Re: [PATCH] Track accurate idle time with tick_sched.idle_sleeptime

[Ingo Molnar <mingo@elte.hu>]

* Andi Kleen <andi@firstfloor.org> wrote:

> Venki Pallipadi <venkatesh.pallipadi@intel.com> writes:
> >  
> > +void tick_nohz_stop_idle(int cpu)
> > +{
> > +	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
> > +
> > +	if (ts->idle_active) {
> > +		ktime_t now, delta;
> > +		now = ktime_get();
> 
> That could be PM timer read costing thousands of cycles.
> 
> For going into idle it's probably not a real problem, but it could 
> seriously affect out of idle latencies.

at least the current out-of-idle code already does what amounts to a 
PM-timer read when exiting from C2 or C3 mode. The 
sched_clock_idle_sleep_event() and sched_clock_idle_wakeup_event() 
callbacks utilize this already existing time information.

> You would really need a cheaper per CPU timer that doesn't stop in 
> idle (which rules out sched_clock). It's all unfortunately quite 
> messy.

yep.

	Ingo