RE: [PATCH] sched: Folding nohz load accounting more accurate

[Peter Zijlstra <peterz@infradead.org>]

On Tue, 2012-06-19 at 08:50 -0700, Doug Smythies wrote:
> My automated 63 hour test has been terminated, the code changed
> and the test re-started. 

There's another boo-boo in there:


> @@ -2369,11 +2493,35 @@ void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
>   */
>  void calc_global_load(unsigned long ticks)
>  {
> -       long active;
> +       long active, delta;
>  
>         if (time_before(jiffies, calc_load_update + 10))
>                 return;
>  
> +       /*
> +        * Fold the 'old' idle-delta to include all NO_HZ cpus.
> +        *
> +        *      cpu0    cpu1    cpu2    ..
> +        *
> +        * >--- [sample A]
> +        *
> +        *                      -> NOHZ
> +        *              -> NOHZ
> +        *      ->NOHZ
> +        *
> +        * >--- [sample B]
> +        *
> +        * >--- [sample C]
> +        *
> +        *      NOHZ-> (here)
> +        *
> +        * Since all CPUs went into NOHZ state, all 'missed' samples (B, C)
> +        * should include the folded idle-delta.
> +        */
> +       delta += calc_load_fold_idle();

This gives a gcc warning about use without init, which is true, that should be:

	delta = calc_load_fold_idle();

> +       if (delta)
> +               atomic_long_add(delta, &calc_load_tasks);