Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757769AbXFMWUp (ORCPT ); Wed, 13 Jun 2007 18:20:45 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1755778AbXFMWUY (ORCPT ); Wed, 13 Jun 2007 18:20:24 -0400 Received: from ebiederm.dsl.xmission.com ([166.70.28.69]:37671 "EHLO ebiederm.dsl.xmission.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755481AbXFMWUW (ORCPT ); Wed, 13 Jun 2007 18:20:22 -0400 From: ebiederm@xmission.com (Eric W. Biederman) To: Jesse Barnes Cc: Jesse Barnes , Andi Kleen , linux-kernel@vger.kernel.org, Justin Piszcz Subject: Re: [PATCH] trim memory not covered by WB MTRRs References: <200706061229.24486.jesse.barnes@intel.com> <200706121929.07527.jbarnes@virtuousgeek.org> Date: Wed, 13 Jun 2007 16:19:04 -0600 In-Reply-To: <200706121929.07527.jbarnes@virtuousgeek.org> (Jesse Barnes's message of "Tue, 12 Jun 2007 19:29:07 -0700") Message-ID: User-Agent: Gnus/5.110006 (No Gnus v0.6) Emacs/21.4 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2398 Lines: 54 Jesse Barnes writes: > On Tuesday, June 12, 2007 6:11:21 Eric W. Biederman wrote: >> Jesse Barnes writes: >> > On some machines, buggy BIOSes don't properly setup WB MTRRs to >> > cover all available RAM, meaning the last few megs (or even gigs) >> > of memory will be marked uncached. Since Linux tends to allocate >> > from high memory addresses first, this causes the machine to be >> > unusably slow as soon as the kernel starts really using memory >> > (i.e. right around init time). >> > >> > This patch works around the problem by scanning the MTRRs at >> > boot and figuring out whether the current end_pfn value (setup >> > by early e820 code) goes beyond the highest WB MTRR range, and >> > if so, trimming it to match. A fairly obnoxious KERN_WARNING >> > is printed too, letting the user know that not all of their >> > memory is available due to a likely BIOS bug. >> >> A quick update. This patch is horribly incorrect on a socket F >> opteron/Athlon 64 with memory above 4GB. >> >> In particular those cpus are capable of mapping all of memory >> above 4GB as write back without using a single MTRR. >> >> So examining MTRRs is insufficient. > > Hm, yuck. What do you suggest? Should we only run this check when Intel > chips are present? Checking only the bottom 4G isn't sufficient since we've > seen platforms that have issues above that range... My gut feel says that we need to call a function that is potentially cpu specific, older AMD cpus and Intel cpus can just use the generic mtrr code. I would also suggest we build a list of ranges of write-back memory. Which until we handle overlapping MTRRs in the generic MTRR case is just the write-back MTRRs. Then we get the data in a linux specific form we can check the linux specific data structure against the e820 map. I don't think that is going to much harder and it allows for creative cpu designers. Although this suggests that we want to worry about all memory holes as well. Because I have seen at least one system which failed to cover the lower 4G with MTRRs. While everything above 4G was fine. Eric - 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/