On Tue, Oct 9, 2018 at 5:14 AM, Liran Alon <liran.alon@xxxxxxxxxx> wrote:
> I dislike the fact that you also changed (1u << nr) to BIT(nr) on this patch.
> All such use-cases currently in vmx.c are of the form (1u << X) and there is no use of BIT() macro.
> I am not against it but I prefer for such a change to happen on a separate patch and modify other similar places as-well.
Okay; I'll revert that part of the change in v2.
>> - if (vcpu->arch.exception.nr == DB_VECTOR &&
>> - (vcpu->arch.dr7 & DR7_GD)) {
>> - vcpu->arch.dr7 &= ~DR7_GD;
>> - kvm_update_dr7(vcpu);
>> + if (vcpu->arch.exception.nr == DB_VECTOR) {
>> + kvm_deliver_exception_payload(vcpu);
>
> I would add a comment here that one should note that once we will modify nSVM to use
> check_nested_events() framework, the call here for kvm_deliver_exception_payload()
> should be moved to svm_check_nested_events().
Actually, if nSVM uses the check_nested_events framework, things are a
little more complicated than that. The DR7 update has to be done in
svm_check_nested_events as well. It would be nice to just move the DR7
update into kvm_deliver_exception_payload, but that would break the
old ABI. I'm not sure that this is the appropriate place for a lengthy
comment about a speculative future code change, but I'll go ahead and
write something up.
