From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754128AbbJSR7P (ORCPT ); Mon, 19 Oct 2015 13:59:15 -0400 Received: from mx1.redhat.com ([209.132.183.28]:49881 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752594AbbJSR7N (ORCPT ); Mon, 19 Oct 2015 13:59:13 -0400 From: Jeff Moyer To: Ming Lei Cc: Jens Axboe , Linux Kernel Mailing List , Ming Lin , Kent Overstreet , Christoph Hellwig Subject: Re: [PATCH v1 5/6] blk-mq: fix for trace_block_plug() References: <1445097154-18229-1-git-send-email-ming.lei@canonical.com> <1445097154-18229-6-git-send-email-ming.lei@canonical.com> X-PGP-KeyID: 1F78E1B4 X-PGP-CertKey: F6FE 280D 8293 F72C 65FD 5A58 1FF8 A7CA 1F78 E1B4 X-PCLoadLetter: What the f**k does that mean? Date: Mon, 19 Oct 2015 13:59:11 -0400 In-Reply-To: (Ming Lei's message of "Mon, 19 Oct 2015 23:58:39 +0800") Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/24.3 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Ming Lei writes: > On Mon, Oct 19, 2015 at 11:38 PM, Jeff Moyer wrote: >> Ming Lei writes: >> >>> The trace point is for tracing plug event of each request >>> queue instead of each task, so we should check the request >>> count in the plug list from current queue instead of >>> current task. >> >> If blk_queue_nomerges returns true, request_count won't be updated, so >> you've essentially broken plugging for those queues. > > Yes, you are right, and blk_queue_bio() has the same problem too, > and looks automatic flush plug can't work for nomerges queue at all. Hi Ming, Something like this would fix it. Feel free to add it into your patch set if you're going to do another spin. Otherwise I'll just send it off separately. -Jeff From: Jeff Moyer Subject: block: fix plug list flushing for nomerge queues Request queues with merging disabled will not flush the plug list after BLK_MAX_REQUEST_COUNT requests have been queued, since the code relies on blk_attempt_plug_merge to compute the request_count. Fix this by computing the number of queued requests even for nomerge queues. Signed-off-by: Jeff Moyer diff --git a/block/blk-core.c b/block/blk-core.c index 2eb722d..f0ae087 100644 --- a/block/blk-core.c +++ b/block/blk-core.c @@ -1594,6 +1594,30 @@ out: return ret; } +unsigned int blk_plug_queued_count(struct request_queue *q) +{ + struct blk_plug *plug; + struct request *rq; + struct list_head *plug_list; + unsigned int ret = 0; + + plug = current->plug; + if (!plug) + goto out; + + if (q->mq_ops) + plug_list = &plug->mq_list; + else + plug_list = &plug->list; + + list_for_each_entry(rq, plug_list, queuelist) { + if (rq->q == q) + ret++; + } +out: + return ret; +} + void init_request_from_bio(struct request *req, struct bio *bio) { req->cmd_type = REQ_TYPE_FS; @@ -1641,9 +1665,11 @@ static void blk_queue_bio(struct request_queue *q, struct bio *bio) * Check if we can merge with the plugged list before grabbing * any locks. */ - if (!blk_queue_nomerges(q) && - blk_attempt_plug_merge(q, bio, &request_count, NULL)) - return; + if (!blk_queue_nomerges(q)) { + if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) + return; + } else + request_count = blk_plug_queued_count(q); spin_lock_irq(q->queue_lock); diff --git a/block/blk-mq.c b/block/blk-mq.c index 7785ae9..604a1f3 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -1267,9 +1267,12 @@ static void blk_mq_make_request(struct request_queue *q, struct bio *bio) blk_queue_split(q, &bio, q->bio_split); - if (!is_flush_fua && !blk_queue_nomerges(q) && - blk_attempt_plug_merge(q, bio, &request_count, &same_queue_rq)) - return; + if (!is_flush_fua && !blk_queue_nomerges(q)) { + if (blk_attempt_plug_merge(q, bio, &request_count, + &same_queue_rq)) + return; + } else + request_count = blk_plug_queued_count(q); rq = blk_mq_map_request(q, bio, &data); if (unlikely(!rq)) diff --git a/block/blk.h b/block/blk.h index 98614ad..aa27d02 100644 --- a/block/blk.h +++ b/block/blk.h @@ -86,6 +86,7 @@ bool bio_attempt_back_merge(struct request_queue *q, struct request *req, bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, unsigned int *request_count, struct request **same_queue_rq); +unsigned int blk_plug_queued_count(struct request_queue *q); void blk_account_io_start(struct request *req, bool new_io); void blk_account_io_completion(struct request *req, unsigned int bytes);