg_nents = 1;
816         }
817 }
818
819 EXPORT_SYMBOL_GPL(ide_map_sg);
820
821 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
822 {
823         ide_hwif_t *hwif = drive->hwif;
824
825         hwif->nsect = hwif->nleft = rq->nr_sectors;
826         hwif->cursg = hwif->cursg_ofs = 0;
827 }
828
829 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
830
831 /**
832  *      execute_drive_command   -       issue special drive command
833  *      @drive: the drive to issue the command on
834  *      @rq: the request structure holding the command
835  *
836  *      execute_drive_cmd() issues a special drive command,  usually 
837  *      initiated by ioctl() from the external hdparm program. The
838  *      command can be a drive command, drive task or taskfile 
839  *      operation. Weirdly you can call it with NULL to wait for
840  *      all commands to finish. Don't do this as that is due to change
841  */
842
843 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
844                 struct request *rq)
845 {
846         ide_hwif_t *hwif = HWIF(drive);
847         if (rq->flags & REQ_DRIVE_TASKFILE) {
848                 ide_task_t *args = rq->special;
849  
850                 if (!args)
851                         goto done;
852
853                 hwif->data_phase = args->data_phase;
854
855                 switch (hwif->data_phase) {
856                 case TASKFILE_MULTI_OUT:
857                 case TASKFILE_OUT:
858                 case TASKFILE_MULTI_IN:
859                 case TASKFILE_IN:
860                         ide_init_sg_cmd(drive, rq);
861                         ide_map_sg(drive, rq);
862                 default:
863                         break;
864                 }
865
866                 if (args->tf_out_flags.all != 0) 
867                         return flagged_taskfile(drive, args);
868                 return do_rw_taskfile(drive, args);
869         } else if (rq->flags & REQ_DRIVE_TASK) {
870                 u8 *args = rq->buffer;
871                 u8 sel;
872  
873                 if (!args)
874                         goto done;
875 #ifdef DEBUG
876                 printk("%s: DRIVE_TASK_CMD ", drive->name);
877                 printk("cmd=0x%02x ", args[0]);
878                 printk("fr=0x%02x ", args[1]);
879                 printk("ns=0x%02x ", args[2]);
880                 printk("sc=0x%02x ", args[3]);
881                 printk("lcyl=0x%02x ", args[4]);
882                 printk("hcyl=0x%02x ", args[5]);
883                 printk("sel=0x%02x\n", args[6]);
884 #endif
885                 hwif->OUTB(args[1], IDE_FEATURE_REG);
886                 hwif->OUTB(args[3], IDE_SECTOR_REG);
887                 hwif->OUTB(args[4], IDE_LCYL_REG);
888                 hwif->OUTB(args[5], IDE_HCYL_REG);
889                 sel = (args[6] & ~0x10);
890                 if (drive->select.b.unit)
891                         sel |= 0x10;
892                 hwif->OUTB(sel, IDE_SELECT_REG);
893                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
894                 return ide_started;
895         } else if (rq->flags & REQ_DRIVE_CMD) {
896                 u8 *args = rq->buffer;
897
898                 if (!args)
899                         goto done;
900 #ifdef DEBUG
901                 printk("%s: DRIVE_CMD ", drive->name);
902                 printk("cmd=0x%02x ", args[0]);
903                 printk("sc=0x%02x ", args[1]);
904                 printk("fr=0x%02x ", args[2]);
905                 printk("xx=0x%02x\n", args[3]);
906 #endif
907                 if (args[0] == WIN_SMART) {
908                         hwif->OUTB(0x4f, IDE_LCYL_REG);
909                         hwif->OUTB(0xc2, IDE_HCYL_REG);
910                         hwif->OUTB(args[2],IDE_FEATURE_REG);
911                         hwif->OUTB(args[1],IDE_SECTOR_REG);
912                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
913                         return ide_started;
914                 }
915                 hwif->OUTB(args[2],IDE_FEATURE_REG);
916                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
917                 return ide_started;
918         }
919
920 done:
921         /*
922          * NULL is actually a valid way of waiting for
923          * all current requests to be flushed from the queue.
924          */
925 #ifdef DEBUG
926         printk("%s: DRIVE_CMD (null)\n", drive->name);
927 #endif
928         ide_end_drive_cmd(drive,
929                         hwif->INB(IDE_STATUS_REG),
930                         hwif->INB(IDE_ERROR_REG));
931         return ide_stopped;
932 }
933
934 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
935 {
936         struct request_pm_state *pm = rq->end_io_data;
937
938         if (blk_pm_suspend_request(rq) &&
939             pm->pm_step == ide_pm_state_start_suspend)
940                 /* Mark drive blocked when starting the suspend sequence. */
941                 drive->blocked = 1;
942         else if (blk_pm_resume_request(rq) &&
943                  pm->pm_step == ide_pm_state_start_resume) {
944                 /* 
945                  * The first thing we do on wakeup is to wait for BSY bit to
946                  * go away (with a looong timeout) as a drive on this hwif may
947                  * just be POSTing itself.
948                  * We do that before even selecting as the "other" device on
949                  * the bus may be broken enough to walk on our toes at this
950                  * point.
951                  */
952                 int rc;
953 #ifdef DEBUG_PM
954                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
955 #endif
956                 rc = ide_wait_not_busy(HWIF(drive), 35000);
957                 if (rc)
958                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
959                 SELECT_DRIVE(drive);
960                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
961                 rc = ide_wait_not_busy(HWIF(drive), 100000);
962                 if (rc)
963                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
964         }
965 }
966
967 /**
968  *      start_request   -       start of I/O and command issuing for IDE
969  *
970  *      start_request() initiates handling of a new I/O request. It
971  *      accepts commands and I/O (read/write) requests. It also does
972  *      the final remapping for weird stuff like EZDrive. Once 
973  *      device mapper can work sector level the EZDrive stuff can go away
974  *
975  *      FIXME: this function needs a rename
976  */
977  
978 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
979 {
980         ide_startstop_t startstop;
981         sector_t block;
982
983         BUG_ON(!(rq->flags & REQ_STARTED));
984
985 #ifdef DEBUG
986         printk("%s: start_request: current=0x%08lx\n",
987                 HWIF(drive)->name, (unsigned long) rq);
988 #endif
989
990         /* bail early if we've exceeded max_failures */
991         if (drive->max_failures && (drive->failures > drive->max_failures)) {
992                 goto kill_rq;
993         }
994
995         block    = rq->sector;
996         if (blk_fs_request(rq) &&
997             (drive->media == ide_disk || drive->media == ide_floppy)) {
998                 block += drive->sect0;
999         }
1000         /* Yecch - this will shift the entire interval,
1001            possibly killing some innocent following sector */
1002         if (block == 0 && drive->remap_0_to_1 == 1)
1003                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
1004
1005         if (blk_pm_request(rq))
1006                 ide_check_pm_state(drive, rq);
1007
1008         SELECT_DRIVE(drive);
1009         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1010                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1011                 return startstop;
1012         }
1013         if (!drive->special.all) {
1014                 ide_driver_t *drv;
1015
1016                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
1017                         return execute_drive_cmd(drive, rq);
1018                 else if (rq->flags & REQ_DRIVE_TASKFILE)
1019                         return execute_drive_cmd(drive, rq);
1020                 else if (blk_pm_request(rq)) {
1021                         struct request_pm_state *pm = rq->end_io_data;
1022 #ifdef DEBUG_PM
1023                         printk("%s: start_power_step(step: %d)\n",
1024                                 drive->name, rq->pm->pm_step);
1025 #endif
1026                         startstop = ide_start_power_step(drive, rq);
1027                         if (startstop == ide_stopped &&
1028                             pm->pm_step == ide_pm_state_completed)
1029                                 ide_complete_pm_request(drive, rq);
1030                         return startstop;
1031                 }
1032
1033                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1034                 return drv->do_request(drive, rq, block);
1035         }
1036         return do_special(drive);
1037 kill_rq:
1038         ide_kill_rq(drive, rq);
1039         return ide_stopped;
1040 }
1041
1042 /**
1043  *      ide_stall_queue         -       pause an IDE device
1044  *      @drive: drive to stall
1045  *      @timeout: time to stall for (jiffies)
1046  *
1047  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1048  *      to the hwgroup by sleeping for timeout jiffies.
1049  */
1050  
1051 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1052 {
1053         if (timeout > WAIT_WORSTCASE)
1054                 timeout = WAIT_WORSTCASE;
1055         drive->sleep = timeout + jiffies;
1056         drive->sleeping = 1;
1057 }
1058
1059 EXPORT_SYMBOL(ide_stall_queue);
1060
1061 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1062
1063 /**
1064  *      choose_drive            -       select a drive to service
1065  *      @hwgroup: hardware group to select on
1066  *
1067  *      choose_drive() selects the next drive which will be serviced.
1068  *      This is necessary because the IDE layer can't issue commands
1069  *      to both drives on the same cable, unlike SCSI.
1070  */
1071  
1072 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1073 {
1074         ide_drive_t *drive, *best;
1075
1076 repeat: 
1077         best = NULL;
1078         drive = hwgroup->drive;
1079
1080         /*
1081          * drive is doing pre-flush, ordered write, post-flush sequence. even
1082          * though that is 3 requests, it must be seen as a single transaction.
1083          * we must not preempt this drive until that is complete
1084          */
1085         if (blk_queue_flushing(drive->queue)) {
1086                 /*
1087                  * small race where queue could get replugged during
1088                  * the 3-request flush cycle, just yank the plug since
1089                  * we want it to finish asap
1090                  */
1091                 blk_remove_plug(drive->queue);
1092                 return drive;
1093         }
1094
1095         do {
1096                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1097                     && !elv_queue_empty(drive->queue)) {
1098                         if (!best
1099                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1100                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1101                         {
1102                                 if (!blk_queue_plugged(drive->queue))
1103                                         best = drive;
1104                         }
1105                 }
1106         } while ((drive = drive->next) != hwgroup->drive);
1107         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1108                 long t = (signed long)(WAKEUP(best) - jiffies);
1109                 if (t >= WAIT_MIN_SLEEP) {
1110                 /*
1111                  * We *may* have some time to spare, but first let's see if
1112                  * someone can potentially benefit from our nice mood today..
1113                  */
1114                         drive = best->next;
1115                         do {
1116                                 if (!drive->sleeping
1117                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1118                                  && time_before(WAKEUP(drive), jiffies + t))
1119                                 {
1120                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1121                                         goto repeat;
1122                                 }
1123                         } while ((drive = drive->next) != best);
1124                 }
1125         }
1126         return best;
1127 }
1128
1129 /*
1130  * Issue a new request to a drive from hwgroup
1131  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1132  *
1133  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1134  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1135  * may have both interfaces in a single hwgroup to "serialize" access.
1136  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1137  * together into one hwgroup for serialized access.
1138  *
1139  * Note also that several hwgroups can end up sharing a single IRQ,
1140  * possibly along with many other devices.  This is especially common in
1141  * PCI-based systems with off-board IDE controller cards.
1142  *
1143  * The IDE driver uses the single global ide_lock spinlock to protect
1144  * access to the request queues, and to protect the hwgroup->busy flag.
1145  *
1146  * The first thread into the driver for a particular hwgroup sets the
1147  * hwgroup->busy flag to indicate that this hwgroup is now active,
1148  * and then initiates processing of the top request from the request queue.
1149  *
1150  * Other threads attempting entry notice the busy setting, and will simply
1151  * queue their new requests and exit immediately.  Note that hwgroup->busy
1152  * remains set even when the driver is merely awaiting the next interrupt.
1153  * Thus, the meaning is "this hwgroup is busy processing a request".
1154  *
1155  * When processing of a request completes, the completing thread or IRQ-handler
1156  * will start the next request from the queue.  If no more work remains,
1157  * the driver will clear the hwgroup->busy flag and exit.
1158  *
1159  * The ide_lock (spinlock) is used to protect all access to the
1160  * hwgroup->busy flag, but is otherwise not needed for most processing in
1161  * the driver.  This makes the driver much more friendlier to shared IRQs
1162  * than previous designs, while remaining 100% (?) SMP safe and capable.
1163  */
1164 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1165 {
1166         ide_drive_t     *drive;
1167         ide_hwif_t      *hwif;
1168         struct request  *rq;
1169         ide_startstop_t startstop;
1170         int             loops = 0;
1171
1172         /* for atari only: POSSIBLY BROKEN HERE(?) */
1173         ide_get_lock(ide_intr, hwgroup);
1174
1175         /* caller must own ide_lock */
1176         BUG_ON(!irqs_disabled());
1177
1178         while (!hwgroup->busy) {
1179                 hwgroup->busy = 1;
1180                 drive = choose_drive(hwgroup);
1181                 if (drive == NULL) {
1182                         int sleeping = 0;
1183                         unsigned long sleep = 0; /* shut up, gcc */
1184                         hwgroup->rq = NULL;
1185                         drive = hwgroup->drive;
1186                         do {
1187                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1188                                         sleeping = 1;
1189                                         sleep = drive->sleep;
1190                                 }
1191                         } while ((drive = drive->next) != hwgroup->drive);
1192                         if (sleeping) {
1193                 /*
1194                  * Take a short snooze, and then wake up this hwgroup again.
1195                  * This gives other hwgroups on the same a chance to
1196                  * play fairly with us, just in case there are big differences
1197                  * in relative throughputs.. don't want to hog the cpu too much.
1198                  */
1199                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1200                                         sleep = jiffies + WAIT_MIN_SLEEP;
1201 #if 1
1202                                 if (timer_pending(&hwgroup->timer))
1203                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1204 #endif
1205                                 /* so that ide_timer_expiry knows what to do */
1206                                 hwgroup->sleeping = 1;
1207                                 mod_timer(&hwgroup->timer, sleep);
1208                                 /* we purposely leave hwgroup->busy==1
1209                                  * while sleeping */
1210                         } else {
1211                                 /* Ugly, but how can we sleep for the lock
1212                                  * otherwise? perhaps from tq_disk?
1213                                  */
1214
1215                                 /* for atari only */
1216                                 ide_release_lock();
1217                                 hwgroup->busy = 0;
1218                         }
1219
1220                         /* no more work for this hwgroup (for now) */
1221                         return;
1222                 }
1223         again:
1224                 hwif = HWIF(drive);
1225                 if (hwgroup->hwif->sharing_irq &&
1226                     hwif != hwgroup->hwif &&
1227                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1228                         /* set nIEN for previous hwif */
1229                         SELECT_INTERRUPT(drive);
1230                 }
1231                 hwgroup->hwif = hwif;
1232                 hwgroup->drive = drive;
1233                 drive->sleeping = 0;
1234                 drive->service_start = jiffies;
1235
1236                 if (blk_queue_plugged(drive->queue)) {
1237                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1238                         break;
1239                 }
1240
1241                 /*
1242                  * we know that the queue isn't empty, but this can happen
1243                  * if the q->prep_rq_fn() decides to kill a request
1244                  */
1245                 rq = elv_next_request(drive->queue);
1246                 if (!rq) {
1247                         hwgroup->busy = 0;
1248                         break;
1249                 }
1250
1251                 /*
1252                  * Sanity: don't accept a request that isn't a PM request
1253                  * if we are currently power managed. This is very important as
1254                  * blk_stop_queue() doesn't prevent the elv_next_request()
1255                  * above to return us whatever is in the queue. Since we call
1256                  * ide_do_request() ourselves, we end up taking requests while
1257                  * the queue is blocked...
1258                  * 
1259                  * We let requests forced at head of queue with ide-preempt
1260                  * though. I hope that doesn't happen too much, hopefully not
1261                  * unless the subdriver triggers such a thing in its own PM
1262                  * state machine.
1263                  *
1264                  * We count how many times we loop here to make sure we service
1265                  * all drives in the hwgroup without looping for ever
1266                  */
1267                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1268                         drive = drive->next ? drive->next : hwgroup->drive;
1269                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1270                                 goto again;
1271                         /* We clear busy, there should be no pending ATA command at this point. */
1272                         hwgroup->busy = 0;
1273                         break;
1274                 }
1275
1276                 hwgroup->rq = rq;
1277
1278                 /*
1279                  * Some systems have trouble with IDE IRQs arriving while
1280                  * the driver is still setting things up.  So, here we disable
1281                  * the IRQ used by this interface while the request is being started.
1282                  * This may look bad at first, but pretty much the same thing
1283                  * happens anyway when any interrupt comes in, IDE or otherwise
1284                  *  -- the kernel masks the IRQ while it is being handled.
1285                  */
1286                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1287                         disable_irq_nosync(hwif->irq);
1288                 spin_unlock(&ide_lock);
1289                 local_irq_enable_in_hardirq();
1290                         /* allow other IRQs while we start this request */
1291                 startstop = start_request(drive, rq);
1292                 spin_lock_irq(&ide_lock);
1293                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1294                         enable_irq(hwif->irq);
1295                 if (startstop == ide_stopped)
1296                         hwgroup->busy = 0;
1297         }
1298 }
1299
1300 /*
1301  * Passes the stuff to ide_do_request
1302  */
1303 void do_ide_request(request_queue_t *q)
1304 {
1305         ide_drive_t *drive = q->queuedata;
1306
1307         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1308 }
1309
1310 /*
1311  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1312  * retry the current request in pio mode instead of risking tossing it
1313  * all away
1314  */
1315 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1316 {
1317         ide_hwif_t *hwif = HWIF(drive);
1318         struct request *rq;
1319         ide_startstop_t ret = ide_stopped;
1320
1321         /*
1322          * end current dma transaction
1323          */
1324
1325         if (error < 0) {
1326                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1327                 (void)HWIF(drive)->ide_dma_end(drive);
1328                 ret = ide_error(drive, "dma timeout error",
1329                                                 hwif->INB(IDE_STATUS_REG));
1330         } else {
1331                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1332                 (void) hwif->ide_dma_timeout(drive);
1333         }
1334
1335         /*
1336          * disable dma for now, but remember that we did so because of
1337          * a timeout -- we'll reenable after we finish this next request
1338          * (or rather the first chunk of it) in pio.
1339          */
1340         drive->retry_pio++;
1341         drive->state = DMA_PIO_RETRY;
1342         (void) hwif->ide_dma_off_quietly(drive);
1343
1344         /*
1345          * un-busy drive etc (hwgroup->busy is cleared on return) and
1346          * make sure request is sane
1347          */
1348         rq = HWGROUP(drive)->rq;
1349         HWGROUP(drive)->rq = NULL;
1350
1351         rq->errors = 0;
1352
1353         if (!rq->bio)
1354                 goto out;
1355
1356         rq->sector = rq->bio->bi_sector;
1357         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1358         rq->hard_cur_sectors = rq->current_nr_sectors;
1359         rq->buffer = bio_data(rq->bio);
1360 out:
1361         return ret;
1362 }
1363
1364 /**
1365  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1366  *      @data: timer callback magic (hwgroup)
1367  *
1368  *      An IDE command has timed out before the expected drive return
1369  *      occurred. At this point we attempt to clean up the current
1370  *      mess. If the current handler includes an expiry handler then
1371  *      we invoke the expiry handler, and providing it is happy the
1372  *      work is done. If that fails we apply generic recovery rules
1373  *      invoking the handler and checking the drive DMA status. We
1374  *      have an excessively incestuous relationship with the DMA
1375  *      logic that wants cleaning up.
1376  */
1377  
1378 void ide_timer_expiry (unsigned long data)
1379 {
1380         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1381         ide_handler_t   *handler;
1382         ide_expiry_t    *expiry;
1383         unsigned long   flags;
1384         unsigned long   wait = -1;
1385
1386         spin_lock_irqsave(&ide_lock, flags);
1387
1388         if ((handler = hwgroup->handler) == NULL) {
1389                 /*
1390                  * Either a marginal timeout occurred
1391                  * (got the interrupt just as timer expired),
1392                  * or we were "sleeping" to give other devices a chance.
1393                  * Either way, we don't really want to complain about anything.
1394                  */
1395                 if (hwgroup->sleeping) {
1396                         hwgroup->sleeping = 0;
1397                         hwgroup->busy = 0;
1398                 }
1399         } else {
1400                 ide_drive_t *drive = hwgroup->drive;
1401                 if (!drive) {
1402                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1403                         hwgroup->handler = NULL;
1404                 } else {
1405                         ide_hwif_t *hwif;
1406                         ide_startstop_t startstop = ide_stopped;
1407                         if (!hwgroup->busy) {
1408                                 hwgroup->busy = 1;      /* paranoia */
1409                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1410                         }
1411                         if ((expiry = hwgroup->expiry) != NULL) {
1412                                 /* continue */
1413                                 if ((wait = expiry(drive)) > 0) {
1414                                         /* reset timer */
1415                                         hwgroup->timer.expires  = jiffies + wait;
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 #if DISABLE_IRQ_NOSYNC
1430                         disable_irq_nosync(hwif->irq);
1431 #else
1432                         /* disable_irq_nosync ?? */
1433                         disable_irq(hwif->irq);
1434 #endif /* DISABLE_IRQ_NOSYNC */
1435                         /* local CPU only,
1436                          * as if we were handling an interrupt */
1437                         local_irq_disable();
1438                         if (hwgroup->polling) {
1439                                 startstop = handler(drive);
1440                         } else if (drive_is_ready(drive)) {
1441                                 if (drive->waiting_for_dma)
1442                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1443                                 (void)ide_ack_intr(hwif);
1444                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1445                                 startstop = handler(drive);
1446                         } else {
1447                                 if (drive->waiting_for_dma) {
1448                                         startstop = ide_dma_timeout_retry(drive, wait);
1449                                 } else
1450                                         startstop =
1451                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1452                         }
1453                         drive->service_time = jiffies - drive->service_start;
1454                         spin_lock_irq(&ide_lock);
1455                         enable_irq(hwif->irq);
1456                         if (startstop == ide_stopped)
1457                                 hwgroup->busy = 0;
1458                 }
1459         }
1460         ide_do_request(hwgroup, IDE_NO_IRQ);
1461         spin_unlock_irqrestore(&ide_lock, flags);
1462 }
1463
1464 /**
1465  *      unexpected_intr         -       handle an unexpected IDE interrupt
1466  *      @irq: interrupt line
1467  *      @hwgroup: hwgroup being processed
1468  *
1469  *      There's nothing really useful we can do with an unexpected interrupt,
1470  *      other than reading the status register (to clear it), and logging it.
1471  *      There should be no way that an irq can happen before we're ready for it,
1472  *      so we needn't worry much about losing an "important" interrupt here.
1473  *
1474  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1475  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1476  *      looks "good", we just ignore the interrupt completely.
1477  *
1478  *      This routine assumes __cli() is in effect when called.
1479  *
1480  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1481  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1482  *      we could screw up by interfering with a new request being set up for 
1483  *      irq15.
1484  *
1485  *      In reality, this is a non-issue.  The new command is not sent unless 
1486  *      the drive is ready to accept one, in which case we know the drive is
1487  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1488  *      before completing the issuance of any new drive command, so we will not
1489  *      be accidentally invoked as a result of any valid command completion
1490  *      interrupt.
1491  *
1492  *      Note that we must walk the entire hwgroup here. We know which hwif
1493  *      is doing the current command, but we don't know which hwif burped
1494  *      mysteriously.
1495  */
1496  
1497 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1498 {
1499         u8 stat;
1500         ide_hwif_t *hwif = hwgroup->hwif;
1501
1502         /*
1503          * handle the unexpected interrupt
1504          */
1505         do {
1506                 if (hwif->irq == irq) {
1507                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1508                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1509                                 /* Try to not flood the console with msgs */
1510                                 static unsigned long last_msgtime, count;
1511                                 ++count;
1512                                 if (time_after(jiffies, last_msgtime + HZ)) {
1513                                         last_msgtime = jiffies;
1514                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1515                                                 "status=0x%02x, count=%ld\n",
1516                                                 hwif->name,
1517                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1518                                 }
1519                         }
1520                 }
1521         } while ((hwif = hwif->next) != hwgroup->hwif);
1522 }
1523
1524 /**
1525  *      ide_intr        -       default IDE interrupt handler
1526  *      @irq: interrupt number
1527  *      @dev_id: hwif group
1528  *      @regs: unused weirdness from the kernel irq layer
1529  *
1530  *      This is the default IRQ handler for the IDE layer. You should
1531  *      not need to override it. If you do be aware it is subtle in
1532  *      places
1533  *
1534  *      hwgroup->hwif is the interface in the group currently performing
1535  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1536  *      the IRQ handler to call. As we issue a command the handlers
1537  *      step through multiple states, reassigning the handler to the
1538  *      next step in the process. Unlike a smart SCSI controller IDE
1539  *      expects the main processor to sequence the various transfer
1540  *      stages. We also manage a poll timer to catch up with most
1541  *      timeout situations. There are still a few where the handlers
1542  *      don't ever decide to give up.
1543  *
1544  *      The handler eventually returns ide_stopped to indicate the
1545  *      request completed. At this point we issue the next request
1546  *      on the hwgroup and the process begins again.
1547  */
1548  
1549 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1550 {
1551         unsigned long flags;
1552         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1553         ide_hwif_t *hwif;
1554         ide_drive_t *drive;
1555         ide_handler_t *handler;
1556         ide_startstop_t startstop;
1557
1558         spin_lock_irqsave(&ide_lock, flags);
1559         hwif = hwgroup->hwif;
1560
1561         if (!ide_ack_intr(hwif)) {
1562                 spin_unlock_irqrestore(&ide_lock, flags);
1563                 return IRQ_NONE;
1564         }
1565
1566         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1567                 /*
1568                  * Not expecting an interrupt from this drive.
1569                  * That means this could be:
1570                  *      (1) an interrupt from another PCI device
1571                  *      sharing the same PCI INT# as us.
1572                  * or   (2) a drive just entered sleep or standby mode,
1573                  *      and is interrupting to let us know.
1574                  * or   (3) a spurious interrupt of unknown origin.
1575                  *
1576                  * For PCI, we cannot tell the difference,
1577                  * so in that case we just ignore it and hope it goes away.
1578                  *
1579                  * FIXME: unexpected_intr should be hwif-> then we can
1580                  * remove all the ifdef PCI crap
1581                  */
1582 #ifdef CONFIG_BLK_DEV_IDEPCI
1583                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1584 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1585                 {
1586                         /*
1587                          * Probably not a shared PCI interrupt,
1588                          * so we can safely try to do something about it:
1589                          */
1590                         unexpected_intr(irq, hwgroup);
1591 #ifdef CONFIG_BLK_DEV_IDEPCI
1592                 } else {
1593                         /*
1594                          * Whack the status register, just in case
1595                          * we have a leftover pending IRQ.
1596                          */
1597                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1598 #endif /* CONFIG_BLK_DEV_IDEPCI */
1599                 }
1600                 spin_unlock_irqrestore(&ide_lock, flags);
1601                 return IRQ_NONE;
1602         }
1603         drive = hwgroup->drive;
1604         if (!drive) {
1605                 /*
1606                  * This should NEVER happen, and there isn't much
1607                  * we could do about it here.
1608                  *
1609                  * [Note - this can occur if the drive is hot unplugged]
1610                  */
1611                 spin_unlock_irqrestore(&ide_lock, flags);
1612                 return IRQ_HANDLED;
1613         }
1614         if (!drive_is_ready(drive)) {
1615                 /*
1616                  * This happens regularly when we share a PCI IRQ with
1617                  * another device.  Unfortunately, it can also happen
1618                  * with some buggy drives that trigger the IRQ before
1619                  * their status register is up to date.  Hopefully we have
1620                  * enough advance overhead that the latter isn't a problem.
1621                  */
1622                 spin_unlock_irqrestore(&ide_lock, flags);
1623                 return IRQ_NONE;
1624         }
1625         if (!hwgroup->busy) {
1626                 hwgroup->busy = 1;      /* paranoia */
1627                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1628         }
1629         hwgroup->handler = NULL;
1630         del_timer(&hwgroup->timer);
1631         spin_unlock(&ide_lock);
1632
1633         if (drive->unmask)
1634                 local_irq_enable_in_hardirq();
1635         /* service this interrupt, may set handler for next interrupt */
1636         startstop = handler(drive);
1637         spin_lock_irq(&ide_lock);
1638
1639         /*
1640          * Note that handler() may have set things up for another
1641          * interrupt to occur soon, but it cannot happen until
1642          * we exit from this routine, because it will be the
1643          * same irq as is currently being serviced here, and Linux
1644          * won't allow another of the same (on any CPU) until we return.
1645          */
1646         drive->service_time = jiffies - drive->service_start;
1647         if (startstop == ide_stopped) {
1648                 if (hwgroup->handler == NULL) { /* paranoia */
1649                         hwgroup->busy = 0;
1650                         ide_do_request(hwgroup, hwif->irq);
1651                 } else {
1652                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1653                                 "on exit\n", drive->name);
1654                 }
1655         }
1656         spin_unlock_irqrestore(&ide_lock, flags);
1657         return IRQ_HANDLED;
1658 }
1659
1660 /**
1661  *      ide_init_drive_cmd      -       initialize a drive command request
1662  *      @rq: request object
1663  *
1664  *      Initialize a request before we fill it in and send it down to
1665  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1666  *      now it doesn't do a lot, but if that changes abusers will have a
1667  *      nasty surprise.
1668  */
1669
1670 void ide_init_drive_cmd (struct request *rq)
1671 {
1672         memset(rq, 0, sizeof(*rq));
1673         rq->flags = REQ_DRIVE_CMD;
1674         rq->ref_count = 1;
1675 }
1676
1677 EXPORT_SYMBOL(ide_init_drive_cmd);
1678
1679 /**
1680  *      ide_do_drive_cmd        -       issue IDE special command
1681  *      @drive: device to issue command
1682  *      @rq: request to issue
1683  *      @action: action for processing
1684  *
1685  *      This function issues a special IDE device request
1686  *      onto the request queue.
1687  *
1688  *      If action is ide_wait, then the rq is queued at the end of the
1689  *      request queue, and the function sleeps until it has been processed.
1690  *      This is for use when invoked from an ioctl handler.
1691  *
1692  *      If action is ide_preempt, then the rq is queued at the head of
1693  *      the request queue, displacing the currently-being-processed
1694  *      request and this function returns immediately without waiting
1695  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1696  *      intended for careful use by the ATAPI tape/cdrom driver code.
1697  *
1698  *      If action is ide_end, then the rq is queued at the end of the
1699  *      request queue, and the function returns immediately without waiting
1700  *      for the new rq to be completed. This is again intended for careful
1701  *      use by the ATAPI tape/cdrom driver code.
1702  */
1703  
1704 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1705 {
1706         unsigned long flags;
1707         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1708         DECLARE_COMPLETION_ONSTACK(wait);
1709         int where = ELEVATOR_INSERT_BACK, err;
1710         int must_wait = (action == ide_wait || action == ide_head_wait);
1711
1712         rq->errors = 0;
1713         rq->rq_status = RQ_ACTIVE;
1714
1715         /*
1716          * we need to hold an extra reference to request for safe inspection
1717          * after completion
1718          */
1719         if (must_wait) {
1720                 rq->ref_count++;
1721                 rq->waiting = &wait;
1722                 rq->end_io = blk_end_sync_rq;
1723         }
1724
1725         spin_lock_irqsave(&ide_lock, flags);
1726         if (action == ide_preempt)
1727                 hwgroup->rq = NULL;
1728         if (action == ide_preempt || action == ide_head_wait) {
1729                 where = ELEVATOR_INSERT_FRONT;
1730                 rq->flags |= REQ_PREEMPT;
1731         }
1732         __elv_add_request(drive->queue, rq, where, 0);
1733         ide_do_request(hwgroup, IDE_NO_IRQ);
1734         spin_unlock_irqrestore(&ide_lock, flags);
1735
1736         err = 0;
1737         if (must_wait) {
1738                 wait_for_completion(&wait);
1739                 rq->waiting = NULL;
1740                 if (rq->errors)
1741                         err = -EIO;
1742
1743                 blk_put_request(rq);
1744         }
1745
1746         return err;
1747 }
1748
1749 EXPORT_SYMBOL(ide_do_drive_cmd);