On Thu, Oct 03, 2019 at 04:34:09PM +0200, Frederic Weisbecker wrote:
> On Wed, Oct 02, 2019 at 06:38:57PM -0700, paulmck@xxxxxxxxxx wrote:
> > From: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxx>
> >
> > CPU-hotplug removal operations run the multi_cpu_stop() function, which
> > relies on the scheduler to gain control from whatever is running on the
> > various online CPUs, including any nohz_full CPUs running long loops in
> > kernel-mode code. Lack of the scheduler-clock interrupt on such CPUs
> > can delay multi_cpu_stop() for several minutes and can also result in
> > RCU CPU stall warnings. This commit therefore causes CPU-hotplug removal
> > operations to enable the scheduler-clock interrupt on all online CPUs.
>
> So, like Peter said back then, there must be an issue in the scheduler
> such as a missing or mishandled preemption point.
Fair enough, but this is useful in the meantime.
> > [ paulmck: Apply Joel Fernandes TICK_DEP_MASK_RCU->TICK_DEP_BIT_RCU fix. ]
> > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxx>
> > ---
> > kernel/rcu/tree.c | 15 +++++++++++++++
> > 1 file changed, 15 insertions(+)
> >
> > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> > index f708d54..74bf5c65 100644
> > --- a/kernel/rcu/tree.c
> > +++ b/kernel/rcu/tree.c
> > @@ -2091,6 +2091,7 @@ static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf)
> > */
> > int rcutree_dead_cpu(unsigned int cpu)
> > {
> > + int c;
> > struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
> > struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */
> >
> > @@ -2101,6 +2102,10 @@ int rcutree_dead_cpu(unsigned int cpu)
> > rcu_boost_kthread_setaffinity(rnp, -1);
> > /* Do any needed no-CB deferred wakeups from this CPU. */
> > do_nocb_deferred_wakeup(per_cpu_ptr(&rcu_data, cpu));
> > +
> > + // Stop-machine done, so allow nohz_full to disable tick.
> > + for_each_online_cpu(c)
> > + tick_dep_clear_cpu(c, TICK_DEP_BIT_RCU);
>
> Just use tick_dep_clear() without for_each_online_cpu().
>
> > return 0;
> > }
> >
> > @@ -3074,6 +3079,7 @@ static void rcutree_affinity_setting(unsigned int cpu, int outgoing)
> > */
> > int rcutree_online_cpu(unsigned int cpu)
> > {
> > + int c;
> > unsigned long flags;
> > struct rcu_data *rdp;
> > struct rcu_node *rnp;
> > @@ -3087,6 +3093,10 @@ int rcutree_online_cpu(unsigned int cpu)
> > return 0; /* Too early in boot for scheduler work. */
> > sync_sched_exp_online_cleanup(cpu);
> > rcutree_affinity_setting(cpu, -1);
> > +
> > + // Stop-machine done, so allow nohz_full to disable tick.
> > + for_each_online_cpu(c)
> > + tick_dep_clear_cpu(c, TICK_DEP_BIT_RCU);
>
> Same here.
>
> > return 0;
> > }
> >
> > @@ -3096,6 +3106,7 @@ int rcutree_online_cpu(unsigned int cpu)
> > */
> > int rcutree_offline_cpu(unsigned int cpu)
> > {
> > + int c;
> > unsigned long flags;
> > struct rcu_data *rdp;
> > struct rcu_node *rnp;
> > @@ -3107,6 +3118,10 @@ int rcutree_offline_cpu(unsigned int cpu)
> > raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
> >
> > rcutree_affinity_setting(cpu, cpu);
> > +
> > + // nohz_full CPUs need the tick for stop-machine to work quickly
> > + for_each_online_cpu(c)
> > + tick_dep_set_cpu(c, TICK_DEP_BIT_RCU);
>
> And here you only need tick_dep_set() without for_each_online_cpu().
Thank you! I applied all three simplifications.
Thanx, Paul
> Thanks.
>
> > return 0;
> > }
> >
> > --
> > 2.9.5
> >
