From patchwork Mon Aug 26 07:48:43 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Yu Kuai X-Patchwork-Id: 13777384 Received: from dggsgout12.his.huawei.com (dggsgout12.his.huawei.com [45.249.212.56]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id A34C113E039; Mon, 26 Aug 2024 07:53:41 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=45.249.212.56 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1724658823; cv=none; b=MzUyseN8i7lk5WWRS74pApqDbJmfeVZGr3tFaGWmB+zBsZ6EUj3pZBD5YmLVsB0chYFMpUZNTj8jJa3141R8NdCTrhUPzyUmhefPi08jPEGD6VMi3fFsSUkiNQERm9j3JbrbP7YxZyzIqSK+bMAiqUgceV2YGfhXQySkuCCtgso= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1724658823; c=relaxed/simple; bh=89S7aqW0XWcT8u4h7lIkLcU4uOiqz5+yPwST6RasXGk=; h=From:To:Cc:Subject:Date:Message-Id:MIME-Version; b=naGwJbDO/DkfLmUxCJcMvj7U/cA13SYQDcujKlYTG+DdExllDr3w3EOD+Xl19CJGIBNvhbg3GXn9OxU8gY1OPRyQUOPAixs7dlf8NWN4LE0kdPPliFQSDvQCeyZldz3UC2IVbGPLX6wLQsF3ZrLa9uSIKbfWP+EDITaZ6vqkm4I= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=none (p=none dis=none) header.from=huaweicloud.com; spf=pass smtp.mailfrom=huaweicloud.com; arc=none smtp.client-ip=45.249.212.56 Authentication-Results: smtp.subspace.kernel.org; dmarc=none (p=none dis=none) header.from=huaweicloud.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=huaweicloud.com Received: from mail.maildlp.com (unknown [172.19.163.235]) by dggsgout12.his.huawei.com (SkyGuard) with ESMTP id 4WsjbS25JTz4f3jLy; Mon, 26 Aug 2024 15:53:24 +0800 (CST) Received: from mail02.huawei.com (unknown [10.116.40.128]) by mail.maildlp.com (Postfix) with ESMTP id B9DFC1A058E; Mon, 26 Aug 2024 15:53:38 +0800 (CST) Received: from huaweicloud.com (unknown [10.175.104.67]) by APP4 (Coremail) with SMTP id gCh0CgB3n4WBNMxmd6QECw--.16182S4; Mon, 26 Aug 2024 15:53:38 +0800 (CST) From: Yu Kuai To: song@kernel.org Cc: linux-raid@vger.kernel.org, linux-kernel@vger.kernel.org, yukuai3@huawei.com, yukuai1@huaweicloud.com, yi.zhang@huawei.com, yangerkun@huawei.com Subject: [PATCH md-6.12] md: remove flush handling Date: Mon, 26 Aug 2024 15:48:43 +0800 Message-Id: <20240826074843.1575099-1-yukuai1@huaweicloud.com> X-Mailer: git-send-email 2.39.2 Precedence: bulk X-Mailing-List: linux-raid@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-CM-TRANSID: gCh0CgB3n4WBNMxmd6QECw--.16182S4 X-Coremail-Antispam: 1UD129KBjvJXoWxtF17Kw43ZF17Ww43GF47twb_yoW3Jr15p3 ySg3ZxJr48XrZ0qw1DJFWkXr1Fgw1xKayDtrW3Z34xZr13Jrn8GayFqryFqr98A3s3urW5 Ww4kt3yDuryjqFUanT9S1TB71UUUUU7qnTZGkaVYY2UrUUUUjbIjqfuFe4nvWSU5nxnvy2 9KBjDU0xBIdaVrnRJUUUkE14x267AKxVW8JVW5JwAFc2x0x2IEx4CE42xK8VAvwI8IcIk0 rVWrJVCq3wAFIxvE14AKwVWUJVWUGwA2ocxC64kIII0Yj41l84x0c7CEw4AK67xGY2AK02 1l84ACjcxK6xIIjxv20xvE14v26F1j6w1UM28EF7xvwVC0I7IYx2IY6xkF7I0E14v26F4U JVW0owA2z4x0Y4vEx4A2jsIE14v26rxl6s0DM28EF7xvwVC2z280aVCY1x0267AKxVW0oV Cq3wAS0I0E0xvYzxvE52x082IY62kv0487Mc02F40EFcxC0VAKzVAqx4xG6I80ewAv7VC0 I7IYx2IY67AKxVWUGVWUXwAv7VC2z280aVAFwI0_Jr0_Gr1lOx8S6xCaFVCjc4AY6r1j6r 4UM4x0Y48IcxkI7VAKI48JM4x0x7Aq67IIx4CEVc8vx2IErcIFxwCY1x0262kKe7AKxVWU AVWUtwCF04k20xvY0x0EwIxGrwCFx2IqxVCFs4IE7xkEbVWUJVW8JwC20s026c02F40E14 v26r1j6r18MI8I3I0E7480Y4vE14v26r106r1rMI8E67AF67kF1VAFwI0_JF0_Jw1lIxkG c2Ij64vIr41lIxAIcVC0I7IYx2IY67AKxVW7JVWDJwCI42IY6xIIjxv20xvEc7CjxVAFwI 0_Gr1j6F4UJwCI42IY6xAIw20EY4v20xvaj40_Jr0_JF4lIxAIcVC2z280aVAFwI0_Gr0_ Cr1lIxAIcVC2z280aVCY1x0267AKxVW8Jr0_Cr1UYxBIdaVFxhVjvjDU0xZFpf9x0JUG0P hUUUUU= X-CM-SenderInfo: 51xn3trlr6x35dzhxuhorxvhhfrp/ From: Yu Kuai For flush request, md has a special flush handling to merge concurrent flush request into single one, however, the whole mechanism is based on a disk level spin_lock 'mddev->lock'. And fsync can be called quite often in some user cases, for consequence, spin lock from IO fast path can cause performance degration. Fortunately, the block layer already have flush handling to merge concurrent flush request, and it only acquire hctx level spin lock(see details in blk-flush.c). This patch remove the flush handling in md, and convert to use general block layer flush handling in underlying disks. Flush test for 4 nvme raid10: start 128 threads to do fsync 100000 times, on arm64, see how long it takes. Test result: about 10 times faster for high concurrency. Before this patch: 50943374 microseconds After this patch: 5096347 microseconds BTW, this patch can fix the same problem as commit 611d5cbc0b35 ("md: fix deadlock between mddev_suspend and flush bio"). Signed-off-by: Yu Kuai --- drivers/md/md.c | 137 ++++-------------------------------------------- drivers/md/md.h | 10 ---- 2 files changed, 11 insertions(+), 136 deletions(-) diff --git a/drivers/md/md.c b/drivers/md/md.c index a38981de8901..4d675f7cc2a7 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -546,137 +546,23 @@ static int mddev_set_closing_and_sync_blockdev(struct mddev *mddev, int opener_n return 0; } -/* - * Generic flush handling for md - */ - -static void md_end_flush(struct bio *bio) -{ - struct md_rdev *rdev = bio->bi_private; - struct mddev *mddev = rdev->mddev; - - bio_put(bio); - - rdev_dec_pending(rdev, mddev); - - if (atomic_dec_and_test(&mddev->flush_pending)) - /* The pre-request flush has finished */ - queue_work(md_wq, &mddev->flush_work); -} - -static void md_submit_flush_data(struct work_struct *ws); - -static void submit_flushes(struct work_struct *ws) +bool md_flush_request(struct mddev *mddev, struct bio *bio) { - struct mddev *mddev = container_of(ws, struct mddev, flush_work); struct md_rdev *rdev; + struct bio *new; - mddev->start_flush = ktime_get_boottime(); - INIT_WORK(&mddev->flush_work, md_submit_flush_data); - atomic_set(&mddev->flush_pending, 1); - rcu_read_lock(); - rdev_for_each_rcu(rdev, mddev) - if (rdev->raid_disk >= 0 && - !test_bit(Faulty, &rdev->flags)) { - struct bio *bi; - - atomic_inc(&rdev->nr_pending); - rcu_read_unlock(); - bi = bio_alloc_bioset(rdev->bdev, 0, - REQ_OP_WRITE | REQ_PREFLUSH, - GFP_NOIO, &mddev->bio_set); - bi->bi_end_io = md_end_flush; - bi->bi_private = rdev; - atomic_inc(&mddev->flush_pending); - submit_bio(bi); - rcu_read_lock(); - } - rcu_read_unlock(); - if (atomic_dec_and_test(&mddev->flush_pending)) - queue_work(md_wq, &mddev->flush_work); -} - -static void md_submit_flush_data(struct work_struct *ws) -{ - struct mddev *mddev = container_of(ws, struct mddev, flush_work); - struct bio *bio = mddev->flush_bio; - - /* - * must reset flush_bio before calling into md_handle_request to avoid a - * deadlock, because other bios passed md_handle_request suspend check - * could wait for this and below md_handle_request could wait for those - * bios because of suspend check - */ - spin_lock_irq(&mddev->lock); - mddev->prev_flush_start = mddev->start_flush; - mddev->flush_bio = NULL; - spin_unlock_irq(&mddev->lock); - wake_up(&mddev->sb_wait); - - if (bio->bi_iter.bi_size == 0) { - /* an empty barrier - all done */ - bio_endio(bio); - } else { - bio->bi_opf &= ~REQ_PREFLUSH; - - /* - * make_requst() will never return error here, it only - * returns error in raid5_make_request() by dm-raid. - * Since dm always splits data and flush operation into - * two separate io, io size of flush submitted by dm - * always is 0, make_request() will not be called here. - */ - if (WARN_ON_ONCE(!mddev->pers->make_request(mddev, bio))) - bio_io_error(bio); - } - - /* The pair is percpu_ref_get() from md_flush_request() */ - percpu_ref_put(&mddev->active_io); -} + rdev_for_each(rdev, mddev) { + if (rdev->raid_disk < 0 || test_bit(Faulty, &rdev->flags)) + continue; -/* - * Manages consolidation of flushes and submitting any flushes needed for - * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is - * being finished in another context. Returns false if the flushing is - * complete but still needs the I/O portion of the bio to be processed. - */ -bool md_flush_request(struct mddev *mddev, struct bio *bio) -{ - ktime_t req_start = ktime_get_boottime(); - spin_lock_irq(&mddev->lock); - /* flush requests wait until ongoing flush completes, - * hence coalescing all the pending requests. - */ - wait_event_lock_irq(mddev->sb_wait, - !mddev->flush_bio || - ktime_before(req_start, mddev->prev_flush_start), - mddev->lock); - /* new request after previous flush is completed */ - if (ktime_after(req_start, mddev->prev_flush_start)) { - WARN_ON(mddev->flush_bio); - /* - * Grab a reference to make sure mddev_suspend() will wait for - * this flush to be done. - * - * md_flush_reqeust() is called under md_handle_request() and - * 'active_io' is already grabbed, hence percpu_ref_is_zero() - * won't pass, percpu_ref_tryget_live() can't be used because - * percpu_ref_kill() can be called by mddev_suspend() - * concurrently. - */ - WARN_ON(percpu_ref_is_zero(&mddev->active_io)); - percpu_ref_get(&mddev->active_io); - mddev->flush_bio = bio; - spin_unlock_irq(&mddev->lock); - INIT_WORK(&mddev->flush_work, submit_flushes); - queue_work(md_wq, &mddev->flush_work); - return true; + new = bio_alloc_bioset(rdev->bdev, 0, + REQ_OP_WRITE | REQ_PREFLUSH, GFP_NOIO, + &mddev->bio_set); + bio_chain(new, bio); + submit_bio(new); } - /* flush was performed for some other bio while we waited. */ - spin_unlock_irq(&mddev->lock); - if (bio->bi_iter.bi_size == 0) { - /* pure flush without data - all done */ + if (bio_sectors(bio) == 0) { bio_endio(bio); return true; } @@ -763,7 +649,6 @@ int mddev_init(struct mddev *mddev) atomic_set(&mddev->openers, 0); atomic_set(&mddev->sync_seq, 0); spin_lock_init(&mddev->lock); - atomic_set(&mddev->flush_pending, 0); init_waitqueue_head(&mddev->sb_wait); init_waitqueue_head(&mddev->recovery_wait); mddev->reshape_position = MaxSector; diff --git a/drivers/md/md.h b/drivers/md/md.h index 1c6a5f41adca..5d2e6bd58e4d 100644 --- a/drivers/md/md.h +++ b/drivers/md/md.h @@ -572,16 +572,6 @@ struct mddev { */ struct bio_set io_clone_set; - /* Generic flush handling. - * The last to finish preflush schedules a worker to submit - * the rest of the request (without the REQ_PREFLUSH flag). - */ - struct bio *flush_bio; - atomic_t flush_pending; - ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed - * flush was started. - */ - struct work_struct flush_work; struct work_struct event_work; /* used by dm to report failure event */ mempool_t *serial_info_pool; void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);