From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S273067AbTG3RVM (ORCPT ); Wed, 30 Jul 2003 13:21:12 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S273070AbTG3RVM (ORCPT ); Wed, 30 Jul 2003 13:21:12 -0400 Received: from nat9.steeleye.com ([65.114.3.137]:53508 "EHLO hancock.sc.steeleye.com") by vger.kernel.org with ESMTP id S273067AbTG3RVL (ORCPT ); Wed, 30 Jul 2003 13:21:11 -0400 Subject: Re: [RFC] block layer support for DMA IOMMU bypass mode II From: James Bottomley To: Andi Kleen Cc: Grant Grundler , "David S. Miller" , Alan Cox , Jens Axboe , suparna@in.ibm.com, Linux Kernel , alex_williamson@hp.com, bjorn_helgaas@hp.com In-Reply-To: <20030730163615.GD17201@wotan.suse.de> References: <20030708213427.39de0195.ak@suse.de> <20030708.150433.104048841.davem@redhat.com> <20030708222545.GC6787@dsl2.external.hp.com> <20030708.152314.115928676.davem@redhat.com> <20030723114006.GA28688@dsl2.external.hp.com> <20030728131513.5d4b1bd3.ak@suse.de> <20030730044256.GA1974@dsl2.external.hp.com> <20030729215118.13a5ac18.davem@redhat.com> <20030730160250.GA16960@dsl2.external.hp.com> <20030730163615.GD17201@wotan.suse.de> Content-Type: text/plain Content-Transfer-Encoding: 7bit X-Mailer: Ximian Evolution 1.0.8 (1.0.8-9) Date: 30 Jul 2003 12:18:46 -0500 Message-Id: <1059585657.1850.197.camel@mulgrave> Mime-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Wed, 2003-07-30 at 11:36, Andi Kleen wrote: > The differences were greater with the mpt fusion driver, maybe it has > more overhead. Or your IO subsystem is significantly different. By and large, these results are more like what I expect. As I've said before, getting SG tables to work efficiently is a core part of getting an I/O board to function. There are two places vmerging can help: 1. Reducing the size of the SG table 2. Increasing the length of the I/O for devices with fixed (but small) SG table lengths. However, it's important to remember that vmerging comes virtually for free in the BIO layer, so the only added cost is the programming of the IOMMU. This isn't an issue on SPARC, PA-RISC and the like where IOMMU programming is required to do I/O, it may be something the IOMMU optional architectures (like IA-64 and AMD-64) should consider, which is where I entered with the IOMMU bypass patch. James