block: avoid hctx spinlock for plug with multiple queues

blk_mq_flush_plug_list() has a fast path if all requests in the plug
are destined for the same request_queue. It calls ->queue_rqs() with the
whole batch of requests, falling back on ->queue_rq() for any requests
not handled by ->queue_rqs(). However, if the requests are destined for
multiple queues, blk_mq_flush_plug_list() has a slow path that calls
blk_mq_dispatch_list() repeatedly to filter the requests by ctx/hctx.
Each queue's requests are inserted into the hctx's dispatch list under a
spinlock, then __blk_mq_sched_dispatch_requests() takes them out of the
dispatch list (taking the spinlock again), and finally
blk_mq_dispatch_rq_list() calls ->queue_rq() on each request.

Acquiring the hctx spinlock twice and calling ->queue_rq() instead of
->queue_rqs() makes the slow path significantly more expensive. Thus,
batching more requests into a single plug (e.g. io_uring_enter syscall)
can counterintuitively hurt performance by causing the plug to span
multiple queues. We have observed 2-3% of CPU time spent acquiring the
hctx spinlock alone on workloads issuing requests to multiple NVMe
devices in the same io_uring SQE batches.

Add a medium path in blk_mq_flush_plug_list() for plugs that don't have
elevators or come from a schedule, but do span multiple queues. Filter
the requests by queue and call ->queue_rqs()/->queue_rq() on the list of
requests destined to each request_queue.

With this change, we no longer see any CPU time spent in _raw_spin_lock
from blk_mq_flush_plug_list and throughput increases accordingly.

Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250426011728.4189119-4-csander@purestorage.com
[axboe: fix whitespace damage]
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/block/blk-mq.c b/block/blk-mq.c
index a777cb361ee3024975f9d53a498825277b35e909..796baeccd37b846bc3d31037e353e2387b10f1e1 100644 (file)
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -2836,6 +2836,35 @@ static void __blk_mq_flush_list(struct request_queue *q, struct rq_list *rqs)
        q->mq_ops->queue_rqs(rqs);
 }
 
+static unsigned blk_mq_extract_queue_requests(struct rq_list *rqs,
+                                             struct rq_list *queue_rqs)
+{
+       struct request *rq = rq_list_pop(rqs);
+       struct request_queue *this_q = rq->q;
+       struct request **prev = &rqs->head;
+       struct rq_list matched_rqs = {};
+       struct request *last = NULL;
+       unsigned depth = 1;
+
+       rq_list_add_tail(&matched_rqs, rq);
+       while ((rq = *prev)) {
+               if (rq->q == this_q) {
+                       /* move rq from rqs to matched_rqs */
+                       *prev = rq->rq_next;
+                       rq_list_add_tail(&matched_rqs, rq);
+                       depth++;
+               } else {
+                       /* leave rq in rqs */
+                       prev = &rq->rq_next;
+                       last = rq;
+               }
+       }
+
+       rqs->tail = last;
+       *queue_rqs = matched_rqs;
+       return depth;
+}
+
 static void blk_mq_dispatch_queue_requests(struct rq_list *rqs, unsigned depth)
 {
        struct request_queue *q = rq_list_peek(rqs)->q;
@@ -2902,6 +2931,19 @@ static void blk_mq_dispatch_list(struct rq_list *rqs, bool from_sched)
        percpu_ref_put(&this_hctx->queue->q_usage_counter);
 }
 
+static void blk_mq_dispatch_multiple_queue_requests(struct rq_list *rqs)
+{
+       do {
+               struct rq_list queue_rqs;
+               unsigned depth;
+
+               depth = blk_mq_extract_queue_requests(rqs, &queue_rqs);
+               blk_mq_dispatch_queue_requests(&queue_rqs, depth);
+               while (!rq_list_empty(&queue_rqs))
+                       blk_mq_dispatch_list(&queue_rqs, false);
+       } while (!rq_list_empty(rqs));
+}
+
 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
 {
        unsigned int depth;
@@ -2918,7 +2960,12 @@ void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule)
        depth = plug->rq_count;
        plug->rq_count = 0;
 
-       if (!plug->multiple_queues && !plug->has_elevator && !from_schedule) {
+       if (!plug->has_elevator && !from_schedule) {
+               if (plug->multiple_queues) {
+                       blk_mq_dispatch_multiple_queue_requests(&plug->mq_list);
+                       return;
+               }
+
                blk_mq_dispatch_queue_requests(&plug->mq_list, depth);
                if (rq_list_empty(&plug->mq_list))
                        return;