Commit 9a54f98e authored Jul 14, 2015 by Paul E. McKenney

rcu: Don't disable CPU hotplug during OOM notifiers

RCU's rcu_oom_notify() disables CPU hotplug in order to stabilize the
list of online CPUs, which it traverses. However, this is completely
pointless because smp_call_function_single() will quietly fail if invoked
on an offline CPU. Because the count of requests is incremented in the
rcu_oom_notify_cpu() function that is remotely invoked, everything works
nicely even in the face of concurrent CPU-hotplug operations.

Furthermore, in recent kernels, invoking get_online_cpus() from an OOM
notifier can result in deadlock. This commit therefore removes the
call to get_online_cpus() and put_online_cpus() from rcu_oom_notify().

Reported-by: Marcin Ślusarz <marcin.slusarz@gmail.com>
Reported-by: David Rientjes <rientjes@google.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: David Rientjes <rientjes@google.com>
Tested-by: Marcin Ślusarz <marcin.slusarz@gmail.com>

parent 3ad81779

kernel/rcu/tree_plugin.h

+0 −2

Original line number	Original line	Diff line number	Diff line
	@@ -1679,12 +1679,10 @@ static int rcu_oom_notify(struct notifier_block *self,
	*/		*/
	atomic_set(&oom_callback_count, 1);		atomic_set(&oom_callback_count, 1);

	get_online_cpus();
	for_each_online_cpu(cpu) {		for_each_online_cpu(cpu) {
	smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1);		smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1);
	cond_resched_rcu_qs();		cond_resched_rcu_qs();
	}		}
	put_online_cpus();

	/* Unconditionally decrement: no need to wake ourselves up. */		/* Unconditionally decrement: no need to wake ourselves up. */
	atomic_dec(&oom_callback_count);		atomic_dec(&oom_callback_count);