From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S964895AbWFNFst (ORCPT ); Wed, 14 Jun 2006 01:48:49 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S964892AbWFNFst (ORCPT ); Wed, 14 Jun 2006 01:48:49 -0400 Received: from cantor2.suse.de ([195.135.220.15]:20164 "EHLO mx2.suse.de") by vger.kernel.org with ESMTP id S964897AbWFNFst (ORCPT ); Wed, 14 Jun 2006 01:48:49 -0400 From: Andi Kleen To: Christoph Lameter Subject: Re: [RFC] Use ZVCs for numa statistics. Date: Wed, 14 Jun 2006 07:48:32 +0200 User-Agent: KMail/1.9.3 Cc: linux-kernel@vger.kernel.org, pj@sgi.com, akpm@osdl.org 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: <200606140748.32032.ak@suse.de> Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Wednesday 14 June 2006 03:13, Christoph Lameter wrote: > The numa statistics are really event counters. But they are per > node and therefore we can use the per zone feature to easily > implement what we need. > > Remove the special statistics for numa and replace them with > zoned vm counters. Looks reasonable. > > This has the side effect that global sums of these events now show up > in /proc/vmstat. > > I am not sure if this is a good approach. It certainly removes a > lot of code and may place the counters more securely into the > pcp cacheline but the implementation may be less efficient than the > original code (not tested yet) since the ZVC require the disabling > of interrupts. I still hope that can be avoided by using local_t properly. -Andi