Date:	Fri, 30 Apr 2010 16:14:37 -0400
From:	Steven Rostedt <rostedt@...dmis.org>
To:	Mathieu Desnoyers <compudj@...stal.dyndns.org>
Cc:	linux-kernel@...r.kernel.org, Ingo Molnar <mingo@...e.hu>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Thomas Gleixner <tglx@...utronix.de>,
	Peter Zijlstra <peterz@...radead.org>,
	Frederic Weisbecker <fweisbec@...il.com>,
	Arnaldo Carvalho de Melo <acme@...hat.com>,
	Lai Jiangshan <laijs@...fujitsu.com>,
	Li Zefan <lizf@...fujitsu.com>,
	Masami Hiramatsu <mhiramat@...hat.com>,
	Christoph Hellwig <hch@....de>
Subject: Re: [PATCH 04/10][RFC] tracing: Remove per event trace registering

On Fri, 2010-04-30 at 16:07 -0400, Mathieu Desnoyers wrote:
> * Steven Rostedt (rostedt@...dmis.org) wrote:
> > On Fri, 2010-04-30 at 15:06 -0400, Mathieu Desnoyers wrote:
> > 
> > > > If it is possible sure, but that's the point. Where do you add the
> > > > check? The typecast is in the C code that is constant for all trace
> > > > events.
> > > 
> > > You can add the call to the static inline type check directly within the
> > > generated probe function, right after the local variable declarations.
> > 
> > Well, one thing, the callback is not going to be the same as the
> > DECLARE_TRACE() because the prototype ends with "void *data", and the
> > function being called actually uses the type of that data.
> > 
> > We now will have:
> > 
> > 	DEFINE_TRACE(mytracepoint, int myarg, myarg);
> > 
> > 	void mycallback(int myarg, struct mystuct *mydata);
> > 
> > 	register_trace_mytracepoint_data(mycallback, mydata)
> > 
> > There's no place in DEFINE_TRACE to be able to test the type of data
> > that is being passed back. I could make the calling function be:
> > 
> > 	void mycallback(int myarg, void *data)
> > 	{
> > 		struct mystruct *mydata = data;
> > 	[...]
> > 
> > Because the data is defined uniquely by the caller that registers a
> > callback. Each function can register its own data type.
> 
> Yep. There would need to be a cast from void * to struct mystruct *
> at the beginning of the callback as you propose here. I prefer this cast
> to be explicit (as proposed here) rather than hidden within the entire
> function call (void *) cast.
> 

OK, so you prefer that, I don't, but I also don't care, so I could
easily change it.


> > Let me explain this again:
> > 
> > 	DECLARE_TRACE(name, proto, args);
> > 
> > Will call the function like:
> > 
> > 	callback(args, data);
> > 
> > The callback will be at best:
> > 
> > 	int callback(proto, void *data);
> > 
> > 
> > because the data being passed in is not defined yet. It is defined at
> > the point of the registering of the callback. You can have two callbacks
> > registered to the same tracepoint with two different types as the data
> > field.
> > 
> > So what is it that this check is testing?
> 
> It's making sure that TRACE_EVENT() creates callbacks with the following
> signature:
> 
>   void callback(proto, void *data)
> 
> rather than
> 
>   void callback(proto, struct somestruct *data)
> 
> and forces the cast to be done within the callback rather than casting
> the whole function pointer type to void *, assuming types to match. I
> prefer to leave the cast outside of the tracepoint infrastructure, so we
> do not obfuscate the fact that an explicit type cast is needed there.

Fine, but I hardly see it as obfuscation. But my question again, even if
we do change this. What is this test testing? To me, it is checking that
CPP works.

-- Steve


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists