Re: [PATCH RT 1/4] rcu: Acquire RCU lock when disabling BHs

[Scott Wood <swood@redhat.com>]

On Thu, 2019-06-20 at 13:53 -0700, Paul E. McKenney wrote:
> On Tue, Jun 18, 2019 at 08:19:05PM -0500, Scott Wood wrote:
> > diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
> > index fb267bc04fdf..aca4e5e25ace 100644
> > --- a/include/linux/rcupdate.h
> > +++ b/include/linux/rcupdate.h
> > @@ -637,10 +637,12 @@ static inline void rcu_read_unlock(void)
> >  static inline void rcu_read_lock_bh(void)
> >  {
> >  	local_bh_disable();
> > +#ifndef CONFIG_PREEMPT_RT_FULL
> 
> How about this instead?
> 
> 	if (IS_ENABLED(CONFIG_PREEMPT_RT_FULL))
> 		return;

OK.

> > @@ -189,8 +193,10 @@ void __local_bh_enable_ip(unsigned long ip,
> > unsigned int cnt)
> >  	WARN_ON_ONCE(count < 0);
> >  	local_irq_enable();
> >  
> > -	if (!in_atomic())
> > +	if (!in_atomic()) {
> > +		rcu_read_unlock();
> >  		local_unlock(bh_lock);
> > +	}
> >  
> >  	preempt_check_resched();
> >  }
> 
> And I have to ask...
> 
> What did you do to test this change to kernel/softirq.c?  My past attempts
> to do this sort of thing have always run afoul of open-coded BH
> transitions.

Mostly rcutorture and loads such as kernel builds, on a debug kernel.  By
"open-coded BH transition" do you mean directly manipulating the preempt
count?  That would already be broken on RT.

-Scott