Re: [PATCH v2 31/32] KVM: nVMX: Don't flush TLB on nested VM transition with EPT enabled

From: Sean Christopherson
Date: Wed Mar 18 2020 - 13:26:17 EST


On Wed, Mar 18, 2020 at 06:11:28PM +0100, Paolo Bonzini wrote:
> On 18/03/20 18:02, Sean Christopherson wrote:
> > So something like this?
> >
> > if (!nested_ept)
> > kvm_mmu_new_cr3(vcpu, cr3, enable_ept ||
> > nested_cpu_has_vpid(vmcs12));
>
> ... which is exactly nested_has_guest_tlb_tag(vcpu). Well, not exactly
> but it's a bug in your code above. :)

I don't think it's a bug, it's intentionally different. When enable_ept=0,
nested_has_guest_tlb_tag() returns true if and only if L1 has enabled VPID
for L2 *and* L2 has been assigned a unique VPID by L0.

For sync purposes, whether or not L2 has been assigned a unique VPID is
irrelevant. L0 needs to invalidate TLB entries to prevent resuing L1's
entries (assuming L1 has been assigned a VPID), but L0 doesn't need to sync
SPTEs because L2 doesn't expect them to be refreshed.

> It completely makes sense to use that as the third argument, and while a
> comment is still needed it will be much smaller.

Ya, agreed.