On Mon, Aug 17, 2020 at 08:31:53AM +0200, Christoph Hellwig wrote:
> On Mon, Aug 17, 2020 at 12:09:18PM +0800, Baolin Wang wrote:
> > unsigned int nr_segs)
> > {
> > @@ -447,7 +425,16 @@ bool __blk_mq_sched_bio_merge(struct request_queue *q,
> > struct bio *bio,
> > !list_empty_careful(&ctx->rq_lists[type])) {
> > /* default per sw-queue merge */
> > spin_lock(&ctx->lock);
> > - ret = blk_mq_attempt_merge(q, hctx, ctx, bio, nr_segs);
> > + /*
> > + * Reverse check our software queue for entries that we could
> > + * potentially merge with. Currently includes a hand-wavy stop
> > + * count of 8, to not spend too much time checking for merges.
> > + */
> > + if (blk_mq_bio_list_merge(q, &ctx->rq_lists[type], bio,
> > nr_segs)) {
> > + ctx->rq_merged++;
> > + ret = true;
> > + }
> > +
> > spin_unlock(&ctx->lock);
>
> This adds an overly long line. That being said the whole thing could
> be nicely simplified to:
>
> ...
>
> if (e && e->type->ops.bio_merge)
> return e->type->ops.bio_merge(hctx, bio, nr_segs);
>
> if (!(hctx->flags & BLK_MQ_F_SHOULD_MERGE) ||
> list_empty_careful(&ctx->rq_lists[hctx->type]))
> return false;
>
> /*
> * Reverse check our software queue for entries that we could
> * potentially merge with. Currently includes a hand-wavy stop count of
> * 8, to not spend too much time checking for merges.
> */
> spin_lock(&ctx->lock);
> ret = blk_mq_bio_list_merge(q, &ctx->rq_lists[type], bio, nr_segs);
> if (ret)
> ctx->rq_merged++;
> spin_unlock(&ctx->lock);
>
> Also I think it would make sense to move the locking into
> blk_mq_bio_list_merge.
Sure, will do in next version.
