Re: [PATCH] KVM: x86: Convert INIT and SIPI signals into synchronously handled events

[Paolo Bonzini <pbonzini@redhat.com>]

Il 12/03/2013 13:46, Jan Kiszka ha scritto:
>>>> >>> @@ -6178,7 +6177,13 @@ int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
>>>> >>>  int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
>>>> >>>  				    struct kvm_mp_state *mp_state)
>>>> >>>  {
>>>> >>> -	vcpu->arch.mp_state = mp_state->mp_state;
>>>> >>> +	if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
>>>> >>> +		if (!kvm_vcpu_has_lapic(vcpu))
>>>> >>> +			return -EINVAL;
>>>> >>> +		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
>>>> >>> +		set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
>>>> >>> +	} else
>>>> >>> +		vcpu->arch.mp_state = mp_state->mp_state;
>>> >>
>>> >> Should INIT_RECEIVED also be invalid without an in-kernel LAPIC?
>> > 
>> > And since migration was brought up yesterday, do we need an interface to
>> > retrieve and set this?
>> > 
>> > And should KVM_GET/SET_VCPU_EVENTS use the new sipi_vector in the APIC
>> > rather than the old one?
> I hope not. The idea is that the APIC events are processed before the
> migration completes. Translating events on get_mpstate should ensure this.

What about persistent state such as "an INIT has been received and
caused a vmexit, but is being latched until vmxoff"?  Perhaps we could
use the top 8 bits of vcpu->arch.sipi_vector for this, they are always
shifted out when sipi_vector is used, and should be zero in all cases.
It would then be possible to reuse KVM_GET/SET_VCPU_EVENTS.

Migration support for nested VMX is still far far away, so perhaps we do
not care, but we still need a way to inject INIT from userspace.

Paolo