virtio_blk: add block topology support
authorChristoph Hellwig <hch@lst.de>
Wed, 24 Feb 2010 20:22:25 +0000 (14:22 -0600)
committerRusty Russell <rusty@rustcorp.com.au>
Wed, 24 Feb 2010 03:52:26 +0000 (14:22 +1030)
Allow reading various alignment values from the config page.  This
allows the guest to much better align I/O requests depending on the
storage topology.

Note that the formats for the config values appear a bit messed up,
but we follow the formats used by ATA and SCSI so they are expected in
the storage world.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
drivers/block/virtio_blk.c
include/linux/virtio_blk.h

index 51042f0..7eff828 100644 (file)
@@ -243,10 +243,12 @@ static int index_to_minor(int index)
 static int __devinit virtblk_probe(struct virtio_device *vdev)
 {
        struct virtio_blk *vblk;
+       struct request_queue *q;
        int err;
        u64 cap;
-       u32 v;
-       u32 blk_size, sg_elems;
+       u32 v, blk_size, sg_elems, opt_io_size;
+       u16 min_io_size;
+       u8 physical_block_exp, alignment_offset;
 
        if (index_to_minor(index) >= 1 << MINORBITS)
                return -ENOSPC;
@@ -293,13 +295,13 @@ static int __devinit virtblk_probe(struct virtio_device *vdev)
                goto out_mempool;
        }
 
-       vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
-       if (!vblk->disk->queue) {
+       q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
+       if (!q) {
                err = -ENOMEM;
                goto out_put_disk;
        }
 
-       vblk->disk->queue->queuedata = vblk;
+       q->queuedata = vblk;
 
        if (index < 26) {
                sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
@@ -323,10 +325,10 @@ static int __devinit virtblk_probe(struct virtio_device *vdev)
 
        /* If barriers are supported, tell block layer that queue is ordered */
        if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
-               blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_DRAIN_FLUSH,
+               blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH,
                                  virtblk_prepare_flush);
        else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
-               blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
+               blk_queue_ordered(q, QUEUE_ORDERED_TAG, NULL);
 
        /* If disk is read-only in the host, the guest should obey */
        if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
@@ -345,14 +347,14 @@ static int __devinit virtblk_probe(struct virtio_device *vdev)
        set_capacity(vblk->disk, cap);
 
        /* We can handle whatever the host told us to handle. */
-       blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2);
-       blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2);
+       blk_queue_max_phys_segments(q, vblk->sg_elems-2);
+       blk_queue_max_hw_segments(q, vblk->sg_elems-2);
 
        /* No need to bounce any requests */
-       blk_queue_bounce_limit(vblk->disk->queue, BLK_BOUNCE_ANY);
+       blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 
        /* No real sector limit. */
-       blk_queue_max_sectors(vblk->disk->queue, -1U);
+       blk_queue_max_sectors(q, -1U);
 
        /* Host can optionally specify maximum segment size and number of
         * segments. */
@@ -360,16 +362,45 @@ static int __devinit virtblk_probe(struct virtio_device *vdev)
                                offsetof(struct virtio_blk_config, size_max),
                                &v);
        if (!err)
-               blk_queue_max_segment_size(vblk->disk->queue, v);
+               blk_queue_max_segment_size(q, v);
        else
-               blk_queue_max_segment_size(vblk->disk->queue, -1U);
+               blk_queue_max_segment_size(q, -1U);
 
        /* Host can optionally specify the block size of the device */
        err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
                                offsetof(struct virtio_blk_config, blk_size),
                                &blk_size);
        if (!err)
-               blk_queue_logical_block_size(vblk->disk->queue, blk_size);
+               blk_queue_logical_block_size(q, blk_size);
+       else
+               blk_size = queue_logical_block_size(q);
+
+       /* Use topology information if available */
+       err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
+                       offsetof(struct virtio_blk_config, physical_block_exp),
+                       &physical_block_exp);
+       if (!err && physical_block_exp)
+               blk_queue_physical_block_size(q,
+                               blk_size * (1 << physical_block_exp));
+
+       err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
+                       offsetof(struct virtio_blk_config, alignment_offset),
+                       &alignment_offset);
+       if (!err && alignment_offset)
+               blk_queue_alignment_offset(q, blk_size * alignment_offset);
+
+       err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
+                       offsetof(struct virtio_blk_config, min_io_size),
+                       &min_io_size);
+       if (!err && min_io_size)
+               blk_queue_io_min(q, blk_size * min_io_size);
+
+       err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
+                       offsetof(struct virtio_blk_config, opt_io_size),
+                       &opt_io_size);
+       if (!err && opt_io_size)
+               blk_queue_io_opt(q, blk_size * opt_io_size);
+
 
        add_disk(vblk->disk);
        return 0;
@@ -412,7 +443,7 @@ static struct virtio_device_id id_table[] = {
 static unsigned int features[] = {
        VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
        VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
-       VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH
+       VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
 };
 
 /*
index fd294c5..e52029e 100644 (file)
@@ -15,6 +15,7 @@
 #define VIRTIO_BLK_F_BLK_SIZE  6       /* Block size of disk is available*/
 #define VIRTIO_BLK_F_SCSI      7       /* Supports scsi command passthru */
 #define VIRTIO_BLK_F_FLUSH     9       /* Cache flush command support */
+#define VIRTIO_BLK_F_TOPOLOGY  10      /* Topology information is available */
 
 struct virtio_blk_config {
        /* The capacity (in 512-byte sectors). */
@@ -29,8 +30,20 @@ struct virtio_blk_config {
                __u8 heads;
                __u8 sectors;
        } geometry;
+
        /* block size of device (if VIRTIO_BLK_F_BLK_SIZE) */
        __u32 blk_size;
+
+       /* the next 4 entries are guarded by VIRTIO_BLK_F_TOPOLOGY  */
+       /* exponent for physical block per logical block. */
+       __u8 physical_block_exp;
+       /* alignment offset in logical blocks. */
+       __u8 alignment_offset;
+       /* minimum I/O size without performance penalty in logical blocks. */
+       __u16 min_io_size;
+       /* optimal sustained I/O size in logical blocks. */
+       __u32 opt_io_size;
+
 } __attribute__((packed));
 
 /*