Re: [PATCH 0/3] tracing: introduce TRACE_EVENT_NONE and use it

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 26 Mar 2019 09:40:04 +0800
Yafang Shao <laoar.shao@gmail.com> wrote:

> In this patchset, I introduce a new macro TRACE_EVENT_NONE(), which will
> define a tracepoint as a do-nothing inline function.
> 	#define TRACE_EVENT_NONE(name, proto)			\
> 		static inline void trace_##name(proto)		\
> 		{ }						\
> 		static inline bool trace_##name##_enabled(void)	\
> 		{						\
> 			return false;				\
> 		}
> 

BTW, I prefer a name of TRACE_EVENT_DISABLED()

Which shows that it is disabled, or TRACE_EVENT_UNDEF(), or
TRACE_EVENT_NOP(). Something that states it is turned off. When I first
saw NONE() I thought it was another attempt to introduce zero parameter
trace events.


> Let's take some example for why this macro is needed.
> 
> - sched
> The tracepoints trace_sched_stat_{iowait, blocked, wait, sleep} should
> be not exposed to user if CONFIG_SCHEDSTATS is not set.
> We can place #ifdef in the kernel/sched/fair.c to fix it, but we
> don't want to sprinkle #ifdef.

But you still sprinkle #ifdef in the header, but the part I don't like
is the need to duplicate the prototypes and the like.


> So with TRACE_EVENT_NONE(), we can fix it in
> include/trace/events/sched.h.
> 
> - rcu
> When CONFIG_RCU_TRACE is not set, some rcu tracepoints are define as
> do-nothing macro without validating arguments, that is not proper.
> We should validate the arguments.
> 
> Yafang Shao (3):
>   tracing: introduce TRACE_EVENT_NONE()
>   sched/fair: do not expose some tracepoints to user if
>     CONFIG_SCHEDSTATS is not set
>   rcu: validate arguments for rcu tracepoints
> 

See my reply in the next patch.

-- Steve