Merge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-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 #include <linux/bitops.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/irq.h>
54 #include <asm/uaccess.h>
55 #include <asm/io.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, unsigned int nr_bytes, int dequeue)
59 {
60         int ret = 1;
61
62         /*
63          * if failfast is set on a request, override number of sectors and
64          * complete the whole request right now
65          */
66         if (blk_noretry_request(rq) && end_io_error(uptodate))
67                 nr_bytes = rq->hard_nr_sectors << 9;
68
69         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
70                 rq->errors = -EIO;
71
72         /*
73          * decide whether to reenable DMA -- 3 is a random magic for now,
74          * if we DMA timeout more than 3 times, just stay in PIO
75          */
76         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
77                 drive->state = 0;
78                 HWGROUP(drive)->hwif->ide_dma_on(drive);
79         }
80
81         if (!end_that_request_chunk(rq, uptodate, nr_bytes)) {
82                 add_disk_randomness(rq->rq_disk);
83                 if (dequeue) {
84                         if (!list_empty(&rq->queuelist))
85                                 blkdev_dequeue_request(rq);
86                         HWGROUP(drive)->rq = NULL;
87                 }
88                 end_that_request_last(rq, uptodate);
89                 ret = 0;
90         }
91
92         return ret;
93 }
94
95 /**
96  *      ide_end_request         -       complete an IDE I/O
97  *      @drive: IDE device for the I/O
98  *      @uptodate:
99  *      @nr_sectors: number of sectors completed
100  *
101  *      This is our end_request wrapper function. We complete the I/O
102  *      update random number input and dequeue the request, which if
103  *      it was tagged may be out of order.
104  */
105
106 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
107 {
108         unsigned int nr_bytes = nr_sectors << 9;
109         struct request *rq;
110         unsigned long flags;
111         int ret = 1;
112
113         /*
114          * room for locking improvements here, the calls below don't
115          * need the queue lock held at all
116          */
117         spin_lock_irqsave(&ide_lock, flags);
118         rq = HWGROUP(drive)->rq;
119
120         if (!nr_bytes) {
121                 if (blk_pc_request(rq))
122                         nr_bytes = rq->data_len;
123                 else
124                         nr_bytes = rq->hard_cur_sectors << 9;
125         }
126
127         ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
128
129         spin_unlock_irqrestore(&ide_lock, flags);
130         return ret;
131 }
132 EXPORT_SYMBOL(ide_end_request);
133
134 /*
135  * Power Management state machine. This one is rather trivial for now,
136  * we should probably add more, like switching back to PIO on suspend
137  * to help some BIOSes, re-do the door locking on resume, etc...
138  */
139
140 enum {
141         ide_pm_flush_cache      = ide_pm_state_start_suspend,
142         idedisk_pm_standby,
143
144         idedisk_pm_restore_pio  = ide_pm_state_start_resume,
145         idedisk_pm_idle,
146         ide_pm_restore_dma,
147 };
148
149 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
150 {
151         struct request_pm_state *pm = rq->data;
152
153         if (drive->media != ide_disk)
154                 return;
155
156         switch (pm->pm_step) {
157         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
158                 if (pm->pm_state == PM_EVENT_FREEZE)
159                         pm->pm_step = ide_pm_state_completed;
160                 else
161                         pm->pm_step = idedisk_pm_standby;
162                 break;
163         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
164                 pm->pm_step = ide_pm_state_completed;
165                 break;
166         case idedisk_pm_restore_pio:    /* Resume step 1 complete */
167                 pm->pm_step = idedisk_pm_idle;
168                 break;
169         case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
170                 pm->pm_step = ide_pm_restore_dma;
171                 break;
172         }
173 }
174
175 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
176 {
177         struct request_pm_state *pm = rq->data;
178         ide_task_t *args = rq->special;
179
180         memset(args, 0, sizeof(*args));
181
182         switch (pm->pm_step) {
183         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
184                 if (drive->media != ide_disk)
185                         break;
186                 /* Not supported? Switch to next step now. */
187                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
188                         ide_complete_power_step(drive, rq, 0, 0);
189                         return ide_stopped;
190                 }
191                 if (ide_id_has_flush_cache_ext(drive->id))
192                         args->tf.command = WIN_FLUSH_CACHE_EXT;
193                 else
194                         args->tf.command = WIN_FLUSH_CACHE;
195                 goto out_do_tf;
196
197         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
198                 args->tf.command = WIN_STANDBYNOW1;
199                 goto out_do_tf;
200
201         case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
202                 ide_set_max_pio(drive);
203                 /*
204                  * skip idedisk_pm_idle for ATAPI devices
205                  */
206                 if (drive->media != ide_disk)
207                         pm->pm_step = ide_pm_restore_dma;
208                 else
209                         ide_complete_power_step(drive, rq, 0, 0);
210                 return ide_stopped;
211
212         case idedisk_pm_idle:           /* Resume step 2 (idle) */
213                 args->tf.command = WIN_IDLEIMMEDIATE;
214                 goto out_do_tf;
215
216         case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
217                 /*
218                  * Right now, all we do is call ide_set_dma(drive),
219                  * we could be smarter and check for current xfer_speed
220                  * in struct drive etc...
221                  */
222                 if (drive->hwif->ide_dma_on == NULL)
223                         break;
224                 /*
225                  * TODO: respect ->using_dma setting
226                  */
227                 ide_set_dma(drive);
228                 break;
229         }
230         pm->pm_step = ide_pm_state_completed;
231         return ide_stopped;
232
233 out_do_tf:
234         args->tf_flags   = IDE_TFLAG_OUT_TF | IDE_TFLAG_OUT_DEVICE;
235         args->data_phase = TASKFILE_NO_DATA;
236         return do_rw_taskfile(drive, args);
237 }
238
239 /**
240  *      ide_end_dequeued_request        -       complete an IDE I/O
241  *      @drive: IDE device for the I/O
242  *      @uptodate:
243  *      @nr_sectors: number of sectors completed
244  *
245  *      Complete an I/O that is no longer on the request queue. This
246  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
247  *      We must still finish the old request but we must not tamper with the
248  *      queue in the meantime.
249  *
250  *      NOTE: This path does not handle barrier, but barrier is not supported
251  *      on ide-cd anyway.
252  */
253
254 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
255                              int uptodate, int nr_sectors)
256 {
257         unsigned long flags;
258         int ret;
259
260         spin_lock_irqsave(&ide_lock, flags);
261         BUG_ON(!blk_rq_started(rq));
262         ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
263         spin_unlock_irqrestore(&ide_lock, flags);
264
265         return ret;
266 }
267 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
268
269
270 /**
271  *      ide_complete_pm_request - end the current Power Management request
272  *      @drive: target drive
273  *      @rq: request
274  *
275  *      This function cleans up the current PM request and stops the queue
276  *      if necessary.
277  */
278 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
279 {
280         unsigned long flags;
281
282 #ifdef DEBUG_PM
283         printk("%s: completing PM request, %s\n", drive->name,
284                blk_pm_suspend_request(rq) ? "suspend" : "resume");
285 #endif
286         spin_lock_irqsave(&ide_lock, flags);
287         if (blk_pm_suspend_request(rq)) {
288                 blk_stop_queue(drive->queue);
289         } else {
290                 drive->blocked = 0;
291                 blk_start_queue(drive->queue);
292         }
293         blkdev_dequeue_request(rq);
294         HWGROUP(drive)->rq = NULL;
295         end_that_request_last(rq, 1);
296         spin_unlock_irqrestore(&ide_lock, flags);
297 }
298
299 void ide_tf_read(ide_drive_t *drive, ide_task_t *task)
300 {
301         ide_hwif_t *hwif = drive->hwif;
302         struct ide_taskfile *tf = &task->tf;
303
304         if (task->tf_flags & IDE_TFLAG_IN_DATA) {
305                 u16 data = hwif->INW(IDE_DATA_REG);
306
307                 tf->data = data & 0xff;
308                 tf->hob_data = (data >> 8) & 0xff;
309         }
310
311         /* be sure we're looking at the low order bits */
312         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
313
314         if (task->tf_flags & IDE_TFLAG_IN_NSECT)
315                 tf->nsect  = hwif->INB(IDE_NSECTOR_REG);
316         if (task->tf_flags & IDE_TFLAG_IN_LBAL)
317                 tf->lbal   = hwif->INB(IDE_SECTOR_REG);
318         if (task->tf_flags & IDE_TFLAG_IN_LBAM)
319                 tf->lbam   = hwif->INB(IDE_LCYL_REG);
320         if (task->tf_flags & IDE_TFLAG_IN_LBAH)
321                 tf->lbah   = hwif->INB(IDE_HCYL_REG);
322         if (task->tf_flags & IDE_TFLAG_IN_DEVICE)
323                 tf->device = hwif->INB(IDE_SELECT_REG);
324
325         if (task->tf_flags & IDE_TFLAG_LBA48) {
326                 hwif->OUTB(drive->ctl | 0x80, IDE_CONTROL_REG);
327
328                 if (task->tf_flags & IDE_TFLAG_IN_HOB_FEATURE)
329                         tf->hob_feature = hwif->INB(IDE_FEATURE_REG);
330                 if (task->tf_flags & IDE_TFLAG_IN_HOB_NSECT)
331                         tf->hob_nsect   = hwif->INB(IDE_NSECTOR_REG);
332                 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAL)
333                         tf->hob_lbal    = hwif->INB(IDE_SECTOR_REG);
334                 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAM)
335                         tf->hob_lbam    = hwif->INB(IDE_LCYL_REG);
336                 if (task->tf_flags & IDE_TFLAG_IN_HOB_LBAH)
337                         tf->hob_lbah    = hwif->INB(IDE_HCYL_REG);
338         }
339 }
340
341 /**
342  *      ide_end_drive_cmd       -       end an explicit drive command
343  *      @drive: command 
344  *      @stat: status bits
345  *      @err: error bits
346  *
347  *      Clean up after success/failure of an explicit drive command.
348  *      These get thrown onto the queue so they are synchronized with
349  *      real I/O operations on the drive.
350  *
351  *      In LBA48 mode we have to read the register set twice to get
352  *      all the extra information out.
353  */
354  
355 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
356 {
357         ide_hwif_t *hwif = HWIF(drive);
358         unsigned long flags;
359         struct request *rq;
360
361         spin_lock_irqsave(&ide_lock, flags);
362         rq = HWGROUP(drive)->rq;
363         spin_unlock_irqrestore(&ide_lock, flags);
364
365         if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
366                 u8 *args = (u8 *) rq->buffer;
367                 if (rq->errors == 0)
368                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
369
370                 if (args) {
371                         args[0] = stat;
372                         args[1] = err;
373                         args[2] = hwif->INB(IDE_NSECTOR_REG);
374                 }
375         } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
376                 ide_task_t *args = (ide_task_t *) rq->special;
377                 if (rq->errors == 0)
378                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
379                         
380                 if (args) {
381                         struct ide_taskfile *tf = &args->tf;
382
383                         tf->error = err;
384                         tf->status = stat;
385
386                         args->tf_flags |= (IDE_TFLAG_IN_TF|IDE_TFLAG_IN_DEVICE);
387                         if (args->tf_flags & IDE_TFLAG_LBA48)
388                                 args->tf_flags |= IDE_TFLAG_IN_HOB;
389
390                         ide_tf_read(drive, args);
391                 }
392         } else if (blk_pm_request(rq)) {
393                 struct request_pm_state *pm = rq->data;
394 #ifdef DEBUG_PM
395                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
396                         drive->name, rq->pm->pm_step, stat, err);
397 #endif
398                 ide_complete_power_step(drive, rq, stat, err);
399                 if (pm->pm_step == ide_pm_state_completed)
400                         ide_complete_pm_request(drive, rq);
401                 return;
402         }
403
404         spin_lock_irqsave(&ide_lock, flags);
405         blkdev_dequeue_request(rq);
406         HWGROUP(drive)->rq = NULL;
407         rq->errors = err;
408         end_that_request_last(rq, !rq->errors);
409         spin_unlock_irqrestore(&ide_lock, flags);
410 }
411
412 EXPORT_SYMBOL(ide_end_drive_cmd);
413
414 /**
415  *      try_to_flush_leftover_data      -       flush junk
416  *      @drive: drive to flush
417  *
418  *      try_to_flush_leftover_data() is invoked in response to a drive
419  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
420  *      resetting the drive, this routine tries to clear the condition
421  *      by read a sector's worth of data from the drive.  Of course,
422  *      this may not help if the drive is *waiting* for data from *us*.
423  */
424 static void try_to_flush_leftover_data (ide_drive_t *drive)
425 {
426         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
427
428         if (drive->media != ide_disk)
429                 return;
430         while (i > 0) {
431                 u32 buffer[16];
432                 u32 wcount = (i > 16) ? 16 : i;
433
434                 i -= wcount;
435                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
436         }
437 }
438
439 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
440 {
441         if (rq->rq_disk) {
442                 ide_driver_t *drv;
443
444                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
445                 drv->end_request(drive, 0, 0);
446         } else
447                 ide_end_request(drive, 0, 0);
448 }
449
450 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
451 {
452         ide_hwif_t *hwif = drive->hwif;
453
454         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
455                 /* other bits are useless when BUSY */
456                 rq->errors |= ERROR_RESET;
457         } else if (stat & ERR_STAT) {
458                 /* err has different meaning on cdrom and tape */
459                 if (err == ABRT_ERR) {
460                         if (drive->select.b.lba &&
461                             /* some newer drives don't support WIN_SPECIFY */
462                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
463                                 return ide_stopped;
464                 } else if ((err & BAD_CRC) == BAD_CRC) {
465                         /* UDMA crc error, just retry the operation */
466                         drive->crc_count++;
467                 } else if (err & (BBD_ERR | ECC_ERR)) {
468                         /* retries won't help these */
469                         rq->errors = ERROR_MAX;
470                 } else if (err & TRK0_ERR) {
471                         /* help it find track zero */
472                         rq->errors |= ERROR_RECAL;
473                 }
474         }
475
476         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ &&
477             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0)
478                 try_to_flush_leftover_data(drive);
479
480         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
481                 ide_kill_rq(drive, rq);
482                 return ide_stopped;
483         }
484
485         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
486                 rq->errors |= ERROR_RESET;
487
488         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
489                 ++rq->errors;
490                 return ide_do_reset(drive);
491         }
492
493         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
494                 drive->special.b.recalibrate = 1;
495
496         ++rq->errors;
497
498         return ide_stopped;
499 }
500
501 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
502 {
503         ide_hwif_t *hwif = drive->hwif;
504
505         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
506                 /* other bits are useless when BUSY */
507                 rq->errors |= ERROR_RESET;
508         } else {
509                 /* add decoding error stuff */
510         }
511
512         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
513                 /* force an abort */
514                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
515
516         if (rq->errors >= ERROR_MAX) {
517                 ide_kill_rq(drive, rq);
518         } else {
519                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
520                         ++rq->errors;
521                         return ide_do_reset(drive);
522                 }
523                 ++rq->errors;
524         }
525
526         return ide_stopped;
527 }
528
529 ide_startstop_t
530 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
531 {
532         if (drive->media == ide_disk)
533                 return ide_ata_error(drive, rq, stat, err);
534         return ide_atapi_error(drive, rq, stat, err);
535 }
536
537 EXPORT_SYMBOL_GPL(__ide_error);
538
539 /**
540  *      ide_error       -       handle an error on the IDE
541  *      @drive: drive the error occurred on
542  *      @msg: message to report
543  *      @stat: status bits
544  *
545  *      ide_error() takes action based on the error returned by the drive.
546  *      For normal I/O that may well include retries. We deal with
547  *      both new-style (taskfile) and old style command handling here.
548  *      In the case of taskfile command handling there is work left to
549  *      do
550  */
551  
552 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
553 {
554         struct request *rq;
555         u8 err;
556
557         err = ide_dump_status(drive, msg, stat);
558
559         if ((rq = HWGROUP(drive)->rq) == NULL)
560                 return ide_stopped;
561
562         /* retry only "normal" I/O: */
563         if (!blk_fs_request(rq)) {
564                 rq->errors = 1;
565                 ide_end_drive_cmd(drive, stat, err);
566                 return ide_stopped;
567         }
568
569         if (rq->rq_disk) {
570                 ide_driver_t *drv;
571
572                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
573                 return drv->error(drive, rq, stat, err);
574         } else
575                 return __ide_error(drive, rq, stat, err);
576 }
577
578 EXPORT_SYMBOL_GPL(ide_error);
579
580 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
581 {
582         if (drive->media != ide_disk)
583                 rq->errors |= ERROR_RESET;
584
585         ide_kill_rq(drive, rq);
586
587         return ide_stopped;
588 }
589
590 EXPORT_SYMBOL_GPL(__ide_abort);
591
592 /**
593  *      ide_abort       -       abort pending IDE operations
594  *      @drive: drive the error occurred on
595  *      @msg: message to report
596  *
597  *      ide_abort kills and cleans up when we are about to do a 
598  *      host initiated reset on active commands. Longer term we
599  *      want handlers to have sensible abort handling themselves
600  *
601  *      This differs fundamentally from ide_error because in 
602  *      this case the command is doing just fine when we
603  *      blow it away.
604  */
605  
606 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
607 {
608         struct request *rq;
609
610         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
611                 return ide_stopped;
612
613         /* retry only "normal" I/O: */
614         if (!blk_fs_request(rq)) {
615                 rq->errors = 1;
616                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
617                 return ide_stopped;
618         }
619
620         if (rq->rq_disk) {
621                 ide_driver_t *drv;
622
623                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
624                 return drv->abort(drive, rq);
625         } else
626                 return __ide_abort(drive, rq);
627 }
628
629 /**
630  *      drive_cmd_intr          -       drive command completion interrupt
631  *      @drive: drive the completion interrupt occurred on
632  *
633  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
634  *      We do any necessary data reading and then wait for the drive to
635  *      go non busy. At that point we may read the error data and complete
636  *      the request
637  */
638  
639 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
640 {
641         struct request *rq = HWGROUP(drive)->rq;
642         ide_hwif_t *hwif = HWIF(drive);
643         u8 *args = (u8 *) rq->buffer;
644         u8 stat = hwif->INB(IDE_STATUS_REG);
645         int retries = 10;
646
647         local_irq_enable_in_hardirq();
648         if (rq->cmd_type == REQ_TYPE_ATA_CMD &&
649             (stat & DRQ_STAT) && args && args[3]) {
650                 u8 io_32bit = drive->io_32bit;
651                 drive->io_32bit = 0;
652                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
653                 drive->io_32bit = io_32bit;
654                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
655                         udelay(100);
656         }
657
658         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
659                 return ide_error(drive, "drive_cmd", stat);
660                 /* calls ide_end_drive_cmd */
661         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
662         return ide_stopped;
663 }
664
665 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
666 {
667         tf->nsect   = drive->sect;
668         tf->lbal    = drive->sect;
669         tf->lbam    = drive->cyl;
670         tf->lbah    = drive->cyl >> 8;
671         tf->device  = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
672         tf->command = WIN_SPECIFY;
673 }
674
675 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
676 {
677         tf->nsect   = drive->sect;
678         tf->command = WIN_RESTORE;
679 }
680
681 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
682 {
683         tf->nsect   = drive->mult_req;
684         tf->command = WIN_SETMULT;
685 }
686
687 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
688 {
689         special_t *s = &drive->special;
690         ide_task_t args;
691
692         memset(&args, 0, sizeof(ide_task_t));
693         args.data_phase = TASKFILE_NO_DATA;
694
695         if (s->b.set_geometry) {
696                 s->b.set_geometry = 0;
697                 ide_tf_set_specify_cmd(drive, &args.tf);
698         } else if (s->b.recalibrate) {
699                 s->b.recalibrate = 0;
700                 ide_tf_set_restore_cmd(drive, &args.tf);
701         } else if (s->b.set_multmode) {
702                 s->b.set_multmode = 0;
703                 if (drive->mult_req > drive->id->max_multsect)
704                         drive->mult_req = drive->id->max_multsect;
705                 ide_tf_set_setmult_cmd(drive, &args.tf);
706         } else if (s->all) {
707                 int special = s->all;
708                 s->all = 0;
709                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
710                 return ide_stopped;
711         }
712
713         args.tf_flags = IDE_TFLAG_OUT_TF | IDE_TFLAG_OUT_DEVICE |
714                         IDE_TFLAG_CUSTOM_HANDLER;
715
716         do_rw_taskfile(drive, &args);
717
718         return ide_started;
719 }
720
721 /*
722  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
723  */
724 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
725 {
726         switch (req_pio) {
727         case 202:
728         case 201:
729         case 200:
730         case 102:
731         case 101:
732         case 100:
733                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
734         case 9:
735         case 8:
736                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
737         case 7:
738         case 6:
739                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
740         default:
741                 return 0;
742         }
743 }
744
745 /**
746  *      do_special              -       issue some special commands
747  *      @drive: drive the command is for
748  *
749  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
750  *      commands to a drive.  It used to do much more, but has been scaled
751  *      back.
752  */
753
754 static ide_startstop_t do_special (ide_drive_t *drive)
755 {
756         special_t *s = &drive->special;
757
758 #ifdef DEBUG
759         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
760 #endif
761         if (s->b.set_tune) {
762                 ide_hwif_t *hwif = drive->hwif;
763                 u8 req_pio = drive->tune_req;
764
765                 s->b.set_tune = 0;
766
767                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
768
769                         if (hwif->set_pio_mode == NULL)
770                                 return ide_stopped;
771
772                         /*
773                          * take ide_lock for drive->[no_]unmask/[no_]io_32bit
774                          */
775                         if (req_pio == 8 || req_pio == 9) {
776                                 unsigned long flags;
777
778                                 spin_lock_irqsave(&ide_lock, flags);
779                                 hwif->set_pio_mode(drive, req_pio);
780                                 spin_unlock_irqrestore(&ide_lock, flags);
781                         } else
782                                 hwif->set_pio_mode(drive, req_pio);
783                 } else {
784                         int keep_dma = drive->using_dma;
785
786                         ide_set_pio(drive, req_pio);
787
788                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
789                                 if (keep_dma)
790                                         hwif->ide_dma_on(drive);
791                         }
792                 }
793
794                 return ide_stopped;
795         } else {
796                 if (drive->media == ide_disk)
797                         return ide_disk_special(drive);
798
799                 s->all = 0;
800                 drive->mult_req = 0;
801                 return ide_stopped;
802         }
803 }
804
805 void ide_map_sg(ide_drive_t *drive, struct request *rq)
806 {
807         ide_hwif_t *hwif = drive->hwif;
808         struct scatterlist *sg = hwif->sg_table;
809
810         if (hwif->sg_mapped)    /* needed by ide-scsi */
811                 return;
812
813         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
814                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
815         } else {
816                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
817                 hwif->sg_nents = 1;
818         }
819 }
820
821 EXPORT_SYMBOL_GPL(ide_map_sg);
822
823 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
824 {
825         ide_hwif_t *hwif = drive->hwif;
826
827         hwif->nsect = hwif->nleft = rq->nr_sectors;
828         hwif->cursg_ofs = 0;
829         hwif->cursg = NULL;
830 }
831
832 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
833
834 /**
835  *      execute_drive_command   -       issue special drive command
836  *      @drive: the drive to issue the command on
837  *      @rq: the request structure holding the command
838  *
839  *      execute_drive_cmd() issues a special drive command,  usually 
840  *      initiated by ioctl() from the external hdparm program. The
841  *      command can be a drive command, drive task or taskfile 
842  *      operation. Weirdly you can call it with NULL to wait for
843  *      all commands to finish. Don't do this as that is due to change
844  */
845
846 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
847                 struct request *rq)
848 {
849         ide_hwif_t *hwif = HWIF(drive);
850         u8 *args = rq->buffer;
851         ide_task_t ltask;
852         struct ide_taskfile *tf = &ltask.tf;
853
854         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
855                 ide_task_t *task = rq->special;
856  
857                 if (task == NULL)
858                         goto done;
859
860                 hwif->data_phase = task->data_phase;
861
862                 switch (hwif->data_phase) {
863                 case TASKFILE_MULTI_OUT:
864                 case TASKFILE_OUT:
865                 case TASKFILE_MULTI_IN:
866                 case TASKFILE_IN:
867                         ide_init_sg_cmd(drive, rq);
868                         ide_map_sg(drive, rq);
869                 default:
870                         break;
871                 }
872
873                 return do_rw_taskfile(drive, task);
874         }
875
876         if (args == NULL)
877                 goto done;
878
879         memset(&ltask, 0, sizeof(ltask));
880         if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
881 #ifdef DEBUG
882                 printk("%s: DRIVE_CMD\n", drive->name);
883 #endif
884                 tf->feature = args[2];
885                 if (args[0] == WIN_SMART) {
886                         tf->nsect = args[3];
887                         tf->lbal  = args[1];
888                         tf->lbam  = 0x4f;
889                         tf->lbah  = 0xc2;
890                         ltask.tf_flags = IDE_TFLAG_OUT_TF;
891                 } else {
892                         tf->nsect = args[1];
893                         ltask.tf_flags = IDE_TFLAG_OUT_FEATURE |
894                                          IDE_TFLAG_OUT_NSECT;
895                 }
896         }
897         tf->command = args[0];
898         ide_tf_load(drive, &ltask);
899         ide_execute_command(drive, args[0], &drive_cmd_intr, WAIT_WORSTCASE, NULL);
900         return ide_started;
901
902 done:
903         /*
904          * NULL is actually a valid way of waiting for
905          * all current requests to be flushed from the queue.
906          */
907 #ifdef DEBUG
908         printk("%s: DRIVE_CMD (null)\n", drive->name);
909 #endif
910         ide_end_drive_cmd(drive,
911                         hwif->INB(IDE_STATUS_REG),
912                         hwif->INB(IDE_ERROR_REG));
913         return ide_stopped;
914 }
915
916 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
917 {
918         struct request_pm_state *pm = rq->data;
919
920         if (blk_pm_suspend_request(rq) &&
921             pm->pm_step == ide_pm_state_start_suspend)
922                 /* Mark drive blocked when starting the suspend sequence. */
923                 drive->blocked = 1;
924         else if (blk_pm_resume_request(rq) &&
925                  pm->pm_step == ide_pm_state_start_resume) {
926                 /* 
927                  * The first thing we do on wakeup is to wait for BSY bit to
928                  * go away (with a looong timeout) as a drive on this hwif may
929                  * just be POSTing itself.
930                  * We do that before even selecting as the "other" device on
931                  * the bus may be broken enough to walk on our toes at this
932                  * point.
933                  */
934                 int rc;
935 #ifdef DEBUG_PM
936                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
937 #endif
938                 rc = ide_wait_not_busy(HWIF(drive), 35000);
939                 if (rc)
940                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
941                 SELECT_DRIVE(drive);
942                 if (IDE_CONTROL_REG)
943                         HWIF(drive)->OUTB(drive->ctl, IDE_CONTROL_REG);
944                 rc = ide_wait_not_busy(HWIF(drive), 100000);
945                 if (rc)
946                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
947         }
948 }
949
950 /**
951  *      start_request   -       start of I/O and command issuing for IDE
952  *
953  *      start_request() initiates handling of a new I/O request. It
954  *      accepts commands and I/O (read/write) requests. It also does
955  *      the final remapping for weird stuff like EZDrive. Once 
956  *      device mapper can work sector level the EZDrive stuff can go away
957  *
958  *      FIXME: this function needs a rename
959  */
960  
961 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
962 {
963         ide_startstop_t startstop;
964         sector_t block;
965
966         BUG_ON(!blk_rq_started(rq));
967
968 #ifdef DEBUG
969         printk("%s: start_request: current=0x%08lx\n",
970                 HWIF(drive)->name, (unsigned long) rq);
971 #endif
972
973         /* bail early if we've exceeded max_failures */
974         if (drive->max_failures && (drive->failures > drive->max_failures)) {
975                 rq->cmd_flags |= REQ_FAILED;
976                 goto kill_rq;
977         }
978
979         block    = rq->sector;
980         if (blk_fs_request(rq) &&
981             (drive->media == ide_disk || drive->media == ide_floppy)) {
982                 block += drive->sect0;
983         }
984         /* Yecch - this will shift the entire interval,
985            possibly killing some innocent following sector */
986         if (block == 0 && drive->remap_0_to_1 == 1)
987                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
988
989         if (blk_pm_request(rq))
990                 ide_check_pm_state(drive, rq);
991
992         SELECT_DRIVE(drive);
993         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
994                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
995                 return startstop;
996         }
997         if (!drive->special.all) {
998                 ide_driver_t *drv;
999
1000                 /*
1001                  * We reset the drive so we need to issue a SETFEATURES.
1002                  * Do it _after_ do_special() restored device parameters.
1003                  */
1004                 if (drive->current_speed == 0xff)
1005                         ide_config_drive_speed(drive, drive->desired_speed);
1006
1007                 if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1008                     rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1009                         return execute_drive_cmd(drive, rq);
1010                 else if (blk_pm_request(rq)) {
1011                         struct request_pm_state *pm = rq->data;
1012 #ifdef DEBUG_PM
1013                         printk("%s: start_power_step(step: %d)\n",
1014                                 drive->name, rq->pm->pm_step);
1015 #endif
1016                         startstop = ide_start_power_step(drive, rq);
1017                         if (startstop == ide_stopped &&
1018                             pm->pm_step == ide_pm_state_completed)
1019                                 ide_complete_pm_request(drive, rq);
1020                         return startstop;
1021                 }
1022
1023                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1024                 return drv->do_request(drive, rq, block);
1025         }
1026         return do_special(drive);
1027 kill_rq:
1028         ide_kill_rq(drive, rq);
1029         return ide_stopped;
1030 }
1031
1032 /**
1033  *      ide_stall_queue         -       pause an IDE device
1034  *      @drive: drive to stall
1035  *      @timeout: time to stall for (jiffies)
1036  *
1037  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1038  *      to the hwgroup by sleeping for timeout jiffies.
1039  */
1040  
1041 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1042 {
1043         if (timeout > WAIT_WORSTCASE)
1044                 timeout = WAIT_WORSTCASE;
1045         drive->sleep = timeout + jiffies;
1046         drive->sleeping = 1;
1047 }
1048
1049 EXPORT_SYMBOL(ide_stall_queue);
1050
1051 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1052
1053 /**
1054  *      choose_drive            -       select a drive to service
1055  *      @hwgroup: hardware group to select on
1056  *
1057  *      choose_drive() selects the next drive which will be serviced.
1058  *      This is necessary because the IDE layer can't issue commands
1059  *      to both drives on the same cable, unlike SCSI.
1060  */
1061  
1062 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1063 {
1064         ide_drive_t *drive, *best;
1065
1066 repeat: 
1067         best = NULL;
1068         drive = hwgroup->drive;
1069
1070         /*
1071          * drive is doing pre-flush, ordered write, post-flush sequence. even
1072          * though that is 3 requests, it must be seen as a single transaction.
1073          * we must not preempt this drive until that is complete
1074          */
1075         if (blk_queue_flushing(drive->queue)) {
1076                 /*
1077                  * small race where queue could get replugged during
1078                  * the 3-request flush cycle, just yank the plug since
1079                  * we want it to finish asap
1080                  */
1081                 blk_remove_plug(drive->queue);
1082                 return drive;
1083         }
1084
1085         do {
1086                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1087                     && !elv_queue_empty(drive->queue)) {
1088                         if (!best
1089                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1090                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1091                         {
1092                                 if (!blk_queue_plugged(drive->queue))
1093                                         best = drive;
1094                         }
1095                 }
1096         } while ((drive = drive->next) != hwgroup->drive);
1097         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1098                 long t = (signed long)(WAKEUP(best) - jiffies);
1099                 if (t >= WAIT_MIN_SLEEP) {
1100                 /*
1101                  * We *may* have some time to spare, but first let's see if
1102                  * someone can potentially benefit from our nice mood today..
1103                  */
1104                         drive = best->next;
1105                         do {
1106                                 if (!drive->sleeping
1107                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1108                                  && time_before(WAKEUP(drive), jiffies + t))
1109                                 {
1110                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1111                                         goto repeat;
1112                                 }
1113                         } while ((drive = drive->next) != best);
1114                 }
1115         }
1116         return best;
1117 }
1118
1119 /*
1120  * Issue a new request to a drive from hwgroup
1121  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1122  *
1123  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1124  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1125  * may have both interfaces in a single hwgroup to "serialize" access.
1126  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1127  * together into one hwgroup for serialized access.
1128  *
1129  * Note also that several hwgroups can end up sharing a single IRQ,
1130  * possibly along with many other devices.  This is especially common in
1131  * PCI-based systems with off-board IDE controller cards.
1132  *
1133  * The IDE driver uses the single global ide_lock spinlock to protect
1134  * access to the request queues, and to protect the hwgroup->busy flag.
1135  *
1136  * The first thread into the driver for a particular hwgroup sets the
1137  * hwgroup->busy flag to indicate that this hwgroup is now active,
1138  * and then initiates processing of the top request from the request queue.
1139  *
1140  * Other threads attempting entry notice the busy setting, and will simply
1141  * queue their new requests and exit immediately.  Note that hwgroup->busy
1142  * remains set even when the driver is merely awaiting the next interrupt.
1143  * Thus, the meaning is "this hwgroup is busy processing a request".
1144  *
1145  * When processing of a request completes, the completing thread or IRQ-handler
1146  * will start the next request from the queue.  If no more work remains,
1147  * the driver will clear the hwgroup->busy flag and exit.
1148  *
1149  * The ide_lock (spinlock) is used to protect all access to the
1150  * hwgroup->busy flag, but is otherwise not needed for most processing in
1151  * the driver.  This makes the driver much more friendlier to shared IRQs
1152  * than previous designs, while remaining 100% (?) SMP safe and capable.
1153  */
1154 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1155 {
1156         ide_drive_t     *drive;
1157         ide_hwif_t      *hwif;
1158         struct request  *rq;
1159         ide_startstop_t startstop;
1160         int             loops = 0;
1161
1162         /* for atari only: POSSIBLY BROKEN HERE(?) */
1163         ide_get_lock(ide_intr, hwgroup);
1164
1165         /* caller must own ide_lock */
1166         BUG_ON(!irqs_disabled());
1167
1168         while (!hwgroup->busy) {
1169                 hwgroup->busy = 1;
1170                 drive = choose_drive(hwgroup);
1171                 if (drive == NULL) {
1172                         int sleeping = 0;
1173                         unsigned long sleep = 0; /* shut up, gcc */
1174                         hwgroup->rq = NULL;
1175                         drive = hwgroup->drive;
1176                         do {
1177                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1178                                         sleeping = 1;
1179                                         sleep = drive->sleep;
1180                                 }
1181                         } while ((drive = drive->next) != hwgroup->drive);
1182                         if (sleeping) {
1183                 /*
1184                  * Take a short snooze, and then wake up this hwgroup again.
1185                  * This gives other hwgroups on the same a chance to
1186                  * play fairly with us, just in case there are big differences
1187                  * in relative throughputs.. don't want to hog the cpu too much.
1188                  */
1189                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1190                                         sleep = jiffies + WAIT_MIN_SLEEP;
1191 #if 1
1192                                 if (timer_pending(&hwgroup->timer))
1193                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1194 #endif
1195                                 /* so that ide_timer_expiry knows what to do */
1196                                 hwgroup->sleeping = 1;
1197                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1198                                 mod_timer(&hwgroup->timer, sleep);
1199                                 /* we purposely leave hwgroup->busy==1
1200                                  * while sleeping */
1201                         } else {
1202                                 /* Ugly, but how can we sleep for the lock
1203                                  * otherwise? perhaps from tq_disk?
1204                                  */
1205
1206                                 /* for atari only */
1207                                 ide_release_lock();
1208                                 hwgroup->busy = 0;
1209                         }
1210
1211                         /* no more work for this hwgroup (for now) */
1212                         return;
1213                 }
1214         again:
1215                 hwif = HWIF(drive);
1216                 if (hwgroup->hwif->sharing_irq &&
1217                     hwif != hwgroup->hwif &&
1218                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1219                         /*
1220                          * set nIEN for previous hwif, drives in the
1221                          * quirk_list may not like intr setups/cleanups
1222                          */
1223                         if (drive->quirk_list != 1)
1224                                 hwif->OUTB(drive->ctl | 2, IDE_CONTROL_REG);
1225                 }
1226                 hwgroup->hwif = hwif;
1227                 hwgroup->drive = drive;
1228                 drive->sleeping = 0;
1229                 drive->service_start = jiffies;
1230
1231                 if (blk_queue_plugged(drive->queue)) {
1232                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1233                         break;
1234                 }
1235
1236                 /*
1237                  * we know that the queue isn't empty, but this can happen
1238                  * if the q->prep_rq_fn() decides to kill a request
1239                  */
1240                 rq = elv_next_request(drive->queue);
1241                 if (!rq) {
1242                         hwgroup->busy = 0;
1243                         break;
1244                 }
1245
1246                 /*
1247                  * Sanity: don't accept a request that isn't a PM request
1248                  * if we are currently power managed. This is very important as
1249                  * blk_stop_queue() doesn't prevent the elv_next_request()
1250                  * above to return us whatever is in the queue. Since we call
1251                  * ide_do_request() ourselves, we end up taking requests while
1252                  * the queue is blocked...
1253                  * 
1254                  * We let requests forced at head of queue with ide-preempt
1255                  * though. I hope that doesn't happen too much, hopefully not
1256                  * unless the subdriver triggers such a thing in its own PM
1257                  * state machine.
1258                  *
1259                  * We count how many times we loop here to make sure we service
1260                  * all drives in the hwgroup without looping for ever
1261                  */
1262                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1263                         drive = drive->next ? drive->next : hwgroup->drive;
1264                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1265                                 goto again;
1266                         /* We clear busy, there should be no pending ATA command at this point. */
1267                         hwgroup->busy = 0;
1268                         break;
1269                 }
1270
1271                 hwgroup->rq = rq;
1272
1273                 /*
1274                  * Some systems have trouble with IDE IRQs arriving while
1275                  * the driver is still setting things up.  So, here we disable
1276                  * the IRQ used by this interface while the request is being started.
1277                  * This may look bad at first, but pretty much the same thing
1278                  * happens anyway when any interrupt comes in, IDE or otherwise
1279                  *  -- the kernel masks the IRQ while it is being handled.
1280                  */
1281                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1282                         disable_irq_nosync(hwif->irq);
1283                 spin_unlock(&ide_lock);
1284                 local_irq_enable_in_hardirq();
1285                         /* allow other IRQs while we start this request */
1286                 startstop = start_request(drive, rq);
1287                 spin_lock_irq(&ide_lock);
1288                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1289                         enable_irq(hwif->irq);
1290                 if (startstop == ide_stopped)
1291                         hwgroup->busy = 0;
1292         }
1293 }
1294
1295 /*
1296  * Passes the stuff to ide_do_request
1297  */
1298 void do_ide_request(struct request_queue *q)
1299 {
1300         ide_drive_t *drive = q->queuedata;
1301
1302         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1303 }
1304
1305 /*
1306  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1307  * retry the current request in pio mode instead of risking tossing it
1308  * all away
1309  */
1310 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1311 {
1312         ide_hwif_t *hwif = HWIF(drive);
1313         struct request *rq;
1314         ide_startstop_t ret = ide_stopped;
1315
1316         /*
1317          * end current dma transaction
1318          */
1319
1320         if (error < 0) {
1321                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1322                 (void)HWIF(drive)->ide_dma_end(drive);
1323                 ret = ide_error(drive, "dma timeout error",
1324                                                 hwif->INB(IDE_STATUS_REG));
1325         } else {
1326                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1327                 hwif->dma_timeout(drive);
1328         }
1329
1330         /*
1331          * disable dma for now, but remember that we did so because of
1332          * a timeout -- we'll reenable after we finish this next request
1333          * (or rather the first chunk of it) in pio.
1334          */
1335         drive->retry_pio++;
1336         drive->state = DMA_PIO_RETRY;
1337         hwif->dma_off_quietly(drive);
1338
1339         /*
1340          * un-busy drive etc (hwgroup->busy is cleared on return) and
1341          * make sure request is sane
1342          */
1343         rq = HWGROUP(drive)->rq;
1344
1345         if (!rq)
1346                 goto out;
1347
1348         HWGROUP(drive)->rq = NULL;
1349
1350         rq->errors = 0;
1351
1352         if (!rq->bio)
1353                 goto out;
1354
1355         rq->sector = rq->bio->bi_sector;
1356         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1357         rq->hard_cur_sectors = rq->current_nr_sectors;
1358         rq->buffer = bio_data(rq->bio);
1359 out:
1360         return ret;
1361 }
1362
1363 /**
1364  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1365  *      @data: timer callback magic (hwgroup)
1366  *
1367  *      An IDE command has timed out before the expected drive return
1368  *      occurred. At this point we attempt to clean up the current
1369  *      mess. If the current handler includes an expiry handler then
1370  *      we invoke the expiry handler, and providing it is happy the
1371  *      work is done. If that fails we apply generic recovery rules
1372  *      invoking the handler and checking the drive DMA status. We
1373  *      have an excessively incestuous relationship with the DMA
1374  *      logic that wants cleaning up.
1375  */
1376  
1377 void ide_timer_expiry (unsigned long data)
1378 {
1379         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1380         ide_handler_t   *handler;
1381         ide_expiry_t    *expiry;
1382         unsigned long   flags;
1383         unsigned long   wait = -1;
1384
1385         spin_lock_irqsave(&ide_lock, flags);
1386
1387         if (((handler = hwgroup->handler) == NULL) ||
1388             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
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                                         hwgroup->req_gen_timer = hwgroup->req_gen;
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                         /* disable_irq_nosync ?? */
1431                         disable_irq(hwif->irq);
1432                         /* local CPU only,
1433                          * as if we were handling an interrupt */
1434                         local_irq_disable();
1435                         if (hwgroup->polling) {
1436                                 startstop = handler(drive);
1437                         } else if (drive_is_ready(drive)) {
1438                                 if (drive->waiting_for_dma)
1439                                         hwgroup->hwif->dma_lost_irq(drive);
1440                                 (void)ide_ack_intr(hwif);
1441                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1442                                 startstop = handler(drive);
1443                         } else {
1444                                 if (drive->waiting_for_dma) {
1445                                         startstop = ide_dma_timeout_retry(drive, wait);
1446                                 } else
1447                                         startstop =
1448                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1449                         }
1450                         drive->service_time = jiffies - drive->service_start;
1451                         spin_lock_irq(&ide_lock);
1452                         enable_irq(hwif->irq);
1453                         if (startstop == ide_stopped)
1454                                 hwgroup->busy = 0;
1455                 }
1456         }
1457         ide_do_request(hwgroup, IDE_NO_IRQ);
1458         spin_unlock_irqrestore(&ide_lock, flags);
1459 }
1460
1461 /**
1462  *      unexpected_intr         -       handle an unexpected IDE interrupt
1463  *      @irq: interrupt line
1464  *      @hwgroup: hwgroup being processed
1465  *
1466  *      There's nothing really useful we can do with an unexpected interrupt,
1467  *      other than reading the status register (to clear it), and logging it.
1468  *      There should be no way that an irq can happen before we're ready for it,
1469  *      so we needn't worry much about losing an "important" interrupt here.
1470  *
1471  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1472  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1473  *      looks "good", we just ignore the interrupt completely.
1474  *
1475  *      This routine assumes __cli() is in effect when called.
1476  *
1477  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1478  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1479  *      we could screw up by interfering with a new request being set up for 
1480  *      irq15.
1481  *
1482  *      In reality, this is a non-issue.  The new command is not sent unless 
1483  *      the drive is ready to accept one, in which case we know the drive is
1484  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1485  *      before completing the issuance of any new drive command, so we will not
1486  *      be accidentally invoked as a result of any valid command completion
1487  *      interrupt.
1488  *
1489  *      Note that we must walk the entire hwgroup here. We know which hwif
1490  *      is doing the current command, but we don't know which hwif burped
1491  *      mysteriously.
1492  */
1493  
1494 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1495 {
1496         u8 stat;
1497         ide_hwif_t *hwif = hwgroup->hwif;
1498
1499         /*
1500          * handle the unexpected interrupt
1501          */
1502         do {
1503                 if (hwif->irq == irq) {
1504                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1505                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1506                                 /* Try to not flood the console with msgs */
1507                                 static unsigned long last_msgtime, count;
1508                                 ++count;
1509                                 if (time_after(jiffies, last_msgtime + HZ)) {
1510                                         last_msgtime = jiffies;
1511                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1512                                                 "status=0x%02x, count=%ld\n",
1513                                                 hwif->name,
1514                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1515                                 }
1516                         }
1517                 }
1518         } while ((hwif = hwif->next) != hwgroup->hwif);
1519 }
1520
1521 /**
1522  *      ide_intr        -       default IDE interrupt handler
1523  *      @irq: interrupt number
1524  *      @dev_id: hwif group
1525  *      @regs: unused weirdness from the kernel irq layer
1526  *
1527  *      This is the default IRQ handler for the IDE layer. You should
1528  *      not need to override it. If you do be aware it is subtle in
1529  *      places
1530  *
1531  *      hwgroup->hwif is the interface in the group currently performing
1532  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1533  *      the IRQ handler to call. As we issue a command the handlers
1534  *      step through multiple states, reassigning the handler to the
1535  *      next step in the process. Unlike a smart SCSI controller IDE
1536  *      expects the main processor to sequence the various transfer
1537  *      stages. We also manage a poll timer to catch up with most
1538  *      timeout situations. There are still a few where the handlers
1539  *      don't ever decide to give up.
1540  *
1541  *      The handler eventually returns ide_stopped to indicate the
1542  *      request completed. At this point we issue the next request
1543  *      on the hwgroup and the process begins again.
1544  */
1545  
1546 irqreturn_t ide_intr (int irq, void *dev_id)
1547 {
1548         unsigned long flags;
1549         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1550         ide_hwif_t *hwif;
1551         ide_drive_t *drive;
1552         ide_handler_t *handler;
1553         ide_startstop_t startstop;
1554
1555         spin_lock_irqsave(&ide_lock, flags);
1556         hwif = hwgroup->hwif;
1557
1558         if (!ide_ack_intr(hwif)) {
1559                 spin_unlock_irqrestore(&ide_lock, flags);
1560                 return IRQ_NONE;
1561         }
1562
1563         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1564                 /*
1565                  * Not expecting an interrupt from this drive.
1566                  * That means this could be:
1567                  *      (1) an interrupt from another PCI device
1568                  *      sharing the same PCI INT# as us.
1569                  * or   (2) a drive just entered sleep or standby mode,
1570                  *      and is interrupting to let us know.
1571                  * or   (3) a spurious interrupt of unknown origin.
1572                  *
1573                  * For PCI, we cannot tell the difference,
1574                  * so in that case we just ignore it and hope it goes away.
1575                  *
1576                  * FIXME: unexpected_intr should be hwif-> then we can
1577                  * remove all the ifdef PCI crap
1578                  */
1579 #ifdef CONFIG_BLK_DEV_IDEPCI
1580                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1581 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1582                 {
1583                         /*
1584                          * Probably not a shared PCI interrupt,
1585                          * so we can safely try to do something about it:
1586                          */
1587                         unexpected_intr(irq, hwgroup);
1588 #ifdef CONFIG_BLK_DEV_IDEPCI
1589                 } else {
1590                         /*
1591                          * Whack the status register, just in case
1592                          * we have a leftover pending IRQ.
1593                          */
1594                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1595 #endif /* CONFIG_BLK_DEV_IDEPCI */
1596                 }
1597                 spin_unlock_irqrestore(&ide_lock, flags);
1598                 return IRQ_NONE;
1599         }
1600         drive = hwgroup->drive;
1601         if (!drive) {
1602                 /*
1603                  * This should NEVER happen, and there isn't much
1604                  * we could do about it here.
1605                  *
1606                  * [Note - this can occur if the drive is hot unplugged]
1607                  */
1608                 spin_unlock_irqrestore(&ide_lock, flags);
1609                 return IRQ_HANDLED;
1610         }
1611         if (!drive_is_ready(drive)) {
1612                 /*
1613                  * This happens regularly when we share a PCI IRQ with
1614                  * another device.  Unfortunately, it can also happen
1615                  * with some buggy drives that trigger the IRQ before
1616                  * their status register is up to date.  Hopefully we have
1617                  * enough advance overhead that the latter isn't a problem.
1618                  */
1619                 spin_unlock_irqrestore(&ide_lock, flags);
1620                 return IRQ_NONE;
1621         }
1622         if (!hwgroup->busy) {
1623                 hwgroup->busy = 1;      /* paranoia */
1624                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1625         }
1626         hwgroup->handler = NULL;
1627         hwgroup->req_gen++;
1628         del_timer(&hwgroup->timer);
1629         spin_unlock(&ide_lock);
1630
1631         /* Some controllers might set DMA INTR no matter DMA or PIO;
1632          * bmdma status might need to be cleared even for
1633          * PIO interrupts to prevent spurious/lost irq.
1634          */
1635         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1636                 /* ide_dma_end() needs bmdma status for error checking.
1637                  * So, skip clearing bmdma status here and leave it
1638                  * to ide_dma_end() if this is dma interrupt.
1639                  */
1640                 hwif->ide_dma_clear_irq(drive);
1641
1642         if (drive->unmask)
1643                 local_irq_enable_in_hardirq();
1644         /* service this interrupt, may set handler for next interrupt */
1645         startstop = handler(drive);
1646         spin_lock_irq(&ide_lock);
1647
1648         /*
1649          * Note that handler() may have set things up for another
1650          * interrupt to occur soon, but it cannot happen until
1651          * we exit from this routine, because it will be the
1652          * same irq as is currently being serviced here, and Linux
1653          * won't allow another of the same (on any CPU) until we return.
1654          */
1655         drive->service_time = jiffies - drive->service_start;
1656         if (startstop == ide_stopped) {
1657                 if (hwgroup->handler == NULL) { /* paranoia */
1658                         hwgroup->busy = 0;
1659                         ide_do_request(hwgroup, hwif->irq);
1660                 } else {
1661                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1662                                 "on exit\n", drive->name);
1663                 }
1664         }
1665         spin_unlock_irqrestore(&ide_lock, flags);
1666         return IRQ_HANDLED;
1667 }
1668
1669 /**
1670  *      ide_init_drive_cmd      -       initialize a drive command request
1671  *      @rq: request object
1672  *
1673  *      Initialize a request before we fill it in and send it down to
1674  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1675  *      now it doesn't do a lot, but if that changes abusers will have a
1676  *      nasty surprise.
1677  */
1678
1679 void ide_init_drive_cmd (struct request *rq)
1680 {
1681         memset(rq, 0, sizeof(*rq));
1682         rq->cmd_type = REQ_TYPE_ATA_CMD;
1683         rq->ref_count = 1;
1684 }
1685
1686 EXPORT_SYMBOL(ide_init_drive_cmd);
1687
1688 /**
1689  *      ide_do_drive_cmd        -       issue IDE special command
1690  *      @drive: device to issue command
1691  *      @rq: request to issue
1692  *      @action: action for processing
1693  *
1694  *      This function issues a special IDE device request
1695  *      onto the request queue.
1696  *
1697  *      If action is ide_wait, then the rq is queued at the end of the
1698  *      request queue, and the function sleeps until it has been processed.
1699  *      This is for use when invoked from an ioctl handler.
1700  *
1701  *      If action is ide_preempt, then the rq is queued at the head of
1702  *      the request queue, displacing the currently-being-processed
1703  *      request and this function returns immediately without waiting
1704  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1705  *      intended for careful use by the ATAPI tape/cdrom driver code.
1706  *
1707  *      If action is ide_end, then the rq is queued at the end of the
1708  *      request queue, and the function returns immediately without waiting
1709  *      for the new rq to be completed. This is again intended for careful
1710  *      use by the ATAPI tape/cdrom driver code.
1711  */
1712  
1713 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1714 {
1715         unsigned long flags;
1716         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1717         DECLARE_COMPLETION_ONSTACK(wait);
1718         int where = ELEVATOR_INSERT_BACK, err;
1719         int must_wait = (action == ide_wait || action == ide_head_wait);
1720
1721         rq->errors = 0;
1722
1723         /*
1724          * we need to hold an extra reference to request for safe inspection
1725          * after completion
1726          */
1727         if (must_wait) {
1728                 rq->ref_count++;
1729                 rq->end_io_data = &wait;
1730                 rq->end_io = blk_end_sync_rq;
1731         }
1732
1733         spin_lock_irqsave(&ide_lock, flags);
1734         if (action == ide_preempt)
1735                 hwgroup->rq = NULL;
1736         if (action == ide_preempt || action == ide_head_wait) {
1737                 where = ELEVATOR_INSERT_FRONT;
1738                 rq->cmd_flags |= REQ_PREEMPT;
1739         }
1740         __elv_add_request(drive->queue, rq, where, 0);
1741         ide_do_request(hwgroup, IDE_NO_IRQ);
1742         spin_unlock_irqrestore(&ide_lock, flags);
1743
1744         err = 0;
1745         if (must_wait) {
1746                 wait_for_completion(&wait);
1747                 if (rq->errors)
1748                         err = -EIO;
1749
1750                 blk_put_request(rq);
1751         }
1752
1753         return err;
1754 }
1755
1756 EXPORT_SYMBOL(ide_do_drive_cmd);
1757
1758 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1759 {
1760         ide_task_t task;
1761
1762         memset(&task, 0, sizeof(task));
1763         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1764                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1765         task.tf.feature = dma;          /* Use PIO/DMA */
1766         task.tf.lbam    = bcount & 0xff;
1767         task.tf.lbah    = (bcount >> 8) & 0xff;
1768
1769         ide_tf_load(drive, &task);
1770 }
1771
1772 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);