Re: [PATCH 1/7] posix-cpu-timers: don't account cpu timer after stopped thread runtime accounting

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, May 06, 2013 at 11:16:40PM -0400, Olivier Langlois wrote:
> On Tue, 2013-05-07 at 01:47 +0200, Frederic Weisbecker wrote:
> > On Fri, May 03, 2013 at 12:47:42AM -0400, kosaki.motohiro@gmail.com wrote:
> > > From: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> > > 
> > > When tsk->signal->cputimer->running is 1, signal->cputimer and
> > > tsk->sum_sched_runtime increase at the same pace because update_curr()
> > > increases both accounting.
> > > 
> > > However, there is one exception. When thread exiting, __exit_signal() turns
> > > over task's sum_shced_runtime to sig->sum_sched_runtime, but it doesn't stop
> > > signal->cputimer accounting.
> > > 
> > > This inconsistency makes POSIX timer wake up too early. This patch fixes it.
> > > 
> > > Cc: Thomas Gleixner <tglx@linutronix.de>
> > > Cc: Frederic Weisbecker <fweisbec@gmail.com>
> > > Cc: Ingo Molnar <mingo@kernel.org>
> > > Acked-by: Peter Zijlstra <peterz@infradead.org>
> > > Signed-off-by: Olivier Langlois <olivier@trillion01.com>
> > > Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> > > ---
> > >  kernel/sched/stats.h |    7 +++++++
> > >  1 files changed, 7 insertions(+), 0 deletions(-)
> > > 
> > > diff --git a/kernel/sched/stats.h b/kernel/sched/stats.h
> > > index 2ef90a5..5a0cfc4 100644
> > > --- a/kernel/sched/stats.h
> > > +++ b/kernel/sched/stats.h
> > > @@ -225,6 +225,13 @@ static inline void account_group_exec_runtime(struct task_struct *tsk,
> > >       if (!cputimer->running)
> > >               return;
> > >  
> > > +     /*
> > > +      * After turning over se.sum_exec_runtime to sig->sum_sched_runtime
> > > +      * in __exit_signal(), we must not account exec_runtime for consistency.
> > > +      */

Please just precise the nature of that consistency: the fact we want CLOCK_PROCESS_CPUTIME_ID
clock and timer to be consistent.

> > > +     if (unlikely(!tsk->sighand))
> > > +             return;
> > 
> > Ok, if we want the clock and timer to be consistent, do we also want the same check in
> > account_group_user_time() and account_group_system_time()? The task can still account
> > a tick after autoreaping itself between release_task() and the final schedule().
> > 
> If I can reply, I believe that Kosaki refactoring is superior to my
> initial proposal because:
> 
> 1. Condition will only be evaluated when cputimer is on (and not slow
> down scheduler when it is off)
> 2. It handles all the cases when this function is called. There are few
> other places outside update_curr() that this function is called.

Agreed. Thanks!

Acked-by: Frederic Weisbecker <fweisbec@gmail.com>