[PATCH 1/3] block: introduce blk_queue_nr_active()

[Ming Lei]

On Wed, Sep 27, 2023 at 09:36:11AM +0200, Hannes Reinecke wrote:
> On 9/25/23 18:31, Ewan D. Milne wrote:
> > Returns a count of the total number of active requests
> > in a queue.  For non-shared tags (the usual case) this is
> > the sum of nr_active from all of the hctxs.
> > 
> Bit of an exaggeration here.
> Shared tags are in use if the hardware supports only a global tag space
> (ie basically all SCSI and FC HBAs).
> 
> > Signed-off-by: Ewan D. Milne <emilne at redhat.com>
> > ---
> >   block/blk-mq.h         |  5 -----
> >   include/linux/blk-mq.h | 33 ++++++++++++++++++++++++++-------
> >   2 files changed, 26 insertions(+), 12 deletions(-)
> > 
> > diff --git a/block/blk-mq.h b/block/blk-mq.h
> > index 1743857e0b01..fbc65eefa017 100644
> > --- a/block/blk-mq.h
> > +++ b/block/blk-mq.h
> > @@ -214,11 +214,6 @@ static inline bool blk_mq_tag_is_reserved(struct blk_mq_tags *tags,
> >   	return tag < tags->nr_reserved_tags;
> >   }
> > -static inline bool blk_mq_is_shared_tags(unsigned int flags)
> > -{
> > -	return flags & BLK_MQ_F_TAG_HCTX_SHARED;
> > -}
> > -
> >   static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data)
> >   {
> >   	if (data->rq_flags & RQF_SCHED_TAGS)
> > diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
> > index 01e8c31db665..c921ae5236ab 100644
> > --- a/include/linux/blk-mq.h
> > +++ b/include/linux/blk-mq.h
> > @@ -716,6 +716,32 @@ int blk_rq_poll(struct request *rq, struct io_comp_batch *iob,
> >   bool blk_mq_queue_inflight(struct request_queue *q);
> > +#define queue_for_each_hw_ctx(q, hctx, i)				\
> > +	xa_for_each(&(q)->hctx_table, (i), (hctx))
> > +
> > +#define hctx_for_each_ctx(hctx, ctx, i)					\
> > +	for ((i) = 0; (i) < (hctx)->nr_ctx &&				\
> > +	     ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
> > +
> > +static inline bool blk_mq_is_shared_tags(unsigned int flags)
> > +{
> > +	return flags & BLK_MQ_F_TAG_HCTX_SHARED;
> > +}
> > +
> > +static inline unsigned int blk_mq_queue_nr_active(struct request_queue *q)
> > +{
> > +	unsigned int nr_active = 0;
> > +	struct blk_mq_hw_ctx *hctx;
> > +	unsigned long i;
> > +
> > +	queue_for_each_hw_ctx(q, hctx, i) {
> > +		if (unlikely(blk_mq_is_shared_tags(hctx->flags)))
> > +			return atomic_read(&q->nr_active_requests_shared_tags);
> > +		nr_active += atomic_read(&hctx->nr_active);
> > +	}
> > +	return nr_active;
> > +}
> > +
> >   enum {
> >   	/* return when out of requests */
> >   	BLK_MQ_REQ_NOWAIT	= (__force blk_mq_req_flags_t)(1 << 0),
> > @@ -941,13 +967,6 @@ static inline void *blk_mq_rq_to_pdu(struct request *rq)
> >   	return rq + 1;
> >   }
> > -#define queue_for_each_hw_ctx(q, hctx, i)				\
> > -	xa_for_each(&(q)->hctx_table, (i), (hctx))
> > -
> > -#define hctx_for_each_ctx(hctx, ctx, i)					\
> > -	for ((i) = 0; (i) < (hctx)->nr_ctx &&				\
> > -	     ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
> > -
> >   static inline void blk_mq_cleanup_rq(struct request *rq)
> >   {
> >   	if (rq->q->mq_ops->cleanup_rq)
> 
> Well. As discussed, using xarray on 'small' arrays is horrible for
> performance. We really should revert the patch from Ming to
> turn it back into a simple array; that'll make traversing much faster.

But queue_for_each_hw_ctx() isn't supposed for fast path, and it is always
expensive to run cross-queue things.

It also means any way running cross-queue thing in fast path may not be
one good idea.

Thanks,
Ming