Re: [PATCH 1/4] tracing: move __DO_TRACE out of line

[Mathieu Desnoyers]

* Ingo Molnar (mingo@xxxxxxx) wrote:
> 
> * Jeremy Fitzhardinge <jeremy@xxxxxxxx> wrote:
> 
> > From: Jeremy Fitzhardinge <jeremy.fitzhardinge@xxxxxxxxxx>
> > 
> > Mainly simplify linux/tracepoint.h's include dependencies (removes 
> > rcupdate.h), but it can't help with icache locality, since it 
> > definitely moves the code out of line, rather than relying on gcc 
> > to do it.
> 
> > --- a/include/linux/tracepoint.h
> > +++ b/include/linux/tracepoint.h
> > @@ -15,7 +15,6 @@
> >   */
> >  
> >  #include <linux/types.h>
> > -#include <linux/rcupdate.h>
> 
> nice!
> 
> > +#define DEFINE_DO_TRACE(name, proto, args)				\
> > +	void __do_trace_##name(struct tracepoint *tp, TP_PROTO(proto))	\
> > +	{								\
> 
> that needs to be marked notrace, otherwise the function tracer 
> becomes noisy. (or even lockupy.)
> 

I guess I'll have to put it more clearly : I am all for minimizing
tracepoint header dependency, but I'll be nacking this kind of
out-of-lining patch. Taking a function call, and moving it out-of-line
(thus duplicating the function call for nothing) seems *really*
pointless and will hurt tracer performance.

If thread_info.h is now so big that it needs a cleanup, I guess we'll
just have to do it.

Mathieu

> 	Ingo

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/