Re: [PATCH] KVM: apic: reuse smp_wmb() in kvm_make_request()

[Sean Christopherson]

On Thu, Feb 06, 2020 at 11:47:02AM +0100, Vitaly Kuznetsov wrote:
> linmiaohe <linmiaohe@xxxxxxxxxx> writes:
> 
> > From: Miaohe Lin <linmiaohe@xxxxxxxxxx>
> >
> > There is already an smp_mb() barrier in kvm_make_request(). We reuse it
> > here.
> >
> > Signed-off-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
> > ---
> >  arch/x86/kvm/lapic.c | 9 ++++++---
> >  1 file changed, 6 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> > index eafc631d305c..ea871206a370 100644
> > --- a/arch/x86/kvm/lapic.c
> > +++ b/arch/x86/kvm/lapic.c
> > @@ -1080,9 +1080,12 @@ static int __apic_accept_irq(struct kvm_lapic *apic, int delivery_mode,
> >  			result = 1;
> >  			/* assumes that there are only KVM_APIC_INIT/SIPI */
> >  			apic->pending_events = (1UL << KVM_APIC_INIT);
> > -			/* make sure pending_events is visible before sending
> > -			 * the request */
> > -			smp_wmb();
> > +			/*
> > +			 * Make sure pending_events is visible before sending
> > +			 * the request.
> > +			 * There is already an smp_wmb() in kvm_make_request(),
> > +			 * we reuse that barrier here.
> > +			 */
> 
> Let me suggest an alternative wording,
> 
> "kvm_make_request() provides smp_wmb() so pending_events changes are
> guaranteed to be visible"
> 
> But there is nothing wrong with yours, it's just longer than it could be
> :-)

I usually lean in favor of more comments, but in thise case I'd vote to
drop the comment altogether.  There are lots of places that rely on the
smp_wmb() in kvm_make_request() without a comment, e.g. the cases for
APIC_DM_STARTUP and APIC_DM_REMRD in this same switch, kvm_inject_nmi(),
etc...  One might wonder what makes INIT special.

And on the flip side, APIC_DM_STARTUP is a good example of when a
smp_wmb()/smp_rmb() is needed and commented correctly (though calling out
the exactly location of the other half would be helpful).

> Reviewed-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
> 
> >  			kvm_make_request(KVM_REQ_EVENT, vcpu);
> >  			kvm_vcpu_kick(vcpu);
> >  		}
> 
> -- 
> Vitaly
>