[PATCH] block: early return from blk_queue_split() if q->bio_split is NULL

[PATCH] block: early return from blk_queue_split() if q->bio_split is NULL

[Mike Snitzer]

DM appears to be the only block driver that doesn't lean on the block
core's bio splitting.  My hope is to fix that but in the meantime it
doesn't make sense for a device that doesn't need blk_queue_split() to
go through the associated work.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
---
 block/blk-merge.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/block/blk-merge.c b/block/blk-merge.c
index f5dedd5..212004c 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -186,6 +186,9 @@ void blk_queue_split(struct request_queue *q, struct bio **bio)
 	struct bio *split, *res;
 	unsigned nsegs;
 
+	if (!q->bio_split)
+		return;
+
 	switch (bio_op(*bio)) {
 	case REQ_OP_DISCARD:
 	case REQ_OP_SECURE_ERASE:
-- 
2.10.1

Re: block: early return from blk_queue_split() if q->bio_split is NULL

[Mike Snitzer]

On Mon, Nov 20 2017 at 11:54am -0500,
Mike Snitzer <snitzer@redhat.com> wrote:

> DM appears to be the only block driver that doesn't lean on the block
> core's bio splitting.  My hope is to fix that but in the meantime it
> doesn't make sense for a device that doesn't need blk_queue_split() to
> go through the associated work.
> 
> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
> ---
>  block/blk-merge.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/block/blk-merge.c b/block/blk-merge.c
> index f5dedd5..212004c 100644
> --- a/block/blk-merge.c
> +++ b/block/blk-merge.c
> @@ -186,6 +186,9 @@ void blk_queue_split(struct request_queue *q, struct bio **bio)
>  	struct bio *split, *res;
>  	unsigned nsegs;
>  
> +	if (!q->bio_split)
> +		return;
> +
>  	switch (bio_op(*bio)) {
>  	case REQ_OP_DISCARD:
>  	case REQ_OP_SECURE_ERASE:
> -- 
> 2.10.1
> 

Please ignore this patch.  While the patch is perfectly valid, it
doesn't have an existing consumer.  Ming pointed out that, in constrat
to blk_queue_bio(), bio-based DM's dm_make_request() makes it so that
DM never calls blk_queue_split().

Mike