From: Wanpeng Li <wanpengli@tencent.com>

The timer is disarmed when switching between TSC deadline and other modes, 
however, the pending timer is still in-flight, so let's accurately set 
everything to a disarmed state.

Fixes: 4427593258 (KVM: x86: thoroughly disarm LAPIC timer around TSC deadline switch)
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
---
 arch/x86/kvm/lapic.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 66b0eb0bda94..0274d17d91c2 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -1562,6 +1562,7 @@ static void apic_update_lvtt(struct kvm_lapic *apic)
 			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
 			apic->lapic_timer.period = 0;
 			apic->lapic_timer.tscdeadline = 0;
+			atomic_set(&apic->lapic_timer.pending, 0);
 		}
 		apic->lapic_timer.timer_mode = timer_mode;
 		limit_periodic_timer_frequency(apic);
-- 
2.25.1

On Tue, May 10, 2022, Wanpeng Li wrote:
> From: Wanpeng Li <wanpengli@tencent.com>
> 
> The timer is disarmed when switching between TSC deadline and other modes, 
> however, the pending timer is still in-flight, so let's accurately set 
> everything to a disarmed state.
> 
> Fixes: 4427593258 (KVM: x86: thoroughly disarm LAPIC timer around TSC deadline switch)
> Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> ---
>  arch/x86/kvm/lapic.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index 66b0eb0bda94..0274d17d91c2 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -1562,6 +1562,7 @@ static void apic_update_lvtt(struct kvm_lapic *apic)
>  			kvm_lapic_set_reg(apic, APIC_TMICT, 0);
>  			apic->lapic_timer.period = 0;
>  			apic->lapic_timer.tscdeadline = 0;
> +			atomic_set(&apic->lapic_timer.pending, 0);

What about doing this in cancel_apic_timer()?  That seems to be a more natural
place to clear pending.  It's somewhat redundant since the other two callers of
cancel_apic_timer() start the timer immediately after, i.e. clear pending anyways,
but IMO it's odd to leave pending set when canceling the timer.

>  		}
>  		apic->lapic_timer.timer_mode = timer_mode;
>  		limit_periodic_timer_frequency(apic);
> -- 
> 2.25.1
>

On Tue, 10 May 2022 at 21:43, Sean Christopherson <seanjc@google.com> wrote:
>
> On Tue, May 10, 2022, Wanpeng Li wrote:
> > From: Wanpeng Li <wanpengli@tencent.com>
> >
> > The timer is disarmed when switching between TSC deadline and other modes,
> > however, the pending timer is still in-flight, so let's accurately set
> > everything to a disarmed state.
> >
> > Fixes: 4427593258 (KVM: x86: thoroughly disarm LAPIC timer around TSC deadline switch)
> > Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> > ---
> >  arch/x86/kvm/lapic.c | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> > index 66b0eb0bda94..0274d17d91c2 100644
> > --- a/arch/x86/kvm/lapic.c
> > +++ b/arch/x86/kvm/lapic.c
> > @@ -1562,6 +1562,7 @@ static void apic_update_lvtt(struct kvm_lapic *apic)
> >                       kvm_lapic_set_reg(apic, APIC_TMICT, 0);
> >                       apic->lapic_timer.period = 0;
> >                       apic->lapic_timer.tscdeadline = 0;
> > +                     atomic_set(&apic->lapic_timer.pending, 0);
>
> What about doing this in cancel_apic_timer()?  That seems to be a more natural
> place to clear pending.  It's somewhat redundant since the other two callers of
> cancel_apic_timer() start the timer immediately after, i.e. clear pending anyways,
> but IMO it's odd to leave pending set when canceling the timer.

Do it in v2.

    Wanpeng