Paolo Bonzini <pbonzini@xxxxxxxxxx> writes:
> On 18/03/21 15:09, Vitaly Kuznetsov wrote:
>> +static inline void check_tsc_msr_tsc_page(struct ms_hyperv_tsc_page *tsc_page)
>> +{
>> + u64 r1, r2, t1, t2;
>> + s64 delta_ns;
>> +
>> + /* Compare TSC page clocksource with HV_X64_MSR_TIME_REF_COUNT */
>> + t1 = mul_u64_u64_shr64(rdtsc(), tsc_page->tsc_scale) + tsc_page->tsc_offset;
>> + r1 = rdmsr(HV_X64_MSR_TIME_REF_COUNT);
>> + nop_loop();
>> + t2 = mul_u64_u64_shr64(rdtsc(), tsc_page->tsc_scale) + tsc_page->tsc_offset;
>> + r2 = rdmsr(HV_X64_MSR_TIME_REF_COUNT);
>> +
>> + delta_ns = ((r2 - r1) - (t2 - t1)) * 100;
>> + if (delta_ns < 0)
>> + delta_ns = -delta_ns;
>> +
>> + /* 1% tolerance */
>> + GUEST_ASSERT(delta_ns * 100 < (t2 - t1) * 100);
>> +}
>> +
>
> I think you should also be able to check r1 and r2 individually, not
> just r1 and r2. Is that correct?
Right, we could've checked r1 == t1 and r2 == t2 actually (with some
tiny margin of course). Let me try that.
--
Vitaly
