ide: make void and rename ide_dma_timeout() method
[linux-2.6] / drivers / ide / ide-io.c
1 /*
2  *      IDE I/O functions
3  *
4  *      Basic PIO and command management functionality.
5  *
6  * This code was split off from ide.c. See ide.c for history and original
7  * copyrights.
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the
11  * Free Software Foundation; either version 2, or (at your option) any
12  * later version.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * For the avoidance of doubt the "preferred form" of this code is one which
20  * is in an open non patent encumbered format. Where cryptographic key signing
21  * forms part of the process of creating an executable the information
22  * including keys needed to generate an equivalently functional executable
23  * are deemed to be part of the source code.
24  */
25  
26  
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/string.h>
30 #include <linux/kernel.h>
31 #include <linux/timer.h>
32 #include <linux/mm.h>
33 #include <linux/interrupt.h>
34 #include <linux/major.h>
35 #include <linux/errno.h>
36 #include <linux/genhd.h>
37 #include <linux/blkpg.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/ide.h>
43 #include <linux/completion.h>
44 #include <linux/reboot.h>
45 #include <linux/cdrom.h>
46 #include <linux/seq_file.h>
47 #include <linux/device.h>
48 #include <linux/kmod.h>
49 #include <linux/scatterlist.h>
50
51 #include <asm/byteorder.h>
52 #include <asm/irq.h>
53 #include <asm/uaccess.h>
54 #include <asm/io.h>
55 #include <asm/bitops.h>
56
57 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
58                              int uptodate, int nr_sectors)
59 {
60         int ret = 1;
61
62         /*
63          * if failfast is set on a request, override number of sectors and
64          * complete the whole request right now
65          */
66         if (blk_noretry_request(rq) && end_io_error(uptodate))
67                 nr_sectors = rq->hard_nr_sectors;
68
69         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
70                 rq->errors = -EIO;
71
72         /*
73          * decide whether to reenable DMA -- 3 is a random magic for now,
74          * if we DMA timeout more than 3 times, just stay in PIO
75          */
76         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
77                 drive->state = 0;
78                 HWGROUP(drive)->hwif->ide_dma_on(drive);
79         }
80
81         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
82                 add_disk_randomness(rq->rq_disk);
83                 if (!list_empty(&rq->queuelist))
84                         blkdev_dequeue_request(rq);
85                 HWGROUP(drive)->rq = NULL;
86                 end_that_request_last(rq, uptodate);
87                 ret = 0;
88         }
89
90         return ret;
91 }
92
93 /**
94  *      ide_end_request         -       complete an IDE I/O
95  *      @drive: IDE device for the I/O
96  *      @uptodate:
97  *      @nr_sectors: number of sectors completed
98  *
99  *      This is our end_request wrapper function. We complete the I/O
100  *      update random number input and dequeue the request, which if
101  *      it was tagged may be out of order.
102  */
103
104 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
105 {
106         struct request *rq;
107         unsigned long flags;
108         int ret = 1;
109
110         /*
111          * room for locking improvements here, the calls below don't
112          * need the queue lock held at all
113          */
114         spin_lock_irqsave(&ide_lock, flags);
115         rq = HWGROUP(drive)->rq;
116
117         if (!nr_sectors)
118                 nr_sectors = rq->hard_cur_sectors;
119
120         ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
121
122         spin_unlock_irqrestore(&ide_lock, flags);
123         return ret;
124 }
125 EXPORT_SYMBOL(ide_end_request);
126
127 /*
128  * Power Management state machine. This one is rather trivial for now,
129  * we should probably add more, like switching back to PIO on suspend
130  * to help some BIOSes, re-do the door locking on resume, etc...
131  */
132
133 enum {
134         ide_pm_flush_cache      = ide_pm_state_start_suspend,
135         idedisk_pm_standby,
136
137         idedisk_pm_restore_pio  = ide_pm_state_start_resume,
138         idedisk_pm_idle,
139         ide_pm_restore_dma,
140 };
141
142 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
143 {
144         struct request_pm_state *pm = rq->data;
145
146         if (drive->media != ide_disk)
147                 return;
148
149         switch (pm->pm_step) {
150         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
151                 if (pm->pm_state == PM_EVENT_FREEZE)
152                         pm->pm_step = ide_pm_state_completed;
153                 else
154                         pm->pm_step = idedisk_pm_standby;
155                 break;
156         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
157                 pm->pm_step = ide_pm_state_completed;
158                 break;
159         case idedisk_pm_restore_pio:    /* Resume step 1 complete */
160                 pm->pm_step = idedisk_pm_idle;
161                 break;
162         case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
163                 pm->pm_step = ide_pm_restore_dma;
164                 break;
165         }
166 }
167
168 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
169 {
170         struct request_pm_state *pm = rq->data;
171         ide_task_t *args = rq->special;
172
173         memset(args, 0, sizeof(*args));
174
175         switch (pm->pm_step) {
176         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
177                 if (drive->media != ide_disk)
178                         break;
179                 /* Not supported? Switch to next step now. */
180                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
181                         ide_complete_power_step(drive, rq, 0, 0);
182                         return ide_stopped;
183                 }
184                 if (ide_id_has_flush_cache_ext(drive->id))
185                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
186                 else
187                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
188                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
189                 args->handler      = &task_no_data_intr;
190                 return do_rw_taskfile(drive, args);
191
192         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
193                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
194                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
195                 args->handler      = &task_no_data_intr;
196                 return do_rw_taskfile(drive, args);
197
198         case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
199                 if (drive->hwif->tuneproc != NULL)
200                         drive->hwif->tuneproc(drive, 255);
201                 /*
202                  * skip idedisk_pm_idle for ATAPI devices
203                  */
204                 if (drive->media != ide_disk)
205                         pm->pm_step = ide_pm_restore_dma;
206                 else
207                         ide_complete_power_step(drive, rq, 0, 0);
208                 return ide_stopped;
209
210         case idedisk_pm_idle:           /* Resume step 2 (idle) */
211                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
212                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
213                 args->handler = task_no_data_intr;
214                 return do_rw_taskfile(drive, args);
215
216         case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
217                 /*
218                  * Right now, all we do is call hwif->ide_dma_check(drive),
219                  * we could be smarter and check for current xfer_speed
220                  * in struct drive etc...
221                  */
222                 if ((drive->id->capability & 1) == 0)
223                         break;
224                 if (drive->hwif->ide_dma_check == NULL)
225                         break;
226                 drive->hwif->dma_off_quietly(drive);
227                 ide_set_dma(drive);
228                 break;
229         }
230         pm->pm_step = ide_pm_state_completed;
231         return ide_stopped;
232 }
233
234 /**
235  *      ide_end_dequeued_request        -       complete an IDE I/O
236  *      @drive: IDE device for the I/O
237  *      @uptodate:
238  *      @nr_sectors: number of sectors completed
239  *
240  *      Complete an I/O that is no longer on the request queue. This
241  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
242  *      We must still finish the old request but we must not tamper with the
243  *      queue in the meantime.
244  *
245  *      NOTE: This path does not handle barrier, but barrier is not supported
246  *      on ide-cd anyway.
247  */
248
249 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
250                              int uptodate, int nr_sectors)
251 {
252         unsigned long flags;
253         int ret = 1;
254
255         spin_lock_irqsave(&ide_lock, flags);
256
257         BUG_ON(!blk_rq_started(rq));
258
259         /*
260          * if failfast is set on a request, override number of sectors and
261          * complete the whole request right now
262          */
263         if (blk_noretry_request(rq) && end_io_error(uptodate))
264                 nr_sectors = rq->hard_nr_sectors;
265
266         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
267                 rq->errors = -EIO;
268
269         /*
270          * decide whether to reenable DMA -- 3 is a random magic for now,
271          * if we DMA timeout more than 3 times, just stay in PIO
272          */
273         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
274                 drive->state = 0;
275                 HWGROUP(drive)->hwif->ide_dma_on(drive);
276         }
277
278         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
279                 add_disk_randomness(rq->rq_disk);
280                 if (blk_rq_tagged(rq))
281                         blk_queue_end_tag(drive->queue, rq);
282                 end_that_request_last(rq, uptodate);
283                 ret = 0;
284         }
285         spin_unlock_irqrestore(&ide_lock, flags);
286         return ret;
287 }
288 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
289
290
291 /**
292  *      ide_complete_pm_request - end the current Power Management request
293  *      @drive: target drive
294  *      @rq: request
295  *
296  *      This function cleans up the current PM request and stops the queue
297  *      if necessary.
298  */
299 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
300 {
301         unsigned long flags;
302
303 #ifdef DEBUG_PM
304         printk("%s: completing PM request, %s\n", drive->name,
305                blk_pm_suspend_request(rq) ? "suspend" : "resume");
306 #endif
307         spin_lock_irqsave(&ide_lock, flags);
308         if (blk_pm_suspend_request(rq)) {
309                 blk_stop_queue(drive->queue);
310         } else {
311                 drive->blocked = 0;
312                 blk_start_queue(drive->queue);
313         }
314         blkdev_dequeue_request(rq);
315         HWGROUP(drive)->rq = NULL;
316         end_that_request_last(rq, 1);
317         spin_unlock_irqrestore(&ide_lock, flags);
318 }
319
320 /*
321  * FIXME: probably move this somewhere else, name is bad too :)
322  */
323 u64 ide_get_error_location(ide_drive_t *drive, char *args)
324 {
325         u32 high, low;
326         u8 hcyl, lcyl, sect;
327         u64 sector;
328
329         high = 0;
330         hcyl = args[5];
331         lcyl = args[4];
332         sect = args[3];
333
334         if (ide_id_has_flush_cache_ext(drive->id)) {
335                 low = (hcyl << 16) | (lcyl << 8) | sect;
336                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
337                 high = ide_read_24(drive);
338         } else {
339                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
340                 if (cur & 0x40) {
341                         high = cur & 0xf;
342                         low = (hcyl << 16) | (lcyl << 8) | sect;
343                 } else {
344                         low = hcyl * drive->head * drive->sect;
345                         low += lcyl * drive->sect;
346                         low += sect - 1;
347                 }
348         }
349
350         sector = ((u64) high << 24) | low;
351         return sector;
352 }
353 EXPORT_SYMBOL(ide_get_error_location);
354
355 /**
356  *      ide_end_drive_cmd       -       end an explicit drive command
357  *      @drive: command 
358  *      @stat: status bits
359  *      @err: error bits
360  *
361  *      Clean up after success/failure of an explicit drive command.
362  *      These get thrown onto the queue so they are synchronized with
363  *      real I/O operations on the drive.
364  *
365  *      In LBA48 mode we have to read the register set twice to get
366  *      all the extra information out.
367  */
368  
369 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
370 {
371         ide_hwif_t *hwif = HWIF(drive);
372         unsigned long flags;
373         struct request *rq;
374
375         spin_lock_irqsave(&ide_lock, flags);
376         rq = HWGROUP(drive)->rq;
377         spin_unlock_irqrestore(&ide_lock, flags);
378
379         if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
380                 u8 *args = (u8 *) rq->buffer;
381                 if (rq->errors == 0)
382                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
383
384                 if (args) {
385                         args[0] = stat;
386                         args[1] = err;
387                         args[2] = hwif->INB(IDE_NSECTOR_REG);
388                 }
389         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
390                 u8 *args = (u8 *) rq->buffer;
391                 if (rq->errors == 0)
392                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
393
394                 if (args) {
395                         args[0] = stat;
396                         args[1] = err;
397                         args[2] = hwif->INB(IDE_NSECTOR_REG);
398                         args[3] = hwif->INB(IDE_SECTOR_REG);
399                         args[4] = hwif->INB(IDE_LCYL_REG);
400                         args[5] = hwif->INB(IDE_HCYL_REG);
401                         args[6] = hwif->INB(IDE_SELECT_REG);
402                 }
403         } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
404                 ide_task_t *args = (ide_task_t *) rq->special;
405                 if (rq->errors == 0)
406                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
407                         
408                 if (args) {
409                         if (args->tf_in_flags.b.data) {
410                                 u16 data                                = hwif->INW(IDE_DATA_REG);
411                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
412                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
413                         }
414                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
415                         /* be sure we're looking at the low order bits */
416                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
417                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
418                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
419                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
420                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
421                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
422                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
423
424                         if (drive->addressing == 1) {
425                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
426                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
427                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
428                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
429                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
430                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
431                         }
432                 }
433         } else if (blk_pm_request(rq)) {
434                 struct request_pm_state *pm = rq->data;
435 #ifdef DEBUG_PM
436                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
437                         drive->name, rq->pm->pm_step, stat, err);
438 #endif
439                 ide_complete_power_step(drive, rq, stat, err);
440                 if (pm->pm_step == ide_pm_state_completed)
441                         ide_complete_pm_request(drive, rq);
442                 return;
443         }
444
445         spin_lock_irqsave(&ide_lock, flags);
446         blkdev_dequeue_request(rq);
447         HWGROUP(drive)->rq = NULL;
448         rq->errors = err;
449         end_that_request_last(rq, !rq->errors);
450         spin_unlock_irqrestore(&ide_lock, flags);
451 }
452
453 EXPORT_SYMBOL(ide_end_drive_cmd);
454
455 /**
456  *      try_to_flush_leftover_data      -       flush junk
457  *      @drive: drive to flush
458  *
459  *      try_to_flush_leftover_data() is invoked in response to a drive
460  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
461  *      resetting the drive, this routine tries to clear the condition
462  *      by read a sector's worth of data from the drive.  Of course,
463  *      this may not help if the drive is *waiting* for data from *us*.
464  */
465 static void try_to_flush_leftover_data (ide_drive_t *drive)
466 {
467         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
468
469         if (drive->media != ide_disk)
470                 return;
471         while (i > 0) {
472                 u32 buffer[16];
473                 u32 wcount = (i > 16) ? 16 : i;
474
475                 i -= wcount;
476                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
477         }
478 }
479
480 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
481 {
482         if (rq->rq_disk) {
483                 ide_driver_t *drv;
484
485                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
486                 drv->end_request(drive, 0, 0);
487         } else
488                 ide_end_request(drive, 0, 0);
489 }
490
491 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
492 {
493         ide_hwif_t *hwif = drive->hwif;
494
495         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
496                 /* other bits are useless when BUSY */
497                 rq->errors |= ERROR_RESET;
498         } else if (stat & ERR_STAT) {
499                 /* err has different meaning on cdrom and tape */
500                 if (err == ABRT_ERR) {
501                         if (drive->select.b.lba &&
502                             /* some newer drives don't support WIN_SPECIFY */
503                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
504                                 return ide_stopped;
505                 } else if ((err & BAD_CRC) == BAD_CRC) {
506                         /* UDMA crc error, just retry the operation */
507                         drive->crc_count++;
508                 } else if (err & (BBD_ERR | ECC_ERR)) {
509                         /* retries won't help these */
510                         rq->errors = ERROR_MAX;
511                 } else if (err & TRK0_ERR) {
512                         /* help it find track zero */
513                         rq->errors |= ERROR_RECAL;
514                 }
515         }
516
517         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ && hwif->err_stops_fifo == 0)
518                 try_to_flush_leftover_data(drive);
519
520         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
521                 ide_kill_rq(drive, rq);
522                 return ide_stopped;
523         }
524
525         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
526                 rq->errors |= ERROR_RESET;
527
528         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
529                 ++rq->errors;
530                 return ide_do_reset(drive);
531         }
532
533         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
534                 drive->special.b.recalibrate = 1;
535
536         ++rq->errors;
537
538         return ide_stopped;
539 }
540
541 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
542 {
543         ide_hwif_t *hwif = drive->hwif;
544
545         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
546                 /* other bits are useless when BUSY */
547                 rq->errors |= ERROR_RESET;
548         } else {
549                 /* add decoding error stuff */
550         }
551
552         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
553                 /* force an abort */
554                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
555
556         if (rq->errors >= ERROR_MAX) {
557                 ide_kill_rq(drive, rq);
558         } else {
559                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
560                         ++rq->errors;
561                         return ide_do_reset(drive);
562                 }
563                 ++rq->errors;
564         }
565
566         return ide_stopped;
567 }
568
569 ide_startstop_t
570 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
571 {
572         if (drive->media == ide_disk)
573                 return ide_ata_error(drive, rq, stat, err);
574         return ide_atapi_error(drive, rq, stat, err);
575 }
576
577 EXPORT_SYMBOL_GPL(__ide_error);
578
579 /**
580  *      ide_error       -       handle an error on the IDE
581  *      @drive: drive the error occurred on
582  *      @msg: message to report
583  *      @stat: status bits
584  *
585  *      ide_error() takes action based on the error returned by the drive.
586  *      For normal I/O that may well include retries. We deal with
587  *      both new-style (taskfile) and old style command handling here.
588  *      In the case of taskfile command handling there is work left to
589  *      do
590  */
591  
592 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
593 {
594         struct request *rq;
595         u8 err;
596
597         err = ide_dump_status(drive, msg, stat);
598
599         if ((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, stat, err);
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->error(drive, rq, stat, err);
614         } else
615                 return __ide_error(drive, rq, stat, err);
616 }
617
618 EXPORT_SYMBOL_GPL(ide_error);
619
620 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
621 {
622         if (drive->media != ide_disk)
623                 rq->errors |= ERROR_RESET;
624
625         ide_kill_rq(drive, rq);
626
627         return ide_stopped;
628 }
629
630 EXPORT_SYMBOL_GPL(__ide_abort);
631
632 /**
633  *      ide_abort       -       abort pending IDE operations
634  *      @drive: drive the error occurred on
635  *      @msg: message to report
636  *
637  *      ide_abort kills and cleans up when we are about to do a 
638  *      host initiated reset on active commands. Longer term we
639  *      want handlers to have sensible abort handling themselves
640  *
641  *      This differs fundamentally from ide_error because in 
642  *      this case the command is doing just fine when we
643  *      blow it away.
644  */
645  
646 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
647 {
648         struct request *rq;
649
650         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
651                 return ide_stopped;
652
653         /* retry only "normal" I/O: */
654         if (!blk_fs_request(rq)) {
655                 rq->errors = 1;
656                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
657                 return ide_stopped;
658         }
659
660         if (rq->rq_disk) {
661                 ide_driver_t *drv;
662
663                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
664                 return drv->abort(drive, rq);
665         } else
666                 return __ide_abort(drive, rq);
667 }
668
669 /**
670  *      ide_cmd         -       issue a simple drive command
671  *      @drive: drive the command is for
672  *      @cmd: command byte
673  *      @nsect: sector byte
674  *      @handler: handler for the command completion
675  *
676  *      Issue a simple drive command with interrupts.
677  *      The drive must be selected beforehand.
678  */
679
680 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
681                 ide_handler_t *handler)
682 {
683         ide_hwif_t *hwif = HWIF(drive);
684         if (IDE_CONTROL_REG)
685                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
686         SELECT_MASK(drive,0);
687         hwif->OUTB(nsect,IDE_NSECTOR_REG);
688         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
689 }
690
691 /**
692  *      drive_cmd_intr          -       drive command completion interrupt
693  *      @drive: drive the completion interrupt occurred on
694  *
695  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
696  *      We do any necessary data reading and then wait for the drive to
697  *      go non busy. At that point we may read the error data and complete
698  *      the request
699  */
700  
701 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
702 {
703         struct request *rq = HWGROUP(drive)->rq;
704         ide_hwif_t *hwif = HWIF(drive);
705         u8 *args = (u8 *) rq->buffer;
706         u8 stat = hwif->INB(IDE_STATUS_REG);
707         int retries = 10;
708
709         local_irq_enable_in_hardirq();
710         if ((stat & DRQ_STAT) && args && args[3]) {
711                 u8 io_32bit = drive->io_32bit;
712                 drive->io_32bit = 0;
713                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
714                 drive->io_32bit = io_32bit;
715                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
716                         udelay(100);
717         }
718
719         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
720                 return ide_error(drive, "drive_cmd", stat);
721                 /* calls ide_end_drive_cmd */
722         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
723         return ide_stopped;
724 }
725
726 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
727 {
728         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
729         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
730         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
731         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
732         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
733         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
734
735         task->handler = &set_geometry_intr;
736 }
737
738 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
739 {
740         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
741         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
742
743         task->handler = &recal_intr;
744 }
745
746 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
747 {
748         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
749         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
750
751         task->handler = &set_multmode_intr;
752 }
753
754 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
755 {
756         special_t *s = &drive->special;
757         ide_task_t args;
758
759         memset(&args, 0, sizeof(ide_task_t));
760         args.command_type = IDE_DRIVE_TASK_NO_DATA;
761
762         if (s->b.set_geometry) {
763                 s->b.set_geometry = 0;
764                 ide_init_specify_cmd(drive, &args);
765         } else if (s->b.recalibrate) {
766                 s->b.recalibrate = 0;
767                 ide_init_restore_cmd(drive, &args);
768         } else if (s->b.set_multmode) {
769                 s->b.set_multmode = 0;
770                 if (drive->mult_req > drive->id->max_multsect)
771                         drive->mult_req = drive->id->max_multsect;
772                 ide_init_setmult_cmd(drive, &args);
773         } else if (s->all) {
774                 int special = s->all;
775                 s->all = 0;
776                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
777                 return ide_stopped;
778         }
779
780         do_rw_taskfile(drive, &args);
781
782         return ide_started;
783 }
784
785 /**
786  *      do_special              -       issue some special commands
787  *      @drive: drive the command is for
788  *
789  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
790  *      commands to a drive.  It used to do much more, but has been scaled
791  *      back.
792  */
793
794 static ide_startstop_t do_special (ide_drive_t *drive)
795 {
796         special_t *s = &drive->special;
797
798 #ifdef DEBUG
799         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
800 #endif
801         if (s->b.set_tune) {
802                 s->b.set_tune = 0;
803                 if (HWIF(drive)->tuneproc != NULL)
804                         HWIF(drive)->tuneproc(drive, drive->tune_req);
805                 return ide_stopped;
806         } else {
807                 if (drive->media == ide_disk)
808                         return ide_disk_special(drive);
809
810                 s->all = 0;
811                 drive->mult_req = 0;
812                 return ide_stopped;
813         }
814 }
815
816 void ide_map_sg(ide_drive_t *drive, struct request *rq)
817 {
818         ide_hwif_t *hwif = drive->hwif;
819         struct scatterlist *sg = hwif->sg_table;
820
821         if (hwif->sg_mapped)    /* needed by ide-scsi */
822                 return;
823
824         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
825                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
826         } else {
827                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
828                 hwif->sg_nents = 1;
829         }
830 }
831
832 EXPORT_SYMBOL_GPL(ide_map_sg);
833
834 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
835 {
836         ide_hwif_t *hwif = drive->hwif;
837
838         hwif->nsect = hwif->nleft = rq->nr_sectors;
839         hwif->cursg = hwif->cursg_ofs = 0;
840 }
841
842 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
843
844 /**
845  *      execute_drive_command   -       issue special drive command
846  *      @drive: the drive to issue the command on
847  *      @rq: the request structure holding the command
848  *
849  *      execute_drive_cmd() issues a special drive command,  usually 
850  *      initiated by ioctl() from the external hdparm program. The
851  *      command can be a drive command, drive task or taskfile 
852  *      operation. Weirdly you can call it with NULL to wait for
853  *      all commands to finish. Don't do this as that is due to change
854  */
855
856 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
857                 struct request *rq)
858 {
859         ide_hwif_t *hwif = HWIF(drive);
860         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
861                 ide_task_t *args = rq->special;
862  
863                 if (!args)
864                         goto done;
865
866                 hwif->data_phase = args->data_phase;
867
868                 switch (hwif->data_phase) {
869                 case TASKFILE_MULTI_OUT:
870                 case TASKFILE_OUT:
871                 case TASKFILE_MULTI_IN:
872                 case TASKFILE_IN:
873                         ide_init_sg_cmd(drive, rq);
874                         ide_map_sg(drive, rq);
875                 default:
876                         break;
877                 }
878
879                 if (args->tf_out_flags.all != 0) 
880                         return flagged_taskfile(drive, args);
881                 return do_rw_taskfile(drive, args);
882         } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
883                 u8 *args = rq->buffer;
884                 u8 sel;
885  
886                 if (!args)
887                         goto done;
888 #ifdef DEBUG
889                 printk("%s: DRIVE_TASK_CMD ", drive->name);
890                 printk("cmd=0x%02x ", args[0]);
891                 printk("fr=0x%02x ", args[1]);
892                 printk("ns=0x%02x ", args[2]);
893                 printk("sc=0x%02x ", args[3]);
894                 printk("lcyl=0x%02x ", args[4]);
895                 printk("hcyl=0x%02x ", args[5]);
896                 printk("sel=0x%02x\n", args[6]);
897 #endif
898                 hwif->OUTB(args[1], IDE_FEATURE_REG);
899                 hwif->OUTB(args[3], IDE_SECTOR_REG);
900                 hwif->OUTB(args[4], IDE_LCYL_REG);
901                 hwif->OUTB(args[5], IDE_HCYL_REG);
902                 sel = (args[6] & ~0x10);
903                 if (drive->select.b.unit)
904                         sel |= 0x10;
905                 hwif->OUTB(sel, IDE_SELECT_REG);
906                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
907                 return ide_started;
908         } else if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
909                 u8 *args = rq->buffer;
910
911                 if (!args)
912                         goto done;
913 #ifdef DEBUG
914                 printk("%s: DRIVE_CMD ", drive->name);
915                 printk("cmd=0x%02x ", args[0]);
916                 printk("sc=0x%02x ", args[1]);
917                 printk("fr=0x%02x ", args[2]);
918                 printk("xx=0x%02x\n", args[3]);
919 #endif
920                 if (args[0] == WIN_SMART) {
921                         hwif->OUTB(0x4f, IDE_LCYL_REG);
922                         hwif->OUTB(0xc2, IDE_HCYL_REG);
923                         hwif->OUTB(args[2],IDE_FEATURE_REG);
924                         hwif->OUTB(args[1],IDE_SECTOR_REG);
925                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
926                         return ide_started;
927                 }
928                 hwif->OUTB(args[2],IDE_FEATURE_REG);
929                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
930                 return ide_started;
931         }
932
933 done:
934         /*
935          * NULL is actually a valid way of waiting for
936          * all current requests to be flushed from the queue.
937          */
938 #ifdef DEBUG
939         printk("%s: DRIVE_CMD (null)\n", drive->name);
940 #endif
941         ide_end_drive_cmd(drive,
942                         hwif->INB(IDE_STATUS_REG),
943                         hwif->INB(IDE_ERROR_REG));
944         return ide_stopped;
945 }
946
947 static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
948 {
949         struct request_pm_state *pm = rq->data;
950
951         if (blk_pm_suspend_request(rq) &&
952             pm->pm_step == ide_pm_state_start_suspend)
953                 /* Mark drive blocked when starting the suspend sequence. */
954                 drive->blocked = 1;
955         else if (blk_pm_resume_request(rq) &&
956                  pm->pm_step == ide_pm_state_start_resume) {
957                 /* 
958                  * The first thing we do on wakeup is to wait for BSY bit to
959                  * go away (with a looong timeout) as a drive on this hwif may
960                  * just be POSTing itself.
961                  * We do that before even selecting as the "other" device on
962                  * the bus may be broken enough to walk on our toes at this
963                  * point.
964                  */
965                 int rc;
966 #ifdef DEBUG_PM
967                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
968 #endif
969                 rc = ide_wait_not_busy(HWIF(drive), 35000);
970                 if (rc)
971                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
972                 SELECT_DRIVE(drive);
973                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
974                 rc = ide_wait_not_busy(HWIF(drive), 100000);
975                 if (rc)
976                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
977         }
978 }
979
980 /**
981  *      start_request   -       start of I/O and command issuing for IDE
982  *
983  *      start_request() initiates handling of a new I/O request. It
984  *      accepts commands and I/O (read/write) requests. It also does
985  *      the final remapping for weird stuff like EZDrive. Once 
986  *      device mapper can work sector level the EZDrive stuff can go away
987  *
988  *      FIXME: this function needs a rename
989  */
990  
991 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
992 {
993         ide_startstop_t startstop;
994         sector_t block;
995
996         BUG_ON(!blk_rq_started(rq));
997
998 #ifdef DEBUG
999         printk("%s: start_request: current=0x%08lx\n",
1000                 HWIF(drive)->name, (unsigned long) rq);
1001 #endif
1002
1003         /* bail early if we've exceeded max_failures */
1004         if (drive->max_failures && (drive->failures > drive->max_failures)) {
1005                 goto kill_rq;
1006         }
1007
1008         block    = rq->sector;
1009         if (blk_fs_request(rq) &&
1010             (drive->media == ide_disk || drive->media == ide_floppy)) {
1011                 block += drive->sect0;
1012         }
1013         /* Yecch - this will shift the entire interval,
1014            possibly killing some innocent following sector */
1015         if (block == 0 && drive->remap_0_to_1 == 1)
1016                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
1017
1018         if (blk_pm_request(rq))
1019                 ide_check_pm_state(drive, rq);
1020
1021         SELECT_DRIVE(drive);
1022         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
1023                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
1024                 return startstop;
1025         }
1026         if (!drive->special.all) {
1027                 ide_driver_t *drv;
1028
1029                 /*
1030                  * We reset the drive so we need to issue a SETFEATURES.
1031                  * Do it _after_ do_special() restored device parameters.
1032                  */
1033                 if (drive->current_speed == 0xff)
1034                         ide_config_drive_speed(drive, drive->desired_speed);
1035
1036                 if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
1037                     rq->cmd_type == REQ_TYPE_ATA_TASK ||
1038                     rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
1039                         return execute_drive_cmd(drive, rq);
1040                 else if (blk_pm_request(rq)) {
1041                         struct request_pm_state *pm = rq->data;
1042 #ifdef DEBUG_PM
1043                         printk("%s: start_power_step(step: %d)\n",
1044                                 drive->name, rq->pm->pm_step);
1045 #endif
1046                         startstop = ide_start_power_step(drive, rq);
1047                         if (startstop == ide_stopped &&
1048                             pm->pm_step == ide_pm_state_completed)
1049                                 ide_complete_pm_request(drive, rq);
1050                         return startstop;
1051                 }
1052
1053                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
1054                 return drv->do_request(drive, rq, block);
1055         }
1056         return do_special(drive);
1057 kill_rq:
1058         ide_kill_rq(drive, rq);
1059         return ide_stopped;
1060 }
1061
1062 /**
1063  *      ide_stall_queue         -       pause an IDE device
1064  *      @drive: drive to stall
1065  *      @timeout: time to stall for (jiffies)
1066  *
1067  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1068  *      to the hwgroup by sleeping for timeout jiffies.
1069  */
1070  
1071 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1072 {
1073         if (timeout > WAIT_WORSTCASE)
1074                 timeout = WAIT_WORSTCASE;
1075         drive->sleep = timeout + jiffies;
1076         drive->sleeping = 1;
1077 }
1078
1079 EXPORT_SYMBOL(ide_stall_queue);
1080
1081 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1082
1083 /**
1084  *      choose_drive            -       select a drive to service
1085  *      @hwgroup: hardware group to select on
1086  *
1087  *      choose_drive() selects the next drive which will be serviced.
1088  *      This is necessary because the IDE layer can't issue commands
1089  *      to both drives on the same cable, unlike SCSI.
1090  */
1091  
1092 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1093 {
1094         ide_drive_t *drive, *best;
1095
1096 repeat: 
1097         best = NULL;
1098         drive = hwgroup->drive;
1099
1100         /*
1101          * drive is doing pre-flush, ordered write, post-flush sequence. even
1102          * though that is 3 requests, it must be seen as a single transaction.
1103          * we must not preempt this drive until that is complete
1104          */
1105         if (blk_queue_flushing(drive->queue)) {
1106                 /*
1107                  * small race where queue could get replugged during
1108                  * the 3-request flush cycle, just yank the plug since
1109                  * we want it to finish asap
1110                  */
1111                 blk_remove_plug(drive->queue);
1112                 return drive;
1113         }
1114
1115         do {
1116                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1117                     && !elv_queue_empty(drive->queue)) {
1118                         if (!best
1119                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1120                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1121                         {
1122                                 if (!blk_queue_plugged(drive->queue))
1123                                         best = drive;
1124                         }
1125                 }
1126         } while ((drive = drive->next) != hwgroup->drive);
1127         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1128                 long t = (signed long)(WAKEUP(best) - jiffies);
1129                 if (t >= WAIT_MIN_SLEEP) {
1130                 /*
1131                  * We *may* have some time to spare, but first let's see if
1132                  * someone can potentially benefit from our nice mood today..
1133                  */
1134                         drive = best->next;
1135                         do {
1136                                 if (!drive->sleeping
1137                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1138                                  && time_before(WAKEUP(drive), jiffies + t))
1139                                 {
1140                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1141                                         goto repeat;
1142                                 }
1143                         } while ((drive = drive->next) != best);
1144                 }
1145         }
1146         return best;
1147 }
1148
1149 /*
1150  * Issue a new request to a drive from hwgroup
1151  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1152  *
1153  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1154  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1155  * may have both interfaces in a single hwgroup to "serialize" access.
1156  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1157  * together into one hwgroup for serialized access.
1158  *
1159  * Note also that several hwgroups can end up sharing a single IRQ,
1160  * possibly along with many other devices.  This is especially common in
1161  * PCI-based systems with off-board IDE controller cards.
1162  *
1163  * The IDE driver uses the single global ide_lock spinlock to protect
1164  * access to the request queues, and to protect the hwgroup->busy flag.
1165  *
1166  * The first thread into the driver for a particular hwgroup sets the
1167  * hwgroup->busy flag to indicate that this hwgroup is now active,
1168  * and then initiates processing of the top request from the request queue.
1169  *
1170  * Other threads attempting entry notice the busy setting, and will simply
1171  * queue their new requests and exit immediately.  Note that hwgroup->busy
1172  * remains set even when the driver is merely awaiting the next interrupt.
1173  * Thus, the meaning is "this hwgroup is busy processing a request".
1174  *
1175  * When processing of a request completes, the completing thread or IRQ-handler
1176  * will start the next request from the queue.  If no more work remains,
1177  * the driver will clear the hwgroup->busy flag and exit.
1178  *
1179  * The ide_lock (spinlock) is used to protect all access to the
1180  * hwgroup->busy flag, but is otherwise not needed for most processing in
1181  * the driver.  This makes the driver much more friendlier to shared IRQs
1182  * than previous designs, while remaining 100% (?) SMP safe and capable.
1183  */
1184 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1185 {
1186         ide_drive_t     *drive;
1187         ide_hwif_t      *hwif;
1188         struct request  *rq;
1189         ide_startstop_t startstop;
1190         int             loops = 0;
1191
1192         /* for atari only: POSSIBLY BROKEN HERE(?) */
1193         ide_get_lock(ide_intr, hwgroup);
1194
1195         /* caller must own ide_lock */
1196         BUG_ON(!irqs_disabled());
1197
1198         while (!hwgroup->busy) {
1199                 hwgroup->busy = 1;
1200                 drive = choose_drive(hwgroup);
1201                 if (drive == NULL) {
1202                         int sleeping = 0;
1203                         unsigned long sleep = 0; /* shut up, gcc */
1204                         hwgroup->rq = NULL;
1205                         drive = hwgroup->drive;
1206                         do {
1207                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1208                                         sleeping = 1;
1209                                         sleep = drive->sleep;
1210                                 }
1211                         } while ((drive = drive->next) != hwgroup->drive);
1212                         if (sleeping) {
1213                 /*
1214                  * Take a short snooze, and then wake up this hwgroup again.
1215                  * This gives other hwgroups on the same a chance to
1216                  * play fairly with us, just in case there are big differences
1217                  * in relative throughputs.. don't want to hog the cpu too much.
1218                  */
1219                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1220                                         sleep = jiffies + WAIT_MIN_SLEEP;
1221 #if 1
1222                                 if (timer_pending(&hwgroup->timer))
1223                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1224 #endif
1225                                 /* so that ide_timer_expiry knows what to do */
1226                                 hwgroup->sleeping = 1;
1227                                 hwgroup->req_gen_timer = hwgroup->req_gen;
1228                                 mod_timer(&hwgroup->timer, sleep);
1229                                 /* we purposely leave hwgroup->busy==1
1230                                  * while sleeping */
1231                         } else {
1232                                 /* Ugly, but how can we sleep for the lock
1233                                  * otherwise? perhaps from tq_disk?
1234                                  */
1235
1236                                 /* for atari only */
1237                                 ide_release_lock();
1238                                 hwgroup->busy = 0;
1239                         }
1240
1241                         /* no more work for this hwgroup (for now) */
1242                         return;
1243                 }
1244         again:
1245                 hwif = HWIF(drive);
1246                 if (hwgroup->hwif->sharing_irq &&
1247                     hwif != hwgroup->hwif &&
1248                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1249                         /* set nIEN for previous hwif */
1250                         SELECT_INTERRUPT(drive);
1251                 }
1252                 hwgroup->hwif = hwif;
1253                 hwgroup->drive = drive;
1254                 drive->sleeping = 0;
1255                 drive->service_start = jiffies;
1256
1257                 if (blk_queue_plugged(drive->queue)) {
1258                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1259                         break;
1260                 }
1261
1262                 /*
1263                  * we know that the queue isn't empty, but this can happen
1264                  * if the q->prep_rq_fn() decides to kill a request
1265                  */
1266                 rq = elv_next_request(drive->queue);
1267                 if (!rq) {
1268                         hwgroup->busy = 0;
1269                         break;
1270                 }
1271
1272                 /*
1273                  * Sanity: don't accept a request that isn't a PM request
1274                  * if we are currently power managed. This is very important as
1275                  * blk_stop_queue() doesn't prevent the elv_next_request()
1276                  * above to return us whatever is in the queue. Since we call
1277                  * ide_do_request() ourselves, we end up taking requests while
1278                  * the queue is blocked...
1279                  * 
1280                  * We let requests forced at head of queue with ide-preempt
1281                  * though. I hope that doesn't happen too much, hopefully not
1282                  * unless the subdriver triggers such a thing in its own PM
1283                  * state machine.
1284                  *
1285                  * We count how many times we loop here to make sure we service
1286                  * all drives in the hwgroup without looping for ever
1287                  */
1288                 if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
1289                         drive = drive->next ? drive->next : hwgroup->drive;
1290                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1291                                 goto again;
1292                         /* We clear busy, there should be no pending ATA command at this point. */
1293                         hwgroup->busy = 0;
1294                         break;
1295                 }
1296
1297                 hwgroup->rq = rq;
1298
1299                 /*
1300                  * Some systems have trouble with IDE IRQs arriving while
1301                  * the driver is still setting things up.  So, here we disable
1302                  * the IRQ used by this interface while the request is being started.
1303                  * This may look bad at first, but pretty much the same thing
1304                  * happens anyway when any interrupt comes in, IDE or otherwise
1305                  *  -- the kernel masks the IRQ while it is being handled.
1306                  */
1307                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1308                         disable_irq_nosync(hwif->irq);
1309                 spin_unlock(&ide_lock);
1310                 local_irq_enable_in_hardirq();
1311                         /* allow other IRQs while we start this request */
1312                 startstop = start_request(drive, rq);
1313                 spin_lock_irq(&ide_lock);
1314                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1315                         enable_irq(hwif->irq);
1316                 if (startstop == ide_stopped)
1317                         hwgroup->busy = 0;
1318         }
1319 }
1320
1321 /*
1322  * Passes the stuff to ide_do_request
1323  */
1324 void do_ide_request(request_queue_t *q)
1325 {
1326         ide_drive_t *drive = q->queuedata;
1327
1328         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1329 }
1330
1331 /*
1332  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1333  * retry the current request in pio mode instead of risking tossing it
1334  * all away
1335  */
1336 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1337 {
1338         ide_hwif_t *hwif = HWIF(drive);
1339         struct request *rq;
1340         ide_startstop_t ret = ide_stopped;
1341
1342         /*
1343          * end current dma transaction
1344          */
1345
1346         if (error < 0) {
1347                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1348                 (void)HWIF(drive)->ide_dma_end(drive);
1349                 ret = ide_error(drive, "dma timeout error",
1350                                                 hwif->INB(IDE_STATUS_REG));
1351         } else {
1352                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1353                 hwif->dma_timeout(drive);
1354         }
1355
1356         /*
1357          * disable dma for now, but remember that we did so because of
1358          * a timeout -- we'll reenable after we finish this next request
1359          * (or rather the first chunk of it) in pio.
1360          */
1361         drive->retry_pio++;
1362         drive->state = DMA_PIO_RETRY;
1363         hwif->dma_off_quietly(drive);
1364
1365         /*
1366          * un-busy drive etc (hwgroup->busy is cleared on return) and
1367          * make sure request is sane
1368          */
1369         rq = HWGROUP(drive)->rq;
1370
1371         if (!rq)
1372                 goto out;
1373
1374         HWGROUP(drive)->rq = NULL;
1375
1376         rq->errors = 0;
1377
1378         if (!rq->bio)
1379                 goto out;
1380
1381         rq->sector = rq->bio->bi_sector;
1382         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1383         rq->hard_cur_sectors = rq->current_nr_sectors;
1384         rq->buffer = bio_data(rq->bio);
1385 out:
1386         return ret;
1387 }
1388
1389 /**
1390  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1391  *      @data: timer callback magic (hwgroup)
1392  *
1393  *      An IDE command has timed out before the expected drive return
1394  *      occurred. At this point we attempt to clean up the current
1395  *      mess. If the current handler includes an expiry handler then
1396  *      we invoke the expiry handler, and providing it is happy the
1397  *      work is done. If that fails we apply generic recovery rules
1398  *      invoking the handler and checking the drive DMA status. We
1399  *      have an excessively incestuous relationship with the DMA
1400  *      logic that wants cleaning up.
1401  */
1402  
1403 void ide_timer_expiry (unsigned long data)
1404 {
1405         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1406         ide_handler_t   *handler;
1407         ide_expiry_t    *expiry;
1408         unsigned long   flags;
1409         unsigned long   wait = -1;
1410
1411         spin_lock_irqsave(&ide_lock, flags);
1412
1413         if (((handler = hwgroup->handler) == NULL) ||
1414             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
1415                 /*
1416                  * Either a marginal timeout occurred
1417                  * (got the interrupt just as timer expired),
1418                  * or we were "sleeping" to give other devices a chance.
1419                  * Either way, we don't really want to complain about anything.
1420                  */
1421                 if (hwgroup->sleeping) {
1422                         hwgroup->sleeping = 0;
1423                         hwgroup->busy = 0;
1424                 }
1425         } else {
1426                 ide_drive_t *drive = hwgroup->drive;
1427                 if (!drive) {
1428                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1429                         hwgroup->handler = NULL;
1430                 } else {
1431                         ide_hwif_t *hwif;
1432                         ide_startstop_t startstop = ide_stopped;
1433                         if (!hwgroup->busy) {
1434                                 hwgroup->busy = 1;      /* paranoia */
1435                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1436                         }
1437                         if ((expiry = hwgroup->expiry) != NULL) {
1438                                 /* continue */
1439                                 if ((wait = expiry(drive)) > 0) {
1440                                         /* reset timer */
1441                                         hwgroup->timer.expires  = jiffies + wait;
1442                                         hwgroup->req_gen_timer = hwgroup->req_gen;
1443                                         add_timer(&hwgroup->timer);
1444                                         spin_unlock_irqrestore(&ide_lock, flags);
1445                                         return;
1446                                 }
1447                         }
1448                         hwgroup->handler = NULL;
1449                         /*
1450                          * We need to simulate a real interrupt when invoking
1451                          * the handler() function, which means we need to
1452                          * globally mask the specific IRQ:
1453                          */
1454                         spin_unlock(&ide_lock);
1455                         hwif  = HWIF(drive);
1456 #if DISABLE_IRQ_NOSYNC
1457                         disable_irq_nosync(hwif->irq);
1458 #else
1459                         /* disable_irq_nosync ?? */
1460                         disable_irq(hwif->irq);
1461 #endif /* DISABLE_IRQ_NOSYNC */
1462                         /* local CPU only,
1463                          * as if we were handling an interrupt */
1464                         local_irq_disable();
1465                         if (hwgroup->polling) {
1466                                 startstop = handler(drive);
1467                         } else if (drive_is_ready(drive)) {
1468                                 if (drive->waiting_for_dma)
1469                                         hwgroup->hwif->dma_lost_irq(drive);
1470                                 (void)ide_ack_intr(hwif);
1471                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1472                                 startstop = handler(drive);
1473                         } else {
1474                                 if (drive->waiting_for_dma) {
1475                                         startstop = ide_dma_timeout_retry(drive, wait);
1476                                 } else
1477                                         startstop =
1478                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1479                         }
1480                         drive->service_time = jiffies - drive->service_start;
1481                         spin_lock_irq(&ide_lock);
1482                         enable_irq(hwif->irq);
1483                         if (startstop == ide_stopped)
1484                                 hwgroup->busy = 0;
1485                 }
1486         }
1487         ide_do_request(hwgroup, IDE_NO_IRQ);
1488         spin_unlock_irqrestore(&ide_lock, flags);
1489 }
1490
1491 /**
1492  *      unexpected_intr         -       handle an unexpected IDE interrupt
1493  *      @irq: interrupt line
1494  *      @hwgroup: hwgroup being processed
1495  *
1496  *      There's nothing really useful we can do with an unexpected interrupt,
1497  *      other than reading the status register (to clear it), and logging it.
1498  *      There should be no way that an irq can happen before we're ready for it,
1499  *      so we needn't worry much about losing an "important" interrupt here.
1500  *
1501  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1502  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1503  *      looks "good", we just ignore the interrupt completely.
1504  *
1505  *      This routine assumes __cli() is in effect when called.
1506  *
1507  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1508  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1509  *      we could screw up by interfering with a new request being set up for 
1510  *      irq15.
1511  *
1512  *      In reality, this is a non-issue.  The new command is not sent unless 
1513  *      the drive is ready to accept one, in which case we know the drive is
1514  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1515  *      before completing the issuance of any new drive command, so we will not
1516  *      be accidentally invoked as a result of any valid command completion
1517  *      interrupt.
1518  *
1519  *      Note that we must walk the entire hwgroup here. We know which hwif
1520  *      is doing the current command, but we don't know which hwif burped
1521  *      mysteriously.
1522  */
1523  
1524 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1525 {
1526         u8 stat;
1527         ide_hwif_t *hwif = hwgroup->hwif;
1528
1529         /*
1530          * handle the unexpected interrupt
1531          */
1532         do {
1533                 if (hwif->irq == irq) {
1534                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1535                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1536                                 /* Try to not flood the console with msgs */
1537                                 static unsigned long last_msgtime, count;
1538                                 ++count;
1539                                 if (time_after(jiffies, last_msgtime + HZ)) {
1540                                         last_msgtime = jiffies;
1541                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1542                                                 "status=0x%02x, count=%ld\n",
1543                                                 hwif->name,
1544                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1545                                 }
1546                         }
1547                 }
1548         } while ((hwif = hwif->next) != hwgroup->hwif);
1549 }
1550
1551 /**
1552  *      ide_intr        -       default IDE interrupt handler
1553  *      @irq: interrupt number
1554  *      @dev_id: hwif group
1555  *      @regs: unused weirdness from the kernel irq layer
1556  *
1557  *      This is the default IRQ handler for the IDE layer. You should
1558  *      not need to override it. If you do be aware it is subtle in
1559  *      places
1560  *
1561  *      hwgroup->hwif is the interface in the group currently performing
1562  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1563  *      the IRQ handler to call. As we issue a command the handlers
1564  *      step through multiple states, reassigning the handler to the
1565  *      next step in the process. Unlike a smart SCSI controller IDE
1566  *      expects the main processor to sequence the various transfer
1567  *      stages. We also manage a poll timer to catch up with most
1568  *      timeout situations. There are still a few where the handlers
1569  *      don't ever decide to give up.
1570  *
1571  *      The handler eventually returns ide_stopped to indicate the
1572  *      request completed. At this point we issue the next request
1573  *      on the hwgroup and the process begins again.
1574  */
1575  
1576 irqreturn_t ide_intr (int irq, void *dev_id)
1577 {
1578         unsigned long flags;
1579         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1580         ide_hwif_t *hwif;
1581         ide_drive_t *drive;
1582         ide_handler_t *handler;
1583         ide_startstop_t startstop;
1584
1585         spin_lock_irqsave(&ide_lock, flags);
1586         hwif = hwgroup->hwif;
1587
1588         if (!ide_ack_intr(hwif)) {
1589                 spin_unlock_irqrestore(&ide_lock, flags);
1590                 return IRQ_NONE;
1591         }
1592
1593         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1594                 /*
1595                  * Not expecting an interrupt from this drive.
1596                  * That means this could be:
1597                  *      (1) an interrupt from another PCI device
1598                  *      sharing the same PCI INT# as us.
1599                  * or   (2) a drive just entered sleep or standby mode,
1600                  *      and is interrupting to let us know.
1601                  * or   (3) a spurious interrupt of unknown origin.
1602                  *
1603                  * For PCI, we cannot tell the difference,
1604                  * so in that case we just ignore it and hope it goes away.
1605                  *
1606                  * FIXME: unexpected_intr should be hwif-> then we can
1607                  * remove all the ifdef PCI crap
1608                  */
1609 #ifdef CONFIG_BLK_DEV_IDEPCI
1610                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1611 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1612                 {
1613                         /*
1614                          * Probably not a shared PCI interrupt,
1615                          * so we can safely try to do something about it:
1616                          */
1617                         unexpected_intr(irq, hwgroup);
1618 #ifdef CONFIG_BLK_DEV_IDEPCI
1619                 } else {
1620                         /*
1621                          * Whack the status register, just in case
1622                          * we have a leftover pending IRQ.
1623                          */
1624                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1625 #endif /* CONFIG_BLK_DEV_IDEPCI */
1626                 }
1627                 spin_unlock_irqrestore(&ide_lock, flags);
1628                 return IRQ_NONE;
1629         }
1630         drive = hwgroup->drive;
1631         if (!drive) {
1632                 /*
1633                  * This should NEVER happen, and there isn't much
1634                  * we could do about it here.
1635                  *
1636                  * [Note - this can occur if the drive is hot unplugged]
1637                  */
1638                 spin_unlock_irqrestore(&ide_lock, flags);
1639                 return IRQ_HANDLED;
1640         }
1641         if (!drive_is_ready(drive)) {
1642                 /*
1643                  * This happens regularly when we share a PCI IRQ with
1644                  * another device.  Unfortunately, it can also happen
1645                  * with some buggy drives that trigger the IRQ before
1646                  * their status register is up to date.  Hopefully we have
1647                  * enough advance overhead that the latter isn't a problem.
1648                  */
1649                 spin_unlock_irqrestore(&ide_lock, flags);
1650                 return IRQ_NONE;
1651         }
1652         if (!hwgroup->busy) {
1653                 hwgroup->busy = 1;      /* paranoia */
1654                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1655         }
1656         hwgroup->handler = NULL;
1657         hwgroup->req_gen++;
1658         del_timer(&hwgroup->timer);
1659         spin_unlock(&ide_lock);
1660
1661         /* Some controllers might set DMA INTR no matter DMA or PIO;
1662          * bmdma status might need to be cleared even for
1663          * PIO interrupts to prevent spurious/lost irq.
1664          */
1665         if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
1666                 /* ide_dma_end() needs bmdma status for error checking.
1667                  * So, skip clearing bmdma status here and leave it
1668                  * to ide_dma_end() if this is dma interrupt.
1669                  */
1670                 hwif->ide_dma_clear_irq(drive);
1671
1672         if (drive->unmask)
1673                 local_irq_enable_in_hardirq();
1674         /* service this interrupt, may set handler for next interrupt */
1675         startstop = handler(drive);
1676         spin_lock_irq(&ide_lock);
1677
1678         /*
1679          * Note that handler() may have set things up for another
1680          * interrupt to occur soon, but it cannot happen until
1681          * we exit from this routine, because it will be the
1682          * same irq as is currently being serviced here, and Linux
1683          * won't allow another of the same (on any CPU) until we return.
1684          */
1685         drive->service_time = jiffies - drive->service_start;
1686         if (startstop == ide_stopped) {
1687                 if (hwgroup->handler == NULL) { /* paranoia */
1688                         hwgroup->busy = 0;
1689                         ide_do_request(hwgroup, hwif->irq);
1690                 } else {
1691                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1692                                 "on exit\n", drive->name);
1693                 }
1694         }
1695         spin_unlock_irqrestore(&ide_lock, flags);
1696         return IRQ_HANDLED;
1697 }
1698
1699 /**
1700  *      ide_init_drive_cmd      -       initialize a drive command request
1701  *      @rq: request object
1702  *
1703  *      Initialize a request before we fill it in and send it down to
1704  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1705  *      now it doesn't do a lot, but if that changes abusers will have a
1706  *      nasty surprise.
1707  */
1708
1709 void ide_init_drive_cmd (struct request *rq)
1710 {
1711         memset(rq, 0, sizeof(*rq));
1712         rq->cmd_type = REQ_TYPE_ATA_CMD;
1713         rq->ref_count = 1;
1714 }
1715
1716 EXPORT_SYMBOL(ide_init_drive_cmd);
1717
1718 /**
1719  *      ide_do_drive_cmd        -       issue IDE special command
1720  *      @drive: device to issue command
1721  *      @rq: request to issue
1722  *      @action: action for processing
1723  *
1724  *      This function issues a special IDE device request
1725  *      onto the request queue.
1726  *
1727  *      If action is ide_wait, then the rq is queued at the end of the
1728  *      request queue, and the function sleeps until it has been processed.
1729  *      This is for use when invoked from an ioctl handler.
1730  *
1731  *      If action is ide_preempt, then the rq is queued at the head of
1732  *      the request queue, displacing the currently-being-processed
1733  *      request and this function returns immediately without waiting
1734  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1735  *      intended for careful use by the ATAPI tape/cdrom driver code.
1736  *
1737  *      If action is ide_end, then the rq is queued at the end of the
1738  *      request queue, and the function returns immediately without waiting
1739  *      for the new rq to be completed. This is again intended for careful
1740  *      use by the ATAPI tape/cdrom driver code.
1741  */
1742  
1743 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1744 {
1745         unsigned long flags;
1746         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1747         DECLARE_COMPLETION_ONSTACK(wait);
1748         int where = ELEVATOR_INSERT_BACK, err;
1749         int must_wait = (action == ide_wait || action == ide_head_wait);
1750
1751         rq->errors = 0;
1752
1753         /*
1754          * we need to hold an extra reference to request for safe inspection
1755          * after completion
1756          */
1757         if (must_wait) {
1758                 rq->ref_count++;
1759                 rq->end_io_data = &wait;
1760                 rq->end_io = blk_end_sync_rq;
1761         }
1762
1763         spin_lock_irqsave(&ide_lock, flags);
1764         if (action == ide_preempt)
1765                 hwgroup->rq = NULL;
1766         if (action == ide_preempt || action == ide_head_wait) {
1767                 where = ELEVATOR_INSERT_FRONT;
1768                 rq->cmd_flags |= REQ_PREEMPT;
1769         }
1770         __elv_add_request(drive->queue, rq, where, 0);
1771         ide_do_request(hwgroup, IDE_NO_IRQ);
1772         spin_unlock_irqrestore(&ide_lock, flags);
1773
1774         err = 0;
1775         if (must_wait) {
1776                 wait_for_completion(&wait);
1777                 if (rq->errors)
1778                         err = -EIO;
1779
1780                 blk_put_request(rq);
1781         }
1782
1783         return err;
1784 }
1785
1786 EXPORT_SYMBOL(ide_do_drive_cmd);