Re: [PATCH 0/4] kvmclock: improve accuracy

[Marcelo Tosatti]

On Mon, Feb 08, 2016 at 04:18:27PM +0100, Paolo Bonzini wrote:
> Currently kvmclock is obtaining the multiplier and shift value from
> the TSC kHz.  These however are less accurate than the host kernel's
> clock, which includes corrections made through NTP.
> 
> These patches change kvmclock to tick at the same frequency as the
> host kernel's clocksource (as obtained through the pvclock_gtod
> notifier).  This is precise enough that the Hyper-V clock can be
> implemented on top of it.

Note the current option to sync kvmclock periodically every 300 seconds:

#define KVMCLOCK_SYNC_PERIOD (300 * HZ)

static void kvmclock_sync_fn(struct work_struct *work)
{
        struct delayed_work *dwork = to_delayed_work(work);
        struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
                                           kvmclock_sync_work);
        struct kvm *kvm = container_of(ka, struct kvm, arch);

        schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
        schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
                                        KVMCLOCK_SYNC_PERIOD);
}

How large is the error when the corrections are applied every 5 minutes? 

Does this match the errors you are seeing?