Re: [PATCH v2] rcu: Allow to eliminate softirq processing from rcutree

["Paul E. McKenney" <paulmck@linux.ibm.com>]

On Wed, Mar 20, 2019 at 08:21:46AM -0700, Paul E. McKenney wrote:
> On Wed, Mar 20, 2019 at 12:32:19PM +0100, Sebastian Andrzej Siewior wrote:
> > On 2019-03-19 12:44:19 [+0100], To Paul E. McKenney wrote:
> > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > > index 0f31b79eb6761..0a719f726e149 100644
> > > --- a/kernel/rcu/tree.c
> > > +++ b/kernel/rcu/tree.c
> > …
> > > +/*
> > > + * Spawn per-CPU RCU core processing kthreads.
> > > + */
> > > +static int __init rcu_spawn_core_kthreads(void)
> > > +{
> > > +	int cpu;
> > > +
> > > +	for_each_possible_cpu(cpu)
> > > +		per_cpu(rcu_data.rcu_cpu_has_work, cpu) = 0;
> > > +	if (!IS_ENABLED(CONFIG_RCU_BOOST) && !rcu_softirq_enabled)
> > 
> > and this needs to become
> > -       if (!IS_ENABLED(CONFIG_RCU_BOOST) && !rcu_softirq_enabled)
> > +       if (!IS_ENABLED(CONFIG_RCU_BOOST) && rcu_softirq_enabled)
> > 
> > With this change and hunk that I just sent to Joel I get thee three RCU
> > modes with and without BOOST booted.
> > 
> > Unless there is something (and Paul agrees that the Joel hunk is
> > correct) I would post a v3 with those changes included.
> 
> Well, I did get a bunch of grace-period hangs of various sorts when I
> ran rcutorture like this:
> 
> 	tools/testing/selftests/rcutorture/bin/kvm.sh --memory 1G --bootargs rcunosoftirq
> 
> I also got a silent hard hang on TREE03.  And an rcutorture forward-progress
> failure, which goes along with the grace-period hangs.
> 
> So something does need adjustment.  I will try again with this change.

And it does seem to work better.  I will give it more intense testing
later on, but in the meantime I have merged this change into your
earlier patch.

We will see whether or not I am able summon up the courage to push it
into v5.2, however.  ;-)

							Thanx, Paul

> > > +		return 0;
> > > +	WARN_ONCE(smpboot_register_percpu_thread(&rcu_cpu_thread_spec), "%s: Could not start rcub kthread, OOM is now expected behavior\n", __func__);
> > > +	return 0;
> > > +}
> > > +early_initcall(rcu_spawn_core_kthreads);
> > 
> > Sebastian
> >