Re: [PATCH RT v3 3/5] sched: migrate_dis/enable: Use rt_invol_sleep

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

On 2019-09-23 11:59:23 [-0500], Scott Wood wrote:
> On Tue, 2019-09-17 at 09:06 -0500, Scott Wood wrote:
> > On Tue, 2019-09-17 at 09:59 +0200, Sebastian Andrzej Siewior wrote:
> > > On 2019-09-11 17:57:27 [+0100], Scott Wood wrote:
> > > > diff --git a/kernel/cpu.c b/kernel/cpu.c
> > > > index 885a195dfbe0..32c6175b63b6 100644
> > > > --- a/kernel/cpu.c
> > > > +++ b/kernel/cpu.c
> > > > @@ -308,7 +308,9 @@ void pin_current_cpu(void)
> > > >  	preempt_lazy_enable();
> > > >  	preempt_enable();
> > > >  
> > > > +	rt_invol_sleep_inc();
> > > >  	__read_rt_lock(cpuhp_pin);
> > > > +	rt_invol_sleep_dec();
> > > >  
> > > >  	preempt_disable();
> > > >  	preempt_lazy_disable();
> > > 
> > > I understand the other one. But now looking at it, both end up in
> > > rt_spin_lock_slowlock_locked() which would be the proper place to do
> > > that annotation. Okay.
> > 
> > FWIW, if my lazy migrate patchset is accepted, then there will be no users
> > of __read_rt_lock() outside rwlock-rt.c and it'll be moot.
> 
> I missed the "both" -- which is the "other one" that ends up in
> rt_spin_lock_slowlock_locked()?  stop_one_cpu() doesn't...

That one used here:
 __read_rt_lock()
    -> rt_spin_lock_slowlock_locked()

The official one (including the write part):
 rt_read_lock() (annotation)
   -> do_read_rt_lock()
     -> __read_rt_lock()
       -> rt_spin_lock_slowlock_locked()


and the only missing to the party of sleeping locks is:
rt_spin_lock() (annotation)
  -> rt_spin_lock_slowlock()
    -> rt_spin_lock_slowlock_locked()

> -Scott

Sebastian