From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754208AbYEZJMR (ORCPT ); Mon, 26 May 2008 05:12:17 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751457AbYEZJMJ (ORCPT ); Mon, 26 May 2008 05:12:09 -0400 Received: from mx3.mail.elte.hu ([157.181.1.138]:40698 "EHLO mx3.mail.elte.hu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751298AbYEZJMI (ORCPT ); Mon, 26 May 2008 05:12:08 -0400 Date: Mon, 26 May 2008 11:11:55 +0200 From: Ingo Molnar To: Vegard Nossum Cc: Pekka Enberg , linux-kernel@vger.kernel.org, Pekka Paalanen Subject: Re: [PATCH] kmemcheck: SMP support Message-ID: <20080526091155.GE13529@elte.hu> References: <20080523141759.GA1833@damson.getinternet.no> <20080523150622.GA9782@elte.hu> <19f34abd0805230830o3af93956h8152de3f6e350a09@mail.gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <19f34abd0805230830o3af93956h8152de3f6e350a09@mail.gmail.com> User-Agent: Mutt/1.5.18 (2008-05-17) X-ELTE-VirusStatus: clean X-ELTE-SpamScore: -1.5 X-ELTE-SpamLevel: X-ELTE-SpamCheck: no X-ELTE-SpamVersion: ELTE 2.0 X-ELTE-SpamCheck-Details: score=-1.5 required=5.9 tests=BAYES_00 autolearn=no SpamAssassin version=3.2.3 -1.5 BAYES_00 BODY: Bayesian spam probability is 0 to 1% [score: 0.0000] Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Vegard Nossum wrote: > On Fri, May 23, 2008 at 5:06 PM, Ingo Molnar wrote: > > > > Vegard, wanna have a look at introducing per CPU kernel pagetables? I > > tried that once in the past and it wasnt too horrible. (the patches are > > gone though) We could do it before bringing other CPUs online, i.e. much > > of the really yucky boot time pagetable juggling phase would be over > > already. Hm? > > Ingo. > > It really doesn't matter how easy it was for you. > > You're one of the x86 maintainers. > > And I think you're forgetting how hard these things are for a newbie. > I don't even know which one comes first of pmds and puds. hey i kept mixing that up myself, until i had to fix a bug in it once. > Per-cpu page tables sounds about on the same scale of as, say, > rewriting the VM or some other major subsystem. Epic! just to make sure: this is an optional "would be really cool" thing. It does not impact the mergability of kmemcheck in any way. [ It impacts the _usability_ of it :-) ] > I'm glad to hear from you, though. > > Pekka suggested that per-cpu page tables might help NUMA systems too. > Does that sound right to you? Would anybody else benefit from having > per-CPU page tables? If not, I have a hard time believing it will ever > get merged. i tried to implement it once for 4G:4G split kernel. My early results indicated that especially on PAE there was a significant SMP speedup from it. I have not gone down that path so i dont know it for sure, but perhaps it was because the CPU sets the accessed bit in kernel PTEs and dirties those cachelines which creates overhead on SMP. That happened even if there was no kernel pagetable remapping activities. (That was many years ago and newer CPUs might be better at it though.) > (Oh. mmio-trace might. But that's also a hacking tool and doesn't > really count.) even if mmiotrace were just a hacking too (but mmiotrace is a system diagnostics tool as well), that could still be a strong reason to do it. Ingo