Re: tip.today - scheduler bam boom crash (cpu hotplug)

[Wanpeng Li]

2017-02-27 20:43 GMT+08:00 Peter Zijlstra <peterz@xxxxxxxxxxxxx>:
> On Mon, Feb 27, 2017 at 08:30:11PM +0800, Wanpeng Li wrote:
>
>> > diff --git a/arch/x86/kernel/kvmclock.c b/arch/x86/kernel/kvmclock.c
>> > index 2a5cafd..542710b 100644
>> > --- a/arch/x86/kernel/kvmclock.c
>> > +++ b/arch/x86/kernel/kvmclock.c
>> > @@ -107,12 +107,12 @@ static inline void kvm_sched_clock_init(bool stable)
>> >  {
>> >         if (!stable) {
>> >                 pv_time_ops.sched_clock = kvm_clock_read;
>> > +               clear_sched_clock_stable();
>> >                 return;
>> >         }
>> >
>> >         kvm_sched_clock_offset = kvm_clock_read();
>> >         pv_time_ops.sched_clock = kvm_sched_clock_read;
>> > -       set_sched_clock_stable();
>>
>> This results in sched clock always unstable for kvm guest since there
>> is no invariant tsc cpuid bit exposed for kvm guest currently.
>
> What the heck is KVM_FEATURE_CLOCKSOURCE_STABLE_BIT /
> PVCLOCK_TSC_STABLE_BIT about then?
>
>> How about something like below(untested):
>
> I have no clue what that tries to do.

The patch tries to mark sched clock stable if
KVM_FEATURE_CLOCKSOURCE_STABLE_BIT / PVCLOCK_TSC_STABLE_BIT is set in
kvm guest even if there is no invariant tsc cpuid bit.

Regards,
Wanpeng Li