Re: [PATCH] trace: add ability to set a target task for events (v2)

[Frederic Weisbecker <fweisbec@gmail.com>]

On Wed, Jul 11, 2012 at 04:55:08PM +0200, Peter Zijlstra wrote:
> On Wed, 2012-07-11 at 16:48 +0200, Frederic Weisbecker wrote:
> > On Wed, Jul 11, 2012 at 04:38:19PM +0200, Peter Zijlstra wrote:
> > > On Wed, 2012-07-11 at 16:36 +0200, Frederic Weisbecker wrote:
> > > > 
> > > > In this case he can just record sched wakeup as well. With sched_switch
> > > > + sched_wakeup, he'll unlikely lose events.
> > > > 
> > > > With sched_stat_sleep he will lose events, unless we fix this period
> > > > demux thing. 
> > > 
> > > But without this patch, the sched_wakeup will belong to another task, so
> > > if you trace task A, and B wakes you, you'll never see the wakeup.
> > 
> > Ah so the goal is to minimize the amount of events by only tracing task A?
> 
> Right, or just not having sufficient privs to trace the world. And a
> wakeup of A is very much also part of A, not only the task doing the
> wakeup.
> 
> Hence the proposed mechanism.

Yeah that's fair.

> 
> > Ok then. Still we need to fix these events that use __perf_count() because
> > wide tracing of sched_switch/wake_up still generate less events than
> > sched stat sleep.
> > 
> > I believe:
> > 
> > 	perf record -e sched:sched_stat_sleep sleep 1
> > 
> > produces 1 billion events because we sleep 1 billion nanosecs. Or
> > something like that.
> 
> Right.. back when I did that the plan was to make PERF_SAMPLE_PERIOD fix
> that, of course that never seemed to have happened.
> 
> With PERF_SAMPLE_PERIOD you can simply write the 1b into the period of 1
> event and be done with it.

I believe the perf tools handle pretty well variable periods of an event
on top of PERF_SAMPLE_PERIOD. We just need to tweak the maths in
perf_swevent_overflow() I think...