upgrade to linux 2.6.10-1.12_FC2

[linux-2.6.git] / drivers / ide / ide-io.c

/*
 *      IDE I/O functions
 *
 *      Basic PIO and command management functionality.
 *
 * This code was split off from ide.c. See ide.c for history and original
 * copyrights.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the
 * Free Software Foundation; either version 2, or (at your option) any
 * later version.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * For the avoidance of doubt the "preferred form" of this code is one which
 * is in an open non patent encumbered format. Where cryptographic key signing
 * forms part of the process of creating an executable the information
 * including keys needed to generate an equivalently functional executable
 * are deemed to be part of the source code.
 */
 
 
#include <linux/config.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/string.h>
#include <linux/kernel.h>
#include <linux/timer.h>
#include <linux/mm.h>
#include <linux/interrupt.h>
#include <linux/major.h>
#include <linux/errno.h>
#include <linux/genhd.h>
#include <linux/blkpg.h>
#include <linux/slab.h>
#include <linux/init.h>
#include <linux/pci.h>
#include <linux/delay.h>
#include <linux/ide.h>
#include <linux/completion.h>
#include <linux/reboot.h>
#include <linux/cdrom.h>
#include <linux/seq_file.h>
#include <linux/device.h>
#include <linux/kmod.h>
#include <linux/scatterlist.h>

#include <asm/byteorder.h>
#include <asm/irq.h>
#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/bitops.h>

static void ide_fill_flush_cmd(ide_drive_t *drive, struct request *rq)
{
        char *buf = rq->cmd;

        /*
         * reuse cdb space for ata command
         */
        memset(buf, 0, sizeof(rq->cmd));

        rq->flags |= REQ_DRIVE_TASK | REQ_STARTED;
        rq->buffer = buf;
        rq->buffer[0] = WIN_FLUSH_CACHE;

        if (ide_id_has_flush_cache_ext(drive->id) &&
            (drive->capacity64 >= (1UL << 28)))
                rq->buffer[0] = WIN_FLUSH_CACHE_EXT;
}

/*
 * preempt pending requests, and store this cache flush for immediate
 * execution
 */
static struct request *ide_queue_flush_cmd(ide_drive_t *drive,
                                           struct request *rq, int post)
{
        struct request *flush_rq = &HWGROUP(drive)->wrq;

        /*
         * write cache disabled, clear the barrier bit and treat it like
         * an ordinary write
         */
        if (!drive->wcache) {
                rq->flags |= REQ_BAR_PREFLUSH;
                return rq;
        }

        ide_init_drive_cmd(flush_rq);
        ide_fill_flush_cmd(drive, flush_rq);

        flush_rq->special = rq;
        flush_rq->nr_sectors = rq->nr_sectors;

        if (!post) {
                drive->doing_barrier = 1;
                flush_rq->flags |= REQ_BAR_PREFLUSH;
                blkdev_dequeue_request(rq);
        } else
                flush_rq->flags |= REQ_BAR_POSTFLUSH;

        __elv_add_request(drive->queue, flush_rq, ELEVATOR_INSERT_FRONT, 0);
        HWGROUP(drive)->rq = NULL;
        return flush_rq;
}

static int __ide_end_request(ide_drive_t *drive, struct request *rq,
                             int uptodate, int nr_sectors)
{
        int ret = 1;

        BUG_ON(!(rq->flags & REQ_STARTED));

        /*
         * if failfast is set on a request, override number of sectors and
         * complete the whole request right now
         */
        if (blk_noretry_request(rq) && end_io_error(uptodate))
                nr_sectors = rq->hard_nr_sectors;

        if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
                rq->errors = -EIO;

        /*
         * decide whether to reenable DMA -- 3 is a random magic for now,
         * if we DMA timeout more than 3 times, just stay in PIO
         */
        if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
                drive->state = 0;
                HWGROUP(drive)->hwif->ide_dma_on(drive);
        }

        if (!end_that_request_first(rq, uptodate, nr_sectors)) {
                add_disk_randomness(rq->rq_disk);

                if (blk_rq_tagged(rq))
                        blk_queue_end_tag(drive->queue, rq);

                blkdev_dequeue_request(rq);
                HWGROUP(drive)->rq = NULL;
                end_that_request_last(rq);
                ret = 0;
        }
        return ret;
}

/**
 *      ide_end_request         -       complete an IDE I/O
 *      @drive: IDE device for the I/O
 *      @uptodate:
 *      @nr_sectors: number of sectors completed
 *
 *      This is our end_request wrapper function. We complete the I/O
 *      update random number input and dequeue the request, which if
 *      it was tagged may be out of order.
 */

int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
{
        struct request *rq;
        unsigned long flags;
        int ret = 1;

        spin_lock_irqsave(&ide_lock, flags);
        rq = HWGROUP(drive)->rq;

        if (!nr_sectors)
                nr_sectors = rq->hard_cur_sectors;

        if (!blk_barrier_rq(rq) || !drive->wcache)
                ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
        else {
                struct request *flush_rq = &HWGROUP(drive)->wrq;

                flush_rq->nr_sectors -= nr_sectors;
                if (!flush_rq->nr_sectors) {
                        ide_queue_flush_cmd(drive, rq, 1);
                        ret = 0;
                }
        }

        spin_unlock_irqrestore(&ide_lock, flags);
        return ret;
}
EXPORT_SYMBOL(ide_end_request);

/**
 *      ide_end_dequeued_request        -       complete an IDE I/O
 *      @drive: IDE device for the I/O
 *      @uptodate:
 *      @nr_sectors: number of sectors completed
 *
 *      Complete an I/O that is no longer on the request queue. This 
 *      typically occurs when we pull the request and issue a REQUEST_SENSE.
 *      We must still finish the old request but we must not tamper with the
 *      queue in the meantime.
 *
 *      NOTE: This path does not handle barrier, but barrier is not supported
 *      on ide-cd anyway.
 */

int ide_end_dequeued_request(ide_drive_t *drive, struct request *rq,
                             int uptodate, int nr_sectors)
{
        unsigned long flags;
        int ret = 1;

        spin_lock_irqsave(&ide_lock, flags);

        BUG_ON(!(rq->flags & REQ_STARTED));

        /*
         * if failfast is set on a request, override number of sectors and
         * complete the whole request right now
         */
        if (blk_noretry_request(rq) && end_io_error(uptodate))
                nr_sectors = rq->hard_nr_sectors;

        if (!blk_fs_request(rq) && end_io_error(uptodate) && !rq->errors)
                rq->errors = -EIO;

        /*
         * decide whether to reenable DMA -- 3 is a random magic for now,
         * if we DMA timeout more than 3 times, just stay in PIO
         */
        if (drive->state == DMA_PIO_RETRY && drive->retry_pio <= 3) {
                drive->state = 0;
                HWGROUP(drive)->hwif->ide_dma_on(drive);
        }

        if (!end_that_request_first(rq, uptodate, nr_sectors)) {
                add_disk_randomness(rq->rq_disk);
                if (blk_rq_tagged(rq))
                        blk_queue_end_tag(drive->queue, rq);
                end_that_request_last(rq);
                ret = 0;
        }
        spin_unlock_irqrestore(&ide_lock, flags);
        return ret;
}
EXPORT_SYMBOL_GPL(ide_end_dequeued_request);

/**
 *      ide_complete_pm_request - end the current Power Management request
 *      @drive: target drive
 *      @rq: request
 *
 *      This function cleans up the current PM request and stops the queue
 *      if necessary.
 */
static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
{
        unsigned long flags;

#ifdef DEBUG_PM
        printk("%s: completing PM request, %s\n", drive->name,
               blk_pm_suspend_request(rq) ? "suspend" : "resume");
#endif
        spin_lock_irqsave(&ide_lock, flags);
        if (blk_pm_suspend_request(rq)) {
                blk_stop_queue(drive->queue);
        } else {
                drive->blocked = 0;
                blk_start_queue(drive->queue);
        }
        blkdev_dequeue_request(rq);
        HWGROUP(drive)->rq = NULL;
        end_that_request_last(rq);
        spin_unlock_irqrestore(&ide_lock, flags);
}

/*
 * FIXME: probably move this somewhere else, name is bad too :)
 */
u64 ide_get_error_location(ide_drive_t *drive, char *args)
{
        u32 high, low;
        u8 hcyl, lcyl, sect;
        u64 sector;

        high = 0;
        hcyl = args[5];
        lcyl = args[4];
        sect = args[3];

        if (ide_id_has_flush_cache_ext(drive->id)) {
                low = (hcyl << 16) | (lcyl << 8) | sect;
                HWIF(drive)->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
                high = ide_read_24(drive);
        } else {
                u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
                if (cur & 0x40)
                        low = (hcyl << 16) | (lcyl << 8) | sect;
                else {
                        low = hcyl * drive->head * drive->sect;
                        low += lcyl * drive->sect;
                        low += sect - 1;
                }
        }

        sector = ((u64) high << 24) | low;
        return sector;
}
EXPORT_SYMBOL(ide_get_error_location);

static void ide_complete_barrier(ide_drive_t *drive, struct request *rq,
                                 int error)
{
        struct request *real_rq = rq->special;
        int good_sectors, bad_sectors;
        sector_t sector;

        if (!error) {
                if (blk_barrier_postflush(rq)) {
                        /*
                         * this completes the barrier write
                         */
                        __ide_end_request(drive, real_rq, 1, real_rq->hard_nr_sectors);
                        drive->doing_barrier = 0;
                } else {
                        /*
                         * just indicate that we did the pre flush
                         */
                        real_rq->flags |= REQ_BAR_PREFLUSH;
                        elv_requeue_request(drive->queue, real_rq);
                }
                /*
                 * all is fine, return
                 */
                return;
        }

        /*
         * we need to end real_rq, but it's not on the queue currently.
         * put it back on the queue, so we don't have to special case
         * anything else for completing it
         */
        if (!blk_barrier_postflush(rq))
                elv_requeue_request(drive->queue, real_rq);

        /*
         * drive aborted flush command, assume FLUSH_CACHE_* doesn't
         * work and disable barrier support
         */
        if (error & ABRT_ERR) {
                printk(KERN_ERR "%s: barrier support doesn't work\n", drive->name);
                __ide_end_request(drive, real_rq, -EOPNOTSUPP, real_rq->hard_nr_sectors);
                blk_queue_ordered(drive->queue, 0);
                blk_queue_issue_flush_fn(drive->queue, NULL);
        } else {
                /*
                 * find out what part of the request failed
                 */
                good_sectors = 0;
                if (blk_barrier_postflush(rq)) {
                        sector = ide_get_error_location(drive, rq->buffer);

                        if ((sector >= real_rq->hard_sector) &&
                            (sector < real_rq->hard_sector + real_rq->hard_nr_sectors))
                                good_sectors = sector - real_rq->hard_sector;
                } else
                        sector = real_rq->hard_sector;

                bad_sectors = real_rq->hard_nr_sectors - good_sectors;
                if (good_sectors)
                        __ide_end_request(drive, real_rq, 1, good_sectors);
                if (bad_sectors)
                        __ide_end_request(drive, real_rq, 0, bad_sectors);

                printk(KERN_ERR "%s: failed barrier write: "
                                "sector=%Lx(good=%d/bad=%d)\n",
                                drive->name, (unsigned long long)sector,
                                good_sectors, bad_sectors);
        }

        drive->doing_barrier = 0;
}

/**
 *      ide_end_drive_cmd       -       end an explicit drive command
 *      @drive: command 
 *      @stat: status bits
 *      @err: error bits
 *
 *      Clean up after success/failure of an explicit drive command.
 *      These get thrown onto the queue so they are synchronized with
 *      real I/O operations on the drive.
 *
 *      In LBA48 mode we have to read the register set twice to get
 *      all the extra information out.
 */
 
void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
{
        ide_hwif_t *hwif = HWIF(drive);
        unsigned long flags;
        struct request *rq;

        spin_lock_irqsave(&ide_lock, flags);
        rq = HWGROUP(drive)->rq;
        spin_unlock_irqrestore(&ide_lock, flags);

        if (rq->flags & REQ_DRIVE_CMD) {
                u8 *args = (u8 *) rq->buffer;
                if (rq->errors == 0)
                        rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);

                if (args) {
                        args[0] = stat;
                        args[1] = err;
                        args[2] = hwif->INB(IDE_NSECTOR_REG);
                }
        } else if (rq->flags & REQ_DRIVE_TASK) {
                u8 *args = (u8 *) rq->buffer;
                if (rq->errors == 0)
                        rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);

                if (args) {
                        args[0] = stat;
                        args[1] = err;
                        args[2] = hwif->INB(IDE_NSECTOR_REG);
                        args[3] = hwif->INB(IDE_SECTOR_REG);
                        args[4] = hwif->INB(IDE_LCYL_REG);
                        args[5] = hwif->INB(IDE_HCYL_REG);
                        args[6] = hwif->INB(IDE_SELECT_REG);
                }
        } else if (rq->flags & REQ_DRIVE_TASKFILE) {
                ide_task_t *args = (ide_task_t *) rq->special;
                if (rq->errors == 0)
                        rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
                        
                if (args) {
                        if (args->tf_in_flags.b.data) {
                                u16 data                                = hwif->INW(IDE_DATA_REG);
                                args->tfRegister[IDE_DATA_OFFSET]       = (data) & 0xFF;
                                args->hobRegister[IDE_DATA_OFFSET]      = (data >> 8) & 0xFF;
                        }
                        args->tfRegister[IDE_ERROR_OFFSET]   = err;
                        /* be sure we're looking at the low order bits */
                        hwif->OUTB(drive->ctl & ~0x80, IDE_CONTROL_REG);
                        args->tfRegister[IDE_NSECTOR_OFFSET] = hwif->INB(IDE_NSECTOR_REG);
                        args->tfRegister[IDE_SECTOR_OFFSET]  = hwif->INB(IDE_SECTOR_REG);
                        args->tfRegister[IDE_LCYL_OFFSET]    = hwif->INB(IDE_LCYL_REG);
                        args->tfRegister[IDE_HCYL_OFFSET]    = hwif->INB(IDE_HCYL_REG);
                        args->tfRegister[IDE_SELECT_OFFSET]  = hwif->INB(IDE_SELECT_REG);
                        args->tfRegister[IDE_STATUS_OFFSET]  = stat;

                        if (drive->addressing == 1) {
                                hwif->OUTB(drive->ctl|0x80, IDE_CONTROL_REG);
                                args->hobRegister[IDE_FEATURE_OFFSET]   = hwif->INB(IDE_FEATURE_REG);
                                args->hobRegister[IDE_NSECTOR_OFFSET]   = hwif->INB(IDE_NSECTOR_REG);
                                args->hobRegister[IDE_SECTOR_OFFSET]    = hwif->INB(IDE_SECTOR_REG);
                                args->hobRegister[IDE_LCYL_OFFSET]      = hwif->INB(IDE_LCYL_REG);
                                args->hobRegister[IDE_HCYL_OFFSET]      = hwif->INB(IDE_HCYL_REG);
                        }
                }
        } else if (blk_pm_request(rq)) {
#ifdef DEBUG_PM
                printk("%s: complete_power_step(step: %d, stat: %x, err: %x)\n",
                        drive->name, rq->pm->pm_step, stat, err);
#endif
                DRIVER(drive)->complete_power_step(drive, rq, stat, err);
                if (rq->pm->pm_step == ide_pm_state_completed)
                        ide_complete_pm_request(drive, rq);
                return;
        }

        spin_lock_irqsave(&ide_lock, flags);
        blkdev_dequeue_request(rq);

        if (blk_barrier_preflush(rq) || blk_barrier_postflush(rq))
                ide_complete_barrier(drive, rq, err);

        HWGROUP(drive)->rq = NULL;
        end_that_request_last(rq);
        spin_unlock_irqrestore(&ide_lock, flags);
}

EXPORT_SYMBOL(ide_end_drive_cmd);

/**
 *      try_to_flush_leftover_data      -       flush junk
 *      @drive: drive to flush
 *
 *      try_to_flush_leftover_data() is invoked in response to a drive
 *      unexpectedly having its DRQ_STAT bit set.  As an alternative to
 *      resetting the drive, this routine tries to clear the condition
 *      by read a sector's worth of data from the drive.  Of course,
 *      this may not help if the drive is *waiting* for data from *us*.
 */
void try_to_flush_leftover_data (ide_drive_t *drive)
{
        int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;

        if (drive->media != ide_disk)
                return;
        while (i > 0) {
                u32 buffer[16];
                u32 wcount = (i > 16) ? 16 : i;

                i -= wcount;
                HWIF(drive)->ata_input_data(drive, buffer, wcount);
        }
}

EXPORT_SYMBOL(try_to_flush_leftover_data);

/*
 * FIXME Add an ATAPI error
 */

/**
 *      ide_error       -       handle an error on the IDE
 *      @drive: drive the error occurred on
 *      @msg: message to report
 *      @stat: status bits
 *
 *      ide_error() takes action based on the error returned by the drive.
 *      For normal I/O that may well include retries. We deal with
 *      both new-style (taskfile) and old style command handling here.
 *      In the case of taskfile command handling there is work left to
 *      do
 */
 
ide_startstop_t ide_error (ide_drive_t *drive, const char *msg, u8 stat)
{
        ide_hwif_t *hwif;
        struct request *rq;
        u8 err;

        err = ide_dump_status(drive, msg, stat);
        if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL)
                return ide_stopped;

        hwif = HWIF(drive);
        /* retry only "normal" I/O: */
        if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
                rq->errors = 1;
                ide_end_drive_cmd(drive, stat, err);
                return ide_stopped;
        }
        if (rq->flags & REQ_DRIVE_TASKFILE) {
                rq->errors = 1;
                ide_end_drive_cmd(drive, stat, err);
                return ide_stopped;
        }

        if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
                 /* other bits are useless when BUSY */
                rq->errors |= ERROR_RESET;
        } else {
                if (drive->media != ide_disk)
                        goto media_out;

                if (stat & ERR_STAT) {
                        /* err has different meaning on cdrom and tape */
                        if (err == ABRT_ERR) {
                                if (drive->select.b.lba &&
                                    (hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY))
                                        /* some newer drives don't
                                         * support WIN_SPECIFY
                                         */
                                        return ide_stopped;
                        } else if ((err & BAD_CRC) == BAD_CRC) {
                                drive->crc_count++;
                                /* UDMA crc error -- just retry the operation */
                        } else if (err & (BBD_ERR | ECC_ERR)) {
                                /* retries won't help these */
                                rq->errors = ERROR_MAX;
                        } else if (err & TRK0_ERR) {
                                /* help it find track zero */
                                rq->errors |= ERROR_RECAL;
                        }
                }
media_out:
                if ((stat & DRQ_STAT) && rq_data_dir(rq) != WRITE)
                        try_to_flush_leftover_data(drive);
        }
        if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT)) {
                /* force an abort */
                hwif->OUTB(WIN_IDLEIMMEDIATE,IDE_COMMAND_REG);
        }
        if (rq->errors >= ERROR_MAX) {
                DRIVER(drive)->end_request(drive, 0, 0);
        } else {
                if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
                        ++rq->errors;
                        return ide_do_reset(drive);
                }
                if ((rq->errors & ERROR_RECAL) == ERROR_RECAL)
                        drive->special.b.recalibrate = 1;
                ++rq->errors;
        }
        return ide_stopped;
}

/**
 *      ide_abort       -       abort pending IDE operatins
 *      @drive: drive the error occurred on
 *      @msg: message to report
 *
 *      ide_abort kills and cleans up when we are about to do a 
 *      host initiated reset on active commands. Longer term we
 *      want handlers to have sensible abort handling themselves
 *
 *      This differs fundamentally from ide_error because in 
 *      this case the command is doing just fine when we
 *      blow it away.
 *
 *      FIXME: need to fix locking corner cases
 */
 
ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
{
        ide_hwif_t *hwif;
        struct request *rq;
        unsigned long flags;

        spin_lock_irqsave(&ide_lock, flags);
        if (drive == NULL || (rq = HWGROUP(drive)->rq) == NULL) {
                spin_unlock_irqrestore(&ide_lock, flags);
                return ide_stopped;
        }
        hwif = HWIF(drive);
        /* retry only "normal" I/O: */
        if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK)) {
                rq->errors = 1;
                spin_unlock_irqrestore(&ide_lock, flags);
                ide_end_drive_cmd(drive, BUSY_STAT, 0);
                return ide_stopped;
        }
        if (rq->flags & REQ_DRIVE_TASKFILE) {
                rq->errors = 1;
                spin_unlock_irqrestore(&ide_lock, flags);
                ide_end_drive_cmd(drive, BUSY_STAT, 0);
                return ide_stopped;
        }

        rq->errors |= ERROR_RESET;
        spin_unlock_irqrestore(&ide_lock, flags);
        DRIVER(drive)->end_request(drive, 0, 0);
        return ide_stopped;
}

/**
 *      ide_cmd         -       issue a simple drive command
 *      @drive: drive the command is for
 *      @cmd: command byte
 *      @nsect: sector byte
 *      @handler: handler for the command completion
 *
 *      Issue a simple drive command with interrupts.
 *      The drive must be selected beforehand.
 */

static void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect,
                ide_handler_t *handler)
{
        ide_hwif_t *hwif = HWIF(drive);
        if (IDE_CONTROL_REG)
                hwif->OUTB(drive->ctl,IDE_CONTROL_REG); /* clear nIEN */
        SELECT_MASK(drive,0);
        hwif->OUTB(nsect,IDE_NSECTOR_REG);
        ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
}

/**
 *      drive_cmd_intr          -       drive command completion interrupt
 *      @drive: drive the completion interrupt occurred on
 *
 *      drive_cmd_intr() is invoked on completion of a special DRIVE_CMD.
 *      We do any necessary daya reading and then wait for the drive to
 *      go non busy. At that point we may read the error data and complete
 *      the request
 */
 
static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
{
        struct request *rq = HWGROUP(drive)->rq;
        ide_hwif_t *hwif = HWIF(drive);
        u8 *args = (u8 *) rq->buffer;
        u8 stat = hwif->INB(IDE_STATUS_REG);
        int retries = 10;

        local_irq_enable();
        if ((stat & DRQ_STAT) && args && args[3]) {
                u8 io_32bit = drive->io_32bit;
                drive->io_32bit = 0;
                hwif->ata_input_data(drive, &args[4], args[3] * SECTOR_WORDS);
                drive->io_32bit = io_32bit;
                while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
                        udelay(100);
        }

        if (!OK_STAT(stat, READY_STAT, BAD_STAT) && DRIVER(drive) != NULL)
                return DRIVER(drive)->error(drive, "drive_cmd", stat);
                /* calls ide_end_drive_cmd */
        ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
        return ide_stopped;
}

/**
 *      do_special              -       issue some special commands
 *      @drive: drive the command is for
 *
 *      do_special() is used to issue WIN_SPECIFY, WIN_RESTORE, and WIN_SETMULT
 *      commands to a drive.  It used to do much more, but has been scaled
 *      back.
 */

static ide_startstop_t do_special (ide_drive_t *drive)
{
        special_t *s = &drive->special;

#ifdef DEBUG
        printk("%s: do_special: 0x%02x\n", drive->name, s->all);
#endif
        if (s->b.set_tune) {
                s->b.set_tune = 0;
                if (HWIF(drive)->tuneproc != NULL)
                        HWIF(drive)->tuneproc(drive, drive->tune_req);
                return ide_stopped;
        }
        else
                return DRIVER(drive)->special(drive);
}

void ide_map_sg(ide_drive_t *drive, struct request *rq)
{
        ide_hwif_t *hwif = drive->hwif;
        struct scatterlist *sg = hwif->sg_table;

        if (hwif->sg_mapped)    /* needed by ide-scsi */
                return;

        if ((rq->flags & REQ_DRIVE_TASKFILE) == 0) {
                hwif->sg_nents = blk_rq_map_sg(drive->queue, rq, sg);
        } else {
                sg_init_one(sg, rq->buffer, rq->nr_sectors * SECTOR_SIZE);
                hwif->sg_nents = 1;
        }
}

EXPORT_SYMBOL_GPL(ide_map_sg);

void ide_init_sg_cmd(ide_drive_t *drive, struct request *rq)
{
        ide_hwif_t *hwif = drive->hwif;

        hwif->nsect = hwif->nleft = rq->nr_sectors;
        hwif->cursg = hwif->cursg_ofs = 0;
}

EXPORT_SYMBOL_GPL(ide_init_sg_cmd);

/**
 *      execute_drive_command   -       issue special drive command
 *      @drive: the drive to issue th command on
 *      @rq: the request structure holding the command
 *
 *      execute_drive_cmd() issues a special drive command,  usually 
 *      initiated by ioctl() from the external hdparm program. The
 *      command can be a drive command, drive task or taskfile 
 *      operation. Weirdly you can call it with NULL to wait for
 *      all commands to finish. Don't do this as that is due to change
 */

static ide_startstop_t execute_drive_cmd (ide_drive_t *drive,
                struct request *rq)
{
        ide_hwif_t *hwif = HWIF(drive);
        if (rq->flags & REQ_DRIVE_TASKFILE) {
                ide_task_t *args = rq->special;
 
                if (!args)
                        goto done;

                hwif->data_phase = args->data_phase;

                switch (hwif->data_phase) {
                case TASKFILE_MULTI_OUT:
                case TASKFILE_OUT:
                case TASKFILE_MULTI_IN:
                case TASKFILE_IN:
                        ide_init_sg_cmd(drive, rq);
                        ide_map_sg(drive, rq);
                default:
                        break;
                }

                if (args->tf_out_flags.all != 0) 
                        return flagged_taskfile(drive, args);
                return do_rw_taskfile(drive, args);
        } else if (rq->flags & REQ_DRIVE_TASK) {
                u8 *args = rq->buffer;
                u8 sel;
 
                if (!args)
                        goto done;
#ifdef DEBUG
                printk("%s: DRIVE_TASK_CMD ", drive->name);
                printk("cmd=0x%02x ", args[0]);
                printk("fr=0x%02x ", args[1]);
                printk("ns=0x%02x ", args[2]);
                printk("sc=0x%02x ", args[3]);
                printk("lcyl=0x%02x ", args[4]);
                printk("hcyl=0x%02x ", args[5]);
                printk("sel=0x%02x\n", args[6]);
#endif
                hwif->OUTB(args[1], IDE_FEATURE_REG);
                hwif->OUTB(args[3], IDE_SECTOR_REG);
                hwif->OUTB(args[4], IDE_LCYL_REG);
                hwif->OUTB(args[5], IDE_HCYL_REG);
                sel = (args[6] & ~0x10);
                if (drive->select.b.unit)
                        sel |= 0x10;
                hwif->OUTB(sel, IDE_SELECT_REG);
                ide_cmd(drive, args[0], args[2], &drive_cmd_intr);
                return ide_started;
        } else if (rq->flags & REQ_DRIVE_CMD) {
                u8 *args = rq->buffer;

                if (!args)
                        goto done;
#ifdef DEBUG
                printk("%s: DRIVE_CMD ", drive->name);
                printk("cmd=0x%02x ", args[0]);
                printk("sc=0x%02x ", args[1]);
                printk("fr=0x%02x ", args[2]);
                printk("xx=0x%02x\n", args[3]);
#endif
                if (args[0] == WIN_SMART) {
                        hwif->OUTB(0x4f, IDE_LCYL_REG);
                        hwif->OUTB(0xc2, IDE_HCYL_REG);
                        hwif->OUTB(args[2],IDE_FEATURE_REG);
                        hwif->OUTB(args[1],IDE_SECTOR_REG);
                        ide_cmd(drive, args[0], args[3], &drive_cmd_intr);
                        return ide_started;
                }
                hwif->OUTB(args[2],IDE_FEATURE_REG);
                ide_cmd(drive, args[0], args[1], &drive_cmd_intr);
                return ide_started;
        }

done:
        /*
         * NULL is actually a valid way of waiting for
         * all current requests to be flushed from the queue.
         */
#ifdef DEBUG
        printk("%s: DRIVE_CMD (null)\n", drive->name);
#endif
        ide_end_drive_cmd(drive,
                        hwif->INB(IDE_STATUS_REG),
                        hwif->INB(IDE_ERROR_REG));
        return ide_stopped;
}

/**
 *      start_request   -       start of I/O and command issuing for IDE
 *
 *      start_request() initiates handling of a new I/O request. It
 *      accepts commands and I/O (read/write) requests. It also does
 *      the final remapping for weird stuff like EZDrive. Once 
 *      device mapper can work sector level the EZDrive stuff can go away
 *
 *      FIXME: this function needs a rename
 */
 
static ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
{
        ide_startstop_t startstop;
        sector_t block;

        BUG_ON(!(rq->flags & REQ_STARTED));

#ifdef DEBUG
        printk("%s: start_request: current=0x%08lx\n",
                HWIF(drive)->name, (unsigned long) rq);
#endif

        /* bail early if we've exceeded max_failures */
        if (drive->max_failures && (drive->failures > drive->max_failures)) {
                goto kill_rq;
        }

        /*
         * bail early if we've sent a device to sleep, however how to wake
         * this needs to be a masked flag.  FIXME for proper operations.
         */
        if (drive->suspend_reset)
                goto kill_rq;

        block    = rq->sector;
        if (blk_fs_request(rq) &&
            (drive->media == ide_disk || drive->media == ide_floppy)) {
                block += drive->sect0;
        }
        /* Yecch - this will shift the entire interval,
           possibly killing some innocent following sector */
        if (block == 0 && drive->remap_0_to_1 == 1)
                block = 1;  /* redirect MBR access to EZ-Drive partn table */

        if (blk_pm_suspend_request(rq) &&
            rq->pm->pm_step == ide_pm_state_start_suspend)
                /* Mark drive blocked when starting the suspend sequence. */
                drive->blocked = 1;
        else if (blk_pm_resume_request(rq) &&
                 rq->pm->pm_step == ide_pm_state_start_resume) {
                /* 
                 * The first thing we do on wakeup is to wait for BSY bit to
                 * go away (with a looong timeout) as a drive on this hwif may
                 * just be POSTing itself.
                 * We do that before even selecting as the "other" device on
                 * the bus may be broken enough to walk on our toes at this
                 * point.
                 */
                int rc;
#ifdef DEBUG_PM
                printk("%s: Wakeup request inited, waiting for !BSY...\n", drive->name);
#endif
                rc = ide_wait_not_busy(HWIF(drive), 35000);
                if (rc)
                        printk(KERN_WARNING "%s: bus not ready on wakeup\n", drive->name);
                SELECT_DRIVE(drive);
                HWIF(drive)->OUTB(8, HWIF(drive)->io_ports[IDE_CONTROL_OFFSET]);
                rc = ide_wait_not_busy(HWIF(drive), 10000);
                if (rc)
                        printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
        }

        SELECT_DRIVE(drive);
        if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
                printk(KERN_ERR "%s: drive not ready for command\n", drive->name);
                return startstop;
        }
        if (!drive->special.all) {
                if (rq->flags & (REQ_DRIVE_CMD | REQ_DRIVE_TASK))
                        return execute_drive_cmd(drive, rq);
                else if (rq->flags & REQ_DRIVE_TASKFILE)
                        return execute_drive_cmd(drive, rq);
                else if (blk_pm_request(rq)) {
#ifdef DEBUG_PM
                        printk("%s: start_power_step(step: %d)\n",
                                drive->name, rq->pm->pm_step);
#endif
                        startstop = DRIVER(drive)->start_power_step(drive, rq);
                        if (startstop == ide_stopped &&
                            rq->pm->pm_step == ide_pm_state_completed)
                                ide_complete_pm_request(drive, rq);
                        return startstop;
                }
                return (DRIVER(drive)->do_request(drive, rq, block));
        }
        return do_special(drive);
kill_rq:
        DRIVER(drive)->end_request(drive, 0, 0);
        return ide_stopped;
}

/**
 *      ide_stall_queue         -       pause an IDE device
 *      @drive: drive to stall
 *      @timeout: time to stall for (jiffies)
 *
 *      ide_stall_queue() can be used by a drive to give excess bandwidth back
 *      to the hwgroup by sleeping for timeout jiffies.
 */
 
void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
{
        if (timeout > WAIT_WORSTCASE)
                timeout = WAIT_WORSTCASE;
        drive->sleep = timeout + jiffies;
}

EXPORT_SYMBOL(ide_stall_queue);

#define WAKEUP(drive)   ((drive)->service_start + 2 * (drive)->service_time)

/**
 *      choose_drive            -       select a drive to service
 *      @hwgroup: hardware group to select on
 *
 *      choose_drive() selects the next drive which will be serviced.
 *      This is necessary because the IDE layer can't issue commands
 *      to both drives on the same cable, unlike SCSI.
 */
 
static inline ide_drive_t *choose_drive (ide_hwgroup_t *hwgroup)
{
        ide_drive_t *drive, *best;

repeat: 
        best = NULL;
        drive = hwgroup->drive;

        /*
         * drive is doing pre-flush, ordered write, post-flush sequence. even
         * though that is 3 requests, it must be seen as a single transaction.
         * we must not preempt this drive until that is complete
         */
        if (drive->doing_barrier) {
                /*
                 * small race where queue could get replugged during
                 * the 3-request flush cycle, just yank the plug since
                 * we want it to finish asap
                 */
                blk_remove_plug(drive->queue);
                return drive;
        }

        do {
                if ((!drive->sleep || time_after_eq(jiffies, drive->sleep))
                    && !elv_queue_empty(drive->queue)) {
                        if (!best
                         || (drive->sleep && (!best->sleep || 0 < (signed long)(best->sleep - drive->sleep)))
                         || (!best->sleep && 0 < (signed long)(WAKEUP(best) - WAKEUP(drive))))
                        {
                                if (!blk_queue_plugged(drive->queue))
                                        best = drive;
                        }
                }
        } while ((drive = drive->next) != hwgroup->drive);
        if (best && best->nice1 && !best->sleep && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
                long t = (signed long)(WAKEUP(best) - jiffies);
                if (t >= WAIT_MIN_SLEEP) {
                /*
                 * We *may* have some time to spare, but first let's see if
                 * someone can potentially benefit from our nice mood today..
                 */
                        drive = best->next;
                        do {
                                if (!drive->sleep
                                /* FIXME: use time_before */
                                 && 0 < (signed long)(WAKEUP(drive) - (jiffies - best->service_time))
                                 && 0 < (signed long)((jiffies + t) - WAKEUP(drive)))
                                {
                                        ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
                                        goto repeat;
                                }
                        } while ((drive = drive->next) != best);
                }
        }
        return best;
}

/*
 * Issue a new request to a drive from hwgroup
 * Caller must have already done spin_lock_irqsave(&ide_lock, ..);
 *
 * A hwgroup is a serialized group of IDE interfaces.  Usually there is
 * exactly one hwif (interface) per hwgroup, but buggy controllers (eg. CMD640)
 * may have both interfaces in a single hwgroup to "serialize" access.
 * Or possibly multiple ISA interfaces can share a common IRQ by being grouped
 * together into one hwgroup for serialized access.
 *
 * Note also that several hwgroups can end up sharing a single IRQ,
 * possibly along with many other devices.  This is especially common in
 * PCI-based systems with off-board IDE controller cards.
 *
 * The IDE driver uses the single global ide_lock spinlock to protect
 * access to the request queues, and to protect the hwgroup->busy flag.
 *
 * The first thread into the driver for a particular hwgroup sets the
 * hwgroup->busy flag to indicate that this hwgroup is now active,
 * and then initiates processing of the top request from the request queue.
 *
 * Other threads attempting entry notice the busy setting, and will simply
 * queue their new requests and exit immediately.  Note that hwgroup->busy
 * remains set even when the driver is merely awaiting the next interrupt.
 * Thus, the meaning is "this hwgroup is busy processing a request".
 *
 * When processing of a request completes, the completing thread or IRQ-handler
 * will start the next request from the queue.  If no more work remains,
 * the driver will clear the hwgroup->busy flag and exit.
 *
 * The ide_lock (spinlock) is used to protect all access to the
 * hwgroup->busy flag, but is otherwise not needed for most processing in
 * the driver.  This makes the driver much more friendlier to shared IRQs
 * than previous designs, while remaining 100% (?) SMP safe and capable.
 */
static void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
{
        ide_drive_t     *drive;
        ide_hwif_t      *hwif;
        struct request  *rq;
        ide_startstop_t startstop;

        /* for atari only: POSSIBLY BROKEN HERE(?) */
        ide_get_lock(ide_intr, hwgroup);

        /* caller must own ide_lock */
        BUG_ON(!irqs_disabled());

        while (!hwgroup->busy) {
                hwgroup->busy = 1;
                drive = choose_drive(hwgroup);
                if (drive == NULL) {
                        unsigned long sleep = 0;
                        hwgroup->rq = NULL;
                        drive = hwgroup->drive;
                        do {
                                if (drive->sleep && (!sleep || 0 < (signed long)(sleep - drive->sleep)))
                                        sleep = drive->sleep;
                        } while ((drive = drive->next) != hwgroup->drive);
                        if (sleep) {
                /*
                 * Take a short snooze, and then wake up this hwgroup again.
                 * This gives other hwgroups on the same a chance to
                 * play fairly with us, just in case there are big differences
                 * in relative throughputs.. don't want to hog the cpu too much.
                 */
                                if (time_before(sleep, jiffies + WAIT_MIN_SLEEP))
                                        sleep = jiffies + WAIT_MIN_SLEEP;
#if 1
                                if (timer_pending(&hwgroup->timer))
                                        printk(KERN_CRIT "ide_set_handler: timer already active\n");
#endif
                                /* so that ide_timer_expiry knows what to do */
                                hwgroup->sleeping = 1;
                                mod_timer(&hwgroup->timer, sleep);
                                /* we purposely leave hwgroup->busy==1
                                 * while sleeping */
                        } else {
                                /* Ugly, but how can we sleep for the lock
                                 * otherwise? perhaps from tq_disk?
                                 */

                                /* for atari only */
                                ide_release_lock();
                                hwgroup->busy = 0;
                        }

                        /* no more work for this hwgroup (for now) */
                        return;
                }
                hwif = HWIF(drive);
                if (hwgroup->hwif->sharing_irq &&
                    hwif != hwgroup->hwif &&
                    hwif->io_ports[IDE_CONTROL_OFFSET]) {
                        /* set nIEN for previous hwif */
                        SELECT_INTERRUPT(drive);
                }
                hwgroup->hwif = hwif;
                hwgroup->drive = drive;
                drive->sleep = 0;
                drive->service_start = jiffies;

                if (blk_queue_plugged(drive->queue)) {
                        printk(KERN_ERR "ide: huh? queue was plugged!\n");
                        break;
                }

                /*
                 * we know that the queue isn't empty, but this can happen
                 * if the q->prep_rq_fn() decides to kill a request
                 */
                rq = elv_next_request(drive->queue);
                if (!rq) {
                        hwgroup->busy = 0;
                        break;
                }

                /*
                 * if rq is a barrier write, issue pre cache flush if not
                 * already done
                 */
                if (blk_barrier_rq(rq) && !blk_barrier_preflush(rq))
                        rq = ide_queue_flush_cmd(drive, rq, 0);

                /*
                 * Sanity: don't accept a request that isn't a PM request
                 * if we are currently power managed. This is very important as
                 * blk_stop_queue() doesn't prevent the elv_next_request()
                 * above to return us whatever is in the queue. Since we call
                 * ide_do_request() ourselves, we end up taking requests while
                 * the queue is blocked...
                 * 
                 * We let requests forced at head of queue with ide-preempt
                 * though. I hope that doesn't happen too much, hopefully not
                 * unless the subdriver triggers such a thing in its own PM
                 * state machine.
                 */
                if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
                        /* We clear busy, there should be no pending ATA command at this point. */
                        hwgroup->busy = 0;
                        break;
                }

                hwgroup->rq = rq;

                /*
                 * Some systems have trouble with IDE IRQs arriving while
                 * the driver is still setting things up.  So, here we disable
                 * the IRQ used by this interface while the request is being started.
                 * This may look bad at first, but pretty much the same thing
                 * happens anyway when any interrupt comes in, IDE or otherwise
                 *  -- the kernel masks the IRQ while it is being handled.
                 */
                if (hwif->irq != masked_irq)
                        disable_irq_nosync(hwif->irq);
                spin_unlock(&ide_lock);
                local_irq_enable();
                        /* allow other IRQs while we start this request */
                startstop = start_request(drive, rq);
                spin_lock_irq(&ide_lock);
                if (hwif->irq != masked_irq)
                        enable_irq(hwif->irq);
                if (startstop == ide_stopped)
                        hwgroup->busy = 0;
        }
}

/*
 * Passes the stuff to ide_do_request
 */
void do_ide_request(request_queue_t *q)
{
        ide_drive_t *drive = q->queuedata;

        ide_do_request(HWGROUP(drive), IDE_NO_IRQ);
}

/*
 * un-busy the hwgroup etc, and clear any pending DMA status. we want to
 * retry the current request in pio mode instead of risking tossing it
 * all away
 */
static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
{
        ide_hwif_t *hwif = HWIF(drive);
        struct request *rq;
        ide_startstop_t ret = ide_stopped;

        /*
         * end current dma transaction
         */

        if (error < 0) {
                printk(KERN_WARNING "%s: DMA timeout error\n", drive->name);
                (void)HWIF(drive)->ide_dma_end(drive);
                ret = DRIVER(drive)->error(drive, "dma timeout error",
                                                hwif->INB(IDE_STATUS_REG));
        } else {
                printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
                (void) hwif->ide_dma_timeout(drive);
        }

        /*
         * disable dma for now, but remember that we did so because of
         * a timeout -- we'll reenable after we finish this next request
         * (or rather the first chunk of it) in pio.
         */
        drive->retry_pio++;
        drive->state = DMA_PIO_RETRY;
        (void) hwif->ide_dma_off_quietly(drive);

        /*
         * un-busy drive etc (hwgroup->busy is cleared on return) and
         * make sure request is sane
         */
        rq = HWGROUP(drive)->rq;
        HWGROUP(drive)->rq = NULL;

        rq->errors = 0;

        if (!rq->bio)
                goto out;

        rq->sector = rq->bio->bi_sector;
        rq->current_nr_sectors = bio_iovec(rq->bio)->bv_len >> 9;
        rq->hard_cur_sectors = rq->current_nr_sectors;
        rq->buffer = bio_data(rq->bio);
out:
        return ret;
}

/**
 *      ide_timer_expiry        -       handle lack of an IDE interrupt
 *      @data: timer callback magic (hwgroup)
 *
 *      An IDE command has timed out before the expected drive return
 *      occurred. At this point we attempt to clean up the current
 *      mess. If the current handler includes an expiry handler then
 *      we invoke the expiry handler, and providing it is happy the
 *      work is done. If that fails we apply generic recovery rules
 *      invoking the handler and checking the drive DMA status. We
 *      have an excessively incestuous relationship with the DMA
 *      logic that wants cleaning up.
 */
 
void ide_timer_expiry (unsigned long data)
{
        ide_hwgroup_t   *hwgroup = (ide_hwgroup_t *) data;
        ide_handler_t   *handler;
        ide_expiry_t    *expiry;
        unsigned long   flags;
        unsigned long   wait = -1;

        spin_lock_irqsave(&ide_lock, flags);

        if ((handler = hwgroup->handler) == NULL) {
                /*
                 * Either a marginal timeout occurred
                 * (got the interrupt just as timer expired),
                 * or we were "sleeping" to give other devices a chance.
                 * Either way, we don't really want to complain about anything.
                 */
                if (hwgroup->sleeping) {
                        hwgroup->sleeping = 0;
                        hwgroup->busy = 0;
                }
        } else {
                ide_drive_t *drive = hwgroup->drive;
                if (!drive) {
                        printk(KERN_ERR "ide_timer_expiry: hwgroup->drive was NULL\n");
                        hwgroup->handler = NULL;
                } else {
                        ide_hwif_t *hwif;
                        ide_startstop_t startstop = ide_stopped;
                        if (!hwgroup->busy) {
                                hwgroup->busy = 1;      /* paranoia */
                                printk(KERN_ERR "%s: ide_timer_expiry: hwgroup->busy was 0 ??\n", drive->name);
                        }
                        if ((expiry = hwgroup->expiry) != NULL) {
                                /* continue */
                                if ((wait = expiry(drive)) > 0) {
                                        /* reset timer */
                                        hwgroup->timer.expires  = jiffies + wait;
                                        add_timer(&hwgroup->timer);
                                        spin_unlock_irqrestore(&ide_lock, flags);
                                        return;
                                }
                        }
                        hwgroup->handler = NULL;
                        /*
                         * We need to simulate a real interrupt when invoking
                         * the handler() function, which means we need to
                         * globally mask the specific IRQ:
                         */
                        spin_unlock(&ide_lock);
                        hwif  = HWIF(drive);
#if DISABLE_IRQ_NOSYNC
                        disable_irq_nosync(hwif->irq);
#else
                        /* disable_irq_nosync ?? */
                        disable_irq(hwif->irq);
#endif /* DISABLE_IRQ_NOSYNC */
                        /* local CPU only,
                         * as if we were handling an interrupt */
                        local_irq_disable();
                        if (hwgroup->poll_timeout != 0) {
                                startstop = handler(drive);
                        } else if (drive_is_ready(drive)) {
                                if (drive->waiting_for_dma)
                                        (void) hwgroup->hwif->ide_dma_lostirq(drive);
                                (void)ide_ack_intr(hwif);
                                printk(KERN_WARNING "%s: lost interrupt\n", drive->name);
                                startstop = handler(drive);
                        } else {
                                if (drive->waiting_for_dma) {
                                        startstop = ide_dma_timeout_retry(drive, wait);
                                } else
                                        startstop =
                                        DRIVER(drive)->error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
                        }
                        drive->service_time = jiffies - drive->service_start;
                        spin_lock_irq(&ide_lock);
                        enable_irq(hwif->irq);
                        if (startstop == ide_stopped)
                                hwgroup->busy = 0;
                }
        }
        ide_do_request(hwgroup, IDE_NO_IRQ);
        spin_unlock_irqrestore(&ide_lock, flags);
}

/**
 *      unexpected_intr         -       handle an unexpected IDE interrupt
 *      @irq: interrupt line
 *      @hwgroup: hwgroup being processed
 *
 *      There's nothing really useful we can do with an unexpected interrupt,
 *      other than reading the status register (to clear it), and logging it.
 *      There should be no way that an irq can happen before we're ready for it,
 *      so we needn't worry much about losing an "important" interrupt here.
 *
 *      On laptops (and "green" PCs), an unexpected interrupt occurs whenever
 *      the drive enters "idle", "standby", or "sleep" mode, so if the status
 *      looks "good", we just ignore the interrupt completely.
 *
 *      This routine assumes __cli() is in effect when called.
 *
 *      If an unexpected interrupt happens on irq15 while we are handling irq14
 *      and if the two interfaces are "serialized" (CMD640), then it looks like
 *      we could screw up by interfering with a new request being set up for 
 *      irq15.
 *
 *      In reality, this is a non-issue.  The new command is not sent unless 
 *      the drive is ready to accept one, in which case we know the drive is
 *      not trying to interrupt us.  And ide_set_handler() is always invoked
 *      before completing the issuance of any new drive command, so we will not
 *      be accidentally invoked as a result of any valid command completion
 *      interrupt.
 *
 *      Note that we must walk the entire hwgroup here. We know which hwif
 *      is doing the current command, but we don't know which hwif burped
 *      mysteriously.
 */
 
static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
{
        u8 stat;
        ide_hwif_t *hwif = hwgroup->hwif;

        /*
         * handle the unexpected interrupt
         */
        do {
                if (hwif->irq == irq) {
                        stat = hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
                        if (!OK_STAT(stat, READY_STAT, BAD_STAT)) {
                                /* Try to not flood the console with msgs */
                                static unsigned long last_msgtime, count;
                                ++count;
                                if (time_after(jiffies, last_msgtime + HZ)) {
                                        last_msgtime = jiffies;
                                        printk(KERN_ERR "%s%s: unexpected interrupt, "
                                                "status=0x%02x, count=%ld\n",
                                                hwif->name,
                                                (hwif->next==hwgroup->hwif) ? "" : "(?)", stat, count);
                                }
                        }
                }
        } while ((hwif = hwif->next) != hwgroup->hwif);
}

/**
 *      ide_intr        -       default IDE interrupt handler
 *      @irq: interrupt number
 *      @dev_id: hwif group
 *      @regs: unused weirdness from the kernel irq layer
 *
 *      This is the default IRQ handler for the IDE layer. You should
 *      not need to override it. If you do be aware it is subtle in
 *      places
 *
 *      hwgroup->hwif is the interface in the group currently performing
 *      a command. hwgroup->drive is the drive and hwgroup->handler is
 *      the IRQ handler to call. As we issue a command the handlers
 *      step through multiple states, reassigning the handler to the
 *      next step in the process. Unlike a smart SCSI controller IDE
 *      expects the main processor to sequence the various transfer
 *      stages. We also manage a poll timer to catch up with most
 *      timeout situations. There are still a few where the handlers
 *      don't ever decide to give up.
 *
 *      The handler eventually returns ide_stopped to indicate the
 *      request completed. At this point we issue the next request
 *      on the hwgroup and the process begins again.
 */
 
irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
{
        unsigned long flags;
        ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
        ide_hwif_t *hwif;
        ide_drive_t *drive;
        ide_handler_t *handler;
        ide_startstop_t startstop;

        spin_lock_irqsave(&ide_lock, flags);
        hwif = hwgroup->hwif;

        if (!ide_ack_intr(hwif)) {
                spin_unlock_irqrestore(&ide_lock, flags);
                return IRQ_NONE;
        }

        if (hwif->polling) {
                /* We took an interrupt during a polled drive retune. 
                   This should go away eventually when that code uses
                   the polling logic like do_reset1 */
                spin_unlock_irqrestore(&ide_lock, flags);
                return IRQ_HANDLED;
        }
        if ((handler = hwgroup->handler) == NULL ||
            hwgroup->poll_timeout != 0) {
                /*
                 * Not expecting an interrupt from this drive.
                 * That means this could be:
                 *      (1) an interrupt from another PCI device
                 *      sharing the same PCI INT# as us.
                 * or   (2) a drive just entered sleep or standby mode,
                 *      and is interrupting to let us know.
                 * or   (3) a spurious interrupt of unknown origin.
                 *
                 * For PCI, we cannot tell the difference,
                 * so in that case we just ignore it and hope it goes away.
                 *
                 * FIXME: unexpected_intr should be hwif-> then we can
                 * remove all the ifdef PCI crap
                 */
#ifdef CONFIG_BLK_DEV_IDEPCI
                if (hwif->pci_dev && !hwif->pci_dev->vendor)
#endif  /* CONFIG_BLK_DEV_IDEPCI */
                {
                        /*
                         * Probably not a shared PCI interrupt,
                         * so we can safely try to do something about it:
                         */
                        unexpected_intr(irq, hwgroup);
#ifdef CONFIG_BLK_DEV_IDEPCI
                } else {
                        /*
                         * Whack the status register, just in case
                         * we have a leftover pending IRQ.
                         */
                        (void) hwif->INB(hwif->io_ports[IDE_STATUS_OFFSET]);
#endif /* CONFIG_BLK_DEV_IDEPCI */
                }
                spin_unlock_irqrestore(&ide_lock, flags);
                return IRQ_NONE;
        }
        drive = hwgroup->drive;
        if (!drive) {
                /*
                 * This should NEVER happen, and there isn't much
                 * we could do about it here.
                 *
                 * [Note - this can occur if the drive is hot unplugged]
                 */
                spin_unlock_irqrestore(&ide_lock, flags);
                return IRQ_HANDLED;
        }
        if (!drive_is_ready(drive)) {
                /*
                 * This happens regularly when we share a PCI IRQ with
                 * another device.  Unfortunately, it can also happen
                 * with some buggy drives that trigger the IRQ before
                 * their status register is up to date.  Hopefully we have
                 * enough advance overhead that the latter isn't a problem.
                 */
                spin_unlock_irqrestore(&ide_lock, flags);
                return IRQ_NONE;
        }
        if (!hwgroup->busy) {
                hwgroup->busy = 1;      /* paranoia */
                printk(KERN_ERR "%s: ide_intr: hwgroup->busy was 0 ??\n", drive->name);
        }
        hwgroup->handler = NULL;
        del_timer(&hwgroup->timer);
        spin_unlock(&ide_lock);

        if (drive->unmask)
                local_irq_enable();
        /* service this interrupt, may set handler for next interrupt */
        startstop = handler(drive);
        spin_lock_irq(&ide_lock);

        /*
         * Note that handler() may have set things up for another
         * interrupt to occur soon, but it cannot happen until
         * we exit from this routine, because it will be the
         * same irq as is currently being serviced here, and Linux
         * won't allow another of the same (on any CPU) until we return.
         */
        drive->service_time = jiffies - drive->service_start;
        if (startstop == ide_stopped) {
                if (hwgroup->handler == NULL) { /* paranoia */
                        hwgroup->busy = 0;
                        ide_do_request(hwgroup, hwif->irq);
                } else {
                        printk(KERN_ERR "%s: ide_intr: huh? expected NULL handler "
                                "on exit\n", drive->name);
                }
        }
        spin_unlock_irqrestore(&ide_lock, flags);
        return IRQ_HANDLED;
}

/**
 *      ide_init_drive_cmd      -       initialize a drive command request
 *      @rq: request object
 *
 *      Initialize a request before we fill it in and send it down to
 *      ide_do_drive_cmd. Commands must be set up by this function. Right
 *      now it doesn't do a lot, but if that changes abusers will have a
 *      nasty suprise.
 */

void ide_init_drive_cmd (struct request *rq)
{
        memset(rq, 0, sizeof(*rq));
        rq->flags = REQ_DRIVE_CMD;
        rq->ref_count = 1;
}

EXPORT_SYMBOL(ide_init_drive_cmd);

/**
 *      ide_do_drive_cmd        -       issue IDE special command
 *      @drive: device to issue command
 *      @rq: request to issue
 *      @action: action for processing
 *
 *      This function issues a special IDE device request
 *      onto the request queue.
 *
 *      If action is ide_wait, then the rq is queued at the end of the
 *      request queue, and the function sleeps until it has been processed.
 *      This is for use when invoked from an ioctl handler.
 *
 *      If action is ide_preempt, then the rq is queued at the head of
 *      the request queue, displacing the currently-being-processed
 *      request and this function returns immediately without waiting
 *      for the new rq to be completed.  This is VERY DANGEROUS, and is
 *      intended for careful use by the ATAPI tape/cdrom driver code.
 *
 *      If action is ide_next, then the rq is queued immediately after
 *      the currently-being-processed-request (if any), and the function
 *      returns without waiting for the new rq to be completed.  As above,
 *      This is VERY DANGEROUS, and is intended for careful use by the
 *      ATAPI tape/cdrom driver code.
 *
 *      If action is ide_end, then the rq is queued at the end of the
 *      request queue, and the function returns immediately without waiting
 *      for the new rq to be completed. This is again intended for careful
 *      use by the ATAPI tape/cdrom driver code.
 */
 
int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t action)
{
        unsigned long flags;
        ide_hwgroup_t *hwgroup = HWGROUP(drive);
        DECLARE_COMPLETION(wait);
        int where = ELEVATOR_INSERT_BACK, err;
        int must_wait = (action == ide_wait || action == ide_head_wait);

        rq->errors = 0;
        rq->rq_status = RQ_ACTIVE;

        rq->rq_disk = drive->disk;

        /*
         * we need to hold an extra reference to request for safe inspection
         * after completion
         */
        if (must_wait) {
                rq->ref_count++;
                rq->waiting = &wait;
        }

        spin_lock_irqsave(&ide_lock, flags);
        if (action == ide_preempt)
                hwgroup->rq = NULL;
        if (action == ide_preempt || action == ide_head_wait) {
                where = ELEVATOR_INSERT_FRONT;
                rq->flags |= REQ_PREEMPT;
        }
        __elv_add_request(drive->queue, rq, where, 0);
        ide_do_request(hwgroup, IDE_NO_IRQ);
        spin_unlock_irqrestore(&ide_lock, flags);

        err = 0;
        if (must_wait) {
                wait_for_completion(&wait);
                rq->waiting = NULL;
                if (rq->errors)
                        err = -EIO;

                blk_put_request(rq);
        }

        return err;
}

EXPORT_SYMBOL(ide_do_drive_cmd);