Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Wed, May 20, 2015 at 5:44 AM, Christoph Hellwig <hch at lst.de> wrote: > This needs the patch below for the blk-mq case with multiple queues. Will add it. > > Also I wonder why we really want to push this below ->make_request. > Moving it just outside ->make_request would make the callchain more > obvious and avoid the recursion into the driver. Are you saying move the split to generic_make_request()? Below is an off-list discussion. Ming Lei asked: > > I am wondering why the bio isn't splitted just before q->make_request_fn > > is called in generic_make_request()? By this way, drivers won't need > > to call blk_queue_split() at all. Is it because performance reason? or > > others? Kent replied: > > The reasoning was because blk_queue_split() was only an interim solution - > > there's actually no inherent need to split the bios at all, but not splitting > > them takes (a small amount of) driver modification. You just have the driver > > consume the bios incrementally, advancing the iterator and using bi_remaining > > when they have multiple tags pointing to the same bio. > > > diff --git a/block/blk-mq.c b/block/blk-mq.c > index e9bbc04..e7fae76 100644 > --- a/block/blk-mq.c > +++ b/block/blk-mq.c > @@ -1256,6 +1256,8 @@ static void blk_mq_make_request(struct request_queue *q, struct bio *bio) > return; > } > > + blk_queue_split(q, &bio, q->bio_split); > + > rq = blk_mq_map_request(q, bio, &data); > if (unlikely(!rq)) > return;