From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1761239AbXLNIiR (ORCPT ); Fri, 14 Dec 2007 03:38:17 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1753806AbXLNIiA (ORCPT ); Fri, 14 Dec 2007 03:38:00 -0500 Received: from public.id2-vpn.continvity.gns.novell.com ([195.33.99.129]:29699 "EHLO public.id2-vpn.continvity.gns.novell.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752800AbXLNIiA convert rfc822-to-8bit (ORCPT ); Fri, 14 Dec 2007 03:38:00 -0500 Message-Id: <47624F26.76E4.0078.0@novell.com> X-Mailer: Novell GroupWise Internet Agent 7.0.2 HP Date: Fri, 14 Dec 2007 08:38:46 +0000 From: "Jan Beulich" To: "Jeremy Fitzhardinge" Cc: , , "Andi Kleen" , , Subject: Re: [PATCH] x86: fix ref-counting bug in change_page_attr() References: <47610AB5.76E4.0078.0@novell.com> <47622CC5.3060903@goop.org> In-Reply-To: <47622CC5.3060903@goop.org> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 8BIT Content-Disposition: inline Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org >>> Jeremy Fitzhardinge 14.12.07 08:12 >>> >Jan Beulich wrote: >> When either calling change_page_attr() with the default attributes >> pages in the direct mapping have and a page's attributes already were >> set to the default or when changing the attributes from one non-default >> value to another, the reference counting broke, leading to either >> premature restoration of a large page or missing the opportunity to do >> so. >> >> At the same time, make __PHYSICAL_MASK_SHIFT on 64-bits the value it >> architecturally ought to have. >> > >Could you put this in a separate patch? I have a bunch of page*.h and >pgtable*.h refactoring patches which will conflict with this. I doesn't seem logical to do so: The patch needs to introduce the definitions for 32-bits (in order to define pte_pgprot()), and not doing the adjustment for 64-bits here means (a) becoming inconsistent and (b) the pte_pgprot() there would be incorrect. So such a split out patch would need to be a pre- requisite to the one here, which wouldn't help avoiding the collisions with your unification patches. Jan