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