linux-kernel - [PATCH] block/mq-deadline: Speed up the dispatch of low-priority requests

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20210826144039.2143-1-thunder.leizhen@huawei.com>
Date:   Thu, 26 Aug 2021 22:40:39 +0800
From:   Zhen Lei <thunder.leizhen@...wei.com>
To:     Jens Axboe <axboe@...nel.dk>,
        linux-block <linux-block@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
CC:     Zhen Lei <thunder.leizhen@...wei.com>,
        Damien Le Moal <damien.lemoal@....com>,
        Bart Van Assche <bvanassche@....org>
Subject: [PATCH] block/mq-deadline: Speed up the dispatch of low-priority requests

dd_queued() traverses the percpu variable for summation. The more cores,
the higher the performance overhead. I currently have a 128-core board and
this function takes 2.5 us. If the number of high-priority requests is
small and the number of low- and medium-priority requests is large, the
performance impact is significant.

Let's maintain a non-percpu member variable 'nr_queued', which is
incremented by 1 immediately following "inserted++" and decremented by 1
immediately following "completed++". Because both the judgment dd_queued()
in dd_dispatch_request() and operation "inserted++" in dd_insert_request()
are protected by dd->lock, lock protection needs to be added only in
dd_finish_request(), which is unlikely to cause significant performance
side effects.

Tested on my 128-core board with two ssd disks.
fio bs=4k rw=read iodepth=128 cpus_allowed=0-95 <others>
Before:
[183K/0/0 iops]
[172K/0/0 iops]

After:
[258K/0/0 iops]
[258K/0/0 iops]

Fixes: fb926032b320 ("block/mq-deadline: Prioritize high-priority requests")
Signed-off-by: Zhen Lei <thunder.leizhen@...wei.com>
---
 block/mq-deadline.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/block/mq-deadline.c b/block/mq-deadline.c
index a09761cbdf12e58..d8f6aa12de80049 100644
--- a/block/mq-deadline.c
+++ b/block/mq-deadline.c
@@ -79,6 +79,7 @@ struct dd_per_prio {
 	struct list_head fifo_list[DD_DIR_COUNT];
 	/* Next request in FIFO order. Read, write or both are NULL. */
 	struct request *next_rq[DD_DIR_COUNT];
+	unsigned int nr_queued;
 };
 
 struct deadline_data {
@@ -277,9 +278,9 @@ deadline_move_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
 }
 
 /* Number of requests queued for a given priority level. */
-static u32 dd_queued(struct deadline_data *dd, enum dd_prio prio)
+static __always_inline u32 dd_queued(struct deadline_data *dd, enum dd_prio prio)
 {
-	return dd_sum(dd, inserted, prio) - dd_sum(dd, completed, prio);
+	return dd->per_prio[prio].nr_queued;
 }
 
 /*
@@ -711,6 +712,8 @@ static void dd_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
 
 	prio = ioprio_class_to_prio[ioprio_class];
 	dd_count(dd, inserted, prio);
+	per_prio = &dd->per_prio[prio];
+	per_prio->nr_queued++;
 
 	if (blk_mq_sched_try_insert_merge(q, rq, &free)) {
 		blk_mq_free_requests(&free);
@@ -719,7 +722,6 @@ static void dd_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
 
 	trace_block_rq_insert(rq);
 
-	per_prio = &dd->per_prio[prio];
 	if (at_head) {
 		list_add(&rq->queuelist, &per_prio->dispatch);
 	} else {
@@ -790,12 +792,14 @@ static void dd_finish_request(struct request *rq)
 	const u8 ioprio_class = dd_rq_ioclass(rq);
 	const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
 	struct dd_per_prio *per_prio = &dd->per_prio[prio];
+	unsigned long flags;
 
 	dd_count(dd, completed, prio);
+	spin_lock_irqsave(&dd->lock, flags);
+	per_prio->nr_queued--;
+	spin_unlock_irqrestore(&dd->lock, flags);
 
 	if (blk_queue_is_zoned(q)) {
-		unsigned long flags;
-
 		spin_lock_irqsave(&dd->zone_lock, flags);
 		blk_req_zone_write_unlock(rq);
 		if (!list_empty(&per_prio->fifo_list[DD_WRITE]))
-- 
2.25.1