[PATCH 1/2] sched: Some sched trace points should depend on SCHEDSTATS

[PATCH 1/2] sched: Some sched trace points should depend on SCHEDSTATS

[yangoliver]

When SCHEDSTATS is not enabled, follow trace points shouldn't be
accessed by default,

	sched:sched_stat_wait
	sched:sched_stat_sleep
	sched:sched_stat_iowait
	sched:sched_stat_blocked

These trace points should only be available when SCHEDSTATS is
enabled.

Signed-off-by: Yong Yang <yangoliver@gmail.com>
---
 include/trace/events/sched.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
index d57a575..5b34309 100644
--- a/include/trace/events/sched.h
+++ b/include/trace/events/sched.h
@@ -334,6 +334,7 @@ DECLARE_EVENT_CLASS(sched_stat_template,
 );
 
 
+#ifdef CONFIG_SCHEDSTATS
 /*
  * Tracepoint for accounting wait time (time the task is runnable
  * but not actually running due to scheduler contention).
@@ -364,6 +365,7 @@ DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
 DEFINE_EVENT(sched_stat_template, sched_stat_blocked,
 	     TP_PROTO(struct task_struct *tsk, u64 delay),
 	     TP_ARGS(tsk, delay));
+#endif
 
 /*
  * Tracepoint for accounting runtime (time the task is executing
-- 
2.4.0

Re: [PATCH 1/2] sched: Some sched trace points should depend on SCHEDSTATS

[Steven Rostedt]

On Mon, 27 Jul 2015 09:11:51 -0400
yangoliver <yang_oliver@hotmail.com> wrote:

> When SCHEDSTATS is not enabled, follow trace points shouldn't be
> accessed by default,
> 
> 	sched:sched_stat_wait
> 	sched:sched_stat_sleep
> 	sched:sched_stat_iowait
> 	sched:sched_stat_blocked
> 
> These trace points should only be available when SCHEDSTATS is
> enabled.
> 
> Signed-off-by: Yong Yang <yangoliver@gmail.com>
> ---
>  include/trace/events/sched.h | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
> index d57a575..5b34309 100644
> --- a/include/trace/events/sched.h
> +++ b/include/trace/events/sched.h
> @@ -334,6 +334,7 @@ DECLARE_EVENT_CLASS(sched_stat_template,
>  );
>  
>  
> +#ifdef CONFIG_SCHEDSTATS

Why not also surround the sched_stat_template too? That also creates
code even if no event uses it.

-- Steve

>  /*
>   * Tracepoint for accounting wait time (time the task is runnable
>   * but not actually running due to scheduler contention).
> @@ -364,6 +365,7 @@ DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
>  DEFINE_EVENT(sched_stat_template, sched_stat_blocked,
>  	     TP_PROTO(struct task_struct *tsk, u64 delay),
>  	     TP_ARGS(tsk, delay));
> +#endif
>  
>  /*
>   * Tracepoint for accounting runtime (time the task is executing

Re: [PATCH 1/2] sched: Some sched trace points should depend on SCHEDSTATS

[Oliver Yang]

On 2015/8/4 1:34, Steven Rostedt wrote:
> On Mon, 27 Jul 2015 09:11:51 -0400
> yangoliver <yang_oliver@hotmail.com> wrote:
>
>> When SCHEDSTATS is not enabled, follow trace points shouldn't be
>> accessed by default,
>>
>> 	sched:sched_stat_wait
>> 	sched:sched_stat_sleep
>> 	sched:sched_stat_iowait
>> 	sched:sched_stat_blocked
>>
>> These trace points should only be available when SCHEDSTATS is
>> enabled.
>>
>> Signed-off-by: Yong Yang <yangoliver@gmail.com>
>> ---
>>  include/trace/events/sched.h | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
>> index d57a575..5b34309 100644
>> --- a/include/trace/events/sched.h
>> +++ b/include/trace/events/sched.h
>> @@ -334,6 +334,7 @@ DECLARE_EVENT_CLASS(sched_stat_template,
>>  );
>>  
>>  
>> +#ifdef CONFIG_SCHEDSTATS
> Why not also surround the sched_stat_template too? That also creates
> code even if no event uses it.
Steve,

Because the sched_stat_template is used by sched_stat_runtime.
The sched_stat_runtime is used event without SCHEDSTATS option enabled.

And the se.sum_exec_runtime counter is also similar. I don't think this counter is
defined for SCHEDSTATS.

Yes. I just try to make our trace points design be similar with SCHEDSTATS counters
design.

> -- Steve
>
>>  /*
>>   * Tracepoint for accounting wait time (time the task is runnable
>>   * but not actually running due to scheduler contention).
>> @@ -364,6 +365,7 @@ DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
>>  DEFINE_EVENT(sched_stat_template, sched_stat_blocked,
>>  	     TP_PROTO(struct task_struct *tsk, u64 delay),
>>  	     TP_ARGS(tsk, delay));
>> +#endif
>>  
>>  /*
>>   * Tracepoint for accounting runtime (time the task is executing

Re: [PATCH 1/2] sched: Some sched trace points should depend on SCHEDSTATS

[yangoliver]

Steve,

Any comments about this patch?

I think I addressed your question about why sched_stat_template is still 
required. Because another tracepoint depends on it, and that tracepoint
doesn't depend on SCHEDSTATS.

If you thought this patch is valid, I could re-submit it with a separte 
patch.

On Tue, 4 Aug 2015, Oliver Yang wrote:

> 
> 
> On 2015/8/4 1:34, Steven Rostedt wrote:
> > On Mon, 27 Jul 2015 09:11:51 -0400
> > yangoliver <yang_oliver@hotmail.com> wrote:
> >
> >> When SCHEDSTATS is not enabled, follow trace points shouldn't be
> >> accessed by default,
> >>
> >> 	sched:sched_stat_wait
> >> 	sched:sched_stat_sleep
> >> 	sched:sched_stat_iowait
> >> 	sched:sched_stat_blocked
> >>
> >> These trace points should only be available when SCHEDSTATS is
> >> enabled.
> >>
> >> Signed-off-by: Yong Yang <yangoliver@gmail.com>
> >> ---
> >>  include/trace/events/sched.h | 2 ++
> >>  1 file changed, 2 insertions(+)
> >>
> >> diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
> >> index d57a575..5b34309 100644
> >> --- a/include/trace/events/sched.h
> >> +++ b/include/trace/events/sched.h
> >> @@ -334,6 +334,7 @@ DECLARE_EVENT_CLASS(sched_stat_template,
> >>  );
> >>  
> >>  
> >> +#ifdef CONFIG_SCHEDSTATS
> > Why not also surround the sched_stat_template too? That also creates
> > code even if no event uses it.
> Steve,
> 
> Because the sched_stat_template is used by sched_stat_runtime.
> The sched_stat_runtime is used event without SCHEDSTATS option enabled.
> 
> And the se.sum_exec_runtime counter is also similar. I don't think this counter is
> defined for SCHEDSTATS.
> 
> Yes. I just try to make our trace points design be similar with SCHEDSTATS counters
> design.
> 
> > -- Steve
> >
> >>  /*
> >>   * Tracepoint for accounting wait time (time the task is runnable
> >>   * but not actually running due to scheduler contention).
> >> @@ -364,6 +365,7 @@ DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
> >>  DEFINE_EVENT(sched_stat_template, sched_stat_blocked,
> >>  	     TP_PROTO(struct task_struct *tsk, u64 delay),
> >>  	     TP_ARGS(tsk, delay));
> >> +#endif
> >>  
> >>  /*
> >>   * Tracepoint for accounting runtime (time the task is executing
> 
>