drivers/ide/ide-io.c

   1 /*
   2  *      IDE I/O functions
   3  *
   4  *      Basic PIO and command management functionality.
   5  *
   6  * This code was split off from ide.c. See ide.c for history and original
   7  * copyrights.
   8  *
   9  * This program is free software; you can redistribute it and/or modify it
  10  * under the terms of the GNU General Public License as published by the
  11  * Free Software Foundation; either version 2, or (at your option) any
  12  * later version.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  *
  19  * For the avoidance of doubt the "preferred form" of this code is one which
  20  * is in an open non patent encumbered format. Where cryptographic key signing
  21  * forms part of the process of creating an executable the information
  22  * including keys needed to generate an equivalently functional executable
  23  * are deemed to be part of the source code.
  24  */
  25
  26
  27 #include <linux/config.h>
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/string.h>
  31 #include <linux/kernel.h>
  32 #include <linux/timer.h>
  33 #include <linux/mm.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/major.h>
  36 #include <linux/errno.h>
  37 #include <linux/genhd.h>
  38 #include <linux/blkpg.h>
  39 #include <linux/slab.h>
  40 #include <linux/init.h>
  41 #include <linux/pci.h>
  42 #include <linux/delay.h>
  43 #include <linux/ide.h>
  44 #include <linux/completion.h>
  45 #include <linux/reboot.h>
  46 #include <linux/cdrom.h>
  47 #include <linux/seq_file.h>
  48 #include <linux/device.h>
  49 #include <linux/kmod.h>
  50 #include <linux/scatterlist.h>
  51
  52 #include <asm/byteorder.h>
  53 #include <asm/irq.h>
  54 #include <asm/uaccess.h>
  55 #include <asm/io.h>
  56 #include <asm/bitops.h>
  57
  58 void ide_softirq_done(struct request *rq)
  59 {
  60         request_queue_t *q = rq->q;
  61
  62         add_disk_randomness(rq->rq_disk);
  63         end_that_request_chunk(rq, rq->errors, rq->data_len);
  64
  65         spin_lock_irq(q->queue_lock);
  66         end_that_request_last(rq, rq->errors);
  67         spin_unlock_irq(q->queue_lock);
  68 }
  69
  70 int __ide_end_request(ide_drive_t *drive, struct request *rq, int uptodate,
  71                       int nr_sectors)
  72 {
  73         unsigned int nbytes;
  74         int ret = 1;
  75
  76         BUG_ON(!(rq->flags & REQ_STARTED));
  77
  78         /*
  79          * if failfast is set on a request, override number of sectors and
  80          * complete the whole request right now
  81          */
  82         if (blk_noretry_request(rq) && end_io_error(uptodate))
  83                 nr_sectors = rq->hard_nr_sectors;
  84
  85         if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
  86                 rq->errors = -EIO;
  87
  88         /*
  89          * decide whether to reenable DMA -- 3 is a random magic for now,
  90          * if we DMA timeout more than 3 times, just stay in PIO
  91          */
  92         if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
  93                 drive->state = 0;
  94                 HWGROUP(drive)->hwif->ide_dma_on(drive);
  95         }
  96
  97         /*
  98          * For partial completions (or non fs/pc requests), use the regular
  99          * direct completion path.
 100          */
 101         nbytes = nr_sectors << 9;
 102         if (rq_all_done(rq, nbytes)) {
 103                 rq->errors = uptodate;
 104                 rq->data_len = nbytes;
 105                 blkdev_dequeue_request(rq);
 106                 HWGROUP(drive)->rq = NULL;
 107                 blk_complete_request(rq);
 108                 ret = 0;
 109         } else {
 110                 if (!end_that_request_first(rq, uptodate, nr_sectors)) {
 111                         add_disk_randomness(rq->rq_disk);
 112                         blkdev_dequeue_request(rq);
 113                         HWGROUP(drive)->rq = NULL;
 114                         end_that_request_last(rq, uptodate);
 115                         ret = 0;
 116                 }
 117         }
 118
 119         return ret;
 120 }
 121 EXPORT_SYMBOL(__ide_end_request);
 122
 123 /**
 124  *      ide_end_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  *      This is our end_request wrapper function. We complete the I/O
 130  *      update random number input and dequeue the request, which if
 131  *      it was tagged may be out of order.
 132  */
 133
 134 int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
 135 {
 136         struct request *rq;
 137         unsigned long flags;
 138         int ret = 1;
 139
 140         /*
 141          * room for locking improvements here, the calls below don't
 142          * need the queue lock held at all
 143          */
 144         spin_lock_irqsave(&ide_lock, flags);
 145         rq = HWGROUP(drive)->rq;
 146
 147         if (!nr_sectors)
 148                 nr_sectors = rq->hard_cur_sectors;
 149
 150         ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
 151
 152         spin_unlock_irqrestore(&ide_lock, flags);
 153         return ret;
 154 }
 155 EXPORT_SYMBOL(ide_end_request);
 156
 157 /*
 158  * Power Management state machine. This one is rather trivial for now,
 159  * we should probably add more, like switching back to PIO on suspend
 160  * to help some BIOSes, re-do the door locking on resume, etc...
 161  */
 162
 163 enum {
 164         ide_pm_flush_cache      = ide_pm_state_start_suspend,
 165         idedisk_pm_standby,
 166
 167         idedisk_pm_idle         = ide_pm_state_start_resume,
 168         ide_pm_restore_dma,
 169 };
 170
 171 static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
 172 {
 173         if (drive->media != ide_disk)
 174                 return;
 175
 176         switch (rq->pm->pm_step) {
 177         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
 178                 if (rq->pm->pm_state == PM_EVENT_FREEZE)
 179                         rq->pm->pm_step = ide_pm_state_completed;
 180                 else
 181                         rq->pm->pm_step = idedisk_pm_standby;
 182                 break;
 183         case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
 184                 rq->pm->pm_step = ide_pm_state_completed;
 185                 break;
 186         case idedisk_pm_idle:           /* Resume step 1 (idle) complete */
 187                 rq->pm->pm_step = ide_pm_restore_dma;
 188                 break;
 189         }
 190 }
 191
 192 static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
 193 {
 194         ide_task_t *args = rq->special;
 195
 196         memset(args, 0, sizeof(*args));
 197
 198         if (drive->media != ide_disk) {
 199                 /* skip idedisk_pm_idle for ATAPI devices */
 200                 if (rq->pm->pm_step == idedisk_pm_idle)
 201                         rq->pm->pm_step = ide_pm_restore_dma;
 202         }
 203
 204         switch (rq->pm->pm_step) {
 205         case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
 206                 if (drive->media != ide_disk)
 207                         break;
 208                 /* Not supported? Switch to next step now. */
 209                 if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
 210                         ide_complete_power_step(drive, rq, 0, 0);
 211                         return ide_stopped;
 212                 }
 213                 if (ide_id_has_flush_cache_ext(drive->id))
 214                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
 215                 else
 216                         args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
 217                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 218                 args->handler      = &task_no_data_intr;
 219                 return do_rw_taskfile(drive, args);
 220
 221         case idedisk_pm_standby:        /* Suspend step 2 (standby) */
 222                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
 223                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 224                 args->handler      = &task_no_data_intr;
 225                 return do_rw_taskfile(drive, args);
 226
 227         case idedisk_pm_idle:           /* Resume step 1 (idle) */
 228                 args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
 229                 args->command_type = IDE_DRIVE_TASK_NO_DATA;
 230                 args->handler = task_no_data_intr;
 231                 return do_rw_taskfile(drive, args);
 232
 233         case ide_pm_restore_dma:        /* Resume step 2 (restore DMA) */
 234                 /*
 235                  * Right now, all we do is call hwif->ide_dma_check(drive),
 236                  * we could be smarter and check for current xfer_speed
 237                  * in struct drive etc...
 238                  */
 239                 if ((drive->id->capability & 1) == 0)
 240                         break;
 241                 if (drive->hwif->ide_dma_check == NULL)
 242                         break;
 243                 drive->hwif->ide_dma_check(drive);
 244                 break;
 245         }
 246         rq->pm->pm_step = ide_pm_state_completed;
 247         return ide_stopped;
 248 }
 249
 250 /**
 251  *      ide_complete_pm_request - end the current Power Management request
 252  *      @drive: target drive
 253  *      @rq: request
 254  *
 255  *      This function cleans up the current PM request and stops the queue
 256  *      if necessary.
 257  */
 258 static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
 259 {
 260         unsigned long flags;
 261
 262 #ifdef DEBUG_PM
 263         printk("%s: completing PM request, %s\n", drive->name,
 264                blk_pm_suspend_request(rq) ? "suspend" : "resume");
 265 #endif
 266         spin_lock_irqsave(&ide_lock, flags);
 267         if (blk_pm_suspend_request(rq)) {
 268                 blk_stop_queue(drive->queue);
 269         } else {
 270                 drive->blocked = 0;
 271                 blk_start_queue(drive->queue);
 272         }
 273         blkdev_dequeue_request(rq);
 274         HWGROUP(drive)->rq = NULL;
 275         end_that_request_last(rq, 1);
 276         spin_unlock_irqrestore(&ide_lock, flags);
 277 }
 278
 279 /*
 280  * FIXME: probably move this somewhere else, name is bad too :)
 281  */
 282 u64 ide_get_error_location(ide_drive_t *drive, char *args)
 283 {
 284         u32 high, low;
 285         u8 hcyl, lcyl, sect;
 286         u64 sector;
 287
 288         high = 0;
 289         hcyl = args[5];
 290         lcyl = args[4];
 291         sect = args[3];
 292
 293         if (ide_id_has_flush_cache_ext(drive->id)) {
 294                 low = (hcyl << 16) | (lcyl << 8) | sect;
 295                 HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 296                 high = ide_read_24(drive);
 297         } else {
 298                 u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
 299                 if (cur & 0x40) {
 300                         high = cur & 0xf;
 301                         low = (hcyl << 16) | (lcyl << 8) | sect;
 302                 } else {
 303                         low = hcyl * drive->head * drive->sect;
 304                         low += lcyl * drive->sect;
 305                         low += sect - 1;
 306                 }
 307         }
 308
 309         sector = ((u64) high << 24) | low;
 310         return sector;
 311 }
 312 EXPORT_SYMBOL(ide_get_error_location);
 313
 314 /**
 315  *      ide_end_drive_cmd       -       end an explicit drive command
 316  *      @drive: command
 317  *      @stat: status bits
 318  *      @err: error bits
 319  *
 320  *      Clean up after success/failure of an explicit drive command.
 321  *      These get thrown onto the queue so they are synchronized with
 322  *      real I/O operations on the drive.
 323  *
 324  *      In LBA48 mode we have to read the register set twice to get
 325  *      all the extra information out.
 326  */
 327
 328 void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
 329 {
 330         ide_hwif_t *hwif = HWIF(drive);
 331         unsigned long flags;
 332         struct request *rq;
 333
 334         spin_lock_irqsave(&ide_lock, flags);
 335         rq = HWGROUP(drive)->rq;
 336         spin_unlock_irqrestore(&ide_lock, flags);
 337
 338         if (rq->flags & REQ_DRIVE_CMD) {
 339                 u8 *args = (u8 *) rq->buffer;
 340                 if (rq->errors == 0)
 341                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 342
 343                 if (args) {
 344                         args[0] = stat;
 345                         args[1] = err;
 346                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 347                 }
 348         } else if (rq->flags & REQ_DRIVE_TASK) {
 349                 u8 *args = (u8 *) rq->buffer;
 350                 if (rq->errors == 0)
 351                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 352
 353                 if (args) {
 354                         args[0] = stat;
 355                         args[1] = err;
 356                         args[2] = hwif->INB(IDE_NSECTOR_REG);
 357                         args[3] = hwif->INB(IDE_SECTOR_REG);
 358                         args[4] = hwif->INB(IDE_LCYL_REG);
 359                         args[5] = hwif->INB(IDE_HCYL_REG);
 360                         args[6] = hwif->INB(IDE_SELECT_REG);
 361                 }
 362         } else if (rq->flags & REQ_DRIVE_TASKFILE) {
 363                 ide_task_t *args = (ide_task_t *) rq->special;
 364                 if (rq->errors == 0)
 365                         rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
 366
 367                 if (args) {
 368                         if (args->tf_in_flags.b.data) {
 369                                 u16 data                                = hwif->INW(IDE_DATA_REG);
 370                                 args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
 371                                 args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
 372                         }
 373                         args->tfRegister[IDE_ERROR_OFFSET]   = err;
 374                         /* be sure we're looking at the low order bits */
 375                         hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
 376                         args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
 377                         args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
 378                         args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
 379                         args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
 380                         args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
 381                         args->tfRegister[IDE_STATUS_OFFSET]  = stat;
 382
 383                         if (drive->addressing == 1) {
 384                                 hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
 385                                 args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
 386                                 args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
 387                                 args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
 388                                 args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
 389                                 args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
 390                         }
 391                 }
 392         } else if (blk_pm_request(rq)) {
 393 #ifdef DEBUG_PM
 394                 printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
 395                         drive->name, rq->pm->pm_step, stat, err);
 396 #endif
 397                 ide_complete_power_step(drive, rq, stat, err);
 398                 if (rq->pm->pm_step == ide_pm_state_completed)
 399                         ide_complete_pm_request(drive, rq);
 400                 return;
 401         }
 402
 403         spin_lock_irqsave(&ide_lock, flags);
 404         blkdev_dequeue_request(rq);
 405         HWGROUP(drive)->rq = NULL;
 406         rq->errors = err;
 407         end_that_request_last(rq, !rq->errors);
 408         spin_unlock_irqrestore(&ide_lock, flags);
 409 }
 410
 411 EXPORT_SYMBOL(ide_end_drive_cmd);
 412
 413 /**
 414  *      try_to_flush_leftover_data      -       flush junk
 415  *      @drive: drive to flush
 416  *
 417  *      try_to_flush_leftover_data() is invoked in response to a drive
 418  *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
 419  *      resetting the drive, this routine tries to clear the condition
 420  *      by read a sector's worth of data from the drive.  Of course,
 421  *      this may not help if the drive is *waiting* for data from *us*.
 422  */
 423 static void try_to_flush_leftover_data (ide_drive_t *drive)
 424 {
 425         int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
 426
 427         if (drive->media != ide_disk)
 428                 return;
 429         while (i > 0) {
 430                 u32 buffer[16];
 431                 u32 wcount = (i > 16) ? 16 : i;
 432
 433                 i -= wcount;
 434                 HWIF(drive)->ata_input_data(drive, buffer, wcount);
 435         }
 436 }
 437
 438 static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
 439 {
 440         if (rq->rq_disk) {
 441                 ide_driver_t *drv;
 442
 443                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 444                 drv->end_request(drive, 0, 0);
 445         } else
 446                 ide_end_request(drive, 0, 0);
 447 }
 448
 449 static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 450 {
 451         ide_hwif_t *hwif = drive->hwif;
 452
 453         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 454                 /* other bits are useless when BUSY */
 455                 rq->errors |= ERROR_RESET;
 456         } else if (stat & ERR_STAT) {
 457                 /* err has different meaning on cdrom and tape */
 458                 if (err == ABRT_ERR) {
 459                         if (drive->select.b.lba &&
 460                             /* some newer drives don't support WIN_SPECIFY */
 461                             hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
 462                                 return ide_stopped;
 463                 } else if ((err & BAD_CRC) == BAD_CRC) {
 464                         /* UDMA crc error, just retry the operation */
 465                         drive->crc_count++;
 466                 } else if (err & (BBD_ERR | ECC_ERR)) {
 467                         /* retries won't help these */
 468                         rq->errors = ERROR_MAX;
 469                 } else if (err & TRK0_ERR) {
 470                         /* help it find track zero */
 471                         rq->errors |= ERROR_RECAL;
 472                 }
 473         }
 474
 475         if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ)
 476                 try_to_flush_leftover_data(drive);
 477
 478         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 479                 /* force an abort */
 480                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 481
 482         if (rq->errors >= ERROR_MAX || blk_noretry_request(rq))
 483                 ide_kill_rq(drive, rq);
 484         else {
 485                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 486                         ++rq->errors;
 487                         return ide_do_reset(drive);
 488                 }
 489                 if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
 490                         drive->special.b.recalibrate = 1;
 491                 ++rq->errors;
 492         }
 493         return ide_stopped;
 494 }
 495
 496 static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 497 {
 498         ide_hwif_t *hwif = drive->hwif;
 499
 500         if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
 501                 /* other bits are useless when BUSY */
 502                 rq->errors |= ERROR_RESET;
 503         } else {
 504                 /* add decoding error stuff */
 505         }
 506
 507         if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
 508                 /* force an abort */
 509                 hwif->OUTB(WIN_IDLEIMMEDIATE, IDE_COMMAND_REG);
 510
 511         if (rq->errors >= ERROR_MAX) {
 512                 ide_kill_rq(drive, rq);
 513         } else {
 514                 if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
 515                         ++rq->errors;
 516                         return ide_do_reset(drive);
 517                 }
 518                 ++rq->errors;
 519         }
 520
 521         return ide_stopped;
 522 }
 523
 524 ide_startstop_t
 525 __ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
 526 {
 527         if (drive->media == ide_disk)
 528                 return ide_ata_error(drive, rq, stat, err);
 529         return ide_atapi_error(drive, rq, stat, err);
 530 }
 531
 532 EXPORT_SYMBOL_GPL(__ide_error);
 533
 534 /**
 535  *      ide_error       -       handle an error on the IDE
 536  *      @drive: drive the error occurred on
 537  *      @msg: message to report
 538  *      @stat: status bits
 539  *
 540  *      ide_error() takes action based on the error returned by the drive.
 541  *      For normal I/O that may well include retries. We deal with
 542  *      both new-style (taskfile) and old style command handling here.
 543  *      In the case of taskfile command handling there is work left to
 544  *      do
 545  */
 546
 547 ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
 548 {
 549         struct request *rq;
 550         u8 err;
 551
 552         err = ide_dump_status(drive, msg, stat);
 553
 554         if ((rq = HWGROUP(drive)->rq) == NULL)
 555                 return ide_stopped;
 556
 557         /* retry only "normal" I/O: */
 558         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 559                 rq->errors = 1;
 560                 ide_end_drive_cmd(drive, stat, err);
 561                 return ide_stopped;
 562         }
 563
 564         if (rq->rq_disk) {
 565                 ide_driver_t *drv;
 566
 567                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 568                 return drv->error(drive, rq, stat, err);
 569         } else
 570                 return __ide_error(drive, rq, stat, err);
 571 }
 572
 573 EXPORT_SYMBOL_GPL(ide_error);
 574
 575 ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
 576 {
 577         if (drive->media != ide_disk)
 578                 rq->errors |= ERROR_RESET;
 579
 580         ide_kill_rq(drive, rq);
 581
 582         return ide_stopped;
 583 }
 584
 585 EXPORT_SYMBOL_GPL(__ide_abort);
 586
 587 /**
 588  *      ide_abort       -       abort pending IDE operations
 589  *      @drive: drive the error occurred on
 590  *      @msg: message to report
 591  *
 592  *      ide_abort kills and cleans up when we are about to do a
 593  *      host initiated reset on active commands. Longer term we
 594  *      want handlers to have sensible abort handling themselves
 595  *
 596  *      This differs fundamentally from ide_error because in
 597  *      this case the command is doing just fine when we
 598  *      blow it away.
 599  */
 600
 601 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
 602 {
 603         struct request *rq;
 604
 605         if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
 606                 return ide_stopped;
 607
 608         /* retry only "normal" I/O: */
 609         if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK | REQ_DRIVE_TASKFILE)) {
 610                 rq->errors = 1;
 611                 ide_end_drive_cmd(drive, BUSY_STAT, 0);
 612                 return ide_stopped;
 613         }
 614
 615         if (rq->rq_disk) {
 616                 ide_driver_t *drv;
 617
 618                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 619                 return drv->abort(drive, rq);
 620         } else
 621                 return __ide_abort(drive, rq);
 622 }
 623
 624 /**
 625  *      ide_cmd         -       issue a simple drive command
 626  *      @drive: drive the command is for
 627  *      @cmd: command byte
 628  *      @nsect: sector byte
 629  *      @handler: handler for the command completion
 630  *
 631  *      Issue a simple drive command with interrupts.
 632  *      The drive must be selected beforehand.
 633  */
 634
 635 static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
 636                 ide_handler_t *handler)
 637 {
 638         ide_hwif_t *hwif = HWIF(drive);
 639         if (IDE_CONTROL_REG)
 640                 hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
 641         SELECT_MASK(drive,0);
 642         hwif->OUTB(nsect,IDE_NSECTOR_REG);
 643         ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
 644 }
 645
 646 /**
 647  *      drive_cmd_intr          -       drive command completion interrupt
 648  *      @drive: drive the completion interrupt occurred on
 649  *
 650  *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
 651  *      We do any necessary data reading and then wait for the drive to
 652  *      go non busy. At that point we may read the error data and complete
 653  *      the request
 654  */
 655
 656 static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
 657 {
 658         struct request *rq = HWGROUP(drive)->rq;
 659         ide_hwif_t *hwif = HWIF(drive);
 660         u8 *args = (u8 *) rq->buffer;
 661         u8 stat = hwif->INB(IDE_STATUS_REG);
 662         int retries = 10;
 663
 664         local_irq_enable();
 665         if ((stat & DRQ_STAT) && args && args[3]) {
 666                 u8 io_32bit = drive->io_32bit;
 667                 drive->io_32bit = 0;
 668                 hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
 669                 drive->io_32bit = io_32bit;
 670                 while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
 671                         udelay(100);
 672         }
 673
 674         if (!OK_STAT(stat, READY_STAT, BAD_STAT))
 675                 return ide_error(drive, "drive_cmd", stat);
 676                 /* calls ide_end_drive_cmd */
 677         ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
 678         return ide_stopped;
 679 }
 680
 681 static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
 682 {
 683         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
 684         task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
 685         task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
 686         task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
 687         task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
 688         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
 689
 690         task->handler = &set_geometry_intr;
 691 }
 692
 693 static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
 694 {
 695         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
 696         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
 697
 698         task->handler = &recal_intr;
 699 }
 700
 701 static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
 702 {
 703         task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
 704         task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
 705
 706         task->handler = &set_multmode_intr;
 707 }
 708
 709 static ide_startstop_t ide_disk_special(ide_drive_t *drive)
 710 {
 711         special_t *s = &drive->special;
 712         ide_task_t args;
 713
 714         memset(&args, 0, sizeof(ide_task_t));
 715         args.command_type = IDE_DRIVE_TASK_NO_DATA;
 716
 717         if (s->b.set_geometry) {
 718                 s->b.set_geometry = 0;
 719                 ide_init_specify_cmd(drive, &args);
 720         } else if (s->b.recalibrate) {
 721                 s->b.recalibrate = 0;
 722                 ide_init_restore_cmd(drive, &args);
 723         } else if (s->b.set_multmode) {
 724                 s->b.set_multmode = 0;
 725                 if (drive->mult_req > drive->id->max_multsect)
 726                         drive->mult_req = drive->id->max_multsect;
 727                 ide_init_setmult_cmd(drive, &args);
 728         } else if (s->all) {
 729                 int special = s->all;
 730                 s->all = 0;
 731                 printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
 732                 return ide_stopped;
 733         }
 734
 735         do_rw_taskfile(drive, &args);
 736
 737         return ide_started;
 738 }
 739
 740 /**
 741  *      do_special              -       issue some special commands
 742  *      @drive: drive the command is for
 743  *
 744  *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
 745  *      commands to a drive.  It used to do much more, but has been scaled
 746  *      back.
 747  */
 748
 749 static ide_startstop_t do_special (ide_drive_t *drive)
 750 {
 751         special_t *s = &drive->special;
 752
 753 #ifdef DEBUG
 754         printk("%s: do_special: 0x%02x\n", drive->name, s->all);
 755 #endif
 756         if (s->b.set_tune) {
 757                 s->b.set_tune = 0;
 758                 if (HWIF(drive)->tuneproc != NULL)
 759                         HWIF(drive)->tuneproc(drive, drive->tune_req);
 760                 return ide_stopped;
 761         } else {
 762                 if (drive->media == ide_disk)
 763                         return ide_disk_special(drive);
 764
 765                 s->all = 0;
 766                 drive->mult_req = 0;
 767                 return ide_stopped;
 768         }
 769 }
 770
 771 void ide_map_sg(ide_drive_t *drive, struct request *rq)
 772 {
 773         ide_hwif_t *hwif = drive->hwif;
 774         struct scatterlist *sg = hwif->sg_table;
 775
 776         if (hwif->sg_mapped)    /* needed by ide-scsi */
 777                 return;
 778
 779         if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
 780                 hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
 781         } else {
 782                 sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
 783                 hwif->sg_nents = 1;
 784         }
 785 }
 786
 787 EXPORT_SYMBOL_GPL(ide_map_sg);
 788
 789 void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
 790 {
 791         ide_hwif_t *hwif = drive->hwif;
 792
 793         hwif->nsect = hwif->nleft = rq->nr_sectors;
 794         hwif->cursg = hwif->cursg_ofs = 0;
 795 }
 796
 797 EXPORT_SYMBOL_GPL(ide_init_sg_cmd);
 798
 799 /**
 800  *      execute_drive_command   -       issue special drive command
 801  *      @drive: the drive to issue the command on
 802  *      @rq: the request structure holding the command
 803  *
 804  *      execute_drive_cmd() issues a special drive command,  usually
 805  *      initiated by ioctl() from the external hdparm program. The
 806  *      command can be a drive command, drive task or taskfile
 807  *      operation. Weirdly you can call it with NULL to wait for
 808  *      all commands to finish. Don't do this as that is due to change
 809  */
 810
 811 static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
 812                 struct request *rq)
 813 {
 814         ide_hwif_t *hwif = HWIF(drive);
 815         if (rq->flags & REQ_DRIVE_TASKFILE) {
 816                 ide_task_t *args = rq->special;
 817
 818                 if (!args)
 819                         goto done;
 820
 821                 hwif->data_phase = args->data_phase;
 822
 823                 switch (hwif->data_phase) {
 824                 case TASKFILE_MULTI_OUT:
 825                 case TASKFILE_OUT:
 826                 case TASKFILE_MULTI_IN:
 827                 case TASKFILE_IN:
 828                         ide_init_sg_cmd(drive, rq);
 829                         ide_map_sg(drive, rq);
 830                 default:
 831                         break;
 832                 }
 833
 834                 if (args->tf_out_flags.all != 0)
 835                         return flagged_taskfile(drive, args);
 836                 return do_rw_taskfile(drive, args);
 837         } else if (rq->flags & REQ_DRIVE_TASK) {
 838                 u8 *args = rq->buffer;
 839                 u8 sel;
 840
 841                 if (!args)
 842                         goto done;
 843 #ifdef DEBUG
 844                 printk("%s: DRIVE_TASK_CMD ", drive->name);
 845                 printk("cmd=0x%02x ", args[0]);
 846                 printk("fr=0x%02x ", args[1]);
 847                 printk("ns=0x%02x ", args[2]);
 848                 printk("sc=0x%02x ", args[3]);
 849                 printk("lcyl=0x%02x ", args[4]);
 850                 printk("hcyl=0x%02x ", args[5]);
 851                 printk("sel=0x%02x\n", args[6]);
 852 #endif
 853                 hwif->OUTB(args[1], IDE_FEATURE_REG);
 854                 hwif->OUTB(args[3], IDE_SECTOR_REG);
 855                 hwif->OUTB(args[4], IDE_LCYL_REG);
 856                 hwif->OUTB(args[5], IDE_HCYL_REG);
 857                 sel = (args[6] & ~0x10);
 858                 if (drive->select.b.unit)
 859                         sel |= 0x10;
 860                 hwif->OUTB(sel, IDE_SELECT_REG);
 861                 ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
 862                 return ide_started;
 863         } else if (rq->flags & REQ_DRIVE_CMD) {
 864                 u8 *args = rq->buffer;
 865
 866                 if (!args)
 867                         goto done;
 868 #ifdef DEBUG
 869                 printk("%s: DRIVE_CMD ", drive->name);
 870                 printk("cmd=0x%02x ", args[0]);
 871                 printk("sc=0x%02x ", args[1]);
 872                 printk("fr=0x%02x ", args[2]);
 873                 printk("xx=0x%02x\n", args[3]);
 874 #endif
 875                 if (args[0] == WIN_SMART) {
 876                         hwif->OUTB(0x4f, IDE_LCYL_REG);
 877                         hwif->OUTB(0xc2, IDE_HCYL_REG);
 878                         hwif->OUTB(args[2],IDE_FEATURE_REG);
 879                         hwif->OUTB(args[1],IDE_SECTOR_REG);
 880                         ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
 881                         return ide_started;
 882                 }
 883                 hwif->OUTB(args[2],IDE_FEATURE_REG);
 884                 ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
 885                 return ide_started;
 886         }
 887
 888 done:
 889         /*
 890          * NULL is actually a valid way of waiting for
 891          * all current requests to be flushed from the queue.
 892          */
 893 #ifdef DEBUG
 894         printk("%s: DRIVE_CMD (null)\n", drive->name);
 895 #endif
 896         ide_end_drive_cmd(drive,
 897                         hwif->INB(IDE_STATUS_REG),
 898                         hwif->INB(IDE_ERROR_REG));
 899         return ide_stopped;
 900 }
 901
 902 /**
 903  *      start_request   -       start of I/O and command issuing for IDE
 904  *
 905  *      start_request() initiates handling of a new I/O request. It
 906  *      accepts commands and I/O (read/write) requests. It also does
 907  *      the final remapping for weird stuff like EZDrive. Once
 908  *      device mapper can work sector level the EZDrive stuff can go away
 909  *
 910  *      FIXME: this function needs a rename
 911  */
 912
 913 static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
 914 {
 915         ide_startstop_t startstop;
 916         sector_t block;
 917
 918         BUG_ON(!(rq->flags & REQ_STARTED));
 919
 920 #ifdef DEBUG
 921         printk("%s: start_request: current=0x%08lx\n",
 922                 HWIF(drive)->name, (unsigned long) rq);
 923 #endif
 924
 925         /* bail early if we've exceeded max_failures */
 926         if (drive->max_failures && (drive->failures > drive->max_failures)) {
 927                 goto kill_rq;
 928         }
 929
 930         block    = rq->sector;
 931         if (blk_fs_request(rq) &&
 932             (drive->media == ide_disk || drive->media == ide_floppy)) {
 933                 block += drive->sect0;
 934         }
 935         /* Yecch - this will shift the entire interval,
 936            possibly killing some innocent following sector */
 937         if (block == 0 && drive->remap_0_to_1 == 1)
 938                 block = 1;  /* redirect MBR access to EZ-Drive partn table */
 939
 940         if (blk_pm_suspend_request(rq) &&
 941             rq->pm->pm_step == ide_pm_state_start_suspend)
 942                 /* Mark drive blocked when starting the suspend sequence. */
 943                 drive->blocked = 1;
 944         else if (blk_pm_resume_request(rq) &&
 945                  rq->pm->pm_step == ide_pm_state_start_resume) {
 946                 /*
 947                  * The first thing we do on wakeup is to wait for BSY bit to
 948                  * go away (with a looong timeout) as a drive on this hwif may
 949                  * just be POSTing itself.
 950                  * We do that before even selecting as the "other" device on
 951                  * the bus may be broken enough to walk on our toes at this
 952                  * point.
 953                  */
 954                 int rc;
 955 #ifdef DEBUG_PM
 956                 printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
 957 #endif
 958                 rc = ide_wait_not_busy(HWIF(drive), 35000);
 959                 if (rc)
 960                         printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
 961                 SELECT_DRIVE(drive);
 962                 HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
 963                 rc = ide_wait_not_busy(HWIF(drive), 10000);
 964                 if (rc)
 965                         printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
 966         }
 967
 968         SELECT_DRIVE(drive);
 969         if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
 970                 printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
 971                 return startstop;
 972         }
 973         if (!drive->special.all) {
 974                 ide_driver_t *drv;
 975
 976                 if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
 977                         return execute_drive_cmd(drive, rq);
 978                 else if (rq->flags & REQ_DRIVE_TASKFILE)
 979                         return execute_drive_cmd(drive, rq);
 980                 else if (blk_pm_request(rq)) {
 981 #ifdef DEBUG_PM
 982                         printk("%s: start_power_step(step: %d)\n",
 983                                 drive->name, rq->pm->pm_step);
 984 #endif
 985                         startstop = ide_start_power_step(drive, rq);
 986                         if (startstop == ide_stopped &&
 987                             rq->pm->pm_step == ide_pm_state_completed)
 988                                 ide_complete_pm_request(drive, rq);
 989                         return startstop;
 990                 }
 991
 992                 drv = *(ide_driver_t **)rq->rq_disk->private_data;
 993                 return drv->do_request(drive, rq, block);
 994         }
 995         return do_special(drive);
 996 kill_rq:
 997         ide_kill_rq(drive, rq);
 998         return ide_stopped;
 999 }
1000
1001 /**
1002  *      ide_stall_queue         -       pause an IDE device
1003  *      @drive: drive to stall
1004  *      @timeout: time to stall for (jiffies)
1005  *
1006  *      ide_stall_queue() can be used by a drive to give excess bandwidth back
1007  *      to the hwgroup by sleeping for timeout jiffies.
1008  */
1009
1010 void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
1011 {
1012         if (timeout > WAIT_WORSTCASE)
1013                 timeout = WAIT_WORSTCASE;
1014         drive->sleep = timeout + jiffies;
1015         drive->sleeping = 1;
1016 }
1017
1018 EXPORT_SYMBOL(ide_stall_queue);
1019
1020 #define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)
1021
1022 /**
1023  *      choose_drive            -       select a drive to service
1024  *      @hwgroup: hardware group to select on
1025  *
1026  *      choose_drive() selects the next drive which will be serviced.
1027  *      This is necessary because the IDE layer can't issue commands
1028  *      to both drives on the same cable, unlike SCSI.
1029  */
1030
1031 static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
1032 {
1033         ide_drive_t *drive, *best;
1034
1035 repeat:
1036         best = NULL;
1037         drive = hwgroup->drive;
1038
1039         /*
1040          * drive is doing pre-flush, ordered write, post-flush sequence. even
1041          * though that is 3 requests, it must be seen as a single transaction.
1042          * we must not preempt this drive until that is complete
1043          */
1044         if (blk_queue_flushing(drive->queue)) {
1045                 /*
1046                  * small race where queue could get replugged during
1047                  * the 3-request flush cycle, just yank the plug since
1048                  * we want it to finish asap
1049                  */
1050                 blk_remove_plug(drive->queue);
1051                 return drive;
1052         }
1053
1054         do {
1055                 if ((!drive->sleeping || time_after_eq(jiffies, drive->sleep))
1056                     && !elv_queue_empty(drive->queue)) {
1057                         if (!best
1058                          || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
1059                          || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
1060                         {
1061                                 if (!blk_queue_plugged(drive->queue))
1062                                         best = drive;
1063                         }
1064                 }
1065         } while ((drive = drive->next) != hwgroup->drive);
1066         if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
1067                 long t = (signed long)(WAKEUP(best) - jiffies);
1068                 if (t >= WAIT_MIN_SLEEP) {
1069                 /*
1070                  * We *may* have some time to spare, but first let's see if
1071                  * someone can potentially benefit from our nice mood today..
1072                  */
1073                         drive = best->next;
1074                         do {
1075                                 if (!drive->sleeping
1076                                  && time_before(jiffies - best->service_time, WAKEUP(drive))
1077                                  && time_before(WAKEUP(drive), jiffies + t))
1078                                 {
1079                                         ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
1080                                         goto repeat;
1081                                 }
1082                         } while ((drive = drive->next) != best);
1083                 }
1084         }
1085         return best;
1086 }
1087
1088 /*
1089  * Issue a new request to a drive from hwgroup
1090  * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
1091  *
1092  * A hwgroup is a serialized group of IDE interfaces.  Usually there is
1093  * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
1094  * may have both interfaces in a single hwgroup to "serialize" access.
1095  * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
1096  * together into one hwgroup for serialized access.
1097  *
1098  * Note also that several hwgroups can end up sharing a single IRQ,
1099  * possibly along with many other devices.  This is especially common in
1100  * PCI-based systems with off-board IDE controller cards.
1101  *
1102  * The IDE driver uses the single global ide_lock spinlock to protect
1103  * access to the request queues, and to protect the hwgroup->busy flag.
1104  *
1105  * The first thread into the driver for a particular hwgroup sets the
1106  * hwgroup->busy flag to indicate that this hwgroup is now active,
1107  * and then initiates processing of the top request from the request queue.
1108  *
1109  * Other threads attempting entry notice the busy setting, and will simply
1110  * queue their new requests and exit immediately.  Note that hwgroup->busy
1111  * remains set even when the driver is merely awaiting the next interrupt.
1112  * Thus, the meaning is "this hwgroup is busy processing a request".
1113  *
1114  * When processing of a request completes, the completing thread or IRQ-handler
1115  * will start the next request from the queue.  If no more work remains,
1116  * the driver will clear the hwgroup->busy flag and exit.
1117  *
1118  * The ide_lock (spinlock) is used to protect all access to the
1119  * hwgroup->busy flag, but is otherwise not needed for most processing in
1120  * the driver.  This makes the driver much more friendlier to shared IRQs
1121  * than previous designs, while remaining 100% (?) SMP safe and capable.
1122  */
1123 static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
1124 {
1125         ide_drive_t     *drive;
1126         ide_hwif_t      *hwif;
1127         struct request  *rq;
1128         ide_startstop_t startstop;
1129         int             loops = 0;
1130
1131         /* for atari only: POSSIBLY BROKEN HERE(?) */
1132         ide_get_lock(ide_intr, hwgroup);
1133
1134         /* caller must own ide_lock */
1135         BUG_ON(!irqs_disabled());
1136
1137         while (!hwgroup->busy) {
1138                 hwgroup->busy = 1;
1139                 drive = choose_drive(hwgroup);
1140                 if (drive == NULL) {
1141                         int sleeping = 0;
1142                         unsigned long sleep = 0; /* shut up, gcc */
1143                         hwgroup->rq = NULL;
1144                         drive = hwgroup->drive;
1145                         do {
1146                                 if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
1147                                         sleeping = 1;
1148                                         sleep = drive->sleep;
1149                                 }
1150                         } while ((drive = drive->next) != hwgroup->drive);
1151                         if (sleeping) {
1152                 /*
1153                  * Take a short snooze, and then wake up this hwgroup again.
1154                  * This gives other hwgroups on the same a chance to
1155                  * play fairly with us, just in case there are big differences
1156                  * in relative throughputs.. don't want to hog the cpu too much.
1157                  */
1158                                 if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
1159                                         sleep = jiffies + WAIT_MIN_SLEEP;
1160 #if 1
1161                                 if (timer_pending(&hwgroup->timer))
1162                                         printk(KERN_CRIT "ide_set_handler: timer already active\n");
1163 #endif
1164                                 /* so that ide_timer_expiry knows what to do */
1165                                 hwgroup->sleeping = 1;
1166                                 mod_timer(&hwgroup->timer, sleep);
1167                                 /* we purposely leave hwgroup->busy==1
1168                                  * while sleeping */
1169                         } else {
1170                                 /* Ugly, but how can we sleep for the lock
1171                                  * otherwise? perhaps from tq_disk?
1172                                  */
1173
1174                                 /* for atari only */
1175                                 ide_release_lock();
1176                                 hwgroup->busy = 0;
1177                         }
1178
1179                         /* no more work for this hwgroup (for now) */
1180                         return;
1181                 }
1182         again:
1183                 hwif = HWIF(drive);
1184                 if (hwgroup->hwif->sharing_irq &&
1185                     hwif != hwgroup->hwif &&
1186                     hwif->io_ports[IDE_CONTROL_OFFSET]) {
1187                         /* set nIEN for previous hwif */
1188                         SELECT_INTERRUPT(drive);
1189                 }
1190                 hwgroup->hwif = hwif;
1191                 hwgroup->drive = drive;
1192                 drive->sleeping = 0;
1193                 drive->service_start = jiffies;
1194
1195                 if (blk_queue_plugged(drive->queue)) {
1196                         printk(KERN_ERR "ide: huh? queue was plugged!\n");
1197                         break;
1198                 }
1199
1200                 /*
1201                  * we know that the queue isn't empty, but this can happen
1202                  * if the q->prep_rq_fn() decides to kill a request
1203                  */
1204                 rq = elv_next_request(drive->queue);
1205                 if (!rq) {
1206                         hwgroup->busy = 0;
1207                         break;
1208                 }
1209
1210                 /*
1211                  * Sanity: don't accept a request that isn't a PM request
1212                  * if we are currently power managed. This is very important as
1213                  * blk_stop_queue() doesn't prevent the elv_next_request()
1214                  * above to return us whatever is in the queue. Since we call
1215                  * ide_do_request() ourselves, we end up taking requests while
1216                  * the queue is blocked...
1217                  *
1218                  * We let requests forced at head of queue with ide-preempt
1219                  * though. I hope that doesn't happen too much, hopefully not
1220                  * unless the subdriver triggers such a thing in its own PM
1221                  * state machine.
1222                  *
1223                  * We count how many times we loop here to make sure we service
1224                  * all drives in the hwgroup without looping for ever
1225                  */
1226                 if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
1227                         drive = drive->next ? drive->next : hwgroup->drive;
1228                         if (loops++ < 4 && !blk_queue_plugged(drive->queue))
1229                                 goto again;
1230                         /* We clear busy, there should be no pending ATA command at this point. */
1231                         hwgroup->busy = 0;
1232                         break;
1233                 }
1234
1235                 hwgroup->rq = rq;
1236
1237                 /*
1238                  * Some systems have trouble with IDE IRQs arriving while
1239                  * the driver is still setting things up.  So, here we disable
1240                  * the IRQ used by this interface while the request is being started.
1241                  * This may look bad at first, but pretty much the same thing
1242                  * happens anyway when any interrupt comes in, IDE or otherwise
1243                  *  -- the kernel masks the IRQ while it is being handled.
1244                  */
1245                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1246                         disable_irq_nosync(hwif->irq);
1247                 spin_unlock(&ide_lock);
1248                 local_irq_enable();
1249                         /* allow other IRQs while we start this request */
1250                 startstop = start_request(drive, rq);
1251                 spin_lock_irq(&ide_lock);
1252                 if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
1253                         enable_irq(hwif->irq);
1254                 if (startstop == ide_stopped)
1255                         hwgroup->busy = 0;
1256         }
1257 }
1258
1259 /*
1260  * Passes the stuff to ide_do_request
1261  */
1262 void do_ide_request(request_queue_t *q)
1263 {
1264         ide_drive_t *drive = q->queuedata;
1265
1266         ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
1267 }
1268
1269 /*
1270  * un-busy the hwgroup etc, and clear any pending DMA status. we want to
1271  * retry the current request in pio mode instead of risking tossing it
1272  * all away
1273  */
1274 static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
1275 {
1276         ide_hwif_t *hwif = HWIF(drive);
1277         struct request *rq;
1278         ide_startstop_t ret = ide_stopped;
1279
1280         /*
1281          * end current dma transaction
1282          */
1283
1284         if (error < 0) {
1285                 printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
1286                 (void)HWIF(drive)->ide_dma_end(drive);
1287                 ret = ide_error(drive, "dma timeout error",
1288                                                 hwif->INB(IDE_STATUS_REG));
1289         } else {
1290                 printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
1291                 (void) hwif->ide_dma_timeout(drive);
1292         }
1293
1294         /*
1295          * disable dma for now, but remember that we did so because of
1296          * a timeout -- we'll reenable after we finish this next request
1297          * (or rather the first chunk of it) in pio.
1298          */
1299         drive->retry_pio++;
1300         drive->state = DMA_PIO_RETRY;
1301         (void) hwif->ide_dma_off_quietly(drive);
1302
1303         /*
1304          * un-busy drive etc (hwgroup->busy is cleared on return) and
1305          * make sure request is sane
1306          */
1307         rq = HWGROUP(drive)->rq;
1308         HWGROUP(drive)->rq = NULL;
1309
1310         rq->errors = 0;
1311
1312         if (!rq->bio)
1313                 goto out;
1314
1315         rq->sector = rq->bio->bi_sector;
1316         rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
1317         rq->hard_cur_sectors = rq->current_nr_sectors;
1318         rq->buffer = bio_data(rq->bio);
1319 out:
1320         return ret;
1321 }
1322
1323 /**
1324  *      ide_timer_expiry        -       handle lack of an IDE interrupt
1325  *      @data: timer callback magic (hwgroup)
1326  *
1327  *      An IDE command has timed out before the expected drive return
1328  *      occurred. At this point we attempt to clean up the current
1329  *      mess. If the current handler includes an expiry handler then
1330  *      we invoke the expiry handler, and providing it is happy the
1331  *      work is done. If that fails we apply generic recovery rules
1332  *      invoking the handler and checking the drive DMA status. We
1333  *      have an excessively incestuous relationship with the DMA
1334  *      logic that wants cleaning up.
1335  */
1336
1337 void ide_timer_expiry (unsigned long data)
1338 {
1339         ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
1340         ide_handler_t   *handler;
1341         ide_expiry_t    *expiry;
1342         unsigned long   flags;
1343         unsigned long   wait = -1;
1344
1345         spin_lock_irqsave(&ide_lock, flags);
1346
1347         if ((handler = hwgroup->handler) == NULL) {
1348                 /*
1349                  * Either a marginal timeout occurred
1350                  * (got the interrupt just as timer expired),
1351                  * or we were "sleeping" to give other devices a chance.
1352                  * Either way, we don't really want to complain about anything.
1353                  */
1354                 if (hwgroup->sleeping) {
1355                         hwgroup->sleeping = 0;
1356                         hwgroup->busy = 0;
1357                 }
1358         } else {
1359                 ide_drive_t *drive = hwgroup->drive;
1360                 if (!drive) {
1361                         printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
1362                         hwgroup->handler = NULL;
1363                 } else {
1364                         ide_hwif_t *hwif;
1365                         ide_startstop_t startstop = ide_stopped;
1366                         if (!hwgroup->busy) {
1367                                 hwgroup->busy = 1;      /* paranoia */
1368                                 printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
1369                         }
1370                         if ((expiry = hwgroup->expiry) != NULL) {
1371                                 /* continue */
1372                                 if ((wait = expiry(drive)) > 0) {
1373                                         /* reset timer */
1374                                         hwgroup->timer.expires  = jiffies + wait;
1375                                         add_timer(&hwgroup->timer);
1376                                         spin_unlock_irqrestore(&ide_lock, flags);
1377                                         return;
1378                                 }
1379                         }
1380                         hwgroup->handler = NULL;
1381                         /*
1382                          * We need to simulate a real interrupt when invoking
1383                          * the handler() function, which means we need to
1384                          * globally mask the specific IRQ:
1385                          */
1386                         spin_unlock(&ide_lock);
1387                         hwif  = HWIF(drive);
1388 #if DISABLE_IRQ_NOSYNC
1389                         disable_irq_nosync(hwif->irq);
1390 #else
1391                         /* disable_irq_nosync ?? */
1392                         disable_irq(hwif->irq);
1393 #endif /* DISABLE_IRQ_NOSYNC */
1394                         /* local CPU only,
1395                          * as if we were handling an interrupt */
1396                         local_irq_disable();
1397                         if (hwgroup->polling) {
1398                                 startstop = handler(drive);
1399                         } else if (drive_is_ready(drive)) {
1400                                 if (drive->waiting_for_dma)
1401                                         (void) hwgroup->hwif->ide_dma_lostirq(drive);
1402                                 (void)ide_ack_intr(hwif);
1403                                 printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
1404                                 startstop = handler(drive);
1405                         } else {
1406                                 if (drive->waiting_for_dma) {
1407                                         startstop = ide_dma_timeout_retry(drive, wait);
1408                                 } else
1409                                         startstop =
1410                                         ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
1411                         }
1412                         drive->service_time = jiffies - drive->service_start;
1413                         spin_lock_irq(&ide_lock);
1414                         enable_irq(hwif->irq);
1415                         if (startstop == ide_stopped)
1416                                 hwgroup->busy = 0;
1417                 }
1418         }
1419         ide_do_request(hwgroup, IDE_NO_IRQ);
1420         spin_unlock_irqrestore(&ide_lock, flags);
1421 }
1422
1423 /**
1424  *      unexpected_intr         -       handle an unexpected IDE interrupt
1425  *      @irq: interrupt line
1426  *      @hwgroup: hwgroup being processed
1427  *
1428  *      There's nothing really useful we can do with an unexpected interrupt,
1429  *      other than reading the status register (to clear it), and logging it.
1430  *      There should be no way that an irq can happen before we're ready for it,
1431  *      so we needn't worry much about losing an "important" interrupt here.
1432  *
1433  *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
1434  *      the drive enters "idle", "standby", or "sleep" mode, so if the status
1435  *      looks "good", we just ignore the interrupt completely.
1436  *
1437  *      This routine assumes __cli() is in effect when called.
1438  *
1439  *      If an unexpected interrupt happens on irq15 while we are handling irq14
1440  *      and if the two interfaces are "serialized" (CMD640), then it looks like
1441  *      we could screw up by interfering with a new request being set up for
1442  *      irq15.
1443  *
1444  *      In reality, this is a non-issue.  The new command is not sent unless
1445  *      the drive is ready to accept one, in which case we know the drive is
1446  *      not trying to interrupt us.  And ide_set_handler() is always invoked
1447  *      before completing the issuance of any new drive command, so we will not
1448  *      be accidentally invoked as a result of any valid command completion
1449  *      interrupt.
1450  *
1451  *      Note that we must walk the entire hwgroup here. We know which hwif
1452  *      is doing the current command, but we don't know which hwif burped
1453  *      mysteriously.
1454  */
1455
1456 static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
1457 {
1458         u8 stat;
1459         ide_hwif_t *hwif = hwgroup->hwif;
1460
1461         /*
1462          * handle the unexpected interrupt
1463          */
1464         do {
1465                 if (hwif->irq == irq) {
1466                         stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1467                         if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
1468                                 /* Try to not flood the console with msgs */
1469                                 static unsigned long last_msgtime, count;
1470                                 ++count;
1471                                 if (time_after(jiffies, last_msgtime + HZ)) {
1472                                         last_msgtime = jiffies;
1473                                         printk(KERN_ERR "%s%s: unexpected interrupt, "
1474                                                 "status=0x%02x, count=%ld\n",
1475                                                 hwif->name,
1476                                                 (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
1477                                 }
1478                         }
1479                 }
1480         } while ((hwif = hwif->next) != hwgroup->hwif);
1481 }
1482
1483 /**
1484  *      ide_intr        -       default IDE interrupt handler
1485  *      @irq: interrupt number
1486  *      @dev_id: hwif group
1487  *      @regs: unused weirdness from the kernel irq layer
1488  *
1489  *      This is the default IRQ handler for the IDE layer. You should
1490  *      not need to override it. If you do be aware it is subtle in
1491  *      places
1492  *
1493  *      hwgroup->hwif is the interface in the group currently performing
1494  *      a command. hwgroup->drive is the drive and hwgroup->handler is
1495  *      the IRQ handler to call. As we issue a command the handlers
1496  *      step through multiple states, reassigning the handler to the
1497  *      next step in the process. Unlike a smart SCSI controller IDE
1498  *      expects the main processor to sequence the various transfer
1499  *      stages. We also manage a poll timer to catch up with most
1500  *      timeout situations. There are still a few where the handlers
1501  *      don't ever decide to give up.
1502  *
1503  *      The handler eventually returns ide_stopped to indicate the
1504  *      request completed. At this point we issue the next request
1505  *      on the hwgroup and the process begins again.
1506  */
1507
1508 irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
1509 {
1510         unsigned long flags;
1511         ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
1512         ide_hwif_t *hwif;
1513         ide_drive_t *drive;
1514         ide_handler_t *handler;
1515         ide_startstop_t startstop;
1516
1517         spin_lock_irqsave(&ide_lock, flags);
1518         hwif = hwgroup->hwif;
1519
1520         if (!ide_ack_intr(hwif)) {
1521                 spin_unlock_irqrestore(&ide_lock, flags);
1522                 return IRQ_NONE;
1523         }
1524
1525         if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
1526                 /*
1527                  * Not expecting an interrupt from this drive.
1528                  * That means this could be:
1529                  *      (1) an interrupt from another PCI device
1530                  *      sharing the same PCI INT# as us.
1531                  * or   (2) a drive just entered sleep or standby mode,
1532                  *      and is interrupting to let us know.
1533                  * or   (3) a spurious interrupt of unknown origin.
1534                  *
1535                  * For PCI, we cannot tell the difference,
1536                  * so in that case we just ignore it and hope it goes away.
1537                  *
1538                  * FIXME: unexpected_intr should be hwif-> then we can
1539                  * remove all the ifdef PCI crap
1540                  */
1541 #ifdef CONFIG_BLK_DEV_IDEPCI
1542                 if (hwif->pci_dev && !hwif->pci_dev->vendor)
1543 #endif  /* CONFIG_BLK_DEV_IDEPCI */
1544                 {
1545                         /*
1546                          * Probably not a shared PCI interrupt,
1547                          * so we can safely try to do something about it:
1548                          */
1549                         unexpected_intr(irq, hwgroup);
1550 #ifdef CONFIG_BLK_DEV_IDEPCI
1551                 } else {
1552                         /*
1553                          * Whack the status register, just in case
1554                          * we have a leftover pending IRQ.
1555                          */
1556                         (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
1557 #endif /* CONFIG_BLK_DEV_IDEPCI */
1558                 }
1559                 spin_unlock_irqrestore(&ide_lock, flags);
1560                 return IRQ_NONE;
1561         }
1562         drive = hwgroup->drive;
1563         if (!drive) {
1564                 /*
1565                  * This should NEVER happen, and there isn't much
1566                  * we could do about it here.
1567                  *
1568                  * [Note - this can occur if the drive is hot unplugged]
1569                  */
1570                 spin_unlock_irqrestore(&ide_lock, flags);
1571                 return IRQ_HANDLED;
1572         }
1573         if (!drive_is_ready(drive)) {
1574                 /*
1575                  * This happens regularly when we share a PCI IRQ with
1576                  * another device.  Unfortunately, it can also happen
1577                  * with some buggy drives that trigger the IRQ before
1578                  * their status register is up to date.  Hopefully we have
1579                  * enough advance overhead that the latter isn't a problem.
1580                  */
1581                 spin_unlock_irqrestore(&ide_lock, flags);
1582                 return IRQ_NONE;
1583         }
1584         if (!hwgroup->busy) {
1585                 hwgroup->busy = 1;      /* paranoia */
1586                 printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
1587         }
1588         hwgroup->handler = NULL;
1589         del_timer(&hwgroup->timer);
1590         spin_unlock(&ide_lock);
1591
1592         if (drive->unmask)
1593                 local_irq_enable();
1594         /* service this interrupt, may set handler for next interrupt */
1595         startstop = handler(drive);
1596         spin_lock_irq(&ide_lock);
1597
1598         /*
1599          * Note that handler() may have set things up for another
1600          * interrupt to occur soon, but it cannot happen until
1601          * we exit from this routine, because it will be the
1602          * same irq as is currently being serviced here, and Linux
1603          * won't allow another of the same (on any CPU) until we return.
1604          */
1605         drive->service_time = jiffies - drive->service_start;
1606         if (startstop == ide_stopped) {
1607                 if (hwgroup->handler == NULL) { /* paranoia */
1608                         hwgroup->busy = 0;
1609                         ide_do_request(hwgroup, hwif->irq);
1610                 } else {
1611                         printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
1612                                 "on exit\n", drive->name);
1613                 }
1614         }
1615         spin_unlock_irqrestore(&ide_lock, flags);
1616         return IRQ_HANDLED;
1617 }
1618
1619 /**
1620  *      ide_init_drive_cmd      -       initialize a drive command request
1621  *      @rq: request object
1622  *
1623  *      Initialize a request before we fill it in and send it down to
1624  *      ide_do_drive_cmd. Commands must be set up by this function. Right
1625  *      now it doesn't do a lot, but if that changes abusers will have a
1626  *      nasty suprise.
1627  */
1628
1629 void ide_init_drive_cmd (struct request *rq)
1630 {
1631         memset(rq, 0, sizeof(*rq));
1632         rq->flags = REQ_DRIVE_CMD;
1633         rq->ref_count = 1;
1634 }
1635
1636 EXPORT_SYMBOL(ide_init_drive_cmd);
1637
1638 /**
1639  *      ide_do_drive_cmd        -       issue IDE special command
1640  *      @drive: device to issue command
1641  *      @rq: request to issue
1642  *      @action: action for processing
1643  *
1644  *      This function issues a special IDE device request
1645  *      onto the request queue.
1646  *
1647  *      If action is ide_wait, then the rq is queued at the end of the
1648  *      request queue, and the function sleeps until it has been processed.
1649  *      This is for use when invoked from an ioctl handler.
1650  *
1651  *      If action is ide_preempt, then the rq is queued at the head of
1652  *      the request queue, displacing the currently-being-processed
1653  *      request and this function returns immediately without waiting
1654  *      for the new rq to be completed.  This is VERY DANGEROUS, and is
1655  *      intended for careful use by the ATAPI tape/cdrom driver code.
1656  *
1657  *      If action is ide_end, then the rq is queued at the end of the
1658  *      request queue, and the function returns immediately without waiting
1659  *      for the new rq to be completed. This is again intended for careful
1660  *      use by the ATAPI tape/cdrom driver code.
1661  */
1662
1663 int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
1664 {
1665         unsigned long flags;
1666         ide_hwgroup_t *hwgroup = HWGROUP(drive);
1667         DECLARE_COMPLETION(wait);
1668         int where = ELEVATOR_INSERT_BACK, err;
1669         int must_wait = (action == ide_wait || action == ide_head_wait);
1670
1671         rq->errors = 0;
1672         rq->rq_status = RQ_ACTIVE;
1673
1674         /*
1675          * we need to hold an extra reference to request for safe inspection
1676          * after completion
1677          */
1678         if (must_wait) {
1679                 rq->ref_count++;
1680                 rq->waiting = &wait;
1681                 rq->end_io = blk_end_sync_rq;
1682         }
1683
1684         spin_lock_irqsave(&ide_lock, flags);
1685         if (action == ide_preempt)
1686                 hwgroup->rq = NULL;
1687         if (action == ide_preempt || action == ide_head_wait) {
1688                 where = ELEVATOR_INSERT_FRONT;
1689                 rq->flags |= REQ_PREEMPT;
1690         }
1691         __elv_add_request(drive->queue, rq, where, 0);
1692         ide_do_request(hwgroup, IDE_NO_IRQ);
1693         spin_unlock_irqrestore(&ide_lock, flags);
1694
1695         err = 0;
1696         if (must_wait) {
1697                 wait_for_completion(&wait);
1698                 rq->waiting = NULL;
1699                 if (rq->errors)
1700                         err = -EIO;
1701
1702                 blk_put_request(rq);
1703         }
1704
1705         return err;
1706 }
1707
1708 EXPORT_SYMBOL(ide_do_drive_cmd);