sched: restore deterministic CPU accounting on powerpc
[linux-2.6] / drivers / ide / ide-io.c
1 /*
2  *      IDE I/O functions
3  *
4  *      Basic PIO and command management functionality.
5  *
6  * This code was split off from ide.c. See ide.c for history and original
7  * copyrights.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the
11  * Free Software Foundation; either version 2, or (at your option) any
12  * later version.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * For the avoidance of doubt the "preferred form" of this code is one which
20  * is in an open non patent encumbered format. Where cryptographic key signing
21  * forms part of the process of creating an executable the information
22  * including keys needed to generate an equivalently functional executable
23  * are deemed to be part of the source code.
24  */
25  
26  
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
32 #include <linux/mm.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/completion.h>
44 #include <linux/reboot.h>
45 #include <linux/cdrom.h>
46 #include <linux/seq_file.h>
47 #include <linux/device.h>
48 #include <linux/kmod.h>
49 #include <linux/scatterlist.h>
50 #include <linux/bitops.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/irq.h>
54 #include <asm/uaccess.h>
55 #include <asm/io.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, unsigned int nr_bytes, int dequeue)
59 {
60         int ret = 1;
61
62         /*
63          * if failfast is set on a request, override number of sectors and
64          * complete the whole request right now
65          */
66         if (blk_noretry_request(rq) && end_io_error(uptodate))
67                 nr_bytes = rq->hard_nr_sectors << 9;
68
69         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
70                 rq->errors = -EIO;
71
72         /*
73          * decide whether to reenable DMA -- 3 is a random magic for now,
74          * if we DMA timeout more than 3 times, just stay in PIO
75          */
76         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
77                 drive->state = 0;
78                 HWGROUP(drive)->hwif->ide_dma_on(drive);
79         }
80
81         if (!end_that_request_chunk(rq, uptodate, nr_bytes)) {
82                 add_disk_randomness(rq->rq_disk);
83                 if (dequeue) {
84                         if (!list_empty(&rq->queuelist))
85                                 blkdev_dequeue_request(rq);
86                         HWGROUP(drive)->rq = NULL;
87                 }
88                 end_that_request_last(rq, uptodate);
89                 ret = 0;
90         }
91
92         return ret;
93 }
94
95 /**
96  *      ide_end_request         -       complete an IDE I/O
97  *      @drive: IDE device for the I/O
98  *      @uptodate:
99  *      @nr_sectors: number of sectors completed
100  *
101  *      This is our end_request wrapper function. We complete the I/O
102  *      update random number input and dequeue the request, which if
103  *      it was tagged may be out of order.
104  */
105
106 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
107 {
108         unsigned int nr_bytes = nr_sectors << 9;
109         struct request *rq;
110         unsigned long flags;
111         int ret = 1;
112
113         /*
114          * room for locking improvements here, the calls below don't
115          * need the queue lock held at all
116          */
117         spin_lock_irqsave(&ide_lock, flags);
118         rq = HWGROUP(drive)->rq;
119
120         if (!nr_bytes) {
121                 if (blk_pc_request(rq))
122                         nr_bytes = rq->data_len;
123                 else
124                         nr_bytes = rq->hard_cur_sectors << 9;
125         }
126
127         ret = __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
128
129         spin_unlock_irqrestore(&ide_lock, flags);
130         return ret;
131 }
132 EXPORT_SYMBOL(ide_end_request);
133
134 /*
135  * Power Management state machine. This one is rather trivial for now,
136  * we should probably add more, like switching back to PIO on suspend
137  * to help some BIOSes, re-do the door locking on resume, etc...
138  */
139
140 enum {
141         ide_pm_flush_cache      = ide_pm_state_start_suspend,
142         idedisk_pm_standby,
143
144         idedisk_pm_restore_pio  = ide_pm_state_start_resume,
145         idedisk_pm_idle,
146         ide_pm_restore_dma,
147 };
148
149 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
150 {
151         struct request_pm_state *pm = rq->data;
152
153         if (drive->media != ide_disk)
154                 return;
155
156         switch (pm->pm_step) {
157         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
158                 if (pm->pm_state == PM_EVENT_FREEZE)
159                         pm->pm_step = ide_pm_state_completed;
160                 else
161                         pm->pm_step = idedisk_pm_standby;
162                 break;
163         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
164                 pm->pm_step = ide_pm_state_completed;
165                 break;
166         case idedisk_pm_restore_pio:    /* Resume step 1 complete */
167                 pm->pm_step = idedisk_pm_idle;
168                 break;
169         case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
170                 pm->pm_step = ide_pm_restore_dma;
171                 break;
172         }
173 }
174
175 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
176 {
177         struct request_pm_state *pm = rq->data;
178         ide_task_t *args = rq->special;
179
180         memset(args, 0, sizeof(*args));
181
182         switch (pm->pm_step) {
183         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
184                 if (drive->media != ide_disk)
185                         break;
186                 /* Not supported? Switch to next step now. */
187                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
188                         ide_complete_power_step(drive, rq, 0, 0);
189                         return ide_stopped;
190                 }
191                 if (ide_id_has_flush_cache_ext(drive->id))
192                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
193                 else
194                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
195                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
196                 args->handler      = &task_no_data_intr;
197                 return do_rw_taskfile(drive, args);
198
199         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
200                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
201                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
202                 args->handler      = &task_no_data_intr;
203                 return do_rw_taskfile(drive, args);
204
205         case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
206                 ide_set_max_pio(drive);
207                 /*
208                  * skip idedisk_pm_idle for ATAPI devices
209                  */
210                 if (drive->media != ide_disk)
211                         pm->pm_step = ide_pm_restore_dma;
212                 else
213                         ide_complete_power_step(drive, rq, 0, 0);
214                 return ide_stopped;
215
216         case idedisk_pm_idle:           /* Resume step 2 (idle) */
217                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
218                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
219                 args->handler = task_no_data_intr;
220                 return do_rw_taskfile(drive, args);
221
222         case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
223                 /*
224                  * Right now, all we do is call ide_set_dma(drive),
225                  * we could be smarter and check for current xfer_speed
226                  * in struct drive etc...
227                  */
228                 if (drive->hwif->ide_dma_on == NULL)
229                         break;
230                 drive->hwif->dma_off_quietly(drive);
231                 /*
232                  * TODO: respect ->using_dma setting
233                  */
234                 ide_set_dma(drive);
235                 break;
236         }
237         pm->pm_step = ide_pm_state_completed;
238         return ide_stopped;
239 }
240
241 /**
242  *      ide_end_dequeued_request        -       complete an IDE I/O
243  *      @drive: IDE device for the I/O
244  *      @uptodate:
245  *      @nr_sectors: number of sectors completed
246  *
247  *      Complete an I/O that is no longer on the request queue. This
248  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
249  *      We must still finish the old request but we must not tamper with the
250  *      queue in the meantime.
251  *
252  *      NOTE: This path does not handle barrier, but barrier is not supported
253  *      on ide-cd anyway.
254  */
255
256 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
257                              int uptodate, int nr_sectors)
258 {
259         unsigned long flags;
260         int ret;
261
262         spin_lock_irqsave(&ide_lock, flags);
263         BUG_ON(!blk_rq_started(rq));
264         ret = __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
265         spin_unlock_irqrestore(&ide_lock, flags);
266
267         return ret;
268 }
269 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
270
271
272 /**
273  *      ide_complete_pm_request - end the current Power Management request
274  *      @drive: target drive
275  *      @rq: request
276  *
277  *      This function cleans up the current PM request and stops the queue
278  *      if necessary.
279  */
280 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
281 {
282         unsigned long flags;
283
284 #ifdef DEBUG_PM
285         printk("%s: completing PM request, %s\n", drive->name,
286                blk_pm_suspend_request(rq) ? "suspend" : "resume");
287 #endif
288         spin_lock_irqsave(&ide_lock, flags);
289         if (blk_pm_suspend_request(rq)) {
290                 blk_stop_queue(drive->queue);
291         } else {
292                 drive->blocked = 0;
293                 blk_start_queue(drive->queue);
294         }
295         blkdev_dequeue_request(rq);
296         HWGROUP(drive)->rq = NULL;
297         end_that_request_last(rq, 1);
298         spin_unlock_irqrestore(&ide_lock, flags);
299 }
300
301 /**
302  *      ide_end_drive_cmd       -       end an explicit drive command
303  *      @drive: command 
304  *      @stat: status bits
305  *      @err: error bits
306  *
307  *      Clean up after success/failure of an explicit drive command.
308  *      These get thrown onto the queue so they are synchronized with
309  *      real I/O operations on the drive.
310  *
311  *      In LBA48 mode we have to read the register set twice to get
312  *      all the extra information out.
313  */
314  
315 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
316 {
317         ide_hwif_t *hwif = HWIF(drive);
318         unsigned long flags;
319         struct request *rq;
320
321         spin_lock_irqsave(&ide_lock, flags);
322         rq = HWGROUP(drive)->rq;
323         spin_unlock_irqrestore(&ide_lock, flags);
324
325         if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
326                 u8 *args = (u8 *) rq->buffer;
327                 if (rq->errors == 0)
328                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
329
330                 if (args) {
331                         args[0] = stat;
332                         args[1] = err;
333                         args[2] = hwif->INB(IDE_NSECTOR_REG);
334                 }
335         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
336                 u8 *args = (u8 *) rq->buffer;
337                 if (rq->errors == 0)
338                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
339
340                 if (args) {
341                         args[0] = stat;
342                         args[1] = err;
343                         /* be sure we're looking at the low order bits */
344                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
345                         args[2] = hwif->INB(IDE_NSECTOR_REG);
346                         args[3] = hwif->INB(IDE_SECTOR_REG);
347                         args[4] = hwif->INB(IDE_LCYL_REG);
348                         args[5] = hwif->INB(IDE_HCYL_REG);
349                         args[6] = hwif->INB(IDE_SELECT_REG);
350                 }
351         } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
352                 ide_task_t *args = (ide_task_t *) rq->special;
353                 if (rq->errors == 0)
354                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
355                         
356                 if (args) {
357                         if (args->tf_in_flags.b.data) {
358                                 u16 data                                = hwif->INW(IDE_DATA_REG);
359                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
360                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
361                         }
362                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
363                         /* be sure we're looking at the low order bits */
364                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
365                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
366                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
367                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
368                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
369                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
370                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
371
372                         if (drive->addressing == 1) {
373                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
374                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
375                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
376                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
377                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
378                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
379                         }
380                 }
381         } else if (blk_pm_request(rq)) {
382                 struct request_pm_state *pm = rq->data;
383 #ifdef DEBUG_PM
384                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
385                         drive->name, rq->pm->pm_step, stat, err);
386 #endif
387                 ide_complete_power_step(drive, rq, stat, err);
388                 if (pm->pm_step == ide_pm_state_completed)
389                         ide_complete_pm_request(drive, rq);
390                 return;
391         }
392
393         spin_lock_irqsave(&ide_lock, flags);
394         blkdev_dequeue_request(rq);
395         HWGROUP(drive)->rq = NULL;
396         rq->errors = err;
397         end_that_request_last(rq, !rq->errors);
398         spin_unlock_irqrestore(&ide_lock, flags);
399 }
400
401 EXPORT_SYMBOL(ide_end_drive_cmd);
402
403 /**
404  *      try_to_flush_leftover_data      -       flush junk
405  *      @drive: drive to flush
406  *
407  *      try_to_flush_leftover_data() is invoked in response to a drive
408  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
409  *      resetting the drive, this routine tries to clear the condition
410  *      by read a sector's worth of data from the drive.  Of course,
411  *      this may not help if the drive is *waiting* for data from *us*.
412  */
413 static void try_to_flush_leftover_data (ide_drive_t *drive)
414 {
415         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
416
417         if (drive->media != ide_disk)
418                 return;
419         while (i > 0) {
420                 u32 buffer[16];
421                 u32 wcount = (i > 16) ? 16 : i;
422
423                 i -= wcount;
424                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
425         }
426 }
427
428 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
429 {
430         if (rq->rq_disk) {
431                 ide_driver_t *drv;
432
433                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
434                 drv->end_request(drive, 0, 0);
435         } else
436                 ide_end_request(drive, 0, 0);
437 }
438
439 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
440 {
441         ide_hwif_t *hwif = drive->hwif;
442
443         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
444                 /* other bits are useless when BUSY */
445                 rq->errors |= ERROR_RESET;
446         } else if (stat & ERR_STAT) {
447                 /* err has different meaning on cdrom and tape */
448                 if (err == ABRT_ERR) {
449                         if (drive->select.b.lba &&
450                             /* some newer drives don't support WIN_SPECIFY */
451                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
452                                 return ide_stopped;
453                 } else if ((err & BAD_CRC) == BAD_CRC) {
454                         /* UDMA crc error, just retry the operation */
455                         drive->crc_count++;
456                 } else if (err & (BBD_ERR | ECC_ERR)) {
457                         /* retries won't help these */
458                         rq->errors = ERROR_MAX;
459                 } else if (err & TRK0_ERR) {
460                         /* help it find track zero */
461                         rq->errors |= ERROR_RECAL;
462                 }
463         }
464
465         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ &&
466             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0)
467                 try_to_flush_leftover_data(drive);
468
469         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
470                 ide_kill_rq(drive, rq);
471                 return ide_stopped;
472         }
473
474         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
475                 rq->errors |= ERROR_RESET;
476
477         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
478                 ++rq->errors;
479                 return ide_do_reset(drive);
480         }
481
482         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
483                 drive->special.b.recalibrate = 1;
484
485         ++rq->errors;
486
487         return ide_stopped;
488 }
489
490 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
491 {
492         ide_hwif_t *hwif = drive->hwif;
493
494         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
495                 /* other bits are useless when BUSY */
496                 rq->errors |= ERROR_RESET;
497         } else {
498                 /* add decoding error stuff */
499         }
500
501         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
502                 /* force an abort */
503                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
504
505         if (rq->errors >= ERROR_MAX) {
506                 ide_kill_rq(drive, rq);
507         } else {
508                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
509                         ++rq->errors;
510                         return ide_do_reset(drive);
511                 }
512                 ++rq->errors;
513         }
514
515         return ide_stopped;
516 }
517
518 ide_startstop_t
519 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
520 {
521         if (drive->media == ide_disk)
522                 return ide_ata_error(drive, rq, stat, err);
523         return ide_atapi_error(drive, rq, stat, err);
524 }
525
526 EXPORT_SYMBOL_GPL(__ide_error);
527
528 /**
529  *      ide_error       -       handle an error on the IDE
530  *      @drive: drive the error occurred on
531  *      @msg: message to report
532  *      @stat: status bits
533  *
534  *      ide_error() takes action based on the error returned by the drive.
535  *      For normal I/O that may well include retries. We deal with
536  *      both new-style (taskfile) and old style command handling here.
537  *      In the case of taskfile command handling there is work left to
538  *      do
539  */
540  
541 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
542 {
543         struct request *rq;
544         u8 err;
545
546         err = ide_dump_status(drive, msg, stat);
547
548         if ((rq = HWGROUP(drive)->rq) == NULL)
549                 return ide_stopped;
550
551         /* retry only "normal" I/O: */
552         if (!blk_fs_request(rq)) {
553                 rq->errors = 1;
554                 ide_end_drive_cmd(drive, stat, err);
555                 return ide_stopped;
556         }
557
558         if (rq->rq_disk) {
559                 ide_driver_t *drv;
560
561                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
562                 return drv->error(drive, rq, stat, err);
563         } else
564                 return __ide_error(drive, rq, stat, err);
565 }
566
567 EXPORT_SYMBOL_GPL(ide_error);
568
569 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
570 {
571         if (drive->media != ide_disk)
572                 rq->errors |= ERROR_RESET;
573
574         ide_kill_rq(drive, rq);
575
576         return ide_stopped;
577 }
578
579 EXPORT_SYMBOL_GPL(__ide_abort);
580
581 /**
582  *      ide_abort       -       abort pending IDE operations
583  *      @drive: drive the error occurred on
584  *      @msg: message to report
585  *
586  *      ide_abort kills and cleans up when we are about to do a 
587  *      host initiated reset on active commands. Longer term we
588  *      want handlers to have sensible abort handling themselves
589  *
590  *      This differs fundamentally from ide_error because in 
591  *      this case the command is doing just fine when we
592  *      blow it away.
593  */
594  
595 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
596 {
597         struct request *rq;
598
599         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
600                 return ide_stopped;
601
602         /* retry only "normal" I/O: */
603         if (!blk_fs_request(rq)) {
604                 rq->errors = 1;
605                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
606                 return ide_stopped;
607         }
608
609         if (rq->rq_disk) {
610                 ide_driver_t *drv;
611
612                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
613                 return drv->abort(drive, rq);
614         } else
615                 return __ide_abort(drive, rq);
616 }
617
618 /**
619  *      ide_cmd         -       issue a simple drive command
620  *      @drive: drive the command is for
621  *      @cmd: command byte
622  *      @nsect: sector byte
623  *      @handler: handler for the command completion
624  *
625  *      Issue a simple drive command with interrupts.
626  *      The drive must be selected beforehand.
627  */
628
629 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
630                 ide_handler_t *handler)
631 {
632         ide_hwif_t *hwif = HWIF(drive);
633         if (IDE_CONTROL_REG)
634                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
635         SELECT_MASK(drive,0);
636         hwif->OUTB(nsect,IDE_NSECTOR_REG);
637         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
638 }
639
640 /**
641  *      drive_cmd_intr          -       drive command completion interrupt
642  *      @drive: drive the completion interrupt occurred on
643  *
644  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
645  *      We do any necessary data reading and then wait for the drive to
646  *      go non busy. At that point we may read the error data and complete
647  *      the request
648  */
649  
650 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
651 {
652         struct request *rq = HWGROUP(drive)->rq;
653         ide_hwif_t *hwif = HWIF(drive);
654         u8 *args = (u8 *) rq->buffer;
655         u8 stat = hwif->INB(IDE_STATUS_REG);
656         int retries = 10;
657
658         local_irq_enable_in_hardirq();
659         if (rq->cmd_type == REQ_TYPE_ATA_CMD &&
660             (stat & DRQ_STAT) && args && args[3]) {
661                 u8 io_32bit = drive->io_32bit;
662                 drive->io_32bit = 0;
663                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
664                 drive->io_32bit = io_32bit;
665                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
666                         udelay(100);
667         }
668
669         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
670                 return ide_error(drive, "drive_cmd", stat);
671                 /* calls ide_end_drive_cmd */
672         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
673         return ide_stopped;
674 }
675
676 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
677 {
678         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
679         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
680         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
681         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
682         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
683         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
684
685         task->handler = &set_geometry_intr;
686 }
687
688 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
689 {
690         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
691         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
692
693         task->handler = &recal_intr;
694 }
695
696 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
697 {
698         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
699         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
700
701         task->handler = &set_multmode_intr;
702 }
703
704 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
705 {
706         special_t *s = &drive->special;
707         ide_task_t args;
708
709         memset(&args, 0, sizeof(ide_task_t));
710         args.command_type = IDE_DRIVE_TASK_NO_DATA;
711
712         if (s->b.set_geometry) {
713                 s->b.set_geometry = 0;
714                 ide_init_specify_cmd(drive, &args);
715         } else if (s->b.recalibrate) {
716                 s->b.recalibrate = 0;
717                 ide_init_restore_cmd(drive, &args);
718         } else if (s->b.set_multmode) {
719                 s->b.set_multmode = 0;
720                 if (drive->mult_req > drive->id->max_multsect)
721                         drive->mult_req = drive->id->max_multsect;
722                 ide_init_setmult_cmd(drive, &args);
723         } else if (s->all) {
724                 int special = s->all;
725                 s->all = 0;
726                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
727                 return ide_stopped;
728         }
729
730         do_rw_taskfile(drive, &args);
731
732         return ide_started;
733 }
734
735 /*
736  * handle HDIO_SET_PIO_MODE ioctl abusers here, eventually it will go away
737  */
738 static int set_pio_mode_abuse(ide_hwif_t *hwif, u8 req_pio)
739 {
740         switch (req_pio) {
741         case 202:
742         case 201:
743         case 200:
744         case 102:
745         case 101:
746         case 100:
747                 return (hwif->host_flags & IDE_HFLAG_ABUSE_DMA_MODES) ? 1 : 0;
748         case 9:
749         case 8:
750                 return (hwif->host_flags & IDE_HFLAG_ABUSE_PREFETCH) ? 1 : 0;
751         case 7:
752         case 6:
753                 return (hwif->host_flags & IDE_HFLAG_ABUSE_FAST_DEVSEL) ? 1 : 0;
754         default:
755                 return 0;
756         }
757 }
758
759 /**
760  *      do_special              -       issue some special commands
761  *      @drive: drive the command is for
762  *
763  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
764  *      commands to a drive.  It used to do much more, but has been scaled
765  *      back.
766  */
767
768 static ide_startstop_t do_special (ide_drive_t *drive)
769 {
770         special_t *s = &drive->special;
771
772 #ifdef DEBUG
773         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
774 #endif
775         if (s->b.set_tune) {
776                 ide_hwif_t *hwif = drive->hwif;
777                 u8 req_pio = drive->tune_req;
778
779                 s->b.set_tune = 0;
780
781                 if (set_pio_mode_abuse(drive->hwif, req_pio)) {
782
783                         if (hwif->set_pio_mode == NULL)
784                                 return ide_stopped;
785
786                         /*
787                          * take ide_lock for drive->[no_]unmask/[no_]io_32bit
788                          */
789                         if (req_pio == 8 || req_pio == 9) {
790                                 unsigned long flags;
791
792                                 spin_lock_irqsave(&ide_lock, flags);
793                                 hwif->set_pio_mode(drive, req_pio);
794                                 spin_unlock_irqrestore(&ide_lock, flags);
795                         } else
796                                 hwif->set_pio_mode(drive, req_pio);
797                 } else {
798                         int keep_dma = drive->using_dma;
799
800                         ide_set_pio(drive, req_pio);
801
802                         if (hwif->host_flags & IDE_HFLAG_SET_PIO_MODE_KEEP_DMA) {
803                                 if (keep_dma)
804                                         hwif->ide_dma_on(drive);
805                         }
806                 }
807
808                 return ide_stopped;
809         } else {
810                 if (drive->media == ide_disk)
811                         return ide_disk_special(drive);
812
813                 s->all = 0;
814                 drive->mult_req = 0;
815                 return ide_stopped;
816         }
817 }
818
819 void ide_map_sg(ide_drive_t *drive, struct request *rq)
820 {
821         ide_hwif_t *hwif = drive->hwif;
822         struct scatterlist *sg = hwif->sg_table;
823
824         if (hwif->sg_mapped)    /* needed by ide-scsi */
825                 return;
826
827         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
828                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
829         } else {
830                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
831                 hwif->sg_nents = 1;
832         }
833 }
834
835 EXPORT_SYMBOL_GPL(ide_map_sg);
836
837 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
838 {
839         ide_hwif_t *hwif = drive->hwif;
840
841         hwif->nsect = hwif->nleft = rq->nr_sectors;
842         hwif->cursg_ofs = 0;
843         hwif->cursg = NULL;
844 }
845
846 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
847
848 /**
849  *      execute_drive_command   -       issue special drive command
850  *      @drive: the drive to issue the command on
851  *      @rq: the request structure holding the command
852  *
853  *      execute_drive_cmd() issues a special drive command,  usually 
854  *      initiated by ioctl() from the external hdparm program. The
855  *      command can be a drive command, drive task or taskfile 
856  *      operation. Weirdly you can call it with NULL to wait for
857  *      all commands to finish. Don't do this as that is due to change
858  */
859
860 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
861                 struct request *rq)
862 {
863         ide_hwif_t *hwif = HWIF(drive);
864         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
865                 ide_task_t *args = rq->special;
866  
867                 if (!args)
868                         goto done;
869
870                 hwif->data_phase = args->data_phase;
871
872                 switch (hwif->data_phase) {
873                 case TASKFILE_MULTI_OUT:
874                 case TASKFILE_OUT:
875                 case TASKFILE_MULTI_IN:
876                 case TASKFILE_IN:
877                         ide_init_sg_cmd(drive, rq);
878                         ide_map_sg(drive, rq);
879                 default:
880                         break;
881                 }
882
883                 if (args->tf_out_flags.all != 0) 
884                         return flagged_taskfile(drive, args);
885                 return do_rw_taskfile(drive, args);
886         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
887                 u8 *args = rq->buffer;
888                 u8 sel;
889  
890                 if (!args)
891                         goto done;
892 #ifdef DEBUG
893                 printk("%s: DRIVE_TASK_CMD ", drive->name);
894                 printk("cmd=0x%02x ", args[0]);
895                 printk("fr=0x%02x ", args[1]);
896                 printk("ns=0x%02x ", args[2]);
897                 printk("sc=0x%02x ", args[3]);
898                 printk("lcyl=0x%02x ", args[4]);
899                 printk("hcyl=0x%02x ", args[5]);
900                 printk("sel=0x%02x\n", args[6]);
901 #endif
902                 hwif->OUTB(args[1], IDE_FEATURE_REG);
903                 hwif->OUTB(args[3], IDE_SECTOR_REG);
904                 hwif->OUTB(args[4], IDE_LCYL_REG);
905                 hwif->OUTB(args[5], IDE_HCYL_REG);
906                 sel = (args[6] & ~0x10);
907                 if (drive->select.b.unit)
908                         sel |= 0x10;
909                 hwif->OUTB(sel, IDE_SELECT_REG);
910                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
911                 return ide_started;
912         } else if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
913                 u8 *args = rq->buffer;
914
915                 if (!args)
916                         goto done;
917 #ifdef DEBUG
918                 printk("%s: DRIVE_CMD ", drive->name);
919                 printk("cmd=0x%02x ", args[0]);
920                 printk("sc=0x%02x ", args[1]);
921                 printk("fr=0x%02x ", args[2]);
922                 printk("xx=0x%02x\n", args[3]);
923 #endif
924                 if (args[0] == WIN_SMART) {
925                         hwif->OUTB(0x4f, IDE_LCYL_REG);
926                         hwif->OUTB(0xc2, IDE_HCYL_REG);
927                         hwif->OUTB(args[2],IDE_FEATURE_REG);
928                         hwif->OUTB(args[1],IDE_SECTOR_REG);
929                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
930                         return ide_started;
931                 }
932                 hwif->OUTB(args[2],IDE_FEATURE_REG);
933                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
934                 return ide_started;
935         }
936
937 done:
938         /*
939          * NULL is actually a valid way of waiting for
940          * all current requests to be flushed from the queue.
941          */
942 #ifdef DEBUG
943         printk("%s: DRIVE_CMD (null)\n", drive->name);
944 #endif
945         ide_end_drive_cmd(drive,
946                         hwif->INB(IDE_STATUS_REG),
947                         hwif->INB(IDE_ERROR_REG));
948         return ide_stopped;
949 }
950
951 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
952 {
953         struct request_pm_state *pm = rq->data;
954
955         if (blk_pm_suspend_request(rq) &&
956             pm->pm_step == ide_pm_state_start_suspend)
957                 /* Mark drive blocked when starting the suspend sequence. */
958                 drive->blocked = 1;
959         else if (blk_pm_resume_request(rq) &&
960                  pm->pm_step == ide_pm_state_start_resume) {
961                 /* 
962                  * The first thing we do on wakeup is to wait for BSY bit to
963                  * go away (with a looong timeout) as a drive on this hwif may
964                  * just be POSTing itself.
965                  * We do that before even selecting as the "other" device on
966                  * the bus may be broken enough to walk on our toes at this
967                  * point.
968                  */
969                 int rc;
970 #ifdef DEBUG_PM
971                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
972 #endif
973                 rc = ide_wait_not_busy(HWIF(drive), 35000);
974                 if (rc)
975                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
976                 SELECT_DRIVE(drive);
977                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
978                 rc = ide_wait_not_busy(HWIF(drive), 100000);
979                 if (rc)
980                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
981         }
982 }
983
984 /**
985  *      start_request   -       start of I/O and command issuing for IDE
986  *
987  *      start_request() initiates handling of a new I/O request. It
988  *      accepts commands and I/O (read/write) requests. It also does
989  *      the final remapping for weird stuff like EZDrive. Once 
990  *      device mapper can work sector level the EZDrive stuff can go away
991  *
992  *      FIXME: this function needs a rename
993  */
994  
995 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
996 {
997         ide_startstop_t startstop;
998         sector_t block;
999
1000         BUG_ON(!blk_rq_started(rq));
1001
1002 #ifdef DEBUG
1003         printk("%s: start_request: current=0x%08lx\n",
1004                 HWIF(drive)->name, (unsigned long) rq);
1005 #endif
1006
1007         /* bail early if we've exceeded max_failures */
1008         if (drive->max_failures && (drive->failures > drive->max_failures)) {
1009                 goto kill_rq;
1010         }
1011
1012         block    = rq->sector;
1013         if (blk_fs_request(rq) &&
1014             (drive->media == ide_disk || drive->media == ide_floppy)) {
1015                 block += drive->sect0;
1016         }
1017         /* Yecch - this will shift the entire interval,
1018            possibly killing some innocent following sector */
1019         if (block == 0 && drive->remap_0_to_1 == 1)
1020                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
1021
1022         if (blk_pm_request(rq))
1023                 ide_check_pm_state(drive, rq);
1024
1025         SELECT_DRIVE(drive);
1026         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1027                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1028                 return startstop;
1029         }
1030         if (!drive->special.all) {
1031                 ide_driver_t *drv;
1032
1033                 /*
1034                  * We reset the drive so we need to issue a SETFEATURES.
1035                  * Do it _after_ do_special() restored device parameters.
1036                  */
1037                 if (drive->current_speed == 0xff)
1038                         ide_config_drive_speed(drive, drive->desired_speed);
1039
1040                 if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1041                     rq->cmd_type == REQ_TYPE_ATA_TASK ||
1042                     rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1043                         return execute_drive_cmd(drive, rq);
1044                 else if (blk_pm_request(rq)) {
1045                         struct request_pm_state *pm = rq->data;
1046 #ifdef DEBUG_PM
1047                         printk("%s: start_power_step(step: %d)\n",
1048                                 drive->name, rq->pm->pm_step);
1049 #endif
1050                         startstop = ide_start_power_step(drive, rq);
1051                         if (startstop == ide_stopped &&
1052                             pm->pm_step == ide_pm_state_completed)
1053                                 ide_complete_pm_request(drive, rq);
1054                         return startstop;
1055                 }
1056
1057                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1058                 return drv->do_request(drive, rq, block);
1059         }
1060         return do_special(drive);
1061 kill_rq:
1062         ide_kill_rq(drive, rq);
1063         return ide_stopped;
1064 }
1065
1066 /**
1067  *      ide_stall_queue         -       pause an IDE device
1068  *      @drive: drive to stall
1069  *      @timeout: time to stall for (jiffies)
1070  *
1071  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1072  *      to the hwgroup by sleeping for timeout jiffies.
1073  */
1074  
1075 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1076 {
1077         if (timeout > WAIT_WORSTCASE)
1078                 timeout = WAIT_WORSTCASE;
1079         drive->sleep = timeout + jiffies;
1080         drive->sleeping = 1;
1081 }
1082
1083 EXPORT_SYMBOL(ide_stall_queue);
1084
1085 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1086
1087 /**
1088  *      choose_drive            -       select a drive to service
1089  *      @hwgroup: hardware group to select on
1090  *
1091  *      choose_drive() selects the next drive which will be serviced.
1092  *      This is necessary because the IDE layer can't issue commands
1093  *      to both drives on the same cable, unlike SCSI.
1094  */
1095  
1096 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1097 {
1098         ide_drive_t *drive, *best;
1099
1100 repeat: 
1101         best = NULL;
1102         drive = hwgroup->drive;
1103
1104         /*
1105          * drive is doing pre-flush, ordered write, post-flush sequence. even
1106          * though that is 3 requests, it must be seen as a single transaction.
1107          * we must not preempt this drive until that is complete
1108          */
1109         if (blk_queue_flushing(drive->queue)) {
1110                 /*
1111                  * small race where queue could get replugged during
1112                  * the 3-request flush cycle, just yank the plug since
1113                  * we want it to finish asap
1114                  */
1115                 blk_remove_plug(drive->queue);
1116                 return drive;
1117         }
1118
1119         do {
1120                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1121                     && !elv_queue_empty(drive->queue)) {
1122                         if (!best
1123                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1124                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1125                         {
1126                                 if (!blk_queue_plugged(drive->queue))
1127                                         best = drive;
1128                         }
1129                 }
1130         } while ((drive = drive->next) != hwgroup->drive);
1131         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1132                 long t = (signed long)(WAKEUP(best) - jiffies);
1133                 if (t >= WAIT_MIN_SLEEP) {
1134                 /*
1135                  * We *may* have some time to spare, but first let's see if
1136                  * someone can potentially benefit from our nice mood today..
1137                  */
1138                         drive = best->next;
1139                         do {
1140                                 if (!drive->sleeping
1141                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1142                                  && time_before(WAKEUP(drive), jiffies + t))
1143                                 {
1144                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1145                                         goto repeat;
1146                                 }
1147                         } while ((drive = drive->next) != best);
1148                 }
1149         }
1150         return best;
1151 }
1152
1153 /*
1154  * Issue a new request to a drive from hwgroup
1155  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1156  *
1157  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1158  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1159  * may have both interfaces in a single hwgroup to "serialize" access.
1160  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1161  * together into one hwgroup for serialized access.
1162  *
1163  * Note also that several hwgroups can end up sharing a single IRQ,
1164  * possibly along with many other devices.  This is especially common in
1165  * PCI-based systems with off-board IDE controller cards.
1166  *
1167  * The IDE driver uses the single global ide_lock spinlock to protect
1168  * access to the request queues, and to protect the hwgroup->busy flag.
1169  *
1170  * The first thread into the driver for a particular hwgroup sets the
1171  * hwgroup->busy flag to indicate that this hwgroup is now active,
1172  * and then initiates processing of the top request from the request queue.
1173  *
1174  * Other threads attempting entry notice the busy setting, and will simply
1175  * queue their new requests and exit immediately.  Note that hwgroup->busy
1176  * remains set even when the driver is merely awaiting the next interrupt.
1177  * Thus, the meaning is "this hwgroup is busy processing a request".
1178  *
1179  * When processing of a request completes, the completing thread or IRQ-handler
1180  * will start the next request from the queue.  If no more work remains,
1181  * the driver will clear the hwgroup->busy flag and exit.
1182  *
1183  * The ide_lock (spinlock) is used to protect all access to the
1184  * hwgroup->busy flag, but is otherwise not needed for most processing in
1185  * the driver.  This makes the driver much more friendlier to shared IRQs
1186  * than previous designs, while remaining 100% (?) SMP safe and capable.
1187  */
1188 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1189 {
1190         ide_drive_t     *drive;
1191         ide_hwif_t      *hwif;
1192         struct request  *rq;
1193         ide_startstop_t startstop;
1194         int             loops = 0;
1195
1196         /* for atari only: POSSIBLY BROKEN HERE(?) */
1197         ide_get_lock(ide_intr, hwgroup);
1198
1199         /* caller must own ide_lock */
1200         BUG_ON(!irqs_disabled());
1201
1202         while (!hwgroup->busy) {
1203                 hwgroup->busy = 1;
1204                 drive = choose_drive(hwgroup);
1205                 if (drive == NULL) {
1206                         int sleeping = 0;
1207                         unsigned long sleep = 0; /* shut up, gcc */
1208                         hwgroup->rq = NULL;
1209                         drive = hwgroup->drive;
1210                         do {
1211                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1212                                         sleeping = 1;
1213                                         sleep = drive->sleep;
1214                                 }
1215                         } while ((drive = drive->next) != hwgroup->drive);
1216                         if (sleeping) {
1217                 /*
1218                  * Take a short snooze, and then wake up this hwgroup again.
1219                  * This gives other hwgroups on the same a chance to
1220                  * play fairly with us, just in case there are big differences
1221                  * in relative throughputs.. don't want to hog the cpu too much.
1222                  */
1223                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1224                                         sleep = jiffies + WAIT_MIN_SLEEP;
1225 #if 1
1226                                 if (timer_pending(&hwgroup->timer))
1227                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1228 #endif
1229                                 /* so that ide_timer_expiry knows what to do */
1230                                 hwgroup->sleeping = 1;
1231                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1232                                 mod_timer(&hwgroup->timer, sleep);
1233                                 /* we purposely leave hwgroup->busy==1
1234                                  * while sleeping */
1235                         } else {
1236                                 /* Ugly, but how can we sleep for the lock
1237                                  * otherwise? perhaps from tq_disk?
1238                                  */
1239
1240                                 /* for atari only */
1241                                 ide_release_lock();
1242                                 hwgroup->busy = 0;
1243                         }
1244
1245                         /* no more work for this hwgroup (for now) */
1246                         return;
1247                 }
1248         again:
1249                 hwif = HWIF(drive);
1250                 if (hwgroup->hwif->sharing_irq &&
1251                     hwif != hwgroup->hwif &&
1252                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1253                         /* set nIEN for previous hwif */
1254                         SELECT_INTERRUPT(drive);
1255                 }
1256                 hwgroup->hwif = hwif;
1257                 hwgroup->drive = drive;
1258                 drive->sleeping = 0;
1259                 drive->service_start = jiffies;
1260
1261                 if (blk_queue_plugged(drive->queue)) {
1262                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1263                         break;
1264                 }
1265
1266                 /*
1267                  * we know that the queue isn't empty, but this can happen
1268                  * if the q->prep_rq_fn() decides to kill a request
1269                  */
1270                 rq = elv_next_request(drive->queue);
1271                 if (!rq) {
1272                         hwgroup->busy = 0;
1273                         break;
1274                 }
1275
1276                 /*
1277                  * Sanity: don't accept a request that isn't a PM request
1278                  * if we are currently power managed. This is very important as
1279                  * blk_stop_queue() doesn't prevent the elv_next_request()
1280                  * above to return us whatever is in the queue. Since we call
1281                  * ide_do_request() ourselves, we end up taking requests while
1282                  * the queue is blocked...
1283                  * 
1284                  * We let requests forced at head of queue with ide-preempt
1285                  * though. I hope that doesn't happen too much, hopefully not
1286                  * unless the subdriver triggers such a thing in its own PM
1287                  * state machine.
1288                  *
1289                  * We count how many times we loop here to make sure we service
1290                  * all drives in the hwgroup without looping for ever
1291                  */
1292                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1293                         drive = drive->next ? drive->next : hwgroup->drive;
1294                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1295                                 goto again;
1296                         /* We clear busy, there should be no pending ATA command at this point. */
1297                         hwgroup->busy = 0;
1298                         break;
1299                 }
1300
1301                 hwgroup->rq = rq;
1302
1303                 /*
1304                  * Some systems have trouble with IDE IRQs arriving while
1305                  * the driver is still setting things up.  So, here we disable
1306                  * the IRQ used by this interface while the request is being started.
1307                  * This may look bad at first, but pretty much the same thing
1308                  * happens anyway when any interrupt comes in, IDE or otherwise
1309                  *  -- the kernel masks the IRQ while it is being handled.
1310                  */
1311                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1312                         disable_irq_nosync(hwif->irq);
1313                 spin_unlock(&ide_lock);
1314                 local_irq_enable_in_hardirq();
1315                         /* allow other IRQs while we start this request */
1316                 startstop = start_request(drive, rq);
1317                 spin_lock_irq(&ide_lock);
1318                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1319                         enable_irq(hwif->irq);
1320                 if (startstop == ide_stopped)
1321                         hwgroup->busy = 0;
1322         }
1323 }
1324
1325 /*
1326  * Passes the stuff to ide_do_request
1327  */
1328 void do_ide_request(struct request_queue *q)
1329 {
1330         ide_drive_t *drive = q->queuedata;
1331
1332         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1333 }
1334
1335 /*
1336  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1337  * retry the current request in pio mode instead of risking tossing it
1338  * all away
1339  */
1340 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1341 {
1342         ide_hwif_t *hwif = HWIF(drive);
1343         struct request *rq;
1344         ide_startstop_t ret = ide_stopped;
1345
1346         /*
1347          * end current dma transaction
1348          */
1349
1350         if (error < 0) {
1351                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1352                 (void)HWIF(drive)->ide_dma_end(drive);
1353                 ret = ide_error(drive, "dma timeout error",
1354                                                 hwif->INB(IDE_STATUS_REG));
1355         } else {
1356                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1357                 hwif->dma_timeout(drive);
1358         }
1359
1360         /*
1361          * disable dma for now, but remember that we did so because of
1362          * a timeout -- we'll reenable after we finish this next request
1363          * (or rather the first chunk of it) in pio.
1364          */
1365         drive->retry_pio++;
1366         drive->state = DMA_PIO_RETRY;
1367         hwif->dma_off_quietly(drive);
1368
1369         /*
1370          * un-busy drive etc (hwgroup->busy is cleared on return) and
1371          * make sure request is sane
1372          */
1373         rq = HWGROUP(drive)->rq;
1374
1375         if (!rq)
1376                 goto out;
1377
1378         HWGROUP(drive)->rq = NULL;
1379
1380         rq->errors = 0;
1381
1382         if (!rq->bio)
1383                 goto out;
1384
1385         rq->sector = rq->bio->bi_sector;
1386         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1387         rq->hard_cur_sectors = rq->current_nr_sectors;
1388         rq->buffer = bio_data(rq->bio);
1389 out:
1390         return ret;
1391 }
1392
1393 /**
1394  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1395  *      @data: timer callback magic (hwgroup)
1396  *
1397  *      An IDE command has timed out before the expected drive return
1398  *      occurred. At this point we attempt to clean up the current
1399  *      mess. If the current handler includes an expiry handler then
1400  *      we invoke the expiry handler, and providing it is happy the
1401  *      work is done. If that fails we apply generic recovery rules
1402  *      invoking the handler and checking the drive DMA status. We
1403  *      have an excessively incestuous relationship with the DMA
1404  *      logic that wants cleaning up.
1405  */
1406  
1407 void ide_timer_expiry (unsigned long data)
1408 {
1409         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1410         ide_handler_t   *handler;
1411         ide_expiry_t    *expiry;
1412         unsigned long   flags;
1413         unsigned long   wait = -1;
1414
1415         spin_lock_irqsave(&ide_lock, flags);
1416
1417         if (((handler = hwgroup->handler) == NULL) ||
1418             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1419                 /*
1420                  * Either a marginal timeout occurred
1421                  * (got the interrupt just as timer expired),
1422                  * or we were "sleeping" to give other devices a chance.
1423                  * Either way, we don't really want to complain about anything.
1424                  */
1425                 if (hwgroup->sleeping) {
1426                         hwgroup->sleeping = 0;
1427                         hwgroup->busy = 0;
1428                 }
1429         } else {
1430                 ide_drive_t *drive = hwgroup->drive;
1431                 if (!drive) {
1432                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1433                         hwgroup->handler = NULL;
1434                 } else {
1435                         ide_hwif_t *hwif;
1436                         ide_startstop_t startstop = ide_stopped;
1437                         if (!hwgroup->busy) {
1438                                 hwgroup->busy = 1;      /* paranoia */
1439                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1440                         }
1441                         if ((expiry = hwgroup->expiry) != NULL) {
1442                                 /* continue */
1443                                 if ((wait = expiry(drive)) > 0) {
1444                                         /* reset timer */
1445                                         hwgroup->timer.expires  = jiffies + wait;
1446                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1447                                         add_timer(&hwgroup->timer);
1448                                         spin_unlock_irqrestore(&ide_lock, flags);
1449                                         return;
1450                                 }
1451                         }
1452                         hwgroup->handler = NULL;
1453                         /*
1454                          * We need to simulate a real interrupt when invoking
1455                          * the handler() function, which means we need to
1456                          * globally mask the specific IRQ:
1457                          */
1458                         spin_unlock(&ide_lock);
1459                         hwif  = HWIF(drive);
1460 #if DISABLE_IRQ_NOSYNC
1461                         disable_irq_nosync(hwif->irq);
1462 #else
1463                         /* disable_irq_nosync ?? */
1464                         disable_irq(hwif->irq);
1465 #endif /* DISABLE_IRQ_NOSYNC */
1466                         /* local CPU only,
1467                          * as if we were handling an interrupt */
1468                         local_irq_disable();
1469                         if (hwgroup->polling) {
1470                                 startstop = handler(drive);
1471                         } else if (drive_is_ready(drive)) {
1472                                 if (drive->waiting_for_dma)
1473                                         hwgroup->hwif->dma_lost_irq(drive);
1474                                 (void)ide_ack_intr(hwif);
1475                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1476                                 startstop = handler(drive);
1477                         } else {
1478                                 if (drive->waiting_for_dma) {
1479                                         startstop = ide_dma_timeout_retry(drive, wait);
1480                                 } else
1481                                         startstop =
1482                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1483                         }
1484                         drive->service_time = jiffies - drive->service_start;
1485                         spin_lock_irq(&ide_lock);
1486                         enable_irq(hwif->irq);
1487                         if (startstop == ide_stopped)
1488                                 hwgroup->busy = 0;
1489                 }
1490         }
1491         ide_do_request(hwgroup, IDE_NO_IRQ);
1492         spin_unlock_irqrestore(&ide_lock, flags);
1493 }
1494
1495 /**
1496  *      unexpected_intr         -       handle an unexpected IDE interrupt
1497  *      @irq: interrupt line
1498  *      @hwgroup: hwgroup being processed
1499  *
1500  *      There's nothing really useful we can do with an unexpected interrupt,
1501  *      other than reading the status register (to clear it), and logging it.
1502  *      There should be no way that an irq can happen before we're ready for it,
1503  *      so we needn't worry much about losing an "important" interrupt here.
1504  *
1505  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1506  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1507  *      looks "good", we just ignore the interrupt completely.
1508  *
1509  *      This routine assumes __cli() is in effect when called.
1510  *
1511  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1512  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1513  *      we could screw up by interfering with a new request being set up for 
1514  *      irq15.
1515  *
1516  *      In reality, this is a non-issue.  The new command is not sent unless 
1517  *      the drive is ready to accept one, in which case we know the drive is
1518  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1519  *      before completing the issuance of any new drive command, so we will not
1520  *      be accidentally invoked as a result of any valid command completion
1521  *      interrupt.
1522  *
1523  *      Note that we must walk the entire hwgroup here. We know which hwif
1524  *      is doing the current command, but we don't know which hwif burped
1525  *      mysteriously.
1526  */
1527  
1528 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1529 {
1530         u8 stat;
1531         ide_hwif_t *hwif = hwgroup->hwif;
1532
1533         /*
1534          * handle the unexpected interrupt
1535          */
1536         do {
1537                 if (hwif->irq == irq) {
1538                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1539                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1540                                 /* Try to not flood the console with msgs */
1541                                 static unsigned long last_msgtime, count;
1542                                 ++count;
1543                                 if (time_after(jiffies, last_msgtime + HZ)) {
1544                                         last_msgtime = jiffies;
1545                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1546                                                 "status=0x%02x, count=%ld\n",
1547                                                 hwif->name,
1548                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1549                                 }
1550                         }
1551                 }
1552         } while ((hwif = hwif->next) != hwgroup->hwif);
1553 }
1554
1555 /**
1556  *      ide_intr        -       default IDE interrupt handler
1557  *      @irq: interrupt number
1558  *      @dev_id: hwif group
1559  *      @regs: unused weirdness from the kernel irq layer
1560  *
1561  *      This is the default IRQ handler for the IDE layer. You should
1562  *      not need to override it. If you do be aware it is subtle in
1563  *      places
1564  *
1565  *      hwgroup->hwif is the interface in the group currently performing
1566  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1567  *      the IRQ handler to call. As we issue a command the handlers
1568  *      step through multiple states, reassigning the handler to the
1569  *      next step in the process. Unlike a smart SCSI controller IDE
1570  *      expects the main processor to sequence the various transfer
1571  *      stages. We also manage a poll timer to catch up with most
1572  *      timeout situations. There are still a few where the handlers
1573  *      don't ever decide to give up.
1574  *
1575  *      The handler eventually returns ide_stopped to indicate the
1576  *      request completed. At this point we issue the next request
1577  *      on the hwgroup and the process begins again.
1578  */
1579  
1580 irqreturn_t ide_intr (int irq, void *dev_id)
1581 {
1582         unsigned long flags;
1583         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1584         ide_hwif_t *hwif;
1585         ide_drive_t *drive;
1586         ide_handler_t *handler;
1587         ide_startstop_t startstop;
1588
1589         spin_lock_irqsave(&ide_lock, flags);
1590         hwif = hwgroup->hwif;
1591
1592         if (!ide_ack_intr(hwif)) {
1593                 spin_unlock_irqrestore(&ide_lock, flags);
1594                 return IRQ_NONE;
1595         }
1596
1597         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1598                 /*
1599                  * Not expecting an interrupt from this drive.
1600                  * That means this could be:
1601                  *      (1) an interrupt from another PCI device
1602                  *      sharing the same PCI INT# as us.
1603                  * or   (2) a drive just entered sleep or standby mode,
1604                  *      and is interrupting to let us know.
1605                  * or   (3) a spurious interrupt of unknown origin.
1606                  *
1607                  * For PCI, we cannot tell the difference,
1608                  * so in that case we just ignore it and hope it goes away.
1609                  *
1610                  * FIXME: unexpected_intr should be hwif-> then we can
1611                  * remove all the ifdef PCI crap
1612                  */
1613 #ifdef CONFIG_BLK_DEV_IDEPCI
1614                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1615 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1616                 {
1617                         /*
1618                          * Probably not a shared PCI interrupt,
1619                          * so we can safely try to do something about it:
1620                          */
1621                         unexpected_intr(irq, hwgroup);
1622 #ifdef CONFIG_BLK_DEV_IDEPCI
1623                 } else {
1624                         /*
1625                          * Whack the status register, just in case
1626                          * we have a leftover pending IRQ.
1627                          */
1628                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1629 #endif /* CONFIG_BLK_DEV_IDEPCI */
1630                 }
1631                 spin_unlock_irqrestore(&ide_lock, flags);
1632                 return IRQ_NONE;
1633         }
1634         drive = hwgroup->drive;
1635         if (!drive) {
1636                 /*
1637                  * This should NEVER happen, and there isn't much
1638                  * we could do about it here.
1639                  *
1640                  * [Note - this can occur if the drive is hot unplugged]
1641                  */
1642                 spin_unlock_irqrestore(&ide_lock, flags);
1643                 return IRQ_HANDLED;
1644         }
1645         if (!drive_is_ready(drive)) {
1646                 /*
1647                  * This happens regularly when we share a PCI IRQ with
1648                  * another device.  Unfortunately, it can also happen
1649                  * with some buggy drives that trigger the IRQ before
1650                  * their status register is up to date.  Hopefully we have
1651                  * enough advance overhead that the latter isn't a problem.
1652                  */
1653                 spin_unlock_irqrestore(&ide_lock, flags);
1654                 return IRQ_NONE;
1655         }
1656         if (!hwgroup->busy) {
1657                 hwgroup->busy = 1;      /* paranoia */
1658                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1659         }
1660         hwgroup->handler = NULL;
1661         hwgroup->req_gen++;
1662         del_timer(&hwgroup->timer);
1663         spin_unlock(&ide_lock);
1664
1665         /* Some controllers might set DMA INTR no matter DMA or PIO;
1666          * bmdma status might need to be cleared even for
1667          * PIO interrupts to prevent spurious/lost irq.
1668          */
1669         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1670                 /* ide_dma_end() needs bmdma status for error checking.
1671                  * So, skip clearing bmdma status here and leave it
1672                  * to ide_dma_end() if this is dma interrupt.
1673                  */
1674                 hwif->ide_dma_clear_irq(drive);
1675
1676         if (drive->unmask)
1677                 local_irq_enable_in_hardirq();
1678         /* service this interrupt, may set handler for next interrupt */
1679         startstop = handler(drive);
1680         spin_lock_irq(&ide_lock);
1681
1682         /*
1683          * Note that handler() may have set things up for another
1684          * interrupt to occur soon, but it cannot happen until
1685          * we exit from this routine, because it will be the
1686          * same irq as is currently being serviced here, and Linux
1687          * won't allow another of the same (on any CPU) until we return.
1688          */
1689         drive->service_time = jiffies - drive->service_start;
1690         if (startstop == ide_stopped) {
1691                 if (hwgroup->handler == NULL) { /* paranoia */
1692                         hwgroup->busy = 0;
1693                         ide_do_request(hwgroup, hwif->irq);
1694                 } else {
1695                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1696                                 "on exit\n", drive->name);
1697                 }
1698         }
1699         spin_unlock_irqrestore(&ide_lock, flags);
1700         return IRQ_HANDLED;
1701 }
1702
1703 /**
1704  *      ide_init_drive_cmd      -       initialize a drive command request
1705  *      @rq: request object
1706  *
1707  *      Initialize a request before we fill it in and send it down to
1708  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1709  *      now it doesn't do a lot, but if that changes abusers will have a
1710  *      nasty surprise.
1711  */
1712
1713 void ide_init_drive_cmd (struct request *rq)
1714 {
1715         memset(rq, 0, sizeof(*rq));
1716         rq->cmd_type = REQ_TYPE_ATA_CMD;
1717         rq->ref_count = 1;
1718 }
1719
1720 EXPORT_SYMBOL(ide_init_drive_cmd);
1721
1722 /**
1723  *      ide_do_drive_cmd        -       issue IDE special command
1724  *      @drive: device to issue command
1725  *      @rq: request to issue
1726  *      @action: action for processing
1727  *
1728  *      This function issues a special IDE device request
1729  *      onto the request queue.
1730  *
1731  *      If action is ide_wait, then the rq is queued at the end of the
1732  *      request queue, and the function sleeps until it has been processed.
1733  *      This is for use when invoked from an ioctl handler.
1734  *
1735  *      If action is ide_preempt, then the rq is queued at the head of
1736  *      the request queue, displacing the currently-being-processed
1737  *      request and this function returns immediately without waiting
1738  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1739  *      intended for careful use by the ATAPI tape/cdrom driver code.
1740  *
1741  *      If action is ide_end, then the rq is queued at the end of the
1742  *      request queue, and the function returns immediately without waiting
1743  *      for the new rq to be completed. This is again intended for careful
1744  *      use by the ATAPI tape/cdrom driver code.
1745  */
1746  
1747 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1748 {
1749         unsigned long flags;
1750         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1751         DECLARE_COMPLETION_ONSTACK(wait);
1752         int where = ELEVATOR_INSERT_BACK, err;
1753         int must_wait = (action == ide_wait || action == ide_head_wait);
1754
1755         rq->errors = 0;
1756
1757         /*
1758          * we need to hold an extra reference to request for safe inspection
1759          * after completion
1760          */
1761         if (must_wait) {
1762                 rq->ref_count++;
1763                 rq->end_io_data = &wait;
1764                 rq->end_io = blk_end_sync_rq;
1765         }
1766
1767         spin_lock_irqsave(&ide_lock, flags);
1768         if (action == ide_preempt)
1769                 hwgroup->rq = NULL;
1770         if (action == ide_preempt || action == ide_head_wait) {
1771                 where = ELEVATOR_INSERT_FRONT;
1772                 rq->cmd_flags |= REQ_PREEMPT;
1773         }
1774         __elv_add_request(drive->queue, rq, where, 0);
1775         ide_do_request(hwgroup, IDE_NO_IRQ);
1776         spin_unlock_irqrestore(&ide_lock, flags);
1777
1778         err = 0;
1779         if (must_wait) {
1780                 wait_for_completion(&wait);
1781                 if (rq->errors)
1782                         err = -EIO;
1783
1784                 blk_put_request(rq);
1785         }
1786
1787         return err;
1788 }
1789
1790 EXPORT_SYMBOL(ide_do_drive_cmd);