sched: Improve wake_up_all_idle_cpus() take #2

commit 96611c2 upstream.

As reported by syzbot and experienced by Pavel, using cpus_read_lock()
in wake_up_all_idle_cpus() generates lock inversion (against mmap_sem
and possibly others).

Instead, shrink the preempt disable region by iterating all CPUs and
checking the online status for each individual CPU while having
preemption disabled.

Fixes: 8850cb6 ("sched: Simplify wake_up_*idle*()")
Reported-by: [email protected]
Reported-by: Pavel Machek <[email protected]>
Reported-by: Qian Cai <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Qian Cai <[email protected]>
(cherry picked from commit 96611c2)

Author: Peter Zijlstra

kernel/smp.c

@@ -1170,14 +1170,12 @@ void wake_up_all_idle_cpus(void)
 {
 	int cpu;
 
-	cpus_read_lock();
-	for_each_online_cpu(cpu) {
-		if (cpu == raw_smp_processor_id())
-			continue;
-
-		wake_up_if_idle(cpu);
+	for_each_possible_cpu(cpu) {
+		preempt_disable();
+		if (cpu != smp_processor_id() && cpu_online(cpu))
+			wake_up_if_idle(cpu);
+		preempt_enable();
 	}
-	cpus_read_unlock();
 }
 EXPORT_SYMBOL_GPL(wake_up_all_idle_cpus);