Pull asus into release branch
[linux-2.6] / drivers / ide / ide-io.c
1 /*
2  *      IDE I/O functions
3  *
4  *      Basic PIO and command management functionality.
5  *
6  * This code was split off from ide.c. See ide.c for history and original
7  * copyrights.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the
11  * Free Software Foundation; either version 2, or (at your option) any
12  * later version.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * For the avoidance of doubt the "preferred form" of this code is one which
20  * is in an open non patent encumbered format. Where cryptographic key signing
21  * forms part of the process of creating an executable the information
22  * including keys needed to generate an equivalently functional executable
23  * are deemed to be part of the source code.
24  */
25  
26  
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
32 #include <linux/mm.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/completion.h>
44 #include <linux/reboot.h>
45 #include <linux/cdrom.h>
46 #include <linux/seq_file.h>
47 #include <linux/device.h>
48 #include <linux/kmod.h>
49 #include <linux/scatterlist.h>
50
51 #include <asm/byteorder.h>
52 #include <asm/irq.h>
53 #include <asm/uaccess.h>
54 #include <asm/io.h>
55 #include <asm/bitops.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, int nr_sectors)
59 {
60         int ret = 1;
61
62         /*
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_sectors = rq->hard_nr_sectors;
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_first(rq, uptodate, nr_sectors)) {
82                 add_disk_randomness(rq->rq_disk);
83                 if (!list_empty(&rq->queuelist))
84                         blkdev_dequeue_request(rq);
85                 HWGROUP(drive)->rq = NULL;
86                 end_that_request_last(rq, uptodate);
87                 ret = 0;
88         }
89
90         return ret;
91 }
92
93 /**
94  *      ide_end_request         -       complete an IDE I/O
95  *      @drive: IDE device for the I/O
96  *      @uptodate:
97  *      @nr_sectors: number of sectors completed
98  *
99  *      This is our end_request wrapper function. We complete the I/O
100  *      update random number input and dequeue the request, which if
101  *      it was tagged may be out of order.
102  */
103
104 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
105 {
106         struct request *rq;
107         unsigned long flags;
108         int ret = 1;
109
110         /*
111          * room for locking improvements here, the calls below don't
112          * need the queue lock held at all
113          */
114         spin_lock_irqsave(&ide_lock, flags);
115         rq = HWGROUP(drive)->rq;
116
117         if (!nr_sectors)
118                 nr_sectors = rq->hard_cur_sectors;
119
120         ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
121
122         spin_unlock_irqrestore(&ide_lock, flags);
123         return ret;
124 }
125 EXPORT_SYMBOL(ide_end_request);
126
127 /*
128  * Power Management state machine. This one is rather trivial for now,
129  * we should probably add more, like switching back to PIO on suspend
130  * to help some BIOSes, re-do the door locking on resume, etc...
131  */
132
133 enum {
134         ide_pm_flush_cache      = ide_pm_state_start_suspend,
135         idedisk_pm_standby,
136
137         idedisk_pm_restore_pio  = ide_pm_state_start_resume,
138         idedisk_pm_idle,
139         ide_pm_restore_dma,
140 };
141
142 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
143 {
144         struct request_pm_state *pm = rq->data;
145
146         if (drive->media != ide_disk)
147                 return;
148
149         switch (pm->pm_step) {
150         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
151                 if (pm->pm_state == PM_EVENT_FREEZE)
152                         pm->pm_step = ide_pm_state_completed;
153                 else
154                         pm->pm_step = idedisk_pm_standby;
155                 break;
156         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
157                 pm->pm_step = ide_pm_state_completed;
158                 break;
159         case idedisk_pm_restore_pio:    /* Resume step 1 complete */
160                 pm->pm_step = idedisk_pm_idle;
161                 break;
162         case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
163                 pm->pm_step = ide_pm_restore_dma;
164                 break;
165         }
166 }
167
168 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
169 {
170         struct request_pm_state *pm = rq->data;
171         ide_task_t *args = rq->special;
172
173         memset(args, 0, sizeof(*args));
174
175         if (drive->media != ide_disk) {
176                 /*
177                  * skip idedisk_pm_restore_pio and idedisk_pm_idle for ATAPI
178                  * devices
179                  */
180                 if (pm->pm_step == idedisk_pm_restore_pio)
181                         pm->pm_step = ide_pm_restore_dma;
182         }
183
184         switch (pm->pm_step) {
185         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
186                 if (drive->media != ide_disk)
187                         break;
188                 /* Not supported? Switch to next step now. */
189                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
190                         ide_complete_power_step(drive, rq, 0, 0);
191                         return ide_stopped;
192                 }
193                 if (ide_id_has_flush_cache_ext(drive->id))
194                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
195                 else
196                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
197                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
198                 args->handler      = &task_no_data_intr;
199                 return do_rw_taskfile(drive, args);
200
201         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
202                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
203                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
204                 args->handler      = &task_no_data_intr;
205                 return do_rw_taskfile(drive, args);
206
207         case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
208                 if (drive->hwif->tuneproc != NULL)
209                         drive->hwif->tuneproc(drive, 255);
210                 ide_complete_power_step(drive, rq, 0, 0);
211                 return ide_stopped;
212
213         case idedisk_pm_idle:           /* Resume step 2 (idle) */
214                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
215                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
216                 args->handler = task_no_data_intr;
217                 return do_rw_taskfile(drive, args);
218
219         case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
220                 /*
221                  * Right now, all we do is call hwif->ide_dma_check(drive),
222                  * we could be smarter and check for current xfer_speed
223                  * in struct drive etc...
224                  */
225                 if ((drive->id->capability & 1) == 0)
226                         break;
227                 if (drive->hwif->ide_dma_check == NULL)
228                         break;
229                 ide_set_dma(drive);
230                 break;
231         }
232         pm->pm_step = ide_pm_state_completed;
233         return ide_stopped;
234 }
235
236 /**
237  *      ide_end_dequeued_request        -       complete an IDE I/O
238  *      @drive: IDE device for the I/O
239  *      @uptodate:
240  *      @nr_sectors: number of sectors completed
241  *
242  *      Complete an I/O that is no longer on the request queue. This
243  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
244  *      We must still finish the old request but we must not tamper with the
245  *      queue in the meantime.
246  *
247  *      NOTE: This path does not handle barrier, but barrier is not supported
248  *      on ide-cd anyway.
249  */
250
251 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
252                              int uptodate, int nr_sectors)
253 {
254         unsigned long flags;
255         int ret = 1;
256
257         spin_lock_irqsave(&ide_lock, flags);
258
259         BUG_ON(!blk_rq_started(rq));
260
261         /*
262          * if failfast is set on a request, override number of sectors and
263          * complete the whole request right now
264          */
265         if (blk_noretry_request(rq) && end_io_error(uptodate))
266                 nr_sectors = rq->hard_nr_sectors;
267
268         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
269                 rq->errors = -EIO;
270
271         /*
272          * decide whether to reenable DMA -- 3 is a random magic for now,
273          * if we DMA timeout more than 3 times, just stay in PIO
274          */
275         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
276                 drive->state = 0;
277                 HWGROUP(drive)->hwif->ide_dma_on(drive);
278         }
279
280         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
281                 add_disk_randomness(rq->rq_disk);
282                 if (blk_rq_tagged(rq))
283                         blk_queue_end_tag(drive->queue, rq);
284                 end_that_request_last(rq, uptodate);
285                 ret = 0;
286         }
287         spin_unlock_irqrestore(&ide_lock, flags);
288         return ret;
289 }
290 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
291
292
293 /**
294  *      ide_complete_pm_request - end the current Power Management request
295  *      @drive: target drive
296  *      @rq: request
297  *
298  *      This function cleans up the current PM request and stops the queue
299  *      if necessary.
300  */
301 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
302 {
303         unsigned long flags;
304
305 #ifdef DEBUG_PM
306         printk("%s: completing PM request, %s\n", drive->name,
307                blk_pm_suspend_request(rq) ? "suspend" : "resume");
308 #endif
309         spin_lock_irqsave(&ide_lock, flags);
310         if (blk_pm_suspend_request(rq)) {
311                 blk_stop_queue(drive->queue);
312         } else {
313                 drive->blocked = 0;
314                 blk_start_queue(drive->queue);
315         }
316         blkdev_dequeue_request(rq);
317         HWGROUP(drive)->rq = NULL;
318         end_that_request_last(rq, 1);
319         spin_unlock_irqrestore(&ide_lock, flags);
320 }
321
322 /*
323  * FIXME: probably move this somewhere else, name is bad too :)
324  */
325 u64 ide_get_error_location(ide_drive_t *drive, char *args)
326 {
327         u32 high, low;
328         u8 hcyl, lcyl, sect;
329         u64 sector;
330
331         high = 0;
332         hcyl = args[5];
333         lcyl = args[4];
334         sect = args[3];
335
336         if (ide_id_has_flush_cache_ext(drive->id)) {
337                 low = (hcyl << 16) | (lcyl << 8) | sect;
338                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
339                 high = ide_read_24(drive);
340         } else {
341                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
342                 if (cur & 0x40) {
343                         high = cur & 0xf;
344                         low = (hcyl << 16) | (lcyl << 8) | sect;
345                 } else {
346                         low = hcyl * drive->head * drive->sect;
347                         low += lcyl * drive->sect;
348                         low += sect - 1;
349                 }
350         }
351
352         sector = ((u64) high << 24) | low;
353         return sector;
354 }
355 EXPORT_SYMBOL(ide_get_error_location);
356
357 /**
358  *      ide_end_drive_cmd       -       end an explicit drive command
359  *      @drive: command 
360  *      @stat: status bits
361  *      @err: error bits
362  *
363  *      Clean up after success/failure of an explicit drive command.
364  *      These get thrown onto the queue so they are synchronized with
365  *      real I/O operations on the drive.
366  *
367  *      In LBA48 mode we have to read the register set twice to get
368  *      all the extra information out.
369  */
370  
371 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
372 {
373         ide_hwif_t *hwif = HWIF(drive);
374         unsigned long flags;
375         struct request *rq;
376
377         spin_lock_irqsave(&ide_lock, flags);
378         rq = HWGROUP(drive)->rq;
379         spin_unlock_irqrestore(&ide_lock, flags);
380
381         if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
382                 u8 *args = (u8 *) rq->buffer;
383                 if (rq->errors == 0)
384                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
385
386                 if (args) {
387                         args[0] = stat;
388                         args[1] = err;
389                         args[2] = hwif->INB(IDE_NSECTOR_REG);
390                 }
391         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
392                 u8 *args = (u8 *) rq->buffer;
393                 if (rq->errors == 0)
394                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
395
396                 if (args) {
397                         args[0] = stat;
398                         args[1] = err;
399                         args[2] = hwif->INB(IDE_NSECTOR_REG);
400                         args[3] = hwif->INB(IDE_SECTOR_REG);
401                         args[4] = hwif->INB(IDE_LCYL_REG);
402                         args[5] = hwif->INB(IDE_HCYL_REG);
403                         args[6] = hwif->INB(IDE_SELECT_REG);
404                 }
405         } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
406                 ide_task_t *args = (ide_task_t *) rq->special;
407                 if (rq->errors == 0)
408                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
409                         
410                 if (args) {
411                         if (args->tf_in_flags.b.data) {
412                                 u16 data                                = hwif->INW(IDE_DATA_REG);
413                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
414                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
415                         }
416                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
417                         /* be sure we're looking at the low order bits */
418                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
419                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
420                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
421                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
422                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
423                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
424                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
425
426                         if (drive->addressing == 1) {
427                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
428                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
429                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
430                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
431                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
432                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
433                         }
434                 }
435         } else if (blk_pm_request(rq)) {
436                 struct request_pm_state *pm = rq->data;
437 #ifdef DEBUG_PM
438                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
439                         drive->name, rq->pm->pm_step, stat, err);
440 #endif
441                 ide_complete_power_step(drive, rq, stat, err);
442                 if (pm->pm_step == ide_pm_state_completed)
443                         ide_complete_pm_request(drive, rq);
444                 return;
445         }
446
447         spin_lock_irqsave(&ide_lock, flags);
448         blkdev_dequeue_request(rq);
449         HWGROUP(drive)->rq = NULL;
450         rq->errors = err;
451         end_that_request_last(rq, !rq->errors);
452         spin_unlock_irqrestore(&ide_lock, flags);
453 }
454
455 EXPORT_SYMBOL(ide_end_drive_cmd);
456
457 /**
458  *      try_to_flush_leftover_data      -       flush junk
459  *      @drive: drive to flush
460  *
461  *      try_to_flush_leftover_data() is invoked in response to a drive
462  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
463  *      resetting the drive, this routine tries to clear the condition
464  *      by read a sector's worth of data from the drive.  Of course,
465  *      this may not help if the drive is *waiting* for data from *us*.
466  */
467 static void try_to_flush_leftover_data (ide_drive_t *drive)
468 {
469         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
470
471         if (drive->media != ide_disk)
472                 return;
473         while (i > 0) {
474                 u32 buffer[16];
475                 u32 wcount = (i > 16) ? 16 : i;
476
477                 i -= wcount;
478                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
479         }
480 }
481
482 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
483 {
484         if (rq->rq_disk) {
485                 ide_driver_t *drv;
486
487                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
488                 drv->end_request(drive, 0, 0);
489         } else
490                 ide_end_request(drive, 0, 0);
491 }
492
493 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
494 {
495         ide_hwif_t *hwif = drive->hwif;
496
497         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
498                 /* other bits are useless when BUSY */
499                 rq->errors |= ERROR_RESET;
500         } else if (stat & ERR_STAT) {
501                 /* err has different meaning on cdrom and tape */
502                 if (err == ABRT_ERR) {
503                         if (drive->select.b.lba &&
504                             /* some newer drives don't support WIN_SPECIFY */
505                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
506                                 return ide_stopped;
507                 } else if ((err & BAD_CRC) == BAD_CRC) {
508                         /* UDMA crc error, just retry the operation */
509                         drive->crc_count++;
510                 } else if (err & (BBD_ERR | ECC_ERR)) {
511                         /* retries won't help these */
512                         rq->errors = ERROR_MAX;
513                 } else if (err & TRK0_ERR) {
514                         /* help it find track zero */
515                         rq->errors |= ERROR_RECAL;
516                 }
517         }
518
519         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ && hwif->err_stops_fifo == 0)
520                 try_to_flush_leftover_data(drive);
521
522         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
523                 ide_kill_rq(drive, rq);
524                 return ide_stopped;
525         }
526
527         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
528                 rq->errors |= ERROR_RESET;
529
530         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
531                 ++rq->errors;
532                 return ide_do_reset(drive);
533         }
534
535         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
536                 drive->special.b.recalibrate = 1;
537
538         ++rq->errors;
539
540         return ide_stopped;
541 }
542
543 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
544 {
545         ide_hwif_t *hwif = drive->hwif;
546
547         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
548                 /* other bits are useless when BUSY */
549                 rq->errors |= ERROR_RESET;
550         } else {
551                 /* add decoding error stuff */
552         }
553
554         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
555                 /* force an abort */
556                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
557
558         if (rq->errors >= ERROR_MAX) {
559                 ide_kill_rq(drive, rq);
560         } else {
561                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
562                         ++rq->errors;
563                         return ide_do_reset(drive);
564                 }
565                 ++rq->errors;
566         }
567
568         return ide_stopped;
569 }
570
571 ide_startstop_t
572 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
573 {
574         if (drive->media == ide_disk)
575                 return ide_ata_error(drive, rq, stat, err);
576         return ide_atapi_error(drive, rq, stat, err);
577 }
578
579 EXPORT_SYMBOL_GPL(__ide_error);
580
581 /**
582  *      ide_error       -       handle an error on the IDE
583  *      @drive: drive the error occurred on
584  *      @msg: message to report
585  *      @stat: status bits
586  *
587  *      ide_error() takes action based on the error returned by the drive.
588  *      For normal I/O that may well include retries. We deal with
589  *      both new-style (taskfile) and old style command handling here.
590  *      In the case of taskfile command handling there is work left to
591  *      do
592  */
593  
594 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
595 {
596         struct request *rq;
597         u8 err;
598
599         err = ide_dump_status(drive, msg, stat);
600
601         if ((rq = HWGROUP(drive)->rq) == NULL)
602                 return ide_stopped;
603
604         /* retry only "normal" I/O: */
605         if (!blk_fs_request(rq)) {
606                 rq->errors = 1;
607                 ide_end_drive_cmd(drive, stat, err);
608                 return ide_stopped;
609         }
610
611         if (rq->rq_disk) {
612                 ide_driver_t *drv;
613
614                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
615                 return drv->error(drive, rq, stat, err);
616         } else
617                 return __ide_error(drive, rq, stat, err);
618 }
619
620 EXPORT_SYMBOL_GPL(ide_error);
621
622 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
623 {
624         if (drive->media != ide_disk)
625                 rq->errors |= ERROR_RESET;
626
627         ide_kill_rq(drive, rq);
628
629         return ide_stopped;
630 }
631
632 EXPORT_SYMBOL_GPL(__ide_abort);
633
634 /**
635  *      ide_abort       -       abort pending IDE operations
636  *      @drive: drive the error occurred on
637  *      @msg: message to report
638  *
639  *      ide_abort kills and cleans up when we are about to do a 
640  *      host initiated reset on active commands. Longer term we
641  *      want handlers to have sensible abort handling themselves
642  *
643  *      This differs fundamentally from ide_error because in 
644  *      this case the command is doing just fine when we
645  *      blow it away.
646  */
647  
648 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
649 {
650         struct request *rq;
651
652         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
653                 return ide_stopped;
654
655         /* retry only "normal" I/O: */
656         if (!blk_fs_request(rq)) {
657                 rq->errors = 1;
658                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
659                 return ide_stopped;
660         }
661
662         if (rq->rq_disk) {
663                 ide_driver_t *drv;
664
665                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
666                 return drv->abort(drive, rq);
667         } else
668                 return __ide_abort(drive, rq);
669 }
670
671 /**
672  *      ide_cmd         -       issue a simple drive command
673  *      @drive: drive the command is for
674  *      @cmd: command byte
675  *      @nsect: sector byte
676  *      @handler: handler for the command completion
677  *
678  *      Issue a simple drive command with interrupts.
679  *      The drive must be selected beforehand.
680  */
681
682 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
683                 ide_handler_t *handler)
684 {
685         ide_hwif_t *hwif = HWIF(drive);
686         if (IDE_CONTROL_REG)
687                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
688         SELECT_MASK(drive,0);
689         hwif->OUTB(nsect,IDE_NSECTOR_REG);
690         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
691 }
692
693 /**
694  *      drive_cmd_intr          -       drive command completion interrupt
695  *      @drive: drive the completion interrupt occurred on
696  *
697  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
698  *      We do any necessary data reading and then wait for the drive to
699  *      go non busy. At that point we may read the error data and complete
700  *      the request
701  */
702  
703 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
704 {
705         struct request *rq = HWGROUP(drive)->rq;
706         ide_hwif_t *hwif = HWIF(drive);
707         u8 *args = (u8 *) rq->buffer;
708         u8 stat = hwif->INB(IDE_STATUS_REG);
709         int retries = 10;
710
711         local_irq_enable_in_hardirq();
712         if ((stat & DRQ_STAT) && args && args[3]) {
713                 u8 io_32bit = drive->io_32bit;
714                 drive->io_32bit = 0;
715                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
716                 drive->io_32bit = io_32bit;
717                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
718                         udelay(100);
719         }
720
721         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
722                 return ide_error(drive, "drive_cmd", stat);
723                 /* calls ide_end_drive_cmd */
724         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
725         return ide_stopped;
726 }
727
728 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
729 {
730         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
731         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
732         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
733         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
734         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
735         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
736
737         task->handler = &set_geometry_intr;
738 }
739
740 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
741 {
742         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
743         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
744
745         task->handler = &recal_intr;
746 }
747
748 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
749 {
750         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
751         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
752
753         task->handler = &set_multmode_intr;
754 }
755
756 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
757 {
758         special_t *s = &drive->special;
759         ide_task_t args;
760
761         memset(&args, 0, sizeof(ide_task_t));
762         args.command_type = IDE_DRIVE_TASK_NO_DATA;
763
764         if (s->b.set_geometry) {
765                 s->b.set_geometry = 0;
766                 ide_init_specify_cmd(drive, &args);
767         } else if (s->b.recalibrate) {
768                 s->b.recalibrate = 0;
769                 ide_init_restore_cmd(drive, &args);
770         } else if (s->b.set_multmode) {
771                 s->b.set_multmode = 0;
772                 if (drive->mult_req > drive->id->max_multsect)
773                         drive->mult_req = drive->id->max_multsect;
774                 ide_init_setmult_cmd(drive, &args);
775         } else if (s->all) {
776                 int special = s->all;
777                 s->all = 0;
778                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
779                 return ide_stopped;
780         }
781
782         do_rw_taskfile(drive, &args);
783
784         return ide_started;
785 }
786
787 /**
788  *      do_special              -       issue some special commands
789  *      @drive: drive the command is for
790  *
791  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
792  *      commands to a drive.  It used to do much more, but has been scaled
793  *      back.
794  */
795
796 static ide_startstop_t do_special (ide_drive_t *drive)
797 {
798         special_t *s = &drive->special;
799
800 #ifdef DEBUG
801         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
802 #endif
803         if (s->b.set_tune) {
804                 s->b.set_tune = 0;
805                 if (HWIF(drive)->tuneproc != NULL)
806                         HWIF(drive)->tuneproc(drive, drive->tune_req);
807                 return ide_stopped;
808         } else {
809                 if (drive->media == ide_disk)
810                         return ide_disk_special(drive);
811
812                 s->all = 0;
813                 drive->mult_req = 0;
814                 return ide_stopped;
815         }
816 }
817
818 void ide_map_sg(ide_drive_t *drive, struct request *rq)
819 {
820         ide_hwif_t *hwif = drive->hwif;
821         struct scatterlist *sg = hwif->sg_table;
822
823         if (hwif->sg_mapped)    /* needed by ide-scsi */
824                 return;
825
826         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
827                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
828         } else {
829                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
830                 hwif->sg_nents = 1;
831         }
832 }
833
834 EXPORT_SYMBOL_GPL(ide_map_sg);
835
836 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
837 {
838         ide_hwif_t *hwif = drive->hwif;
839
840         hwif->nsect = hwif->nleft = rq->nr_sectors;
841         hwif->cursg = hwif->cursg_ofs = 0;
842 }
843
844 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
845
846 /**
847  *      execute_drive_command   -       issue special drive command
848  *      @drive: the drive to issue the command on
849  *      @rq: the request structure holding the command
850  *
851  *      execute_drive_cmd() issues a special drive command,  usually 
852  *      initiated by ioctl() from the external hdparm program. The
853  *      command can be a drive command, drive task or taskfile 
854  *      operation. Weirdly you can call it with NULL to wait for
855  *      all commands to finish. Don't do this as that is due to change
856  */
857
858 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
859                 struct request *rq)
860 {
861         ide_hwif_t *hwif = HWIF(drive);
862         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
863                 ide_task_t *args = rq->special;
864  
865                 if (!args)
866                         goto done;
867
868                 hwif->data_phase = args->data_phase;
869
870                 switch (hwif->data_phase) {
871                 case TASKFILE_MULTI_OUT:
872                 case TASKFILE_OUT:
873                 case TASKFILE_MULTI_IN:
874                 case TASKFILE_IN:
875                         ide_init_sg_cmd(drive, rq);
876                         ide_map_sg(drive, rq);
877                 default:
878                         break;
879                 }
880
881                 if (args->tf_out_flags.all != 0) 
882                         return flagged_taskfile(drive, args);
883                 return do_rw_taskfile(drive, args);
884         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
885                 u8 *args = rq->buffer;
886                 u8 sel;
887  
888                 if (!args)
889                         goto done;
890 #ifdef DEBUG
891                 printk("%s: DRIVE_TASK_CMD ", drive->name);
892                 printk("cmd=0x%02x ", args[0]);
893                 printk("fr=0x%02x ", args[1]);
894                 printk("ns=0x%02x ", args[2]);
895                 printk("sc=0x%02x ", args[3]);
896                 printk("lcyl=0x%02x ", args[4]);
897                 printk("hcyl=0x%02x ", args[5]);
898                 printk("sel=0x%02x\n", args[6]);
899 #endif
900                 hwif->OUTB(args[1], IDE_FEATURE_REG);
901                 hwif->OUTB(args[3], IDE_SECTOR_REG);
902                 hwif->OUTB(args[4], IDE_LCYL_REG);
903                 hwif->OUTB(args[5], IDE_HCYL_REG);
904                 sel = (args[6] & ~0x10);
905                 if (drive->select.b.unit)
906                         sel |= 0x10;
907                 hwif->OUTB(sel, IDE_SELECT_REG);
908                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
909                 return ide_started;
910         } else if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
911                 u8 *args = rq->buffer;
912
913                 if (!args)
914                         goto done;
915 #ifdef DEBUG
916                 printk("%s: DRIVE_CMD ", drive->name);
917                 printk("cmd=0x%02x ", args[0]);
918                 printk("sc=0x%02x ", args[1]);
919                 printk("fr=0x%02x ", args[2]);
920                 printk("xx=0x%02x\n", args[3]);
921 #endif
922                 if (args[0] == WIN_SMART) {
923                         hwif->OUTB(0x4f, IDE_LCYL_REG);
924                         hwif->OUTB(0xc2, IDE_HCYL_REG);
925                         hwif->OUTB(args[2],IDE_FEATURE_REG);
926                         hwif->OUTB(args[1],IDE_SECTOR_REG);
927                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
928                         return ide_started;
929                 }
930                 hwif->OUTB(args[2],IDE_FEATURE_REG);
931                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
932                 return ide_started;
933         }
934
935 done:
936         /*
937          * NULL is actually a valid way of waiting for
938          * all current requests to be flushed from the queue.
939          */
940 #ifdef DEBUG
941         printk("%s: DRIVE_CMD (null)\n", drive->name);
942 #endif
943         ide_end_drive_cmd(drive,
944                         hwif->INB(IDE_STATUS_REG),
945                         hwif->INB(IDE_ERROR_REG));
946         return ide_stopped;
947 }
948
949 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
950 {
951         struct request_pm_state *pm = rq->data;
952
953         if (blk_pm_suspend_request(rq) &&
954             pm->pm_step == ide_pm_state_start_suspend)
955                 /* Mark drive blocked when starting the suspend sequence. */
956                 drive->blocked = 1;
957         else if (blk_pm_resume_request(rq) &&
958                  pm->pm_step == ide_pm_state_start_resume) {
959                 /* 
960                  * The first thing we do on wakeup is to wait for BSY bit to
961                  * go away (with a looong timeout) as a drive on this hwif may
962                  * just be POSTing itself.
963                  * We do that before even selecting as the "other" device on
964                  * the bus may be broken enough to walk on our toes at this
965                  * point.
966                  */
967                 int rc;
968 #ifdef DEBUG_PM
969                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
970 #endif
971                 rc = ide_wait_not_busy(HWIF(drive), 35000);
972                 if (rc)
973                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
974                 SELECT_DRIVE(drive);
975                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
976                 rc = ide_wait_not_busy(HWIF(drive), 100000);
977                 if (rc)
978                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
979         }
980 }
981
982 /**
983  *      start_request   -       start of I/O and command issuing for IDE
984  *
985  *      start_request() initiates handling of a new I/O request. It
986  *      accepts commands and I/O (read/write) requests. It also does
987  *      the final remapping for weird stuff like EZDrive. Once 
988  *      device mapper can work sector level the EZDrive stuff can go away
989  *
990  *      FIXME: this function needs a rename
991  */
992  
993 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
994 {
995         ide_startstop_t startstop;
996         sector_t block;
997
998         BUG_ON(!blk_rq_started(rq));
999
1000 #ifdef DEBUG
1001         printk("%s: start_request: current=0x%08lx\n",
1002                 HWIF(drive)->name, (unsigned long) rq);
1003 #endif
1004
1005         /* bail early if we've exceeded max_failures */
1006         if (drive->max_failures && (drive->failures > drive->max_failures)) {
1007                 goto kill_rq;
1008         }
1009
1010         block    = rq->sector;
1011         if (blk_fs_request(rq) &&
1012             (drive->media == ide_disk || drive->media == ide_floppy)) {
1013                 block += drive->sect0;
1014         }
1015         /* Yecch - this will shift the entire interval,
1016            possibly killing some innocent following sector */
1017         if (block == 0 && drive->remap_0_to_1 == 1)
1018                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
1019
1020         if (blk_pm_request(rq))
1021                 ide_check_pm_state(drive, rq);
1022
1023         SELECT_DRIVE(drive);
1024         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1025                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1026                 return startstop;
1027         }
1028         if (!drive->special.all) {
1029                 ide_driver_t *drv;
1030
1031                 /*
1032                  * We reset the drive so we need to issue a SETFEATURES.
1033                  * Do it _after_ do_special() restored device parameters.
1034                  */
1035                 if (drive->current_speed == 0xff)
1036                         ide_config_drive_speed(drive, drive->desired_speed);
1037
1038                 if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1039                     rq->cmd_type == REQ_TYPE_ATA_TASK ||
1040                     rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1041                         return execute_drive_cmd(drive, rq);
1042                 else if (blk_pm_request(rq)) {
1043                         struct request_pm_state *pm = rq->data;
1044 #ifdef DEBUG_PM
1045                         printk("%s: start_power_step(step: %d)\n",
1046                                 drive->name, rq->pm->pm_step);
1047 #endif
1048                         startstop = ide_start_power_step(drive, rq);
1049                         if (startstop == ide_stopped &&
1050                             pm->pm_step == ide_pm_state_completed)
1051                                 ide_complete_pm_request(drive, rq);
1052                         return startstop;
1053                 }
1054
1055                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1056                 return drv->do_request(drive, rq, block);
1057         }
1058         return do_special(drive);
1059 kill_rq:
1060         ide_kill_rq(drive, rq);
1061         return ide_stopped;
1062 }
1063
1064 /**
1065  *      ide_stall_queue         -       pause an IDE device
1066  *      @drive: drive to stall
1067  *      @timeout: time to stall for (jiffies)
1068  *
1069  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1070  *      to the hwgroup by sleeping for timeout jiffies.
1071  */
1072  
1073 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1074 {
1075         if (timeout > WAIT_WORSTCASE)
1076                 timeout = WAIT_WORSTCASE;
1077         drive->sleep = timeout + jiffies;
1078         drive->sleeping = 1;
1079 }
1080
1081 EXPORT_SYMBOL(ide_stall_queue);
1082
1083 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1084
1085 /**
1086  *      choose_drive            -       select a drive to service
1087  *      @hwgroup: hardware group to select on
1088  *
1089  *      choose_drive() selects the next drive which will be serviced.
1090  *      This is necessary because the IDE layer can't issue commands
1091  *      to both drives on the same cable, unlike SCSI.
1092  */
1093  
1094 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1095 {
1096         ide_drive_t *drive, *best;
1097
1098 repeat: 
1099         best = NULL;
1100         drive = hwgroup->drive;
1101
1102         /*
1103          * drive is doing pre-flush, ordered write, post-flush sequence. even
1104          * though that is 3 requests, it must be seen as a single transaction.
1105          * we must not preempt this drive until that is complete
1106          */
1107         if (blk_queue_flushing(drive->queue)) {
1108                 /*
1109                  * small race where queue could get replugged during
1110                  * the 3-request flush cycle, just yank the plug since
1111                  * we want it to finish asap
1112                  */
1113                 blk_remove_plug(drive->queue);
1114                 return drive;
1115         }
1116
1117         do {
1118                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1119                     && !elv_queue_empty(drive->queue)) {
1120                         if (!best
1121                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1122                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1123                         {
1124                                 if (!blk_queue_plugged(drive->queue))
1125                                         best = drive;
1126                         }
1127                 }
1128         } while ((drive = drive->next) != hwgroup->drive);
1129         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1130                 long t = (signed long)(WAKEUP(best) - jiffies);
1131                 if (t >= WAIT_MIN_SLEEP) {
1132                 /*
1133                  * We *may* have some time to spare, but first let's see if
1134                  * someone can potentially benefit from our nice mood today..
1135                  */
1136                         drive = best->next;
1137                         do {
1138                                 if (!drive->sleeping
1139                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1140                                  && time_before(WAKEUP(drive), jiffies + t))
1141                                 {
1142                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1143                                         goto repeat;
1144                                 }
1145                         } while ((drive = drive->next) != best);
1146                 }
1147         }
1148         return best;
1149 }
1150
1151 /*
1152  * Issue a new request to a drive from hwgroup
1153  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1154  *
1155  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1156  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1157  * may have both interfaces in a single hwgroup to "serialize" access.
1158  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1159  * together into one hwgroup for serialized access.
1160  *
1161  * Note also that several hwgroups can end up sharing a single IRQ,
1162  * possibly along with many other devices.  This is especially common in
1163  * PCI-based systems with off-board IDE controller cards.
1164  *
1165  * The IDE driver uses the single global ide_lock spinlock to protect
1166  * access to the request queues, and to protect the hwgroup->busy flag.
1167  *
1168  * The first thread into the driver for a particular hwgroup sets the
1169  * hwgroup->busy flag to indicate that this hwgroup is now active,
1170  * and then initiates processing of the top request from the request queue.
1171  *
1172  * Other threads attempting entry notice the busy setting, and will simply
1173  * queue their new requests and exit immediately.  Note that hwgroup->busy
1174  * remains set even when the driver is merely awaiting the next interrupt.
1175  * Thus, the meaning is "this hwgroup is busy processing a request".
1176  *
1177  * When processing of a request completes, the completing thread or IRQ-handler
1178  * will start the next request from the queue.  If no more work remains,
1179  * the driver will clear the hwgroup->busy flag and exit.
1180  *
1181  * The ide_lock (spinlock) is used to protect all access to the
1182  * hwgroup->busy flag, but is otherwise not needed for most processing in
1183  * the driver.  This makes the driver much more friendlier to shared IRQs
1184  * than previous designs, while remaining 100% (?) SMP safe and capable.
1185  */
1186 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1187 {
1188         ide_drive_t     *drive;
1189         ide_hwif_t      *hwif;
1190         struct request  *rq;
1191         ide_startstop_t startstop;
1192         int             loops = 0;
1193
1194         /* for atari only: POSSIBLY BROKEN HERE(?) */
1195         ide_get_lock(ide_intr, hwgroup);
1196
1197         /* caller must own ide_lock */
1198         BUG_ON(!irqs_disabled());
1199
1200         while (!hwgroup->busy) {
1201                 hwgroup->busy = 1;
1202                 drive = choose_drive(hwgroup);
1203                 if (drive == NULL) {
1204                         int sleeping = 0;
1205                         unsigned long sleep = 0; /* shut up, gcc */
1206                         hwgroup->rq = NULL;
1207                         drive = hwgroup->drive;
1208                         do {
1209                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1210                                         sleeping = 1;
1211                                         sleep = drive->sleep;
1212                                 }
1213                         } while ((drive = drive->next) != hwgroup->drive);
1214                         if (sleeping) {
1215                 /*
1216                  * Take a short snooze, and then wake up this hwgroup again.
1217                  * This gives other hwgroups on the same a chance to
1218                  * play fairly with us, just in case there are big differences
1219                  * in relative throughputs.. don't want to hog the cpu too much.
1220                  */
1221                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1222                                         sleep = jiffies + WAIT_MIN_SLEEP;
1223 #if 1
1224                                 if (timer_pending(&hwgroup->timer))
1225                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1226 #endif
1227                                 /* so that ide_timer_expiry knows what to do */
1228                                 hwgroup->sleeping = 1;
1229                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1230                                 mod_timer(&hwgroup->timer, sleep);
1231                                 /* we purposely leave hwgroup->busy==1
1232                                  * while sleeping */
1233                         } else {
1234                                 /* Ugly, but how can we sleep for the lock
1235                                  * otherwise? perhaps from tq_disk?
1236                                  */
1237
1238                                 /* for atari only */
1239                                 ide_release_lock();
1240                                 hwgroup->busy = 0;
1241                         }
1242
1243                         /* no more work for this hwgroup (for now) */
1244                         return;
1245                 }
1246         again:
1247                 hwif = HWIF(drive);
1248                 if (hwgroup->hwif->sharing_irq &&
1249                     hwif != hwgroup->hwif &&
1250                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1251                         /* set nIEN for previous hwif */
1252                         SELECT_INTERRUPT(drive);
1253                 }
1254                 hwgroup->hwif = hwif;
1255                 hwgroup->drive = drive;
1256                 drive->sleeping = 0;
1257                 drive->service_start = jiffies;
1258
1259                 if (blk_queue_plugged(drive->queue)) {
1260                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1261                         break;
1262                 }
1263
1264                 /*
1265                  * we know that the queue isn't empty, but this can happen
1266                  * if the q->prep_rq_fn() decides to kill a request
1267                  */
1268                 rq = elv_next_request(drive->queue);
1269                 if (!rq) {
1270                         hwgroup->busy = 0;
1271                         break;
1272                 }
1273
1274                 /*
1275                  * Sanity: don't accept a request that isn't a PM request
1276                  * if we are currently power managed. This is very important as
1277                  * blk_stop_queue() doesn't prevent the elv_next_request()
1278                  * above to return us whatever is in the queue. Since we call
1279                  * ide_do_request() ourselves, we end up taking requests while
1280                  * the queue is blocked...
1281                  * 
1282                  * We let requests forced at head of queue with ide-preempt
1283                  * though. I hope that doesn't happen too much, hopefully not
1284                  * unless the subdriver triggers such a thing in its own PM
1285                  * state machine.
1286                  *
1287                  * We count how many times we loop here to make sure we service
1288                  * all drives in the hwgroup without looping for ever
1289                  */
1290                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1291                         drive = drive->next ? drive->next : hwgroup->drive;
1292                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1293                                 goto again;
1294                         /* We clear busy, there should be no pending ATA command at this point. */
1295                         hwgroup->busy = 0;
1296                         break;
1297                 }
1298
1299                 hwgroup->rq = rq;
1300
1301                 /*
1302                  * Some systems have trouble with IDE IRQs arriving while
1303                  * the driver is still setting things up.  So, here we disable
1304                  * the IRQ used by this interface while the request is being started.
1305                  * This may look bad at first, but pretty much the same thing
1306                  * happens anyway when any interrupt comes in, IDE or otherwise
1307                  *  -- the kernel masks the IRQ while it is being handled.
1308                  */
1309                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1310                         disable_irq_nosync(hwif->irq);
1311                 spin_unlock(&ide_lock);
1312                 local_irq_enable_in_hardirq();
1313                         /* allow other IRQs while we start this request */
1314                 startstop = start_request(drive, rq);
1315                 spin_lock_irq(&ide_lock);
1316                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1317                         enable_irq(hwif->irq);
1318                 if (startstop == ide_stopped)
1319                         hwgroup->busy = 0;
1320         }
1321 }
1322
1323 /*
1324  * Passes the stuff to ide_do_request
1325  */
1326 void do_ide_request(request_queue_t *q)
1327 {
1328         ide_drive_t *drive = q->queuedata;
1329
1330         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1331 }
1332
1333 /*
1334  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1335  * retry the current request in pio mode instead of risking tossing it
1336  * all away
1337  */
1338 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1339 {
1340         ide_hwif_t *hwif = HWIF(drive);
1341         struct request *rq;
1342         ide_startstop_t ret = ide_stopped;
1343
1344         /*
1345          * end current dma transaction
1346          */
1347
1348         if (error < 0) {
1349                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1350                 (void)HWIF(drive)->ide_dma_end(drive);
1351                 ret = ide_error(drive, "dma timeout error",
1352                                                 hwif->INB(IDE_STATUS_REG));
1353         } else {
1354                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1355                 (void) hwif->ide_dma_timeout(drive);
1356         }
1357
1358         /*
1359          * disable dma for now, but remember that we did so because of
1360          * a timeout -- we'll reenable after we finish this next request
1361          * (or rather the first chunk of it) in pio.
1362          */
1363         drive->retry_pio++;
1364         drive->state = DMA_PIO_RETRY;
1365         hwif->dma_off_quietly(drive);
1366
1367         /*
1368          * un-busy drive etc (hwgroup->busy is cleared on return) and
1369          * make sure request is sane
1370          */
1371         rq = HWGROUP(drive)->rq;
1372
1373         if (!rq)
1374                 goto out;
1375
1376         HWGROUP(drive)->rq = NULL;
1377
1378         rq->errors = 0;
1379
1380         if (!rq->bio)
1381                 goto out;
1382
1383         rq->sector = rq->bio->bi_sector;
1384         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1385         rq->hard_cur_sectors = rq->current_nr_sectors;
1386         rq->buffer = bio_data(rq->bio);
1387 out:
1388         return ret;
1389 }
1390
1391 /**
1392  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1393  *      @data: timer callback magic (hwgroup)
1394  *
1395  *      An IDE command has timed out before the expected drive return
1396  *      occurred. At this point we attempt to clean up the current
1397  *      mess. If the current handler includes an expiry handler then
1398  *      we invoke the expiry handler, and providing it is happy the
1399  *      work is done. If that fails we apply generic recovery rules
1400  *      invoking the handler and checking the drive DMA status. We
1401  *      have an excessively incestuous relationship with the DMA
1402  *      logic that wants cleaning up.
1403  */
1404  
1405 void ide_timer_expiry (unsigned long data)
1406 {
1407         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1408         ide_handler_t   *handler;
1409         ide_expiry_t    *expiry;
1410         unsigned long   flags;
1411         unsigned long   wait = -1;
1412
1413         spin_lock_irqsave(&ide_lock, flags);
1414
1415         if (((handler = hwgroup->handler) == NULL) ||
1416             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1417                 /*
1418                  * Either a marginal timeout occurred
1419                  * (got the interrupt just as timer expired),
1420                  * or we were "sleeping" to give other devices a chance.
1421                  * Either way, we don't really want to complain about anything.
1422                  */
1423                 if (hwgroup->sleeping) {
1424                         hwgroup->sleeping = 0;
1425                         hwgroup->busy = 0;
1426                 }
1427         } else {
1428                 ide_drive_t *drive = hwgroup->drive;
1429                 if (!drive) {
1430                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1431                         hwgroup->handler = NULL;
1432                 } else {
1433                         ide_hwif_t *hwif;
1434                         ide_startstop_t startstop = ide_stopped;
1435                         if (!hwgroup->busy) {
1436                                 hwgroup->busy = 1;      /* paranoia */
1437                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1438                         }
1439                         if ((expiry = hwgroup->expiry) != NULL) {
1440                                 /* continue */
1441                                 if ((wait = expiry(drive)) > 0) {
1442                                         /* reset timer */
1443                                         hwgroup->timer.expires  = jiffies + wait;
1444                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1445                                         add_timer(&hwgroup->timer);
1446                                         spin_unlock_irqrestore(&ide_lock, flags);
1447                                         return;
1448                                 }
1449                         }
1450                         hwgroup->handler = NULL;
1451                         /*
1452                          * We need to simulate a real interrupt when invoking
1453                          * the handler() function, which means we need to
1454                          * globally mask the specific IRQ:
1455                          */
1456                         spin_unlock(&ide_lock);
1457                         hwif  = HWIF(drive);
1458 #if DISABLE_IRQ_NOSYNC
1459                         disable_irq_nosync(hwif->irq);
1460 #else
1461                         /* disable_irq_nosync ?? */
1462                         disable_irq(hwif->irq);
1463 #endif /* DISABLE_IRQ_NOSYNC */
1464                         /* local CPU only,
1465                          * as if we were handling an interrupt */
1466                         local_irq_disable();
1467                         if (hwgroup->polling) {
1468                                 startstop = handler(drive);
1469                         } else if (drive_is_ready(drive)) {
1470                                 if (drive->waiting_for_dma)
1471                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1472                                 (void)ide_ack_intr(hwif);
1473                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1474                                 startstop = handler(drive);
1475                         } else {
1476                                 if (drive->waiting_for_dma) {
1477                                         startstop = ide_dma_timeout_retry(drive, wait);
1478                                 } else
1479                                         startstop =
1480                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1481                         }
1482                         drive->service_time = jiffies - drive->service_start;
1483                         spin_lock_irq(&ide_lock);
1484                         enable_irq(hwif->irq);
1485                         if (startstop == ide_stopped)
1486                                 hwgroup->busy = 0;
1487                 }
1488         }
1489         ide_do_request(hwgroup, IDE_NO_IRQ);
1490         spin_unlock_irqrestore(&ide_lock, flags);
1491 }
1492
1493 /**
1494  *      unexpected_intr         -       handle an unexpected IDE interrupt
1495  *      @irq: interrupt line
1496  *      @hwgroup: hwgroup being processed
1497  *
1498  *      There's nothing really useful we can do with an unexpected interrupt,
1499  *      other than reading the status register (to clear it), and logging it.
1500  *      There should be no way that an irq can happen before we're ready for it,
1501  *      so we needn't worry much about losing an "important" interrupt here.
1502  *
1503  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1504  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1505  *      looks "good", we just ignore the interrupt completely.
1506  *
1507  *      This routine assumes __cli() is in effect when called.
1508  *
1509  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1510  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1511  *      we could screw up by interfering with a new request being set up for 
1512  *      irq15.
1513  *
1514  *      In reality, this is a non-issue.  The new command is not sent unless 
1515  *      the drive is ready to accept one, in which case we know the drive is
1516  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1517  *      before completing the issuance of any new drive command, so we will not
1518  *      be accidentally invoked as a result of any valid command completion
1519  *      interrupt.
1520  *
1521  *      Note that we must walk the entire hwgroup here. We know which hwif
1522  *      is doing the current command, but we don't know which hwif burped
1523  *      mysteriously.
1524  */
1525  
1526 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1527 {
1528         u8 stat;
1529         ide_hwif_t *hwif = hwgroup->hwif;
1530
1531         /*
1532          * handle the unexpected interrupt
1533          */
1534         do {
1535                 if (hwif->irq == irq) {
1536                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1537                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1538                                 /* Try to not flood the console with msgs */
1539                                 static unsigned long last_msgtime, count;
1540                                 ++count;
1541                                 if (time_after(jiffies, last_msgtime + HZ)) {
1542                                         last_msgtime = jiffies;
1543                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1544                                                 "status=0x%02x, count=%ld\n",
1545                                                 hwif->name,
1546                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1547                                 }
1548                         }
1549                 }
1550         } while ((hwif = hwif->next) != hwgroup->hwif);
1551 }
1552
1553 /**
1554  *      ide_intr        -       default IDE interrupt handler
1555  *      @irq: interrupt number
1556  *      @dev_id: hwif group
1557  *      @regs: unused weirdness from the kernel irq layer
1558  *
1559  *      This is the default IRQ handler for the IDE layer. You should
1560  *      not need to override it. If you do be aware it is subtle in
1561  *      places
1562  *
1563  *      hwgroup->hwif is the interface in the group currently performing
1564  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1565  *      the IRQ handler to call. As we issue a command the handlers
1566  *      step through multiple states, reassigning the handler to the
1567  *      next step in the process. Unlike a smart SCSI controller IDE
1568  *      expects the main processor to sequence the various transfer
1569  *      stages. We also manage a poll timer to catch up with most
1570  *      timeout situations. There are still a few where the handlers
1571  *      don't ever decide to give up.
1572  *
1573  *      The handler eventually returns ide_stopped to indicate the
1574  *      request completed. At this point we issue the next request
1575  *      on the hwgroup and the process begins again.
1576  */
1577  
1578 irqreturn_t ide_intr (int irq, void *dev_id)
1579 {
1580         unsigned long flags;
1581         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1582         ide_hwif_t *hwif;
1583         ide_drive_t *drive;
1584         ide_handler_t *handler;
1585         ide_startstop_t startstop;
1586
1587         spin_lock_irqsave(&ide_lock, flags);
1588         hwif = hwgroup->hwif;
1589
1590         if (!ide_ack_intr(hwif)) {
1591                 spin_unlock_irqrestore(&ide_lock, flags);
1592                 return IRQ_NONE;
1593         }
1594
1595         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1596                 /*
1597                  * Not expecting an interrupt from this drive.
1598                  * That means this could be:
1599                  *      (1) an interrupt from another PCI device
1600                  *      sharing the same PCI INT# as us.
1601                  * or   (2) a drive just entered sleep or standby mode,
1602                  *      and is interrupting to let us know.
1603                  * or   (3) a spurious interrupt of unknown origin.
1604                  *
1605                  * For PCI, we cannot tell the difference,
1606                  * so in that case we just ignore it and hope it goes away.
1607                  *
1608                  * FIXME: unexpected_intr should be hwif-> then we can
1609                  * remove all the ifdef PCI crap
1610                  */
1611 #ifdef CONFIG_BLK_DEV_IDEPCI
1612                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1613 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1614                 {
1615                         /*
1616                          * Probably not a shared PCI interrupt,
1617                          * so we can safely try to do something about it:
1618                          */
1619                         unexpected_intr(irq, hwgroup);
1620 #ifdef CONFIG_BLK_DEV_IDEPCI
1621                 } else {
1622                         /*
1623                          * Whack the status register, just in case
1624                          * we have a leftover pending IRQ.
1625                          */
1626                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1627 #endif /* CONFIG_BLK_DEV_IDEPCI */
1628                 }
1629                 spin_unlock_irqrestore(&ide_lock, flags);
1630                 return IRQ_NONE;
1631         }
1632         drive = hwgroup->drive;
1633         if (!drive) {
1634                 /*
1635                  * This should NEVER happen, and there isn't much
1636                  * we could do about it here.
1637                  *
1638                  * [Note - this can occur if the drive is hot unplugged]
1639                  */
1640                 spin_unlock_irqrestore(&ide_lock, flags);
1641                 return IRQ_HANDLED;
1642         }
1643         if (!drive_is_ready(drive)) {
1644                 /*
1645                  * This happens regularly when we share a PCI IRQ with
1646                  * another device.  Unfortunately, it can also happen
1647                  * with some buggy drives that trigger the IRQ before
1648                  * their status register is up to date.  Hopefully we have
1649                  * enough advance overhead that the latter isn't a problem.
1650                  */
1651                 spin_unlock_irqrestore(&ide_lock, flags);
1652                 return IRQ_NONE;
1653         }
1654         if (!hwgroup->busy) {
1655                 hwgroup->busy = 1;      /* paranoia */
1656                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1657         }
1658         hwgroup->handler = NULL;
1659         hwgroup->req_gen++;
1660         del_timer(&hwgroup->timer);
1661         spin_unlock(&ide_lock);
1662
1663         /* Some controllers might set DMA INTR no matter DMA or PIO;
1664          * bmdma status might need to be cleared even for
1665          * PIO interrupts to prevent spurious/lost irq.
1666          */
1667         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1668                 /* ide_dma_end() needs bmdma status for error checking.
1669                  * So, skip clearing bmdma status here and leave it
1670                  * to ide_dma_end() if this is dma interrupt.
1671                  */
1672                 hwif->ide_dma_clear_irq(drive);
1673
1674         if (drive->unmask)
1675                 local_irq_enable_in_hardirq();
1676         /* service this interrupt, may set handler for next interrupt */
1677         startstop = handler(drive);
1678         spin_lock_irq(&ide_lock);
1679
1680         /*
1681          * Note that handler() may have set things up for another
1682          * interrupt to occur soon, but it cannot happen until
1683          * we exit from this routine, because it will be the
1684          * same irq as is currently being serviced here, and Linux
1685          * won't allow another of the same (on any CPU) until we return.
1686          */
1687         drive->service_time = jiffies - drive->service_start;
1688         if (startstop == ide_stopped) {
1689                 if (hwgroup->handler == NULL) { /* paranoia */
1690                         hwgroup->busy = 0;
1691                         ide_do_request(hwgroup, hwif->irq);
1692                 } else {
1693                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1694                                 "on exit\n", drive->name);
1695                 }
1696         }
1697         spin_unlock_irqrestore(&ide_lock, flags);
1698         return IRQ_HANDLED;
1699 }
1700
1701 /**
1702  *      ide_init_drive_cmd      -       initialize a drive command request
1703  *      @rq: request object
1704  *
1705  *      Initialize a request before we fill it in and send it down to
1706  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1707  *      now it doesn't do a lot, but if that changes abusers will have a
1708  *      nasty surprise.
1709  */
1710
1711 void ide_init_drive_cmd (struct request *rq)
1712 {
1713         memset(rq, 0, sizeof(*rq));
1714         rq->cmd_type = REQ_TYPE_ATA_CMD;
1715         rq->ref_count = 1;
1716 }
1717
1718 EXPORT_SYMBOL(ide_init_drive_cmd);
1719
1720 /**
1721  *      ide_do_drive_cmd        -       issue IDE special command
1722  *      @drive: device to issue command
1723  *      @rq: request to issue
1724  *      @action: action for processing
1725  *
1726  *      This function issues a special IDE device request
1727  *      onto the request queue.
1728  *
1729  *      If action is ide_wait, then the rq is queued at the end of the
1730  *      request queue, and the function sleeps until it has been processed.
1731  *      This is for use when invoked from an ioctl handler.
1732  *
1733  *      If action is ide_preempt, then the rq is queued at the head of
1734  *      the request queue, displacing the currently-being-processed
1735  *      request and this function returns immediately without waiting
1736  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1737  *      intended for careful use by the ATAPI tape/cdrom driver code.
1738  *
1739  *      If action is ide_end, then the rq is queued at the end of the
1740  *      request queue, and the function returns immediately without waiting
1741  *      for the new rq to be completed. This is again intended for careful
1742  *      use by the ATAPI tape/cdrom driver code.
1743  */
1744  
1745 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1746 {
1747         unsigned long flags;
1748         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1749         DECLARE_COMPLETION_ONSTACK(wait);
1750         int where = ELEVATOR_INSERT_BACK, err;
1751         int must_wait = (action == ide_wait || action == ide_head_wait);
1752
1753         rq->errors = 0;
1754
1755         /*
1756          * we need to hold an extra reference to request for safe inspection
1757          * after completion
1758          */
1759         if (must_wait) {
1760                 rq->ref_count++;
1761                 rq->end_io_data = &wait;
1762                 rq->end_io = blk_end_sync_rq;
1763         }
1764
1765         spin_lock_irqsave(&ide_lock, flags);
1766         if (action == ide_preempt)
1767                 hwgroup->rq = NULL;
1768         if (action == ide_preempt || action == ide_head_wait) {
1769                 where = ELEVATOR_INSERT_FRONT;
1770                 rq->cmd_flags |= REQ_PREEMPT;
1771         }
1772         __elv_add_request(drive->queue, rq, where, 0);
1773         ide_do_request(hwgroup, IDE_NO_IRQ);
1774         spin_unlock_irqrestore(&ide_lock, flags);
1775
1776         err = 0;
1777         if (must_wait) {
1778                 wait_for_completion(&wait);
1779                 if (rq->errors)
1780                         err = -EIO;
1781
1782                 blk_put_request(rq);
1783         }
1784
1785         return err;
1786 }
1787
1788 EXPORT_SYMBOL(ide_do_drive_cmd);