From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1750716AbWDRIuJ (ORCPT ); Tue, 18 Apr 2006 04:50:09 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1750913AbWDRIuI (ORCPT ); Tue, 18 Apr 2006 04:50:08 -0400 Received: from 167.imtp.Ilyichevsk.Odessa.UA ([195.66.192.167]:23451 "HELO ilport.com.ua") by vger.kernel.org with SMTP id S1750716AbWDRIuH (ORCPT ); Tue, 18 Apr 2006 04:50:07 -0400 From: Denis Vlasenko To: Jan Engelhardt Subject: Re: Flexible mem= parameter Date: Tue, 18 Apr 2006 11:49:21 +0300 User-Agent: KMail/1.8.2 Cc: Linux Kernel Mailing List References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200604181149.22108.vda@ilport.com.ua> Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Friday 14 April 2006 01:49, Jan Engelhardt wrote: > Hello, > > > mem= can be used to limit the amount of physical memory Linux uses. I > presume that when specifying mem=256M only the memory from 0x0 to > 0x10000000 is used. Is there a way to tune mem= so that it will use > 0x70000000 to 0x80000000? (A compile-time change would suffice.) On i386 this works: "mem=exactmap mem=640K@0 mem=79M@1M" > Background: I am trying to track down a nondeterministic (but > reproducible) segfault while building big projects like gcc or glibc. I do > not think it's a memory problem since it only showed up since I changed to > gcc 4.1.0 and tweaked it to use -mcpu=ultrasparc (rather than -mcpu=v7) by > default. I had been building gcc4 before with gcc3 a number of times w/o > problems. I did ran a userspace memtester for approx 20 hours (`memtest` > from debian). -- vda