Re: [PATCH 02/11] tracing: Introduce TRACE_EVENT_INJECT

[Steven Rostedt <rostedt@goodmis.org>]

On Sat, 2010-02-06 at 13:20 +0100, Frederic Weisbecker wrote:

> I agree with you that creating a new trace event macro is a bit
> insane. ftrace.h is already a nightmare. I just thought that
> having the injector set inside the same macro that the synchronous
> event is defined helped to make it clear about its nature: that
> it needs a secondary async catch up thing.
> 
> But a register_event_injector thing will work too, we probably
> better want that, given the maintainance pain otherwise.

We can add a register_event_injector later. For now, why not just add
the TRACE_EVENT() and then hook to it in perf using the normal
tracepoint mechanism.

You could add some macro around the trace_init_lock_class() call that
would facilitate finding all the locks you need. This would probably be
a bit more straight forward than to overload TRACE_EVENT() again.

> 
> I really would like to make something useful for everyone, could
> you tell me more about johill needs?

Well, basically he needed a way to cause polling to happen using an
event. From what I understood, the polling called the trace point. Just
enabling the trace point did nothing because the polling did not take
place. I thought about ways to enable this command when the trace point
was enabled. But in the end, it was specific to a driver and another
debugfs file seemed a better fit to initiate it.

-- Steve