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

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

----- On Feb 12, 2020, at 3:02 AM, Peter Zijlstra peterz@infradead.org wrote:

> On Tue, Feb 11, 2020 at 12:35:21PM -0500, Mathieu Desnoyers wrote:
> 
>> Minor nits:
>> 
>> Why not make these an enum ?
>> 
>> > +
>> > +#define trace_rcu_enter()					\
>> > +({								\
>> > +	unsigned long state = 0;				\
>> > +	if (!rcu_is_watching())	{				\
>> > +		if (in_nmi()) {					\
>> > +			state = __TR_NMI;			\
>> > +			rcu_nmi_enter();			\
>> > +		} else {					\
>> > +			state = __TR_IRQ;			\
>> > +			rcu_irq_enter_irqson();			\
>> > +		}						\
>> > +	}							\
>> > +	state;							\
>> > +})
>> > +
>> > +#define trace_rcu_exit(state)					\
>> > +do {								\
>> > +	switch (state) {					\
>> > +	case __TR_IRQ:						\
>> > +		rcu_irq_exit_irqson();				\
>> > +		break;						\
>> > +	case __IRQ_NMI:						\
>> > +		rcu_nmi_exit();					\
>> > +		break;						\
>> > +	default:						\
>> > +		break;						\
>> > +	}							\
>> > +} while (0)
>> 
>> And convert these into static inline functions ?
> 
> Probably the same reason the rest of the file is macros; header hell.
> 
> Now, I could probably make these inlines, but then I'd have to add more
> RCU function declariations to this file (which is outside of
> rcupdate.h). Do we want to do that?

Probably not :) I was just wondering why.

Thanks,

Mathieu

> 
> The reason these were in this file is because I want to keep the comment
> and implementation near to nmi_enter/exit.

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com