cpu/hotplug: Plug death reporting race

Paul noticed that the conversion of the death reporting introduced a race
where the outgoing cpu might be delayed after waking the controll processor,
so it might not be able to call rcu_report_dead() before being physically
removed, leading to RCU stalls.

We cant call complete after rcu_report_dead(), so instead of going back to
busy polling, simply issue a function call to do the completion.

Fixes: 27d50c7eeb0f "rcu: Make CPU_DYING_IDLE an explicit call"
Reported-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20160302201127.GA23440@linux.vnet.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra <peterz@infradead.org>

diff --git a/kernel/cpu.c b/kernel/cpu.c
index ff8059b76a853a28e1514ad3d45b8ded23590156..93e9d89bb0ab80c9c71a8b9c4842ec425244f6c0 100644 (file)
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -755,14 +755,26 @@ static int notify_dead(unsigned int cpu)
        return 0;
 }
 
+static void cpuhp_complete_idle_dead(void *arg)
+{
+       struct cpuhp_cpu_state *st = arg;
+
+       complete(&st->done);
+}
+
 void cpuhp_report_idle_dead(void)
 {
        struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
 
        BUG_ON(st->state != CPUHP_AP_OFFLINE);
-       st->state = CPUHP_AP_IDLE_DEAD;
-       complete(&st->done);
        rcu_report_dead(smp_processor_id());
+       st->state = CPUHP_AP_IDLE_DEAD;
+       /*
+        * We cannot call complete after rcu_report_dead() so we delegate it
+        * to an online cpu.
+        */
+       smp_call_function_single(cpumask_first(cpu_online_mask),
+                                cpuhp_complete_idle_dead, st, 0);
 }
 
 #else