On Fri, Aug 15, 2025 at 05:29:01PM +0530, Nilay Shroff wrote:
>
>
> On 8/15/25 7:02 AM, Yu Kuai wrote:
> > Hi,
> >
> > 在 2025/08/14 20:23, Nilay Shroff 写道:
> >>
> >>
> >> On 8/14/25 9:05 AM, Yu Kuai wrote:
> >>> From: Yu Kuai <yukuai3@xxxxxxxxxx>
> >>>
> >>> 1) queue_requests_store() is the only caller of
> >>> blk_mq_update_nr_requests(), where queue is already freezed, no need to
> >>> check mq_freeze_depth;
> >>> 2) q->tag_set must be set for request_based device, and queue_is_mq() is
> >>> already checked in blk_mq_queue_attr_visible(), no need to check
> >>> q->tag_set.
> >>> 3) During initialization, hctx->tags in initialized before queue
> >>> kobject, and during del_gendisk, queue kobject is deleted before
> >>> exiting hctx, hence checking hctx->tags is useless.
> >>>
> >>> Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
> >>> ---
> >>> block/blk-mq.c | 11 +----------
> >>> 1 file changed, 1 insertion(+), 10 deletions(-)
> >>>
> >>> diff --git a/block/blk-mq.c b/block/blk-mq.c
> >>> index b67d6c02eceb..3a219b7b3688 100644
> >>> --- a/block/blk-mq.c
> >>> +++ b/block/blk-mq.c
> >>> @@ -4921,24 +4921,15 @@ int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr)
> >>> {
> >>> struct blk_mq_tag_set *set = q->tag_set;
> >>> struct blk_mq_hw_ctx *hctx;
> >>> - int ret;
> >>> + int ret = 0;
> >>> unsigned long i;
> >>> - if (WARN_ON_ONCE(!q->mq_freeze_depth))
> >>> - return -EINVAL;
> >>> -
> >>> - if (!set)
> >>> - return -EINVAL;
> >>> -
> >>> if (q->nr_requests == nr)
> >>> return 0;
> >>> blk_mq_quiesce_queue(q);
> >>> - ret = 0;
> >>> queue_for_each_hw_ctx(q, hctx, i) {
> >>> - if (!hctx->tags)
> >>> - continue;
> >> It's possible that hctx->tags is set to NULL in case no software
> >> queues are mapped to the hardware queue. So it seems that this
> >> check is valid. Please see blk_mq_map_swqueue().
> >
> > Ok, thanks for the reviw.
> >
> > I didn't notice this, just wonder how can this happen?
> > nr_hw_queues > NR_CPUS?
> >
> I think typically having nr_hw_queues > NR_CPUS is not allowed.
> But it's possible to have no software queues are mapped to hctx.
> Check this commit 4412efecf7fd ("Revert "blk-mq: remove code for
> dealing with remapping queue")
It is possible in case of poll_queues, for example:
modprobe null_blk submit_queues=$(getconf _NPROCESSORS_ONLN) poll_queues=2
Thanks,
Ming
