On Thu, Oct 27, 2022, Vitaly Kuznetsov wrote:
> Sean Christopherson <seanjc@xxxxxxxxxx> writes:
>
> > On Fri, Oct 21, 2022, Vitaly Kuznetsov wrote:
> >> @@ -1908,6 +1909,12 @@ static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
> >> */
> >> BUILD_BUG_ON(KVM_HV_MAX_SPARSE_VCPU_SET_BITS > 64);
> >>
> >> + if (!hc->fast && is_guest_mode(vcpu)) {
> >
> > Please add a comment explaining why only "slow" hypercalls need to translate the
> > GPA from L2=>L1.
> >
> > With a comment (and assuming this isn't a bug),
>
> This is intended,
>
> For "slow" hypercalls 'hc->ingpa' is the GPA (or an 'nGPA' -- thus the
> patch) in guest memory where hypercall parameters are placed, kvm reads
> them with kvm_read_guest() later. For "fast" hypercalls 'ingpa' is a
> misnomer as it is not an address but the first parameter (in the 'tlb
> flush' case it's 'address space id' which we currently don't
> analyze). We may want to add a union in 'struct kvm_hv_hcall' to make
> this explicit.
Ya, a union would be helpful. I'm pretty sure at some point I knew the "fast"
ingpa isn't actually a GPA, but obviously forgot that detail.
