From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755292AbZCJVTZ (ORCPT ); Tue, 10 Mar 2009 17:19:25 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1754457AbZCJVTP (ORCPT ); Tue, 10 Mar 2009 17:19:15 -0400 Received: from gw.goop.org ([64.81.55.164]:35387 "EHLO mail.goop.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753840AbZCJVTP (ORCPT ); Tue, 10 Mar 2009 17:19:15 -0400 Message-ID: <49B6D94F.9070902@goop.org> Date: Tue, 10 Mar 2009 14:19:11 -0700 From: Jeremy Fitzhardinge User-Agent: Thunderbird 2.0.0.19 (X11/20090105) MIME-Version: 1.0 To: "H. Peter Anvin" CC: the arch/x86 maintainers , Linux Kernel Mailing List , Xen-devel , Jeremy Fitzhardinge Subject: Re: [PATCH] paravirt/xen: add pvop for page_is_ram References: <1235786365-17744-1-git-send-email-jeremy@goop.org> <1235786365-17744-10-git-send-email-jeremy@goop.org> <49B5BD63.1090909@zytor.com> In-Reply-To: <49B5BD63.1090909@zytor.com> X-Enigmail-Version: 0.95.6 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org H. Peter Anvin wrote: > Jeremy Fitzhardinge wrote: > >> From: Jeremy Fitzhardinge >> >> A guest domain may have external pages mapped into its address space, >> in order to share memory with other domains. These shared pages are >> more akin to io mappings than real RAM, and should not pass the >> page_is_ram test. Add a paravirt op for this so that a hypervisor >> backend can validate whether a page should be considered ram or not. >> >> Signed-off-by: Jeremy Fitzhardinge >> >> > > Why are these pages mapped as RAM in the memory map? That is the right > way to handle that, not by adding yet another bloody hook... > Granted pages can turn up anywhere dynamically, since they're pages borrowed from other domains for the purposes of IO. They're not static regions of non-RAM like the other cases page_is_ram() tests for, They can't be mapped via normal pte operations (because they have additional state associated with them, like the grant handle), so /dev/mem can't just create an aliased mapping by copying the pte. page_is_ram is used to: 1. prevent /dev/mem from mapping non-RAM pages 2. prevent ioremap from mapping any RAM pages 3. testing for RAMness in PAT 3) isn't yet relevant to Xen; ioremap can't map granted pages either, so 2) isn't terribly relevent, so the main motivation for this patch is 1). This allows us to reject usermode attempts to map granted pages, rather than oopsing (as a failed set_pte will raise a page fault). So, more cosmetic than essential, but I don't see a better way to implement this functionality if its to be there at all. J