Re: [PATCH 2/2] x86/speculation: Support "Enhanced IBRS" on future CPUs

[David Woodhouse <dwmw2@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 1366 bytes --]

On Tue, 2018-02-13 at 10:58 +0100, Paolo Bonzini wrote:

> > If spectre_v2_ibrs_all() is true then KVM should *never* actually pass
> > through or touch the real MSR.
> 
> That would be nice but unfortunately it's not possible. :(
> 
> The VM might actually not have IBRS_ALL, as usual the reason is
> migration compatibility.  In that case, that no-op fiction would be very
> slow because the VM will actually do a lot of SPEC_CTRL writes.

If the VM *thinks* it's bashing on a real SPEC_CTRL register all the
time, and it's actually just trapping to a no-op, then it's actually
going to be a lot *faster* than the VM expects. We can live with that.

> So the right logic is:
> 
> - if the VM has IBRS_ALL, pass through the MSR when it is zero and
> intercept writes when it is one (no writes should happen)
> 
> - if the VM doesn't have IBRS_ALL, do as we are doing now, independent
> of what the host spectre_v2_ibrs_all() setting is.

We end up having to turn IBRS on again on vmexit then, taking care that
no conditional branch can go round it. So that becomes an
*unconditional* wrmsr or lfence in the vmexit path. We really don't
want that.

If we choose to tell a guest that it doesn't have IBRS_ALL, or if the
guest doesn't use IBRS_ALL and does it the old way, it's OK that it's
trapped. It's still faster than they expected.

[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 5213 bytes --]