Re: [kvm-devel] [ RfC / patch ] kvmclock fixes

[Jeremy F. <je...@go...>]

Gerd Hoffmann wrote:
> +cycle_t pvclock_clocksource_read(struct kvm_vcpu_time_info *src)
> +{
> +	struct pvclock_shadow_time *shadow;
> +	cycle_t ret;
> +	unsigned version;
> +
> +	shadow = &get_cpu_var(shadow_time);
> +	do {
> +		version = pvclock_get_time_values(shadow, src);
> +		barrier();
> +		ret = shadow->system_timestamp + pvclock_get_nsec_offset(shadow);
> +		barrier();
>   

Is barrier() strong enough?  Does kvm guarantee that the per-cpu time 
parameters are only ever updated by that cpu?  I'm pretty sure Xen does, 
so that's OK.

    J