drivers/ide/ide-io.c

   1 /*
   2  *      IDE I/O functions
   3  *
   4  *      Basic PIO and command management functionality.
   5  *
   6  * This code was split off from ide.c. See ide.c for history and original
   7  * copyrights.
   8  *
   9  * This program is free software; you can redistribute it and/or modify it
  10  * under the terms of the GNU General Public License as published by the
  11  * Free Software Foundation; either version 2, or (at your option) any
  12  * later version.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  *
  19  * For the avoidance of doubt the "preferred form" of this code is one which
  20  * is in an open non patent encumbered format. Where cryptographic key signing
  21  * forms part of the process of creating an executable the information
  22  * including keys needed to generate an equivalently functional executable
  23  * are deemed to be part of the source code.
  24  */
  25
  26
  27 #include <linux/config.h>
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/string.h>
  31 #include <linux/kernel.h>
  32 #include <linux/timer.h>
  33 #include <linux/mm.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/major.h>
  36 #include <linux/errno.h>
  37 #include <linux/genhd.h>
  38 #include <linux/blkpg.h>
  39 #include <linux/slab.h>
  40 #include <linux/init.h>
  41 #include <linux/pci.h>
  42 #include <linux/delay.h>
  43 #include <linux/ide.h>
  44 #include <linux/completion.h>
  45 #include <linux/reboot.h>
  46 #include <linux/cdrom.h>
  47 #include <linux/seq_file.h>
  48 #include <linux/device.h>
  49 #include <linux/kmod.h>
  50
  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 void ide_fill_flush_cmd(ide_drive_t *drive, struct request *rq)
  58 {
  59         char *buf = rq->cmd;
  60
  61         /*
  62          * reuse cdb space for ata command
  63          */
  64         memset(buf, 0, sizeof(rq->cmd));
  65
  66         rq->flags |= REQ_DRIVE_TASK | REQ_STARTED;
  67         rq->buffer = buf;
  68         rq->buffer[0] = WIN_FLUSH_CACHE;
  69
  70         if (ide_id_has_flush_cache_ext(drive->id) &&
  71             (drive->capacity64 >= (1UL << 28)))
  72                 rq->buffer[0] = WIN_FLUSH_CACHE_EXT;
  73 }
  74
  75 /*
  76  * preempt pending requests, and store this cache flush for immediate
  77  * execution
  78  */
  79 static struct request *ide_queue_flush_cmd(ide_drive_t *drive,
  80                                            struct request *rq, int post)
  81 {
  82         struct request *flush_rq = &HWGROUP(drive)->wrq;
  83
  84         /*
  85          * write cache disabled, clear the barrier bit and treat it like
  86          * an ordinary write
  87          */
  88         if (!drive->wcache) {
  89                 rq->flags |= REQ_BAR_PREFLUSH;
  90                 return rq;
  91         }
  92
  93         ide_init_drive_cmd(flush_rq);
  94         ide_fill_flush_cmd(drive, flush_rq);
  95
  96         flush_rq->special = rq;
  97         flush_rq->nr_sectors = rq->nr_sectors;
  98
  99         if (!post) {
 100                 drive->doing_barrier = 1;
 101                 flush_rq->flags |= REQ_BAR_PREFLUSH;
 102                 blkdev_dequeue_request(rq);
 103         } else
 104                 flush_rq->flags |= REQ_BAR_POSTFLUSH;
 105
 106         __elv_add_request(drive->queue, flush_rq, ELEVATOR_INSERT_FRONT, 0);
 107         HWGROUP(drive)->rq = NULL;
 108         return flush_rq;
 109 }
 110
 111 static int __ide_end_request(ide_drive_t *drive, struct request *rq,
 112                              int uptodate, int nr_sectors)
 113 {
 114         int ret = 1;
 115
 116         BUG_ON(!(rq->flags & REQ_STARTED));
 117
 118         /*
 119          * if failfast is set on a request, override number of sectors and
 120          * complete the whole request right now
 121          */
 122         if (blk_noretry_request(rq) && end_io_error(uptodate))
 123                 nr_sectors = rq->hard_nr_sectors;
 124
 125         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
 126                 rq->errors = -EIO;
 127
 128         /*
 129          * decide whether to reenable DMA -- 3 is a random magic for now,
 130          * if we DMA timeout more than 3 times, just stay in PIO
 131          */
 132         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
 133                 drive->state = 0;
 134                 HWGROUP(drive)->hwif->ide_dma_on(drive);
 135         }
 136
 137         if (!end_that_request_first(rq, uptodate, nr_sectors)) {
 138                 add_disk_randomness(rq->rq_disk);
 139
 140                 if (blk_rq_tagged(rq))
 141                         blk_queue_end_tag(drive->queue, rq);
 142
 143                 blkdev_dequeue_request(rq);
 144                 HWGROUP(drive)->rq = NULL;
 145                 end_that_request_last(rq);
 146                 ret = 0;
 147         }
 148         return ret;
 149 }
 150
 151 /**
 152  *      ide_end_request         -       complete an IDE I/O
 153  *      @drive: IDE device for the I/O
 154  *      @uptodate:
 155  *      @nr_sectors: number of sectors completed
 156  *
 157  *      This is our end_request wrapper function. We complete the I/O
 158  *      update random number input and dequeue the request, which if
 159  *      it was tagged may be out of order.
 160  */
 161
 162 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
 163 {
 164         struct request *rq;
 165         unsigned long flags;
 166         int ret = 1;
 167
 168         spin_lock_irqsave(&ide_lock, flags);
 169         rq = HWGROUP(drive)->rq;
 170
 171         if (!nr_sectors)
 172                 nr_sectors = rq->hard_cur_sectors;
 173
 174         if (!blk_barrier_rq(rq) || !drive->wcache)
 175                 ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
 176         else {
 177                 struct request *flush_rq = &HWGROUP(drive)->wrq;
 178
 179                 flush_rq->nr_sectors -= nr_sectors;
 180                 if (!flush_rq->nr_sectors) {
 181                         ide_queue_flush_cmd(drive, rq, 1);
 182                         ret = 0;
 183                 }
 184         }
 185
 186         spin_unlock_irqrestore(&ide_lock, flags);
 187         return ret;
 188 }
 189 EXPORT_SYMBOL(ide_end_request);
 190
 191 /**
 192  *      ide_complete_pm_request - end the current Power Management request
 193  *      @drive: target drive
 194  *      @rq: request
 195  *
 196  *      This function cleans up the current PM request and stops the queue
 197  *      if necessary.
 198  */
 199 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
 200 {
 201         unsigned long flags;
 202
 203 #ifdef DEBUG_PM
 204         printk("%s: completing PM request, %s\n", drive->name,
 205                blk_pm_suspend_request(rq) ? "suspend" : "resume");
 206 #endif
 207         spin_lock_irqsave(&ide_lock, flags);
 208         if (blk_pm_suspend_request(rq)) {
 209                 blk_stop_queue(drive->queue);
 210         } else {
 211                 drive->blocked = 0;
 212                 blk_start_queue(drive->queue);
 213         }
 214         blkdev_dequeue_request(rq);
 215         HWGROUP(drive)->rq = NULL;
 216         end_that_request_last(rq);
 217         spin_unlock_irqrestore(&ide_lock, flags);
 218 }
 219
 220 /*
 221  * FIXME: probably move this somewhere else, name is bad too :)
 222  */
 223 u64 ide_get_error_location(ide_drive_t *drive, char *args)
 224 {
 225         u32 high, low;
 226         u8 hcyl, lcyl, sect;
 227         u64 sector;
 228
 229         high = 0;
 230         hcyl = args[5];
 231         lcyl = args[4];
 232         sect = args[3];
 233
 234         if (ide_id_has_flush_cache_ext(drive->id)) {
 235                 low = (hcyl << 16) | (lcyl << 8) | sect;
 236                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 237                 high = ide_read_24(drive);
 238         } else {
 239                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
 240                 if (cur & 0x40)
 241                         low = (hcyl << 16) | (lcyl << 8) | sect;
 242                 else {
 243                         low = hcyl * drive->head * drive->sect;
 244                         low += lcyl * drive->sect;
 245                         low += sect - 1;
 246                 }
 247         }
 248
 249         sector = ((u64) high << 24) | low;
 250         return sector;
 251 }
 252 EXPORT_SYMBOL(ide_get_error_location);
 253
 254 static void ide_complete_barrier(ide_drive_t *drive, struct request *rq,
 255                                  int error)
 256 {
 257         struct request *real_rq = rq->special;
 258         int good_sectors, bad_sectors;
 259         sector_t sector;
 260
 261         if (!error) {
 262                 if (blk_barrier_postflush(rq)) {
 263                         /*
 264                          * this completes the barrier write
 265                          */
 266                         __ide_end_request(drive, real_rq, 1, real_rq->hard_nr_sectors);
 267                         drive->doing_barrier = 0;
 268                 } else {
 269                         /*
 270                          * just indicate that we did the pre flush
 271                          */
 272                         real_rq->flags |= REQ_BAR_PREFLUSH;
 273                         elv_requeue_request(drive->queue, real_rq);
 274                 }
 275                 /*
 276                  * all is fine, return
 277                  */
 278                 return;
 279         }
 280
 281         /*
 282          * we need to end real_rq, but it's not on the queue currently.
 283          * put it back on the queue, so we don't have to special case
 284          * anything else for completing it
 285          */
 286         if (!blk_barrier_postflush(rq))
 287                 elv_requeue_request(drive->queue, real_rq);
 288
 289         /*
 290          * drive aborted flush command, assume FLUSH_CACHE_* doesn't
 291          * work and disable barrier support
 292          */
 293         if (error & ABRT_ERR) {
 294                 printk(KERN_ERR "%s: barrier support doesn't work\n", drive->name);
 295                 __ide_end_request(drive, real_rq, -EOPNOTSUPP, real_rq->hard_nr_sectors);
 296                 blk_queue_ordered(drive->queue, 0);
 297                 blk_queue_issue_flush_fn(drive->queue, NULL);
 298         } else {
 299                 /*
 300                  * find out what part of the request failed
 301                  */
 302                 good_sectors = 0;
 303                 if (blk_barrier_postflush(rq)) {
 304                         sector = ide_get_error_location(drive, rq->buffer);
 305
 306                         if ((sector >= real_rq->hard_sector) &&
 307                             (sector < real_rq->hard_sector + real_rq->hard_nr_sectors))
 308                                 good_sectors = sector - real_rq->hard_sector;
 309                 } else
 310                         sector = real_rq->hard_sector;
 311
 312                 bad_sectors = real_rq->hard_nr_sectors - good_sectors;
 313                 if (good_sectors)
 314                         __ide_end_request(drive, real_rq, 1, good_sectors);
 315                 if (bad_sectors)
 316                         __ide_end_request(drive, real_rq, 0, bad_sectors);
 317
 318                 printk(KERN_ERR "%s: failed barrier write: "
 319                                 "sector=%Lx(good=%d/bad=%d)\n",
 320                                 drive->name, (unsigned long long)sector,
 321                                 good_sectors, bad_sectors);
 322         }
 323
 324         drive->doing_barrier = 0;
 325 }
 326
 327 /**
 328  *      ide_end_drive_cmd       -       end an explicit drive command
 329  *      @drive: command
 330  *      @stat: status bits
 331  *      @err: error bits
 332  *
 333  *      Clean up after success/failure of an explicit drive command.
 334  *      These get thrown onto the queue so they are synchronized with
 335  *      real I/O operations on the drive.
 336  *
 337  *      In LBA48 mode we have to read the register set twice to get
 338  *      all the extra information out.
 339  */
 340
 341 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
 342 {
 343         ide_hwif_t *hwif = HWIF(drive);
 344         unsigned long flags;
 345         struct request *rq;
 346
 347         spin_lock_irqsave(&ide_lock, flags);
 348         rq = HWGROUP(drive)->rq;
 349         spin_unlock_irqrestore(&ide_lock, flags);
 350
 351         if (rq->flags & REQ_DRIVE_CMD) {
 352                 u8 *args = (u8 *) rq->buffer;
 353                 if (rq->errors == 0)
 354                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 355
 356                 if (args) {
 357                         args[0] = stat;
 358                         args[1] = err;
 359                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 360                 }
 361         } else if (rq->flags & REQ_DRIVE_TASK) {
 362                 u8 *args = (u8 *) rq->buffer;
 363                 if (rq->errors == 0)
 364                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 365
 366                 if (args) {
 367                         args[0] = stat;
 368                         args[1] = err;
 369                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 370                         args[3] = hwif->INB(IDE_SECTOR_REG);
 371                         args[4] = hwif->INB(IDE_LCYL_REG);
 372                         args[5] = hwif->INB(IDE_HCYL_REG);
 373                         args[6] = hwif->INB(IDE_SELECT_REG);
 374                 }
 375         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
 376                 ide_task_t *args = (ide_task_t *) rq->special;
 377                 if (rq->errors == 0)
 378                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 379
 380                 if (args) {
 381                         if (args->tf_in_flags.b.data) {
 382                                 u16 data                                = hwif->INW(IDE_DATA_REG);
 383                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
 384                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
 385                         }
 386                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
 387                         /* be sure we're looking at the low order bits */
 388                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
 389                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
 390                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
 391                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
 392                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
 393                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
 394                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
 395
 396                         if (drive->addressing == 1) {
 397                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 398                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
 399                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
 400                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
 401                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
 402                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
 403                         }
 404                 }
 405         } else if (blk_pm_request(rq)) {
 406 #ifdef DEBUG_PM
 407                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
 408                         drive->name, rq->pm->pm_step, stat, err);
 409 #endif
 410                 DRIVER(drive)->complete_power_step(drive, rq, stat, err);
 411                 if (rq->pm->pm_step == ide_pm_state_completed)
 412                         ide_complete_pm_request(drive, rq);
 413                 return;
 414         }
 415
 416         spin_lock_irqsave(&ide_lock, flags);
 417         blkdev_dequeue_request(rq);
 418
 419         if (blk_barrier_preflush(rq) || blk_barrier_postflush(rq))
 420                 ide_complete_barrier(drive, rq, err);
 421
 422         HWGROUP(drive)->rq = NULL;
 423         end_that_request_last(rq);
 424         spin_unlock_irqrestore(&ide_lock, flags);
 425 }
 426
 427 EXPORT_SYMBOL(ide_end_drive_cmd);
 428
 429 /**
 430  *      try_to_flush_leftover_data      -       flush junk
 431  *      @drive: drive to flush
 432  *
 433  *      try_to_flush_leftover_data() is invoked in response to a drive
 434  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
 435  *      resetting the drive, this routine tries to clear the condition
 436  *      by read a sector's worth of data from the drive.  Of course,
 437  *      this may not help if the drive is *waiting* for data from *us*.
 438  */
 439 void try_to_flush_leftover_data (ide_drive_t *drive)
 440 {
 441         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
 442
 443         if (drive->media != ide_disk)
 444                 return;
 445         while (i > 0) {
 446                 u32 buffer[16];
 447                 u32 wcount = (i > 16) ? 16 : i;
 448
 449                 i -= wcount;
 450                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
 451         }
 452 }
 453
 454 EXPORT_SYMBOL(try_to_flush_leftover_data);
 455
 456 /*
 457  * FIXME Add an ATAPI error
 458  */
 459
 460 /**
 461  *      ide_error       -       handle an error on the IDE
 462  *      @drive: drive the error occurred on
 463  *      @msg: message to report
 464  *      @stat: status bits
 465  *
 466  *      ide_error() takes action based on the error returned by the drive.
 467  *      For normal I/O that may well include retries. We deal with
 468  *      both new-style (taskfile) and old style command handling here.
 469  *      In the case of taskfile command handling there is work left to
 470  *      do
 471  */
 472
 473 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
 474 {
 475         ide_hwif_t *hwif;
 476         struct request *rq;
 477         u8 err;
 478
 479         err = ide_dump_status(drive, msg, stat);
 480         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
 481                 return ide_stopped;
 482
 483         hwif = HWIF(drive);
 484         /* retry only "normal" I/O: */
 485         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
 486                 rq->errors = 1;
 487                 ide_end_drive_cmd(drive, stat, err);
 488                 return ide_stopped;
 489         }
 490         if (rq->flags & REQ_DRIVE_TASKFILE) {
 491                 rq->errors = 1;
 492                 ide_end_drive_cmd(drive, stat, err);
 493                 return ide_stopped;
 494         }
 495
 496         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 497                  /* other bits are useless when BUSY */
 498                 rq->errors |= ERROR_RESET;
 499         } else {
 500                 if (drive->media != ide_disk)
 501                         goto media_out;
 502
 503                 if (stat & ERR_STAT) {
 504                         /* err has different meaning on cdrom and tape */
 505                         if (err == ABRT_ERR) {
 506                                 if (drive->select.b.lba &&
 507                                     (hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY))
 508                                         /* some newer drives don't
 509                                          * support WIN_SPECIFY
 510                                          */
 511                                         return ide_stopped;
 512                         } else if ((err & BAD_CRC) == BAD_CRC) {
 513                                 drive->crc_count++;
 514                                 /* UDMA crc error -- just retry the operation */
 515                         } else if (err & (BBD_ERR | ECC_ERR)) {
 516                                 /* retries won't help these */
 517                                 rq->errors = ERROR_MAX;
 518                         } else if (err & TRK0_ERR) {
 519                                 /* help it find track zero */
 520                                 rq->errors |= ERROR_RECAL;
 521                         }
 522                 }
 523 media_out:
 524                 if ((stat & DRQ_STAT) && rq_data_dir(rq) != WRITE)
 525                         try_to_flush_leftover_data(drive);
 526         }
 527         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT)) {
 528                 /* force an abort */
 529                 hwif->OUTB(WIN_IDLEIMMEDIATE,IDE_COMMAND_REG);
 530         }
 531         if (rq->errors >= ERROR_MAX) {
 532                 DRIVER(drive)->end_request(drive, 0, 0);
 533         } else {
 534                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 535                         ++rq->errors;
 536                         return ide_do_reset(drive);
 537                 }
 538                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
 539                         drive->special.b.recalibrate = 1;
 540                 ++rq->errors;
 541         }
 542         return ide_stopped;
 543 }
 544
 545 EXPORT_SYMBOL(ide_error);
 546
 547 /**
 548  *      ide_abort       -       abort pending IDE operatins
 549  *      @drive: drive the error occurred on
 550  *      @msg: message to report
 551  *
 552  *      ide_abort kills and cleans up when we are about to do a
 553  *      host initiated reset on active commands. Longer term we
 554  *      want handlers to have sensible abort handling themselves
 555  *
 556  *      This differs fundamentally from ide_error because in
 557  *      this case the command is doing just fine when we
 558  *      blow it away.
 559  */
 560
 561 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
 562 {
 563         ide_hwif_t *hwif;
 564         struct request *rq;
 565
 566         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
 567                 return ide_stopped;
 568
 569         hwif = HWIF(drive);
 570         /* retry only "normal" I/O: */
 571         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
 572                 rq->errors = 1;
 573                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
 574                 return ide_stopped;
 575         }
 576         if (rq->flags & REQ_DRIVE_TASKFILE) {
 577                 rq->errors = 1;
 578                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
 579                 return ide_stopped;
 580         }
 581
 582         rq->errors |= ERROR_RESET;
 583         DRIVER(drive)->end_request(drive, 0, 0);
 584         return ide_stopped;
 585 }
 586
 587 EXPORT_SYMBOL(ide_abort);
 588
 589 /**
 590  *      ide_cmd         -       issue a simple drive command
 591  *      @drive: drive the command is for
 592  *      @cmd: command byte
 593  *      @nsect: sector byte
 594  *      @handler: handler for the command completion
 595  *
 596  *      Issue a simple drive command with interrupts.
 597  *      The drive must be selected beforehand.
 598  */
 599
 600 void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect, ide_handler_t *handler)
 601 {
 602         ide_hwif_t *hwif = HWIF(drive);
 603         if (IDE_CONTROL_REG)
 604                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
 605         SELECT_MASK(drive,0);
 606         hwif->OUTB(nsect,IDE_NSECTOR_REG);
 607         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
 608 }
 609
 610 EXPORT_SYMBOL(ide_cmd);
 611
 612 /**
 613  *      drive_cmd_intr          -       drive command completion interrupt
 614  *      @drive: drive the completion interrupt occurred on
 615  *
 616  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
 617  *      We do any necessary daya reading and then wait for the drive to
 618  *      go non busy. At that point we may read the error data and complete
 619  *      the request
 620  */
 621
 622 ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
 623 {
 624         struct request *rq = HWGROUP(drive)->rq;
 625         ide_hwif_t *hwif = HWIF(drive);
 626         u8 *args = (u8 *) rq->buffer;
 627         u8 stat = hwif->INB(IDE_STATUS_REG);
 628         int retries = 10;
 629
 630         local_irq_enable();
 631         if ((stat & DRQ_STAT) && args && args[3]) {
 632                 u8 io_32bit = drive->io_32bit;
 633                 drive->io_32bit = 0;
 634                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
 635                 drive->io_32bit = io_32bit;
 636                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
 637                         udelay(100);
 638         }
 639
 640         if (!OK_STAT(stat, READY_STAT, BAD_STAT) && DRIVER(drive) != NULL)
 641                 return DRIVER(drive)->error(drive, "drive_cmd", stat);
 642                 /* calls ide_end_drive_cmd */
 643         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
 644         return ide_stopped;
 645 }
 646
 647 EXPORT_SYMBOL(drive_cmd_intr);
 648
 649 /**
 650  *      do_special              -       issue some special commands
 651  *      @drive: drive the command is for
 652  *
 653  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
 654  *      commands to a drive.  It used to do much more, but has been scaled
 655  *      back.
 656  */
 657
 658 ide_startstop_t do_special (ide_drive_t *drive)
 659 {
 660         special_t *s = &drive->special;
 661
 662 #ifdef DEBUG
 663         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
 664 #endif
 665         if (s->b.set_tune) {
 666                 s->b.set_tune = 0;
 667                 if (HWIF(drive)->tuneproc != NULL)
 668                         HWIF(drive)->tuneproc(drive, drive->tune_req);
 669                 return ide_stopped;
 670         }
 671         else
 672                 return DRIVER(drive)->special(drive);
 673 }
 674
 675 EXPORT_SYMBOL(do_special);
 676
 677 /**
 678  *      execute_drive_command   -       issue special drive command
 679  *      @drive: the drive to issue th command on
 680  *      @rq: the request structure holding the command
 681  *
 682  *      execute_drive_cmd() issues a special drive command,  usually
 683  *      initiated by ioctl() from the external hdparm program. The
 684  *      command can be a drive command, drive task or taskfile
 685  *      operation. Weirdly you can call it with NULL to wait for
 686  *      all commands to finish. Don't do this as that is due to change
 687  */
 688
 689 ide_startstop_t execute_drive_cmd (ide_drive_t *drive, struct request *rq)
 690 {
 691         ide_hwif_t *hwif = HWIF(drive);
 692         if (rq->flags & REQ_DRIVE_TASKFILE) {
 693                 ide_task_t *args = rq->special;
 694
 695                 if (!args)
 696                         goto done;
 697
 698                 hwif->data_phase = args->data_phase;
 699
 700                 if (args->tf_out_flags.all != 0)
 701                         return flagged_taskfile(drive, args);
 702                 return do_rw_taskfile(drive, args);
 703         } else if (rq->flags & REQ_DRIVE_TASK) {
 704                 u8 *args = rq->buffer;
 705                 u8 sel;
 706
 707                 if (!args)
 708                         goto done;
 709 #ifdef DEBUG
 710                 printk("%s: DRIVE_TASK_CMD ", drive->name);
 711                 printk("cmd=0x%02x ", args[0]);
 712                 printk("fr=0x%02x ", args[1]);
 713                 printk("ns=0x%02x ", args[2]);
 714                 printk("sc=0x%02x ", args[3]);
 715                 printk("lcyl=0x%02x ", args[4]);
 716                 printk("hcyl=0x%02x ", args[5]);
 717                 printk("sel=0x%02x\n", args[6]);
 718 #endif
 719                 hwif->OUTB(args[1], IDE_FEATURE_REG);
 720                 hwif->OUTB(args[3], IDE_SECTOR_REG);
 721                 hwif->OUTB(args[4], IDE_LCYL_REG);
 722                 hwif->OUTB(args[5], IDE_HCYL_REG);
 723                 sel = (args[6] & ~0x10);
 724                 if (drive->select.b.unit)
 725                         sel |= 0x10;
 726                 hwif->OUTB(sel, IDE_SELECT_REG);
 727                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
 728                 return ide_started;
 729         } else if (rq->flags & REQ_DRIVE_CMD) {
 730                 u8 *args = rq->buffer;
 731
 732                 if (!args)
 733                         goto done;
 734 #ifdef DEBUG
 735                 printk("%s: DRIVE_CMD ", drive->name);
 736                 printk("cmd=0x%02x ", args[0]);
 737                 printk("sc=0x%02x ", args[1]);
 738                 printk("fr=0x%02x ", args[2]);
 739                 printk("xx=0x%02x\n", args[3]);
 740 #endif
 741                 if (args[0] == WIN_SMART) {
 742                         hwif->OUTB(0x4f, IDE_LCYL_REG);
 743                         hwif->OUTB(0xc2, IDE_HCYL_REG);
 744                         hwif->OUTB(args[2],IDE_FEATURE_REG);
 745                         hwif->OUTB(args[1],IDE_SECTOR_REG);
 746                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
 747                         return ide_started;
 748                 }
 749                 hwif->OUTB(args[2],IDE_FEATURE_REG);
 750                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
 751                 return ide_started;
 752         }
 753
 754 done:
 755         /*
 756          * NULL is actually a valid way of waiting for
 757          * all current requests to be flushed from the queue.
 758          */
 759 #ifdef DEBUG
 760         printk("%s: DRIVE_CMD (null)\n", drive->name);
 761 #endif
 762         ide_end_drive_cmd(drive,
 763                         hwif->INB(IDE_STATUS_REG),
 764                         hwif->INB(IDE_ERROR_REG));
 765         return ide_stopped;
 766 }
 767
 768 EXPORT_SYMBOL(execute_drive_cmd);
 769
 770 /**
 771  *      start_request   -       start of I/O and command issuing for IDE
 772  *
 773  *      start_request() initiates handling of a new I/O request. It
 774  *      accepts commands and I/O (read/write) requests. It also does
 775  *      the final remapping for weird stuff like EZDrive. Once
 776  *      device mapper can work sector level the EZDrive stuff can go away
 777  *
 778  *      FIXME: this function needs a rename
 779  */
 780
 781 ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
 782 {
 783         ide_startstop_t startstop;
 784         sector_t block;
 785
 786         BUG_ON(!(rq->flags & REQ_STARTED));
 787
 788 #ifdef DEBUG
 789         printk("%s: start_request: current=0x%08lx\n",
 790                 HWIF(drive)->name, (unsigned long) rq);
 791 #endif
 792
 793         /* bail early if we've exceeded max_failures */
 794         if (drive->max_failures && (drive->failures > drive->max_failures)) {
 795                 goto kill_rq;
 796         }
 797
 798         /*
 799          * bail early if we've sent a device to sleep, however how to wake
 800          * this needs to be a masked flag.  FIXME for proper operations.
 801          */
 802         if (drive->suspend_reset)
 803                 goto kill_rq;
 804
 805         block    = rq->sector;
 806         if (blk_fs_request(rq) &&
 807             (drive->media == ide_disk || drive->media == ide_floppy)) {
 808                 block += drive->sect0;
 809         }
 810         /* Yecch - this will shift the entire interval,
 811            possibly killing some innocent following sector */
 812         if (block == 0 && drive->remap_0_to_1 == 1)
 813                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
 814
 815         if (blk_pm_suspend_request(rq) &&
 816             rq->pm->pm_step == ide_pm_state_start_suspend)
 817                 /* Mark drive blocked when starting the suspend sequence. */
 818                 drive->blocked = 1;
 819         else if (blk_pm_resume_request(rq) &&
 820                  rq->pm->pm_step == ide_pm_state_start_resume) {
 821                 /*
 822                  * The first thing we do on wakeup is to wait for BSY bit to
 823                  * go away (with a looong timeout) as a drive on this hwif may
 824                  * just be POSTing itself.
 825                  * We do that before even selecting as the "other" device on
 826                  * the bus may be broken enough to walk on our toes at this
 827                  * point.
 828                  */
 829                 int rc;
 830 #ifdef DEBUG_PM
 831                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
 832 #endif
 833                 rc = ide_wait_not_busy(HWIF(drive), 35000);
 834                 if (rc)
 835                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
 836                 SELECT_DRIVE(drive);
 837                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
 838                 rc = ide_wait_not_busy(HWIF(drive), 10000);
 839                 if (rc)
 840                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
 841         }
 842
 843         SELECT_DRIVE(drive);
 844         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
 845                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
 846                 return startstop;
 847         }
 848         if (!drive->special.all) {
 849                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
 850                         return execute_drive_cmd(drive, rq);
 851                 else if (rq->flags & REQ_DRIVE_TASKFILE)
 852                         return execute_drive_cmd(drive, rq);
 853                 else if (blk_pm_request(rq)) {
 854 #ifdef DEBUG_PM
 855                         printk("%s: start_power_step(step: %d)\n",
 856                                 drive->name, rq->pm->pm_step);
 857 #endif
 858                         startstop = DRIVER(drive)->start_power_step(drive, rq);
 859                         if (startstop == ide_stopped &&
 860                             rq->pm->pm_step == ide_pm_state_completed)
 861                                 ide_complete_pm_request(drive, rq);
 862                         return startstop;
 863                 }
 864                 return (DRIVER(drive)->do_request(drive, rq, block));
 865         }
 866         return do_special(drive);
 867 kill_rq:
 868         DRIVER(drive)->end_request(drive, 0, 0);
 869         return ide_stopped;
 870 }
 871
 872 EXPORT_SYMBOL(start_request);
 873
 874 /**
 875  *      ide_stall_queue         -       pause an IDE device
 876  *      @drive: drive to stall
 877  *      @timeout: time to stall for (jiffies)
 878  *
 879  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
 880  *      to the hwgroup by sleeping for timeout jiffies.
 881  */
 882
 883 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
 884 {
 885         if (timeout > WAIT_WORSTCASE)
 886                 timeout = WAIT_WORSTCASE;
 887         drive->sleep = timeout + jiffies;
 888 }
 889
 890 EXPORT_SYMBOL(ide_stall_queue);
 891
 892 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
 893
 894 /**
 895  *      choose_drive            -       select a drive to service
 896  *      @hwgroup: hardware group to select on
 897  *
 898  *      choose_drive() selects the next drive which will be serviced.
 899  *      This is necessary because the IDE layer can't issue commands
 900  *      to both drives on the same cable, unlike SCSI.
 901  */
 902
 903 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
 904 {
 905         ide_drive_t *drive, *best;
 906
 907 repeat:
 908         best = NULL;
 909         drive = hwgroup->drive;
 910
 911         /*
 912          * drive is doing pre-flush, ordered write, post-flush sequence. even
 913          * though that is 3 requests, it must be seen as a single transaction.
 914          * we must not preempt this drive until that is complete
 915          */
 916         if (drive->doing_barrier) {
 917                 /*
 918                  * small race where queue could get replugged during
 919                  * the 3-request flush cycle, just yank the plug since
 920                  * we want it to finish asap
 921                  */
 922                 blk_remove_plug(drive->queue);
 923                 return drive;
 924         }
 925
 926         do {
 927                 if ((!drive->sleep || time_after_eq(jiffies, drive->sleep))
 928                     && !elv_queue_empty(drive->queue)) {
 929                         if (!best
 930                          || (drive->sleep && (!best->sleep || 0 < (signed long)(best->sleep - drive->sleep)))
 931                          || (!best->sleep && 0 < (signed long)(WAKEUP(best) - WAKEUP(drive))))
 932                         {
 933                                 if (!blk_queue_plugged(drive->queue))
 934                                         best = drive;
 935                         }
 936                 }
 937         } while ((drive = drive->next) != hwgroup->drive);
 938         if (best && best->nice1 && !best->sleep && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
 939                 long t = (signed long)(WAKEUP(best) - jiffies);
 940                 if (t >= WAIT_MIN_SLEEP) {
 941                 /*
 942                  * We *may* have some time to spare, but first let's see if
 943                  * someone can potentially benefit from our nice mood today..
 944                  */
 945                         drive = best->next;
 946                         do {
 947                                 if (!drive->sleep
 948                                 /* FIXME: use time_before */
 949                                  && 0 < (signed long)(WAKEUP(drive) - (jiffies - best->service_time))
 950                                  && 0 < (signed long)((jiffies + t) - WAKEUP(drive)))
 951                                 {
 952                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
 953                                         goto repeat;
 954                                 }
 955                         } while ((drive = drive->next) != best);
 956                 }
 957         }
 958         return best;
 959 }
 960
 961 /*
 962  * Issue a new request to a drive from hwgroup
 963  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
 964  *
 965  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
 966  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
 967  * may have both interfaces in a single hwgroup to "serialize" access.
 968  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
 969  * together into one hwgroup for serialized access.
 970  *
 971  * Note also that several hwgroups can end up sharing a single IRQ,
 972  * possibly along with many other devices.  This is especially common in
 973  * PCI-based systems with off-board IDE controller cards.
 974  *
 975  * The IDE driver uses the single global ide_lock spinlock to protect
 976  * access to the request queues, and to protect the hwgroup->busy flag.
 977  *
 978  * The first thread into the driver for a particular hwgroup sets the
 979  * hwgroup->busy flag to indicate that this hwgroup is now active,
 980  * and then initiates processing of the top request from the request queue.
 981  *
 982  * Other threads attempting entry notice the busy setting, and will simply
 983  * queue their new requests and exit immediately.  Note that hwgroup->busy
 984  * remains set even when the driver is merely awaiting the next interrupt.
 985  * Thus, the meaning is "this hwgroup is busy processing a request".
 986  *
 987  * When processing of a request completes, the completing thread or IRQ-handler
 988  * will start the next request from the queue.  If no more work remains,
 989  * the driver will clear the hwgroup->busy flag and exit.
 990  *
 991  * The ide_lock (spinlock) is used to protect all access to the
 992  * hwgroup->busy flag, but is otherwise not needed for most processing in
 993  * the driver.  This makes the driver much more friendlier to shared IRQs
 994  * than previous designs, while remaining 100% (?) SMP safe and capable.
 995  */
 996 /* --BenH: made non-static as ide-pmac.c uses it to kick the hwgroup back
 997  *         into life on wakeup from machine sleep.
 998  */
 999 void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1000 {
1001         ide_drive_t     *drive;
1002         ide_hwif_t      *hwif;
1003         struct request  *rq;
1004         ide_startstop_t startstop;
1005
1006         /* for atari only: POSSIBLY BROKEN HERE(?) */
1007         ide_get_lock(ide_intr, hwgroup);
1008
1009         /* caller must own ide_lock */
1010         BUG_ON(!irqs_disabled());
1011
1012         while (!hwgroup->busy) {
1013                 hwgroup->busy = 1;
1014                 drive = choose_drive(hwgroup);
1015                 if (drive == NULL) {
1016                         unsigned long sleep = 0;
1017                         hwgroup->rq = NULL;
1018                         drive = hwgroup->drive;
1019                         do {
1020                                 if (drive->sleep && (!sleep || 0 < (signed long)(sleep - drive->sleep)))
1021                                         sleep = drive->sleep;
1022                         } while ((drive = drive->next) != hwgroup->drive);
1023                         if (sleep) {
1024                 /*
1025                  * Take a short snooze, and then wake up this hwgroup again.
1026                  * This gives other hwgroups on the same a chance to
1027                  * play fairly with us, just in case there are big differences
1028                  * in relative throughputs.. don't want to hog the cpu too much.
1029                  */
1030                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1031                                         sleep = jiffies + WAIT_MIN_SLEEP;
1032 #if 1
1033                                 if (timer_pending(&hwgroup->timer))
1034                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1035 #endif
1036                                 /* so that ide_timer_expiry knows what to do */
1037                                 hwgroup->sleeping = 1;
1038                                 mod_timer(&hwgroup->timer, sleep);
1039                                 /* we purposely leave hwgroup->busy==1
1040                                  * while sleeping */
1041                         } else {
1042                                 /* Ugly, but how can we sleep for the lock
1043                                  * otherwise? perhaps from tq_disk?
1044                                  */
1045
1046                                 /* for atari only */
1047                                 ide_release_lock();
1048                                 hwgroup->busy = 0;
1049                         }
1050
1051                         /* no more work for this hwgroup (for now) */
1052                         return;
1053                 }
1054                 hwif = HWIF(drive);
1055                 if (hwgroup->hwif->sharing_irq &&
1056                     hwif != hwgroup->hwif &&
1057                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1058                         /* set nIEN for previous hwif */
1059                         SELECT_INTERRUPT(drive);
1060                 }
1061                 hwgroup->hwif = hwif;
1062                 hwgroup->drive = drive;
1063                 drive->sleep = 0;
1064                 drive->service_start = jiffies;
1065
1066                 if (blk_queue_plugged(drive->queue)) {
1067                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1068                         break;
1069                 }
1070
1071                 /*
1072                  * we know that the queue isn't empty, but this can happen
1073                  * if the q->prep_rq_fn() decides to kill a request
1074                  */
1075                 rq = elv_next_request(drive->queue);
1076                 if (!rq) {
1077                         hwgroup->busy = 0;
1078                         break;
1079                 }
1080
1081                 /*
1082                  * if rq is a barrier write, issue pre cache flush if not
1083                  * already done
1084                  */
1085                 if (blk_barrier_rq(rq) && !blk_barrier_preflush(rq))
1086                         rq = ide_queue_flush_cmd(drive, rq, 0);
1087
1088                 /*
1089                  * Sanity: don't accept a request that isn't a PM request
1090                  * if we are currently power managed. This is very important as
1091                  * blk_stop_queue() doesn't prevent the elv_next_request()
1092                  * above to return us whatever is in the queue. Since we call
1093                  * ide_do_request() ourselves, we end up taking requests while
1094                  * the queue is blocked...
1095                  *
1096                  * We let requests forced at head of queue with ide-preempt
1097                  * though. I hope that doesn't happen too much, hopefully not
1098                  * unless the subdriver triggers such a thing in its own PM
1099                  * state machine.
1100                  */
1101                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1102                         /* We clear busy, there should be no pending ATA command at this point. */
1103                         hwgroup->busy = 0;
1104                         break;
1105                 }
1106
1107                 hwgroup->rq = rq;
1108
1109                 /*
1110                  * Some systems have trouble with IDE IRQs arriving while
1111                  * the driver is still setting things up.  So, here we disable
1112                  * the IRQ used by this interface while the request is being started.
1113                  * This may look bad at first, but pretty much the same thing
1114                  * happens anyway when any interrupt comes in, IDE or otherwise
1115                  *  -- the kernel masks the IRQ while it is being handled.
1116                  */
1117                 if (hwif->irq != masked_irq)
1118                         disable_irq_nosync(hwif->irq);
1119                 spin_unlock(&ide_lock);
1120                 local_irq_enable();
1121                         /* allow other IRQs while we start this request */
1122                 startstop = start_request(drive, rq);
1123                 spin_lock_irq(&ide_lock);
1124                 if (hwif->irq != masked_irq)
1125                         enable_irq(hwif->irq);
1126                 if (startstop == ide_stopped)
1127                         hwgroup->busy = 0;
1128         }
1129 }
1130
1131 EXPORT_SYMBOL(ide_do_request);
1132
1133 /*
1134  * Passes the stuff to ide_do_request
1135  */
1136 void do_ide_request(request_queue_t *q)
1137 {
1138         ide_drive_t *drive = q->queuedata;
1139
1140         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1141 }
1142
1143 /*
1144  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1145  * retry the current request in pio mode instead of risking tossing it
1146  * all away
1147  */
1148 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1149 {
1150         ide_hwif_t *hwif = HWIF(drive);
1151         struct request *rq;
1152         ide_startstop_t ret = ide_stopped;
1153
1154         /*
1155          * end current dma transaction
1156          */
1157
1158         if (error < 0) {
1159                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1160                 (void)HWIF(drive)->ide_dma_end(drive);
1161                 ret = DRIVER(drive)->error(drive, "dma timeout error",
1162                                                 hwif->INB(IDE_STATUS_REG));
1163         } else {
1164                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1165                 (void) hwif->ide_dma_timeout(drive);
1166         }
1167
1168         /*
1169          * disable dma for now, but remember that we did so because of
1170          * a timeout -- we'll reenable after we finish this next request
1171          * (or rather the first chunk of it) in pio.
1172          */
1173         drive->retry_pio++;
1174         drive->state = DMA_PIO_RETRY;
1175         (void) hwif->ide_dma_off_quietly(drive);
1176
1177         /*
1178          * un-busy drive etc (hwgroup->busy is cleared on return) and
1179          * make sure request is sane
1180          */
1181         rq = HWGROUP(drive)->rq;
1182         HWGROUP(drive)->rq = NULL;
1183
1184         rq->errors = 0;
1185         rq->sector = rq->bio->bi_sector;
1186         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1187         rq->hard_cur_sectors = rq->current_nr_sectors;
1188         if (rq->bio)
1189                 rq->buffer = NULL;
1190
1191         return ret;
1192 }
1193
1194 /**
1195  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1196  *      @data: timer callback magic (hwgroup)
1197  *
1198  *      An IDE command has timed out before the expected drive return
1199  *      occurred. At this point we attempt to clean up the current
1200  *      mess. If the current handler includes an expiry handler then
1201  *      we invoke the expiry handler, and providing it is happy the
1202  *      work is done. If that fails we apply generic recovery rules
1203  *      invoking the handler and checking the drive DMA status. We
1204  *      have an excessively incestuous relationship with the DMA
1205  *      logic that wants cleaning up.
1206  */
1207
1208 void ide_timer_expiry (unsigned long data)
1209 {
1210         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1211         ide_handler_t   *handler;
1212         ide_expiry_t    *expiry;
1213         unsigned long   flags;
1214         unsigned long   wait = -1;
1215
1216         spin_lock_irqsave(&ide_lock, flags);
1217
1218         if ((handler = hwgroup->handler) == NULL) {
1219                 /*
1220                  * Either a marginal timeout occurred
1221                  * (got the interrupt just as timer expired),
1222                  * or we were "sleeping" to give other devices a chance.
1223                  * Either way, we don't really want to complain about anything.
1224                  */
1225                 if (hwgroup->sleeping) {
1226                         hwgroup->sleeping = 0;
1227                         hwgroup->busy = 0;
1228                 }
1229         } else {
1230                 ide_drive_t *drive = hwgroup->drive;
1231                 if (!drive) {
1232                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1233                         hwgroup->handler = NULL;
1234                 } else {
1235                         ide_hwif_t *hwif;
1236                         ide_startstop_t startstop = ide_stopped;
1237                         if (!hwgroup->busy) {
1238                                 hwgroup->busy = 1;      /* paranoia */
1239                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1240                         }
1241                         if ((expiry = hwgroup->expiry) != NULL) {
1242                                 /* continue */
1243                                 if ((wait = expiry(drive)) > 0) {
1244                                         /* reset timer */
1245                                         hwgroup->timer.expires  = jiffies + wait;
1246                                         add_timer(&hwgroup->timer);
1247                                         spin_unlock_irqrestore(&ide_lock, flags);
1248                                         return;
1249                                 }
1250                         }
1251                         hwgroup->handler = NULL;
1252                         /*
1253                          * We need to simulate a real interrupt when invoking
1254                          * the handler() function, which means we need to
1255                          * globally mask the specific IRQ:
1256                          */
1257                         spin_unlock(&ide_lock);
1258                         hwif  = HWIF(drive);
1259 #if DISABLE_IRQ_NOSYNC
1260                         disable_irq_nosync(hwif->irq);
1261 #else
1262                         /* disable_irq_nosync ?? */
1263                         disable_irq(hwif->irq);
1264 #endif /* DISABLE_IRQ_NOSYNC */
1265                         /* local CPU only,
1266                          * as if we were handling an interrupt */
1267                         local_irq_disable();
1268                         if (hwgroup->poll_timeout != 0) {
1269                                 startstop = handler(drive);
1270                         } else if (drive_is_ready(drive)) {
1271                                 if (drive->waiting_for_dma)
1272                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1273                                 (void)ide_ack_intr(hwif);
1274                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1275                                 startstop = handler(drive);
1276                         } else {
1277                                 if (drive->waiting_for_dma) {
1278                                         startstop = ide_dma_timeout_retry(drive, wait);
1279                                 } else
1280                                         startstop =
1281                                         DRIVER(drive)->error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1282                         }
1283                         drive->service_time = jiffies - drive->service_start;
1284                         spin_lock_irq(&ide_lock);
1285                         enable_irq(hwif->irq);
1286                         if (startstop == ide_stopped)
1287                                 hwgroup->busy = 0;
1288                 }
1289         }
1290         ide_do_request(hwgroup, IDE_NO_IRQ);
1291         spin_unlock_irqrestore(&ide_lock, flags);
1292 }
1293
1294 EXPORT_SYMBOL(ide_timer_expiry);
1295
1296 /**
1297  *      unexpected_intr         -       handle an unexpected IDE interrupt
1298  *      @irq: interrupt line
1299  *      @hwgroup: hwgroup being processed
1300  *
1301  *      There's nothing really useful we can do with an unexpected interrupt,
1302  *      other than reading the status register (to clear it), and logging it.
1303  *      There should be no way that an irq can happen before we're ready for it,
1304  *      so we needn't worry much about losing an "important" interrupt here.
1305  *
1306  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1307  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1308  *      looks "good", we just ignore the interrupt completely.
1309  *
1310  *      This routine assumes __cli() is in effect when called.
1311  *
1312  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1313  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1314  *      we could screw up by interfering with a new request being set up for
1315  *      irq15.
1316  *
1317  *      In reality, this is a non-issue.  The new command is not sent unless
1318  *      the drive is ready to accept one, in which case we know the drive is
1319  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1320  *      before completing the issuance of any new drive command, so we will not
1321  *      be accidentally invoked as a result of any valid command completion
1322  *      interrupt.
1323  *
1324  *      Note that we must walk the entire hwgroup here. We know which hwif
1325  *      is doing the current command, but we don't know which hwif burped
1326  *      mysteriously.
1327  */
1328
1329 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1330 {
1331         u8 stat;
1332         ide_hwif_t *hwif = hwgroup->hwif;
1333
1334         /*
1335          * handle the unexpected interrupt
1336          */
1337         do {
1338                 if (hwif->irq == irq) {
1339                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1340                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1341                                 /* Try to not flood the console with msgs */
1342                                 static unsigned long last_msgtime, count;
1343                                 ++count;
1344                                 if (time_after(jiffies, last_msgtime + HZ)) {
1345                                         last_msgtime = jiffies;
1346                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1347                                                 "status=0x%02x, count=%ld\n",
1348                                                 hwif->name,
1349                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1350                                 }
1351                         }
1352                 }
1353         } while ((hwif = hwif->next) != hwgroup->hwif);
1354 }
1355
1356 /**
1357  *      ide_intr        -       default IDE interrupt handler
1358  *      @irq: interrupt number
1359  *      @dev_id: hwif group
1360  *      @regs: unused weirdness from the kernel irq layer
1361  *
1362  *      This is the default IRQ handler for the IDE layer. You should
1363  *      not need to override it. If you do be aware it is subtle in
1364  *      places
1365  *
1366  *      hwgroup->hwif is the interface in the group currently performing
1367  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1368  *      the IRQ handler to call. As we issue a command the handlers
1369  *      step through multiple states, reassigning the handler to the
1370  *      next step in the process. Unlike a smart SCSI controller IDE
1371  *      expects the main processor to sequence the various transfer
1372  *      stages. We also manage a poll timer to catch up with most
1373  *      timeout situations. There are still a few where the handlers
1374  *      don't ever decide to give up.
1375  *
1376  *      The handler eventually returns ide_stopped to indicate the
1377  *      request completed. At this point we issue the next request
1378  *      on the hwgroup and the process begins again.
1379  */
1380
1381 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1382 {
1383         unsigned long flags;
1384         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1385         ide_hwif_t *hwif;
1386         ide_drive_t *drive;
1387         ide_handler_t *handler;
1388         ide_startstop_t startstop;
1389
1390         spin_lock_irqsave(&ide_lock, flags);
1391         hwif = hwgroup->hwif;
1392
1393         if (!ide_ack_intr(hwif)) {
1394                 spin_unlock_irqrestore(&ide_lock, flags);
1395                 return IRQ_NONE;
1396         }
1397
1398         if ((handler = hwgroup->handler) == NULL ||
1399             hwgroup->poll_timeout != 0) {
1400                 /*
1401                  * Not expecting an interrupt from this drive.
1402                  * That means this could be:
1403                  *      (1) an interrupt from another PCI device
1404                  *      sharing the same PCI INT# as us.
1405                  * or   (2) a drive just entered sleep or standby mode,
1406                  *      and is interrupting to let us know.
1407                  * or   (3) a spurious interrupt of unknown origin.
1408                  *
1409                  * For PCI, we cannot tell the difference,
1410                  * so in that case we just ignore it and hope it goes away.
1411                  *
1412                  * FIXME: unexpected_intr should be hwif-> then we can
1413                  * remove all the ifdef PCI crap
1414                  */
1415 #ifdef CONFIG_BLK_DEV_IDEPCI
1416                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1417 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1418                 {
1419                         /*
1420                          * Probably not a shared PCI interrupt,
1421                          * so we can safely try to do something about it:
1422                          */
1423                         unexpected_intr(irq, hwgroup);
1424 #ifdef CONFIG_BLK_DEV_IDEPCI
1425                 } else {
1426                         /*
1427                          * Whack the status register, just in case
1428                          * we have a leftover pending IRQ.
1429                          */
1430                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1431 #endif /* CONFIG_BLK_DEV_IDEPCI */
1432                 }
1433                 spin_unlock_irqrestore(&ide_lock, flags);
1434                 return IRQ_NONE;
1435         }
1436         drive = hwgroup->drive;
1437         if (!drive) {
1438                 /*
1439                  * This should NEVER happen, and there isn't much
1440                  * we could do about it here.
1441                  *
1442                  * [Note - this can occur if the drive is hot unplugged]
1443                  */
1444                 spin_unlock_irqrestore(&ide_lock, flags);
1445                 return IRQ_HANDLED;
1446         }
1447         if (!drive_is_ready(drive)) {
1448                 /*
1449                  * This happens regularly when we share a PCI IRQ with
1450                  * another device.  Unfortunately, it can also happen
1451                  * with some buggy drives that trigger the IRQ before
1452                  * their status register is up to date.  Hopefully we have
1453                  * enough advance overhead that the latter isn't a problem.
1454                  */
1455                 spin_unlock_irqrestore(&ide_lock, flags);
1456                 return IRQ_NONE;
1457         }
1458         if (!hwgroup->busy) {
1459                 hwgroup->busy = 1;      /* paranoia */
1460                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1461         }
1462         hwgroup->handler = NULL;
1463         del_timer(&hwgroup->timer);
1464         spin_unlock(&ide_lock);
1465
1466         if (drive->unmask)
1467                 local_irq_enable();
1468         /* service this interrupt, may set handler for next interrupt */
1469         startstop = handler(drive);
1470         spin_lock_irq(&ide_lock);
1471
1472         /*
1473          * Note that handler() may have set things up for another
1474          * interrupt to occur soon, but it cannot happen until
1475          * we exit from this routine, because it will be the
1476          * same irq as is currently being serviced here, and Linux
1477          * won't allow another of the same (on any CPU) until we return.
1478          */
1479         drive->service_time = jiffies - drive->service_start;
1480         if (startstop == ide_stopped) {
1481                 if (hwgroup->handler == NULL) { /* paranoia */
1482                         hwgroup->busy = 0;
1483                         ide_do_request(hwgroup, hwif->irq);
1484                 } else {
1485                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1486                                 "on exit\n", drive->name);
1487                 }
1488         }
1489         spin_unlock_irqrestore(&ide_lock, flags);
1490         return IRQ_HANDLED;
1491 }
1492
1493 EXPORT_SYMBOL(ide_intr);
1494
1495 /**
1496  *      ide_init_drive_cmd      -       initialize a drive command request
1497  *      @rq: request object
1498  *
1499  *      Initialize a request before we fill it in and send it down to
1500  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1501  *      now it doesn't do a lot, but if that changes abusers will have a
1502  *      nasty suprise.
1503  */
1504
1505 void ide_init_drive_cmd (struct request *rq)
1506 {
1507         memset(rq, 0, sizeof(*rq));
1508         rq->flags = REQ_DRIVE_CMD;
1509         rq->ref_count = 1;
1510 }
1511
1512 EXPORT_SYMBOL(ide_init_drive_cmd);
1513
1514 /**
1515  *      ide_do_drive_cmd        -       issue IDE special command
1516  *      @drive: device to issue command
1517  *      @rq: request to issue
1518  *      @action: action for processing
1519  *
1520  *      This function issues a special IDE device request
1521  *      onto the request queue.
1522  *
1523  *      If action is ide_wait, then the rq is queued at the end of the
1524  *      request queue, and the function sleeps until it has been processed.
1525  *      This is for use when invoked from an ioctl handler.
1526  *
1527  *      If action is ide_preempt, then the rq is queued at the head of
1528  *      the request queue, displacing the currently-being-processed
1529  *      request and this function returns immediately without waiting
1530  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1531  *      intended for careful use by the ATAPI tape/cdrom driver code.
1532  *
1533  *      If action is ide_next, then the rq is queued immediately after
1534  *      the currently-being-processed-request (if any), and the function
1535  *      returns without waiting for the new rq to be completed.  As above,
1536  *      This is VERY DANGEROUS, and is intended for careful use by the
1537  *      ATAPI tape/cdrom driver code.
1538  *
1539  *      If action is ide_end, then the rq is queued at the end of the
1540  *      request queue, and the function returns immediately without waiting
1541  *      for the new rq to be completed. This is again intended for careful
1542  *      use by the ATAPI tape/cdrom driver code.
1543  */
1544
1545 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1546 {
1547         unsigned long flags;
1548         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1549         DECLARE_COMPLETION(wait);
1550         int where = ELEVATOR_INSERT_BACK, err;
1551         int must_wait = (action == ide_wait || action == ide_head_wait);
1552
1553 #ifdef CONFIG_BLK_DEV_PDC4030
1554         /*
1555          *      FIXME: there should be a drive or hwif->special
1556          *      handler that points here by default, not hacks
1557          *      in the ide-io.c code
1558          *
1559          *      FIXME2: That code breaks power management if used with
1560          *      this chipset, that really doesn't belong here !
1561          */
1562         if (HWIF(drive)->chipset == ide_pdc4030 && rq->buffer != NULL)
1563                 return -ENOSYS;  /* special drive cmds not supported */
1564 #endif
1565         rq->errors = 0;
1566         rq->rq_status = RQ_ACTIVE;
1567
1568         rq->rq_disk = drive->disk;
1569
1570         /*
1571          * we need to hold an extra reference to request for safe inspection
1572          * after completion
1573          */
1574         if (must_wait) {
1575                 rq->ref_count++;
1576                 rq->waiting = &wait;
1577         }
1578
1579         spin_lock_irqsave(&ide_lock, flags);
1580         if (action == ide_preempt)
1581                 hwgroup->rq = NULL;
1582         if (action == ide_preempt || action == ide_head_wait) {
1583                 where = ELEVATOR_INSERT_FRONT;
1584                 rq->flags |= REQ_PREEMPT;
1585         }
1586         __elv_add_request(drive->queue, rq, where, 0);
1587         ide_do_request(hwgroup, IDE_NO_IRQ);
1588         spin_unlock_irqrestore(&ide_lock, flags);
1589
1590         err = 0;
1591         if (must_wait) {
1592                 wait_for_completion(&wait);
1593                 rq->waiting = NULL;
1594                 if (rq->errors)
1595                         err = -EIO;
1596
1597                 blk_put_request(rq);
1598         }
1599
1600         return err;
1601 }
1602
1603 EXPORT_SYMBOL(ide_do_drive_cmd);