Re: [RFC PATCH v2 3/3] sched: introduce synchronized idle injection

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Nov 10, 2015 at 08:28:59AM -0800, Jacob Pan wrote:
> On Tue, 10 Nov 2015 15:58:23 +0100
> Peter Zijlstra <peterz@infradead.org> wrote:
> 
> > On Tue, Nov 10, 2015 at 06:01:16AM -0800, Jacob Pan wrote:
> > > On Tue, 10 Nov 2015 14:23:24 +0100
> > > Peter Zijlstra <peterz@infradead.org> wrote:
> > 
> > > > It looks like what you want is:
> > > > 
> > > > 	hrtimer_forward(hrt, period);
> > > > 
> > > > unconditionally.
> > 
> > > In the ideal world yes. But my thinking was that timers may not be
> > > so accurate to deliver interrupts, over the time the timeout error
> > > may accumulate so that eventually timers will be out of sync.
> > 
> > Timers have a global time base. Even if individual deliveries have an
> > error, there is no accumulated error.
> > 
> great! I can get rid of the ktime_roundup(). It seems to work with
> 	now = hrtimer_cb_get_time(hrt);
> 	if (status)
> 		hrtimer_forward(hrt, now, ms_to_ktime(inject_interval));
> 	else
> 		hrtimer_forward(hrt, now, ms_to_ktime(duration));

We have hrtimer_forward_now() for this ;-)

> The downside is that we need to restart the timers every time if
> user were to change injection parameters, i.e. duration and percent.
> Or do locking which might be too expensive. In the previous approach, it
> will naturally catch up the parameter change.

Why? the timer will fire and observe the new value for reprogramming the
next period. All you need to do is to ensure whole values are
written/read -- ie. avoid load/store tearing.