[PATCH] Fix race in sched_exit()

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

Fix a race on sleep_avg in sched_exit().

The symptom I saw on 64-bit s390 has been a fixpoint divide exception
because sleep_avg had a value > NS_MAX_SLEEP_AVG.  I tracked it down and
the problem is sched_exit which recalculates the parents sleep average
without taking the runqueue lock.  schedule() subtracts run_time from
sleep_avg of the previous process.  This can turn out negative and is
corrected shortly after the subtraction but that is already too late.
sched_exit() already read the negative value an miscalculated the parents
sleep_avg -> bang.

I fixed this by adding task_rq_lock/task_rq_unlock to sched_exit().

kernel/sched.c

@@ -819,6 +819,7 @@ void wake_up_forked_process(task_t * p)
 void sched_exit(task_t * p)
 {
 	unsigned long flags;
+	runqueue_t *rq;
 
 	local_irq_save(flags);
 	if (p->first_time_slice) {
@@ -831,10 +832,12 @@ void sched_exit(task_t * p)
 	 * If the child was a (relative-) CPU hog then decrease
 	 * the sleep_avg of the parent as well.
 	 */
+	rq = task_rq_lock(p->parent, &flags);
 	if (p->sleep_avg < p->parent->sleep_avg)
 		p->parent->sleep_avg = p->parent->sleep_avg /
 		(EXIT_WEIGHT + 1) * EXIT_WEIGHT + p->sleep_avg /
 		(EXIT_WEIGHT + 1);
+	task_rq_unlock(rq, &flags);
 }
 
 /**