fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / drivers / ide / ide-io.c
index d20cd1e..d6ff4f2 100644 (file)
@@ -24,7 +24,6 @@
  */
  
  
-#include <linux/config.h>
 #include <linux/module.h>
 #include <linux/types.h>
 #include <linux/string.h>
@@ -47,6 +46,7 @@
 #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/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
@@ -136,15 +80,13 @@ static int __ide_end_request(ide_drive_t *drive, struct request *rq,
 
        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);
+               if (!list_empty(&rq->queuelist))
+                       blkdev_dequeue_request(rq);
                HWGROUP(drive)->rq = NULL;
-               end_that_request_last(rq);
+               end_that_request_last(rq, uptodate);
                ret = 0;
        }
+
        return ret;
 }
 
@@ -165,28 +107,188 @@ int ide_end_request (ide_drive_t *drive, int uptodate, int nr_sectors)
        unsigned long flags;
        int ret = 1;
 
+       /*
+        * room for locking improvements here, the calls below don't
+        * need the queue lock held at all
+        */
        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;
+       ret = __ide_end_request(drive, rq, uptodate, nr_sectors);
+
+       spin_unlock_irqrestore(&ide_lock, flags);
+       return ret;
+}
+EXPORT_SYMBOL(ide_end_request);
+
+/*
+ * Power Management state machine. This one is rather trivial for now,
+ * we should probably add more, like switching back to PIO on suspend
+ * to help some BIOSes, re-do the door locking on resume, etc...
+ */
+
+enum {
+       ide_pm_flush_cache      = ide_pm_state_start_suspend,
+       idedisk_pm_standby,
 
-               flush_rq->nr_sectors -= nr_sectors;
-               if (!flush_rq->nr_sectors) {
-                       ide_queue_flush_cmd(drive, rq, 1);
-                       ret = 0;
+       idedisk_pm_restore_pio  = ide_pm_state_start_resume,
+       idedisk_pm_idle,
+       ide_pm_restore_dma,
+};
+
+static void ide_complete_power_step(ide_drive_t *drive, struct request *rq, u8 stat, u8 error)
+{
+       struct request_pm_state *pm = rq->data;
+
+       if (drive->media != ide_disk)
+               return;
+
+       switch (pm->pm_step) {
+       case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) complete */
+               if (pm->pm_state == PM_EVENT_FREEZE)
+                       pm->pm_step = ide_pm_state_completed;
+               else
+                       pm->pm_step = idedisk_pm_standby;
+               break;
+       case idedisk_pm_standby:        /* Suspend step 2 (standby) complete */
+               pm->pm_step = ide_pm_state_completed;
+               break;
+       case idedisk_pm_restore_pio:    /* Resume step 1 complete */
+               pm->pm_step = idedisk_pm_idle;
+               break;
+       case idedisk_pm_idle:           /* Resume step 2 (idle) complete */
+               pm->pm_step = ide_pm_restore_dma;
+               break;
+       }
+}
+
+static ide_startstop_t ide_start_power_step(ide_drive_t *drive, struct request *rq)
+{
+       struct request_pm_state *pm = rq->data;
+       ide_task_t *args = rq->special;
+
+       memset(args, 0, sizeof(*args));
+
+       if (drive->media != ide_disk) {
+               /*
+                * skip idedisk_pm_restore_pio and idedisk_pm_idle for ATAPI
+                * devices
+                */
+               if (pm->pm_step == idedisk_pm_restore_pio)
+                       pm->pm_step = ide_pm_restore_dma;
+       }
+
+       switch (pm->pm_step) {
+       case ide_pm_flush_cache:        /* Suspend step 1 (flush cache) */
+               if (drive->media != ide_disk)
+                       break;
+               /* Not supported? Switch to next step now. */
+               if (!drive->wcache || !ide_id_has_flush_cache(drive->id)) {
+                       ide_complete_power_step(drive, rq, 0, 0);
+                       return ide_stopped;
                }
+               if (ide_id_has_flush_cache_ext(drive->id))
+                       args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE_EXT;
+               else
+                       args->tfRegister[IDE_COMMAND_OFFSET] = WIN_FLUSH_CACHE;
+               args->command_type = IDE_DRIVE_TASK_NO_DATA;
+               args->handler      = &task_no_data_intr;
+               return do_rw_taskfile(drive, args);
+
+       case idedisk_pm_standby:        /* Suspend step 2 (standby) */
+               args->tfRegister[IDE_COMMAND_OFFSET] = WIN_STANDBYNOW1;
+               args->command_type = IDE_DRIVE_TASK_NO_DATA;
+               args->handler      = &task_no_data_intr;
+               return do_rw_taskfile(drive, args);
+
+       case idedisk_pm_restore_pio:    /* Resume step 1 (restore PIO) */
+               if (drive->hwif->tuneproc != NULL)
+                       drive->hwif->tuneproc(drive, 255);
+               ide_complete_power_step(drive, rq, 0, 0);
+               return ide_stopped;
+
+       case idedisk_pm_idle:           /* Resume step 2 (idle) */
+               args->tfRegister[IDE_COMMAND_OFFSET] = WIN_IDLEIMMEDIATE;
+               args->command_type = IDE_DRIVE_TASK_NO_DATA;
+               args->handler = task_no_data_intr;
+               return do_rw_taskfile(drive, args);
+
+       case ide_pm_restore_dma:        /* Resume step 3 (restore DMA) */
+               /*
+                * Right now, all we do is call hwif->ide_dma_check(drive),
+                * we could be smarter and check for current xfer_speed
+                * in struct drive etc...
+                */
+               if ((drive->id->capability & 1) == 0)
+                       break;
+               if (drive->hwif->ide_dma_check == NULL)
+                       break;
+               drive->hwif->ide_dma_check(drive);
+               break;
+       }
+       pm->pm_step = ide_pm_state_completed;
+       return ide_stopped;
+}
+
+/**
+ *     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(!blk_rq_started(rq));
+
+       /*
+        * 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, uptodate);
+               ret = 0;
+       }
        spin_unlock_irqrestore(&ide_lock, flags);
        return ret;
 }
-EXPORT_SYMBOL(ide_end_request);
+EXPORT_SYMBOL_GPL(ide_end_dequeued_request);
+
 
 /**
  *     ide_complete_pm_request - end the current Power Management request
@@ -213,7 +315,7 @@ static void ide_complete_pm_request (ide_drive_t *drive, struct request *rq)
        }
        blkdev_dequeue_request(rq);
        HWGROUP(drive)->rq = NULL;
-       end_that_request_last(rq);
+       end_that_request_last(rq, 1);
        spin_unlock_irqrestore(&ide_lock, flags);
 }
 
@@ -237,9 +339,10 @@ u64 ide_get_error_location(ide_drive_t *drive, char *args)
                high = ide_read_24(drive);
        } else {
                u8 cur = HWIF(drive)->INB(IDE_SELECT_REG);
-               if (cur & 0x40)
+               if (cur & 0x40) {
+                       high = cur & 0xf;
                        low = (hcyl << 16) | (lcyl << 8) | sect;
-               else {
+               else {
                        low = hcyl * drive->head * drive->sect;
                        low += lcyl * drive->sect;
                        low += sect - 1;
@@ -251,79 +354,6 @@ u64 ide_get_error_location(ide_drive_t *drive, char *args)
 }
 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 
@@ -348,7 +378,7 @@ void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
        rq = HWGROUP(drive)->rq;
        spin_unlock_irqrestore(&ide_lock, flags);
 
-       if (rq->flags & REQ_DRIVE_CMD) {
+       if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
                u8 *args = (u8 *) rq->buffer;
                if (rq->errors == 0)
                        rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
@@ -358,7 +388,7 @@ void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
                        args[1] = err;
                        args[2] = hwif->INB(IDE_NSECTOR_REG);
                }
-       } else if (rq->flags & REQ_DRIVE_TASK) {
+       } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
                u8 *args = (u8 *) rq->buffer;
                if (rq->errors == 0)
                        rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
@@ -372,7 +402,7 @@ void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
                        args[5] = hwif->INB(IDE_HCYL_REG);
                        args[6] = hwif->INB(IDE_SELECT_REG);
                }
-       } else if (rq->flags & REQ_DRIVE_TASKFILE) {
+       } else if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
                ide_task_t *args = (ide_task_t *) rq->special;
                if (rq->errors == 0)
                        rq->errors = !OK_STAT(stat,READY_STAT,BAD_STAT);
@@ -403,24 +433,22 @@ void ide_end_drive_cmd (ide_drive_t *drive, u8 stat, u8 err)
                        }
                }
        } else if (blk_pm_request(rq)) {
+               struct request_pm_state *pm = rq->data;
 #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_power_step(drive, rq, stat, err);
+               if (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);
+       rq->errors = err;
+       end_that_request_last(rq, !rq->errors);
        spin_unlock_irqrestore(&ide_lock, flags);
 }
 
@@ -436,7 +464,7 @@ EXPORT_SYMBOL(ide_end_drive_cmd);
  *     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)
+static void try_to_flush_leftover_data (ide_drive_t *drive)
 {
        int i = (drive->mult_count ? drive->mult_count : 1) * SECTOR_WORDS;
 
@@ -451,11 +479,104 @@ void try_to_flush_leftover_data (ide_drive_t *drive)
        }
 }
 
-EXPORT_SYMBOL(try_to_flush_leftover_data);
+static void ide_kill_rq(ide_drive_t *drive, struct request *rq)
+{
+       if (rq->rq_disk) {
+               ide_driver_t *drv;
 
-/*
- * FIXME Add an ATAPI error
- */
+               drv = *(ide_driver_t **)rq->rq_disk->private_data;
+               drv->end_request(drive, 0, 0);
+       } else
+               ide_end_request(drive, 0, 0);
+}
+
+static ide_startstop_t ide_ata_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
+{
+       ide_hwif_t *hwif = drive->hwif;
+
+       if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
+               /* other bits are useless when BUSY */
+               rq->errors |= ERROR_RESET;
+       } else if (stat & ERR_STAT) {
+               /* err has different meaning on cdrom and tape */
+               if (err == ABRT_ERR) {
+                       if (drive->select.b.lba &&
+                           /* some newer drives don't support WIN_SPECIFY */
+                           hwif->INB(IDE_COMMAND_REG) == WIN_SPECIFY)
+                               return ide_stopped;
+               } else if ((err & BAD_CRC) == BAD_CRC) {
+                       /* UDMA crc error, just retry the operation */
+                       drive->crc_count++;
+               } 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;
+               }
+       }
+
+       if ((stat & DRQ_STAT) && rq_data_dir(rq) == READ && hwif->err_stops_fifo == 0)
+               try_to_flush_leftover_data(drive);
+
+       if (rq->errors >= ERROR_MAX || blk_noretry_request(rq)) {
+               ide_kill_rq(drive, rq);
+               return ide_stopped;
+       }
+
+       if (hwif->INB(IDE_STATUS_REG) & (BUSY_STAT|DRQ_STAT))
+               rq->errors |= ERROR_RESET;
+
+       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;
+}
+
+static ide_startstop_t ide_atapi_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
+{
+       ide_hwif_t *hwif = drive->hwif;
+
+       if (stat & BUSY_STAT || ((stat & WRERR_STAT) && !drive->nowerr)) {
+               /* other bits are useless when BUSY */
+               rq->errors |= ERROR_RESET;
+       } else {
+               /* add decoding error stuff */
+       }
+
+       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) {
+               ide_kill_rq(drive, rq);
+       } else {
+               if ((rq->errors & ERROR_RESET) == ERROR_RESET) {
+                       ++rq->errors;
+                       return ide_do_reset(drive);
+               }
+               ++rq->errors;
+       }
+
+       return ide_stopped;
+}
+
+ide_startstop_t
+__ide_error(ide_drive_t *drive, struct request *rq, u8 stat, u8 err)
+{
+       if (drive->media == ide_disk)
+               return ide_ata_error(drive, rq, stat, err);
+       return ide_atapi_error(drive, rq, stat, err);
+}
+
+EXPORT_SYMBOL_GPL(__ide_error);
 
 /**
  *     ide_error       -       handle an error on the IDE
@@ -472,80 +593,46 @@ EXPORT_SYMBOL(try_to_flush_leftover_data);
  
 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)
+
+       if ((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) {
+       if (!blk_fs_request(rq)) {
                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 */
+       if (rq->rq_disk) {
+               ide_driver_t *drv;
+
+               drv = *(ide_driver_t **)rq->rq_disk->private_data;
+               return drv->error(drive, rq, stat, err);
+       } else
+               return __ide_error(drive, rq, stat, err);
+}
+
+EXPORT_SYMBOL_GPL(ide_error);
+
+ide_startstop_t __ide_abort(ide_drive_t *drive, struct request *rq)
+{
+       if (drive->media != ide_disk)
                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;
-       }
+
+       ide_kill_rq(drive, rq);
+
        return ide_stopped;
 }
 
-EXPORT_SYMBOL(ide_error);
+EXPORT_SYMBOL_GPL(__ide_abort);
 
 /**
- *     ide_abort       -       abort pending IDE operatins
+ *     ide_abort       -       abort pending IDE operations
  *     @drive: drive the error occurred on
  *     @msg: message to report
  *
@@ -560,31 +647,26 @@ EXPORT_SYMBOL(ide_error);
  
 ide_startstop_t ide_abort(ide_drive_t *drive, const char *msg)
 {
-       ide_hwif_t *hwif;
        struct request *rq;
 
        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, BUSY_STAT, 0);
-               return ide_stopped;
-       }
-       if (rq->flags & REQ_DRIVE_TASKFILE) {
+       if (!blk_fs_request(rq)) {
                rq->errors = 1;
                ide_end_drive_cmd(drive, BUSY_STAT, 0);
                return ide_stopped;
        }
 
-       rq->errors |= ERROR_RESET;
-       DRIVER(drive)->end_request(drive, 0, 0);
-       return ide_stopped;
-}
+       if (rq->rq_disk) {
+               ide_driver_t *drv;
 
-EXPORT_SYMBOL(ide_abort);
+               drv = *(ide_driver_t **)rq->rq_disk->private_data;
+               return drv->abort(drive, rq);
+       } else
+               return __ide_abort(drive, rq);
+}
 
 /**
  *     ide_cmd         -       issue a simple drive command
@@ -597,7 +679,8 @@ EXPORT_SYMBOL(ide_abort);
  *     The drive must be selected beforehand.
  */
 
-void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect, ide_handler_t *handler)
+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)
@@ -607,19 +690,17 @@ void ide_cmd (ide_drive_t *drive, u8 cmd, u8 nsect, ide_handler_t *handler)
        ide_execute_command(drive, cmd, handler, WAIT_CMD, NULL);
 }
 
-EXPORT_SYMBOL(ide_cmd);
-
 /**
  *     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
+ *     We do any necessary data reading and then wait for the drive to
  *     go non busy. At that point we may read the error data and complete
  *     the request
  */
  
-ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
+static ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
 {
        struct request *rq = HWGROUP(drive)->rq;
        ide_hwif_t *hwif = HWIF(drive);
@@ -627,7 +708,7 @@ ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
        u8 stat = hwif->INB(IDE_STATUS_REG);
        int retries = 10;
 
-       local_irq_enable();
+       local_irq_enable_in_hardirq();
        if ((stat & DRQ_STAT) && args && args[3]) {
                u8 io_32bit = drive->io_32bit;
                drive->io_32bit = 0;
@@ -637,14 +718,71 @@ ide_startstop_t drive_cmd_intr (ide_drive_t *drive)
                        udelay(100);
        }
 
-       if (!OK_STAT(stat, READY_STAT, BAD_STAT) && DRIVER(drive) != NULL)
-               return DRIVER(drive)->error(drive, "drive_cmd", stat);
+       if (!OK_STAT(stat, READY_STAT, BAD_STAT))
+               return ide_error(drive, "drive_cmd", stat);
                /* calls ide_end_drive_cmd */
        ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));
        return ide_stopped;
 }
 
-EXPORT_SYMBOL(drive_cmd_intr);
+static void ide_init_specify_cmd(ide_drive_t *drive, ide_task_t *task)
+{
+       task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
+       task->tfRegister[IDE_SECTOR_OFFSET]  = drive->sect;
+       task->tfRegister[IDE_LCYL_OFFSET]    = drive->cyl;
+       task->tfRegister[IDE_HCYL_OFFSET]    = drive->cyl>>8;
+       task->tfRegister[IDE_SELECT_OFFSET]  = ((drive->head-1)|drive->select.all)&0xBF;
+       task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SPECIFY;
+
+       task->handler = &set_geometry_intr;
+}
+
+static void ide_init_restore_cmd(ide_drive_t *drive, ide_task_t *task)
+{
+       task->tfRegister[IDE_NSECTOR_OFFSET] = drive->sect;
+       task->tfRegister[IDE_COMMAND_OFFSET] = WIN_RESTORE;
+
+       task->handler = &recal_intr;
+}
+
+static void ide_init_setmult_cmd(ide_drive_t *drive, ide_task_t *task)
+{
+       task->tfRegister[IDE_NSECTOR_OFFSET] = drive->mult_req;
+       task->tfRegister[IDE_COMMAND_OFFSET] = WIN_SETMULT;
+
+       task->handler = &set_multmode_intr;
+}
+
+static ide_startstop_t ide_disk_special(ide_drive_t *drive)
+{
+       special_t *s = &drive->special;
+       ide_task_t args;
+
+       memset(&args, 0, sizeof(ide_task_t));
+       args.command_type = IDE_DRIVE_TASK_NO_DATA;
+
+       if (s->b.set_geometry) {
+               s->b.set_geometry = 0;
+               ide_init_specify_cmd(drive, &args);
+       } else if (s->b.recalibrate) {
+               s->b.recalibrate = 0;
+               ide_init_restore_cmd(drive, &args);
+       } else if (s->b.set_multmode) {
+               s->b.set_multmode = 0;
+               if (drive->mult_req > drive->id->max_multsect)
+                       drive->mult_req = drive->id->max_multsect;
+               ide_init_setmult_cmd(drive, &args);
+       } else if (s->all) {
+               int special = s->all;
+               s->all = 0;
+               printk(KERN_ERR "%s: bad special flag: 0x%02x\n", drive->name, special);
+               return ide_stopped;
+       }
+
+       do_rw_taskfile(drive, &args);
+
+       return ide_started;
+}
 
 /**
  *     do_special              -       issue some special commands
@@ -655,7 +793,7 @@ EXPORT_SYMBOL(drive_cmd_intr);
  *     back.
  */
 
-ide_startstop_t do_special (ide_drive_t *drive)
+static ide_startstop_t do_special (ide_drive_t *drive)
 {
        special_t *s = &drive->special;
 
@@ -667,16 +805,47 @@ ide_startstop_t do_special (ide_drive_t *drive)
                if (HWIF(drive)->tuneproc != NULL)
                        HWIF(drive)->tuneproc(drive, drive->tune_req);
                return ide_stopped;
+       } else {
+               if (drive->media == ide_disk)
+                       return ide_disk_special(drive);
+
+               s->all = 0;
+               drive->mult_req = 0;
+               return ide_stopped;
        }
-       else
-               return DRIVER(drive)->special(drive);
 }
 
-EXPORT_SYMBOL(do_special);
+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->cmd_type != REQ_TYPE_ATA_TASKFILE) {
+               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
+ *     @drive: the drive to issue the command on
  *     @rq: the request structure holding the command
  *
  *     execute_drive_cmd() issues a special drive command,  usually 
@@ -686,10 +855,11 @@ EXPORT_SYMBOL(do_special);
  *     all commands to finish. Don't do this as that is due to change
  */
 
-ide_startstop_t execute_drive_cmd (ide_drive_t *drive, struct request *rq)
+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) {
+       if (rq->cmd_type == REQ_TYPE_ATA_TASKFILE) {
                ide_task_t *args = rq->special;
  
                if (!args)
@@ -697,10 +867,21 @@ ide_startstop_t execute_drive_cmd (ide_drive_t *drive, struct request *rq)
 
                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) {
+       } else if (rq->cmd_type == REQ_TYPE_ATA_TASK) {
                u8 *args = rq->buffer;
                u8 sel;
  
@@ -726,7 +907,7 @@ ide_startstop_t execute_drive_cmd (ide_drive_t *drive, struct request *rq)
                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) {
+       } else if (rq->cmd_type == REQ_TYPE_ATA_CMD) {
                u8 *args = rq->buffer;
 
                if (!args)
@@ -765,7 +946,38 @@ done:
        return ide_stopped;
 }
 
-EXPORT_SYMBOL(execute_drive_cmd);
+static void ide_check_pm_state(ide_drive_t *drive, struct request *rq)
+{
+       struct request_pm_state *pm = rq->data;
+
+       if (blk_pm_suspend_request(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) &&
+                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), 100000);
+               if (rc)
+                       printk(KERN_WARNING "%s: drive not ready on wakeup\n", drive->name);
+       }
+}
 
 /**
  *     start_request   -       start of I/O and command issuing for IDE
@@ -778,12 +990,12 @@ EXPORT_SYMBOL(execute_drive_cmd);
  *     FIXME: this function needs a rename
  */
  
-ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
+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));
+       BUG_ON(!blk_rq_started(rq));
 
 #ifdef DEBUG
        printk("%s: start_request: current=0x%08lx\n",
@@ -795,13 +1007,6 @@ ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
                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)) {
@@ -812,33 +1017,8 @@ ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
        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);
-       }
+       if (blk_pm_request(rq))
+               ide_check_pm_state(drive, rq);
 
        SELECT_DRIVE(drive);
        if (ide_wait_stat(&startstop, drive, drive->ready_stat, BUSY_STAT|DRQ_STAT, WAIT_READY)) {
@@ -846,31 +1026,41 @@ ide_startstop_t start_request (ide_drive_t *drive, struct request *rq)
                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)
+               ide_driver_t *drv;
+
+               /*
+                * We reset the drive so we need to issue a SETFEATURES.
+                * Do it _after_ do_special() restored device parameters.
+                */
+               if (drive->current_speed == 0xff)
+                       ide_config_drive_speed(drive, drive->desired_speed);
+
+               if (rq->cmd_type == REQ_TYPE_ATA_CMD ||
+                   rq->cmd_type == REQ_TYPE_ATA_TASK ||
+                   rq->cmd_type == REQ_TYPE_ATA_TASKFILE)
                        return execute_drive_cmd(drive, rq);
                else if (blk_pm_request(rq)) {
+                       struct request_pm_state *pm = rq->data;
 #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);
+                       startstop = ide_start_power_step(drive, rq);
                        if (startstop == ide_stopped &&
-                           rq->pm->pm_step == ide_pm_state_completed)
+                           pm->pm_step == ide_pm_state_completed)
                                ide_complete_pm_request(drive, rq);
                        return startstop;
                }
-               return (DRIVER(drive)->do_request(drive, rq, block));
+
+               drv = *(ide_driver_t **)rq->rq_disk->private_data;
+               return drv->do_request(drive, rq, block);
        }
        return do_special(drive);
 kill_rq:
-       DRIVER(drive)->end_request(drive, 0, 0);
+       ide_kill_rq(drive, rq);
        return ide_stopped;
 }
 
-EXPORT_SYMBOL(start_request);
-
 /**
  *     ide_stall_queue         -       pause an IDE device
  *     @drive: drive to stall
@@ -885,6 +1075,7 @@ void ide_stall_queue (ide_drive_t *drive, unsigned long timeout)
        if (timeout > WAIT_WORSTCASE)
                timeout = WAIT_WORSTCASE;
        drive->sleep = timeout + jiffies;
+       drive->sleeping = 1;
 }
 
 EXPORT_SYMBOL(ide_stall_queue);
@@ -913,7 +1104,7 @@ repeat:
         * 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) {
+       if (blk_queue_flushing(drive->queue)) {
                /*
                 * small race where queue could get replugged during
                 * the 3-request flush cycle, just yank the plug since
@@ -924,18 +1115,18 @@ repeat:
        }
 
        do {
-               if ((!drive->sleep || time_after_eq(jiffies, drive->sleep))
+               if ((!drive->sleeping || 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))))
+                        || (drive->sleeping && (!best->sleeping || time_before(drive->sleep, best->sleep)))
+                        || (!best->sleeping && time_before(WAKEUP(drive), WAKEUP(best))))
                        {
                                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) {
+       if (best && best->nice1 && !best->sleeping && best != hwgroup->drive && best->service_time > WAIT_MIN_SLEEP) {
                long t = (signed long)(WAKEUP(best) - jiffies);
                if (t >= WAIT_MIN_SLEEP) {
                /*
@@ -944,10 +1135,9 @@ repeat:
                 */
                        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)))
+                               if (!drive->sleeping
+                                && time_before(jiffies - best->service_time, WAKEUP(drive))
+                                && time_before(WAKEUP(drive), jiffies + t))
                                {
                                        ide_stall_queue(best, min_t(long, t, 10 * WAIT_MIN_SLEEP));
                                        goto repeat;
@@ -993,15 +1183,13 @@ repeat:
  * the driver.  This makes the driver much more friendlier to shared IRQs
  * than previous designs, while remaining 100% (?) SMP safe and capable.
  */
-/* --BenH: made non-static as ide-pmac.c uses it to kick the hwgroup back
- *         into life on wakeup from machine sleep.
- */ 
-void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
+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;
+       int             loops = 0;
 
        /* for atari only: POSSIBLY BROKEN HERE(?) */
        ide_get_lock(ide_intr, hwgroup);
@@ -1013,14 +1201,17 @@ void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
                hwgroup->busy = 1;
                drive = choose_drive(hwgroup);
                if (drive == NULL) {
-                       unsigned long sleep = 0;
+                       int sleeping = 0;
+                       unsigned long sleep = 0; /* shut up, gcc */
                        hwgroup->rq = NULL;
                        drive = hwgroup->drive;
                        do {
-                               if (drive->sleep && (!sleep || 0 < (signed long)(sleep - drive->sleep)))
+                               if (drive->sleeping && (!sleeping || time_before(drive->sleep, sleep))) {
+                                       sleeping = 1;
                                        sleep = drive->sleep;
+                               }
                        } while ((drive = drive->next) != hwgroup->drive);
-                       if (sleep) {
+                       if (sleeping) {
                /*
                 * Take a short snooze, and then wake up this hwgroup again.
                 * This gives other hwgroups on the same a chance to
@@ -1051,6 +1242,7 @@ void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
                        /* no more work for this hwgroup (for now) */
                        return;
                }
+       again:
                hwif = HWIF(drive);
                if (hwgroup->hwif->sharing_irq &&
                    hwif != hwgroup->hwif &&
@@ -1060,7 +1252,7 @@ void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
                }
                hwgroup->hwif = hwif;
                hwgroup->drive = drive;
-               drive->sleep = 0;
+               drive->sleeping = 0;
                drive->service_start = jiffies;
 
                if (blk_queue_plugged(drive->queue)) {
@@ -1078,13 +1270,6 @@ void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
                        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
@@ -1097,8 +1282,14 @@ void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
                 * though. I hope that doesn't happen too much, hopefully not
                 * unless the subdriver triggers such a thing in its own PM
                 * state machine.
+                *
+                * We count how many times we loop here to make sure we service
+                * all drives in the hwgroup without looping for ever
                 */
-               if (drive->blocked && !blk_pm_request(rq) && !(rq->flags & REQ_PREEMPT)) {
+               if (drive->blocked && !blk_pm_request(rq) && !(rq->cmd_flags & REQ_PREEMPT)) {
+                       drive = drive->next ? drive->next : hwgroup->drive;
+                       if (loops++ < 4 && !blk_queue_plugged(drive->queue))
+                               goto again;
                        /* We clear busy, there should be no pending ATA command at this point. */
                        hwgroup->busy = 0;
                        break;
@@ -1114,22 +1305,20 @@ void ide_do_request (ide_hwgroup_t *hwgroup, int masked_irq)
                 * 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)
+               if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
                        disable_irq_nosync(hwif->irq);
                spin_unlock(&ide_lock);
-               local_irq_enable();
+               local_irq_enable_in_hardirq();
                        /* allow other IRQs while we start this request */
                startstop = start_request(drive, rq);
                spin_lock_irq(&ide_lock);
-               if (hwif->irq != masked_irq)
+               if (masked_irq != IDE_NO_IRQ && hwif->irq != masked_irq)
                        enable_irq(hwif->irq);
                if (startstop == ide_stopped)
                        hwgroup->busy = 0;
        }
 }
 
-EXPORT_SYMBOL(ide_do_request);
-
 /*
  * Passes the stuff to ide_do_request
  */
@@ -1158,7 +1347,7 @@ static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
        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",
+               ret = ide_error(drive, "dma timeout error",
                                                hwif->INB(IDE_STATUS_REG));
        } else {
                printk(KERN_WARNING "%s: DMA timeout retry\n", drive->name);
@@ -1179,15 +1368,22 @@ static ide_startstop_t ide_dma_timeout_retry(ide_drive_t *drive, int error)
         * make sure request is sane
         */
        rq = HWGROUP(drive)->rq;
+
+       if (!rq)
+               goto out;
+
        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;
-       if (rq->bio)
-               rq->buffer = NULL;
-
+       rq->buffer = bio_data(rq->bio);
+out:
        return ret;
 }
 
@@ -1265,7 +1461,7 @@ void ide_timer_expiry (unsigned long data)
                        /* local CPU only,
                         * as if we were handling an interrupt */
                        local_irq_disable();
-                       if (hwgroup->poll_timeout != 0) {
+                       if (hwgroup->polling) {
                                startstop = handler(drive);
                        } else if (drive_is_ready(drive)) {
                                if (drive->waiting_for_dma)
@@ -1278,7 +1474,7 @@ void ide_timer_expiry (unsigned long data)
                                        startstop = ide_dma_timeout_retry(drive, wait);
                                } else
                                        startstop =
-                                       DRIVER(drive)->error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
+                                       ide_error(drive, "irq timeout", hwif->INB(IDE_STATUS_REG));
                        }
                        drive->service_time = jiffies - drive->service_start;
                        spin_lock_irq(&ide_lock);
@@ -1291,8 +1487,6 @@ void ide_timer_expiry (unsigned long data)
        spin_unlock_irqrestore(&ide_lock, flags);
 }
 
-EXPORT_SYMBOL(ide_timer_expiry);
-
 /**
  *     unexpected_intr         -       handle an unexpected IDE interrupt
  *     @irq: interrupt line
@@ -1378,7 +1572,7 @@ static void unexpected_intr (int irq, ide_hwgroup_t *hwgroup)
  *     on the hwgroup and the process begins again.
  */
  
-irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
+irqreturn_t ide_intr (int irq, void *dev_id)
 {
        unsigned long flags;
        ide_hwgroup_t *hwgroup = (ide_hwgroup_t *)dev_id;
@@ -1395,8 +1589,7 @@ irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
                return IRQ_NONE;
        }
 
-       if ((handler = hwgroup->handler) == NULL ||
-           hwgroup->poll_timeout != 0) {
+       if ((handler = hwgroup->handler) == NULL || hwgroup->polling) {
                /*
                 * Not expecting an interrupt from this drive.
                 * That means this could be:
@@ -1463,8 +1656,19 @@ irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
        del_timer(&hwgroup->timer);
        spin_unlock(&ide_lock);
 
+       /* Some controllers might set DMA INTR no matter DMA or PIO;
+        * bmdma status might need to be cleared even for
+        * PIO interrupts to prevent spurious/lost irq.
+        */
+       if (hwif->ide_dma_clear_irq && !(drive->waiting_for_dma))
+               /* ide_dma_end() needs bmdma status for error checking.
+                * So, skip clearing bmdma status here and leave it
+                * to ide_dma_end() if this is dma interrupt.
+                */
+               hwif->ide_dma_clear_irq(drive);
+
        if (drive->unmask)
-               local_irq_enable();
+               local_irq_enable_in_hardirq();
        /* service this interrupt, may set handler for next interrupt */
        startstop = handler(drive);
        spin_lock_irq(&ide_lock);
@@ -1490,8 +1694,6 @@ irqreturn_t ide_intr (int irq, void *dev_id, struct pt_regs *regs)
        return IRQ_HANDLED;
 }
 
-EXPORT_SYMBOL(ide_intr);
-
 /**
  *     ide_init_drive_cmd      -       initialize a drive command request
  *     @rq: request object
@@ -1499,13 +1701,13 @@ EXPORT_SYMBOL(ide_intr);
  *     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.
+ *     nasty surprise.
  */
 
 void ide_init_drive_cmd (struct request *rq)
 {
        memset(rq, 0, sizeof(*rq));
-       rq->flags = REQ_DRIVE_CMD;
+       rq->cmd_type = REQ_TYPE_ATA_CMD;
        rq->ref_count = 1;
 }
 
@@ -1530,12 +1732,6 @@ EXPORT_SYMBOL(ide_init_drive_cmd);
  *     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
@@ -1546,26 +1742,11 @@ int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t actio
 {
        unsigned long flags;
        ide_hwgroup_t *hwgroup = HWGROUP(drive);
-       DECLARE_COMPLETION(wait);
+       DECLARE_COMPLETION_ONSTACK(wait);
        int where = ELEVATOR_INSERT_BACK, err;
        int must_wait = (action == ide_wait || action == ide_head_wait);
 
-#ifdef CONFIG_BLK_DEV_PDC4030
-       /*
-        *      FIXME: there should be a drive or hwif->special
-        *      handler that points here by default, not hacks
-        *      in the ide-io.c code
-        *
-        *      FIXME2: That code breaks power management if used with
-        *      this chipset, that really doesn't belong here !
-        */
-       if (HWIF(drive)->chipset == ide_pdc4030 && rq->buffer != NULL)
-               return -ENOSYS;  /* special drive cmds not supported */
-#endif
        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
@@ -1573,7 +1754,8 @@ int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t actio
         */
        if (must_wait) {
                rq->ref_count++;
-               rq->waiting = &wait;
+               rq->end_io_data = &wait;
+               rq->end_io = blk_end_sync_rq;
        }
 
        spin_lock_irqsave(&ide_lock, flags);
@@ -1581,7 +1763,7 @@ int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t actio
                hwgroup->rq = NULL;
        if (action == ide_preempt || action == ide_head_wait) {
                where = ELEVATOR_INSERT_FRONT;
-               rq->flags |= REQ_PREEMPT;
+               rq->cmd_flags |= REQ_PREEMPT;
        }
        __elv_add_request(drive->queue, rq, where, 0);
        ide_do_request(hwgroup, IDE_NO_IRQ);
@@ -1590,7 +1772,6 @@ int ide_do_drive_cmd (ide_drive_t *drive, struct request *rq, ide_action_t actio
        err = 0;
        if (must_wait) {
                wait_for_completion(&wait);
-               rq->waiting = NULL;
                if (rq->errors)
                        err = -EIO;