Li, Aubrey wrote:
> Li, Aubrey wrote:
>
>> My system is in idle when this report was obtained.
>>
>> I'll investigate it and let you know what's going on.
>>
>
> hmm..., the problem is when pt_cpuidle_stat_collect()
> calculate C0 residency, it always counts all cpus up.
>
> pt_cpuidle_stat_collect(double interval)
> {
> ----snip----
> /*
> * Assume that all the time spent in this interval will
> * be the default "0" state. The DTrace walker will reallocate
> * time out of the default bucket as it processes aggregation
> * records for time spent in other states.
> */
> g_cstate_info[0].total_time = (long)(interval * g_ncpus * 1000);
> ----snip----
> }
>
> When we use -c option, C0 should be calculated only on the specific
> CPU. So, here, g_ncpus is not fit for the -c option.
>
> Does this make sense? Do you always test powertop on one logical
> core system?
Ah, of course. Thanks for pointing that out.
I'm testing the x86 side on a 2 core and on a 2x quad core system.
Thanks,
Rafael
