From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932719AbXGTAoV (ORCPT ); Thu, 19 Jul 2007 20:44:21 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1759894AbXGTAoO (ORCPT ); Thu, 19 Jul 2007 20:44:14 -0400 Received: from holomorphy.com ([66.93.40.71]:53544 "EHLO holomorphy.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758940AbXGTAoN (ORCPT ); Thu, 19 Jul 2007 20:44:13 -0400 Date: Thu, 19 Jul 2007 17:45:59 -0700 From: William Lee Irwin III To: Christoph Hellwig Cc: Andi Kleen , lkml@rtr.ca, linux-kernel@vger.kernel.org Subject: Re: [PATCH for review] [7/48] i386: divorce CONFIG_X86_PAE from CONFIG_HIGHMEM64G Message-ID: <20070720004559.GO11781@holomorphy.com> References: <20070719348.540885000@suse.de> <20070719134836.0C54214E6E@wotan.suse.de> <20070719145229.GA19858@infradead.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20070719145229.GA19858@infradead.org> Organization: The Domain of Holomorphy User-Agent: Mutt/1.5.13 (2006-08-11) Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org From: William Lee Irwin III >> PAE is useful for more than supporting more than 4GB RAM. It supports >> expanded swapspace and NX executable protections. Some users may want NX >> or expanded swapspace support without the overhead or instability of >> highmem. For these reasons, the following patch divorces CONFIG_X86_PAE >> from CONFIG_HIGHMEM64G. On Thu, Jul 19, 2007 at 03:52:29PM +0100, Christoph Hellwig wrote: > What overhead of instability of highmem? Sorry folks but this is utter > bollocks. Back in the Caldera days we did a lot of measurement on highmem > overhead, and CONFIG_HIGHMEM has no measurable overhead at all on a system > that doesn't use it. CONFIG_HIGHMEM64G on the other hand has > a quite visible overhead on small systems, but that's entirely due to the > bigger page table entries that you need for NX. The missing context here is CONFIG_VMSPLIT on laptops. Laptop users, who frequently use CONFIG_VMSPLIT options to avoid highmem, wanted to turn on NX. Prior to the patch, those options were barred for all highmem configurations. In response to those requests, I produced the patch. The overhead and instability derived from tiny zones as opposed to kmap()/kunmap(), or at least such was the case historically. -- wli