Re: [RFC][PATCH 1/2 v2] tracing: Add TRACE_EVENT_CONDITIONAL()

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 2010-12-02 at 23:54 -0500, Mathieu Desnoyers wrote:
> * Steven Rostedt (rostedt@goodmis.org) wrote:
> [...]
> > -#define __DO_TRACE(tp, proto, args)					\
> > +#define __DO_TRACE(tp, proto, args, cond)				\
> >  	do {								\
> >  		struct tracepoint_func *it_func_ptr;			\
> >  		void *it_func;						\
> >  		void *__data;						\
> >  									\
> > +		if (!(cond))						\
> 
> One small documentation-related detail: my guess is that you are leaving
> "cond" without likely/unlikely builtin expect purposefully so that we
> can write, in TP_CONDITION:
> 
>   TP_CONDITION(unlikely(someparam)),

I actually think this is an abuse of "unlikely".

> 
> when we expect the condition to be usually false (and likely() for the
> reverse). Maybe it could be worth documenting that expressions like the
> following are valid :
> 
>   TP_CONDITION((likely(param1) && unlikely(param2)) || likely(param3))
> 
> It's fair to assume that kernel developers know this already, but given
> we plan to re-use TRACE_EVENT() for the user-space tracer soon enough,
> documenting this kind of use-case now can save us the trouble in the
> future.

I would frown on someone using unlikely here. But a TRACE_EVENT()
belongs to the maintainer, not me.

> 
> Other than that,
> 
> Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Thanks!

-- Steve