From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932332AbWFMViR (ORCPT ); Tue, 13 Jun 2006 17:38:17 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932334AbWFMViR (ORCPT ); Tue, 13 Jun 2006 17:38:17 -0400 Received: from e1.ny.us.ibm.com ([32.97.182.141]:8376 "EHLO e1.ny.us.ibm.com") by vger.kernel.org with ESMTP id S932332AbWFMViQ (ORCPT ); Tue, 13 Jun 2006 17:38:16 -0400 Subject: Re: more than 3 GB in userspace (4G/4G patch?) for 2.6.16 From: Dave Hansen To: "H. Peter Anvin" Cc: linux-kernel@vger.kernel.org In-Reply-To: References: <448F207A.6080601@hanno.de> Content-Type: text/plain Date: Tue, 13 Jun 2006 14:37:01 -0700 Message-Id: <1150234621.20249.4.camel@localhost.localdomain> Mime-Version: 1.0 X-Mailer: Evolution 2.4.1 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Tue, 2006-06-13 at 14:30 -0700, H. Peter Anvin wrote: > > A 0.5/3.5 GB split appears to be what I need for my application. > > That is incompatible with PAE. The way PAE works, the split has to be > an even number of gigabytes. Well, there have been ancient patches available in the past. People pop up now and again and ask for them. This is _horribly_ out of date, but it at least indicates that this is possible. The basic idea is that you never share a kernel PMD page. Each process needs to allocate four of its own: http://lkml.org/lkml/2003/5/9/138 -- Dave