Re: [v2 PATCH 3/4] x86/kvm: Add host side support for virtual suspend time injection

[Paolo Bonzini]

On 18/08/21 11:32, Vitaly Kuznetsov wrote:
> On the host side, I'd vote for keeping MSR_KVM_ASYNC_PF_INT for async PF
> mechanism only for two reasons:
> - We may want to use (currently reserved) upper 56 bits of it for new
> asyncPF related features (e.g. flags) and it would be unnatural to add
> them to 'MSR_KVM_HYPERVISOR_CALLBACK_INT'
> - We should probably leave it to the guest if it wants to share 'suspend
> time' notification interrupt with async PF (and if it actually wants to
> get one/another).

I agree that it's fine either way.  That said, more MSRs are more 
complexity and more opportunity for getting things wrong (in either KVM 
or userspace---for example, migration).  There are still 14 free bits in 
MSR_ASYNC_PF_EN (bits 63-52 and 5-4, so it should not be a problem to 
repurpose MSR_ASYNC_PF_INT.

Paolo

> On the guest side, it is perfectly fine to reuse
> HYPERVISOR_CALLBACK_VECTOR for everything.