libata: fix boot panic with SATAPI devices on non-SFF HBAs
[safe/jmp/linux-2.6] / drivers / scsi / sd.c
index bb5b242..5fe7aae 100644 (file)
  *     than the level indicated above to trigger output.       
  */
 
-#include <linux/config.h>
 #include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/kernel.h>
-#include <linux/sched.h>
 #include <linux/mm.h>
 #include <linux/bio.h>
 #include <linux/genhd.h>
@@ -47,8 +45,8 @@
 #include <linux/init.h>
 #include <linux/blkdev.h>
 #include <linux/blkpg.h>
-#include <linux/kref.h>
 #include <linux/delay.h>
+#include <linux/mutex.h>
 #include <asm/uaccess.h>
 
 #include <scsi/scsi.h>
 #include <scsi/scsi_host.h>
 #include <scsi/scsi_ioctl.h>
 #include <scsi/scsicam.h>
+#include <scsi/sd.h>
 
 #include "scsi_logging.h"
 
-/*
- * More than enough for everybody ;)  The huge number of majors
- * is a leftover from 16bit dev_t days, we don't really need that
- * much numberspace.
- */
-#define SD_MAJORS      16
-
-/*
- * This is limited by the naming scheme enforced in sd_probe,
- * add another character to it if you really need more disks.
- */
-#define SD_MAX_DISKS   (((26 * 26) + 26 + 1) * 26)
-
-/*
- * Time out in seconds for disks and Magneto-opticals (which are slower).
- */
-#define SD_TIMEOUT             (30 * HZ)
-#define SD_MOD_TIMEOUT         (75 * HZ)
+MODULE_AUTHOR("Eric Youngdale");
+MODULE_DESCRIPTION("SCSI disk (sd) driver");
+MODULE_LICENSE("GPL");
 
-/*
- * Number of allowed retries
- */
-#define SD_MAX_RETRIES         5
-#define SD_PASSTHROUGH_RETRIES 1
-
-static void scsi_disk_release(struct kref *kref);
-
-struct scsi_disk {
-       struct scsi_driver *driver;     /* always &sd_template */
-       struct scsi_device *device;
-       struct kref     kref;
-       struct gendisk  *disk;
-       unsigned int    openers;        /* protected by BKL for now, yuck */
-       sector_t        capacity;       /* size in 512-byte sectors */
-       u32             index;
-       u8              media_present;
-       u8              write_prot;
-       unsigned        WCE : 1;        /* state of disk WCE bit */
-       unsigned        RCD : 1;        /* state of disk RCD bit, unused */
-};
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
+MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
+MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
+MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
+MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
+
+static int  sd_revalidate_disk(struct gendisk *);
+static int  sd_probe(struct device *);
+static int  sd_remove(struct device *);
+static void sd_shutdown(struct device *);
+static int sd_suspend(struct device *, pm_message_t state);
+static int sd_resume(struct device *);
+static void sd_rescan(struct device *);
+static int sd_done(struct scsi_cmnd *);
+static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
+static void scsi_disk_release(struct class_device *cdev);
+static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
+static void sd_print_result(struct scsi_disk *, int);
 
 static DEFINE_IDR(sd_index_idr);
 static DEFINE_SPINLOCK(sd_index_lock);
@@ -110,21 +105,142 @@ static DEFINE_SPINLOCK(sd_index_lock);
 /* This semaphore is used to mediate the 0->1 reference get in the
  * face of object destruction (i.e. we can't allow a get on an
  * object after last put) */
-static DECLARE_MUTEX(sd_ref_sem);
+static DEFINE_MUTEX(sd_ref_mutex);
 
-static int sd_revalidate_disk(struct gendisk *disk);
-static void sd_rw_intr(struct scsi_cmnd * SCpnt);
+static const char *sd_cache_types[] = {
+       "write through", "none", "write back",
+       "write back, no read (daft)"
+};
 
-static int sd_probe(struct device *);
-static int sd_remove(struct device *);
-static void sd_shutdown(struct device *dev);
-static void sd_rescan(struct device *);
-static int sd_init_command(struct scsi_cmnd *);
-static int sd_issue_flush(struct device *, sector_t *);
-static void sd_end_flush(request_queue_t *, struct request *);
-static int sd_prepare_flush(request_queue_t *, struct request *);
-static void sd_read_capacity(struct scsi_disk *sdkp, char *diskname,
-                            unsigned char *buffer);
+static ssize_t sd_store_cache_type(struct class_device *cdev, const char *buf,
+                                  size_t count)
+{
+       int i, ct = -1, rcd, wce, sp;
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
+       struct scsi_device *sdp = sdkp->device;
+       char buffer[64];
+       char *buffer_data;
+       struct scsi_mode_data data;
+       struct scsi_sense_hdr sshdr;
+       int len;
+
+       if (sdp->type != TYPE_DISK)
+               /* no cache control on RBC devices; theoretically they
+                * can do it, but there's probably so many exceptions
+                * it's not worth the risk */
+               return -EINVAL;
+
+       for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
+               const int len = strlen(sd_cache_types[i]);
+               if (strncmp(sd_cache_types[i], buf, len) == 0 &&
+                   buf[len] == '\n') {
+                       ct = i;
+                       break;
+               }
+       }
+       if (ct < 0)
+               return -EINVAL;
+       rcd = ct & 0x01 ? 1 : 0;
+       wce = ct & 0x02 ? 1 : 0;
+       if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
+                           SD_MAX_RETRIES, &data, NULL))
+               return -EINVAL;
+       len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
+                 data.block_descriptor_length);
+       buffer_data = buffer + data.header_length +
+               data.block_descriptor_length;
+       buffer_data[2] &= ~0x05;
+       buffer_data[2] |= wce << 2 | rcd;
+       sp = buffer_data[0] & 0x80 ? 1 : 0;
+
+       if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
+                            SD_MAX_RETRIES, &data, &sshdr)) {
+               if (scsi_sense_valid(&sshdr))
+                       sd_print_sense_hdr(sdkp, &sshdr);
+               return -EINVAL;
+       }
+       sd_revalidate_disk(sdkp->disk);
+       return count;
+}
+
+static ssize_t sd_store_manage_start_stop(struct class_device *cdev,
+                                         const char *buf, size_t count)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
+       struct scsi_device *sdp = sdkp->device;
+
+       if (!capable(CAP_SYS_ADMIN))
+               return -EACCES;
+
+       sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
+
+       return count;
+}
+
+static ssize_t sd_store_allow_restart(struct class_device *cdev, const char *buf,
+                                     size_t count)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
+       struct scsi_device *sdp = sdkp->device;
+
+       if (!capable(CAP_SYS_ADMIN))
+               return -EACCES;
+
+       if (sdp->type != TYPE_DISK)
+               return -EINVAL;
+
+       sdp->allow_restart = simple_strtoul(buf, NULL, 10);
+
+       return count;
+}
+
+static ssize_t sd_show_cache_type(struct class_device *cdev, char *buf)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
+       int ct = sdkp->RCD + 2*sdkp->WCE;
+
+       return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
+}
+
+static ssize_t sd_show_fua(struct class_device *cdev, char *buf)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
+
+       return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
+}
+
+static ssize_t sd_show_manage_start_stop(struct class_device *cdev, char *buf)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
+       struct scsi_device *sdp = sdkp->device;
+
+       return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
+}
+
+static ssize_t sd_show_allow_restart(struct class_device *cdev, char *buf)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
+
+       return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
+}
+
+static struct class_device_attribute sd_disk_attrs[] = {
+       __ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
+              sd_store_cache_type),
+       __ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
+       __ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
+              sd_store_allow_restart),
+       __ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
+              sd_store_manage_start_stop),
+       __ATTR_NULL,
+};
+
+static struct class sd_disk_class = {
+       .name           = "scsi_disk",
+       .owner          = THIS_MODULE,
+       .release        = scsi_disk_release,
+       .class_dev_attrs = sd_disk_attrs,
+};
 
 static struct scsi_driver sd_template = {
        .owner                  = THIS_MODULE,
@@ -132,13 +248,12 @@ static struct scsi_driver sd_template = {
                .name           = "sd",
                .probe          = sd_probe,
                .remove         = sd_remove,
+               .suspend        = sd_suspend,
+               .resume         = sd_resume,
                .shutdown       = sd_shutdown,
        },
        .rescan                 = sd_rescan,
-       .init_command           = sd_init_command,
-       .issue_flush            = sd_issue_flush,
-       .prepare_flush          = sd_prepare_flush,
-       .end_flush              = sd_end_flush,
+       .done                   = sd_done,
 };
 
 /*
@@ -170,8 +285,6 @@ static int sd_major(int major_idx)
        }
 }
 
-#define to_scsi_disk(obj) container_of(obj,struct scsi_disk,kref)
-
 static inline struct scsi_disk *scsi_disk(struct gendisk *disk)
 {
        return container_of(disk->private_data, struct scsi_disk, driver);
@@ -184,7 +297,7 @@ static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
        if (disk->private_data) {
                sdkp = scsi_disk(disk);
                if (scsi_device_get(sdkp->device) == 0)
-                       kref_get(&sdkp->kref);
+                       class_device_get(&sdkp->cdev);
                else
                        sdkp = NULL;
        }
@@ -195,9 +308,9 @@ static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
 {
        struct scsi_disk *sdkp;
 
-       down(&sd_ref_sem);
+       mutex_lock(&sd_ref_mutex);
        sdkp = __scsi_disk_get(disk);
-       up(&sd_ref_sem);
+       mutex_unlock(&sd_ref_mutex);
        return sdkp;
 }
 
@@ -205,11 +318,11 @@ static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
 {
        struct scsi_disk *sdkp;
 
-       down(&sd_ref_sem);
+       mutex_lock(&sd_ref_mutex);
        sdkp = dev_get_drvdata(dev);
        if (sdkp)
                sdkp = __scsi_disk_get(sdkp->disk);
-       up(&sd_ref_sem);
+       mutex_unlock(&sd_ref_mutex);
        return sdkp;
 }
 
@@ -217,10 +330,10 @@ static void scsi_disk_put(struct scsi_disk *sdkp)
 {
        struct scsi_device *sdev = sdkp->device;
 
-       down(&sd_ref_sem);
-       kref_put(&sdkp->kref, scsi_disk_release);
+       mutex_lock(&sd_ref_mutex);
+       class_device_put(&sdkp->cdev);
        scsi_device_put(sdev);
-       up(&sd_ref_sem);
+       mutex_unlock(&sd_ref_mutex);
 }
 
 /**
@@ -231,61 +344,46 @@ static void scsi_disk_put(struct scsi_disk *sdkp)
  *
  *     Returns 1 if successful and 0 if error (or cannot be done now).
  **/
-static int sd_init_command(struct scsi_cmnd * SCpnt)
+static int sd_prep_fn(struct request_queue *q, struct request *rq)
 {
-       unsigned int this_count, timeout;
-       struct gendisk *disk;
-       sector_t block;
-       struct scsi_device *sdp = SCpnt->device;
-       struct request *rq = SCpnt->request;
-
-       timeout = sdp->timeout;
-
-       /*
-        * SG_IO from block layer already setup, just copy cdb basically
-        */
-       if (blk_pc_request(rq)) {
-               if (sizeof(rq->cmd) > sizeof(SCpnt->cmnd))
-                       return 0;
-
-               memcpy(SCpnt->cmnd, rq->cmd, sizeof(SCpnt->cmnd));
-               SCpnt->cmd_len = rq->cmd_len;
-               if (rq_data_dir(rq) == WRITE)
-                       SCpnt->sc_data_direction = DMA_TO_DEVICE;
-               else if (rq->data_len)
-                       SCpnt->sc_data_direction = DMA_FROM_DEVICE;
-               else
-                       SCpnt->sc_data_direction = DMA_NONE;
-
-               this_count = rq->data_len;
-               if (rq->timeout)
-                       timeout = rq->timeout;
-
-               SCpnt->transfersize = rq->data_len;
-               SCpnt->allowed = SD_PASSTHROUGH_RETRIES;
-               goto queue;
+       struct scsi_cmnd *SCpnt;
+       struct scsi_device *sdp = q->queuedata;
+       struct gendisk *disk = rq->rq_disk;
+       sector_t block = rq->sector;
+       unsigned int this_count = rq->nr_sectors;
+       unsigned int timeout = sdp->timeout;
+       int ret;
+
+       if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
+               ret = scsi_setup_blk_pc_cmnd(sdp, rq);
+               goto out;
+       } else if (rq->cmd_type != REQ_TYPE_FS) {
+               ret = BLKPREP_KILL;
+               goto out;
        }
+       ret = scsi_setup_fs_cmnd(sdp, rq);
+       if (ret != BLKPREP_OK)
+               goto out;
+       SCpnt = rq->special;
 
-       /*
-        * we only do REQ_CMD and REQ_BLOCK_PC
-        */
-       if (!blk_fs_request(rq))
-               return 0;
-
-       disk = rq->rq_disk;
-       block = rq->sector;
-       this_count = SCpnt->request_bufflen >> 9;
+       /* from here on until we're complete, any goto out
+        * is used for a killable error condition */
+       ret = BLKPREP_KILL;
 
-       SCSI_LOG_HLQUEUE(1, printk("sd_init_command: disk=%s, block=%llu, "
-                           "count=%d\n", disk->disk_name,
-                        (unsigned long long)block, this_count));
+       SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
+                                       "sd_init_command: block=%llu, "
+                                       "count=%d\n",
+                                       (unsigned long long)block,
+                                       this_count));
 
        if (!sdp || !scsi_device_online(sdp) ||
            block + rq->nr_sectors > get_capacity(disk)) {
-               SCSI_LOG_HLQUEUE(2, printk("Finishing %ld sectors\n", 
-                                rq->nr_sectors));
-               SCSI_LOG_HLQUEUE(2, printk("Retry with 0x%p\n", SCpnt));
-               return 0;
+               SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
+                                               "Finishing %ld sectors\n",
+                                               rq->nr_sectors));
+               SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
+                                               "Retry with 0x%p\n", SCpnt));
+               goto out;
        }
 
        if (sdp->changed) {
@@ -294,10 +392,20 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
                 * the changed bit has been reset
                 */
                /* printk("SCSI disk has been changed. Prohibiting further I/O.\n"); */
-               return 0;
+               goto out;
        }
-       SCSI_LOG_HLQUEUE(2, printk("%s : block=%llu\n",
-                                  disk->disk_name, (unsigned long long)block));
+
+       /*
+        * Some devices (some sdcards for one) don't like it if the
+        * last sector gets read in a larger then 1 sector read.
+        */
+       if (unlikely(sdp->last_sector_bug &&
+           rq->nr_sectors > sdp->sector_size / 512 &&
+           block + this_count == get_capacity(disk)))
+               this_count -= sdp->sector_size / 512;
+
+       SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
+                                       (unsigned long long)block));
 
        /*
         * If we have a 1K hardware sectorsize, prevent access to single
@@ -312,8 +420,9 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
         */
        if (sdp->sector_size == 1024) {
                if ((block & 1) || (rq->nr_sectors & 1)) {
-                       printk(KERN_ERR "sd: Bad block number requested");
-                       return 0;
+                       scmd_printk(KERN_ERR, SCpnt,
+                                   "Bad block number requested\n");
+                       goto out;
                } else {
                        block = block >> 1;
                        this_count = this_count >> 1;
@@ -321,8 +430,9 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
        }
        if (sdp->sector_size == 2048) {
                if ((block & 3) || (rq->nr_sectors & 3)) {
-                       printk(KERN_ERR "sd: Bad block number requested");
-                       return 0;
+                       scmd_printk(KERN_ERR, SCpnt,
+                                   "Bad block number requested\n");
+                       goto out;
                } else {
                        block = block >> 2;
                        this_count = this_count >> 2;
@@ -330,8 +440,9 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
        }
        if (sdp->sector_size == 4096) {
                if ((block & 7) || (rq->nr_sectors & 7)) {
-                       printk(KERN_ERR "sd: Bad block number requested");
-                       return 0;
+                       scmd_printk(KERN_ERR, SCpnt,
+                                   "Bad block number requested\n");
+                       goto out;
                } else {
                        block = block >> 3;
                        this_count = this_count >> 3;
@@ -339,7 +450,7 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
        }
        if (rq_data_dir(rq) == WRITE) {
                if (!sdp->writeable) {
-                       return 0;
+                       goto out;
                }
                SCpnt->cmnd[0] = WRITE_6;
                SCpnt->sc_data_direction = DMA_TO_DEVICE;
@@ -347,19 +458,21 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
                SCpnt->cmnd[0] = READ_6;
                SCpnt->sc_data_direction = DMA_FROM_DEVICE;
        } else {
-               printk(KERN_ERR "sd: Unknown command %lx\n", rq->flags);
-/* overkill    panic("Unknown sd command %lx\n", rq->flags); */
-               return 0;
+               scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
+               goto out;
        }
 
-       SCSI_LOG_HLQUEUE(2, printk("%s : %s %d/%ld 512 byte blocks.\n", 
-               disk->disk_name, (rq_data_dir(rq) == WRITE) ? 
-               "writing" : "reading", this_count, rq->nr_sectors));
+       SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
+                                       "%s %d/%ld 512 byte blocks.\n",
+                                       (rq_data_dir(rq) == WRITE) ?
+                                       "writing" : "reading", this_count,
+                                       rq->nr_sectors));
 
        SCpnt->cmnd[1] = 0;
        
        if (block > 0xffffffff) {
                SCpnt->cmnd[0] += READ_16 - READ_6;
+               SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
                SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
                SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
                SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
@@ -379,6 +492,7 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
                        this_count = 0xffff;
 
                SCpnt->cmnd[0] += READ_10 - READ_6;
+               SCpnt->cmnd[1] |= blk_fua_rq(rq) ? 0x8 : 0;
                SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
                SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
                SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
@@ -387,14 +501,25 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
                SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
                SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
        } else {
+               if (unlikely(blk_fua_rq(rq))) {
+                       /*
+                        * This happens only if this drive failed
+                        * 10byte rw command with ILLEGAL_REQUEST
+                        * during operation and thus turned off
+                        * use_10_for_rw.
+                        */
+                       scmd_printk(KERN_ERR, SCpnt,
+                                   "FUA write on READ/WRITE(6) drive\n");
+                       goto out;
+               }
+
                SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
                SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
                SCpnt->cmnd[3] = (unsigned char) block & 0xff;
                SCpnt->cmnd[4] = (unsigned char) this_count;
                SCpnt->cmnd[5] = 0;
        }
-       SCpnt->request_bufflen = SCpnt->bufflen =
-                       this_count * sdp->sector_size;
+       SCpnt->sdb.length = this_count * sdp->sector_size;
 
        /*
         * We shouldn't disconnect in the middle of a sector, so with a dumb
@@ -404,21 +529,15 @@ static int sd_init_command(struct scsi_cmnd * SCpnt)
        SCpnt->transfersize = sdp->sector_size;
        SCpnt->underflow = this_count << 9;
        SCpnt->allowed = SD_MAX_RETRIES;
-
-queue:
        SCpnt->timeout_per_command = timeout;
 
        /*
-        * This is the completion routine we use.  This is matched in terms
-        * of capability to this function.
-        */
-       SCpnt->done = sd_rw_intr;
-
-       /*
         * This indicates that the command is ready from our end to be
         * queued.
         */
-       return 1;
+       ret = BLKPREP_OK;
+ out:
+       return scsi_prep_return(q, rq, ret);
 }
 
 /**
@@ -445,7 +564,7 @@ static int sd_open(struct inode *inode, struct file *filp)
                return -ENXIO;
 
 
-       SCSI_LOG_HLQUEUE(3, printk("sd_open: disk=%s\n", disk->disk_name));
+       SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
 
        sdev = sdkp->device;
 
@@ -515,7 +634,7 @@ static int sd_release(struct inode *inode, struct file *filp)
        struct scsi_disk *sdkp = scsi_disk(disk);
        struct scsi_device *sdev = sdkp->device;
 
-       SCSI_LOG_HLQUEUE(3, printk("sd_release: disk=%s\n", disk->disk_name));
+       SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
 
        if (!--sdkp->openers && sdev->removable) {
                if (scsi_block_when_processing_errors(sdev))
@@ -530,7 +649,7 @@ static int sd_release(struct inode *inode, struct file *filp)
        return 0;
 }
 
-static int sd_hdio_getgeo(struct block_device *bdev, struct hd_geometry __user *loc)
+static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 {
        struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
        struct scsi_device *sdp = sdkp->device;
@@ -548,15 +667,9 @@ static int sd_hdio_getgeo(struct block_device *bdev, struct hd_geometry __user *
        else
                scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
 
-       if (put_user(diskinfo[0], &loc->heads))
-               return -EFAULT;
-       if (put_user(diskinfo[1], &loc->sectors))
-               return -EFAULT;
-       if (put_user(diskinfo[2], &loc->cylinders))
-               return -EFAULT;
-       if (put_user((unsigned)get_start_sect(bdev),
-                    (unsigned long __user *)&loc->start))
-               return -EFAULT;
+       geo->heads = diskinfo[0];
+       geo->sectors = diskinfo[1];
+       geo->cylinders = diskinfo[2];
        return 0;
 }
 
@@ -572,7 +685,7 @@ static int sd_hdio_getgeo(struct block_device *bdev, struct hd_geometry __user *
  *     success as well). Returns a negated errno value in case of error.
  *
  *     Note: most ioctls are forward onto the block subsystem or further
- *     down in the scsi subsytem.
+ *     down in the scsi subsystem.
  **/
 static int sd_ioctl(struct inode * inode, struct file * filp, 
                    unsigned int cmd, unsigned long arg)
@@ -596,12 +709,6 @@ static int sd_ioctl(struct inode * inode, struct file * filp,
        if (!scsi_block_when_processing_errors(sdp) || !error)
                return error;
 
-       if (cmd == HDIO_GETGEO) {
-               if (!arg)
-                       return -EINVAL;
-               return sd_hdio_getgeo(bdev, p);
-       }
-
        /*
         * Send SCSI addressing ioctls directly to mid level, send other
         * ioctls to block level and then onto mid level if they can't be
@@ -612,7 +719,7 @@ static int sd_ioctl(struct inode * inode, struct file * filp,
                case SCSI_IOCTL_GET_BUS_NUMBER:
                        return scsi_ioctl(sdp, cmd, p);
                default:
-                       error = scsi_cmd_ioctl(filp, disk, cmd, p);
+                       error = scsi_cmd_ioctl(filp, disk->queue, disk, cmd, p);
                        if (error != -ENOTTY)
                                return error;
        }
@@ -638,10 +745,10 @@ static int sd_media_changed(struct gendisk *disk)
 {
        struct scsi_disk *sdkp = scsi_disk(disk);
        struct scsi_device *sdp = sdkp->device;
+       struct scsi_sense_hdr *sshdr = NULL;
        int retval;
 
-       SCSI_LOG_HLQUEUE(3, printk("sd_media_changed: disk=%s\n",
-                                               disk->disk_name));
+       SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_media_changed\n"));
 
        if (!sdp->removable)
                return 0;
@@ -652,8 +759,11 @@ static int sd_media_changed(struct gendisk *disk)
         * can deal with it then.  It is only because of unrecoverable errors
         * that we would ever take a device offline in the first place.
         */
-       if (!scsi_device_online(sdp))
-               goto not_present;
+       if (!scsi_device_online(sdp)) {
+               set_media_not_present(sdkp);
+               retval = 1;
+               goto out;
+       }
 
        /*
         * Using TEST_UNIT_READY enables differentiation between drive with
@@ -665,8 +775,12 @@ static int sd_media_changed(struct gendisk *disk)
         * sd_revalidate() is called.
         */
        retval = -ENODEV;
-       if (scsi_block_when_processing_errors(sdp))
-               retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES);
+
+       if (scsi_block_when_processing_errors(sdp)) {
+               sshdr  = kzalloc(sizeof(*sshdr), GFP_KERNEL);
+               retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
+                                             sshdr);
+       }
 
        /*
         * Unable to test, unit probably not ready.   This usually
@@ -674,8 +788,13 @@ static int sd_media_changed(struct gendisk *disk)
         * and we will figure it out later once the drive is
         * available again.
         */
-       if (retval)
-                goto not_present;
+       if (retval || (scsi_sense_valid(sshdr) &&
+                      /* 0x3a is medium not present */
+                      sshdr->asc == 0x3a)) {
+               set_media_not_present(sdkp);
+               retval = 1;
+               goto out;
+       }
 
        /*
         * For removable scsi disk we have to recognise the presence
@@ -686,17 +805,18 @@ static int sd_media_changed(struct gendisk *disk)
 
        retval = sdp->changed;
        sdp->changed = 0;
-
+out:
+       if (retval != sdkp->previous_state)
+               sdev_evt_send_simple(sdp, SDEV_EVT_MEDIA_CHANGE, GFP_KERNEL);
+       sdkp->previous_state = retval;
+       kfree(sshdr);
        return retval;
-
-not_present:
-       set_media_not_present(sdkp);
-       return 1;
 }
 
-static int sd_sync_cache(struct scsi_device *sdp)
+static int sd_sync_cache(struct scsi_disk *sdkp)
 {
        int retries, res;
+       struct scsi_device *sdp = sdkp->device;
        struct scsi_sense_hdr sshdr;
 
        if (!scsi_device_online(sdp))
@@ -717,72 +837,24 @@ static int sd_sync_cache(struct scsi_device *sdp)
                        break;
        }
 
-       if (res) {              printk(KERN_WARNING "FAILED\n  status = %x, message = %02x, "
-                                   "host = %d, driver = %02x\n  ",
-                                   status_byte(res), msg_byte(res),
-                                   host_byte(res), driver_byte(res));
-                       if (driver_byte(res) & DRIVER_SENSE)
-                               scsi_print_sense_hdr("sd", &sshdr);
+       if (res) {
+               sd_print_result(sdkp, res);
+               if (driver_byte(res) & DRIVER_SENSE)
+                       sd_print_sense_hdr(sdkp, &sshdr);
        }
 
-       return res;
-}
-
-static int sd_issue_flush(struct device *dev, sector_t *error_sector)
-{
-       int ret = 0;
-       struct scsi_device *sdp = to_scsi_device(dev);
-       struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
-
-       if (!sdkp)
-               return -ENODEV;
-
-       if (sdkp->WCE)
-               ret = sd_sync_cache(sdp);
-       scsi_disk_put(sdkp);
-       return ret;
+       if (res)
+               return -EIO;
+       return 0;
 }
 
-static void sd_end_flush(request_queue_t *q, struct request *flush_rq)
+static void sd_prepare_flush(struct request_queue *q, struct request *rq)
 {
-       struct request *rq = flush_rq->end_io_data;
-       struct scsi_cmnd *cmd = rq->special;
-       unsigned int bytes = rq->hard_nr_sectors << 9;
-
-       if (!flush_rq->errors) {
-               spin_unlock(q->queue_lock);
-               scsi_io_completion(cmd, bytes, 0);
-               spin_lock(q->queue_lock);
-       } else if (blk_barrier_postflush(rq)) {
-               spin_unlock(q->queue_lock);
-               scsi_io_completion(cmd, 0, bytes);
-               spin_lock(q->queue_lock);
-       } else {
-               /*
-                * force journal abort of barriers
-                */
-               end_that_request_first(rq, -EOPNOTSUPP, rq->hard_nr_sectors);
-               end_that_request_last(rq);
-       }
-}
-
-static int sd_prepare_flush(request_queue_t *q, struct request *rq)
-{
-       struct scsi_device *sdev = q->queuedata;
-       struct scsi_disk *sdkp = scsi_disk_get_from_dev(&sdev->sdev_gendev);
-       int ret = 0;
-
-       if (sdkp) {
-               if (sdkp->WCE) {
-                       memset(rq->cmd, 0, sizeof(rq->cmd));
-                       rq->flags |= REQ_BLOCK_PC | REQ_SOFTBARRIER;
-                       rq->timeout = SD_TIMEOUT;
-                       rq->cmd[0] = SYNCHRONIZE_CACHE;
-                       ret = 1;
-               }
-               scsi_disk_put(sdkp);
-       }
-       return ret;
+       memset(rq->cmd, 0, sizeof(rq->cmd));
+       rq->cmd_type = REQ_TYPE_BLOCK_PC;
+       rq->timeout = SD_TIMEOUT;
+       rq->cmd[0] = SYNCHRONIZE_CACHE;
+       rq->cmd_len = 10;
 }
 
 static void sd_rescan(struct device *dev)
@@ -803,7 +875,7 @@ static void sd_rescan(struct device *dev)
  */
 static long sd_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
-       struct block_device *bdev = file->f_dentry->d_inode->i_bdev;
+       struct block_device *bdev = file->f_path.dentry->d_inode->i_bdev;
        struct gendisk *disk = bdev->bd_disk;
        struct scsi_device *sdev = scsi_disk(disk)->device;
 
@@ -836,6 +908,7 @@ static struct block_device_operations sd_fops = {
        .open                   = sd_open,
        .release                = sd_release,
        .ioctl                  = sd_ioctl,
+       .getgeo                 = sd_getgeo,
 #ifdef CONFIG_COMPAT
        .compat_ioctl           = sd_compat_ioctl,
 #endif
@@ -844,20 +917,20 @@ static struct block_device_operations sd_fops = {
 };
 
 /**
- *     sd_rw_intr - bottom half handler: called when the lower level
+ *     sd_done - bottom half handler: called when the lower level
  *     driver has completed (successfully or otherwise) a scsi command.
  *     @SCpnt: mid-level's per command structure.
  *
  *     Note: potentially run from within an ISR. Must not block.
  **/
-static void sd_rw_intr(struct scsi_cmnd * SCpnt)
+static int sd_done(struct scsi_cmnd *SCpnt)
 {
        int result = SCpnt->result;
-       int this_count = SCpnt->bufflen;
-       int good_bytes = (result == 0 ? this_count : 0);
-       sector_t block_sectors = 1;
-       u64 first_err_block;
-       sector_t error_sector;
+       unsigned int xfer_size = scsi_bufflen(SCpnt);
+       unsigned int good_bytes = result ? 0 : xfer_size;
+       u64 start_lba = SCpnt->request->sector;
+       u64 end_lba = SCpnt->request->sector + (xfer_size / 512);
+       u64 bad_lba;
        struct scsi_sense_hdr sshdr;
        int sense_valid = 0;
        int sense_deferred = 0;
@@ -868,107 +941,80 @@ static void sd_rw_intr(struct scsi_cmnd * SCpnt)
                if (sense_valid)
                        sense_deferred = scsi_sense_is_deferred(&sshdr);
        }
-
 #ifdef CONFIG_SCSI_LOGGING
-       SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: %s: res=0x%x\n", 
-                               SCpnt->request->rq_disk->disk_name, result));
+       SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
        if (sense_valid) {
-               SCSI_LOG_HLCOMPLETE(1, printk("sd_rw_intr: sb[respc,sk,asc,"
-                               "ascq]=%x,%x,%x,%x\n", sshdr.response_code,
-                               sshdr.sense_key, sshdr.asc, sshdr.ascq));
+               SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
+                                                  "sd_done: sb[respc,sk,asc,"
+                                                  "ascq]=%x,%x,%x,%x\n",
+                                                  sshdr.response_code,
+                                                  sshdr.sense_key, sshdr.asc,
+                                                  sshdr.ascq));
        }
 #endif
-       /*
-          Handle MEDIUM ERRORs that indicate partial success.  Since this is a
-          relatively rare error condition, no care is taken to avoid
-          unnecessary additional work such as memcpy's that could be avoided.
-        */
-
-       /* 
-        * If SG_IO from block layer then set good_bytes to stop retries;
-        * else if errors, check them, and if necessary prepare for
-        * (partial) retries.
-        */
-       if (blk_pc_request(SCpnt->request))
-               good_bytes = this_count;
-       else if (driver_byte(result) != 0 &&
-                sense_valid && !sense_deferred) {
-               switch (sshdr.sense_key) {
-               case MEDIUM_ERROR:
-                       if (!blk_fs_request(SCpnt->request))
-                               break;
-                       info_valid = scsi_get_sense_info_fld(
-                               SCpnt->sense_buffer, SCSI_SENSE_BUFFERSIZE,
-                               &first_err_block);
-                       /*
-                        * May want to warn and skip if following cast results
-                        * in actual truncation (if sector_t < 64 bits)
-                        */
-                       error_sector = (sector_t)first_err_block;
-                       if (SCpnt->request->bio != NULL)
-                               block_sectors = bio_sectors(SCpnt->request->bio);
-                       switch (SCpnt->device->sector_size) {
-                       case 1024:
-                               error_sector <<= 1;
-                               if (block_sectors < 2)
-                                       block_sectors = 2;
-                               break;
-                       case 2048:
-                               error_sector <<= 2;
-                               if (block_sectors < 4)
-                                       block_sectors = 4;
-                               break;
-                       case 4096:
-                               error_sector <<=3;
-                               if (block_sectors < 8)
-                                       block_sectors = 8;
-                               break;
-                       case 256:
-                               error_sector >>= 1;
-                               break;
-                       default:
-                               break;
-                       }
+       if (driver_byte(result) != DRIVER_SENSE &&
+           (!sense_valid || sense_deferred))
+               goto out;
 
-                       error_sector &= ~(block_sectors - 1);
-                       good_bytes = (error_sector - SCpnt->request->sector) << 9;
-                       if (good_bytes < 0 || good_bytes >= this_count)
-                               good_bytes = 0;
-                       break;
+       switch (sshdr.sense_key) {
+       case HARDWARE_ERROR:
+       case MEDIUM_ERROR:
+               if (!blk_fs_request(SCpnt->request))
+                       goto out;
+               info_valid = scsi_get_sense_info_fld(SCpnt->sense_buffer,
+                                                    SCSI_SENSE_BUFFERSIZE,
+                                                    &bad_lba);
+               if (!info_valid)
+                       goto out;
+               if (xfer_size <= SCpnt->device->sector_size)
+                       goto out;
+               if (SCpnt->device->sector_size < 512) {
+                       /* only legitimate sector_size here is 256 */
+                       start_lba <<= 1;
+                       end_lba <<= 1;
+               } else {
+                       /* be careful ... don't want any overflows */
+                       u64 factor = SCpnt->device->sector_size / 512;
+                       do_div(start_lba, factor);
+                       do_div(end_lba, factor);
+               }
 
-               case RECOVERED_ERROR: /* an error occurred, but it recovered */
-               case NO_SENSE: /* LLDD got sense data */
-                       /*
-                        * Inform the user, but make sure that it's not treated
-                        * as a hard error.
+               if (bad_lba < start_lba  || bad_lba >= end_lba)
+                       /* the bad lba was reported incorrectly, we have
+                        * no idea where the error is
                         */
-                       scsi_print_sense("sd", SCpnt);
-                       SCpnt->result = 0;
-                       memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
-                       good_bytes = this_count;
-                       break;
+                       goto out;
 
-               case ILLEGAL_REQUEST:
-                       if (SCpnt->device->use_10_for_rw &&
-                           (SCpnt->cmnd[0] == READ_10 ||
-                            SCpnt->cmnd[0] == WRITE_10))
-                               SCpnt->device->use_10_for_rw = 0;
-                       if (SCpnt->device->use_10_for_ms &&
-                           (SCpnt->cmnd[0] == MODE_SENSE_10 ||
-                            SCpnt->cmnd[0] == MODE_SELECT_10))
-                               SCpnt->device->use_10_for_ms = 0;
-                       break;
-
-               default:
-                       break;
-               }
+               /* This computation should always be done in terms of
+                * the resolution of the device's medium.
+                */
+               good_bytes = (bad_lba - start_lba)*SCpnt->device->sector_size;
+               break;
+       case RECOVERED_ERROR:
+       case NO_SENSE:
+               /* Inform the user, but make sure that it's not treated
+                * as a hard error.
+                */
+               scsi_print_sense("sd", SCpnt);
+               SCpnt->result = 0;
+               memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
+               good_bytes = xfer_size;
+               break;
+       case ILLEGAL_REQUEST:
+               if (SCpnt->device->use_10_for_rw &&
+                   (SCpnt->cmnd[0] == READ_10 ||
+                    SCpnt->cmnd[0] == WRITE_10))
+                       SCpnt->device->use_10_for_rw = 0;
+               if (SCpnt->device->use_10_for_ms &&
+                   (SCpnt->cmnd[0] == MODE_SENSE_10 ||
+                    SCpnt->cmnd[0] == MODE_SELECT_10))
+                       SCpnt->device->use_10_for_ms = 0;
+               break;
+       default:
+               break;
        }
-       /*
-        * This calls the generic completion function, now that we know
-        * how many actual sectors finished, and how many sectors we need
-        * to say have failed.
-        */
-       scsi_io_completion(SCpnt, good_bytes, block_sectors << 9);
+ out:
+       return good_bytes;
 }
 
 static int media_not_present(struct scsi_disk *sdkp,
@@ -992,7 +1038,7 @@ static int media_not_present(struct scsi_disk *sdkp,
  * spinup disk - called only in sd_revalidate_disk()
  */
 static void
-sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
+sd_spinup_disk(struct scsi_disk *sdkp)
 {
        unsigned char cmd[10];
        unsigned long spintime_expire = 0;
@@ -1017,6 +1063,14 @@ sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
                                                      &sshdr, SD_TIMEOUT,
                                                      SD_MAX_RETRIES);
 
+                       /*
+                        * If the drive has indicated to us that it
+                        * doesn't have any media in it, don't bother
+                        * with any more polling.
+                        */
+                       if (media_not_present(sdkp, &sshdr))
+                               return;
+
                        if (the_result)
                                sense_valid = scsi_sense_valid(&sshdr);
                        retries++;
@@ -1025,20 +1079,13 @@ sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
                          ((driver_byte(the_result) & DRIVER_SENSE) &&
                          sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
 
-               /*
-                * If the drive has indicated to us that it doesn't have
-                * any media in it, don't bother with any of the rest of
-                * this crap.
-                */
-               if (media_not_present(sdkp, &sshdr))
-                       return;
-
                if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
                        /* no sense, TUR either succeeded or failed
                         * with a status error */
-                       if(!spintime && !scsi_status_is_good(the_result))
-                               printk(KERN_NOTICE "%s: Unit Not Ready, "
-                                      "error = 0x%x\n", diskname, the_result);
+                       if(!spintime && !scsi_status_is_good(the_result)) {
+                               sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
+                               sd_print_result(sdkp, the_result);
+                       }
                        break;
                }
                                        
@@ -1063,8 +1110,7 @@ sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
                 */
                } else if (sense_valid && sshdr.sense_key == NOT_READY) {
                        if (!spintime) {
-                               printk(KERN_NOTICE "%s: Spinning up disk...",
-                                      diskname);
+                               sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
                                cmd[0] = START_STOP;
                                cmd[1] = 1;     /* Return immediately */
                                memset((void *) &cmd[2], 0, 8);
@@ -1097,9 +1143,8 @@ sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
                        /* we don't understand the sense code, so it's
                         * probably pointless to loop */
                        if(!spintime) {
-                               printk(KERN_NOTICE "%s: Unit Not Ready, "
-                                       "sense:\n", diskname);
-                               scsi_print_sense_hdr("", &sshdr);
+                               sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
+                               sd_print_sense_hdr(sdkp, &sshdr);
                        }
                        break;
                }
@@ -1118,8 +1163,7 @@ sd_spinup_disk(struct scsi_disk *sdkp, char *diskname)
  * read disk capacity
  */
 static void
-sd_read_capacity(struct scsi_disk *sdkp, char *diskname,
-                unsigned char *buffer)
+sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
 {
        unsigned char cmd[16];
        int the_result, retries;
@@ -1158,18 +1202,12 @@ repeat:
        } while (the_result && retries);
 
        if (the_result && !longrc) {
-               printk(KERN_NOTICE "%s : READ CAPACITY failed.\n"
-                      "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
-                      diskname, diskname,
-                      status_byte(the_result),
-                      msg_byte(the_result),
-                      host_byte(the_result),
-                      driver_byte(the_result));
-
+               sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
+               sd_print_result(sdkp, the_result);
                if (driver_byte(the_result) & DRIVER_SENSE)
-                       scsi_print_sense_hdr("sd", &sshdr);
+                       sd_print_sense_hdr(sdkp, &sshdr);
                else
-                       printk("%s : sense not available. \n", diskname);
+                       sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
 
                /* Set dirty bit for removable devices if not ready -
                 * sometimes drives will not report this properly. */
@@ -1180,21 +1218,15 @@ repeat:
                /* Either no media are present but the drive didn't tell us,
                   or they are present but the read capacity command fails */
                /* sdkp->media_present = 0; -- not always correct */
-               sdkp->capacity = 0x200000; /* 1 GB - random */
+               sdkp->capacity = 0; /* unknown mapped to zero - as usual */
 
                return;
        } else if (the_result && longrc) {
                /* READ CAPACITY(16) has been failed */
-               printk(KERN_NOTICE "%s : READ CAPACITY(16) failed.\n"
-                      "%s : status=%x, message=%02x, host=%d, driver=%02x \n",
-                      diskname, diskname,
-                      status_byte(the_result),
-                      msg_byte(the_result),
-                      host_byte(the_result),
-                      driver_byte(the_result));
-               printk(KERN_NOTICE "%s : use 0xffffffff as device size\n",
-                      diskname);
-               
+               sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
+               sd_print_result(sdkp, the_result);
+               sd_printk(KERN_NOTICE, sdkp, "Use 0xffffffff as device size\n");
+
                sdkp->capacity = 1 + (sector_t) 0xffffffff;             
                goto got_data;
        }       
@@ -1205,14 +1237,14 @@ repeat:
                if (buffer[0] == 0xff && buffer[1] == 0xff &&
                    buffer[2] == 0xff && buffer[3] == 0xff) {
                        if(sizeof(sdkp->capacity) > 4) {
-                               printk(KERN_NOTICE "%s : very big device. try to use"
-                                      " READ CAPACITY(16).\n", diskname);
+                               sd_printk(KERN_NOTICE, sdkp, "Very big device. "
+                                         "Trying to use READ CAPACITY(16).\n");
                                longrc = 1;
                                goto repeat;
                        }
-                       printk(KERN_ERR "%s: too big for this kernel.  Use a "
-                              "kernel compiled with support for large block "
-                              "devices.\n", diskname);
+                       sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use "
+                                 "a kernel compiled with support for large "
+                                 "block devices.\n");
                        sdkp->capacity = 0;
                        goto got_data;
                }
@@ -1236,14 +1268,23 @@ repeat:
 
        /* Some devices return the total number of sectors, not the
         * highest sector number.  Make the necessary adjustment. */
-       if (sdp->fix_capacity)
+       if (sdp->fix_capacity) {
                --sdkp->capacity;
 
+       /* Some devices have version which report the correct sizes
+        * and others which do not. We guess size according to a heuristic
+        * and err on the side of lowering the capacity. */
+       } else {
+               if (sdp->guess_capacity)
+                       if (sdkp->capacity & 0x01) /* odd sizes are odd */
+                               --sdkp->capacity;
+       }
+
 got_data:
        if (sector_size == 0) {
                sector_size = 512;
-               printk(KERN_NOTICE "%s : sector size 0 reported, "
-                      "assuming 512.\n", diskname);
+               sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
+                         "assuming 512.\n");
        }
 
        if (sector_size != 512 &&
@@ -1251,8 +1292,8 @@ got_data:
            sector_size != 2048 &&
            sector_size != 4096 &&
            sector_size != 256) {
-               printk(KERN_NOTICE "%s : unsupported sector size "
-                      "%d.\n", diskname, sector_size);
+               sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
+                         sector_size);
                /*
                 * The user might want to re-format the drive with
                 * a supported sectorsize.  Once this happens, it
@@ -1276,7 +1317,7 @@ got_data:
                 */
                int hard_sector = sector_size;
                sector_t sz = (sdkp->capacity/2) * (hard_sector/256);
-               request_queue_t *queue = sdp->request_queue;
+               struct request_queue *queue = sdp->request_queue;
                sector_t mb = sz;
 
                blk_queue_hardsect_size(queue, hard_sector);
@@ -1285,10 +1326,10 @@ got_data:
                mb -= sz - 974;
                sector_div(mb, 1950);
 
-               printk(KERN_NOTICE "SCSI device %s: "
-                      "%llu %d-byte hdwr sectors (%llu MB)\n",
-                      diskname, (unsigned long long)sdkp->capacity,
-                      hard_sector, (unsigned long long)mb);
+               sd_printk(KERN_NOTICE, sdkp,
+                         "%llu %d-byte hardware sectors (%llu MB)\n",
+                         (unsigned long long)sdkp->capacity,
+                         hard_sector, (unsigned long long)mb);
        }
 
        /* Rescale capacity to 512-byte units */
@@ -1317,11 +1358,10 @@ sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
 
 /*
  * read write protect setting, if possible - called only in sd_revalidate_disk()
- * called with buffer of length 512
+ * called with buffer of length SD_BUF_SIZE
  */
 static void
-sd_read_write_protect_flag(struct scsi_disk *sdkp, char *diskname,
-                          unsigned char *buffer)
+sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
 {
        int res;
        struct scsi_device *sdp = sdkp->device;
@@ -1329,7 +1369,7 @@ sd_read_write_protect_flag(struct scsi_disk *sdkp, char *diskname,
 
        set_disk_ro(sdkp->disk, 0);
        if (sdp->skip_ms_page_3f) {
-               printk(KERN_NOTICE "%s: assuming Write Enabled\n", diskname);
+               sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
                return;
        }
 
@@ -1361,25 +1401,25 @@ sd_read_write_protect_flag(struct scsi_disk *sdkp, char *diskname,
        }
 
        if (!scsi_status_is_good(res)) {
-               printk(KERN_WARNING
-                      "%s: test WP failed, assume Write Enabled\n", diskname);
+               sd_printk(KERN_WARNING, sdkp,
+                         "Test WP failed, assume Write Enabled\n");
        } else {
                sdkp->write_prot = ((data.device_specific & 0x80) != 0);
                set_disk_ro(sdkp->disk, sdkp->write_prot);
-               printk(KERN_NOTICE "%s: Write Protect is %s\n", diskname,
-                      sdkp->write_prot ? "on" : "off");
-               printk(KERN_DEBUG "%s: Mode Sense: %02x %02x %02x %02x\n",
-                      diskname, buffer[0], buffer[1], buffer[2], buffer[3]);
+               sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
+                         sdkp->write_prot ? "on" : "off");
+               sd_printk(KERN_DEBUG, sdkp,
+                         "Mode Sense: %02x %02x %02x %02x\n",
+                         buffer[0], buffer[1], buffer[2], buffer[3]);
        }
 }
 
 /*
  * sd_read_cache_type - called only from sd_revalidate_disk()
- * called with buffer of length 512
+ * called with buffer of length SD_BUF_SIZE
  */
 static void
-sd_read_cache_type(struct scsi_disk *sdkp, char *diskname,
-                  unsigned char *buffer)
+sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
 {
        int len = 0, res;
        struct scsi_device *sdp = sdkp->device;
@@ -1406,6 +1446,11 @@ sd_read_cache_type(struct scsi_disk *sdkp, char *diskname,
        if (!scsi_status_is_good(res))
                goto bad_sense;
 
+       if (!data.header_length) {
+               modepage = 6;
+               sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
+       }
+
        /* that went OK, now ask for the proper length */
        len = data.length;
 
@@ -1420,20 +1465,22 @@ sd_read_cache_type(struct scsi_disk *sdkp, char *diskname,
 
        /* Take headers and block descriptors into account */
        len += data.header_length + data.block_descriptor_length;
+       if (len > SD_BUF_SIZE)
+               goto bad_sense;
 
        /* Get the data */
        res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, &data, &sshdr);
 
        if (scsi_status_is_good(res)) {
-               const char *types[] = {
-                       "write through", "none", "write back",
-                       "write back, no read (daft)"
-               };
-               int ct = 0;
                int offset = data.header_length + data.block_descriptor_length;
 
+               if (offset >= SD_BUF_SIZE - 2) {
+                       sd_printk(KERN_ERR, sdkp, "Malformed MODE SENSE response\n");
+                       goto defaults;
+               }
+
                if ((buffer[offset] & 0x3f) != modepage) {
-                       printk(KERN_ERR "%s: got wrong page\n", diskname);
+                       sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
                        goto defaults;
                }
 
@@ -1445,10 +1492,19 @@ sd_read_cache_type(struct scsi_disk *sdkp, char *diskname,
                        sdkp->RCD = 0;
                }
 
-               ct =  sdkp->RCD + 2*sdkp->WCE;
+               sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
+               if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
+                       sd_printk(KERN_NOTICE, sdkp,
+                                 "Uses READ/WRITE(6), disabling FUA\n");
+                       sdkp->DPOFUA = 0;
+               }
 
-               printk(KERN_NOTICE "SCSI device %s: drive cache: %s\n",
-                      diskname, types[ct]);
+               sd_printk(KERN_NOTICE, sdkp,
+                      "Write cache: %s, read cache: %s, %s\n",
+                      sdkp->WCE ? "enabled" : "disabled",
+                      sdkp->RCD ? "disabled" : "enabled",
+                      sdkp->DPOFUA ? "supports DPO and FUA"
+                      : "doesn't support DPO or FUA");
 
                return;
        }
@@ -1457,17 +1513,16 @@ bad_sense:
        if (scsi_sense_valid(&sshdr) &&
            sshdr.sense_key == ILLEGAL_REQUEST &&
            sshdr.asc == 0x24 && sshdr.ascq == 0x0)
-               printk(KERN_NOTICE "%s: cache data unavailable\n",
-                      diskname);       /* Invalid field in CDB */
+               /* Invalid field in CDB */
+               sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
        else
-               printk(KERN_ERR "%s: asking for cache data failed\n",
-                      diskname);
+               sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
 
 defaults:
-       printk(KERN_ERR "%s: assuming drive cache: write through\n",
-              diskname);
+       sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
        sdkp->WCE = 0;
        sdkp->RCD = 0;
+       sdkp->DPOFUA = 0;
 }
 
 /**
@@ -1480,8 +1535,10 @@ static int sd_revalidate_disk(struct gendisk *disk)
        struct scsi_disk *sdkp = scsi_disk(disk);
        struct scsi_device *sdp = sdkp->device;
        unsigned char *buffer;
+       unsigned ordered;
 
-       SCSI_LOG_HLQUEUE(3, printk("sd_revalidate_disk: disk=%s\n", disk->disk_name));
+       SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
+                                     "sd_revalidate_disk\n"));
 
        /*
         * If the device is offline, don't try and read capacity or any
@@ -1490,10 +1547,10 @@ static int sd_revalidate_disk(struct gendisk *disk)
        if (!scsi_device_online(sdp))
                goto out;
 
-       buffer = kmalloc(512, GFP_KERNEL | __GFP_DMA);
+       buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
        if (!buffer) {
-               printk(KERN_WARNING "(sd_revalidate_disk:) Memory allocation "
-                      "failure.\n");
+               sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
+                         "allocation failure.\n");
                goto out;
        }
 
@@ -1505,20 +1562,32 @@ static int sd_revalidate_disk(struct gendisk *disk)
        sdkp->WCE = 0;
        sdkp->RCD = 0;
 
-       sd_spinup_disk(sdkp, disk->disk_name);
+       sd_spinup_disk(sdkp);
 
        /*
         * Without media there is no reason to ask; moreover, some devices
         * react badly if we do.
         */
        if (sdkp->media_present) {
-               sd_read_capacity(sdkp, disk->disk_name, buffer);
-               if (sdp->removable)
-                       sd_read_write_protect_flag(sdkp, disk->disk_name,
-                                                  buffer);
-               sd_read_cache_type(sdkp, disk->disk_name, buffer);
+               sd_read_capacity(sdkp, buffer);
+               sd_read_write_protect_flag(sdkp, buffer);
+               sd_read_cache_type(sdkp, buffer);
        }
-               
+
+       /*
+        * We now have all cache related info, determine how we deal
+        * with ordered requests.  Note that as the current SCSI
+        * dispatch function can alter request order, we cannot use
+        * QUEUE_ORDERED_TAG_* even when ordered tag is supported.
+        */
+       if (sdkp->WCE)
+               ordered = sdkp->DPOFUA
+                       ? QUEUE_ORDERED_DRAIN_FUA : QUEUE_ORDERED_DRAIN_FLUSH;
+       else
+               ordered = QUEUE_ORDERED_DRAIN;
+
+       blk_queue_ordered(sdkp->disk->queue, ordered, sd_prepare_flush);
+
        set_capacity(disk, sdkp->capacity);
        kfree(buffer);
 
@@ -1560,13 +1629,10 @@ static int sd_probe(struct device *dev)
                                        "sd_attach\n"));
 
        error = -ENOMEM;
-       sdkp = kmalloc(sizeof(*sdkp), GFP_KERNEL);
+       sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
        if (!sdkp)
                goto out;
 
-       memset (sdkp, 0, sizeof(*sdkp));
-       kref_init(&sdkp->kref);
-
        gd = alloc_disk(16);
        if (!gd)
                goto out_free;
@@ -1583,12 +1649,12 @@ static int sd_probe(struct device *dev)
        if (error)
                goto out_put;
 
-       get_device(&sdp->sdev_gendev);
        sdkp->device = sdp;
        sdkp->driver = &sd_template;
        sdkp->disk = gd;
        sdkp->index = index;
        sdkp->openers = 0;
+       sdkp->previous_state = 1;
 
        if (!sdp->timeout) {
                if (sdp->type != TYPE_MOD)
@@ -1597,6 +1663,16 @@ static int sd_probe(struct device *dev)
                        sdp->timeout = SD_MOD_TIMEOUT;
        }
 
+       class_device_initialize(&sdkp->cdev);
+       sdkp->cdev.dev = &sdp->sdev_gendev;
+       sdkp->cdev.class = &sd_disk_class;
+       strncpy(sdkp->cdev.class_id, sdp->sdev_gendev.bus_id, BUS_ID_SIZE);
+
+       if (class_device_add(&sdkp->cdev))
+               goto out_put;
+
+       get_device(&sdp->sdev_gendev);
+
        gd->major = sd_major((index & 0xf0) >> 4);
        gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
        gd->minors = 16;
@@ -1615,31 +1691,31 @@ static int sd_probe(struct device *dev)
                        'a' + m1, 'a' + m2, 'a' + m3);
        }
 
-       strcpy(gd->devfs_name, sdp->devfs_name);
-
        gd->private_data = &sdkp->driver;
+       gd->queue = sdkp->device->request_queue;
 
        sd_revalidate_disk(gd);
 
+       blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
+
        gd->driverfs_dev = &sdp->sdev_gendev;
        gd->flags = GENHD_FL_DRIVERFS;
        if (sdp->removable)
                gd->flags |= GENHD_FL_REMOVABLE;
-       gd->queue = sdkp->device->request_queue;
 
        dev_set_drvdata(dev, sdkp);
        add_disk(gd);
 
-       sdev_printk(KERN_NOTICE, sdp, "Attached scsi %sdisk %s\n",
-                   sdp->removable ? "removable " : "", gd->disk_name);
+       sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
+                 sdp->removable ? "removable " : "");
 
        return 0;
 
-out_put:
+ out_put:
        put_disk(gd);
-out_free:
+ out_free:
        kfree(sdkp);
-out:
+ out:
        return error;
 }
 
@@ -1658,29 +1734,30 @@ static int sd_remove(struct device *dev)
 {
        struct scsi_disk *sdkp = dev_get_drvdata(dev);
 
+       class_device_del(&sdkp->cdev);
        del_gendisk(sdkp->disk);
        sd_shutdown(dev);
 
-       down(&sd_ref_sem);
+       mutex_lock(&sd_ref_mutex);
        dev_set_drvdata(dev, NULL);
-       kref_put(&sdkp->kref, scsi_disk_release);
-       up(&sd_ref_sem);
+       class_device_put(&sdkp->cdev);
+       mutex_unlock(&sd_ref_mutex);
 
        return 0;
 }
 
 /**
  *     scsi_disk_release - Called to free the scsi_disk structure
- *     @kref: pointer to embedded kref
+ *     @cdev: pointer to embedded class device
  *
- *     sd_ref_sem must be held entering this routine.  Because it is
+ *     sd_ref_mutex must be held entering this routine.  Because it is
  *     called on last put, you should always use the scsi_disk_get()
  *     scsi_disk_put() helpers which manipulate the semaphore directly
- *     and never do a direct kref_put().
+ *     and never do a direct class_device_put().
  **/
-static void scsi_disk_release(struct kref *kref)
+static void scsi_disk_release(struct class_device *cdev)
 {
-       struct scsi_disk *sdkp = to_scsi_disk(kref);
+       struct scsi_disk *sdkp = to_scsi_disk(cdev);
        struct gendisk *disk = sdkp->disk;
        
        spin_lock(&sd_index_lock);
@@ -1694,6 +1771,31 @@ static void scsi_disk_release(struct kref *kref)
        kfree(sdkp);
 }
 
+static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
+{
+       unsigned char cmd[6] = { START_STOP };  /* START_VALID */
+       struct scsi_sense_hdr sshdr;
+       struct scsi_device *sdp = sdkp->device;
+       int res;
+
+       if (start)
+               cmd[4] |= 1;    /* START */
+
+       if (!scsi_device_online(sdp))
+               return -ENODEV;
+
+       res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
+                              SD_TIMEOUT, SD_MAX_RETRIES);
+       if (res) {
+               sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
+               sd_print_result(sdkp, res);
+               if (driver_byte(res) & DRIVER_SENSE)
+                       sd_print_sense_hdr(sdkp, &sshdr);
+       }
+
+       return res;
+}
+
 /*
  * Send a SYNCHRONIZE CACHE instruction down to the device through
  * the normal SCSI command structure.  Wait for the command to
@@ -1701,20 +1803,65 @@ static void scsi_disk_release(struct kref *kref)
  */
 static void sd_shutdown(struct device *dev)
 {
-       struct scsi_device *sdp = to_scsi_device(dev);
        struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
 
        if (!sdkp)
                return;         /* this can happen */
 
        if (sdkp->WCE) {
-               printk(KERN_NOTICE "Synchronizing SCSI cache for disk %s: \n",
-                               sdkp->disk->disk_name);
-               sd_sync_cache(sdp);
+               sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
+               sd_sync_cache(sdkp);
        }
+
+       if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
+               sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
+               sd_start_stop_device(sdkp, 0);
+       }
+
        scsi_disk_put(sdkp);
 }
 
+static int sd_suspend(struct device *dev, pm_message_t mesg)
+{
+       struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
+       int ret = 0;
+
+       if (!sdkp)
+               return 0;       /* this can happen */
+
+       if (sdkp->WCE) {
+               sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
+               ret = sd_sync_cache(sdkp);
+               if (ret)
+                       goto done;
+       }
+
+       if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
+               sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
+               ret = sd_start_stop_device(sdkp, 0);
+       }
+
+done:
+       scsi_disk_put(sdkp);
+       return ret;
+}
+
+static int sd_resume(struct device *dev)
+{
+       struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
+       int ret = 0;
+
+       if (!sdkp->device->manage_start_stop)
+               goto done;
+
+       sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
+       ret = sd_start_stop_device(sdkp, 1);
+
+done:
+       scsi_disk_put(sdkp);
+       return ret;
+}
+
 /**
  *     init_sd - entry point for this driver (both when built in or when
  *     a module).
@@ -1723,7 +1870,7 @@ static void sd_shutdown(struct device *dev)
  **/
 static int __init init_sd(void)
 {
-       int majors = 0, i;
+       int majors = 0, i, err;
 
        SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
 
@@ -1734,7 +1881,22 @@ static int __init init_sd(void)
        if (!majors)
                return -ENODEV;
 
-       return scsi_register_driver(&sd_template.gendrv);
+       err = class_register(&sd_disk_class);
+       if (err)
+               goto err_out;
+
+       err = scsi_register_driver(&sd_template.gendrv);
+       if (err)
+               goto err_out_class;
+
+       return 0;
+
+err_out_class:
+       class_unregister(&sd_disk_class);
+err_out:
+       for (i = 0; i < SD_MAJORS; i++)
+               unregister_blkdev(sd_major(i), "sd");
+       return err;
 }
 
 /**
@@ -1749,13 +1911,27 @@ static void __exit exit_sd(void)
        SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
 
        scsi_unregister_driver(&sd_template.gendrv);
+       class_unregister(&sd_disk_class);
+
        for (i = 0; i < SD_MAJORS; i++)
                unregister_blkdev(sd_major(i), "sd");
 }
 
-MODULE_LICENSE("GPL");
-MODULE_AUTHOR("Eric Youngdale");
-MODULE_DESCRIPTION("SCSI disk (sd) driver");
-
 module_init(init_sd);
 module_exit(exit_sd);
+
+static void sd_print_sense_hdr(struct scsi_disk *sdkp,
+                              struct scsi_sense_hdr *sshdr)
+{
+       sd_printk(KERN_INFO, sdkp, "");
+       scsi_show_sense_hdr(sshdr);
+       sd_printk(KERN_INFO, sdkp, "");
+       scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
+}
+
+static void sd_print_result(struct scsi_disk *sdkp, int result)
+{
+       sd_printk(KERN_INFO, sdkp, "");
+       scsi_show_result(result);
+}
+