Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1750753AbVLXXhV (ORCPT ); Sat, 24 Dec 2005 18:37:21 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1750756AbVLXXhV (ORCPT ); Sat, 24 Dec 2005 18:37:21 -0500 Received: from shawidc-mo1.cg.shawcable.net ([24.71.223.10]:44930 "EHLO pd2mo2so.prod.shaw.ca") by vger.kernel.org with ESMTP id S1750753AbVLXXhU (ORCPT ); Sat, 24 Dec 2005 18:37:20 -0500 Date: Sat, 24 Dec 2005 17:37:07 -0600 From: Robert Hancock Subject: Re: pci-dma disables iommu on nforce4 motherboards? In-reply-to: <5njvv-JD-11@gated-at.bofh.it> To: linux-kernel Cc: safemode@comcast.net Message-id: <43ADDBA3.9010701@shaw.ca> MIME-version: 1.0 Content-type: text/plain; charset=ISO-8859-1; format=flowed Content-transfer-encoding: 7bit X-Accept-Language: en-us, en References: <5njvv-JD-11@gated-at.bofh.it> User-Agent: Mozilla Thunderbird 1.0.7 (Windows/20050923) Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1589 Lines: 33 Ed Sweetman wrote: > I have an asus A8N-E motherboard and recieve the following message on boot. > PCI-DMA: Disabling IOMMU. > > I have no issues with anything not functioning. I guess i'm just > curious as to why this is done and if i'm missing out on any sort of > performance gain by not using the iommu. I have less than 4GB of ram, > would that be why it's disabled (which is why i think it is)? - The IOMMU is not needed if your RAM all lies below 4GB (note that due to memory space used for PCI and PCI-E resources, even with only 4GB of memory, some may end up above 4GB). The purpose of the IOMMU is to allow 32-bit devices which cannot access memory above 4GB to read from such memory. If the system does not have an IOMMU (i.e. the Intel CPUs with EM64T) then bounce buffers must be used when these devices want to perform DMA to memory above 4GB, which reduces performance. On some platforms the IOMMU can be used to remap memory such that a discontigous memory region appears contiguous to the device, so that it can perform DMA transfers in larger chunks. I suspect the performance benefit of this is somewhat negated by the time to set up the IOMMU mapping, however. -- Robert Hancock Saskatoon, SK, Canada To email, remove "nospam" from hancockr@nospamshaw.ca Home Page: http://www.roberthancock.com/ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/