From mboxrd@z Thu Jan 1 00:00:00 1970 From: "H. Peter Anvin" Subject: Re: [PATCH 4/4] x86, efi: Map runtime services 1:1 Date: Thu, 06 Jun 2013 12:38:10 -0700 Message-ID: <51B0E522.4070005@zytor.com> References: <1370177770-26661-1-git-send-email-bp@alien8.de> <1370177770-26661-5-git-send-email-bp@alien8.de> <20130606131439.GI30420@console-pimps.org> <20130606132908.GE20972@pd.tnic> <20130606155822.GH20972@pd.tnic> <51B0E2D4.3040007@zytor.com> <20130606193612.GM20972@pd.tnic> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20130606193612.GM20972-fF5Pk5pvG8Y@public.gmane.org> Sender: linux-efi-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org To: Borislav Petkov Cc: Matt Fleming , Linux EFI , Matthew Garrett , Jiri Kosina , X86-ML , LKML , Borislav Petkov List-Id: linux-efi@vger.kernel.org On 06/06/2013 12:36 PM, Borislav Petkov wrote: > On Thu, Jun 06, 2013 at 12:28:20PM -0700, H. Peter Anvin wrote: >> Or we could materialize mappings for this specific PGD. However, >> adding a read of %cr3 in __do_page_fault sounds expensive. > > Yes, I think we want to make sure all mappings are there when we do an > EFI runtime call so that we never #PF while it executes. > > Matt mentioned on IRC that the it could be that his EFI runtime is > referencing EFI_RESERVED area which we don't map. However, we need to > confirm/disprove that first, as it is currently only a hunch. If it is true, then maybe a #PF handler which materializes mappings on demand might be the right thing to do. Again, it would be interesting to stop a WinX VM in the process of doing an EFI call and see exactly what it has set up in its page tables. -hpa