Re: [PATCH 09/13] KVM: nVMX: Prioritize SMI over nested IRQ/NMI

[Sean Christopherson]

On Tue, Apr 28, 2020 at 03:04:02PM -0700, Jim Mattson wrote:
> On Wed, Apr 22, 2020 at 7:26 PM Sean Christopherson
> <sean.j.christopherson@xxxxxxxxx> wrote:
> >
> > Check for an unblocked SMI in vmx_check_nested_events() so that pending
> > SMIs are correctly prioritized over IRQs and NMIs when the latter events
> > will trigger VM-Exit.  This also fixes an issue where an SMI that was
> > marked pending while processing a nested VM-Enter wouldn't trigger an
> > immediate exit, i.e. would be incorrectly delayed until L2 happened to
> > take a VM-Exit.
> >
> > Fixes: 64d6067057d96 ("KVM: x86: stubs for SMM support")
> > Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> > ---
> >  arch/x86/kvm/vmx/nested.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> >
> > diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
> > index 1fdaca5fd93d..8c16b190816b 100644
> > --- a/arch/x86/kvm/vmx/nested.c
> > +++ b/arch/x86/kvm/vmx/nested.c
> > @@ -3750,6 +3750,12 @@ static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
> >                 return 0;
> >         }
> >
> > +       if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
> > +               if (block_nested_events)
> > +                       return -EBUSY;
> > +               goto no_vmexit;
> > +       }
> > +
> 
> From the SDM, volume 3:
> 
> â System-management interrupts (SMIs), INIT signals, and higher
> priority events take priority over MTF VM exits.
> 
> I think this block needs to be moved up.

Hrm.  It definitely needs to be moved above the preemption timer, though I
can't find any public documentation about the preemption timer's priority.
Preemption timer is lower priority than MTF, ergo it's not in the same
class as SMI.

Regarding SMI vs. MTF and #DB trap, to actually prioritize SMIs above MTF
and #DBs, we'd need to save/restore MTF and pending #DBs via SMRAM.  I
think it makes sense to take the easy road and keep SMI after the traps,
with a comment to say it's technically wrong but not worth fixing.