On Tue, 2022-08-09 at 14:40 +0000, Sean Christopherson wrote:
> I wanted to separate the "init" from the "stop+start", e.g. if there were a more
> appropriate place for invoking kvm_xen_init_timer() I would have suggested moving
> the call outside of KVM_XEN_VCPU_ATTR_TYPE_TIMER entirely.

Yeah, there's nowhere sensible that would apply to only Xen guests. I
looked at kvm_xen_init_vcpu() but that's unconditional.

I do note that we're now calling kvm_xen_init_timer() even when an
*unset* timer is being restored after live update/live migration. But I
think that's OK.