block: split bios to max possible length

This splits bio in the middle of a vector to form the largest possible
bio at the h/w's desired alignment, and guarantees the bio being split
will have some data.

The criteria for splitting is changed from the max sectors to the h/w's
optimal sector alignment if it is provided. For h/w that advertise their
block storage's underlying chunk size, it's a big performance win to not
submit commands that cross them. If sector alignment is not provided,
this patch uses the max sectors as before.

This addresses the performance issue commit d380561113 attempted to
fix, but was reverted due to splitting logic error.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Ming Lei <tom.leiming@gmail.com>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Cc: <stable@vger.kernel.org> # 4.4.x-
Signed-off-by: Jens Axboe <axboe@fb.com>

1 files changed, 16 insertions, 3 deletions

diff --git a/block/blk-merge.c b/block/blk-merge.c
index 0e5643a..237b087 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -83,9 +83,6 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
 	struct bio *new = NULL;
 
 	bio_for_each_segment(bv, bio, iter) {
-		if (sectors + (bv.bv_len >> 9) > queue_max_sectors(q))
-			goto split;
-
 		/*
 		 * If the queue doesn't support SG gaps and adding this
 		 * offset would create a gap, disallow it.
@@ -93,6 +90,22 @@ static struct bio *blk_bio_segment_split(struct request_queue *q,
 		if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
 			goto split;
 
+		if (sectors + (bv.bv_len >> 9) >
+				blk_max_size_offset(q, bio->bi_iter.bi_sector)) {
+			/*
+			 * Consider this a new segment if we're splitting in
+			 * the middle of this vector.
+			 */
+			if (nsegs < queue_max_segments(q) &&
+			    sectors < blk_max_size_offset(q,
+						bio->bi_iter.bi_sector)) {
+				nsegs++;
+				sectors = blk_max_size_offset(q,
+						bio->bi_iter.bi_sector);
+			}
+			goto split;
+		}
+
 		if (bvprvp && blk_queue_cluster(q)) {
 			if (seg_size + bv.bv_len > queue_max_segment_size(q))
 				goto new_segment;