| Message ID | 20210630100342.1100-1-hare@suse.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | blk-mq: don't allocate requests when all cpus in a hctx are offline | expand |
On Wed, Jun 30, 2021 at 12:03:42PM +0200, Hannes Reinecke wrote: > When all CPUs in a hctx are offline in blk_mq_alloc_request_hctx() we should > not try to allocate the request, as we'll fail later on in blk_mq_get_tag() anyway. > > Signed-off-by: Hannes Reinecke <hare@suse.de> > --- Please see my yesterday's reply why this way isn't good: https://lore.kernel.org/lkml/YNrwnWfsxf8cJcoe@T590/ https://lore.kernel.org/lkml/YNrhXFgv%2FgEWbhbl@T590/ Thanks, Ming
diff --git a/block/blk-mq.c b/block/blk-mq.c index f11d4018ce2e..1622d0cabe3a 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -468,13 +468,16 @@ struct request *blk_mq_alloc_request_hctx(struct request_queue *q, data.hctx = q->queue_hw_ctx[hctx_idx]; if (!blk_mq_hw_queue_mapped(data.hctx)) goto out_queue_exit; + ret = -EWOULDBLOCK; + /* all cpus in the hctx are offline, don't try to allocate a tag */ cpu = cpumask_first_and(data.hctx->cpumask, cpu_online_mask); + if (unlikely(cpu >= nr_cpu_ids)) + goto out_queue_exit; data.ctx = __blk_mq_get_ctx(q, cpu); if (!q->elevator) blk_mq_tag_busy(data.hctx); - ret = -EWOULDBLOCK; tag = blk_mq_get_tag(&data); if (tag == BLK_MQ_NO_TAG) goto out_queue_exit;
When all CPUs in a hctx are offline in blk_mq_alloc_request_hctx() we should not try to allocate the request, as we'll fail later on in blk_mq_get_tag() anyway. Signed-off-by: Hannes Reinecke <hare@suse.de> --- block/blk-mq.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)