Merge git://git.kernel.org/pub/scm/linux/kernel/git/bunk/trivial
[linux-2.6.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/config.h>
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/string.h>
31 #include <linux/kernel.h>
32 #include <linux/timer.h>
33 #include <linux/mm.h>
34 #include <linux/interrupt.h>
35 #include <linux/major.h>
36 #include <linux/errno.h>
37 #include <linux/genhd.h>
38 #include <linux/blkpg.h>
39 #include <linux/slab.h>
40 #include <linux/init.h>
41 #include <linux/pci.h>
42 #include <linux/delay.h>
43 #include <linux/ide.h>
44 #include <linux/completion.h>
45 #include <linux/reboot.h>
46 #include <linux/cdrom.h>
47 #include <linux/seq_file.h>
48 #include <linux/device.h>
49 #include <linux/kmod.h>
50 #include <linux/scatterlist.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/irq.h>
54 #include <asm/uaccess.h>
55 #include <asm/io.h>
56 #include <asm/bitops.h>
57
58 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
59                              int uptodate, int nr_sectors)
60 {
61         int ret = 1;
62
63         BUG_ON(!(rq->flags & REQ_STARTED));
64
65         /*
66          * if failfast is set on a request, override number of sectors and
67          * complete the whole request right now
68          */
69         if (blk_noretry_request(rq) && end_io_error(uptodate))
70                 nr_sectors = rq->hard_nr_sectors;
71
72         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
73                 rq->errors = -EIO;
74
75         /*
76          * decide whether to reenable DMA -- 3 is a random magic for now,
77          * if we DMA timeout more than 3 times, just stay in PIO
78          */
79         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
80                 drive->state = 0;
81                 HWGROUP(drive)->hwif->ide_dma_on(drive);
82         }
83
84         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
85                 add_disk_randomness(rq->rq_disk);
86                 blkdev_dequeue_request(rq);
87                 HWGROUP(drive)->rq = NULL;
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         struct request *rq;
109         unsigned long flags;
110         int ret = 1;
111
112         /*
113          * room for locking improvements here, the calls below don't
114          * need the queue lock held at all
115          */
116         spin_lock_irqsave(&ide_lock, flags);
117         rq = HWGROUP(drive)->rq;
118
119         if (!nr_sectors)
120                 nr_sectors = rq->hard_cur_sectors;
121
122         ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
123
124         spin_unlock_irqrestore(&ide_lock, flags);
125         return ret;
126 }
127 EXPORT_SYMBOL(ide_end_request);
128
129 /*
130  * Power Management state machine. This one is rather trivial for now,
131  * we should probably add more, like switching back to PIO on suspend
132  * to help some BIOSes, re-do the door locking on resume, etc...
133  */
134
135 enum {
136         ide_pm_flush_cache      = ide_pm_state_start_suspend,
137         idedisk_pm_standby,
138
139         idedisk_pm_idle         = ide_pm_state_start_resume,
140         ide_pm_restore_dma,
141 };
142
143 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
144 {
145         struct request_pm_state *pm = rq->end_io_data;
146
147         if (drive->media != ide_disk)
148                 return;
149
150         switch (pm->pm_step) {
151         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
152                 if (pm->pm_state == PM_EVENT_FREEZE)
153                         pm->pm_step = ide_pm_state_completed;
154                 else
155                         pm->pm_step = idedisk_pm_standby;
156                 break;
157         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
158                 pm->pm_step = ide_pm_state_completed;
159                 break;
160         case idedisk_pm_idle:           /* Resume step 1 (idle) complete */
161                 pm->pm_step = ide_pm_restore_dma;
162                 break;
163         }
164 }
165
166 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
167 {
168         struct request_pm_state *pm = rq->end_io_data;
169         ide_task_t *args = rq->special;
170
171         memset(args, 0, sizeof(*args));
172
173         if (drive->media != ide_disk) {
174                 /* skip idedisk_pm_idle for ATAPI devices */
175                 if (pm->pm_step == idedisk_pm_idle)
176                         pm->pm_step = ide_pm_restore_dma;
177         }
178
179         switch (pm->pm_step) {
180         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
181                 if (drive->media != ide_disk)
182                         break;
183                 /* Not supported? Switch to next step now. */
184                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
185                         ide_complete_power_step(drive, rq, 0, 0);
186                         return ide_stopped;
187                 }
188                 if (ide_id_has_flush_cache_ext(drive->id))
189                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
190                 else
191                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
192                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
193                 args->handler      = &task_no_data_intr;
194                 return do_rw_taskfile(drive, args);
195
196         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
197                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
198                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
199                 args->handler      = &task_no_data_intr;
200                 return do_rw_taskfile(drive, args);
201
202         case idedisk_pm_idle:           /* Resume step 1 (idle) */
203                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
204                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
205                 args->handler = task_no_data_intr;
206                 return do_rw_taskfile(drive, args);
207
208         case ide_pm_restore_dma:        /* Resume step 2 (restore DMA) */
209                 /*
210                  * Right now, all we do is call hwif->ide_dma_check(drive),
211                  * we could be smarter and check for current xfer_speed
212                  * in struct drive etc...
213                  */
214                 if ((drive->id->capability & 1) == 0)
215                         break;
216                 if (drive->hwif->ide_dma_check == NULL)
217                         break;
218                 drive->hwif->ide_dma_check(drive);
219                 break;
220         }
221         pm->pm_step = ide_pm_state_completed;
222         return ide_stopped;
223 }
224
225 /**
226  *      ide_end_dequeued_request        -       complete an IDE I/O
227  *      @drive: IDE device for the I/O
228  *      @uptodate:
229  *      @nr_sectors: number of sectors completed
230  *
231  *      Complete an I/O that is no longer on the request queue. This
232  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
233  *      We must still finish the old request but we must not tamper with the
234  *      queue in the meantime.
235  *
236  *      NOTE: This path does not handle barrier, but barrier is not supported
237  *      on ide-cd anyway.
238  */
239
240 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
241                              int uptodate, int nr_sectors)
242 {
243         unsigned long flags;
244         int ret = 1;
245
246         spin_lock_irqsave(&ide_lock, flags);
247
248         BUG_ON(!(rq->flags & REQ_STARTED));
249
250         /*
251          * if failfast is set on a request, override number of sectors and
252          * complete the whole request right now
253          */
254         if (blk_noretry_request(rq) && end_io_error(uptodate))
255                 nr_sectors = rq->hard_nr_sectors;
256
257         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
258                 rq->errors = -EIO;
259
260         /*
261          * decide whether to reenable DMA -- 3 is a random magic for now,
262          * if we DMA timeout more than 3 times, just stay in PIO
263          */
264         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
265                 drive->state = 0;
266                 HWGROUP(drive)->hwif->ide_dma_on(drive);
267         }
268
269         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
270                 add_disk_randomness(rq->rq_disk);
271                 if (blk_rq_tagged(rq))
272                         blk_queue_end_tag(drive->queue, rq);
273                 end_that_request_last(rq, uptodate);
274                 ret = 0;
275         }
276         spin_unlock_irqrestore(&ide_lock, flags);
277         return ret;
278 }
279 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
280
281
282 /**
283  *      ide_complete_pm_request - end the current Power Management request
284  *      @drive: target drive
285  *      @rq: request
286  *
287  *      This function cleans up the current PM request and stops the queue
288  *      if necessary.
289  */
290 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
291 {
292         unsigned long flags;
293
294 #ifdef DEBUG_PM
295         printk("%s: completing PM request, %s\n", drive->name,
296                blk_pm_suspend_request(rq) ? "suspend" : "resume");
297 #endif
298         spin_lock_irqsave(&ide_lock, flags);
299         if (blk_pm_suspend_request(rq)) {
300                 blk_stop_queue(drive->queue);
301         } else {
302                 drive->blocked = 0;
303                 blk_start_queue(drive->queue);
304         }
305         blkdev_dequeue_request(rq);
306         HWGROUP(drive)->rq = NULL;
307         end_that_request_last(rq, 1);
308         spin_unlock_irqrestore(&ide_lock, flags);
309 }
310
311 /*
312  * FIXME: probably move this somewhere else, name is bad too :)
313  */
314 u64 ide_get_error_location(ide_drive_t *drive, char *args)
315 {
316         u32 high, low;
317         u8 hcyl, lcyl, sect;
318         u64 sector;
319
320         high = 0;
321         hcyl = args[5];
322         lcyl = args[4];
323         sect = args[3];
324
325         if (ide_id_has_flush_cache_ext(drive->id)) {
326                 low = (hcyl << 16) | (lcyl << 8) | sect;
327                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
328                 high = ide_read_24(drive);
329         } else {
330                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
331                 if (cur & 0x40) {
332                         high = cur & 0xf;
333                         low = (hcyl << 16) | (lcyl << 8) | sect;
334                 } else {
335                         low = hcyl * drive->head * drive->sect;
336                         low += lcyl * drive->sect;
337                         low += sect - 1;
338                 }
339         }
340
341         sector = ((u64) high << 24) | low;
342         return sector;
343 }
344 EXPORT_SYMBOL(ide_get_error_location);
345
346 /**
347  *      ide_end_drive_cmd       -       end an explicit drive command
348  *      @drive: command 
349  *      @stat: status bits
350  *      @err: error bits
351  *
352  *      Clean up after success/failure of an explicit drive command.
353  *      These get thrown onto the queue so they are synchronized with
354  *      real I/O operations on the drive.
355  *
356  *      In LBA48 mode we have to read the register set twice to get
357  *      all the extra information out.
358  */
359  
360 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
361 {
362         ide_hwif_t *hwif = HWIF(drive);
363         unsigned long flags;
364         struct request *rq;
365
366         spin_lock_irqsave(&ide_lock, flags);
367         rq = HWGROUP(drive)->rq;
368         spin_unlock_irqrestore(&ide_lock, flags);
369
370         if (rq->flags & REQ_DRIVE_CMD) {
371                 u8 *args = (u8 *) rq->buffer;
372                 if (rq->errors == 0)
373                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
374
375                 if (args) {
376                         args[0] = stat;
377                         args[1] = err;
378                         args[2] = hwif->INB(IDE_NSECTOR_REG);
379                 }
380         } else if (rq->flags & REQ_DRIVE_TASK) {
381                 u8 *args = (u8 *) rq->buffer;
382                 if (rq->errors == 0)
383                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
384
385                 if (args) {
386                         args[0] = stat;
387                         args[1] = err;
388                         args[2] = hwif->INB(IDE_NSECTOR_REG);
389                         args[3] = hwif->INB(IDE_SECTOR_REG);
390                         args[4] = hwif->INB(IDE_LCYL_REG);
391                         args[5] = hwif->INB(IDE_HCYL_REG);
392                         args[6] = hwif->INB(IDE_SELECT_REG);
393                 }
394         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
395                 ide_task_t *args = (ide_task_t *) rq->special;
396                 if (rq->errors == 0)
397                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
398                         
399                 if (args) {
400                         if (args->tf_in_flags.b.data) {
401                                 u16 data                                = hwif->INW(IDE_DATA_REG);
402                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
403                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
404                         }
405                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
406                         /* be sure we're looking at the low order bits */
407                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
408                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
409                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
410                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
411                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
412                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
413                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
414
415                         if (drive->addressing == 1) {
416                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
417                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
418                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
419                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
420                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
421                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
422                         }
423                 }
424         } else if (blk_pm_request(rq)) {
425                 struct request_pm_state *pm = rq->end_io_data;
426 #ifdef DEBUG_PM
427                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
428                         drive->name, rq->pm->pm_step, stat, err);
429 #endif
430                 ide_complete_power_step(drive, rq, stat, err);
431                 if (pm->pm_step == ide_pm_state_completed)
432                         ide_complete_pm_request(drive, rq);
433                 return;
434         }
435
436         spin_lock_irqsave(&ide_lock, flags);
437         blkdev_dequeue_request(rq);
438         HWGROUP(drive)->rq = NULL;
439         rq->errors = err;
440         end_that_request_last(rq, !rq->errors);
441         spin_unlock_irqrestore(&ide_lock, flags);
442 }
443
444 EXPORT_SYMBOL(ide_end_drive_cmd);
445
446 /**
447  *      try_to_flush_leftover_data      -       flush junk
448  *      @drive: drive to flush
449  *
450  *      try_to_flush_leftover_data() is invoked in response to a drive
451  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
452  *      resetting the drive, this routine tries to clear the condition
453  *      by read a sector's worth of data from the drive.  Of course,
454  *      this may not help if the drive is *waiting* for data from *us*.
455  */
456 static void try_to_flush_leftover_data (ide_drive_t *drive)
457 {
458         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
459
460         if (drive->media != ide_disk)
461                 return;
462         while (i > 0) {
463                 u32 buffer[16];
464                 u32 wcount = (i > 16) ? 16 : i;
465
466                 i -= wcount;
467                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
468         }
469 }
470
471 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
472 {
473         if (rq->rq_disk) {
474                 ide_driver_t *drv;
475
476                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
477                 drv->end_request(drive, 0, 0);
478         } else
479                 ide_end_request(drive, 0, 0);
480 }
481
482 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
483 {
484         ide_hwif_t *hwif = drive->hwif;
485
486         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
487                 /* other bits are useless when BUSY */
488                 rq->errors |= ERROR_RESET;
489         } else if (stat & ERR_STAT) {
490                 /* err has different meaning on cdrom and tape */
491                 if (err == ABRT_ERR) {
492                         if (drive->select.b.lba &&
493                             /* some newer drives don't support WIN_SPECIFY */
494                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
495                                 return ide_stopped;
496                 } else if ((err & BAD_CRC) == BAD_CRC) {
497                         /* UDMA crc error, just retry the operation */
498                         drive->crc_count++;
499                 } else if (err & (BBD_ERR | ECC_ERR)) {
500                         /* retries won't help these */
501                         rq->errors = ERROR_MAX;
502                 } else if (err & TRK0_ERR) {
503                         /* help it find track zero */
504                         rq->errors |= ERROR_RECAL;
505                 }
506         }
507
508         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ)
509                 try_to_flush_leftover_data(drive);
510
511         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
512                 /* force an abort */
513                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
514
515         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
516                 ide_kill_rq(drive, rq);
517         else {
518                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
519                         ++rq->errors;
520                         return ide_do_reset(drive);
521                 }
522                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
523                         drive->special.b.recalibrate = 1;
524                 ++rq->errors;
525         }
526         return ide_stopped;
527 }
528
529 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
530 {
531         ide_hwif_t *hwif = drive->hwif;
532
533         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
534                 /* other bits are useless when BUSY */
535                 rq->errors |= ERROR_RESET;
536         } else {
537                 /* add decoding error stuff */
538         }
539
540         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
541                 /* force an abort */
542                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
543
544         if (rq->errors >= ERROR_MAX) {
545                 ide_kill_rq(drive, rq);
546         } else {
547                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
548                         ++rq->errors;
549                         return ide_do_reset(drive);
550                 }
551                 ++rq->errors;
552         }
553
554         return ide_stopped;
555 }
556
557 ide_startstop_t
558 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
559 {
560         if (drive->media == ide_disk)
561                 return ide_ata_error(drive, rq, stat, err);
562         return ide_atapi_error(drive, rq, stat, err);
563 }
564
565 EXPORT_SYMBOL_GPL(__ide_error);
566
567 /**
568  *      ide_error       -       handle an error on the IDE
569  *      @drive: drive the error occurred on
570  *      @msg: message to report
571  *      @stat: status bits
572  *
573  *      ide_error() takes action based on the error returned by the drive.
574  *      For normal I/O that may well include retries. We deal with
575  *      both new-style (taskfile) and old style command handling here.
576  *      In the case of taskfile command handling there is work left to
577  *      do
578  */
579  
580 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
581 {
582         struct request *rq;
583         u8 err;
584
585         err = ide_dump_status(drive, msg, stat);
586
587         if ((rq = HWGROUP(drive)->rq) == NULL)
588                 return ide_stopped;
589
590         /* retry only "normal" I/O: */
591         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
592                 rq->errors = 1;
593                 ide_end_drive_cmd(drive, stat, err);
594                 return ide_stopped;
595         }
596
597         if (rq->rq_disk) {
598                 ide_driver_t *drv;
599
600                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
601                 return drv->error(drive, rq, stat, err);
602         } else
603                 return __ide_error(drive, rq, stat, err);
604 }
605
606 EXPORT_SYMBOL_GPL(ide_error);
607
608 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
609 {
610         if (drive->media != ide_disk)
611                 rq->errors |= ERROR_RESET;
612
613         ide_kill_rq(drive, rq);
614
615         return ide_stopped;
616 }
617
618 EXPORT_SYMBOL_GPL(__ide_abort);
619
620 /**
621  *      ide_abort       -       abort pending IDE operations
622  *      @drive: drive the error occurred on
623  *      @msg: message to report
624  *
625  *      ide_abort kills and cleans up when we are about to do a 
626  *      host initiated reset on active commands. Longer term we
627  *      want handlers to have sensible abort handling themselves
628  *
629  *      This differs fundamentally from ide_error because in 
630  *      this case the command is doing just fine when we
631  *      blow it away.
632  */
633  
634 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
635 {
636         struct request *rq;
637
638         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
639                 return ide_stopped;
640
641         /* retry only "normal" I/O: */
642         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
643                 rq->errors = 1;
644                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
645                 return ide_stopped;
646         }
647
648         if (rq->rq_disk) {
649                 ide_driver_t *drv;
650
651                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
652                 return drv->abort(drive, rq);
653         } else
654                 return __ide_abort(drive, rq);
655 }
656
657 /**
658  *      ide_cmd         -       issue a simple drive command
659  *      @drive: drive the command is for
660  *      @cmd: command byte
661  *      @nsect: sector byte
662  *      @handler: handler for the command completion
663  *
664  *      Issue a simple drive command with interrupts.
665  *      The drive must be selected beforehand.
666  */
667
668 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
669                 ide_handler_t *handler)
670 {
671         ide_hwif_t *hwif = HWIF(drive);
672         if (IDE_CONTROL_REG)
673                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
674         SELECT_MASK(drive,0);
675         hwif->OUTB(nsect,IDE_NSECTOR_REG);
676         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
677 }
678
679 /**
680  *      drive_cmd_intr          -       drive command completion interrupt
681  *      @drive: drive the completion interrupt occurred on
682  *
683  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
684  *      We do any necessary data reading and then wait for the drive to
685  *      go non busy. At that point we may read the error data and complete
686  *      the request
687  */
688  
689 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
690 {
691         struct request *rq = HWGROUP(drive)->rq;
692         ide_hwif_t *hwif = HWIF(drive);
693         u8 *args = (u8 *) rq->buffer;
694         u8 stat = hwif->INB(IDE_STATUS_REG);
695         int retries = 10;
696
697         local_irq_enable();
698         if ((stat & DRQ_STAT) && args && args[3]) {
699                 u8 io_32bit = drive->io_32bit;
700                 drive->io_32bit = 0;
701                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
702                 drive->io_32bit = io_32bit;
703                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
704                         udelay(100);
705         }
706
707         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
708                 return ide_error(drive, "drive_cmd", stat);
709                 /* calls ide_end_drive_cmd */
710         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
711         return ide_stopped;
712 }
713
714 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
715 {
716         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
717         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
718         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
719         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
720         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
721         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
722
723         task->handler = &set_geometry_intr;
724 }
725
726 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
727 {
728         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
729         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
730
731         task->handler = &recal_intr;
732 }
733
734 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
735 {
736         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
737         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
738
739         task->handler = &set_multmode_intr;
740 }
741
742 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
743 {
744         special_t *s = &drive->special;
745         ide_task_t args;
746
747         memset(&args, 0, sizeof(ide_task_t));
748         args.command_type = IDE_DRIVE_TASK_NO_DATA;
749
750         if (s->b.set_geometry) {
751                 s->b.set_geometry = 0;
752                 ide_init_specify_cmd(drive, &args);
753         } else if (s->b.recalibrate) {
754                 s->b.recalibrate = 0;
755                 ide_init_restore_cmd(drive, &args);
756         } else if (s->b.set_multmode) {
757                 s->b.set_multmode = 0;
758                 if (drive->mult_req > drive->id->max_multsect)
759                         drive->mult_req = drive->id->max_multsect;
760                 ide_init_setmult_cmd(drive, &args);
761         } else if (s->all) {
762                 int special = s->all;
763                 s->all = 0;
764                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
765                 return ide_stopped;
766         }
767
768         do_rw_taskfile(drive, &args);
769
770         return ide_started;
771 }
772
773 /**
774  *      do_special              -       issue some special commands
775  *      @drive: drive the command is for
776  *
777  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
778  *      commands to a drive.  It used to do much more, but has been scaled
779  *      back.
780  */
781
782 static ide_startstop_t do_special (ide_drive_t *drive)
783 {
784         special_t *s = &drive->special;
785
786 #ifdef DEBUG
787         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
788 #endif
789         if (s->b.set_tune) {
790                 s->b.set_tune = 0;
791                 if (HWIF(drive)->tuneproc != NULL)
792                         HWIF(drive)->tuneproc(drive, drive->tune_req);
793                 return ide_stopped;
794         } else {
795                 if (drive->media == ide_disk)
796                         return ide_disk_special(drive);
797
798                 s->all = 0;
799                 drive->mult_req = 0;
800                 return ide_stopped;
801         }
802 }
803
804 void ide_map_sg(ide_drive_t *drive, struct request *rq)
805 {
806         ide_hwif_t *hwif = drive->hwif;
807         struct scatterlist *sg = hwif->sg_table;
808
809         if (hwif->sg_mapped)    /* needed by ide-scsi */
810                 return;
811
812         if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
813                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
814         } else {
815                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
816                 hwif->sg_nents = 1;
817         }
818 }
819
820 EXPORT_SYMBOL_GPL(ide_map_sg);
821
822 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
823 {
824         ide_hwif_t *hwif = drive->hwif;
825
826         hwif->nsect = hwif->nleft = rq->nr_sectors;
827         hwif->cursg = hwif->cursg_ofs = 0;
828 }
829
830 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
831
832 /**
833  *      execute_drive_command   -       issue special drive command
834  *      @drive: the drive to issue the command on
835  *      @rq: the request structure holding the command
836  *
837  *      execute_drive_cmd() issues a special drive command,  usually 
838  *      initiated by ioctl() from the external hdparm program. The
839  *      command can be a drive command, drive task or taskfile 
840  *      operation. Weirdly you can call it with NULL to wait for
841  *      all commands to finish. Don't do this as that is due to change
842  */
843
844 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
845                 struct request *rq)
846 {
847         ide_hwif_t *hwif = HWIF(drive);
848         if (rq->flags & REQ_DRIVE_TASKFILE) {
849                 ide_task_t *args = rq->special;
850  
851                 if (!args)
852                         goto done;
853
854                 hwif->data_phase = args->data_phase;
855
856                 switch (hwif->data_phase) {
857                 case TASKFILE_MULTI_OUT:
858                 case TASKFILE_OUT:
859                 case TASKFILE_MULTI_IN:
860                 case TASKFILE_IN:
861                         ide_init_sg_cmd(drive, rq);
862                         ide_map_sg(drive, rq);
863                 default:
864                         break;
865                 }
866
867                 if (args->tf_out_flags.all != 0) 
868                         return flagged_taskfile(drive, args);
869                 return do_rw_taskfile(drive, args);
870         } else if (rq->flags & REQ_DRIVE_TASK) {
871                 u8 *args = rq->buffer;
872                 u8 sel;
873  
874                 if (!args)
875                         goto done;
876 #ifdef DEBUG
877                 printk("%s: DRIVE_TASK_CMD ", drive->name);
878                 printk("cmd=0x%02x ", args[0]);
879                 printk("fr=0x%02x ", args[1]);
880                 printk("ns=0x%02x ", args[2]);
881                 printk("sc=0x%02x ", args[3]);
882                 printk("lcyl=0x%02x ", args[4]);
883                 printk("hcyl=0x%02x ", args[5]);
884                 printk("sel=0x%02x\n", args[6]);
885 #endif
886                 hwif->OUTB(args[1], IDE_FEATURE_REG);
887                 hwif->OUTB(args[3], IDE_SECTOR_REG);
888                 hwif->OUTB(args[4], IDE_LCYL_REG);
889                 hwif->OUTB(args[5], IDE_HCYL_REG);
890                 sel = (args[6] & ~0x10);
891                 if (drive->select.b.unit)
892                         sel |= 0x10;
893                 hwif->OUTB(sel, IDE_SELECT_REG);
894                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
895                 return ide_started;
896         } else if (rq->flags & REQ_DRIVE_CMD) {
897                 u8 *args = rq->buffer;
898
899                 if (!args)
900                         goto done;
901 #ifdef DEBUG
902                 printk("%s: DRIVE_CMD ", drive->name);
903                 printk("cmd=0x%02x ", args[0]);
904                 printk("sc=0x%02x ", args[1]);
905                 printk("fr=0x%02x ", args[2]);
906                 printk("xx=0x%02x\n", args[3]);
907 #endif
908                 if (args[0] == WIN_SMART) {
909                         hwif->OUTB(0x4f, IDE_LCYL_REG);
910                         hwif->OUTB(0xc2, IDE_HCYL_REG);
911                         hwif->OUTB(args[2],IDE_FEATURE_REG);
912                         hwif->OUTB(args[1],IDE_SECTOR_REG);
913                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
914                         return ide_started;
915                 }
916                 hwif->OUTB(args[2],IDE_FEATURE_REG);
917                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
918                 return ide_started;
919         }
920
921 done:
922         /*
923          * NULL is actually a valid way of waiting for
924          * all current requests to be flushed from the queue.
925          */
926 #ifdef DEBUG
927         printk("%s: DRIVE_CMD (null)\n", drive->name);
928 #endif
929         ide_end_drive_cmd(drive,
930                         hwif->INB(IDE_STATUS_REG),
931                         hwif->INB(IDE_ERROR_REG));
932         return ide_stopped;
933 }
934
935 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
936 {
937         struct request_pm_state *pm = rq->end_io_data;
938
939         if (blk_pm_suspend_request(rq) &&
940             pm->pm_step == ide_pm_state_start_suspend)
941                 /* Mark drive blocked when starting the suspend sequence. */
942                 drive->blocked = 1;
943         else if (blk_pm_resume_request(rq) &&
944                  pm->pm_step == ide_pm_state_start_resume) {
945                 /* 
946                  * The first thing we do on wakeup is to wait for BSY bit to
947                  * go away (with a looong timeout) as a drive on this hwif may
948                  * just be POSTing itself.
949                  * We do that before even selecting as the "other" device on
950                  * the bus may be broken enough to walk on our toes at this
951                  * point.
952                  */
953                 int rc;
954 #ifdef DEBUG_PM
955                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
956 #endif
957                 rc = ide_wait_not_busy(HWIF(drive), 35000);
958                 if (rc)
959                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
960                 SELECT_DRIVE(drive);
961                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
962                 rc = ide_wait_not_busy(HWIF(drive), 100000);
963                 if (rc)
964                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
965         }
966 }
967
968 /**
969  *      start_request   -       start of I/O and command issuing for IDE
970  *
971  *      start_request() initiates handling of a new I/O request. It
972  *      accepts commands and I/O (read/write) requests. It also does
973  *      the final remapping for weird stuff like EZDrive. Once 
974  *      device mapper can work sector level the EZDrive stuff can go away
975  *
976  *      FIXME: this function needs a rename
977  */
978  
979 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
980 {
981         ide_startstop_t startstop;
982         sector_t block;
983
984         BUG_ON(!(rq->flags & REQ_STARTED));
985
986 #ifdef DEBUG
987         printk("%s: start_request: current=0x%08lx\n",
988                 HWIF(drive)->name, (unsigned long) rq);
989 #endif
990
991         /* bail early if we've exceeded max_failures */
992         if (drive->max_failures && (drive->failures > drive->max_failures)) {
993                 goto kill_rq;
994         }
995
996         block    = rq->sector;
997         if (blk_fs_request(rq) &&
998             (drive->media == ide_disk || drive->media == ide_floppy)) {
999                 block += drive->sect0;
1000         }
1001         /* Yecch - this will shift the entire interval,
1002            possibly killing some innocent following sector */
1003         if (block == 0 && drive->remap_0_to_1 == 1)
1004                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
1005
1006         if (blk_pm_request(rq))
1007                 ide_check_pm_state(drive, rq);
1008
1009         SELECT_DRIVE(drive);
1010         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1011                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1012                 return startstop;
1013         }
1014         if (!drive->special.all) {
1015                 ide_driver_t *drv;
1016
1017                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
1018                         return execute_drive_cmd(drive, rq);
1019                 else if (rq->flags & REQ_DRIVE_TASKFILE)
1020                         return execute_drive_cmd(drive, rq);
1021                 else if (blk_pm_request(rq)) {
1022                         struct request_pm_state *pm = rq->end_io_data;
1023 #ifdef DEBUG_PM
1024                         printk("%s: start_power_step(step: %d)\n",
1025                                 drive->name, rq->pm->pm_step);
1026 #endif
1027                         startstop = ide_start_power_step(drive, rq);
1028                         if (startstop == ide_stopped &&
1029                             pm->pm_step == ide_pm_state_completed)
1030                                 ide_complete_pm_request(drive, rq);
1031                         return startstop;
1032                 }
1033
1034                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1035                 return drv->do_request(drive, rq, block);
1036         }
1037         return do_special(drive);
1038 kill_rq:
1039         ide_kill_rq(drive, rq);
1040         return ide_stopped;
1041 }
1042
1043 /**
1044  *      ide_stall_queue         -       pause an IDE device
1045  *      @drive: drive to stall
1046  *      @timeout: time to stall for (jiffies)
1047  *
1048  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1049  *      to the hwgroup by sleeping for timeout jiffies.
1050  */
1051  
1052 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1053 {
1054         if (timeout > WAIT_WORSTCASE)
1055                 timeout = WAIT_WORSTCASE;
1056         drive->sleep = timeout + jiffies;
1057         drive->sleeping = 1;
1058 }
1059
1060 EXPORT_SYMBOL(ide_stall_queue);
1061
1062 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1063
1064 /**
1065  *      choose_drive            -       select a drive to service
1066  *      @hwgroup: hardware group to select on
1067  *
1068  *      choose_drive() selects the next drive which will be serviced.
1069  *      This is necessary because the IDE layer can't issue commands
1070  *      to both drives on the same cable, unlike SCSI.
1071  */
1072  
1073 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1074 {
1075         ide_drive_t *drive, *best;
1076
1077 repeat: 
1078         best = NULL;
1079         drive = hwgroup->drive;
1080
1081         /*
1082          * drive is doing pre-flush, ordered write, post-flush sequence. even
1083          * though that is 3 requests, it must be seen as a single transaction.
1084          * we must not preempt this drive until that is complete
1085          */
1086         if (blk_queue_flushing(drive->queue)) {
1087                 /*
1088                  * small race where queue could get replugged during
1089                  * the 3-request flush cycle, just yank the plug since
1090                  * we want it to finish asap
1091                  */
1092                 blk_remove_plug(drive->queue);
1093                 return drive;
1094         }
1095
1096         do {
1097                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1098                     && !elv_queue_empty(drive->queue)) {
1099                         if (!best
1100                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1101                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1102                         {
1103                                 if (!blk_queue_plugged(drive->queue))
1104                                         best = drive;
1105                         }
1106                 }
1107         } while ((drive = drive->next) != hwgroup->drive);
1108         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1109                 long t = (signed long)(WAKEUP(best) - jiffies);
1110                 if (t >= WAIT_MIN_SLEEP) {
1111                 /*
1112                  * We *may* have some time to spare, but first let's see if
1113                  * someone can potentially benefit from our nice mood today..
1114                  */
1115                         drive = best->next;
1116                         do {
1117                                 if (!drive->sleeping
1118                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1119                                  && time_before(WAKEUP(drive), jiffies + t))
1120                                 {
1121                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1122                                         goto repeat;
1123                                 }
1124                         } while ((drive = drive->next) != best);
1125                 }
1126         }
1127         return best;
1128 }
1129
1130 /*
1131  * Issue a new request to a drive from hwgroup
1132  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1133  *
1134  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1135  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1136  * may have both interfaces in a single hwgroup to "serialize" access.
1137  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1138  * together into one hwgroup for serialized access.
1139  *
1140  * Note also that several hwgroups can end up sharing a single IRQ,
1141  * possibly along with many other devices.  This is especially common in
1142  * PCI-based systems with off-board IDE controller cards.
1143  *
1144  * The IDE driver uses the single global ide_lock spinlock to protect
1145  * access to the request queues, and to protect the hwgroup->busy flag.
1146  *
1147  * The first thread into the driver for a particular hwgroup sets the
1148  * hwgroup->busy flag to indicate that this hwgroup is now active,
1149  * and then initiates processing of the top request from the request queue.
1150  *
1151  * Other threads attempting entry notice the busy setting, and will simply
1152  * queue their new requests and exit immediately.  Note that hwgroup->busy
1153  * remains set even when the driver is merely awaiting the next interrupt.
1154  * Thus, the meaning is "this hwgroup is busy processing a request".
1155  *
1156  * When processing of a request completes, the completing thread or IRQ-handler
1157  * will start the next request from the queue.  If no more work remains,
1158  * the driver will clear the hwgroup->busy flag and exit.
1159  *
1160  * The ide_lock (spinlock) is used to protect all access to the
1161  * hwgroup->busy flag, but is otherwise not needed for most processing in
1162  * the driver.  This makes the driver much more friendlier to shared IRQs
1163  * than previous designs, while remaining 100% (?) SMP safe and capable.
1164  */
1165 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1166 {
1167         ide_drive_t     *drive;
1168         ide_hwif_t      *hwif;
1169         struct request  *rq;
1170         ide_startstop_t startstop;
1171         int             loops = 0;
1172
1173         /* for atari only: POSSIBLY BROKEN HERE(?) */
1174         ide_get_lock(ide_intr, hwgroup);
1175
1176         /* caller must own ide_lock */
1177         BUG_ON(!irqs_disabled());
1178
1179         while (!hwgroup->busy) {
1180                 hwgroup->busy = 1;
1181                 drive = choose_drive(hwgroup);
1182                 if (drive == NULL) {
1183                         int sleeping = 0;
1184                         unsigned long sleep = 0; /* shut up, gcc */
1185                         hwgroup->rq = NULL;
1186                         drive = hwgroup->drive;
1187                         do {
1188                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1189                                         sleeping = 1;
1190                                         sleep = drive->sleep;
1191                                 }
1192                         } while ((drive = drive->next) != hwgroup->drive);
1193                         if (sleeping) {
1194                 /*
1195                  * Take a short snooze, and then wake up this hwgroup again.
1196                  * This gives other hwgroups on the same a chance to
1197                  * play fairly with us, just in case there are big differences
1198                  * in relative throughputs.. don't want to hog the cpu too much.
1199                  */
1200                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1201                                         sleep = jiffies + WAIT_MIN_SLEEP;
1202 #if 1
1203                                 if (timer_pending(&hwgroup->timer))
1204                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1205 #endif
1206                                 /* so that ide_timer_expiry knows what to do */
1207                                 hwgroup->sleeping = 1;
1208                                 mod_timer(&hwgroup->timer, sleep);
1209                                 /* we purposely leave hwgroup->busy==1
1210                                  * while sleeping */
1211                         } else {
1212                                 /* Ugly, but how can we sleep for the lock
1213                                  * otherwise? perhaps from tq_disk?
1214                                  */
1215
1216                                 /* for atari only */
1217                                 ide_release_lock();
1218                                 hwgroup->busy = 0;
1219                         }
1220
1221                         /* no more work for this hwgroup (for now) */
1222                         return;
1223                 }
1224         again:
1225                 hwif = HWIF(drive);
1226                 if (hwgroup->hwif->sharing_irq &&
1227                     hwif != hwgroup->hwif &&
1228                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1229                         /* set nIEN for previous hwif */
1230                         SELECT_INTERRUPT(drive);
1231                 }
1232                 hwgroup->hwif = hwif;
1233                 hwgroup->drive = drive;
1234                 drive->sleeping = 0;
1235                 drive->service_start = jiffies;
1236
1237                 if (blk_queue_plugged(drive->queue)) {
1238                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1239                         break;
1240                 }
1241
1242                 /*
1243                  * we know that the queue isn't empty, but this can happen
1244                  * if the q->prep_rq_fn() decides to kill a request
1245                  */
1246                 rq = elv_next_request(drive->queue);
1247                 if (!rq) {
1248                         hwgroup->busy = 0;
1249                         break;
1250                 }
1251
1252                 /*
1253                  * Sanity: don't accept a request that isn't a PM request
1254                  * if we are currently power managed. This is very important as
1255                  * blk_stop_queue() doesn't prevent the elv_next_request()
1256                  * above to return us whatever is in the queue. Since we call
1257                  * ide_do_request() ourselves, we end up taking requests while
1258                  * the queue is blocked...
1259                  * 
1260                  * We let requests forced at head of queue with ide-preempt
1261                  * though. I hope that doesn't happen too much, hopefully not
1262                  * unless the subdriver triggers such a thing in its own PM
1263                  * state machine.
1264                  *
1265                  * We count how many times we loop here to make sure we service
1266                  * all drives in the hwgroup without looping for ever
1267                  */
1268                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1269                         drive = drive->next ? drive->next : hwgroup->drive;
1270                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1271                                 goto again;
1272                         /* We clear busy, there should be no pending ATA command at this point. */
1273                         hwgroup->busy = 0;
1274                         break;
1275                 }
1276
1277                 hwgroup->rq = rq;
1278
1279                 /*
1280                  * Some systems have trouble with IDE IRQs arriving while
1281                  * the driver is still setting things up.  So, here we disable
1282                  * the IRQ used by this interface while the request is being started.
1283                  * This may look bad at first, but pretty much the same thing
1284                  * happens anyway when any interrupt comes in, IDE or otherwise
1285                  *  -- the kernel masks the IRQ while it is being handled.
1286                  */
1287                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1288                         disable_irq_nosync(hwif->irq);
1289                 spin_unlock(&ide_lock);
1290                 local_irq_enable();
1291                         /* allow other IRQs while we start this request */
1292                 startstop = start_request(drive, rq);
1293                 spin_lock_irq(&ide_lock);
1294                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1295                         enable_irq(hwif->irq);
1296                 if (startstop == ide_stopped)
1297                         hwgroup->busy = 0;
1298         }
1299 }
1300
1301 /*
1302  * Passes the stuff to ide_do_request
1303  */
1304 void do_ide_request(request_queue_t *q)
1305 {
1306         ide_drive_t *drive = q->queuedata;
1307
1308         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1309 }
1310
1311 /*
1312  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1313  * retry the current request in pio mode instead of risking tossing it
1314  * all away
1315  */
1316 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1317 {
1318         ide_hwif_t *hwif = HWIF(drive);
1319         struct request *rq;
1320         ide_startstop_t ret = ide_stopped;
1321
1322         /*
1323          * end current dma transaction
1324          */
1325
1326         if (error < 0) {
1327                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1328                 (void)HWIF(drive)->ide_dma_end(drive);
1329                 ret = ide_error(drive, "dma timeout error",
1330                                                 hwif->INB(IDE_STATUS_REG));
1331         } else {
1332                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1333                 (void) hwif->ide_dma_timeout(drive);
1334         }
1335
1336         /*
1337          * disable dma for now, but remember that we did so because of
1338          * a timeout -- we'll reenable after we finish this next request
1339          * (or rather the first chunk of it) in pio.
1340          */
1341         drive->retry_pio++;
1342         drive->state = DMA_PIO_RETRY;
1343         (void) hwif->ide_dma_off_quietly(drive);
1344
1345         /*
1346          * un-busy drive etc (hwgroup->busy is cleared on return) and
1347          * make sure request is sane
1348          */
1349         rq = HWGROUP(drive)->rq;
1350         HWGROUP(drive)->rq = NULL;
1351
1352         rq->errors = 0;
1353
1354         if (!rq->bio)
1355                 goto out;
1356
1357         rq->sector = rq->bio->bi_sector;
1358         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1359         rq->hard_cur_sectors = rq->current_nr_sectors;
1360         rq->buffer = bio_data(rq->bio);
1361 out:
1362         return ret;
1363 }
1364
1365 /**
1366  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1367  *      @data: timer callback magic (hwgroup)
1368  *
1369  *      An IDE command has timed out before the expected drive return
1370  *      occurred. At this point we attempt to clean up the current
1371  *      mess. If the current handler includes an expiry handler then
1372  *      we invoke the expiry handler, and providing it is happy the
1373  *      work is done. If that fails we apply generic recovery rules
1374  *      invoking the handler and checking the drive DMA status. We
1375  *      have an excessively incestuous relationship with the DMA
1376  *      logic that wants cleaning up.
1377  */
1378  
1379 void ide_timer_expiry (unsigned long data)
1380 {
1381         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1382         ide_handler_t   *handler;
1383         ide_expiry_t    *expiry;
1384         unsigned long   flags;
1385         unsigned long   wait = -1;
1386
1387         spin_lock_irqsave(&ide_lock, flags);
1388
1389         if ((handler = hwgroup->handler) == NULL) {
1390                 /*
1391                  * Either a marginal timeout occurred
1392                  * (got the interrupt just as timer expired),
1393                  * or we were "sleeping" to give other devices a chance.
1394                  * Either way, we don't really want to complain about anything.
1395                  */
1396                 if (hwgroup->sleeping) {
1397                         hwgroup->sleeping = 0;
1398                         hwgroup->busy = 0;
1399                 }
1400         } else {
1401                 ide_drive_t *drive = hwgroup->drive;
1402                 if (!drive) {
1403                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1404                         hwgroup->handler = NULL;
1405                 } else {
1406                         ide_hwif_t *hwif;
1407                         ide_startstop_t startstop = ide_stopped;
1408                         if (!hwgroup->busy) {
1409                                 hwgroup->busy = 1;      /* paranoia */
1410                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1411                         }
1412                         if ((expiry = hwgroup->expiry) != NULL) {
1413                                 /* continue */
1414                                 if ((wait = expiry(drive)) > 0) {
1415                                         /* reset timer */
1416                                         hwgroup->timer.expires  = jiffies + wait;
1417                                         add_timer(&hwgroup->timer);
1418                                         spin_unlock_irqrestore(&ide_lock, flags);
1419                                         return;
1420                                 }
1421                         }
1422                         hwgroup->handler = NULL;
1423                         /*
1424                          * We need to simulate a real interrupt when invoking
1425                          * the handler() function, which means we need to
1426                          * globally mask the specific IRQ:
1427                          */
1428                         spin_unlock(&ide_lock);
1429                         hwif  = HWIF(drive);
1430 #if DISABLE_IRQ_NOSYNC
1431                         disable_irq_nosync(hwif->irq);
1432 #else
1433                         /* disable_irq_nosync ?? */
1434                         disable_irq(hwif->irq);
1435 #endif /* DISABLE_IRQ_NOSYNC */
1436                         /* local CPU only,
1437                          * as if we were handling an interrupt */
1438                         local_irq_disable();
1439                         if (hwgroup->polling) {
1440                                 startstop = handler(drive);
1441                         } else if (drive_is_ready(drive)) {
1442                                 if (drive->waiting_for_dma)
1443                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1444                                 (void)ide_ack_intr(hwif);
1445                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1446                                 startstop = handler(drive);
1447                         } else {
1448                                 if (drive->waiting_for_dma) {
1449                                         startstop = ide_dma_timeout_retry(drive, wait);
1450                                 } else
1451                                         startstop =
1452                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1453                         }
1454                         drive->service_time = jiffies - drive->service_start;
1455                         spin_lock_irq(&ide_lock);
1456                         enable_irq(hwif->irq);
1457                         if (startstop == ide_stopped)
1458                                 hwgroup->busy = 0;
1459                 }
1460         }
1461         ide_do_request(hwgroup, IDE_NO_IRQ);
1462         spin_unlock_irqrestore(&ide_lock, flags);
1463 }
1464
1465 /**
1466  *      unexpected_intr         -       handle an unexpected IDE interrupt
1467  *      @irq: interrupt line
1468  *      @hwgroup: hwgroup being processed
1469  *
1470  *      There's nothing really useful we can do with an unexpected interrupt,
1471  *      other than reading the status register (to clear it), and logging it.
1472  *      There should be no way that an irq can happen before we're ready for it,
1473  *      so we needn't worry much about losing an "important" interrupt here.
1474  *
1475  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1476  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1477  *      looks "good", we just ignore the interrupt completely.
1478  *
1479  *      This routine assumes __cli() is in effect when called.
1480  *
1481  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1482  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1483  *      we could screw up by interfering with a new request being set up for 
1484  *      irq15.
1485  *
1486  *      In reality, this is a non-issue.  The new command is not sent unless 
1487  *      the drive is ready to accept one, in which case we know the drive is
1488  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1489  *      before completing the issuance of any new drive command, so we will not
1490  *      be accidentally invoked as a result of any valid command completion
1491  *      interrupt.
1492  *
1493  *      Note that we must walk the entire hwgroup here. We know which hwif
1494  *      is doing the current command, but we don't know which hwif burped
1495  *      mysteriously.
1496  */
1497  
1498 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1499 {
1500         u8 stat;
1501         ide_hwif_t *hwif = hwgroup->hwif;
1502
1503         /*
1504          * handle the unexpected interrupt
1505          */
1506         do {
1507                 if (hwif->irq == irq) {
1508                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1509                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1510                                 /* Try to not flood the console with msgs */
1511                                 static unsigned long last_msgtime, count;
1512                                 ++count;
1513                                 if (time_after(jiffies, last_msgtime + HZ)) {
1514                                         last_msgtime = jiffies;
1515                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1516                                                 "status=0x%02x, count=%ld\n",
1517                                                 hwif->name,
1518                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1519                                 }
1520                         }
1521                 }
1522         } while ((hwif = hwif->next) != hwgroup->hwif);
1523 }
1524
1525 /**
1526  *      ide_intr        -       default IDE interrupt handler
1527  *      @irq: interrupt number
1528  *      @dev_id: hwif group
1529  *      @regs: unused weirdness from the kernel irq layer
1530  *
1531  *      This is the default IRQ handler for the IDE layer. You should
1532  *      not need to override it. If you do be aware it is subtle in
1533  *      places
1534  *
1535  *      hwgroup->hwif is the interface in the group currently performing
1536  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1537  *      the IRQ handler to call. As we issue a command the handlers
1538  *      step through multiple states, reassigning the handler to the
1539  *      next step in the process. Unlike a smart SCSI controller IDE
1540  *      expects the main processor to sequence the various transfer
1541  *      stages. We also manage a poll timer to catch up with most
1542  *      timeout situations. There are still a few where the handlers
1543  *      don't ever decide to give up.
1544  *
1545  *      The handler eventually returns ide_stopped to indicate the
1546  *      request completed. At this point we issue the next request
1547  *      on the hwgroup and the process begins again.
1548  */
1549  
1550 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1551 {
1552         unsigned long flags;
1553         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1554         ide_hwif_t *hwif;
1555         ide_drive_t *drive;
1556         ide_handler_t *handler;
1557         ide_startstop_t startstop;
1558
1559         spin_lock_irqsave(&ide_lock, flags);
1560         hwif = hwgroup->hwif;
1561
1562         if (!ide_ack_intr(hwif)) {
1563                 spin_unlock_irqrestore(&ide_lock, flags);
1564                 return IRQ_NONE;
1565         }
1566
1567         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1568                 /*
1569                  * Not expecting an interrupt from this drive.
1570                  * That means this could be:
1571                  *      (1) an interrupt from another PCI device
1572                  *      sharing the same PCI INT# as us.
1573                  * or   (2) a drive just entered sleep or standby mode,
1574                  *      and is interrupting to let us know.
1575                  * or   (3) a spurious interrupt of unknown origin.
1576                  *
1577                  * For PCI, we cannot tell the difference,
1578                  * so in that case we just ignore it and hope it goes away.
1579                  *
1580                  * FIXME: unexpected_intr should be hwif-> then we can
1581                  * remove all the ifdef PCI crap
1582                  */
1583 #ifdef CONFIG_BLK_DEV_IDEPCI
1584                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1585 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1586                 {
1587                         /*
1588                          * Probably not a shared PCI interrupt,
1589                          * so we can safely try to do something about it:
1590                          */
1591                         unexpected_intr(irq, hwgroup);
1592 #ifdef CONFIG_BLK_DEV_IDEPCI
1593                 } else {
1594                         /*
1595                          * Whack the status register, just in case
1596                          * we have a leftover pending IRQ.
1597                          */
1598                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1599 #endif /* CONFIG_BLK_DEV_IDEPCI */
1600                 }
1601                 spin_unlock_irqrestore(&ide_lock, flags);
1602                 return IRQ_NONE;
1603         }
1604         drive = hwgroup->drive;
1605         if (!drive) {
1606                 /*
1607                  * This should NEVER happen, and there isn't much
1608                  * we could do about it here.
1609                  *
1610                  * [Note - this can occur if the drive is hot unplugged]
1611                  */
1612                 spin_unlock_irqrestore(&ide_lock, flags);
1613                 return IRQ_HANDLED;
1614         }
1615         if (!drive_is_ready(drive)) {
1616                 /*
1617                  * This happens regularly when we share a PCI IRQ with
1618                  * another device.  Unfortunately, it can also happen
1619                  * with some buggy drives that trigger the IRQ before
1620                  * their status register is up to date.  Hopefully we have
1621                  * enough advance overhead that the latter isn't a problem.
1622                  */
1623                 spin_unlock_irqrestore(&ide_lock, flags);
1624                 return IRQ_NONE;
1625         }
1626         if (!hwgroup->busy) {
1627                 hwgroup->busy = 1;      /* paranoia */
1628                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1629         }
1630         hwgroup->handler = NULL;
1631         del_timer(&hwgroup->timer);
1632         spin_unlock(&ide_lock);
1633
1634         if (drive->unmask)
1635                 local_irq_enable();
1636         /* service this interrupt, may set handler for next interrupt */
1637         startstop = handler(drive);
1638         spin_lock_irq(&ide_lock);
1639
1640         /*
1641          * Note that handler() may have set things up for another
1642          * interrupt to occur soon, but it cannot happen until
1643          * we exit from this routine, because it will be the
1644          * same irq as is currently being serviced here, and Linux
1645          * won't allow another of the same (on any CPU) until we return.
1646          */
1647         drive->service_time = jiffies - drive->service_start;
1648         if (startstop == ide_stopped) {
1649                 if (hwgroup->handler == NULL) { /* paranoia */
1650                         hwgroup->busy = 0;
1651                         ide_do_request(hwgroup, hwif->irq);
1652                 } else {
1653                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1654                                 "on exit\n", drive->name);
1655                 }
1656         }
1657         spin_unlock_irqrestore(&ide_lock, flags);
1658         return IRQ_HANDLED;
1659 }
1660
1661 /**
1662  *      ide_init_drive_cmd      -       initialize a drive command request
1663  *      @rq: request object
1664  *
1665  *      Initialize a request before we fill it in and send it down to
1666  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1667  *      now it doesn't do a lot, but if that changes abusers will have a
1668  *      nasty surprise.
1669  */
1670
1671 void ide_init_drive_cmd (struct request *rq)
1672 {
1673         memset(rq, 0, sizeof(*rq));
1674         rq->flags = REQ_DRIVE_CMD;
1675         rq->ref_count = 1;
1676 }
1677
1678 EXPORT_SYMBOL(ide_init_drive_cmd);
1679
1680 /**
1681  *      ide_do_drive_cmd        -       issue IDE special command
1682  *      @drive: device to issue command
1683  *      @rq: request to issue
1684  *      @action: action for processing
1685  *
1686  *      This function issues a special IDE device request
1687  *      onto the request queue.
1688  *
1689  *      If action is ide_wait, then the rq is queued at the end of the
1690  *      request queue, and the function sleeps until it has been processed.
1691  *      This is for use when invoked from an ioctl handler.
1692  *
1693  *      If action is ide_preempt, then the rq is queued at the head of
1694  *      the request queue, displacing the currently-being-processed
1695  *      request and this function returns immediately without waiting
1696  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1697  *      intended for careful use by the ATAPI tape/cdrom driver code.
1698  *
1699  *      If action is ide_end, then the rq is queued at the end of the
1700  *      request queue, and the function returns immediately without waiting
1701  *      for the new rq to be completed. This is again intended for careful
1702  *      use by the ATAPI tape/cdrom driver code.
1703  */
1704  
1705 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1706 {
1707         unsigned long flags;
1708         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1709         DECLARE_COMPLETION(wait);
1710         int where = ELEVATOR_INSERT_BACK, err;
1711         int must_wait = (action == ide_wait || action == ide_head_wait);
1712
1713         rq->errors = 0;
1714         rq->rq_status = RQ_ACTIVE;
1715
1716         /*
1717          * we need to hold an extra reference to request for safe inspection
1718          * after completion
1719          */
1720         if (must_wait) {
1721                 rq->ref_count++;
1722                 rq->waiting = &wait;
1723                 rq->end_io = blk_end_sync_rq;
1724         }
1725
1726         spin_lock_irqsave(&ide_lock, flags);
1727         if (action == ide_preempt)
1728                 hwgroup->rq = NULL;
1729         if (action == ide_preempt || action == ide_head_wait) {
1730                 where = ELEVATOR_INSERT_FRONT;
1731                 rq->flags |= REQ_PREEMPT;
1732         }
1733         __elv_add_request(drive->queue, rq, where, 0);
1734         ide_do_request(hwgroup, IDE_NO_IRQ);
1735         spin_unlock_irqrestore(&ide_lock, flags);
1736
1737         err = 0;
1738         if (must_wait) {
1739                 wait_for_completion(&wait);
1740                 rq->waiting = NULL;
1741                 if (rq->errors)
1742                         err = -EIO;
1743
1744                 blk_put_request(rq);
1745         }
1746
1747         return err;
1748 }
1749
1750 EXPORT_SYMBOL(ide_do_drive_cmd);