Re: [PATCH] rcu: Add a warnings in sync_sched_exp_online_cleanup()

From: Paul E. McKenney
Date: Mon Jun 27 2022 - 20:40:29 EST


On Tue, Jun 28, 2022 at 12:31:12AM +0000, Zhang, Qiang1 wrote:
> >
> > Add Cc
> >
> > Currently, the sync_sched_exp_online_cleanup() is invoked in
> > cpuhp per-cpu kthreads when CPU is going online, so the CPU id
> > obtained by get_cpu() should always be equal to the CPU id of
> > the passed parameter, that is to say, the smp_call_function_single()
> > never be invoked, if be invoked, there may be problem with cpu-hotplug,
> > this commit add WARN_ON_ONCE() to remind everyone.
> >
> > Signed-off-by: Zqiang <qiang1.zhang@xxxxxxxxx>
> > ---
> > kernel/rcu/tree_exp.h | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/kernel/rcu/tree_exp.h b/kernel/rcu/tree_exp.h
> > index be667583a554..ae8dcfd4486c 100644
> > --- a/kernel/rcu/tree_exp.h
> > +++ b/kernel/rcu/tree_exp.h
> > @@ -865,6 +865,8 @@ static void sync_sched_exp_online_cleanup(int cpu)
> > put_cpu();
> > return;
> > }
> > +
> > + WARN_ON_ONCE(my_cpu != cpu);
>
> >If we are going to add this sort of warning, why not instead add it
> >to rcutree_online_cpu()?
> >
> >The reason the warning has not been present is the prospect of concurrent
> >onlining at boot time, which might have rcutree_online_cpu() invoked
> >from CPU 0 for multiple CPUs at boot. However, the for_each_online_cpu()
> >loop has recently been removed from rcu_init().
>
> The rcutree_online_cpu() is always invoked in per-cpu cpuhp kthreads,
> and even if it is called directly through the for_each_online_cpu() loop in
> rcu_init(), since the rcu_init() takes precedence over smp initialization,
> only the CPU0 is online at this time, still only called on CPU0.

And sync_sched_exp_online_cleanup() is only ever invoked from
rcutree_online_cpu(), correct?

Thanx, Paul

> Thanks
> Zqiang
>
> >
> >But I would like to hear what others think. Would such a warning
> >significantly help debugging?
> >
> > Thanx, Paul
> >
> > /* Quiescent state needed on some other CPU, send IPI. */
> > ret = smp_call_function_single(cpu, rcu_exp_handler, NULL, 0);
> > put_cpu();
> > --
> > 2.25.1
> >