sched/uclamp: Fix incorrect condition

uclamp_update_active() should perform the update when
p->uclamp[clamp_id].active is true. But when the logic was inverted in
[1], the if condition wasn't inverted correctly too.

[1] https://lore.kernel.org/lkml/20190902073836.GO2369@hirez.programming.kicks-ass.net/

Reported-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Qais Yousef <qais.yousef@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Ben Segall <bsegall@google.com>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Patrick Bellasi <patrick.bellasi@matbug.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: babbe17 ("sched/uclamp: Update CPU's refcount on TG's clamp changes")
Link: https://lkml.kernel.org/r/20191114211052.15116-1-qais.yousef@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

kernel/sched/core.c

@@ -1065,7 +1065,7 @@ uclamp_update_active(struct task_struct *p, enum uclamp_id clamp_id)
 	 * affecting a valid clamp bucket, the next time it's enqueued,
 	 * it will already see the updated clamp bucket value.
 	 */
-	if (!p->uclamp[clamp_id].active) {
+	if (p->uclamp[clamp_id].active) {
 		uclamp_rq_dec_id(rq, p, clamp_id);
 		uclamp_rq_inc_id(rq, p, clamp_id);
 	}