Re: [PATCH v2] tracing/perf: Move rcu_irq_enter/exit_irqson() to perf trace point hook

[Peter Zijlstra]

On Tue, Feb 11, 2020 at 10:54:57AM -0800, Paul E. McKenney wrote:
> On Tue, Feb 11, 2020 at 06:32:13PM +0100, Peter Zijlstra wrote:
> > On Tue, Feb 11, 2020 at 06:29:52PM +0100, Peter Zijlstra wrote:
> > > +#define trace_rcu_enter()					\
> > > +({								\
> > > +	unsigned long state = 0;				\
> > > +	if (!rcu_is_watching())	{				\
> > 
> > Also, afaict rcu_is_watching() itself needs more love, the functio has
> > notrace, but calls other stuff that does not have notrace or inline.
> 
> Good catch!  Like this?
> 
> 							Thanx, Paul
> 
> ------------------------------------------------------------------------
> 
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index 1f5fdf7..51616e72 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -295,7 +295,7 @@ static void rcu_dynticks_eqs_online(void)
>   *
>   * No ordering, as we are sampling CPU-local information.
>   */
> -static bool rcu_dynticks_curr_cpu_in_eqs(void)
> +static bool notrace rcu_dynticks_curr_cpu_in_eqs(void)

Right, except that, given the size of the thing, I'd opt for 'inline'
over 'notrace'.