From: Nick Piggin Subject: Re: Could we do immediate pte zaps in vunmap? Date: Sat, 27 Nov 2010 21:36:56 +1100 Message-ID: <20101127103656.GA6884@amd> References: <4CEF6B8B.8080206@goop.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <4CEF6B8B.8080206@goop.org> Sender: linux-kernel-owner@vger.kernel.org To: Jeremy Fitzhardinge Cc: Nick Piggin , "Xen-devel@lists.xensource.com" , Linux Kernel Mailing List List-Id: xen-devel@lists.xenproject.org On Fri, Nov 26, 2010 at 12:10:51AM -0800, Jeremy Fitzhardinge wrote: > What if vm_unmap_ram() and co. immediately zeroed out the ptes, but > lazily deferred the tlb flushes? It seems to me there's no benefit in > batching up the pte clearing since that can't be amortized like the tlb > flush. Yes that can be done. > I think that would solve the problem we have with the interactions > between lazy unmap and Xen. The issue is having stray pte entries > around (because Xen keeps track of those as part of its page-type > mechanism), but stale tlb entries are no problem. It would be great if that solves it.