Re: [PATCH] posix-cpu-timers: clock_gettime(CLOCK_*_CPUTIME_ID) goes backward

[Ingo Molnar <mingo@elte.hu>]

* Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:

> On Fri, 2008-12-26 at 09:43 +0100, Ingo Molnar wrote:
> > * Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com> wrote:
> > 
> > > @@ -321,7 +287,7 @@ static int cpu_clock_sample_group(const clockid_t which_clock,
> > >  		cpu->cpu = cputime.utime;
> > >  		break;
> > >  	case CPUCLOCK_SCHED:
> > > -		cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
> > > +		cpu->sched = cputime.sum_exec_runtime;
> > >  		break;
> > >  	}
> > 
> > hm, doesnt this regress precision?
> 
> No, he folds it into:
> 
> > +void thread_group_cputime(struct task_struct *p, struct task_cputime *times)
> > +{
> > +       unsigned long flags;
> > +       struct rq *rq;
> > +       u64 delta_exec = 0;
> > +       struct task_cputime *tot;
> > +       struct signal_struct *sig;
> > +       int i;
> > +
> > +       sig = p->signal;
> > +       if (unlikely(!sig) || !sig->cputime.totals) {
> > +               times->utime = p->utime;
> > +               times->stime = p->stime;
> > +               times->sum_exec_runtime = task_total_exec(p);
> > +               return;
> > +       }
> > +
> > +       times->stime = times->utime = cputime_zero;
> > +       times->sum_exec_runtime = 0;
> > +
> > +       rq = task_rq_lock(p, &flags);
> > +
> > +       if (task_current(rq, p)) {
> > +               update_rq_clock(rq);
> > +               delta_exec = rq->clock - p->se.exec_start;
> > +               if ((s64)delta_exec < 0)
> > +                       delta_exec = 0;
> > +       }
> > +
> > +       for_each_possible_cpu(i) {
> > +               tot = per_cpu_ptr(p->signal->cputime.totals, i);
> > +               times->utime = cputime_add(times->utime, tot->utime);
> > +               times->stime = cputime_add(times->stime, tot->stime);
> > +               times->sum_exec_runtime += tot->sum_exec_runtime;
> > +       }
> > +       times->sum_exec_runtime += delta_exec;
> > +
> > +       task_rq_unlock(rq, &flags);
> > +}
> 
> Which reminds me, why do we still have this crap in the kernel? I 
> thought we pretty much showed the per-cpu itimer thing was utter crap? 
> -- can we pretty please either revert that or apply 
> http://lkml.org/lkml/2008/11/24/183 ?

yes, i agree with that patch - could you please send it with a signoff?

> Also, I really don't like the above, we now do the per-cpu loop with the 
> RQ lock held...

yes.

	Ingo