Re: [Qemu-devel] [PATCH] spapr: Don't set the TM ibm, pa-features bit in PR KVM mode

[David Gibson]

On Tue, Apr 05, 2016 at 12:12:01PM +1000, Paul Mackerras wrote:
> On Mon, Apr 04, 2016 at 09:09:28PM +1000, Anton Blanchard wrote:
> > We don't support transactional memory in PR KVM, so don't tell
> > the OS that we do.
> 
> This assumes PR KVM won't ever support TM, which is hopefully not
> true.  If PR KVM does get TM support in future, then QEMU will have no
> clear way to know whether it needs to clear the pa-features bit or
> not.  I think we need to define some way for the KVM implementation to
> tell qemu which of these kinds of CPU features it supports.

Yeah, I think we need some sort of capability flag for this.  We also
need to isolate this KVM capability testing better into the KVM code,
so we won't break things on TCG.

Speaking of which... I don't imagine we implement TM instructions in
TCG either, so we should probably make sure TM isn't advertised there
either.

> However, we could defer implementing that mechanism until PR KVM does
> get support for TM, I guess.  In that case this patch could go in now,
> though it seems slightly icky to be using the pvinfo stuff to
> distinguish PR from HV.

It is icky, but it's the best idiom we have for distinguishing PR
KVM.  We try to only do it as a fallback when we can't determine
availability of the specific feature based on a capability, though.

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature