From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S964842AbVHaPjO (ORCPT ); Wed, 31 Aug 2005 11:39:14 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S964852AbVHaPjN (ORCPT ); Wed, 31 Aug 2005 11:39:13 -0400 Received: from dvhart.com ([64.146.134.43]:63622 "EHLO localhost.localdomain") by vger.kernel.org with ESMTP id S964842AbVHaPjM (ORCPT ); Wed, 31 Aug 2005 11:39:12 -0400 Date: Wed, 31 Aug 2005 08:39:13 -0700 From: "Martin J. Bligh" Reply-To: "Martin J. Bligh" To: Hugh Dickins Cc: Arjan van de Ven , Dave McCracken , Andrew Morton , Linux Kernel , Linux Memory Management Subject: Re: [PATCH 1/1] Implement shared page tables Message-ID: <19490000.1125502753@[10.10.2.4]> In-Reply-To: References: <7C49DFF721CB4E671DB260F9@[10.1.1.4]><1125489077.3213.12.camel@laptopd505.fenrus.org><16640000.1125498711@[10.10.2.4]> X-Mailer: Mulberry/2.2.1 (Linux/x86) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Content-Disposition: inline Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org >> They're incompatible, but you could be left to choose one or the other >> via config option. > > Wouldn't need config option: there's /proc/sys/kernel/randomize_va_space > for the whole running system, compatibility check on the ELFs run, and > the infinite stack rlimit: enough ways to suppress randomization if it > doesn't suit you. Even better - much easier to deal with distro stuff if we can do it at runtime. >> 3% on "a certain industry-standard database benchmark" (cough) is huge, >> and we expect the benefit for PPC64 will be larger as we can share the >> underlying hardware PTEs without TLB flushing as well. > > Okay - and you're implying that 3% comes from _using_ the shared page > tables, rather than from avoiding the fork/exit overhead of setting > them up and tearing them down. And it can't use huge TLB pages > because... fragmentation? Yes - as I understand it, that was a straight measurement with/without the patch, and the shmem segment was already using hugetlb (in both cases). Yes, I find that a bit odd as to why as well - they are still trying to get some detailed profiling to explain. M.