From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S964835AbXBPNfw (ORCPT ); Fri, 16 Feb 2007 08:35:52 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S964840AbXBPNfw (ORCPT ); Fri, 16 Feb 2007 08:35:52 -0500 Received: from cantor2.suse.de ([195.135.220.15]:47844 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S964835AbXBPNfv (ORCPT ); Fri, 16 Feb 2007 08:35:51 -0500 Date: Fri, 16 Feb 2007 14:35:46 +0100 From: Nick Piggin To: Lee Schermerhorn Cc: Linux Kernel Mailing List , Linux Memory Management List Subject: Re: [patch] mm: NUMA replicated pagecache Message-ID: <20070216133545.GB3036@wotan.suse.de> References: <20070213060924.GB20644@wotan.suse.de> <1171485124.5099.43.camel@localhost> <20070215003810.GE29797@wotan.suse.de> <1171582169.5114.86.camel@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1171582169.5114.86.camel@localhost> User-Agent: Mutt/1.5.9i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Feb 15, 2007 at 06:29:29PM -0500, Lee Schermerhorn wrote: > > I've attached another patch that closes one race and fixes a context > problem [irq/preemption state] in __unreplicate_page_range(). This > makes the locking even uglier :-(. > > I get further with this patch. Boot all the way up and can run fine > with page replication. However, I still get a NULL pcd in > find_get_page_readonly() when attempting a highly parallel kernel build > [16cpu/4node numa platform]. I'm still trying to track that down. OK, before you get further with your testing, I have done a rework. Sorry you had to wade through that last lot of uncommented spaghetti. This upcoming version should actually be a reasonable base to do testing and development on. > Question about locking: looks like the pcache_descriptor members are > protected by the tree_lock of the mapping, right? Yes. I figured that this would be cleanest and simplest for now. This locking model is retained in the rework. Nick