Date: Tue, 9 Jan 2018 13:03:55 +0100 (CET)
From: Thomas Gleixner <tglx@linutronix.de>
To: Paolo Bonzini <pbonzini@redhat.com>
cc: LKML <linux-kernel@vger.kernel.org>, kvm@vger.kernel.org,
        jmattson@google.com, aliguori@amazon.com, thomas.lendacky@amd.com,
        dwmw@amazon.co.uk, Borislav Petkov <bp@alien8.de>,
        Peter Zijlstra <peterz@infradead.org>
Subject: Re: [PATCH 0/7] KVM: x86: expose CVE-2017-5715 ("Spectre variant
 2") mitigations to guest
In-Reply-To: <d608fd65-ccbe-d87b-33b1-20b3a0c8cea7@redhat.com>
Message-ID: <alpine.DEB.2.20.1801091257060.1766@nanos>
References: <1515434925-10250-1-git-send-email-pbonzini@redhat.com> <alpine.DEB.2.20.1801091113210.1766@nanos> <d608fd65-ccbe-d87b-33b1-20b3a0c8cea7@redhat.com>
User-Agent: Alpine 2.20 (DEB 67 2015-01-07)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org

On Tue, 9 Jan 2018, Paolo Bonzini wrote:
> On 09/01/2018 11:15, Thomas Gleixner wrote:
> > On Mon, 8 Jan 2018, Paolo Bonzini wrote:
> > 
> >> This series allows guests to use the MSR_IA32_SPEC_CTRL and
> >> MSR_IA32_PRED_CMD model specific registers that were added as mitigations
> >> for CVE-2017-5715.
> >>
> >> These are only the KVM specific parts of the fix.  It does *not* yet
> >> include any protection for reading host memory from the guest, because
> >> that would be done in the same way as the rest of Linux.  So there is no
> >> IBRS *usage* here, no retpolines, no stuffing of the return stack buffer.
> >> (KVM already includes a fix to clear all registers on vmexit, which is
> >> enough to block Google Project Zero's PoC exploit).
> >>
> >> However, I am including the changes to use IBPB (indirect branch
> >> predictor barrier) if available.  That occurs only when there is a VCPU
> >> switch on a physical CPU, thus it has a small impact on performance.
> >>
> >> The patches are a bit hackish because the relevant cpufeatures have
> >> not been included yet, and because I wanted to make the patches easier
> >> to backport to distro kernels if desired, but I would still like to
> >> have them in 4.16.

We really want to coordinate that proper with the ongoing integration of
the IB** for bare metal.

And that stuff really does not need to be hackish at all. We've spent a lot
of effort keeping all of it clean _AND_ available for 4.14 stable
consumption. Everything before 4.9 is a big fricking and incompatible mess
anyway.

> >> Please review.
> > 
> > CC'ing x86@kernel.org on this would have been asked too much, right?
> 
> Sorry, my mistake.  I'll CC you on v2.

All good ...

Please add the crowd which has been involved in the bare metal IBRS stuff
as well.

Thanks,

	tglx