xen/blkback: Seperate the bio allocation and the bio submission.
authorKonrad Rzeszutek Wilk <konrad@kernel.org>
Fri, 15 Apr 2011 14:51:27 +0000 (10:51 -0400)
committerKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Fri, 15 Apr 2011 14:54:42 +0000 (10:54 -0400)
We seperate the bio allocation (bio_alloc) from the bio submission so
that the error paths are much easier, and also so that the bio
submission can be done in one tight loop. It also makes the
plug/unplug calls much much easier.

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
drivers/xen/blkback/blkback.c

index 464f2e0b5a61ac83f6f637fc72107f0fb84a3b1b..3c10499d61a7480b2cdf9e21ee4132e20570d5fc 100644 (file)
@@ -421,7 +421,8 @@ static void dispatch_rw_block_io(struct blkif_st *blkif,
        } seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
        unsigned int nseg;
        struct bio *bio = NULL;
-       int ret, i;
+       struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST];
+       int ret, i, nbio = 0;
        int operation;
        struct blk_plug plug;
        struct request_queue *q;
@@ -529,14 +530,7 @@ static void dispatch_rw_block_io(struct blkif_st *blkif,
                goto fail_flush;
        }
 
-       /* Get a reference count for the disk queue and start sending I/O */
-       blk_get_queue(q);
-       blk_start_plug(&plug);
-
-       /* We set it one so that the last submit_bio does not have to call
-        * atomic_inc.
-        */
-       atomic_set(&pending_req->pendcnt, 1);
+       /* This corresponding blkif_put is done in __end_block_io_op */
        blkif_get(blkif);
 
        for (i = 0; i < nseg; i++) {
@@ -552,12 +546,8 @@ static void dispatch_rw_block_io(struct blkif_st *blkif,
                                     blkbk->pending_page(pending_req, i),
                                     seg[i].nsec << 9,
                                     seg[i].buf & ~PAGE_MASK) == 0)) {
-                       if (bio) {
-                               atomic_inc(&pending_req->pendcnt);
-                               submit_bio(operation, bio);
-                       }
 
-                       bio = bio_alloc(GFP_KERNEL, nseg-i);
+                       bio = biolist[nbio++] = bio_alloc(GFP_KERNEL, nseg-i);
                        if (unlikely(bio == NULL))
                                goto fail_put_bio;
 
@@ -573,7 +563,7 @@ static void dispatch_rw_block_io(struct blkif_st *blkif,
        /* This will be hit if the operation was a barrier. */
        if (!bio) {
                BUG_ON(operation != WRITE_BARRIER);
-               bio = bio_alloc(GFP_KERNEL, 0);
+               bio = biolist[nbio++] = bio_alloc(GFP_KERNEL, 0);
                if (unlikely(bio == NULL))
                        goto fail_put_bio;
 
@@ -583,15 +573,28 @@ static void dispatch_rw_block_io(struct blkif_st *blkif,
                bio->bi_sector  = -1;
        }
 
-       submit_bio(operation, bio);
+
+       /* We set it one so that the last submit_bio does not have to call
+        * atomic_inc.
+        */
+       atomic_set(&pending_req->pendcnt, nbio);
+
+       /* Get a reference count for the disk queue and start sending I/O */
+       blk_get_queue(q);
+       blk_start_plug(&plug);
+
+       for (i = 0; i < nbio; i++)
+               submit_bio(operation, biolist[i]);
+
+       blk_finish_plug(&plug);
+       /* Let the I/Os go.. */
+       blk_put_queue(q);
 
        if (operation == READ)
                blkif->st_rd_sect += preq.nr_sects;
        else if (operation == WRITE || operation == WRITE_BARRIER)
                blkif->st_wr_sect += preq.nr_sects;
 
-       blk_finish_plug(&plug);
-       blk_put_queue(q);
        return;
 
  fail_flush:
@@ -604,11 +607,9 @@ static void dispatch_rw_block_io(struct blkif_st *blkif,
        return;
 
  fail_put_bio:
+       for (i = 0; i < (nbio-1); i++)
+               bio_put(biolist[i]);
        __end_block_io_op(pending_req, -EINVAL);
-       if (bio)
-               bio_put(bio);
-       blk_finish_plug(&plug);
-       blk_put_queue(q);
        msleep(1); /* back off a bit */
        return;
 }