]> pilppa.org Git - linux-2.6-omap-h63xx.git/blob - drivers/ide/ide-io.c
ide: execute_drive_cmd() cleanup
[linux-2.6-omap-h63xx.git] / drivers / ide / ide-io.c
1 /*
2  *      IDE I/O functions
3  *
4  *      Basic PIO and command management functionality.
5  *
6  * This code was split off from ide.c. See ide.c for history and original
7  * copyrights.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the
11  * Free Software Foundation; either version 2, or (at your option) any
12  * later version.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * For the avoidance of doubt the "preferred form" of this code is one which
20  * is in an open non patent encumbered format. Where cryptographic key signing
21  * forms part of the process of creating an executable the information
22  * including keys needed to generate an equivalently functional executable
23  * are deemed to be part of the source code.
24  */
25  
26  
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
32 #include <linux/mm.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/completion.h>
44 #include <linux/reboot.h>
45 #include <linux/cdrom.h>
46 #include <linux/seq_file.h>
47 #include <linux/device.h>
48 #include <linux/kmod.h>
49 #include <linux/scatterlist.h>
50 #include <linux/bitops.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/irq.h>
54 #include <asm/uaccess.h>
55 #include <asm/io.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, unsigned int nr_bytes, int dequeue)
59 {
60         int ret = 1;
61
62         /*
63          * if failfast is set on a request, override number of sectors and
64          * complete the whole request right now
65          */
66         if (blk_noretry_request(rq) && end_io_error(uptodate))
67                 nr_bytes = rq->hard_nr_sectors << 9;
68
69         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
70                 rq->errors = -EIO;
71
72         /*
73          * decide whether to reenable DMA -- 3 is a random magic for now,
74          * if we DMA timeout more than 3 times, just stay in PIO
75          */
76         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
77                 drive->state = 0;
78                 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                 drive->hwif->dma_off_quietly(drive);
225                 /*
226                  * TODO: respect ->using_dma setting
227                  */
228                 ide_set_dma(drive);
229                 break;
230         }
231         pm->pm_step = ide_pm_state_completed;
232         return ide_stopped;
233
234 out_do_tf:
235         args->tf_flags = IDE_TFLAG_OUT_TF;
236         if (drive->addressing == 1)
237                 args->tf_flags |= (IDE_TFLAG_LBA48 | IDE_TFLAG_OUT_HOB);
238         args->command_type = IDE_DRIVE_TASK_NO_DATA;
239         args->handler      = task_no_data_intr;
240         return do_rw_taskfile(drive, args);
241 }
242
243 /**
244  *      ide_end_dequeued_request        -       complete an IDE I/O
245  *      @drive: IDE device for the I/O
246  *      @uptodate:
247  *      @nr_sectors: number of sectors completed
248  *
249  *      Complete an I/O that is no longer on the request queue. This
250  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
251  *      We must still finish the old request but we must not tamper with the
252  *      queue in the meantime.
253  *
254  *      NOTE: This path does not handle barrier, but barrier is not supported
255  *      on ide-cd anyway.
256  */
257
258 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
259                              int uptodate, int nr_sectors)
260 {
261         unsigned long flags;
262         int ret;
263
264         spin_lock_irqsave(&ide_lock, flags);
265         BUG_ON(!blk_rq_started(rq));
266         ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
267         spin_unlock_irqrestore(&ide_lock, flags);
268
269         return ret;
270 }
271 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
272
273
274 /**
275  *      ide_complete_pm_request - end the current Power Management request
276  *      @drive: target drive
277  *      @rq: request
278  *
279  *      This function cleans up the current PM request and stops the queue
280  *      if necessary.
281  */
282 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
283 {
284         unsigned long flags;
285
286 #ifdef DEBUG_PM
287         printk("%s: completing PM request, %s\n", drive->name,
288                blk_pm_suspend_request(rq) ? "suspend" : "resume");
289 #endif
290         spin_lock_irqsave(&ide_lock, flags);
291         if (blk_pm_suspend_request(rq)) {
292                 blk_stop_queue(drive->queue);
293         } else {
294                 drive->blocked = 0;
295                 blk_start_queue(drive->queue);
296         }
297         blkdev_dequeue_request(rq);
298         HWGROUP(drive)->rq = NULL;
299         end_that_request_last(rq, 1);
300         spin_unlock_irqrestore(&ide_lock, flags);
301 }
302
303 /**
304  *      ide_end_drive_cmd       -       end an explicit drive command
305  *      @drive: command 
306  *      @stat: status bits
307  *      @err: error bits
308  *
309  *      Clean up after success/failure of an explicit drive command.
310  *      These get thrown onto the queue so they are synchronized with
311  *      real I/O operations on the drive.
312  *
313  *      In LBA48 mode we have to read the register set twice to get
314  *      all the extra information out.
315  */
316  
317 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
318 {
319         ide_hwif_t *hwif = HWIF(drive);
320         unsigned long flags;
321         struct request *rq;
322
323         spin_lock_irqsave(&ide_lock, flags);
324         rq = HWGROUP(drive)->rq;
325         spin_unlock_irqrestore(&ide_lock, flags);
326
327         if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
328                 u8 *args = (u8 *) rq->buffer;
329                 if (rq->errors == 0)
330                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
331
332                 if (args) {
333                         args[0] = stat;
334                         args[1] = err;
335                         args[2] = hwif->INB(IDE_NSECTOR_REG);
336                 }
337         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
338                 u8 *args = (u8 *) rq->buffer;
339                 if (rq->errors == 0)
340                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
341
342                 if (args) {
343                         args[0] = stat;
344                         args[1] = err;
345                         /* be sure we're looking at the low order bits */
346                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
347                         args[2] = hwif->INB(IDE_NSECTOR_REG);
348                         args[3] = hwif->INB(IDE_SECTOR_REG);
349                         args[4] = hwif->INB(IDE_LCYL_REG);
350                         args[5] = hwif->INB(IDE_HCYL_REG);
351                         args[6] = hwif->INB(IDE_SELECT_REG);
352                 }
353         } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
354                 ide_task_t *args = (ide_task_t *) rq->special;
355                 if (rq->errors == 0)
356                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
357                         
358                 if (args) {
359                         struct ide_taskfile *tf = &args->tf;
360
361                         if (args->tf_in_flags.b.data) {
362                                 u16 data = hwif->INW(IDE_DATA_REG);
363
364                                 tf->data = data & 0xff;
365                                 tf->hob_data = (data >> 8) & 0xff;
366                         }
367                         tf->error = err;
368                         /* be sure we're looking at the low order bits */
369                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
370                         tf->nsect  = hwif->INB(IDE_NSECTOR_REG);
371                         tf->lbal   = hwif->INB(IDE_SECTOR_REG);
372                         tf->lbam   = hwif->INB(IDE_LCYL_REG);
373                         tf->lbah   = hwif->INB(IDE_HCYL_REG);
374                         tf->device = hwif->INB(IDE_SELECT_REG);
375                         tf->status = stat;
376
377                         if (drive->addressing == 1) {
378                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
379                                 tf->hob_feature = hwif->INB(IDE_FEATURE_REG);
380                                 tf->hob_nsect   = hwif->INB(IDE_NSECTOR_REG);
381                                 tf->hob_lbal    = hwif->INB(IDE_SECTOR_REG);
382                                 tf->hob_lbam    = hwif->INB(IDE_LCYL_REG);
383                                 tf->hob_lbah    = hwif->INB(IDE_HCYL_REG);
384                         }
385                 }
386         } else if (blk_pm_request(rq)) {
387                 struct request_pm_state *pm = rq->data;
388 #ifdef DEBUG_PM
389                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
390                         drive->name, rq->pm->pm_step, stat, err);
391 #endif
392                 ide_complete_power_step(drive, rq, stat, err);
393                 if (pm->pm_step == ide_pm_state_completed)
394                         ide_complete_pm_request(drive, rq);
395                 return;
396         }
397
398         spin_lock_irqsave(&ide_lock, flags);
399         blkdev_dequeue_request(rq);
400         HWGROUP(drive)->rq = NULL;
401         rq->errors = err;
402         end_that_request_last(rq, !rq->errors);
403         spin_unlock_irqrestore(&ide_lock, flags);
404 }
405
406 EXPORT_SYMBOL(ide_end_drive_cmd);
407
408 /**
409  *      try_to_flush_leftover_data      -       flush junk
410  *      @drive: drive to flush
411  *
412  *      try_to_flush_leftover_data() is invoked in response to a drive
413  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
414  *      resetting the drive, this routine tries to clear the condition
415  *      by read a sector's worth of data from the drive.  Of course,
416  *      this may not help if the drive is *waiting* for data from *us*.
417  */
418 static void try_to_flush_leftover_data (ide_drive_t *drive)
419 {
420         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
421
422         if (drive->media != ide_disk)
423                 return;
424         while (i > 0) {
425                 u32 buffer[16];
426                 u32 wcount = (i > 16) ? 16 : i;
427
428                 i -= wcount;
429                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
430         }
431 }
432
433 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
434 {
435         if (rq->rq_disk) {
436                 ide_driver_t *drv;
437
438                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
439                 drv->end_request(drive, 0, 0);
440         } else
441                 ide_end_request(drive, 0, 0);
442 }
443
444 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
445 {
446         ide_hwif_t *hwif = drive->hwif;
447
448         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
449                 /* other bits are useless when BUSY */
450                 rq->errors |= ERROR_RESET;
451         } else if (stat & ERR_STAT) {
452                 /* err has different meaning on cdrom and tape */
453                 if (err == ABRT_ERR) {
454                         if (drive->select.b.lba &&
455                             /* some newer drives don't support WIN_SPECIFY */
456                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
457                                 return ide_stopped;
458                 } else if ((err & BAD_CRC) == BAD_CRC) {
459                         /* UDMA crc error, just retry the operation */
460                         drive->crc_count++;
461                 } else if (err & (BBD_ERR | ECC_ERR)) {
462                         /* retries won't help these */
463                         rq->errors = ERROR_MAX;
464                 } else if (err & TRK0_ERR) {
465                         /* help it find track zero */
466                         rq->errors |= ERROR_RECAL;
467                 }
468         }
469
470         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ &&
471             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0)
472                 try_to_flush_leftover_data(drive);
473
474         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
475                 ide_kill_rq(drive, rq);
476                 return ide_stopped;
477         }
478
479         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
480                 rq->errors |= ERROR_RESET;
481
482         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
483                 ++rq->errors;
484                 return ide_do_reset(drive);
485         }
486
487         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
488                 drive->special.b.recalibrate = 1;
489
490         ++rq->errors;
491
492         return ide_stopped;
493 }
494
495 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
496 {
497         ide_hwif_t *hwif = drive->hwif;
498
499         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
500                 /* other bits are useless when BUSY */
501                 rq->errors |= ERROR_RESET;
502         } else {
503                 /* add decoding error stuff */
504         }
505
506         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
507                 /* force an abort */
508                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
509
510         if (rq->errors >= ERROR_MAX) {
511                 ide_kill_rq(drive, rq);
512         } else {
513                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
514                         ++rq->errors;
515                         return ide_do_reset(drive);
516                 }
517                 ++rq->errors;
518         }
519
520         return ide_stopped;
521 }
522
523 ide_startstop_t
524 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
525 {
526         if (drive->media == ide_disk)
527                 return ide_ata_error(drive, rq, stat, err);
528         return ide_atapi_error(drive, rq, stat, err);
529 }
530
531 EXPORT_SYMBOL_GPL(__ide_error);
532
533 /**
534  *      ide_error       -       handle an error on the IDE
535  *      @drive: drive the error occurred on
536  *      @msg: message to report
537  *      @stat: status bits
538  *
539  *      ide_error() takes action based on the error returned by the drive.
540  *      For normal I/O that may well include retries. We deal with
541  *      both new-style (taskfile) and old style command handling here.
542  *      In the case of taskfile command handling there is work left to
543  *      do
544  */
545  
546 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
547 {
548         struct request *rq;
549         u8 err;
550
551         err = ide_dump_status(drive, msg, stat);
552
553         if ((rq = HWGROUP(drive)->rq) == NULL)
554                 return ide_stopped;
555
556         /* retry only "normal" I/O: */
557         if (!blk_fs_request(rq)) {
558                 rq->errors = 1;
559                 ide_end_drive_cmd(drive, stat, err);
560                 return ide_stopped;
561         }
562
563         if (rq->rq_disk) {
564                 ide_driver_t *drv;
565
566                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
567                 return drv->error(drive, rq, stat, err);
568         } else
569                 return __ide_error(drive, rq, stat, err);
570 }
571
572 EXPORT_SYMBOL_GPL(ide_error);
573
574 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
575 {
576         if (drive->media != ide_disk)
577                 rq->errors |= ERROR_RESET;
578
579         ide_kill_rq(drive, rq);
580
581         return ide_stopped;
582 }
583
584 EXPORT_SYMBOL_GPL(__ide_abort);
585
586 /**
587  *      ide_abort       -       abort pending IDE operations
588  *      @drive: drive the error occurred on
589  *      @msg: message to report
590  *
591  *      ide_abort kills and cleans up when we are about to do a 
592  *      host initiated reset on active commands. Longer term we
593  *      want handlers to have sensible abort handling themselves
594  *
595  *      This differs fundamentally from ide_error because in 
596  *      this case the command is doing just fine when we
597  *      blow it away.
598  */
599  
600 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
601 {
602         struct request *rq;
603
604         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
605                 return ide_stopped;
606
607         /* retry only "normal" I/O: */
608         if (!blk_fs_request(rq)) {
609                 rq->errors = 1;
610                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
611                 return ide_stopped;
612         }
613
614         if (rq->rq_disk) {
615                 ide_driver_t *drv;
616
617                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
618                 return drv->abort(drive, rq);
619         } else
620                 return __ide_abort(drive, rq);
621 }
622
623 /**
624  *      ide_cmd         -       issue a simple drive command
625  *      @drive: drive the command is for
626  *      @cmd: command byte
627  *      @handler: handler for the command completion
628  *
629  *      Issue a simple drive command with interrupts.
630  *      The drive must be selected beforehand.
631  */
632
633 static void ide_cmd(ide_drive_t *drive, u8 cmd, ide_handler_t *handler)
634 {
635         ide_hwif_t *hwif = HWIF(drive);
636         if (IDE_CONTROL_REG)
637                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
638         SELECT_MASK(drive,0);
639         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
640 }
641
642 /**
643  *      drive_cmd_intr          -       drive command completion interrupt
644  *      @drive: drive the completion interrupt occurred on
645  *
646  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
647  *      We do any necessary data reading and then wait for the drive to
648  *      go non busy. At that point we may read the error data and complete
649  *      the request
650  */
651  
652 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
653 {
654         struct request *rq = HWGROUP(drive)->rq;
655         ide_hwif_t *hwif = HWIF(drive);
656         u8 *args = (u8 *) rq->buffer;
657         u8 stat = hwif->INB(IDE_STATUS_REG);
658         int retries = 10;
659
660         local_irq_enable_in_hardirq();
661         if (rq->cmd_type == REQ_TYPE_ATA_CMD &&
662             (stat & DRQ_STAT) && args && args[3]) {
663                 u8 io_32bit = drive->io_32bit;
664                 drive->io_32bit = 0;
665                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
666                 drive->io_32bit = io_32bit;
667                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
668                         udelay(100);
669         }
670
671         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
672                 return ide_error(drive, "drive_cmd", stat);
673                 /* calls ide_end_drive_cmd */
674         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
675         return ide_stopped;
676 }
677
678 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
679 {
680         task->tf.nsect   = drive->sect;
681         task->tf.lbal    = drive->sect;
682         task->tf.lbam    = drive->cyl;
683         task->tf.lbah    = drive->cyl >> 8;
684         task->tf.device  = ((drive->head - 1) | drive->select.all) & ~ATA_LBA;
685         task->tf.command = WIN_SPECIFY;
686
687         task->handler = &set_geometry_intr;
688 }
689
690 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
691 {
692         task->tf.nsect   = drive->sect;
693         task->tf.command = WIN_RESTORE;
694
695         task->handler = &recal_intr;
696 }
697
698 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
699 {
700         task->tf.nsect   = drive->mult_req;
701         task->tf.command = WIN_SETMULT;
702
703         task->handler = &set_multmode_intr;
704 }
705
706 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
707 {
708         special_t *s = &drive->special;
709         ide_task_t args;
710
711         memset(&args, 0, sizeof(ide_task_t));
712         args.command_type = IDE_DRIVE_TASK_NO_DATA;
713
714         if (s->b.set_geometry) {
715                 s->b.set_geometry = 0;
716                 ide_init_specify_cmd(drive, &args);
717         } else if (s->b.recalibrate) {
718                 s->b.recalibrate = 0;
719                 ide_init_restore_cmd(drive, &args);
720         } else if (s->b.set_multmode) {
721                 s->b.set_multmode = 0;
722                 if (drive->mult_req > drive->id->max_multsect)
723                         drive->mult_req = drive->id->max_multsect;
724                 ide_init_setmult_cmd(drive, &args);
725         } else if (s->all) {
726                 int special = s->all;
727                 s->all = 0;
728                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
729                 return ide_stopped;
730         }
731
732         args.tf_flags = IDE_TFLAG_OUT_TF;
733         if (drive->addressing == 1)
734                 args.tf_flags |= (IDE_TFLAG_LBA48 | IDE_TFLAG_OUT_HOB);
735
736         do_rw_taskfile(drive, &args);
737
738         return ide_started;
739 }
740
741 /*
742  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
743  */
744 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
745 {
746         switch (req_pio) {
747         case 202:
748         case 201:
749         case 200:
750         case 102:
751         case 101:
752         case 100:
753                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
754         case 9:
755         case 8:
756                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
757         case 7:
758         case 6:
759                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
760         default:
761                 return 0;
762         }
763 }
764
765 /**
766  *      do_special              -       issue some special commands
767  *      @drive: drive the command is for
768  *
769  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
770  *      commands to a drive.  It used to do much more, but has been scaled
771  *      back.
772  */
773
774 static ide_startstop_t do_special (ide_drive_t *drive)
775 {
776         special_t *s = &drive->special;
777
778 #ifdef DEBUG
779         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
780 #endif
781         if (s->b.set_tune) {
782                 ide_hwif_t *hwif = drive->hwif;
783                 u8 req_pio = drive->tune_req;
784
785                 s->b.set_tune = 0;
786
787                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
788
789                         if (hwif->set_pio_mode == NULL)
790                                 return ide_stopped;
791
792                         /*
793                          * take ide_lock for drive->[no_]unmask/[no_]io_32bit
794                          */
795                         if (req_pio == 8 || req_pio == 9) {
796                                 unsigned long flags;
797
798                                 spin_lock_irqsave(&ide_lock, flags);
799                                 hwif->set_pio_mode(drive, req_pio);
800                                 spin_unlock_irqrestore(&ide_lock, flags);
801                         } else
802                                 hwif->set_pio_mode(drive, req_pio);
803                 } else {
804                         int keep_dma = drive->using_dma;
805
806                         ide_set_pio(drive, req_pio);
807
808                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
809                                 if (keep_dma)
810                                         hwif->ide_dma_on(drive);
811                         }
812                 }
813
814                 return ide_stopped;
815         } else {
816                 if (drive->media == ide_disk)
817                         return ide_disk_special(drive);
818
819                 s->all = 0;
820                 drive->mult_req = 0;
821                 return ide_stopped;
822         }
823 }
824
825 void ide_map_sg(ide_drive_t *drive, struct request *rq)
826 {
827         ide_hwif_t *hwif = drive->hwif;
828         struct scatterlist *sg = hwif->sg_table;
829
830         if (hwif->sg_mapped)    /* needed by ide-scsi */
831                 return;
832
833         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
834                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
835         } else {
836                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
837                 hwif->sg_nents = 1;
838         }
839 }
840
841 EXPORT_SYMBOL_GPL(ide_map_sg);
842
843 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
844 {
845         ide_hwif_t *hwif = drive->hwif;
846
847         hwif->nsect = hwif->nleft = rq->nr_sectors;
848         hwif->cursg_ofs = 0;
849         hwif->cursg = NULL;
850 }
851
852 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
853
854 /**
855  *      execute_drive_command   -       issue special drive command
856  *      @drive: the drive to issue the command on
857  *      @rq: the request structure holding the command
858  *
859  *      execute_drive_cmd() issues a special drive command,  usually 
860  *      initiated by ioctl() from the external hdparm program. The
861  *      command can be a drive command, drive task or taskfile 
862  *      operation. Weirdly you can call it with NULL to wait for
863  *      all commands to finish. Don't do this as that is due to change
864  */
865
866 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
867                 struct request *rq)
868 {
869         ide_hwif_t *hwif = HWIF(drive);
870         u8 *args = rq->buffer;
871
872         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
873                 ide_task_t *task = rq->special;
874  
875                 if (task == NULL)
876                         goto done;
877
878                 hwif->data_phase = task->data_phase;
879
880                 switch (hwif->data_phase) {
881                 case TASKFILE_MULTI_OUT:
882                 case TASKFILE_OUT:
883                 case TASKFILE_MULTI_IN:
884                 case TASKFILE_IN:
885                         ide_init_sg_cmd(drive, rq);
886                         ide_map_sg(drive, rq);
887                 default:
888                         break;
889                 }
890
891                 if (task->tf_flags & IDE_TFLAG_FLAGGED)
892                         return flagged_taskfile(drive, task);
893
894                 task->tf_flags |= IDE_TFLAG_OUT_TF;
895                 if (drive->addressing == 1)
896                         task->tf_flags |= (IDE_TFLAG_LBA48 | IDE_TFLAG_OUT_HOB);
897
898                 return do_rw_taskfile(drive, task);
899         }
900
901         if (args == NULL)
902                 goto done;
903
904         if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
905 #ifdef DEBUG
906                 printk("%s: DRIVE_TASK_CMD ", drive->name);
907                 printk("cmd=0x%02x ", args[0]);
908                 printk("fr=0x%02x ", args[1]);
909                 printk("ns=0x%02x ", args[2]);
910                 printk("sc=0x%02x ", args[3]);
911                 printk("lcyl=0x%02x ", args[4]);
912                 printk("hcyl=0x%02x ", args[5]);
913                 printk("sel=0x%02x\n", args[6]);
914 #endif
915                 hwif->OUTB(args[1], IDE_FEATURE_REG);
916                 hwif->OUTB(args[2], IDE_NSECTOR_REG);
917                 hwif->OUTB(args[3], IDE_SECTOR_REG);
918                 hwif->OUTB(args[4], IDE_LCYL_REG);
919                 hwif->OUTB(args[5], IDE_HCYL_REG);
920                 hwif->OUTB((args[6] & 0xEF)|drive->select.all, IDE_SELECT_REG);
921         } else { /* rq->cmd_type == REQ_TYPE_ATA_CMD */
922 #ifdef DEBUG
923                 printk("%s: DRIVE_CMD ", drive->name);
924                 printk("cmd=0x%02x ", args[0]);
925                 printk("sc=0x%02x ", args[1]);
926                 printk("fr=0x%02x ", args[2]);
927                 printk("xx=0x%02x\n", args[3]);
928 #endif
929                 hwif->OUTB(args[2], IDE_FEATURE_REG);
930                 if (args[0] == WIN_SMART) {
931                         hwif->OUTB(args[3],IDE_NSECTOR_REG);
932                         hwif->OUTB(args[1],IDE_SECTOR_REG);
933                         hwif->OUTB(0x4f, IDE_LCYL_REG);
934                         hwif->OUTB(0xc2, IDE_HCYL_REG);
935                 } else
936                         hwif->OUTB(args[1], IDE_NSECTOR_REG);
937         }
938
939         ide_cmd(drive, args[0], &drive_cmd_intr);
940         return ide_started;
941
942 done:
943         /*
944          * NULL is actually a valid way of waiting for
945          * all current requests to be flushed from the queue.
946          */
947 #ifdef DEBUG
948         printk("%s: DRIVE_CMD (null)\n", drive->name);
949 #endif
950         ide_end_drive_cmd(drive,
951                         hwif->INB(IDE_STATUS_REG),
952                         hwif->INB(IDE_ERROR_REG));
953         return ide_stopped;
954 }
955
956 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
957 {
958         struct request_pm_state *pm = rq->data;
959
960         if (blk_pm_suspend_request(rq) &&
961             pm->pm_step == ide_pm_state_start_suspend)
962                 /* Mark drive blocked when starting the suspend sequence. */
963                 drive->blocked = 1;
964         else if (blk_pm_resume_request(rq) &&
965                  pm->pm_step == ide_pm_state_start_resume) {
966                 /* 
967                  * The first thing we do on wakeup is to wait for BSY bit to
968                  * go away (with a looong timeout) as a drive on this hwif may
969                  * just be POSTing itself.
970                  * We do that before even selecting as the "other" device on
971                  * the bus may be broken enough to walk on our toes at this
972                  * point.
973                  */
974                 int rc;
975 #ifdef DEBUG_PM
976                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
977 #endif
978                 rc = ide_wait_not_busy(HWIF(drive), 35000);
979                 if (rc)
980                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
981                 SELECT_DRIVE(drive);
982                 if (IDE_CONTROL_REG)
983                         HWIF(drive)->OUTB(drive->ctl, IDE_CONTROL_REG);
984                 rc = ide_wait_not_busy(HWIF(drive), 100000);
985                 if (rc)
986                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
987         }
988 }
989
990 /**
991  *      start_request   -       start of I/O and command issuing for IDE
992  *
993  *      start_request() initiates handling of a new I/O request. It
994  *      accepts commands and I/O (read/write) requests. It also does
995  *      the final remapping for weird stuff like EZDrive. Once 
996  *      device mapper can work sector level the EZDrive stuff can go away
997  *
998  *      FIXME: this function needs a rename
999  */
1000  
1001 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
1002 {
1003         ide_startstop_t startstop;
1004         sector_t block;
1005
1006         BUG_ON(!blk_rq_started(rq));
1007
1008 #ifdef DEBUG
1009         printk("%s: start_request: current=0x%08lx\n",
1010                 HWIF(drive)->name, (unsigned long) rq);
1011 #endif
1012
1013         /* bail early if we've exceeded max_failures */
1014         if (drive->max_failures && (drive->failures > drive->max_failures)) {
1015                 rq->cmd_flags |= REQ_FAILED;
1016                 goto kill_rq;
1017         }
1018
1019         block    = rq->sector;
1020         if (blk_fs_request(rq) &&
1021             (drive->media == ide_disk || drive->media == ide_floppy)) {
1022                 block += drive->sect0;
1023         }
1024         /* Yecch - this will shift the entire interval,
1025            possibly killing some innocent following sector */
1026         if (block == 0 && drive->remap_0_to_1 == 1)
1027                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
1028
1029         if (blk_pm_request(rq))
1030                 ide_check_pm_state(drive, rq);
1031
1032         SELECT_DRIVE(drive);
1033         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1034                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1035                 return startstop;
1036         }
1037         if (!drive->special.all) {
1038                 ide_driver_t *drv;
1039
1040                 /*
1041                  * We reset the drive so we need to issue a SETFEATURES.
1042                  * Do it _after_ do_special() restored device parameters.
1043                  */
1044                 if (drive->current_speed == 0xff)
1045                         ide_config_drive_speed(drive, drive->desired_speed);
1046
1047                 if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1048                     rq->cmd_type == REQ_TYPE_ATA_TASK ||
1049                     rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1050                         return execute_drive_cmd(drive, rq);
1051                 else if (blk_pm_request(rq)) {
1052                         struct request_pm_state *pm = rq->data;
1053 #ifdef DEBUG_PM
1054                         printk("%s: start_power_step(step: %d)\n",
1055                                 drive->name, rq->pm->pm_step);
1056 #endif
1057                         startstop = ide_start_power_step(drive, rq);
1058                         if (startstop == ide_stopped &&
1059                             pm->pm_step == ide_pm_state_completed)
1060                                 ide_complete_pm_request(drive, rq);
1061                         return startstop;
1062                 }
1063
1064                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1065                 return drv->do_request(drive, rq, block);
1066         }
1067         return do_special(drive);
1068 kill_rq:
1069         ide_kill_rq(drive, rq);
1070         return ide_stopped;
1071 }
1072
1073 /**
1074  *      ide_stall_queue         -       pause an IDE device
1075  *      @drive: drive to stall
1076  *      @timeout: time to stall for (jiffies)
1077  *
1078  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1079  *      to the hwgroup by sleeping for timeout jiffies.
1080  */
1081  
1082 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1083 {
1084         if (timeout > WAIT_WORSTCASE)
1085                 timeout = WAIT_WORSTCASE;
1086         drive->sleep = timeout + jiffies;
1087         drive->sleeping = 1;
1088 }
1089
1090 EXPORT_SYMBOL(ide_stall_queue);
1091
1092 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1093
1094 /**
1095  *      choose_drive            -       select a drive to service
1096  *      @hwgroup: hardware group to select on
1097  *
1098  *      choose_drive() selects the next drive which will be serviced.
1099  *      This is necessary because the IDE layer can't issue commands
1100  *      to both drives on the same cable, unlike SCSI.
1101  */
1102  
1103 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1104 {
1105         ide_drive_t *drive, *best;
1106
1107 repeat: 
1108         best = NULL;
1109         drive = hwgroup->drive;
1110
1111         /*
1112          * drive is doing pre-flush, ordered write, post-flush sequence. even
1113          * though that is 3 requests, it must be seen as a single transaction.
1114          * we must not preempt this drive until that is complete
1115          */
1116         if (blk_queue_flushing(drive->queue)) {
1117                 /*
1118                  * small race where queue could get replugged during
1119                  * the 3-request flush cycle, just yank the plug since
1120                  * we want it to finish asap
1121                  */
1122                 blk_remove_plug(drive->queue);
1123                 return drive;
1124         }
1125
1126         do {
1127                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1128                     && !elv_queue_empty(drive->queue)) {
1129                         if (!best
1130                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1131                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1132                         {
1133                                 if (!blk_queue_plugged(drive->queue))
1134                                         best = drive;
1135                         }
1136                 }
1137         } while ((drive = drive->next) != hwgroup->drive);
1138         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1139                 long t = (signed long)(WAKEUP(best) - jiffies);
1140                 if (t >= WAIT_MIN_SLEEP) {
1141                 /*
1142                  * We *may* have some time to spare, but first let's see if
1143                  * someone can potentially benefit from our nice mood today..
1144                  */
1145                         drive = best->next;
1146                         do {
1147                                 if (!drive->sleeping
1148                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1149                                  && time_before(WAKEUP(drive), jiffies + t))
1150                                 {
1151                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1152                                         goto repeat;
1153                                 }
1154                         } while ((drive = drive->next) != best);
1155                 }
1156         }
1157         return best;
1158 }
1159
1160 /*
1161  * Issue a new request to a drive from hwgroup
1162  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1163  *
1164  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1165  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1166  * may have both interfaces in a single hwgroup to "serialize" access.
1167  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1168  * together into one hwgroup for serialized access.
1169  *
1170  * Note also that several hwgroups can end up sharing a single IRQ,
1171  * possibly along with many other devices.  This is especially common in
1172  * PCI-based systems with off-board IDE controller cards.
1173  *
1174  * The IDE driver uses the single global ide_lock spinlock to protect
1175  * access to the request queues, and to protect the hwgroup->busy flag.
1176  *
1177  * The first thread into the driver for a particular hwgroup sets the
1178  * hwgroup->busy flag to indicate that this hwgroup is now active,
1179  * and then initiates processing of the top request from the request queue.
1180  *
1181  * Other threads attempting entry notice the busy setting, and will simply
1182  * queue their new requests and exit immediately.  Note that hwgroup->busy
1183  * remains set even when the driver is merely awaiting the next interrupt.
1184  * Thus, the meaning is "this hwgroup is busy processing a request".
1185  *
1186  * When processing of a request completes, the completing thread or IRQ-handler
1187  * will start the next request from the queue.  If no more work remains,
1188  * the driver will clear the hwgroup->busy flag and exit.
1189  *
1190  * The ide_lock (spinlock) is used to protect all access to the
1191  * hwgroup->busy flag, but is otherwise not needed for most processing in
1192  * the driver.  This makes the driver much more friendlier to shared IRQs
1193  * than previous designs, while remaining 100% (?) SMP safe and capable.
1194  */
1195 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1196 {
1197         ide_drive_t     *drive;
1198         ide_hwif_t      *hwif;
1199         struct request  *rq;
1200         ide_startstop_t startstop;
1201         int             loops = 0;
1202
1203         /* for atari only: POSSIBLY BROKEN HERE(?) */
1204         ide_get_lock(ide_intr, hwgroup);
1205
1206         /* caller must own ide_lock */
1207         BUG_ON(!irqs_disabled());
1208
1209         while (!hwgroup->busy) {
1210                 hwgroup->busy = 1;
1211                 drive = choose_drive(hwgroup);
1212                 if (drive == NULL) {
1213                         int sleeping = 0;
1214                         unsigned long sleep = 0; /* shut up, gcc */
1215                         hwgroup->rq = NULL;
1216                         drive = hwgroup->drive;
1217                         do {
1218                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1219                                         sleeping = 1;
1220                                         sleep = drive->sleep;
1221                                 }
1222                         } while ((drive = drive->next) != hwgroup->drive);
1223                         if (sleeping) {
1224                 /*
1225                  * Take a short snooze, and then wake up this hwgroup again.
1226                  * This gives other hwgroups on the same a chance to
1227                  * play fairly with us, just in case there are big differences
1228                  * in relative throughputs.. don't want to hog the cpu too much.
1229                  */
1230                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1231                                         sleep = jiffies + WAIT_MIN_SLEEP;
1232 #if 1
1233                                 if (timer_pending(&hwgroup->timer))
1234                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1235 #endif
1236                                 /* so that ide_timer_expiry knows what to do */
1237                                 hwgroup->sleeping = 1;
1238                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1239                                 mod_timer(&hwgroup->timer, sleep);
1240                                 /* we purposely leave hwgroup->busy==1
1241                                  * while sleeping */
1242                         } else {
1243                                 /* Ugly, but how can we sleep for the lock
1244                                  * otherwise? perhaps from tq_disk?
1245                                  */
1246
1247                                 /* for atari only */
1248                                 ide_release_lock();
1249                                 hwgroup->busy = 0;
1250                         }
1251
1252                         /* no more work for this hwgroup (for now) */
1253                         return;
1254                 }
1255         again:
1256                 hwif = HWIF(drive);
1257                 if (hwgroup->hwif->sharing_irq &&
1258                     hwif != hwgroup->hwif &&
1259                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1260                         /* set nIEN for previous hwif */
1261                         SELECT_INTERRUPT(drive);
1262                 }
1263                 hwgroup->hwif = hwif;
1264                 hwgroup->drive = drive;
1265                 drive->sleeping = 0;
1266                 drive->service_start = jiffies;
1267
1268                 if (blk_queue_plugged(drive->queue)) {
1269                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1270                         break;
1271                 }
1272
1273                 /*
1274                  * we know that the queue isn't empty, but this can happen
1275                  * if the q->prep_rq_fn() decides to kill a request
1276                  */
1277                 rq = elv_next_request(drive->queue);
1278                 if (!rq) {
1279                         hwgroup->busy = 0;
1280                         break;
1281                 }
1282
1283                 /*
1284                  * Sanity: don't accept a request that isn't a PM request
1285                  * if we are currently power managed. This is very important as
1286                  * blk_stop_queue() doesn't prevent the elv_next_request()
1287                  * above to return us whatever is in the queue. Since we call
1288                  * ide_do_request() ourselves, we end up taking requests while
1289                  * the queue is blocked...
1290                  * 
1291                  * We let requests forced at head of queue with ide-preempt
1292                  * though. I hope that doesn't happen too much, hopefully not
1293                  * unless the subdriver triggers such a thing in its own PM
1294                  * state machine.
1295                  *
1296                  * We count how many times we loop here to make sure we service
1297                  * all drives in the hwgroup without looping for ever
1298                  */
1299                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1300                         drive = drive->next ? drive->next : hwgroup->drive;
1301                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1302                                 goto again;
1303                         /* We clear busy, there should be no pending ATA command at this point. */
1304                         hwgroup->busy = 0;
1305                         break;
1306                 }
1307
1308                 hwgroup->rq = rq;
1309
1310                 /*
1311                  * Some systems have trouble with IDE IRQs arriving while
1312                  * the driver is still setting things up.  So, here we disable
1313                  * the IRQ used by this interface while the request is being started.
1314                  * This may look bad at first, but pretty much the same thing
1315                  * happens anyway when any interrupt comes in, IDE or otherwise
1316                  *  -- the kernel masks the IRQ while it is being handled.
1317                  */
1318                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1319                         disable_irq_nosync(hwif->irq);
1320                 spin_unlock(&ide_lock);
1321                 local_irq_enable_in_hardirq();
1322                         /* allow other IRQs while we start this request */
1323                 startstop = start_request(drive, rq);
1324                 spin_lock_irq(&ide_lock);
1325                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1326                         enable_irq(hwif->irq);
1327                 if (startstop == ide_stopped)
1328                         hwgroup->busy = 0;
1329         }
1330 }
1331
1332 /*
1333  * Passes the stuff to ide_do_request
1334  */
1335 void do_ide_request(struct request_queue *q)
1336 {
1337         ide_drive_t *drive = q->queuedata;
1338
1339         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1340 }
1341
1342 /*
1343  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1344  * retry the current request in pio mode instead of risking tossing it
1345  * all away
1346  */
1347 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1348 {
1349         ide_hwif_t *hwif = HWIF(drive);
1350         struct request *rq;
1351         ide_startstop_t ret = ide_stopped;
1352
1353         /*
1354          * end current dma transaction
1355          */
1356
1357         if (error < 0) {
1358                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1359                 (void)HWIF(drive)->ide_dma_end(drive);
1360                 ret = ide_error(drive, "dma timeout error",
1361                                                 hwif->INB(IDE_STATUS_REG));
1362         } else {
1363                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1364                 hwif->dma_timeout(drive);
1365         }
1366
1367         /*
1368          * disable dma for now, but remember that we did so because of
1369          * a timeout -- we'll reenable after we finish this next request
1370          * (or rather the first chunk of it) in pio.
1371          */
1372         drive->retry_pio++;
1373         drive->state = DMA_PIO_RETRY;
1374         hwif->dma_off_quietly(drive);
1375
1376         /*
1377          * un-busy drive etc (hwgroup->busy is cleared on return) and
1378          * make sure request is sane
1379          */
1380         rq = HWGROUP(drive)->rq;
1381
1382         if (!rq)
1383                 goto out;
1384
1385         HWGROUP(drive)->rq = NULL;
1386
1387         rq->errors = 0;
1388
1389         if (!rq->bio)
1390                 goto out;
1391
1392         rq->sector = rq->bio->bi_sector;
1393         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1394         rq->hard_cur_sectors = rq->current_nr_sectors;
1395         rq->buffer = bio_data(rq->bio);
1396 out:
1397         return ret;
1398 }
1399
1400 /**
1401  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1402  *      @data: timer callback magic (hwgroup)
1403  *
1404  *      An IDE command has timed out before the expected drive return
1405  *      occurred. At this point we attempt to clean up the current
1406  *      mess. If the current handler includes an expiry handler then
1407  *      we invoke the expiry handler, and providing it is happy the
1408  *      work is done. If that fails we apply generic recovery rules
1409  *      invoking the handler and checking the drive DMA status. We
1410  *      have an excessively incestuous relationship with the DMA
1411  *      logic that wants cleaning up.
1412  */
1413  
1414 void ide_timer_expiry (unsigned long data)
1415 {
1416         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1417         ide_handler_t   *handler;
1418         ide_expiry_t    *expiry;
1419         unsigned long   flags;
1420         unsigned long   wait = -1;
1421
1422         spin_lock_irqsave(&ide_lock, flags);
1423
1424         if (((handler = hwgroup->handler) == NULL) ||
1425             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1426                 /*
1427                  * Either a marginal timeout occurred
1428                  * (got the interrupt just as timer expired),
1429                  * or we were "sleeping" to give other devices a chance.
1430                  * Either way, we don't really want to complain about anything.
1431                  */
1432                 if (hwgroup->sleeping) {
1433                         hwgroup->sleeping = 0;
1434                         hwgroup->busy = 0;
1435                 }
1436         } else {
1437                 ide_drive_t *drive = hwgroup->drive;
1438                 if (!drive) {
1439                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1440                         hwgroup->handler = NULL;
1441                 } else {
1442                         ide_hwif_t *hwif;
1443                         ide_startstop_t startstop = ide_stopped;
1444                         if (!hwgroup->busy) {
1445                                 hwgroup->busy = 1;      /* paranoia */
1446                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1447                         }
1448                         if ((expiry = hwgroup->expiry) != NULL) {
1449                                 /* continue */
1450                                 if ((wait = expiry(drive)) > 0) {
1451                                         /* reset timer */
1452                                         hwgroup->timer.expires  = jiffies + wait;
1453                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1454                                         add_timer(&hwgroup->timer);
1455                                         spin_unlock_irqrestore(&ide_lock, flags);
1456                                         return;
1457                                 }
1458                         }
1459                         hwgroup->handler = NULL;
1460                         /*
1461                          * We need to simulate a real interrupt when invoking
1462                          * the handler() function, which means we need to
1463                          * globally mask the specific IRQ:
1464                          */
1465                         spin_unlock(&ide_lock);
1466                         hwif  = HWIF(drive);
1467                         /* disable_irq_nosync ?? */
1468                         disable_irq(hwif->irq);
1469                         /* local CPU only,
1470                          * as if we were handling an interrupt */
1471                         local_irq_disable();
1472                         if (hwgroup->polling) {
1473                                 startstop = handler(drive);
1474                         } else if (drive_is_ready(drive)) {
1475                                 if (drive->waiting_for_dma)
1476                                         hwgroup->hwif->dma_lost_irq(drive);
1477                                 (void)ide_ack_intr(hwif);
1478                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1479                                 startstop = handler(drive);
1480                         } else {
1481                                 if (drive->waiting_for_dma) {
1482                                         startstop = ide_dma_timeout_retry(drive, wait);
1483                                 } else
1484                                         startstop =
1485                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1486                         }
1487                         drive->service_time = jiffies - drive->service_start;
1488                         spin_lock_irq(&ide_lock);
1489                         enable_irq(hwif->irq);
1490                         if (startstop == ide_stopped)
1491                                 hwgroup->busy = 0;
1492                 }
1493         }
1494         ide_do_request(hwgroup, IDE_NO_IRQ);
1495         spin_unlock_irqrestore(&ide_lock, flags);
1496 }
1497
1498 /**
1499  *      unexpected_intr         -       handle an unexpected IDE interrupt
1500  *      @irq: interrupt line
1501  *      @hwgroup: hwgroup being processed
1502  *
1503  *      There's nothing really useful we can do with an unexpected interrupt,
1504  *      other than reading the status register (to clear it), and logging it.
1505  *      There should be no way that an irq can happen before we're ready for it,
1506  *      so we needn't worry much about losing an "important" interrupt here.
1507  *
1508  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1509  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1510  *      looks "good", we just ignore the interrupt completely.
1511  *
1512  *      This routine assumes __cli() is in effect when called.
1513  *
1514  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1515  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1516  *      we could screw up by interfering with a new request being set up for 
1517  *      irq15.
1518  *
1519  *      In reality, this is a non-issue.  The new command is not sent unless 
1520  *      the drive is ready to accept one, in which case we know the drive is
1521  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1522  *      before completing the issuance of any new drive command, so we will not
1523  *      be accidentally invoked as a result of any valid command completion
1524  *      interrupt.
1525  *
1526  *      Note that we must walk the entire hwgroup here. We know which hwif
1527  *      is doing the current command, but we don't know which hwif burped
1528  *      mysteriously.
1529  */
1530  
1531 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1532 {
1533         u8 stat;
1534         ide_hwif_t *hwif = hwgroup->hwif;
1535
1536         /*
1537          * handle the unexpected interrupt
1538          */
1539         do {
1540                 if (hwif->irq == irq) {
1541                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1542                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1543                                 /* Try to not flood the console with msgs */
1544                                 static unsigned long last_msgtime, count;
1545                                 ++count;
1546                                 if (time_after(jiffies, last_msgtime + HZ)) {
1547                                         last_msgtime = jiffies;
1548                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1549                                                 "status=0x%02x, count=%ld\n",
1550                                                 hwif->name,
1551                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1552                                 }
1553                         }
1554                 }
1555         } while ((hwif = hwif->next) != hwgroup->hwif);
1556 }
1557
1558 /**
1559  *      ide_intr        -       default IDE interrupt handler
1560  *      @irq: interrupt number
1561  *      @dev_id: hwif group
1562  *      @regs: unused weirdness from the kernel irq layer
1563  *
1564  *      This is the default IRQ handler for the IDE layer. You should
1565  *      not need to override it. If you do be aware it is subtle in
1566  *      places
1567  *
1568  *      hwgroup->hwif is the interface in the group currently performing
1569  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1570  *      the IRQ handler to call. As we issue a command the handlers
1571  *      step through multiple states, reassigning the handler to the
1572  *      next step in the process. Unlike a smart SCSI controller IDE
1573  *      expects the main processor to sequence the various transfer
1574  *      stages. We also manage a poll timer to catch up with most
1575  *      timeout situations. There are still a few where the handlers
1576  *      don't ever decide to give up.
1577  *
1578  *      The handler eventually returns ide_stopped to indicate the
1579  *      request completed. At this point we issue the next request
1580  *      on the hwgroup and the process begins again.
1581  */
1582  
1583 irqreturn_t ide_intr (int irq, void *dev_id)
1584 {
1585         unsigned long flags;
1586         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1587         ide_hwif_t *hwif;
1588         ide_drive_t *drive;
1589         ide_handler_t *handler;
1590         ide_startstop_t startstop;
1591
1592         spin_lock_irqsave(&ide_lock, flags);
1593         hwif = hwgroup->hwif;
1594
1595         if (!ide_ack_intr(hwif)) {
1596                 spin_unlock_irqrestore(&ide_lock, flags);
1597                 return IRQ_NONE;
1598         }
1599
1600         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1601                 /*
1602                  * Not expecting an interrupt from this drive.
1603                  * That means this could be:
1604                  *      (1) an interrupt from another PCI device
1605                  *      sharing the same PCI INT# as us.
1606                  * or   (2) a drive just entered sleep or standby mode,
1607                  *      and is interrupting to let us know.
1608                  * or   (3) a spurious interrupt of unknown origin.
1609                  *
1610                  * For PCI, we cannot tell the difference,
1611                  * so in that case we just ignore it and hope it goes away.
1612                  *
1613                  * FIXME: unexpected_intr should be hwif-> then we can
1614                  * remove all the ifdef PCI crap
1615                  */
1616 #ifdef CONFIG_BLK_DEV_IDEPCI
1617                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1618 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1619                 {
1620                         /*
1621                          * Probably not a shared PCI interrupt,
1622                          * so we can safely try to do something about it:
1623                          */
1624                         unexpected_intr(irq, hwgroup);
1625 #ifdef CONFIG_BLK_DEV_IDEPCI
1626                 } else {
1627                         /*
1628                          * Whack the status register, just in case
1629                          * we have a leftover pending IRQ.
1630                          */
1631                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1632 #endif /* CONFIG_BLK_DEV_IDEPCI */
1633                 }
1634                 spin_unlock_irqrestore(&ide_lock, flags);
1635                 return IRQ_NONE;
1636         }
1637         drive = hwgroup->drive;
1638         if (!drive) {
1639                 /*
1640                  * This should NEVER happen, and there isn't much
1641                  * we could do about it here.
1642                  *
1643                  * [Note - this can occur if the drive is hot unplugged]
1644                  */
1645                 spin_unlock_irqrestore(&ide_lock, flags);
1646                 return IRQ_HANDLED;
1647         }
1648         if (!drive_is_ready(drive)) {
1649                 /*
1650                  * This happens regularly when we share a PCI IRQ with
1651                  * another device.  Unfortunately, it can also happen
1652                  * with some buggy drives that trigger the IRQ before
1653                  * their status register is up to date.  Hopefully we have
1654                  * enough advance overhead that the latter isn't a problem.
1655                  */
1656                 spin_unlock_irqrestore(&ide_lock, flags);
1657                 return IRQ_NONE;
1658         }
1659         if (!hwgroup->busy) {
1660                 hwgroup->busy = 1;      /* paranoia */
1661                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1662         }
1663         hwgroup->handler = NULL;
1664         hwgroup->req_gen++;
1665         del_timer(&hwgroup->timer);
1666         spin_unlock(&ide_lock);
1667
1668         /* Some controllers might set DMA INTR no matter DMA or PIO;
1669          * bmdma status might need to be cleared even for
1670          * PIO interrupts to prevent spurious/lost irq.
1671          */
1672         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1673                 /* ide_dma_end() needs bmdma status for error checking.
1674                  * So, skip clearing bmdma status here and leave it
1675                  * to ide_dma_end() if this is dma interrupt.
1676                  */
1677                 hwif->ide_dma_clear_irq(drive);
1678
1679         if (drive->unmask)
1680                 local_irq_enable_in_hardirq();
1681         /* service this interrupt, may set handler for next interrupt */
1682         startstop = handler(drive);
1683         spin_lock_irq(&ide_lock);
1684
1685         /*
1686          * Note that handler() may have set things up for another
1687          * interrupt to occur soon, but it cannot happen until
1688          * we exit from this routine, because it will be the
1689          * same irq as is currently being serviced here, and Linux
1690          * won't allow another of the same (on any CPU) until we return.
1691          */
1692         drive->service_time = jiffies - drive->service_start;
1693         if (startstop == ide_stopped) {
1694                 if (hwgroup->handler == NULL) { /* paranoia */
1695                         hwgroup->busy = 0;
1696                         ide_do_request(hwgroup, hwif->irq);
1697                 } else {
1698                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1699                                 "on exit\n", drive->name);
1700                 }
1701         }
1702         spin_unlock_irqrestore(&ide_lock, flags);
1703         return IRQ_HANDLED;
1704 }
1705
1706 /**
1707  *      ide_init_drive_cmd      -       initialize a drive command request
1708  *      @rq: request object
1709  *
1710  *      Initialize a request before we fill it in and send it down to
1711  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1712  *      now it doesn't do a lot, but if that changes abusers will have a
1713  *      nasty surprise.
1714  */
1715
1716 void ide_init_drive_cmd (struct request *rq)
1717 {
1718         memset(rq, 0, sizeof(*rq));
1719         rq->cmd_type = REQ_TYPE_ATA_CMD;
1720         rq->ref_count = 1;
1721 }
1722
1723 EXPORT_SYMBOL(ide_init_drive_cmd);
1724
1725 /**
1726  *      ide_do_drive_cmd        -       issue IDE special command
1727  *      @drive: device to issue command
1728  *      @rq: request to issue
1729  *      @action: action for processing
1730  *
1731  *      This function issues a special IDE device request
1732  *      onto the request queue.
1733  *
1734  *      If action is ide_wait, then the rq is queued at the end of the
1735  *      request queue, and the function sleeps until it has been processed.
1736  *      This is for use when invoked from an ioctl handler.
1737  *
1738  *      If action is ide_preempt, then the rq is queued at the head of
1739  *      the request queue, displacing the currently-being-processed
1740  *      request and this function returns immediately without waiting
1741  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1742  *      intended for careful use by the ATAPI tape/cdrom driver code.
1743  *
1744  *      If action is ide_end, then the rq is queued at the end of the
1745  *      request queue, and the function returns immediately without waiting
1746  *      for the new rq to be completed. This is again intended for careful
1747  *      use by the ATAPI tape/cdrom driver code.
1748  */
1749  
1750 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1751 {
1752         unsigned long flags;
1753         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1754         DECLARE_COMPLETION_ONSTACK(wait);
1755         int where = ELEVATOR_INSERT_BACK, err;
1756         int must_wait = (action == ide_wait || action == ide_head_wait);
1757
1758         rq->errors = 0;
1759
1760         /*
1761          * we need to hold an extra reference to request for safe inspection
1762          * after completion
1763          */
1764         if (must_wait) {
1765                 rq->ref_count++;
1766                 rq->end_io_data = &wait;
1767                 rq->end_io = blk_end_sync_rq;
1768         }
1769
1770         spin_lock_irqsave(&ide_lock, flags);
1771         if (action == ide_preempt)
1772                 hwgroup->rq = NULL;
1773         if (action == ide_preempt || action == ide_head_wait) {
1774                 where = ELEVATOR_INSERT_FRONT;
1775                 rq->cmd_flags |= REQ_PREEMPT;
1776         }
1777         __elv_add_request(drive->queue, rq, where, 0);
1778         ide_do_request(hwgroup, IDE_NO_IRQ);
1779         spin_unlock_irqrestore(&ide_lock, flags);
1780
1781         err = 0;
1782         if (must_wait) {
1783                 wait_for_completion(&wait);
1784                 if (rq->errors)
1785                         err = -EIO;
1786
1787                 blk_put_request(rq);
1788         }
1789
1790         return err;
1791 }
1792
1793 EXPORT_SYMBOL(ide_do_drive_cmd);