From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-2.5 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, MAILING_LIST_MULTI,SPF_PASS,USER_AGENT_MUTT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id C2625C43381 for ; Mon, 18 Mar 2019 15:16:30 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 9E4A22085A for ; Mon, 18 Mar 2019 15:16:30 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726844AbfCRPQa (ORCPT ); Mon, 18 Mar 2019 11:16:30 -0400 Received: from mx1.redhat.com ([209.132.183.28]:49846 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726837AbfCRPQ3 (ORCPT ); Mon, 18 Mar 2019 11:16:29 -0400 Received: from smtp.corp.redhat.com (int-mx05.intmail.prod.int.phx2.redhat.com [10.5.11.15]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 9AC5F4E908; Mon, 18 Mar 2019 15:16:29 +0000 (UTC) Received: from ming.t460p (ovpn-8-17.pek2.redhat.com [10.72.8.17]) by smtp.corp.redhat.com (Postfix) with ESMTPS id 481C45D70E; Mon, 18 Mar 2019 15:16:23 +0000 (UTC) Date: Mon, 18 Mar 2019 23:16:19 +0800 From: Ming Lei To: Bart Van Assche Cc: Jens Axboe , linux-block@vger.kernel.org, Christoph Hellwig , linux-nvme@lists.infradead.org Subject: Re: [PATCH 1/2] blk-mq: introduce blk_mq_complete_request_sync() Message-ID: <20190318151618.GA20371@ming.t460p> References: <20190318032950.17770-1-ming.lei@redhat.com> <20190318032950.17770-2-ming.lei@redhat.com> <20190318073826.GA29746@ming.t460p> <1552921495.152266.8.camel@acm.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1552921495.152266.8.camel@acm.org> User-Agent: Mutt/1.9.1 (2017-09-22) X-Scanned-By: MIMEDefang 2.79 on 10.5.11.15 X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.38]); Mon, 18 Mar 2019 15:16:29 +0000 (UTC) Sender: linux-block-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org On Mon, Mar 18, 2019 at 08:04:55AM -0700, Bart Van Assche wrote: > On Mon, 2019-03-18 at 15:38 +0800, Ming Lei wrote: > > On Sun, Mar 17, 2019 at 09:09:09PM -0700, Bart Van Assche wrote: > > > On 3/17/19 8:29 PM, Ming Lei wrote: > > > > In NVMe's error handler, follows the typical steps for tearing down > > > > hardware: > > > > > > > > 1) stop blk_mq hw queues > > > > 2) stop the real hw queues > > > > 3) cancel in-flight requests via > > > > blk_mq_tagset_busy_iter(tags, cancel_request, ...) > > > > cancel_request(): > > > > mark the request as abort > > > > blk_mq_complete_request(req); > > > > 4) destroy real hw queues > > > > > > > > However, there may be race between #3 and #4, because blk_mq_complete_request() > > > > actually completes the request asynchronously. > > > > > > > > This patch introduces blk_mq_complete_request_sync() for fixing the > > > > above race. > > > > > > Other block drivers wait until outstanding requests have completed by > > > calling blk_cleanup_queue() before hardware queues are destroyed. Why can't > > > the NVMe driver follow that approach? > > > > The tearing down of controller can be done in error handler, in which > > the request queues may not be cleaned up, almost all kinds of NVMe > > controller's error handling follows the above steps, such as: > > > > nvme_rdma_error_recovery_work() > > ->nvme_rdma_teardown_io_queues() > > > > nvme_timeout() > > ->nvme_dev_disable > > Hi Ming, > > This makes me wonder whether the current design of the NVMe core is the best > design we can come up with? The structure of e.g. the SRP initiator and target > drivers is similar to the NVMeOF drivers. However, there is no need in the SRP > initiator driver to terminate requests synchronously. Is this due to I am not familiar with SRP, could you explain what SRP initiator driver will do when the controller is in bad state? Especially about dealing with in-flight IO requests under this situation. > differences in the error handling approaches in the SCSI and NVMe core drivers? As far as I can tell, I don't see obvious design issue in NVMe host drivers, which tries best to recover controller and retries to complete all in-flight IO. Thanks, Ming From mboxrd@z Thu Jan 1 00:00:00 1970 From: ming.lei@redhat.com (Ming Lei) Date: Mon, 18 Mar 2019 23:16:19 +0800 Subject: [PATCH 1/2] blk-mq: introduce blk_mq_complete_request_sync() In-Reply-To: <1552921495.152266.8.camel@acm.org> References: <20190318032950.17770-1-ming.lei@redhat.com> <20190318032950.17770-2-ming.lei@redhat.com> <20190318073826.GA29746@ming.t460p> <1552921495.152266.8.camel@acm.org> Message-ID: <20190318151618.GA20371@ming.t460p> On Mon, Mar 18, 2019@08:04:55AM -0700, Bart Van Assche wrote: > On Mon, 2019-03-18@15:38 +0800, Ming Lei wrote: > > On Sun, Mar 17, 2019@09:09:09PM -0700, Bart Van Assche wrote: > > > On 3/17/19 8:29 PM, Ming Lei wrote: > > > > In NVMe's error handler, follows the typical steps for tearing down > > > > hardware: > > > > > > > > 1) stop blk_mq hw queues > > > > 2) stop the real hw queues > > > > 3) cancel in-flight requests via > > > > blk_mq_tagset_busy_iter(tags, cancel_request, ...) > > > > cancel_request(): > > > > mark the request as abort > > > > blk_mq_complete_request(req); > > > > 4) destroy real hw queues > > > > > > > > However, there may be race between #3 and #4, because blk_mq_complete_request() > > > > actually completes the request asynchronously. > > > > > > > > This patch introduces blk_mq_complete_request_sync() for fixing the > > > > above race. > > > > > > Other block drivers wait until outstanding requests have completed by > > > calling blk_cleanup_queue() before hardware queues are destroyed. Why can't > > > the NVMe driver follow that approach? > > > > The tearing down of controller can be done in error handler, in which > > the request queues may not be cleaned up, almost all kinds of NVMe > > controller's error handling follows the above steps, such as: > > > > nvme_rdma_error_recovery_work() > > ->nvme_rdma_teardown_io_queues() > > > > nvme_timeout() > > ->nvme_dev_disable > > Hi Ming, > > This makes me wonder whether the current design of the NVMe core is the best > design we can come up with? The structure of e.g. the SRP initiator and target > drivers is similar to the NVMeOF drivers. However, there is no need in the SRP > initiator driver to terminate requests synchronously. Is this due to I am not familiar with SRP, could you explain what SRP initiator driver will do when the controller is in bad state? Especially about dealing with in-flight IO requests under this situation. > differences in the error handling approaches in the SCSI and NVMe core drivers? As far as I can tell, I don't see obvious design issue in NVMe host drivers, which tries best to recover controller and retries to complete all in-flight IO. Thanks, Ming