On Thu, 10 Dec 2020 14:54:08 +1100
David Gibson <david@gibson.dropbear.id.au> wrote:

> On Wed, Dec 09, 2020 at 06:00:52PM +0100, Greg Kurz wrote:
> > kvm_handle_nmi() directly calls spapr_mce_req_event() which is machine
> > level code. Apart from being ugly, this forces spapr_mce_req_event()
> > to rely on qdev_get_machine() to get a pointer to the machine state.
> > This is a bit unfortunate since POWER CPUs have a backlink to the
> > virtual hypervisor, which happens to be the machine itself with
> > sPAPR.
> > 
> > Turn spapr_mce_req_event() into a PPC virtual hypervisor operation,
> > and adapt kvm_handle_nmi() to call it as such.
> > 
> > Signed-off-by: Greg Kurz <groug@kaod.org>
> 
> I have somewhat mixed thoughts on this.  Putting it in vhyp makes a
> certain sense.  But on the other hand, the MCE event from KVM is an
> explicitly PAPR specific interface, so it can't really go to any other
> implementation.
> 

True. Same thing goest for the hypercalls actually. So I guess it's
better to keep this dependency explicit, as long as we don't have
to support non-PAPR KVM guests. Please ignore this patch.