From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753976AbbDMJtr (ORCPT ); Mon, 13 Apr 2015 05:49:47 -0400 Received: from verein.lst.de ([213.95.11.211]:60225 "EHLO newverein.lst.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753510AbbDMJtp (ORCPT ); Mon, 13 Apr 2015 05:49:45 -0400 Date: Mon, 13 Apr 2015 11:49:44 +0200 From: Christoph Hellwig To: Jens Axboe Cc: axboe@kernel.dk, linux-kernel@vger.kernel.org, linux-scsi@vger.kernel.org, hch@lst.de Subject: Re: [PATCH 4/6] mpt2sas: store scsi io tracker data in the scsi request Message-ID: <20150413094944.GE5259@lst.de> References: <1428427120-20971-1-git-send-email-axboe@fb.com> <1428427120-20971-5-git-send-email-axboe@fb.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1428427120-20971-5-git-send-email-axboe@fb.com> User-Agent: Mutt/1.5.17 (2007-11-01) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tue, Apr 07, 2015 at 11:18:38AM -0600, Jens Axboe wrote: > Instead of storing the IO tracker structure in a separate list > that we need to pop/push to on every submit and complete (and > lock), store it in the pdu associated with a request. This is > possible on scsi-mq only, and further cuts the spinlock associated > time for higher IOPS IO workloads. At 100K IOPS, this effectively > cuts the locking time in half. As mentioned before please call scsi_init_shared_tag_map and use your new code path for the !mq case as well.