Merge branch 'i2c-next' of git://aeryn.fluff.org.uk/bjdooks/linux
[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/hdreg.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 #include <linux/bitops.h>
52
53 #include <asm/byteorder.h>
54 #include <asm/irq.h>
55 #include <asm/uaccess.h>
56 #include <asm/io.h>
57
58 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
59                              int uptodate, unsigned int nr_bytes, int dequeue)
60 {
61         int ret = 1;
62         int error = 0;
63
64         if (uptodate <= 0)
65                 error = uptodate ? uptodate : -EIO;
66
67         /*
68          * if failfast is set on a request, override number of sectors and
69          * complete the whole request right now
70          */
71         if (blk_noretry_request(rq) && error)
72                 nr_bytes = rq->hard_nr_sectors << 9;
73
74         if (!blk_fs_request(rq) && error && !rq->errors)
75                 rq->errors = -EIO;
76
77         /*
78          * decide whether to reenable DMA -- 3 is a random magic for now,
79          * if we DMA timeout more than 3 times, just stay in PIO
80          */
81         if ((drive->dev_flags & IDE_DFLAG_DMA_PIO_RETRY) &&
82             drive->retry_pio <= 3) {
83                 drive->dev_flags &= ~IDE_DFLAG_DMA_PIO_RETRY;
84                 ide_dma_on(drive);
85         }
86
87         if (!blk_end_request(rq, error, nr_bytes))
88                 ret = 0;
89
90         if (ret == 0 && dequeue)
91                 drive->hwif->hwgroup->rq = NULL;
92
93         return ret;
94 }
95
96 /**
97  *      ide_end_request         -       complete an IDE I/O
98  *      @drive: IDE device for the I/O
99  *      @uptodate:
100  *      @nr_sectors: number of sectors completed
101  *
102  *      This is our end_request wrapper function. We complete the I/O
103  *      update random number input and dequeue the request, which if
104  *      it was tagged may be out of order.
105  */
106
107 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
108 {
109         unsigned int nr_bytes = nr_sectors << 9;
110         struct request *rq = drive->hwif->hwgroup->rq;
111
112         if (!nr_bytes) {
113                 if (blk_pc_request(rq))
114                         nr_bytes = rq->data_len;
115                 else
116                         nr_bytes = rq->hard_cur_sectors << 9;
117         }
118
119         return __ide_end_request(drive, rq, uptodate, nr_bytes, 1);
120 }
121 EXPORT_SYMBOL(ide_end_request);
122
123 /**
124  *      ide_end_dequeued_request        -       complete an IDE I/O
125  *      @drive: IDE device for the I/O
126  *      @uptodate:
127  *      @nr_sectors: number of sectors completed
128  *
129  *      Complete an I/O that is no longer on the request queue. This
130  *      typically occurs when we pull the request and issue a REQUEST_SENSE.
131  *      We must still finish the old request but we must not tamper with the
132  *      queue in the meantime.
133  *
134  *      NOTE: This path does not handle barrier, but barrier is not supported
135  *      on ide-cd anyway.
136  */
137
138 int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
139                              int uptodate, int nr_sectors)
140 {
141         BUG_ON(!blk_rq_started(rq));
142
143         return __ide_end_request(drive, rq, uptodate, nr_sectors << 9, 0);
144 }
145 EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
146
147 /**
148  *      ide_end_drive_cmd       -       end an explicit drive command
149  *      @drive: command 
150  *      @stat: status bits
151  *      @err: error bits
152  *
153  *      Clean up after success/failure of an explicit drive command.
154  *      These get thrown onto the queue so they are synchronized with
155  *      real I/O operations on the drive.
156  *
157  *      In LBA48 mode we have to read the register set twice to get
158  *      all the extra information out.
159  */
160  
161 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
162 {
163         ide_hwgroup_t *hwgroup = drive->hwif->hwgroup;
164         struct request *rq = hwgroup->rq;
165
166         if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
167                 ide_task_t *task = (ide_task_t *)rq->special;
168
169                 if (task) {
170                         struct ide_taskfile *tf = &task->tf;
171
172                         tf->error = err;
173                         tf->status = stat;
174
175                         drive->hwif->tp_ops->tf_read(drive, task);
176
177                         if (task->tf_flags & IDE_TFLAG_DYN)
178                                 kfree(task);
179                 }
180         } else if (blk_pm_request(rq)) {
181                 struct request_pm_state *pm = rq->data;
182
183                 ide_complete_power_step(drive, rq);
184                 if (pm->pm_step == IDE_PM_COMPLETED)
185                         ide_complete_pm_request(drive, rq);
186                 return;
187         }
188
189         hwgroup->rq = NULL;
190
191         rq->errors = err;
192
193         if (unlikely(blk_end_request(rq, (rq->errors ? -EIO : 0),
194                                      blk_rq_bytes(rq))))
195                 BUG();
196 }
197 EXPORT_SYMBOL(ide_end_drive_cmd);
198
199 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
200 {
201         if (rq->rq_disk) {
202                 ide_driver_t *drv;
203
204                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
205                 drv->end_request(drive, 0, 0);
206         } else
207                 ide_end_request(drive, 0, 0);
208 }
209
210 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
211 {
212         ide_hwif_t *hwif = drive->hwif;
213
214         if ((stat & ATA_BUSY) ||
215             ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
216                 /* other bits are useless when BUSY */
217                 rq->errors |= ERROR_RESET;
218         } else if (stat & ATA_ERR) {
219                 /* err has different meaning on cdrom and tape */
220                 if (err == ATA_ABORTED) {
221                         if ((drive->dev_flags & IDE_DFLAG_LBA) &&
222                             /* some newer drives don't support ATA_CMD_INIT_DEV_PARAMS */
223                             hwif->tp_ops->read_status(hwif) == ATA_CMD_INIT_DEV_PARAMS)
224                                 return ide_stopped;
225                 } else if ((err & BAD_CRC) == BAD_CRC) {
226                         /* UDMA crc error, just retry the operation */
227                         drive->crc_count++;
228                 } else if (err & (ATA_BBK | ATA_UNC)) {
229                         /* retries won't help these */
230                         rq->errors = ERROR_MAX;
231                 } else if (err & ATA_TRK0NF) {
232                         /* help it find track zero */
233                         rq->errors |= ERROR_RECAL;
234                 }
235         }
236
237         if ((stat & ATA_DRQ) && rq_data_dir(rq) == READ &&
238             (hwif->host_flags & IDE_HFLAG_ERROR_STOPS_FIFO) == 0) {
239                 int nsect = drive->mult_count ? drive->mult_count : 1;
240
241                 ide_pad_transfer(drive, READ, nsect * SECTOR_SIZE);
242         }
243
244         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
245                 ide_kill_rq(drive, rq);
246                 return ide_stopped;
247         }
248
249         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
250                 rq->errors |= ERROR_RESET;
251
252         if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
253                 ++rq->errors;
254                 return ide_do_reset(drive);
255         }
256
257         if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
258                 drive->special.b.recalibrate = 1;
259
260         ++rq->errors;
261
262         return ide_stopped;
263 }
264
265 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
266 {
267         ide_hwif_t *hwif = drive->hwif;
268
269         if ((stat & ATA_BUSY) ||
270             ((stat & ATA_DF) && (drive->dev_flags & IDE_DFLAG_NOWERR) == 0)) {
271                 /* other bits are useless when BUSY */
272                 rq->errors |= ERROR_RESET;
273         } else {
274                 /* add decoding error stuff */
275         }
276
277         if (hwif->tp_ops->read_status(hwif) & (ATA_BUSY | ATA_DRQ))
278                 /* force an abort */
279                 hwif->tp_ops->exec_command(hwif, ATA_CMD_IDLEIMMEDIATE);
280
281         if (rq->errors >= ERROR_MAX) {
282                 ide_kill_rq(drive, rq);
283         } else {
284                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
285                         ++rq->errors;
286                         return ide_do_reset(drive);
287                 }
288                 ++rq->errors;
289         }
290
291         return ide_stopped;
292 }
293
294 ide_startstop_t
295 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
296 {
297         if (drive->media == ide_disk)
298                 return ide_ata_error(drive, rq, stat, err);
299         return ide_atapi_error(drive, rq, stat, err);
300 }
301
302 EXPORT_SYMBOL_GPL(__ide_error);
303
304 /**
305  *      ide_error       -       handle an error on the IDE
306  *      @drive: drive the error occurred on
307  *      @msg: message to report
308  *      @stat: status bits
309  *
310  *      ide_error() takes action based on the error returned by the drive.
311  *      For normal I/O that may well include retries. We deal with
312  *      both new-style (taskfile) and old style command handling here.
313  *      In the case of taskfile command handling there is work left to
314  *      do
315  */
316  
317 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
318 {
319         struct request *rq;
320         u8 err;
321
322         err = ide_dump_status(drive, msg, stat);
323
324         if ((rq = HWGROUP(drive)->rq) == NULL)
325                 return ide_stopped;
326
327         /* retry only "normal" I/O: */
328         if (!blk_fs_request(rq)) {
329                 rq->errors = 1;
330                 ide_end_drive_cmd(drive, stat, err);
331                 return ide_stopped;
332         }
333
334         if (rq->rq_disk) {
335                 ide_driver_t *drv;
336
337                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
338                 return drv->error(drive, rq, stat, err);
339         } else
340                 return __ide_error(drive, rq, stat, err);
341 }
342
343 EXPORT_SYMBOL_GPL(ide_error);
344
345 static void ide_tf_set_specify_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
346 {
347         tf->nsect   = drive->sect;
348         tf->lbal    = drive->sect;
349         tf->lbam    = drive->cyl;
350         tf->lbah    = drive->cyl >> 8;
351         tf->device  = (drive->head - 1) | drive->select;
352         tf->command = ATA_CMD_INIT_DEV_PARAMS;
353 }
354
355 static void ide_tf_set_restore_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
356 {
357         tf->nsect   = drive->sect;
358         tf->command = ATA_CMD_RESTORE;
359 }
360
361 static void ide_tf_set_setmult_cmd(ide_drive_t *drive, struct ide_taskfile *tf)
362 {
363         tf->nsect   = drive->mult_req;
364         tf->command = ATA_CMD_SET_MULTI;
365 }
366
367 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
368 {
369         special_t *s = &drive->special;
370         ide_task_t args;
371
372         memset(&args, 0, sizeof(ide_task_t));
373         args.data_phase = TASKFILE_NO_DATA;
374
375         if (s->b.set_geometry) {
376                 s->b.set_geometry = 0;
377                 ide_tf_set_specify_cmd(drive, &args.tf);
378         } else if (s->b.recalibrate) {
379                 s->b.recalibrate = 0;
380                 ide_tf_set_restore_cmd(drive, &args.tf);
381         } else if (s->b.set_multmode) {
382                 s->b.set_multmode = 0;
383                 ide_tf_set_setmult_cmd(drive, &args.tf);
384         } else if (s->all) {
385                 int special = s->all;
386                 s->all = 0;
387                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
388                 return ide_stopped;
389         }
390
391         args.tf_flags = IDE_TFLAG_TF | IDE_TFLAG_DEVICE |
392                         IDE_TFLAG_CUSTOM_HANDLER;
393
394         do_rw_taskfile(drive, &args);
395
396         return ide_started;
397 }
398
399 /**
400  *      do_special              -       issue some special commands
401  *      @drive: drive the command is for
402  *
403  *      do_special() is used to issue ATA_CMD_INIT_DEV_PARAMS,
404  *      ATA_CMD_RESTORE and ATA_CMD_SET_MULTI commands to a drive.
405  *
406  *      It used to do much more, but has been scaled back.
407  */
408
409 static ide_startstop_t do_special (ide_drive_t *drive)
410 {
411         special_t *s = &drive->special;
412
413 #ifdef DEBUG
414         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
415 #endif
416         if (drive->media == ide_disk)
417                 return ide_disk_special(drive);
418
419         s->all = 0;
420         drive->mult_req = 0;
421         return ide_stopped;
422 }
423
424 void ide_map_sg(ide_drive_t *drive, struct request *rq)
425 {
426         ide_hwif_t *hwif = drive->hwif;
427         struct scatterlist *sg = hwif->sg_table;
428
429         if (rq->cmd_type != REQ_TYPE_ATA_TASKFILE) {
430                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
431         } else {
432                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
433                 hwif->sg_nents = 1;
434         }
435 }
436
437 EXPORT_SYMBOL_GPL(ide_map_sg);
438
439 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
440 {
441         ide_hwif_t *hwif = drive->hwif;
442
443         hwif->nsect = hwif->nleft = rq->nr_sectors;
444         hwif->cursg_ofs = 0;
445         hwif->cursg = NULL;
446 }
447
448 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
449
450 /**
451  *      execute_drive_command   -       issue special drive command
452  *      @drive: the drive to issue the command on
453  *      @rq: the request structure holding the command
454  *
455  *      execute_drive_cmd() issues a special drive command,  usually 
456  *      initiated by ioctl() from the external hdparm program. The
457  *      command can be a drive command, drive task or taskfile 
458  *      operation. Weirdly you can call it with NULL to wait for
459  *      all commands to finish. Don't do this as that is due to change
460  */
461
462 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
463                 struct request *rq)
464 {
465         ide_hwif_t *hwif = HWIF(drive);
466         ide_task_t *task = rq->special;
467
468         if (task) {
469                 hwif->data_phase = task->data_phase;
470
471                 switch (hwif->data_phase) {
472                 case TASKFILE_MULTI_OUT:
473                 case TASKFILE_OUT:
474                 case TASKFILE_MULTI_IN:
475                 case TASKFILE_IN:
476                         ide_init_sg_cmd(drive, rq);
477                         ide_map_sg(drive, rq);
478                 default:
479                         break;
480                 }
481
482                 return do_rw_taskfile(drive, task);
483         }
484
485         /*
486          * NULL is actually a valid way of waiting for
487          * all current requests to be flushed from the queue.
488          */
489 #ifdef DEBUG
490         printk("%s: DRIVE_CMD (null)\n", drive->name);
491 #endif
492         ide_end_drive_cmd(drive, hwif->tp_ops->read_status(hwif),
493                           ide_read_error(drive));
494
495         return ide_stopped;
496 }
497
498 int ide_devset_execute(ide_drive_t *drive, const struct ide_devset *setting,
499                        int arg)
500 {
501         struct request_queue *q = drive->queue;
502         struct request *rq;
503         int ret = 0;
504
505         if (!(setting->flags & DS_SYNC))
506                 return setting->set(drive, arg);
507
508         rq = blk_get_request(q, READ, __GFP_WAIT);
509         rq->cmd_type = REQ_TYPE_SPECIAL;
510         rq->cmd_len = 5;
511         rq->cmd[0] = REQ_DEVSET_EXEC;
512         *(int *)&rq->cmd[1] = arg;
513         rq->special = setting->set;
514
515         if (blk_execute_rq(q, NULL, rq, 0))
516                 ret = rq->errors;
517         blk_put_request(rq);
518
519         return ret;
520 }
521 EXPORT_SYMBOL_GPL(ide_devset_execute);
522
523 static ide_startstop_t ide_special_rq(ide_drive_t *drive, struct request *rq)
524 {
525         u8 cmd = rq->cmd[0];
526
527         if (cmd == REQ_PARK_HEADS || cmd == REQ_UNPARK_HEADS) {
528                 ide_task_t task;
529                 struct ide_taskfile *tf = &task.tf;
530
531                 memset(&task, 0, sizeof(task));
532                 if (cmd == REQ_PARK_HEADS) {
533                         drive->sleep = *(unsigned long *)rq->special;
534                         drive->dev_flags |= IDE_DFLAG_SLEEPING;
535                         tf->command = ATA_CMD_IDLEIMMEDIATE;
536                         tf->feature = 0x44;
537                         tf->lbal = 0x4c;
538                         tf->lbam = 0x4e;
539                         tf->lbah = 0x55;
540                         task.tf_flags |= IDE_TFLAG_CUSTOM_HANDLER;
541                 } else          /* cmd == REQ_UNPARK_HEADS */
542                         tf->command = ATA_CMD_CHK_POWER;
543
544                 task.tf_flags |= IDE_TFLAG_TF | IDE_TFLAG_DEVICE;
545                 task.rq = rq;
546                 drive->hwif->data_phase = task.data_phase = TASKFILE_NO_DATA;
547                 return do_rw_taskfile(drive, &task);
548         }
549
550         switch (cmd) {
551         case REQ_DEVSET_EXEC:
552         {
553                 int err, (*setfunc)(ide_drive_t *, int) = rq->special;
554
555                 err = setfunc(drive, *(int *)&rq->cmd[1]);
556                 if (err)
557                         rq->errors = err;
558                 else
559                         err = 1;
560                 ide_end_request(drive, err, 0);
561                 return ide_stopped;
562         }
563         case REQ_DRIVE_RESET:
564                 return ide_do_reset(drive);
565         default:
566                 blk_dump_rq_flags(rq, "ide_special_rq - bad request");
567                 ide_end_request(drive, 0, 0);
568                 return ide_stopped;
569         }
570 }
571
572 /**
573  *      start_request   -       start of I/O and command issuing for IDE
574  *
575  *      start_request() initiates handling of a new I/O request. It
576  *      accepts commands and I/O (read/write) requests.
577  *
578  *      FIXME: this function needs a rename
579  */
580  
581 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
582 {
583         ide_startstop_t startstop;
584
585         BUG_ON(!blk_rq_started(rq));
586
587 #ifdef DEBUG
588         printk("%s: start_request: current=0x%08lx\n",
589                 HWIF(drive)->name, (unsigned long) rq);
590 #endif
591
592         /* bail early if we've exceeded max_failures */
593         if (drive->max_failures && (drive->failures > drive->max_failures)) {
594                 rq->cmd_flags |= REQ_FAILED;
595                 goto kill_rq;
596         }
597
598         if (blk_pm_request(rq))
599                 ide_check_pm_state(drive, rq);
600
601         SELECT_DRIVE(drive);
602         if (ide_wait_stat(&startstop, drive, drive->ready_stat,
603                           ATA_BUSY | ATA_DRQ, WAIT_READY)) {
604                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
605                 return startstop;
606         }
607         if (!drive->special.all) {
608                 ide_driver_t *drv;
609
610                 /*
611                  * We reset the drive so we need to issue a SETFEATURES.
612                  * Do it _after_ do_special() restored device parameters.
613                  */
614                 if (drive->current_speed == 0xff)
615                         ide_config_drive_speed(drive, drive->desired_speed);
616
617                 if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
618                         return execute_drive_cmd(drive, rq);
619                 else if (blk_pm_request(rq)) {
620                         struct request_pm_state *pm = rq->data;
621 #ifdef DEBUG_PM
622                         printk("%s: start_power_step(step: %d)\n",
623                                 drive->name, pm->pm_step);
624 #endif
625                         startstop = ide_start_power_step(drive, rq);
626                         if (startstop == ide_stopped &&
627                             pm->pm_step == IDE_PM_COMPLETED)
628                                 ide_complete_pm_request(drive, rq);
629                         return startstop;
630                 } else if (!rq->rq_disk && blk_special_request(rq))
631                         /*
632                          * TODO: Once all ULDs have been modified to
633                          * check for specific op codes rather than
634                          * blindly accepting any special request, the
635                          * check for ->rq_disk above may be replaced
636                          * by a more suitable mechanism or even
637                          * dropped entirely.
638                          */
639                         return ide_special_rq(drive, rq);
640
641                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
642
643                 return drv->do_request(drive, rq, rq->sector);
644         }
645         return do_special(drive);
646 kill_rq:
647         ide_kill_rq(drive, rq);
648         return ide_stopped;
649 }
650
651 /**
652  *      ide_stall_queue         -       pause an IDE device
653  *      @drive: drive to stall
654  *      @timeout: time to stall for (jiffies)
655  *
656  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
657  *      to the hwgroup by sleeping for timeout jiffies.
658  */
659  
660 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
661 {
662         if (timeout > WAIT_WORSTCASE)
663                 timeout = WAIT_WORSTCASE;
664         drive->sleep = timeout + jiffies;
665         drive->dev_flags |= IDE_DFLAG_SLEEPING;
666 }
667 EXPORT_SYMBOL(ide_stall_queue);
668
669 /*
670  * Issue a new request to a drive from hwgroup
671  *
672  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
673  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
674  * may have both interfaces in a single hwgroup to "serialize" access.
675  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
676  * together into one hwgroup for serialized access.
677  *
678  * Note also that several hwgroups can end up sharing a single IRQ,
679  * possibly along with many other devices.  This is especially common in
680  * PCI-based systems with off-board IDE controller cards.
681  *
682  * The IDE driver uses a per-hwgroup lock to protect the hwgroup->busy flag.
683  *
684  * The first thread into the driver for a particular hwgroup sets the
685  * hwgroup->busy flag to indicate that this hwgroup is now active,
686  * and then initiates processing of the top request from the request queue.
687  *
688  * Other threads attempting entry notice the busy setting, and will simply
689  * queue their new requests and exit immediately.  Note that hwgroup->busy
690  * remains set even when the driver is merely awaiting the next interrupt.
691  * Thus, the meaning is "this hwgroup is busy processing a request".
692  *
693  * When processing of a request completes, the completing thread or IRQ-handler
694  * will start the next request from the queue.  If no more work remains,
695  * the driver will clear the hwgroup->busy flag and exit.
696  *
697  * The per-hwgroup spinlock is used to protect all access to the
698  * hwgroup->busy flag, but is otherwise not needed for most processing in
699  * the driver.  This makes the driver much more friendlier to shared IRQs
700  * than previous designs, while remaining 100% (?) SMP safe and capable.
701  */
702 void do_ide_request(struct request_queue *q)
703 {
704         ide_drive_t     *drive = q->queuedata;
705         ide_hwif_t      *hwif = drive->hwif;
706         ide_hwgroup_t   *hwgroup = hwif->hwgroup;
707         struct request  *rq;
708         ide_startstop_t startstop;
709
710         /*
711          * drive is doing pre-flush, ordered write, post-flush sequence. even
712          * though that is 3 requests, it must be seen as a single transaction.
713          * we must not preempt this drive until that is complete
714          */
715         if (blk_queue_flushing(q))
716                 /*
717                  * small race where queue could get replugged during
718                  * the 3-request flush cycle, just yank the plug since
719                  * we want it to finish asap
720                  */
721                 blk_remove_plug(q);
722
723         spin_unlock_irq(q->queue_lock);
724         spin_lock_irq(&hwgroup->lock);
725
726         if (!ide_lock_hwgroup(hwgroup)) {
727 repeat:
728                 hwgroup->rq = NULL;
729
730                 if (drive->dev_flags & IDE_DFLAG_SLEEPING) {
731                         if (time_before(drive->sleep, jiffies)) {
732                                 ide_unlock_hwgroup(hwgroup);
733                                 goto plug_device;
734                         }
735                 }
736
737                 if (hwif != hwgroup->hwif) {
738                         /*
739                          * set nIEN for previous hwif, drives in the
740                          * quirk_list may not like intr setups/cleanups
741                          */
742                         if (drive->quirk_list == 0)
743                                 hwif->tp_ops->set_irq(hwif, 0);
744                 }
745                 hwgroup->hwif = hwif;
746                 hwgroup->drive = drive;
747                 drive->dev_flags &= ~(IDE_DFLAG_SLEEPING | IDE_DFLAG_PARKED);
748
749                 spin_unlock_irq(&hwgroup->lock);
750                 spin_lock_irq(q->queue_lock);
751                 /*
752                  * we know that the queue isn't empty, but this can happen
753                  * if the q->prep_rq_fn() decides to kill a request
754                  */
755                 rq = elv_next_request(drive->queue);
756                 spin_unlock_irq(q->queue_lock);
757                 spin_lock_irq(&hwgroup->lock);
758
759                 if (!rq) {
760                         ide_unlock_hwgroup(hwgroup);
761                         goto out;
762                 }
763
764                 /*
765                  * Sanity: don't accept a request that isn't a PM request
766                  * if we are currently power managed. This is very important as
767                  * blk_stop_queue() doesn't prevent the elv_next_request()
768                  * above to return us whatever is in the queue. Since we call
769                  * ide_do_request() ourselves, we end up taking requests while
770                  * the queue is blocked...
771                  * 
772                  * We let requests forced at head of queue with ide-preempt
773                  * though. I hope that doesn't happen too much, hopefully not
774                  * unless the subdriver triggers such a thing in its own PM
775                  * state machine.
776                  */
777                 if ((drive->dev_flags & IDE_DFLAG_BLOCKED) &&
778                     blk_pm_request(rq) == 0 &&
779                     (rq->cmd_flags & REQ_PREEMPT) == 0) {
780                         /* there should be no pending command at this point */
781                         ide_unlock_hwgroup(hwgroup);
782                         goto plug_device;
783                 }
784
785                 hwgroup->rq = rq;
786
787                 spin_unlock_irq(&hwgroup->lock);
788                 startstop = start_request(drive, rq);
789                 spin_lock_irq(&hwgroup->lock);
790
791                 if (startstop == ide_stopped)
792                         goto repeat;
793         } else
794                 goto plug_device;
795 out:
796         spin_unlock_irq(&hwgroup->lock);
797         spin_lock_irq(q->queue_lock);
798         return;
799
800 plug_device:
801         spin_unlock_irq(&hwgroup->lock);
802         spin_lock_irq(q->queue_lock);
803
804         if (!elv_queue_empty(q))
805                 blk_plug_device(q);
806 }
807
808 /*
809  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
810  * retry the current request in pio mode instead of risking tossing it
811  * all away
812  */
813 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
814 {
815         ide_hwif_t *hwif = HWIF(drive);
816         struct request *rq;
817         ide_startstop_t ret = ide_stopped;
818
819         /*
820          * end current dma transaction
821          */
822
823         if (error < 0) {
824                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
825                 (void)hwif->dma_ops->dma_end(drive);
826                 ret = ide_error(drive, "dma timeout error",
827                                 hwif->tp_ops->read_status(hwif));
828         } else {
829                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
830                 hwif->dma_ops->dma_timeout(drive);
831         }
832
833         /*
834          * disable dma for now, but remember that we did so because of
835          * a timeout -- we'll reenable after we finish this next request
836          * (or rather the first chunk of it) in pio.
837          */
838         drive->dev_flags |= IDE_DFLAG_DMA_PIO_RETRY;
839         drive->retry_pio++;
840         ide_dma_off_quietly(drive);
841
842         /*
843          * un-busy drive etc (hwgroup->busy is cleared on return) and
844          * make sure request is sane
845          */
846         rq = HWGROUP(drive)->rq;
847
848         if (!rq)
849                 goto out;
850
851         HWGROUP(drive)->rq = NULL;
852
853         rq->errors = 0;
854
855         if (!rq->bio)
856                 goto out;
857
858         rq->sector = rq->bio->bi_sector;
859         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
860         rq->hard_cur_sectors = rq->current_nr_sectors;
861         rq->buffer = bio_data(rq->bio);
862 out:
863         return ret;
864 }
865
866 static void ide_plug_device(ide_drive_t *drive)
867 {
868         struct request_queue *q = drive->queue;
869         unsigned long flags;
870
871         spin_lock_irqsave(q->queue_lock, flags);
872         if (!elv_queue_empty(q))
873                 blk_plug_device(q);
874         spin_unlock_irqrestore(q->queue_lock, flags);
875 }
876
877 /**
878  *      ide_timer_expiry        -       handle lack of an IDE interrupt
879  *      @data: timer callback magic (hwgroup)
880  *
881  *      An IDE command has timed out before the expected drive return
882  *      occurred. At this point we attempt to clean up the current
883  *      mess. If the current handler includes an expiry handler then
884  *      we invoke the expiry handler, and providing it is happy the
885  *      work is done. If that fails we apply generic recovery rules
886  *      invoking the handler and checking the drive DMA status. We
887  *      have an excessively incestuous relationship with the DMA
888  *      logic that wants cleaning up.
889  */
890  
891 void ide_timer_expiry (unsigned long data)
892 {
893         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
894         ide_drive_t     *uninitialized_var(drive);
895         ide_handler_t   *handler;
896         ide_expiry_t    *expiry;
897         unsigned long   flags;
898         unsigned long   wait = -1;
899         int             plug_device = 0;
900
901         spin_lock_irqsave(&hwgroup->lock, flags);
902
903         if (((handler = hwgroup->handler) == NULL) ||
904             (hwgroup->req_gen != hwgroup->req_gen_timer)) {
905                 /*
906                  * Either a marginal timeout occurred
907                  * (got the interrupt just as timer expired),
908                  * or we were "sleeping" to give other devices a chance.
909                  * Either way, we don't really want to complain about anything.
910                  */
911         } else {
912                 drive = hwgroup->drive;
913                 if (!drive) {
914                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
915                         hwgroup->handler = NULL;
916                 } else {
917                         ide_hwif_t *hwif;
918                         ide_startstop_t startstop = ide_stopped;
919
920                         if ((expiry = hwgroup->expiry) != NULL) {
921                                 /* continue */
922                                 if ((wait = expiry(drive)) > 0) {
923                                         /* reset timer */
924                                         hwgroup->timer.expires  = jiffies + wait;
925                                         hwgroup->req_gen_timer = hwgroup->req_gen;
926                                         add_timer(&hwgroup->timer);
927                                         spin_unlock_irqrestore(&hwgroup->lock, flags);
928                                         return;
929                                 }
930                         }
931                         hwgroup->handler = NULL;
932                         /*
933                          * We need to simulate a real interrupt when invoking
934                          * the handler() function, which means we need to
935                          * globally mask the specific IRQ:
936                          */
937                         spin_unlock(&hwgroup->lock);
938                         hwif  = HWIF(drive);
939                         /* disable_irq_nosync ?? */
940                         disable_irq(hwif->irq);
941                         /* local CPU only,
942                          * as if we were handling an interrupt */
943                         local_irq_disable();
944                         if (hwgroup->polling) {
945                                 startstop = handler(drive);
946                         } else if (drive_is_ready(drive)) {
947                                 if (drive->waiting_for_dma)
948                                         hwif->dma_ops->dma_lost_irq(drive);
949                                 (void)ide_ack_intr(hwif);
950                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
951                                 startstop = handler(drive);
952                         } else {
953                                 if (drive->waiting_for_dma) {
954                                         startstop = ide_dma_timeout_retry(drive, wait);
955                                 } else
956                                         startstop =
957                                         ide_error(drive, "irq timeout",
958                                                   hwif->tp_ops->read_status(hwif));
959                         }
960                         spin_lock_irq(&hwgroup->lock);
961                         enable_irq(hwif->irq);
962                         if (startstop == ide_stopped) {
963                                 ide_unlock_hwgroup(hwgroup);
964                                 plug_device = 1;
965                         }
966                 }
967         }
968         spin_unlock_irqrestore(&hwgroup->lock, flags);
969
970         if (plug_device)
971                 ide_plug_device(drive);
972 }
973
974 /**
975  *      unexpected_intr         -       handle an unexpected IDE interrupt
976  *      @irq: interrupt line
977  *      @hwgroup: hwgroup being processed
978  *
979  *      There's nothing really useful we can do with an unexpected interrupt,
980  *      other than reading the status register (to clear it), and logging it.
981  *      There should be no way that an irq can happen before we're ready for it,
982  *      so we needn't worry much about losing an "important" interrupt here.
983  *
984  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
985  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
986  *      looks "good", we just ignore the interrupt completely.
987  *
988  *      This routine assumes __cli() is in effect when called.
989  *
990  *      If an unexpected interrupt happens on irq15 while we are handling irq14
991  *      and if the two interfaces are "serialized" (CMD640), then it looks like
992  *      we could screw up by interfering with a new request being set up for 
993  *      irq15.
994  *
995  *      In reality, this is a non-issue.  The new command is not sent unless 
996  *      the drive is ready to accept one, in which case we know the drive is
997  *      not trying to interrupt us.  And ide_set_handler() is always invoked
998  *      before completing the issuance of any new drive command, so we will not
999  *      be accidentally invoked as a result of any valid command completion
1000  *      interrupt.
1001  *
1002  *      Note that we must walk the entire hwgroup here. We know which hwif
1003  *      is doing the current command, but we don't know which hwif burped
1004  *      mysteriously.
1005  */
1006  
1007 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1008 {
1009         u8 stat;
1010         ide_hwif_t *hwif = hwgroup->hwif;
1011
1012         /*
1013          * handle the unexpected interrupt
1014          */
1015         do {
1016                 if (hwif->irq == irq) {
1017                         stat = hwif->tp_ops->read_status(hwif);
1018
1019                         if (!OK_STAT(stat, ATA_DRDY, BAD_STAT)) {
1020                                 /* Try to not flood the console with msgs */
1021                                 static unsigned long last_msgtime, count;
1022                                 ++count;
1023                                 if (time_after(jiffies, last_msgtime + HZ)) {
1024                                         last_msgtime = jiffies;
1025                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1026                                                 "status=0x%02x, count=%ld\n",
1027                                                 hwif->name,
1028                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1029                                 }
1030                         }
1031                 }
1032         } while ((hwif = hwif->next) != hwgroup->hwif);
1033 }
1034
1035 /**
1036  *      ide_intr        -       default IDE interrupt handler
1037  *      @irq: interrupt number
1038  *      @dev_id: hwif group
1039  *      @regs: unused weirdness from the kernel irq layer
1040  *
1041  *      This is the default IRQ handler for the IDE layer. You should
1042  *      not need to override it. If you do be aware it is subtle in
1043  *      places
1044  *
1045  *      hwgroup->hwif is the interface in the group currently performing
1046  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1047  *      the IRQ handler to call. As we issue a command the handlers
1048  *      step through multiple states, reassigning the handler to the
1049  *      next step in the process. Unlike a smart SCSI controller IDE
1050  *      expects the main processor to sequence the various transfer
1051  *      stages. We also manage a poll timer to catch up with most
1052  *      timeout situations. There are still a few where the handlers
1053  *      don't ever decide to give up.
1054  *
1055  *      The handler eventually returns ide_stopped to indicate the
1056  *      request completed. At this point we issue the next request
1057  *      on the hwgroup and the process begins again.
1058  */
1059  
1060 irqreturn_t ide_intr (int irq, void *dev_id)
1061 {
1062         unsigned long flags;
1063         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1064         ide_hwif_t *hwif = hwgroup->hwif;
1065         ide_drive_t *uninitialized_var(drive);
1066         ide_handler_t *handler;
1067         ide_startstop_t startstop;
1068         irqreturn_t irq_ret = IRQ_NONE;
1069         int plug_device = 0;
1070
1071         spin_lock_irqsave(&hwgroup->lock, flags);
1072
1073         if (!ide_ack_intr(hwif))
1074                 goto out;
1075
1076         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1077                 /*
1078                  * Not expecting an interrupt from this drive.
1079                  * That means this could be:
1080                  *      (1) an interrupt from another PCI device
1081                  *      sharing the same PCI INT# as us.
1082                  * or   (2) a drive just entered sleep or standby mode,
1083                  *      and is interrupting to let us know.
1084                  * or   (3) a spurious interrupt of unknown origin.
1085                  *
1086                  * For PCI, we cannot tell the difference,
1087                  * so in that case we just ignore it and hope it goes away.
1088                  *
1089                  * FIXME: unexpected_intr should be hwif-> then we can
1090                  * remove all the ifdef PCI crap
1091                  */
1092 #ifdef CONFIG_BLK_DEV_IDEPCI
1093                 if (hwif->chipset != ide_pci)
1094 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1095                 {
1096                         /*
1097                          * Probably not a shared PCI interrupt,
1098                          * so we can safely try to do something about it:
1099                          */
1100                         unexpected_intr(irq, hwgroup);
1101 #ifdef CONFIG_BLK_DEV_IDEPCI
1102                 } else {
1103                         /*
1104                          * Whack the status register, just in case
1105                          * we have a leftover pending IRQ.
1106                          */
1107                         (void)hwif->tp_ops->read_status(hwif);
1108 #endif /* CONFIG_BLK_DEV_IDEPCI */
1109                 }
1110                 goto out;
1111         }
1112
1113         drive = hwgroup->drive;
1114         if (!drive) {
1115                 /*
1116                  * This should NEVER happen, and there isn't much
1117                  * we could do about it here.
1118                  *
1119                  * [Note - this can occur if the drive is hot unplugged]
1120                  */
1121                 goto out_handled;
1122         }
1123
1124         if (!drive_is_ready(drive))
1125                 /*
1126                  * This happens regularly when we share a PCI IRQ with
1127                  * another device.  Unfortunately, it can also happen
1128                  * with some buggy drives that trigger the IRQ before
1129                  * their status register is up to date.  Hopefully we have
1130                  * enough advance overhead that the latter isn't a problem.
1131                  */
1132                 goto out;
1133
1134         hwgroup->handler = NULL;
1135         hwgroup->req_gen++;
1136         del_timer(&hwgroup->timer);
1137         spin_unlock(&hwgroup->lock);
1138
1139         if (hwif->port_ops && hwif->port_ops->clear_irq)
1140                 hwif->port_ops->clear_irq(drive);
1141
1142         if (drive->dev_flags & IDE_DFLAG_UNMASK)
1143                 local_irq_enable_in_hardirq();
1144
1145         /* service this interrupt, may set handler for next interrupt */
1146         startstop = handler(drive);
1147
1148         spin_lock_irq(&hwgroup->lock);
1149         /*
1150          * Note that handler() may have set things up for another
1151          * interrupt to occur soon, but it cannot happen until
1152          * we exit from this routine, because it will be the
1153          * same irq as is currently being serviced here, and Linux
1154          * won't allow another of the same (on any CPU) until we return.
1155          */
1156         if (startstop == ide_stopped) {
1157                 if (hwgroup->handler == NULL) { /* paranoia */
1158                         ide_unlock_hwgroup(hwgroup);
1159                         plug_device = 1;
1160                 } else
1161                         printk(KERN_ERR "%s: %s: huh? expected NULL handler "
1162                                         "on exit\n", __func__, drive->name);
1163         }
1164 out_handled:
1165         irq_ret = IRQ_HANDLED;
1166 out:
1167         spin_unlock_irqrestore(&hwgroup->lock, flags);
1168
1169         if (plug_device)
1170                 ide_plug_device(drive);
1171
1172         return irq_ret;
1173 }
1174
1175 /**
1176  *      ide_do_drive_cmd        -       issue IDE special command
1177  *      @drive: device to issue command
1178  *      @rq: request to issue
1179  *
1180  *      This function issues a special IDE device request
1181  *      onto the request queue.
1182  *
1183  *      the rq is queued at the head of the request queue, displacing
1184  *      the currently-being-processed request and this function
1185  *      returns immediately without waiting for the new rq to be
1186  *      completed.  This is VERY DANGEROUS, and is intended for
1187  *      careful use by the ATAPI tape/cdrom driver code.
1188  */
1189
1190 void ide_do_drive_cmd(ide_drive_t *drive, struct request *rq)
1191 {
1192         ide_hwgroup_t *hwgroup = drive->hwif->hwgroup;
1193         struct request_queue *q = drive->queue;
1194         unsigned long flags;
1195
1196         hwgroup->rq = NULL;
1197
1198         spin_lock_irqsave(q->queue_lock, flags);
1199         __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0);
1200         blk_start_queueing(q);
1201         spin_unlock_irqrestore(q->queue_lock, flags);
1202 }
1203 EXPORT_SYMBOL(ide_do_drive_cmd);
1204
1205 void ide_pktcmd_tf_load(ide_drive_t *drive, u32 tf_flags, u16 bcount, u8 dma)
1206 {
1207         ide_hwif_t *hwif = drive->hwif;
1208         ide_task_t task;
1209
1210         memset(&task, 0, sizeof(task));
1211         task.tf_flags = IDE_TFLAG_OUT_LBAH | IDE_TFLAG_OUT_LBAM |
1212                         IDE_TFLAG_OUT_FEATURE | tf_flags;
1213         task.tf.feature = dma;          /* Use PIO/DMA */
1214         task.tf.lbam    = bcount & 0xff;
1215         task.tf.lbah    = (bcount >> 8) & 0xff;
1216
1217         ide_tf_dump(drive->name, &task.tf);
1218         hwif->tp_ops->set_irq(hwif, 1);
1219         SELECT_MASK(drive, 0);
1220         hwif->tp_ops->tf_load(drive, &task);
1221 }
1222
1223 EXPORT_SYMBOL_GPL(ide_pktcmd_tf_load);
1224
1225 void ide_pad_transfer(ide_drive_t *drive, int write, int len)
1226 {
1227         ide_hwif_t *hwif = drive->hwif;
1228         u8 buf[4] = { 0 };
1229
1230         while (len > 0) {
1231                 if (write)
1232                         hwif->tp_ops->output_data(drive, NULL, buf, min(4, len));
1233                 else
1234                         hwif->tp_ops->input_data(drive, NULL, buf, min(4, len));
1235                 len -= 4;
1236         }
1237 }
1238 EXPORT_SYMBOL_GPL(ide_pad_transfer);