Re: [RFC][PATCH 3/4] tracing: add per-event filtering

[Tom Zanussi <tzanussi@gmail.com>]

On Tue, 2009-03-17 at 21:37 -0400, Steven Rostedt wrote:
> 
> 
> On Tue, 17 Mar 2009, Tom Zanussi wrote:
> > diff --git a/kernel/trace/trace_events_stage_3.h b/kernel/trace/trace_events_stage_3.h
> > index 30627d8..d556274 100644
> > --- a/kernel/trace/trace_events_stage_3.h
> > +++ b/kernel/trace/trace_events_stage_3.h
> > @@ -154,6 +154,7 @@ static struct ftrace_event_call event_##call;				\
> >  									\
> >  static void ftrace_raw_event_##call(proto)				\
> >  {									\
> > +	struct ftrace_event_call *call = &event_##call;			\
> >  	struct ring_buffer_event *event;				\
> >  	struct ftrace_raw_##call *entry;				\
> >  	unsigned long irq_flags;					\
> > @@ -172,6 +173,9 @@ static void ftrace_raw_event_##call(proto)				\
> >  	assign;								\
> >  									\
> >  	trace_current_buffer_unlock_commit(event, irq_flags, pc);	\
> 
> I have a concern hear about races.  Once the commit happens, there's 
> nothing protecting us from being preempted. The reader could in fact come 
> in and consume the commit, and a new write may happen, and this discard 
> will corrupt the data in the buffer. Also in overwrite mode, the writer 
> itself can over write this.
> 
> Would you be able to move the above unlock_commit after this.
> I'll make the necessary changes in the commit to check for discarded data.
> 

Yeah, I had meant to do that, but was having enough trouble with the
ring-buffer stuff that I wanted to avoid wondering what might happen if
I discarded an uncommitted event at that point.

I'll make that change in the next version.

Tom

> Thanks,
> 
> -- Steve
> 
> 
> 
> > +									
> \ > +	if (call->preds && !filter_match_preds(call, entry))		\
> > +		rb_event_discard(event);				\
> >  }									\
> >  									\
> >  static int ftrace_raw_reg_event_##call(void)				\
> > -- 
> > 1.5.6.3
> > 
> > 
> > 
> > 
> > 
> > 
> >