Re: [RFC PATCH 2/5] sched/events: Introduce cfs_rq load tracking trace event

[Dietmar Eggemann <dietmar.eggemann@arm.com>]

On 03/28/2017 06:44 PM, Peter Zijlstra wrote:
> On Tue, Mar 28, 2017 at 10:46:00AM -0400, Steven Rostedt wrote:
>> On Tue, 28 Mar 2017 07:35:38 +0100
>> Dietmar Eggemann <dietmar.eggemann@arm.com> wrote:

[...]

> I too suggested that; but then I looked again at that code and we can
> actually do this. cfs_rq can be constant propagated and the if
> determined at build time.
>
> Its not immediately obvious from the current code; but if we do
> something like the below, it should be clearer.
>
> ---
> Subject: sched/fair: Explicitly generate __update_load_avg() instances
> From: Peter Zijlstra <peterz@infradead.org>
> Date: Tue Mar 28 11:08:20 CEST 2017
>
> The __update_load_avg() function is an __always_inline because its
> used with constant propagation to generate different variants of the
> code without having to duplicate it (which would be prone to bugs).

Ah, so the if(cfs_rq)/else condition should stay in ___update_load_avg() 
and I shouldn't move the trace events into the 3 variants?

I tried to verify that the if is determined at build time but it's kind 
of hard with trace_events.

> Explicitly instantiate the 3 variants.
>
> Note that most of this is called from rather hot paths, so reducing
> branches is good.
>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -2849,7 +2849,7 @@ static u32 __compute_runnable_contrib(u6
>   *            = u_0 + u_1*y + u_2*y^2 + ... [re-labeling u_i --> u_{i+1}]
>   */
>  static __always_inline int
> -__update_load_avg(u64 now, int cpu, struct sched_avg *sa,
> +___update_load_avg(u64 now, int cpu, struct sched_avg *sa,
>  		  unsigned long weight, int running, struct cfs_rq *cfs_rq)
>  {
>  	u64 delta, scaled_delta, periods;
> @@ -2953,6 +2953,26 @@ __update_load_avg(u64 now, int cpu, stru
>  	return decayed;
>  }
>
> +static int
> +__update_load_avg_blocked_se(u64 now, int cpu, struct sched_avg *sa)
> +{
> +	return ___update_load_avg(now, cpu, sa, 0, 0, NULL);
> +}
> +
> +static int
> +__update_load_avg_se(u64 now, int cpu, struct sched_avg *sa,
> +		     unsigned long weight, int running)
> +{
> +	return ___update_load_avg(now, cpu, sa, weight, running, NULL);
> +}
> +
> +static int
> +__update_load_avg(u64 now, int cpu, struct sched_avg *sa,
> +		  unsigned long weight, int running, struct cfs_rq *cfs_rq)
> +{
> +	return ___update_load_avg(now, cpu, sa, weight, running, cfs_rq);
> +}

Why not reduce the parameter list of these 3 incarnations to 'now, cpu, 
object'?

static int
__update_load_avg_blocked_se(u64 now, int cpu, struct sched_entity *se)

static int
__update_load_avg_se(u64 now, int cpu, struct sched_entity *se)

static int
__update_load_avg_cfs_rq(u64 now, int cpu, struct cfs_rq *cfs_rq)

[...]