Tianyu Lan <ltykernel@xxxxxxxxx> writes: >>> @@ -398,7 +399,8 @@ static u64 notrace read_hv_clock_msr_cs(struct clocksource *arg) >>> >>> static u64 read_hv_sched_clock_msr(void) >>> { >>> - return read_hv_clock_msr() - hv_sched_clock_offset; >>> + return (read_hv_clock_msr() - hv_sched_clock_offset) >>> + * (NSEC_PER_SEC / HV_CLOCK_HZ); >>> } >> >> kvmclock seems to have the same (pre-patch) code ... > > > kvm sched clock gets time from pvclock_clocksource_read() and > the time unit is nanosecond. So there is such issue in KVM code. > Ah, true, kvmclock is always 1Ghz so it's reading 'naturally' converts to ns. -- Vitaly