From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751751Ab3JWHoy (ORCPT ); Wed, 23 Oct 2013 03:44:54 -0400 Received: from mail-ea0-f170.google.com ([209.85.215.170]:57885 "EHLO mail-ea0-f170.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751008Ab3JWHow (ORCPT ); Wed, 23 Oct 2013 03:44:52 -0400 Date: Wed, 23 Oct 2013 09:44:47 +0200 From: Ingo Molnar To: Peter Zijlstra Cc: Linus Torvalds , Don Zickus , Andi Kleen , dave.hansen@linux.intel.com, Stephane Eranian , jmario@redhat.com, Linux Kernel Mailing List , Arnaldo Carvalho de Melo Subject: Re: [PATCH] perf, x86: Optimize intel_pmu_pebs_fixup_ip() Message-ID: <20131023074447.GB9916@gmail.com> References: <20131016210319.GI10651@twins.programming.kicks-ass.net> <20131016230712.GC26785@twins.programming.kicks-ass.net> <20131017094145.GE3364@laptop.programming.kicks-ass.net> <20131017160034.GO227855@redhat.com> <20131017160439.GP227855@redhat.com> <20131017163039.GR10651@twins.programming.kicks-ass.net> <20131017220156.GB10651@twins.programming.kicks-ass.net> <20131022211237.GH2490@laptop.programming.kicks-ass.net> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20131022211237.GH2490@laptop.programming.kicks-ass.net> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Peter Zijlstra wrote: > On Thu, Oct 17, 2013 at 03:27:48PM -0700, Linus Torvalds wrote: > > On Thu, Oct 17, 2013 at 3:01 PM, Peter Zijlstra wrote: > > > > > > Oh wait,.. now that Steven fixed being able to take faults from NMI > > > context; we could actually try copy_from_user_inatomic(). Being able to > > > directly access userspace would make the whole deal a lot easier again. > > > > Careful! There is one magic piece of state that you need to > > save-and-restore if you do this, namely %cr2. Taking a page fault > > always writes to %cr2, and we must *not* corrupt it in the NMI > > handler. > > It looks like this is already dealt with (a similar thing is done > for i386). > commit 7fbb98c5cb07563d3ee08714073a8e5452a96be2 > Author: Steven Rostedt > Date: Thu Jun 7 10:21:21 2012 -0400 > > x86: Save cr2 in NMI in case NMIs take a page fault > > Avi Kivity reported that page faults in NMIs could cause havic if > the NMI preempted another page fault handler: > > The recent changes to NMI allow exceptions to take place in NMI > handlers, but I think that a #PF (say, due to access to vmalloc space) > is still problematic. Consider the sequence > > #PF (cr2 set by processor) > NMI > ... > #PF (cr2 clobbered) > do_page_fault() > IRET > ... > IRET > do_page_fault() > address = read_cr2() > > The last line reads the overwritten cr2 value. > > Originally I wrote a patch to solve this by saving the cr2 on the stack. > Brian Gerst suggested to save it in the r12 register as both r12 and rbx > are saved by the do_nmi handler as required by the C standard. But rbx > is already used for saving if swapgs needs to be run on exit of the NMI > handler. > > Link: http://lkml.kernel.org/r/4FBB8C40.6080304@redhat.com > Link: http://lkml.kernel.org/r/1337763411.13348.140.camel@gandalf.stny.rr.com > > Reported-by: Avi Kivity > Cc: Linus Torvalds > Cc: H. Peter Anvin > Cc: Thomas Gleixner > Suggested-by: Brian Gerst > Signed-off-by: Steven Rostedt > > diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S > index 7d65133..111f6bb 100644 > --- a/arch/x86/kernel/entry_64.S > +++ b/arch/x86/kernel/entry_64.S > @@ -1758,10 +1758,30 @@ ENTRY(nmi) > */ > call save_paranoid > DEFAULT_FRAME 0 > + > + /* > + * Save off the CR2 register. If we take a page fault in the NMI then > + * it could corrupt the CR2 value. If the NMI preempts a page fault > + * handler before it was able to read the CR2 register, and then the > + * NMI itself takes a page fault, the page fault that was preempted > + * will read the information from the NMI page fault and not the > + * origin fault. Save it off and restore it if it changes. > + * Use the r12 callee-saved register. > + */ > + movq %cr2, %r12 > + /* Did the NMI take a page fault? Restore cr2 if it did */ > + movq %cr2, %rcx > + cmpq %rcx, %r12 > + je 1f > + movq %r12, %cr2 > +1: > + Btw., depending on how expensive cr2 is to read this could be optimized a bit, by matching on the race window RIP range instead of saving cr2 unconditionally. It would take a bit of restructuring of the page fault entry path to get the race window into a single RIP range, but that should be fairly trivial and might even speed up the page fault handler as the CPU can start the cr2 load sooner in the page fault handling sequence. It's all a function of how expensive a cr2 read is. If it's 2 cycles then it doesn't matter much. If it's 20 then it might be worthwile. Thanks, Ingo