Re: [PATCH 4/7] rcu: handle quiescent states for PREEMPT_RCU=n, PREEMPT_COUNT=y

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2024-10-10 10:56:36 [-0700], Ankur Arora wrote:
> >
> > PREEMPT_LAZY selects PREEMPT_BUILD which selects PREEMPTION which in
> > turn selects PREEMPT_RCU. So this is not a valid combination. Do you
> > have a different tree than I do? Because maybe I am missing something.
> 
> The second patch in the series?

As long as "PREEMPT_RCU" as no text behind its bool you can't select it
independently.

> >> --- a/kernel/rcu/tree_plugin.h
> >> +++ b/kernel/rcu/tree_plugin.h
> >> @@ -974,13 +974,16 @@ static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
> >>   */
> >>  static void rcu_flavor_sched_clock_irq(int user)
> >>  {
> >> -	if (user || rcu_is_cpu_rrupt_from_idle()) {
> >> +	if (user || rcu_is_cpu_rrupt_from_idle() ||
> >> +	     (IS_ENABLED(CONFIG_PREEMPT_COUNT) &&
> >> +	      !(preempt_count() & (PREEMPT_MASK | SOFTIRQ_MASK)))) {
> >
> > couldn't you use a helper preemptible()?
> 
> Alas no. This check isn't trying to establish preemptibility (this is
> called in irq context so we already know that we aren't preemptible.)
> The check is using the preempt count to see if it can infer the state
> of RCU read side critical section on this CPU.

I see.

Sebastian