Re: [PATCH v2 1/2] KVM: nVMX: use vmx_segment_cache_clear

[Maxim Levitsky <mlevitsk@redhat.com>]

On Tue, 2024-07-16 at 14:07 -0700, Sean Christopherson wrote:
> On Mon, Jul 15, 2024, Maxim Levitsky wrote:
> > In prepare_vmcs02_rare, call vmx_segment_cache_clear, instead
> > of setting the segment_cache.bitmask directly.
> > 
> > No functional change intended.
> > 
> > Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
> > ---
> >  arch/x86/kvm/vmx/nested.c | 5 +++--
> >  arch/x86/kvm/vmx/vmx.c    | 4 ----
> >  arch/x86/kvm/vmx/vmx.h    | 5 +++++
> >  3 files changed, 8 insertions(+), 6 deletions(-)
> > 
> > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> > index 643935a0f70ab..d3ca1a772ae67 100644
> > --- a/arch/x86/kvm/vmx/nested.c
> > +++ b/arch/x86/kvm/vmx/nested.c
> > @@ -2469,6 +2469,9 @@ static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
> >  
> >  	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
> >  			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
> > +
> > +		vmx_segment_cache_clear(vmx);
> > +
> >  		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
> >  		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
> >  		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
> > @@ -2505,8 +2508,6 @@ static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
> >  		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
> >  		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
> >  		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
> > -
> > -		vmx->segment_cache.bitmask = 0;
> 
> This actually exacerbates the bug that you're trying fix in patch 2.  Clearing
> segment_cache.bitmask _after_ writing the relevant state limits the stale data
> to only the accessor that's running in IRQ context (kvm_arch_vcpu_put()).
> 
> Clearing segment_cache.bitmask _before_ writing the relevant statement means
> that kvm_arch_vcpu_put() _and_ all future readers will be exposed to the stale
> data, because the stale data cached by kvm_arch_vcpu_put() won't mark it invalid.
> 
I noticed that after I sent the patch series, this makes sense.

Best regards,
	Maxim Levitsky