Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933426AbdCJVFT (ORCPT ); Fri, 10 Mar 2017 16:05:19 -0500 Received: from mail-ua0-f172.google.com ([209.85.217.172]:32880 "EHLO mail-ua0-f172.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932705AbdCJVFO (ORCPT ); Fri, 10 Mar 2017 16:05:14 -0500 MIME-Version: 1.0 In-Reply-To: <7ea39103-c193-4d6d-572f-a1bdb27c3627@list.ru> References: <20170308003254.27833-1-ricardo.neri-calderon@linux.intel.com> <79ba0fff-4c01-2bfa-06cb-5cfc98dd710c@list.ru> <997ba581-ecfa-b773-a48e-85b92a439836@list.ru> <1489021909.131264.30.camel@ranerica-desktop> <7ea39103-c193-4d6d-572f-a1bdb27c3627@list.ru> From: Andy Lutomirski Date: Fri, 10 Mar 2017 13:04:51 -0800 Message-ID: Subject: Re: [v6 PATCH 00/21] x86: Enable User-Mode Instruction Prevention To: Stas Sergeev Cc: Andy Lutomirski , Ricardo Neri , Ingo Molnar , Thomas Gleixner , "H. Peter Anvin" , Borislav Petkov , Peter Zijlstra , Andrew Morton , Brian Gerst , Chris Metcalf , Dave Hansen , Paolo Bonzini , Masami Hiramatsu , Huang Rui , Jiri Slaby , Jonathan Corbet , "Michael S. Tsirkin" , Paul Gortmaker , Vlastimil Babka , Chen Yucong , Alexandre Julliard , Fenghua Yu , "Ravi V. Shankar" , Shuah Khan , "linux-kernel@vger.kernel.org" , X86 ML , linux-msdos@vger.kernel.org, wine-devel@winehq.org Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by mail.home.local id v2AL5Rs7021201 Content-Length: 2985 Lines: 74 On Fri, Mar 10, 2017 at 2:30 AM, Stas Sergeev wrote: > 10.03.2017 05:41, Andy Lutomirski пишет: > >> On Wed, Mar 8, 2017 at 5:11 PM, Ricardo Neri >> wrote: >>> >>> On Wed, 2017-03-08 at 19:53 +0300, Stas Sergeev wrote: >>>> >>>> 08.03.2017 19:46, Andy Lutomirski пишет: >>>>>> >>>>>> No no, since I meant prot mode, this is not what I need. >>>>>> I would never need to disable UMIP as to allow the >>>>>> prot mode apps to do SLDT. Instead it would be good >>>>>> to have an ability to provide a replacement for the dummy >>>>>> emulation that is currently being proposed for kernel. >>>>>> All is needed for this, is just to deliver a SIGSEGV. >>>>> >>>>> That's what I meant. Turning off FIXUP_UMIP would leave UMIP on but >>>>> turn off the fixup, so you'd get a SIGSEGV indicating #GP (or a vm86 >>>>> GP exit). >>>> >>>> But then I am confused with the word "compat" in >>>> your "COMPAT_MASK0_X86_UMIP_FIXUP" and >>>> "sys_adjust_compat_mask(int op, int word, u32 mask);" >>>> >>>> Leaving UMIP on and only disabling a fixup doesn't >>>> sound like a compat option to me. I would expect >>>> compat to disable it completely. >>> >>> I guess that the _UMIP_FIXUP part makes it clear that emulation, not >>> UMIP is disabled, allowing the SIGSEGV be delivered to the user space >>> program. >>> >>> Would having a COMPAT_MASK0_X86_UMIP_FIXUP to disable emulation and a >>> COMPAT_MASK0_X86_UMIP to disable UMIP make sense? >>> >>> Also, wouldn't having a COMPAT_MASK0_X86_UMIP to disable UMIP defeat its >>> purpose? Applications could simply use this compat mask to bypass UMIP >>> and gain access to the instructions it protects. >>> >> I was obviously extremely unclear. The point of the proposed syscall >> is to let programs opt out of legacy features. > > I guess both "compat" and "legacy" are misleading > here. Maybe these are "x86-specific" or "hypervisor-specific", > but a mere enabling of UMIP doesn't immediately make > the use of SLDT instruction a legacy IMHO. Sure it is. :) Using SLDT from user mode is a legacy ability that just happens to still work on existing CPUs and kernels. Once UMIP goes in, it will officially be obsolete -- it will just be supported for backwards compatibility. New code should opt out and emulate in usermode if needed. (And the vast, vast majority of Linux programs don't use these instructions in the first place.) Similarly, vsyscalls were obsolete the as soon as better alternatives were fully supported and the kernel started making them slow, and the fact that new static glibc programs still used them for a little while didn't make them any less obsolete. > >> I'll ponder this a bit more. > > So if we are to invent something new, it would be nice to > also think up a clear terminology for it. Maybe something > like "X86_FEATURE_xxx_MASK" or alike. But they're misfeatures, not features. --Andy -- Andy Lutomirski AMA Capital Management, LLC