From mboxrd@z Thu Jan 1 00:00:00 1970 Date: Tue, 5 Feb 2008 16:12:22 -0600 From: Robin Holt Subject: Re: [PATCH] mmu notifiers #v5 Message-ID: <20080205221221.GP17211@sgi.com> References: <20080201120955.GX7185@v2.random> <20080203021704.GC7185@v2.random> <20080205052525.GD7441@v2.random> <20080205180802.GE7441@v2.random> <20080205205519.GF7441@v2.random> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: Sender: owner-linux-mm@kvack.org Return-Path: To: Christoph Lameter Cc: Andrea Arcangeli , Robin Holt , Avi Kivity , Izik Eidus , kvm-devel@lists.sourceforge.net, Peter Zijlstra , steiner@sgi.com, linux-kernel@vger.kernel.org, linux-mm@kvack.org, daniel.blueman@quadrics.com List-ID: On Tue, Feb 05, 2008 at 02:06:23PM -0800, Christoph Lameter wrote: > On Tue, 5 Feb 2008, Andrea Arcangeli wrote: > > > On Tue, Feb 05, 2008 at 10:17:41AM -0800, Christoph Lameter wrote: > > > The other approach will not have any remote ptes at that point. Why would > > > there be a coherency issue? > > > > It never happens that two threads writes to two different physical > > pages by working on the same process virtual address. This is an issue > > only for KVM which is probably ok with it but certainly you can't > > consider the dependency on the page-pin less fragile or less complex > > than my PT lock approach. > > You can avoid the page-pin and the pt lock completely by zapping the > mappings at _start and then holding off new references until _end. XPMEM is doing this by putting our equivalent structure of the mm into a recalling state which will cause all future faulters to back off, it then marks any currently active faults in the range as invalid (we have a very small number of possible concurrent faulters for a different reason), proceeds to start remote shoot-downs, waits for those shoot-downs to complete, then returns from the _begin callout with the mm-equiv still in the recalling state. Additional recalls may occur, but no new faults can. The _end callout reduces the number of active recalls until there are none left at which point the faulters are allowed to proceed again. Thanks, Robin -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org