Re: [PATCH v9 4/7] tracepoint: Make rcuidle tracepoint callers use SRCU

[Steven Rostedt]

On Wed, 11 Jul 2018 21:28:25 -0700
Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote:

> On Wed, Jul 11, 2018 at 11:21:20PM -0400, Steven Rostedt wrote:
> > On Wed, 11 Jul 2018 13:52:49 -0700
> > Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote:
> >   
> > > > #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
> > > > 	extern struct tracepoint __tracepoint_##name;			\
> > > > 	static inline void trace_##name(proto)				\
> > > > 	{								\
> > > > 		if (static_key_false(&__tracepoint_##name.key))		\
> > > > 			__DO_TRACE(&__tracepoint_##name,		\
> > > > 				TP_PROTO(data_proto),			\
> > > > 				TP_ARGS(data_args),			\
> > > > 				TP_CONDITION(cond), 0);			\
> > > > 		if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {		\
> > > > 			rcu_read_lock_sched_notrace();			\
> > > > 			rcu_dereference_sched(__tracepoint_##name.funcs);\
> > > > 			rcu_read_unlock_sched_notrace();		\
> > > > 		}							\
> > > > 	}
> > > > 
> > > > Because lockdep would only trigger warnings when the tracepoint was
> > > > enabled and used in a place it shouldn't be, we added the above
> > > > IS_ENABLED(CONFIG_LOCKDEP) part to test regardless if the the
> > > > tracepoint was enabled or not. Because we do this, we don't need to
> > > > have the test in the __DO_TRACE() code itself. That means we can clean
> > > > up the code as per Peter's suggestion.    
> > > 
> > > Sounds good, I'm Ok with making this change.
> > > 
> > > Just to clarify, are you proposing to change the rcu_dereference_sched to
> > > rcu_dereference_raw in both __DECLARE_TRACE and __DO_TRACE?  
> > 
> > No, just in __DO_TRACE(). The rcu_dereference_sched() above in
> > __DECLARE_TRACE() in the if (IS_ENABLED(CONFIG_LOCKDEP) block is
> > required to show the warnings if trace_##name() is used wrong, and is
> > the reason we can use rcu_dereference_raw() in __DO_TRACE() in the
> > first place ;-)
> > 
> > This brings up another point. We should probably add to
> > __DECLARE_TRACE_RCU() this:
> > 
> > #ifndef MODULE
> > #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
> > 	static inline void trace_##name##_rcuidle(proto)		\
> > 	{								\
> > 		if (static_key_false(&__tracepoint_##name.key))		\
> > 			__DO_TRACE(&__tracepoint_##name,		\
> > 				TP_PROTO(data_proto),			\
> > 				TP_ARGS(data_args),			\
> > 				TP_CONDITION(cond), 1);			\
> > +		if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {		\
> > +			int idx;					\
> > +			idx = srcu_read_lock_notrace(&tracepoint_srcu); \
> > +			srcu_dereference_notrace(__tracepoint_##name.funcs, \
> > +						&tracepoint_srcu);	\
> > +			srcu_read_unlock_notrace(&tracepoint_srcu, idx); \
> > +		}							\
> > 	}
> > #else
> > 
> > 
> > So that lockdep works with trace_##name##__rcuidle() when the trace
> > event is not enabled.
> > 
> > But that should be a separate patch and not part of this series. I may
> > write that up tomorrow.  
> 
> Yes, that sounds good to me and would be good to add the safe guard there.
> But you meant srcu_dereference above, not srcu_dereference_notrace right?

We don't need to trace them. I believe that the "srcu_*_notrace" still
performs the lockdep checks. That's what we want. If they don't then we
should not use notrace. But I believe they still do lockdep.

-- Steve