Re: [PATCH] blk-mq: don't fail driver tag allocation because of inactive hctx

[Christoph Hellwig <hch@lst.de>]

On Wed, Jun 03, 2020 at 09:36:08PM +0800, Ming Lei wrote:
> > +bool __blk_mq_get_driver_tag(struct request *rq)
> > +{
> > +	struct sbitmap_queue *bt = &rq->mq_hctx->tags->bitmap_tags;
> > +	unsigned int tag_offset = rq->mq_hctx->tags->nr_reserved_tags;
> > +	bool shared = blk_mq_tag_busy(rq->mq_hctx);
> 
> Not necessary to add 'shared' which is just used once.

blk_mq_tag_busy also increments active_queues, and the existing code
does that before various early returns.  To keep the behavior as-is
the call to blk_mq_tag_busy can't be moved around.

> > +	int tag;
> > +
> > +	if (blk_mq_tag_is_reserved(rq->mq_hctx->sched_tags, rq->internal_tag)) {
> > +		bt = &rq->mq_hctx->tags->breserved_tags;
> 
> Too many rq->mq_hctx->tags, you can add one local variable to store it.

Really just three of them.  And with a local variable confusing it with
rq->mq_hctx->sched_tags becomes much easier.