[BLOCK] add @uptodate to end_that_request_last() and @error to rq_end_io_fn()
[linux-2.6.git] / drivers / ide / ide-io.c
1 /*
2  *      IDE I/O functions
3  *
4  *      Basic PIO and command management functionality.
5  *
6  * This code was split off from ide.c. See ide.c for history and original
7  * copyrights.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the
11  * Free Software Foundation; either version 2, or (at your option) any
12  * later version.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * For the avoidance of doubt the "preferred form" of this code is one which
20  * is in an open non patent encumbered format. Where cryptographic key signing
21  * forms part of the process of creating an executable the information
22  * including keys needed to generate an equivalently functional executable
23  * are deemed to be part of the source code.
24  */
25  
26  
27 #include <linux/config.h>
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/string.h>
31 #include <linux/kernel.h>
32 #include <linux/timer.h>
33 #include <linux/mm.h>
34 #include <linux/interrupt.h>
35 #include <linux/major.h>
36 #include <linux/errno.h>
37 #include <linux/genhd.h>
38 #include <linux/blkpg.h>
39 #include <linux/slab.h>
40 #include <linux/init.h>
41 #include <linux/pci.h>
42 #include <linux/delay.h>
43 #include <linux/ide.h>
44 #include <linux/completion.h>
45 #include <linux/reboot.h>
46 #include <linux/cdrom.h>
47 #include <linux/seq_file.h>
48 #include <linux/device.h>
49 #include <linux/kmod.h>
50 #include <linux/scatterlist.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/irq.h>
54 #include <asm/uaccess.h>
55 #include <asm/io.h>
56 #include <asm/bitops.h>
57
58 int __ide_end_request(ide_drive_t *drive, struct request *rq, int uptodate,
59                       int nr_sectors)
60 {
61         int ret = 1;
62
63         BUG_ON(!(rq->flags & REQ_STARTED));
64
65         /*
66          * if failfast is set on a request, override number of sectors and
67          * complete the whole request right now
68          */
69         if (blk_noretry_request(rq) && end_io_error(uptodate))
70                 nr_sectors = rq->hard_nr_sectors;
71
72         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
73                 rq->errors = -EIO;
74
75         /*
76          * decide whether to reenable DMA -- 3 is a random magic for now,
77          * if we DMA timeout more than 3 times, just stay in PIO
78          */
79         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
80                 drive->state = 0;
81                 HWGROUP(drive)->hwif->ide_dma_on(drive);
82         }
83
84         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
85                 add_disk_randomness(rq->rq_disk);
86
87                 if (blk_rq_tagged(rq))
88                         blk_queue_end_tag(drive->queue, rq);
89
90                 blkdev_dequeue_request(rq);
91                 HWGROUP(drive)->rq = NULL;
92                 end_that_request_last(rq, uptodate);
93                 ret = 0;
94         }
95         return ret;
96 }
97 EXPORT_SYMBOL(__ide_end_request);
98
99 /**
100  *      ide_end_request         -       complete an IDE I/O
101  *      @drive: IDE device for the I/O
102  *      @uptodate:
103  *      @nr_sectors: number of sectors completed
104  *
105  *      This is our end_request wrapper function. We complete the I/O
106  *      update random number input and dequeue the request, which if
107  *      it was tagged may be out of order.
108  */
109
110 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
111 {
112         struct request *rq;
113         unsigned long flags;
114         int ret = 1;
115
116         spin_lock_irqsave(&ide_lock, flags);
117         rq = HWGROUP(drive)->rq;
118
119         if (!nr_sectors)
120                 nr_sectors = rq->hard_cur_sectors;
121
122         if (blk_complete_barrier_rq_locked(drive->queue, rq, nr_sectors))
123                 ret = rq->nr_sectors != 0;
124         else
125                 ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
126
127         spin_unlock_irqrestore(&ide_lock, flags);
128         return ret;
129 }
130 EXPORT_SYMBOL(ide_end_request);
131
132 /*
133  * Power Management state machine. This one is rather trivial for now,
134  * we should probably add more, like switching back to PIO on suspend
135  * to help some BIOSes, re-do the door locking on resume, etc...
136  */
137
138 enum {
139         ide_pm_flush_cache      = ide_pm_state_start_suspend,
140         idedisk_pm_standby,
141
142         idedisk_pm_idle         = ide_pm_state_start_resume,
143         ide_pm_restore_dma,
144 };
145
146 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
147 {
148         if (drive->media != ide_disk)
149                 return;
150
151         switch (rq->pm->pm_step) {
152         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
153                 if (rq->pm->pm_state == PM_EVENT_FREEZE)
154                         rq->pm->pm_step = ide_pm_state_completed;
155                 else
156                         rq->pm->pm_step = idedisk_pm_standby;
157                 break;
158         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
159                 rq->pm->pm_step = ide_pm_state_completed;
160                 break;
161         case idedisk_pm_idle:           /* Resume step 1 (idle) complete */
162                 rq->pm->pm_step = ide_pm_restore_dma;
163                 break;
164         }
165 }
166
167 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
168 {
169         ide_task_t *args = rq->special;
170
171         memset(args, 0, sizeof(*args));
172
173         if (drive->media != ide_disk) {
174                 /* skip idedisk_pm_idle for ATAPI devices */
175                 if (rq->pm->pm_step == idedisk_pm_idle)
176                         rq->pm->pm_step = ide_pm_restore_dma;
177         }
178
179         switch (rq->pm->pm_step) {
180         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
181                 if (drive->media != ide_disk)
182                         break;
183                 /* Not supported? Switch to next step now. */
184                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
185                         ide_complete_power_step(drive, rq, 0, 0);
186                         return ide_stopped;
187                 }
188                 if (ide_id_has_flush_cache_ext(drive->id))
189                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
190                 else
191                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
192                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
193                 args->handler      = &task_no_data_intr;
194                 return do_rw_taskfile(drive, args);
195
196         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
197                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
198                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
199                 args->handler      = &task_no_data_intr;
200                 return do_rw_taskfile(drive, args);
201
202         case idedisk_pm_idle:           /* Resume step 1 (idle) */
203                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
204                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
205                 args->handler = task_no_data_intr;
206                 return do_rw_taskfile(drive, args);
207
208         case ide_pm_restore_dma:        /* Resume step 2 (restore DMA) */
209                 /*
210                  * Right now, all we do is call hwif->ide_dma_check(drive),
211                  * we could be smarter and check for current xfer_speed
212                  * in struct drive etc...
213                  */
214                 if ((drive->id->capability & 1) == 0)
215                         break;
216                 if (drive->hwif->ide_dma_check == NULL)
217                         break;
218                 drive->hwif->ide_dma_check(drive);
219                 break;
220         }
221         rq->pm->pm_step = ide_pm_state_completed;
222         return ide_stopped;
223 }
224
225 /**
226  *      ide_complete_pm_request - end the current Power Management request
227  *      @drive: target drive
228  *      @rq: request
229  *
230  *      This function cleans up the current PM request and stops the queue
231  *      if necessary.
232  */
233 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
234 {
235         unsigned long flags;
236
237 #ifdef DEBUG_PM
238         printk("%s: completing PM request, %s\n", drive->name,
239                blk_pm_suspend_request(rq) ? "suspend" : "resume");
240 #endif
241         spin_lock_irqsave(&ide_lock, flags);
242         if (blk_pm_suspend_request(rq)) {
243                 blk_stop_queue(drive->queue);
244         } else {
245                 drive->blocked = 0;
246                 blk_start_queue(drive->queue);
247         }
248         blkdev_dequeue_request(rq);
249         HWGROUP(drive)->rq = NULL;
250         end_that_request_last(rq, 1);
251         spin_unlock_irqrestore(&ide_lock, flags);
252 }
253
254 /*
255  * FIXME: probably move this somewhere else, name is bad too :)
256  */
257 u64 ide_get_error_location(ide_drive_t *drive, char *args)
258 {
259         u32 high, low;
260         u8 hcyl, lcyl, sect;
261         u64 sector;
262
263         high = 0;
264         hcyl = args[5];
265         lcyl = args[4];
266         sect = args[3];
267
268         if (ide_id_has_flush_cache_ext(drive->id)) {
269                 low = (hcyl << 16) | (lcyl << 8) | sect;
270                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
271                 high = ide_read_24(drive);
272         } else {
273                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
274                 if (cur & 0x40) {
275                         high = cur & 0xf;
276                         low = (hcyl << 16) | (lcyl << 8) | sect;
277                 } else {
278                         low = hcyl * drive->head * drive->sect;
279                         low += lcyl * drive->sect;
280                         low += sect - 1;
281                 }
282         }
283
284         sector = ((u64) high << 24) | low;
285         return sector;
286 }
287 EXPORT_SYMBOL(ide_get_error_location);
288
289 /**
290  *      ide_end_drive_cmd       -       end an explicit drive command
291  *      @drive: command 
292  *      @stat: status bits
293  *      @err: error bits
294  *
295  *      Clean up after success/failure of an explicit drive command.
296  *      These get thrown onto the queue so they are synchronized with
297  *      real I/O operations on the drive.
298  *
299  *      In LBA48 mode we have to read the register set twice to get
300  *      all the extra information out.
301  */
302  
303 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
304 {
305         ide_hwif_t *hwif = HWIF(drive);
306         unsigned long flags;
307         struct request *rq;
308
309         spin_lock_irqsave(&ide_lock, flags);
310         rq = HWGROUP(drive)->rq;
311         spin_unlock_irqrestore(&ide_lock, flags);
312
313         if (rq->flags & REQ_DRIVE_CMD) {
314                 u8 *args = (u8 *) rq->buffer;
315                 if (rq->errors == 0)
316                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
317
318                 if (args) {
319                         args[0] = stat;
320                         args[1] = err;
321                         args[2] = hwif->INB(IDE_NSECTOR_REG);
322                 }
323         } else if (rq->flags & REQ_DRIVE_TASK) {
324                 u8 *args = (u8 *) rq->buffer;
325                 if (rq->errors == 0)
326                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
327
328                 if (args) {
329                         args[0] = stat;
330                         args[1] = err;
331                         args[2] = hwif->INB(IDE_NSECTOR_REG);
332                         args[3] = hwif->INB(IDE_SECTOR_REG);
333                         args[4] = hwif->INB(IDE_LCYL_REG);
334                         args[5] = hwif->INB(IDE_HCYL_REG);
335                         args[6] = hwif->INB(IDE_SELECT_REG);
336                 }
337         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
338                 ide_task_t *args = (ide_task_t *) rq->special;
339                 if (rq->errors == 0)
340                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
341                         
342                 if (args) {
343                         if (args->tf_in_flags.b.data) {
344                                 u16 data                                = hwif->INW(IDE_DATA_REG);
345                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
346                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
347                         }
348                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
349                         /* be sure we're looking at the low order bits */
350                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
351                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
352                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
353                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
354                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
355                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
356                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
357
358                         if (drive->addressing == 1) {
359                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
360                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
361                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
362                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
363                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
364                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
365                         }
366                 }
367         } else if (blk_pm_request(rq)) {
368 #ifdef DEBUG_PM
369                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
370                         drive->name, rq->pm->pm_step, stat, err);
371 #endif
372                 ide_complete_power_step(drive, rq, stat, err);
373                 if (rq->pm->pm_step == ide_pm_state_completed)
374                         ide_complete_pm_request(drive, rq);
375                 return;
376         }
377
378         spin_lock_irqsave(&ide_lock, flags);
379         blkdev_dequeue_request(rq);
380         HWGROUP(drive)->rq = NULL;
381         rq->errors = err;
382         end_that_request_last(rq, !rq->errors);
383         spin_unlock_irqrestore(&ide_lock, flags);
384 }
385
386 EXPORT_SYMBOL(ide_end_drive_cmd);
387
388 /**
389  *      try_to_flush_leftover_data      -       flush junk
390  *      @drive: drive to flush
391  *
392  *      try_to_flush_leftover_data() is invoked in response to a drive
393  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
394  *      resetting the drive, this routine tries to clear the condition
395  *      by read a sector's worth of data from the drive.  Of course,
396  *      this may not help if the drive is *waiting* for data from *us*.
397  */
398 static void try_to_flush_leftover_data (ide_drive_t *drive)
399 {
400         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
401
402         if (drive->media != ide_disk)
403                 return;
404         while (i > 0) {
405                 u32 buffer[16];
406                 u32 wcount = (i > 16) ? 16 : i;
407
408                 i -= wcount;
409                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
410         }
411 }
412
413 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
414 {
415         if (rq->rq_disk) {
416                 ide_driver_t *drv;
417
418                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
419                 drv->end_request(drive, 0, 0);
420         } else
421                 ide_end_request(drive, 0, 0);
422 }
423
424 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
425 {
426         ide_hwif_t *hwif = drive->hwif;
427
428         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
429                 /* other bits are useless when BUSY */
430                 rq->errors |= ERROR_RESET;
431         } else if (stat & ERR_STAT) {
432                 /* err has different meaning on cdrom and tape */
433                 if (err == ABRT_ERR) {
434                         if (drive->select.b.lba &&
435                             /* some newer drives don't support WIN_SPECIFY */
436                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
437                                 return ide_stopped;
438                 } else if ((err & BAD_CRC) == BAD_CRC) {
439                         /* UDMA crc error, just retry the operation */
440                         drive->crc_count++;
441                 } else if (err & (BBD_ERR | ECC_ERR)) {
442                         /* retries won't help these */
443                         rq->errors = ERROR_MAX;
444                 } else if (err & TRK0_ERR) {
445                         /* help it find track zero */
446                         rq->errors |= ERROR_RECAL;
447                 }
448         }
449
450         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ)
451                 try_to_flush_leftover_data(drive);
452
453         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
454                 /* force an abort */
455                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
456
457         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
458                 ide_kill_rq(drive, rq);
459         else {
460                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
461                         ++rq->errors;
462                         return ide_do_reset(drive);
463                 }
464                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
465                         drive->special.b.recalibrate = 1;
466                 ++rq->errors;
467         }
468         return ide_stopped;
469 }
470
471 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
472 {
473         ide_hwif_t *hwif = drive->hwif;
474
475         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
476                 /* other bits are useless when BUSY */
477                 rq->errors |= ERROR_RESET;
478         } else {
479                 /* add decoding error stuff */
480         }
481
482         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
483                 /* force an abort */
484                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
485
486         if (rq->errors >= ERROR_MAX) {
487                 ide_kill_rq(drive, rq);
488         } else {
489                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
490                         ++rq->errors;
491                         return ide_do_reset(drive);
492                 }
493                 ++rq->errors;
494         }
495
496         return ide_stopped;
497 }
498
499 ide_startstop_t
500 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
501 {
502         if (drive->media == ide_disk)
503                 return ide_ata_error(drive, rq, stat, err);
504         return ide_atapi_error(drive, rq, stat, err);
505 }
506
507 EXPORT_SYMBOL_GPL(__ide_error);
508
509 /**
510  *      ide_error       -       handle an error on the IDE
511  *      @drive: drive the error occurred on
512  *      @msg: message to report
513  *      @stat: status bits
514  *
515  *      ide_error() takes action based on the error returned by the drive.
516  *      For normal I/O that may well include retries. We deal with
517  *      both new-style (taskfile) and old style command handling here.
518  *      In the case of taskfile command handling there is work left to
519  *      do
520  */
521  
522 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
523 {
524         struct request *rq;
525         u8 err;
526
527         err = ide_dump_status(drive, msg, stat);
528
529         if ((rq = HWGROUP(drive)->rq) == NULL)
530                 return ide_stopped;
531
532         /* retry only "normal" I/O: */
533         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
534                 rq->errors = 1;
535                 ide_end_drive_cmd(drive, stat, err);
536                 return ide_stopped;
537         }
538
539         if (rq->rq_disk) {
540                 ide_driver_t *drv;
541
542                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
543                 return drv->error(drive, rq, stat, err);
544         } else
545                 return __ide_error(drive, rq, stat, err);
546 }
547
548 EXPORT_SYMBOL_GPL(ide_error);
549
550 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
551 {
552         if (drive->media != ide_disk)
553                 rq->errors |= ERROR_RESET;
554
555         ide_kill_rq(drive, rq);
556
557         return ide_stopped;
558 }
559
560 EXPORT_SYMBOL_GPL(__ide_abort);
561
562 /**
563  *      ide_abort       -       abort pending IDE operations
564  *      @drive: drive the error occurred on
565  *      @msg: message to report
566  *
567  *      ide_abort kills and cleans up when we are about to do a 
568  *      host initiated reset on active commands. Longer term we
569  *      want handlers to have sensible abort handling themselves
570  *
571  *      This differs fundamentally from ide_error because in 
572  *      this case the command is doing just fine when we
573  *      blow it away.
574  */
575  
576 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
577 {
578         struct request *rq;
579
580         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
581                 return ide_stopped;
582
583         /* retry only "normal" I/O: */
584         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
585                 rq->errors = 1;
586                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
587                 return ide_stopped;
588         }
589
590         if (rq->rq_disk) {
591                 ide_driver_t *drv;
592
593                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
594                 return drv->abort(drive, rq);
595         } else
596                 return __ide_abort(drive, rq);
597 }
598
599 /**
600  *      ide_cmd         -       issue a simple drive command
601  *      @drive: drive the command is for
602  *      @cmd: command byte
603  *      @nsect: sector byte
604  *      @handler: handler for the command completion
605  *
606  *      Issue a simple drive command with interrupts.
607  *      The drive must be selected beforehand.
608  */
609
610 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
611                 ide_handler_t *handler)
612 {
613         ide_hwif_t *hwif = HWIF(drive);
614         if (IDE_CONTROL_REG)
615                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
616         SELECT_MASK(drive,0);
617         hwif->OUTB(nsect,IDE_NSECTOR_REG);
618         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
619 }
620
621 /**
622  *      drive_cmd_intr          -       drive command completion interrupt
623  *      @drive: drive the completion interrupt occurred on
624  *
625  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
626  *      We do any necessary data reading and then wait for the drive to
627  *      go non busy. At that point we may read the error data and complete
628  *      the request
629  */
630  
631 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
632 {
633         struct request *rq = HWGROUP(drive)->rq;
634         ide_hwif_t *hwif = HWIF(drive);
635         u8 *args = (u8 *) rq->buffer;
636         u8 stat = hwif->INB(IDE_STATUS_REG);
637         int retries = 10;
638
639         local_irq_enable();
640         if ((stat & DRQ_STAT) && args && args[3]) {
641                 u8 io_32bit = drive->io_32bit;
642                 drive->io_32bit = 0;
643                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
644                 drive->io_32bit = io_32bit;
645                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
646                         udelay(100);
647         }
648
649         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
650                 return ide_error(drive, "drive_cmd", stat);
651                 /* calls ide_end_drive_cmd */
652         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
653         return ide_stopped;
654 }
655
656 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
657 {
658         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
659         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
660         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
661         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
662         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
663         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
664
665         task->handler = &set_geometry_intr;
666 }
667
668 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
669 {
670         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
671         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
672
673         task->handler = &recal_intr;
674 }
675
676 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
677 {
678         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
679         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
680
681         task->handler = &set_multmode_intr;
682 }
683
684 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
685 {
686         special_t *s = &drive->special;
687         ide_task_t args;
688
689         memset(&args, 0, sizeof(ide_task_t));
690         args.command_type = IDE_DRIVE_TASK_NO_DATA;
691
692         if (s->b.set_geometry) {
693                 s->b.set_geometry = 0;
694                 ide_init_specify_cmd(drive, &args);
695         } else if (s->b.recalibrate) {
696                 s->b.recalibrate = 0;
697                 ide_init_restore_cmd(drive, &args);
698         } else if (s->b.set_multmode) {
699                 s->b.set_multmode = 0;
700                 if (drive->mult_req > drive->id->max_multsect)
701                         drive->mult_req = drive->id->max_multsect;
702                 ide_init_setmult_cmd(drive, &args);
703         } else if (s->all) {
704                 int special = s->all;
705                 s->all = 0;
706                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
707                 return ide_stopped;
708         }
709
710         do_rw_taskfile(drive, &args);
711
712         return ide_started;
713 }
714
715 /**
716  *      do_special              -       issue some special commands
717  *      @drive: drive the command is for
718  *
719  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
720  *      commands to a drive.  It used to do much more, but has been scaled
721  *      back.
722  */
723
724 static ide_startstop_t do_special (ide_drive_t *drive)
725 {
726         special_t *s = &drive->special;
727
728 #ifdef DEBUG
729         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
730 #endif
731         if (s->b.set_tune) {
732                 s->b.set_tune = 0;
733                 if (HWIF(drive)->tuneproc != NULL)
734                         HWIF(drive)->tuneproc(drive, drive->tune_req);
735                 return ide_stopped;
736         } else {
737                 if (drive->media == ide_disk)
738                         return ide_disk_special(drive);
739
740                 s->all = 0;
741                 drive->mult_req = 0;
742                 return ide_stopped;
743         }
744 }
745
746 void ide_map_sg(ide_drive_t *drive, struct request *rq)
747 {
748         ide_hwif_t *hwif = drive->hwif;
749         struct scatterlist *sg = hwif->sg_table;
750
751         if (hwif->sg_mapped)    /* needed by ide-scsi */
752                 return;
753
754         if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
755                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
756         } else {
757                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
758                 hwif->sg_nents = 1;
759         }
760 }
761
762 EXPORT_SYMBOL_GPL(ide_map_sg);
763
764 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
765 {
766         ide_hwif_t *hwif = drive->hwif;
767
768         hwif->nsect = hwif->nleft = rq->nr_sectors;
769         hwif->cursg = hwif->cursg_ofs = 0;
770 }
771
772 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
773
774 /**
775  *      execute_drive_command   -       issue special drive command
776  *      @drive: the drive to issue the command on
777  *      @rq: the request structure holding the command
778  *
779  *      execute_drive_cmd() issues a special drive command,  usually 
780  *      initiated by ioctl() from the external hdparm program. The
781  *      command can be a drive command, drive task or taskfile 
782  *      operation. Weirdly you can call it with NULL to wait for
783  *      all commands to finish. Don't do this as that is due to change
784  */
785
786 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
787                 struct request *rq)
788 {
789         ide_hwif_t *hwif = HWIF(drive);
790         if (rq->flags & REQ_DRIVE_TASKFILE) {
791                 ide_task_t *args = rq->special;
792  
793                 if (!args)
794                         goto done;
795
796                 hwif->data_phase = args->data_phase;
797
798                 switch (hwif->data_phase) {
799                 case TASKFILE_MULTI_OUT:
800                 case TASKFILE_OUT:
801                 case TASKFILE_MULTI_IN:
802                 case TASKFILE_IN:
803                         ide_init_sg_cmd(drive, rq);
804                         ide_map_sg(drive, rq);
805                 default:
806                         break;
807                 }
808
809                 if (args->tf_out_flags.all != 0) 
810                         return flagged_taskfile(drive, args);
811                 return do_rw_taskfile(drive, args);
812         } else if (rq->flags & REQ_DRIVE_TASK) {
813                 u8 *args = rq->buffer;
814                 u8 sel;
815  
816                 if (!args)
817                         goto done;
818 #ifdef DEBUG
819                 printk("%s: DRIVE_TASK_CMD ", drive->name);
820                 printk("cmd=0x%02x ", args[0]);
821                 printk("fr=0x%02x ", args[1]);
822                 printk("ns=0x%02x ", args[2]);
823                 printk("sc=0x%02x ", args[3]);
824                 printk("lcyl=0x%02x ", args[4]);
825                 printk("hcyl=0x%02x ", args[5]);
826                 printk("sel=0x%02x\n", args[6]);
827 #endif
828                 hwif->OUTB(args[1], IDE_FEATURE_REG);
829                 hwif->OUTB(args[3], IDE_SECTOR_REG);
830                 hwif->OUTB(args[4], IDE_LCYL_REG);
831                 hwif->OUTB(args[5], IDE_HCYL_REG);
832                 sel = (args[6] & ~0x10);
833                 if (drive->select.b.unit)
834                         sel |= 0x10;
835                 hwif->OUTB(sel, IDE_SELECT_REG);
836                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
837                 return ide_started;
838         } else if (rq->flags & REQ_DRIVE_CMD) {
839                 u8 *args = rq->buffer;
840
841                 if (!args)
842                         goto done;
843 #ifdef DEBUG
844                 printk("%s: DRIVE_CMD ", drive->name);
845                 printk("cmd=0x%02x ", args[0]);
846                 printk("sc=0x%02x ", args[1]);
847                 printk("fr=0x%02x ", args[2]);
848                 printk("xx=0x%02x\n", args[3]);
849 #endif
850                 if (args[0] == WIN_SMART) {
851                         hwif->OUTB(0x4f, IDE_LCYL_REG);
852                         hwif->OUTB(0xc2, IDE_HCYL_REG);
853                         hwif->OUTB(args[2],IDE_FEATURE_REG);
854                         hwif->OUTB(args[1],IDE_SECTOR_REG);
855                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
856                         return ide_started;
857                 }
858                 hwif->OUTB(args[2],IDE_FEATURE_REG);
859                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
860                 return ide_started;
861         }
862
863 done:
864         /*
865          * NULL is actually a valid way of waiting for
866          * all current requests to be flushed from the queue.
867          */
868 #ifdef DEBUG
869         printk("%s: DRIVE_CMD (null)\n", drive->name);
870 #endif
871         ide_end_drive_cmd(drive,
872                         hwif->INB(IDE_STATUS_REG),
873                         hwif->INB(IDE_ERROR_REG));
874         return ide_stopped;
875 }
876
877 /**
878  *      start_request   -       start of I/O and command issuing for IDE
879  *
880  *      start_request() initiates handling of a new I/O request. It
881  *      accepts commands and I/O (read/write) requests. It also does
882  *      the final remapping for weird stuff like EZDrive. Once 
883  *      device mapper can work sector level the EZDrive stuff can go away
884  *
885  *      FIXME: this function needs a rename
886  */
887  
888 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
889 {
890         ide_startstop_t startstop;
891         sector_t block;
892
893         BUG_ON(!(rq->flags & REQ_STARTED));
894
895 #ifdef DEBUG
896         printk("%s: start_request: current=0x%08lx\n",
897                 HWIF(drive)->name, (unsigned long) rq);
898 #endif
899
900         /* bail early if we've exceeded max_failures */
901         if (drive->max_failures && (drive->failures > drive->max_failures)) {
902                 goto kill_rq;
903         }
904
905         block    = rq->sector;
906         if (blk_fs_request(rq) &&
907             (drive->media == ide_disk || drive->media == ide_floppy)) {
908                 block += drive->sect0;
909         }
910         /* Yecch - this will shift the entire interval,
911            possibly killing some innocent following sector */
912         if (block == 0 && drive->remap_0_to_1 == 1)
913                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
914
915         if (blk_pm_suspend_request(rq) &&
916             rq->pm->pm_step == ide_pm_state_start_suspend)
917                 /* Mark drive blocked when starting the suspend sequence. */
918                 drive->blocked = 1;
919         else if (blk_pm_resume_request(rq) &&
920                  rq->pm->pm_step == ide_pm_state_start_resume) {
921                 /* 
922                  * The first thing we do on wakeup is to wait for BSY bit to
923                  * go away (with a looong timeout) as a drive on this hwif may
924                  * just be POSTing itself.
925                  * We do that before even selecting as the "other" device on
926                  * the bus may be broken enough to walk on our toes at this
927                  * point.
928                  */
929                 int rc;
930 #ifdef DEBUG_PM
931                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
932 #endif
933                 rc = ide_wait_not_busy(HWIF(drive), 35000);
934                 if (rc)
935                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
936                 SELECT_DRIVE(drive);
937                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
938                 rc = ide_wait_not_busy(HWIF(drive), 10000);
939                 if (rc)
940                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
941         }
942
943         SELECT_DRIVE(drive);
944         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
945                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
946                 return startstop;
947         }
948         if (!drive->special.all) {
949                 ide_driver_t *drv;
950
951                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
952                         return execute_drive_cmd(drive, rq);
953                 else if (rq->flags & REQ_DRIVE_TASKFILE)
954                         return execute_drive_cmd(drive, rq);
955                 else if (blk_pm_request(rq)) {
956 #ifdef DEBUG_PM
957                         printk("%s: start_power_step(step: %d)\n",
958                                 drive->name, rq->pm->pm_step);
959 #endif
960                         startstop = ide_start_power_step(drive, rq);
961                         if (startstop == ide_stopped &&
962                             rq->pm->pm_step == ide_pm_state_completed)
963                                 ide_complete_pm_request(drive, rq);
964                         return startstop;
965                 }
966
967                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
968                 return drv->do_request(drive, rq, block);
969         }
970         return do_special(drive);
971 kill_rq:
972         ide_kill_rq(drive, rq);
973         return ide_stopped;
974 }
975
976 /**
977  *      ide_stall_queue         -       pause an IDE device
978  *      @drive: drive to stall
979  *      @timeout: time to stall for (jiffies)
980  *
981  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
982  *      to the hwgroup by sleeping for timeout jiffies.
983  */
984  
985 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
986 {
987         if (timeout > WAIT_WORSTCASE)
988                 timeout = WAIT_WORSTCASE;
989         drive->sleep = timeout + jiffies;
990         drive->sleeping = 1;
991 }
992
993 EXPORT_SYMBOL(ide_stall_queue);
994
995 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
996
997 /**
998  *      choose_drive            -       select a drive to service
999  *      @hwgroup: hardware group to select on
1000  *
1001  *      choose_drive() selects the next drive which will be serviced.
1002  *      This is necessary because the IDE layer can't issue commands
1003  *      to both drives on the same cable, unlike SCSI.
1004  */
1005  
1006 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1007 {
1008         ide_drive_t *drive, *best;
1009
1010 repeat: 
1011         best = NULL;
1012         drive = hwgroup->drive;
1013
1014         /*
1015          * drive is doing pre-flush, ordered write, post-flush sequence. even
1016          * though that is 3 requests, it must be seen as a single transaction.
1017          * we must not preempt this drive until that is complete
1018          */
1019         if (blk_queue_flushing(drive->queue)) {
1020                 /*
1021                  * small race where queue could get replugged during
1022                  * the 3-request flush cycle, just yank the plug since
1023                  * we want it to finish asap
1024                  */
1025                 blk_remove_plug(drive->queue);
1026                 return drive;
1027         }
1028
1029         do {
1030                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1031                     && !elv_queue_empty(drive->queue)) {
1032                         if (!best
1033                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1034                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1035                         {
1036                                 if (!blk_queue_plugged(drive->queue))
1037                                         best = drive;
1038                         }
1039                 }
1040         } while ((drive = drive->next) != hwgroup->drive);
1041         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1042                 long t = (signed long)(WAKEUP(best) - jiffies);
1043                 if (t >= WAIT_MIN_SLEEP) {
1044                 /*
1045                  * We *may* have some time to spare, but first let's see if
1046                  * someone can potentially benefit from our nice mood today..
1047                  */
1048                         drive = best->next;
1049                         do {
1050                                 if (!drive->sleeping
1051                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1052                                  && time_before(WAKEUP(drive), jiffies + t))
1053                                 {
1054                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1055                                         goto repeat;
1056                                 }
1057                         } while ((drive = drive->next) != best);
1058                 }
1059         }
1060         return best;
1061 }
1062
1063 /*
1064  * Issue a new request to a drive from hwgroup
1065  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1066  *
1067  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1068  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1069  * may have both interfaces in a single hwgroup to "serialize" access.
1070  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1071  * together into one hwgroup for serialized access.
1072  *
1073  * Note also that several hwgroups can end up sharing a single IRQ,
1074  * possibly along with many other devices.  This is especially common in
1075  * PCI-based systems with off-board IDE controller cards.
1076  *
1077  * The IDE driver uses the single global ide_lock spinlock to protect
1078  * access to the request queues, and to protect the hwgroup->busy flag.
1079  *
1080  * The first thread into the driver for a particular hwgroup sets the
1081  * hwgroup->busy flag to indicate that this hwgroup is now active,
1082  * and then initiates processing of the top request from the request queue.
1083  *
1084  * Other threads attempting entry notice the busy setting, and will simply
1085  * queue their new requests and exit immediately.  Note that hwgroup->busy
1086  * remains set even when the driver is merely awaiting the next interrupt.
1087  * Thus, the meaning is "this hwgroup is busy processing a request".
1088  *
1089  * When processing of a request completes, the completing thread or IRQ-handler
1090  * will start the next request from the queue.  If no more work remains,
1091  * the driver will clear the hwgroup->busy flag and exit.
1092  *
1093  * The ide_lock (spinlock) is used to protect all access to the
1094  * hwgroup->busy flag, but is otherwise not needed for most processing in
1095  * the driver.  This makes the driver much more friendlier to shared IRQs
1096  * than previous designs, while remaining 100% (?) SMP safe and capable.
1097  */
1098 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1099 {
1100         ide_drive_t     *drive;
1101         ide_hwif_t      *hwif;
1102         struct request  *rq;
1103         ide_startstop_t startstop;
1104         int             loops = 0;
1105
1106         /* for atari only: POSSIBLY BROKEN HERE(?) */
1107         ide_get_lock(ide_intr, hwgroup);
1108
1109         /* caller must own ide_lock */
1110         BUG_ON(!irqs_disabled());
1111
1112         while (!hwgroup->busy) {
1113                 hwgroup->busy = 1;
1114                 drive = choose_drive(hwgroup);
1115                 if (drive == NULL) {
1116                         int sleeping = 0;
1117                         unsigned long sleep = 0; /* shut up, gcc */
1118                         hwgroup->rq = NULL;
1119                         drive = hwgroup->drive;
1120                         do {
1121                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1122                                         sleeping = 1;
1123                                         sleep = drive->sleep;
1124                                 }
1125                         } while ((drive = drive->next) != hwgroup->drive);
1126                         if (sleeping) {
1127                 /*
1128                  * Take a short snooze, and then wake up this hwgroup again.
1129                  * This gives other hwgroups on the same a chance to
1130                  * play fairly with us, just in case there are big differences
1131                  * in relative throughputs.. don't want to hog the cpu too much.
1132                  */
1133                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1134                                         sleep = jiffies + WAIT_MIN_SLEEP;
1135 #if 1
1136                                 if (timer_pending(&hwgroup->timer))
1137                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1138 #endif
1139                                 /* so that ide_timer_expiry knows what to do */
1140                                 hwgroup->sleeping = 1;
1141                                 mod_timer(&hwgroup->timer, sleep);
1142                                 /* we purposely leave hwgroup->busy==1
1143                                  * while sleeping */
1144                         } else {
1145                                 /* Ugly, but how can we sleep for the lock
1146                                  * otherwise? perhaps from tq_disk?
1147                                  */
1148
1149                                 /* for atari only */
1150                                 ide_release_lock();
1151                                 hwgroup->busy = 0;
1152                         }
1153
1154                         /* no more work for this hwgroup (for now) */
1155                         return;
1156                 }
1157         again:
1158                 hwif = HWIF(drive);
1159                 if (hwgroup->hwif->sharing_irq &&
1160                     hwif != hwgroup->hwif &&
1161                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1162                         /* set nIEN for previous hwif */
1163                         SELECT_INTERRUPT(drive);
1164                 }
1165                 hwgroup->hwif = hwif;
1166                 hwgroup->drive = drive;
1167                 drive->sleeping = 0;
1168                 drive->service_start = jiffies;
1169
1170                 if (blk_queue_plugged(drive->queue)) {
1171                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1172                         break;
1173                 }
1174
1175                 /*
1176                  * we know that the queue isn't empty, but this can happen
1177                  * if the q->prep_rq_fn() decides to kill a request
1178                  */
1179                 rq = elv_next_request(drive->queue);
1180                 if (!rq) {
1181                         hwgroup->busy = 0;
1182                         break;
1183                 }
1184
1185                 /*
1186                  * Sanity: don't accept a request that isn't a PM request
1187                  * if we are currently power managed. This is very important as
1188                  * blk_stop_queue() doesn't prevent the elv_next_request()
1189                  * above to return us whatever is in the queue. Since we call
1190                  * ide_do_request() ourselves, we end up taking requests while
1191                  * the queue is blocked...
1192                  * 
1193                  * We let requests forced at head of queue with ide-preempt
1194                  * though. I hope that doesn't happen too much, hopefully not
1195                  * unless the subdriver triggers such a thing in its own PM
1196                  * state machine.
1197                  *
1198                  * We count how many times we loop here to make sure we service
1199                  * all drives in the hwgroup without looping for ever
1200                  */
1201                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1202                         drive = drive->next ? drive->next : hwgroup->drive;
1203                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1204                                 goto again;
1205                         /* We clear busy, there should be no pending ATA command at this point. */
1206                         hwgroup->busy = 0;
1207                         break;
1208                 }
1209
1210                 hwgroup->rq = rq;
1211
1212                 /*
1213                  * Some systems have trouble with IDE IRQs arriving while
1214                  * the driver is still setting things up.  So, here we disable
1215                  * the IRQ used by this interface while the request is being started.
1216                  * This may look bad at first, but pretty much the same thing
1217                  * happens anyway when any interrupt comes in, IDE or otherwise
1218                  *  -- the kernel masks the IRQ while it is being handled.
1219                  */
1220                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1221                         disable_irq_nosync(hwif->irq);
1222                 spin_unlock(&ide_lock);
1223                 local_irq_enable();
1224                         /* allow other IRQs while we start this request */
1225                 startstop = start_request(drive, rq);
1226                 spin_lock_irq(&ide_lock);
1227                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1228                         enable_irq(hwif->irq);
1229                 if (startstop == ide_stopped)
1230                         hwgroup->busy = 0;
1231         }
1232 }
1233
1234 /*
1235  * Passes the stuff to ide_do_request
1236  */
1237 void do_ide_request(request_queue_t *q)
1238 {
1239         ide_drive_t *drive = q->queuedata;
1240
1241         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1242 }
1243
1244 /*
1245  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1246  * retry the current request in pio mode instead of risking tossing it
1247  * all away
1248  */
1249 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1250 {
1251         ide_hwif_t *hwif = HWIF(drive);
1252         struct request *rq;
1253         ide_startstop_t ret = ide_stopped;
1254
1255         /*
1256          * end current dma transaction
1257          */
1258
1259         if (error < 0) {
1260                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1261                 (void)HWIF(drive)->ide_dma_end(drive);
1262                 ret = ide_error(drive, "dma timeout error",
1263                                                 hwif->INB(IDE_STATUS_REG));
1264         } else {
1265                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1266                 (void) hwif->ide_dma_timeout(drive);
1267         }
1268
1269         /*
1270          * disable dma for now, but remember that we did so because of
1271          * a timeout -- we'll reenable after we finish this next request
1272          * (or rather the first chunk of it) in pio.
1273          */
1274         drive->retry_pio++;
1275         drive->state = DMA_PIO_RETRY;
1276         (void) hwif->ide_dma_off_quietly(drive);
1277
1278         /*
1279          * un-busy drive etc (hwgroup->busy is cleared on return) and
1280          * make sure request is sane
1281          */
1282         rq = HWGROUP(drive)->rq;
1283         HWGROUP(drive)->rq = NULL;
1284
1285         rq->errors = 0;
1286
1287         if (!rq->bio)
1288                 goto out;
1289
1290         rq->sector = rq->bio->bi_sector;
1291         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1292         rq->hard_cur_sectors = rq->current_nr_sectors;
1293         rq->buffer = bio_data(rq->bio);
1294 out:
1295         return ret;
1296 }
1297
1298 /**
1299  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1300  *      @data: timer callback magic (hwgroup)
1301  *
1302  *      An IDE command has timed out before the expected drive return
1303  *      occurred. At this point we attempt to clean up the current
1304  *      mess. If the current handler includes an expiry handler then
1305  *      we invoke the expiry handler, and providing it is happy the
1306  *      work is done. If that fails we apply generic recovery rules
1307  *      invoking the handler and checking the drive DMA status. We
1308  *      have an excessively incestuous relationship with the DMA
1309  *      logic that wants cleaning up.
1310  */
1311  
1312 void ide_timer_expiry (unsigned long data)
1313 {
1314         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1315         ide_handler_t   *handler;
1316         ide_expiry_t    *expiry;
1317         unsigned long   flags;
1318         unsigned long   wait = -1;
1319
1320         spin_lock_irqsave(&ide_lock, flags);
1321
1322         if ((handler = hwgroup->handler) == NULL) {
1323                 /*
1324                  * Either a marginal timeout occurred
1325                  * (got the interrupt just as timer expired),
1326                  * or we were "sleeping" to give other devices a chance.
1327                  * Either way, we don't really want to complain about anything.
1328                  */
1329                 if (hwgroup->sleeping) {
1330                         hwgroup->sleeping = 0;
1331                         hwgroup->busy = 0;
1332                 }
1333         } else {
1334                 ide_drive_t *drive = hwgroup->drive;
1335                 if (!drive) {
1336                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1337                         hwgroup->handler = NULL;
1338                 } else {
1339                         ide_hwif_t *hwif;
1340                         ide_startstop_t startstop = ide_stopped;
1341                         if (!hwgroup->busy) {
1342                                 hwgroup->busy = 1;      /* paranoia */
1343                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1344                         }
1345                         if ((expiry = hwgroup->expiry) != NULL) {
1346                                 /* continue */
1347                                 if ((wait = expiry(drive)) > 0) {
1348                                         /* reset timer */
1349                                         hwgroup->timer.expires  = jiffies + wait;
1350                                         add_timer(&hwgroup->timer);
1351                                         spin_unlock_irqrestore(&ide_lock, flags);
1352                                         return;
1353                                 }
1354                         }
1355                         hwgroup->handler = NULL;
1356                         /*
1357                          * We need to simulate a real interrupt when invoking
1358                          * the handler() function, which means we need to
1359                          * globally mask the specific IRQ:
1360                          */
1361                         spin_unlock(&ide_lock);
1362                         hwif  = HWIF(drive);
1363 #if DISABLE_IRQ_NOSYNC
1364                         disable_irq_nosync(hwif->irq);
1365 #else
1366                         /* disable_irq_nosync ?? */
1367                         disable_irq(hwif->irq);
1368 #endif /* DISABLE_IRQ_NOSYNC */
1369                         /* local CPU only,
1370                          * as if we were handling an interrupt */
1371                         local_irq_disable();
1372                         if (hwgroup->polling) {
1373                                 startstop = handler(drive);
1374                         } else if (drive_is_ready(drive)) {
1375                                 if (drive->waiting_for_dma)
1376                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1377                                 (void)ide_ack_intr(hwif);
1378                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1379                                 startstop = handler(drive);
1380                         } else {
1381                                 if (drive->waiting_for_dma) {
1382                                         startstop = ide_dma_timeout_retry(drive, wait);
1383                                 } else
1384                                         startstop =
1385                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1386                         }
1387                         drive->service_time = jiffies - drive->service_start;
1388                         spin_lock_irq(&ide_lock);
1389                         enable_irq(hwif->irq);
1390                         if (startstop == ide_stopped)
1391                                 hwgroup->busy = 0;
1392                 }
1393         }
1394         ide_do_request(hwgroup, IDE_NO_IRQ);
1395         spin_unlock_irqrestore(&ide_lock, flags);
1396 }
1397
1398 /**
1399  *      unexpected_intr         -       handle an unexpected IDE interrupt
1400  *      @irq: interrupt line
1401  *      @hwgroup: hwgroup being processed
1402  *
1403  *      There's nothing really useful we can do with an unexpected interrupt,
1404  *      other than reading the status register (to clear it), and logging it.
1405  *      There should be no way that an irq can happen before we're ready for it,
1406  *      so we needn't worry much about losing an "important" interrupt here.
1407  *
1408  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1409  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1410  *      looks "good", we just ignore the interrupt completely.
1411  *
1412  *      This routine assumes __cli() is in effect when called.
1413  *
1414  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1415  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1416  *      we could screw up by interfering with a new request being set up for 
1417  *      irq15.
1418  *
1419  *      In reality, this is a non-issue.  The new command is not sent unless 
1420  *      the drive is ready to accept one, in which case we know the drive is
1421  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1422  *      before completing the issuance of any new drive command, so we will not
1423  *      be accidentally invoked as a result of any valid command completion
1424  *      interrupt.
1425  *
1426  *      Note that we must walk the entire hwgroup here. We know which hwif
1427  *      is doing the current command, but we don't know which hwif burped
1428  *      mysteriously.
1429  */
1430  
1431 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1432 {
1433         u8 stat;
1434         ide_hwif_t *hwif = hwgroup->hwif;
1435
1436         /*
1437          * handle the unexpected interrupt
1438          */
1439         do {
1440                 if (hwif->irq == irq) {
1441                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1442                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1443                                 /* Try to not flood the console with msgs */
1444                                 static unsigned long last_msgtime, count;
1445                                 ++count;
1446                                 if (time_after(jiffies, last_msgtime + HZ)) {
1447                                         last_msgtime = jiffies;
1448                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1449                                                 "status=0x%02x, count=%ld\n",
1450                                                 hwif->name,
1451                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1452                                 }
1453                         }
1454                 }
1455         } while ((hwif = hwif->next) != hwgroup->hwif);
1456 }
1457
1458 /**
1459  *      ide_intr        -       default IDE interrupt handler
1460  *      @irq: interrupt number
1461  *      @dev_id: hwif group
1462  *      @regs: unused weirdness from the kernel irq layer
1463  *
1464  *      This is the default IRQ handler for the IDE layer. You should
1465  *      not need to override it. If you do be aware it is subtle in
1466  *      places
1467  *
1468  *      hwgroup->hwif is the interface in the group currently performing
1469  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1470  *      the IRQ handler to call. As we issue a command the handlers
1471  *      step through multiple states, reassigning the handler to the
1472  *      next step in the process. Unlike a smart SCSI controller IDE
1473  *      expects the main processor to sequence the various transfer
1474  *      stages. We also manage a poll timer to catch up with most
1475  *      timeout situations. There are still a few where the handlers
1476  *      don't ever decide to give up.
1477  *
1478  *      The handler eventually returns ide_stopped to indicate the
1479  *      request completed. At this point we issue the next request
1480  *      on the hwgroup and the process begins again.
1481  */
1482  
1483 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1484 {
1485         unsigned long flags;
1486         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1487         ide_hwif_t *hwif;
1488         ide_drive_t *drive;
1489         ide_handler_t *handler;
1490         ide_startstop_t startstop;
1491
1492         spin_lock_irqsave(&ide_lock, flags);
1493         hwif = hwgroup->hwif;
1494
1495         if (!ide_ack_intr(hwif)) {
1496                 spin_unlock_irqrestore(&ide_lock, flags);
1497                 return IRQ_NONE;
1498         }
1499
1500         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1501                 /*
1502                  * Not expecting an interrupt from this drive.
1503                  * That means this could be:
1504                  *      (1) an interrupt from another PCI device
1505                  *      sharing the same PCI INT# as us.
1506                  * or   (2) a drive just entered sleep or standby mode,
1507                  *      and is interrupting to let us know.
1508                  * or   (3) a spurious interrupt of unknown origin.
1509                  *
1510                  * For PCI, we cannot tell the difference,
1511                  * so in that case we just ignore it and hope it goes away.
1512                  *
1513                  * FIXME: unexpected_intr should be hwif-> then we can
1514                  * remove all the ifdef PCI crap
1515                  */
1516 #ifdef CONFIG_BLK_DEV_IDEPCI
1517                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1518 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1519                 {
1520                         /*
1521                          * Probably not a shared PCI interrupt,
1522                          * so we can safely try to do something about it:
1523                          */
1524                         unexpected_intr(irq, hwgroup);
1525 #ifdef CONFIG_BLK_DEV_IDEPCI
1526                 } else {
1527                         /*
1528                          * Whack the status register, just in case
1529                          * we have a leftover pending IRQ.
1530                          */
1531                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1532 #endif /* CONFIG_BLK_DEV_IDEPCI */
1533                 }
1534                 spin_unlock_irqrestore(&ide_lock, flags);
1535                 return IRQ_NONE;
1536         }
1537         drive = hwgroup->drive;
1538         if (!drive) {
1539                 /*
1540                  * This should NEVER happen, and there isn't much
1541                  * we could do about it here.
1542                  *
1543                  * [Note - this can occur if the drive is hot unplugged]
1544                  */
1545                 spin_unlock_irqrestore(&ide_lock, flags);
1546                 return IRQ_HANDLED;
1547         }
1548         if (!drive_is_ready(drive)) {
1549                 /*
1550                  * This happens regularly when we share a PCI IRQ with
1551                  * another device.  Unfortunately, it can also happen
1552                  * with some buggy drives that trigger the IRQ before
1553                  * their status register is up to date.  Hopefully we have
1554                  * enough advance overhead that the latter isn't a problem.
1555                  */
1556                 spin_unlock_irqrestore(&ide_lock, flags);
1557                 return IRQ_NONE;
1558         }
1559         if (!hwgroup->busy) {
1560                 hwgroup->busy = 1;      /* paranoia */
1561                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1562         }
1563         hwgroup->handler = NULL;
1564         del_timer(&hwgroup->timer);
1565         spin_unlock(&ide_lock);
1566
1567         if (drive->unmask)
1568                 local_irq_enable();
1569         /* service this interrupt, may set handler for next interrupt */
1570         startstop = handler(drive);
1571         spin_lock_irq(&ide_lock);
1572
1573         /*
1574          * Note that handler() may have set things up for another
1575          * interrupt to occur soon, but it cannot happen until
1576          * we exit from this routine, because it will be the
1577          * same irq as is currently being serviced here, and Linux
1578          * won't allow another of the same (on any CPU) until we return.
1579          */
1580         drive->service_time = jiffies - drive->service_start;
1581         if (startstop == ide_stopped) {
1582                 if (hwgroup->handler == NULL) { /* paranoia */
1583                         hwgroup->busy = 0;
1584                         ide_do_request(hwgroup, hwif->irq);
1585                 } else {
1586                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1587                                 "on exit\n", drive->name);
1588                 }
1589         }
1590         spin_unlock_irqrestore(&ide_lock, flags);
1591         return IRQ_HANDLED;
1592 }
1593
1594 /**
1595  *      ide_init_drive_cmd      -       initialize a drive command request
1596  *      @rq: request object
1597  *
1598  *      Initialize a request before we fill it in and send it down to
1599  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1600  *      now it doesn't do a lot, but if that changes abusers will have a
1601  *      nasty suprise.
1602  */
1603
1604 void ide_init_drive_cmd (struct request *rq)
1605 {
1606         memset(rq, 0, sizeof(*rq));
1607         rq->flags = REQ_DRIVE_CMD;
1608         rq->ref_count = 1;
1609 }
1610
1611 EXPORT_SYMBOL(ide_init_drive_cmd);
1612
1613 /**
1614  *      ide_do_drive_cmd        -       issue IDE special command
1615  *      @drive: device to issue command
1616  *      @rq: request to issue
1617  *      @action: action for processing
1618  *
1619  *      This function issues a special IDE device request
1620  *      onto the request queue.
1621  *
1622  *      If action is ide_wait, then the rq is queued at the end of the
1623  *      request queue, and the function sleeps until it has been processed.
1624  *      This is for use when invoked from an ioctl handler.
1625  *
1626  *      If action is ide_preempt, then the rq is queued at the head of
1627  *      the request queue, displacing the currently-being-processed
1628  *      request and this function returns immediately without waiting
1629  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1630  *      intended for careful use by the ATAPI tape/cdrom driver code.
1631  *
1632  *      If action is ide_end, then the rq is queued at the end of the
1633  *      request queue, and the function returns immediately without waiting
1634  *      for the new rq to be completed. This is again intended for careful
1635  *      use by the ATAPI tape/cdrom driver code.
1636  */
1637  
1638 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1639 {
1640         unsigned long flags;
1641         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1642         DECLARE_COMPLETION(wait);
1643         int where = ELEVATOR_INSERT_BACK, err;
1644         int must_wait = (action == ide_wait || action == ide_head_wait);
1645
1646         rq->errors = 0;
1647         rq->rq_status = RQ_ACTIVE;
1648
1649         /*
1650          * we need to hold an extra reference to request for safe inspection
1651          * after completion
1652          */
1653         if (must_wait) {
1654                 rq->ref_count++;
1655                 rq->waiting = &wait;
1656                 rq->end_io = blk_end_sync_rq;
1657         }
1658
1659         spin_lock_irqsave(&ide_lock, flags);
1660         if (action == ide_preempt)
1661                 hwgroup->rq = NULL;
1662         if (action == ide_preempt || action == ide_head_wait) {
1663                 where = ELEVATOR_INSERT_FRONT;
1664                 rq->flags |= REQ_PREEMPT;
1665         }
1666         __elv_add_request(drive->queue, rq, where, 0);
1667         ide_do_request(hwgroup, IDE_NO_IRQ);
1668         spin_unlock_irqrestore(&ide_lock, flags);
1669
1670         err = 0;
1671         if (must_wait) {
1672                 wait_for_completion(&wait);
1673                 rq->waiting = NULL;
1674                 if (rq->errors)
1675                         err = -EIO;
1676
1677                 blk_put_request(rq);
1678         }
1679
1680         return err;
1681 }
1682
1683 EXPORT_SYMBOL(ide_do_drive_cmd);