sched: Don't use tasklist_lock for debug prints

Avoid taking locks from debug prints, this avoids latencies on -rt,
and improves reliability of the debug code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kernel/sched.c

@@ -6021,7 +6021,7 @@ void show_state_filter(unsigned long state_filter)
 	printk(KERN_INFO
 		"  task                        PC stack   pid father\n");
 #endif
-	read_lock(&tasklist_lock);
+	rcu_read_lock();
 	do_each_thread(g, p) {
 		/*
 		 * reset the NMI-timeout, listing all files on a slow
@@ -6037,7 +6037,7 @@ void show_state_filter(unsigned long state_filter)
 #ifdef CONFIG_SCHED_DEBUG
 	sysrq_sched_debug_show();
 #endif
-	read_unlock(&tasklist_lock);
+	rcu_read_unlock();
 	/*
 	 * Only show locks if all tasks are dumped:
 	 */