Re: [PATCH v3 0/4] KVM: Expose speculation control feature to guests

[KarimAllah Ahmed]

On 01/30/2018 10:00 AM, David Woodhouse wrote:
> On Tue, 2018-01-30 at 01:10 +0100, KarimAllah Ahmed wrote:
> > Add direct access to speculation control MSRs for KVM guests. This allows the
> > guest to protect itself against Spectre V2 using IBRS+IBPB instead of a
> > retpoline+IBPB based approach.
> >
> > It also exposes the ARCH_CAPABILITIES MSR which is going to be used by future
> > Intel processors to indicate RDCL_NO and IBRS_ALL.
>
> Thanks. I think you've already fixed the SPEC_CTRL patch in the git
> tree so that it adds F(IBRS) to kvm_cpuid_8000_0008_ebx_x86_features,
> right?
Yup, this is already fixed in the tree.

> The SVM part of Ashok's IBPB patch is still exposing the PRED_CMD MSR
> to guests based on boot_cpu_has(IBPB), not based on the *guest*
> capabilities. Looking back at Paolo's patch set from January 9th, it
> was done differently there but I think it had the same behaviour?
>
> The rest of Paolo's patch set I think has been covered, except 6/8:
>  Âlkml.kernel.org/r/20180109120311.27565-7-pbonzini@xxxxxxxxxx
>
> That exposes SPEC_CTRL for SVM too (since AMD now apparently has it).
> If adding that ends up with duplicate MSR handling for get/set, perhaps
> that wants shifting up into kvm_[sg]et_msr_common()? Although I don't
> see offhand where you'd put the ->spec_ctrl field in that case. It
> doesn't want to live in the generic (even to non-x86) struct kvm_vcpu.
> So maybe a little bit of duplication is the best answer.
>
> Other than those details, I think we're mostly getting close. Do we
> want to add STIBP on top? There is some complexity there which meant I
> was happier getting these first bits ready first, before piling that on
> too.
>
> I believe Ashok sent you a change which made us do IBPB on *every*
> vmexit; I don't think we need that. It's currently done in vcpu_load()
> which means we'll definitely have done it between running one vCPU and
> the next, and when vCPUs are pinned we basically never need to do it.
>
> We know that VMM (e.g. qemu) userspace could be vulnerable to attacks
> from guest ring 3, because there is no flush between the vmexit and the
> host kernel "returning" to the userspace thread. Doing a full IBPB on
> *every* vmexit would protect from that, but it's overkill. If that's
> the reason, let's come up with something better.
Amazon Development Center Germany GmbH
Berlin - Dresden - Aachen
main office: Krausenstr. 38, 10117 Berlin
Geschaeftsfuehrer: Dr. Ralf Herbrich, Christian Schlaeger
Ust-ID: DE289237879
Eingetragen am Amtsgericht Charlottenburg HRB 149173 B