Re: [PATCH] KVM: VMX: call invvpid only when EPT is disabled

[Avi Kivity <avi@redhat.com>]

On 03/15/2012 03:44 AM, Davidlohr Bueso wrote:
> From: Davidlohr Bueso <dave@gnu.org>
>
> With EPT enabled it is not required to explicitly run invvpid to invalidate tagged TLB entries, as
> KVM does not force vm-exits for cr3 writes and invlpg. Run invvpid only when these instructions
> are emulated and shadow pages are used.
>
> Signed-off-by: Davidlohr Bueso <dave@gnu.org>
> ---
>  arch/x86/kvm/vmx.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 2c22fc7..51c7fb9 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -2861,12 +2861,13 @@ static void exit_lmode(struct kvm_vcpu *vcpu)
>  
>  static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
>  {
> -	vpid_sync_context(to_vmx(vcpu));
>  	if (enable_ept) {
>  		if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
>  			return;
>  		ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
>  	}
> +	else
> +		vpid_sync_context(to_vmx(vcpu));
>  }
>  
>  static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)

This is wrong, cr3 can still be written from host userspace.

To perform this optimization you need to separate the concept of
flushing the guest tlb and the host tlb.  I doubt it's worthwhile
though, won't the ept invalidation flush the entire tlb anyway?

-- 
error compiling committee.c: too many arguments to function