From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754355AbXDXVr1 (ORCPT ); Tue, 24 Apr 2007 17:47:27 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1423123AbXDXVr1 (ORCPT ); Tue, 24 Apr 2007 17:47:27 -0400 Received: from 74-93-104-97-Washington.hfc.comcastbusiness.net ([74.93.104.97]:52840 "EHLO sunset.davemloft.net" rhost-flags-OK-FAIL-OK-OK) by vger.kernel.org with ESMTP id S1754180AbXDXVr1 (ORCPT ); Tue, 24 Apr 2007 17:47:27 -0400 Date: Tue, 24 Apr 2007 14:47:33 -0700 (PDT) Message-Id: <20070424.144733.10245273.davem@davemloft.net> To: rdreier@cisco.com Cc: ak@suse.de, ashok.raj@intel.com, linux-kernel@vger.kernel.org, akpm@osdl.org, gregkh@suse.de, muli@il.ibm.com, asit.k.mallick@intel.com, suresh.b.siddha@intel.com, anil.s.keshavamurthy@intel.com, arjan@linux.intel.com, shaohua.li@intel.com Subject: Re: [Intel IOMMU][patch 8/8] Preserve some Virtual Address when devices cannot address entire range. From: David Miller In-Reply-To: References: <200704242312.54738.ak@suse.de> <20070424.142351.62664098.davem@davemloft.net> X-Mailer: Mew version 5.1.52 on Emacs 21.4 / Mule 5.0 (SAKAKI) Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org From: Roland Dreier Date: Tue, 24 Apr 2007 14:32:42 -0700 > > My suggestion would be to allocate top-down in the 32-bit IOMMU space. > > I think that's good for normal things, but it's not unreasonable to > want to map > 4 GB of memory at once for an Infiniband device. That's what the DAC interfaces were for, but people wanted to remoe that since no in-tree users existed. Devices like that want to essentially map the entire address space, via pass-thru, not IOMMU mappings. Clustering cards, such as those made by Dolphin, are another example. I realize that this doesn't work when we absolutely must use an IOMMU such as for virtualization which is the whole impetus of the Intel IOMMU. :-) > So maybe we would want some heuristics about the size of the mapping > being requested or the amount of 32-bit space left to decide whether a > mapping should be below 4 GB. Perhaps, but as I said heuristics might not be enough here.