From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758087Ab0EDME0 (ORCPT ); Tue, 4 May 2010 08:04:26 -0400 Received: from mx1.redhat.com ([209.132.183.28]:59619 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751732Ab0EDMEZ (ORCPT ); Tue, 4 May 2010 08:04:25 -0400 Message-ID: <4BE00D46.6050003@redhat.com> Date: Tue, 04 May 2010 15:04:22 +0300 From: Avi Kivity User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.1.9) Gecko/20100330 Fedora/3.0.4-1.fc12 Thunderbird/3.0.4 MIME-Version: 1.0 To: "Roedel, Joerg" CC: Marcelo Tosatti , "kvm@vger.kernel.org" , "linux-kernel@vger.kernel.org" Subject: Re: [PATCH 16/22] KVM: MMU: Track page fault data in struct vcpu References: <1272364712-17425-1-git-send-email-joerg.roedel@amd.com> <1272364712-17425-17-git-send-email-joerg.roedel@amd.com> <4BD6DF7C.1090203@redhat.com> <20100503163221.GB28950@amd.com> <4BDFD295.7000702@redhat.com> <20100504091157.GC28950@amd.com> <4BDFE6C2.2040601@redhat.com> <20100504093709.GE28950@amd.com> <4BDFECD1.8040109@redhat.com> <4BDFEDFA.3090209@redhat.com> <20100504120039.GH28950@amd.com> In-Reply-To: <20100504120039.GH28950@amd.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 05/04/2010 03:00 PM, Roedel, Joerg wrote: > On Tue, May 04, 2010 at 05:50:50AM -0400, Avi Kivity wrote: > >> On 05/04/2010 12:45 PM, Avi Kivity wrote: >> >>> >>> Just a silly bug. kvm_pdptr_read() can cause a guest memory read on >>> svm, in this case with the mmu lock taken. I'll post something to fix >>> it. >>> >> I guess this was not reported because most svm machines have npt, and >> this requires npt=0 to trigger. Nonpae paging disables npt, so you were >> hit. Interestingly, nsvm makes it more likely to appear, since npt on >> i386+pae will need the pdptrs. >> > Hmm, actually it happened on 32 bit with npt enabled. I think this > can trigger when mmu_alloc_roots is called for an pae guest because it > accidentially tries read the root_gfn from the guest before it figures > out that it runs with tdp and omits the gfn read from the guest. > Yes. I had a patchset which moved the 'direct' calculation before, and skipped root_gfn if it was direct, but it was broken. If you like I can resurrect it, but it may interfere with your work. -- error compiling committee.c: too many arguments to function