sched/deadline: Optimize sched_dl_global_validate()

Under CONFIG_SMP, dl_bw is per root domain, but not per CPU.
When checking or updating dl_bw, currently iterating every CPU is
overdoing, just need iterate each root domain once.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Peng Liu <iwtbavbm@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lkml.kernel.org/r/78d21ee792cc48ff79e8cd62a5f26208463684d6.1602171061.git.iwtbavbm@gmail.com

kernel/sched/deadline.c

@@ -97,6 +97,17 @@ static inline unsigned long dl_bw_capacity(int i)
 		return __dl_bw_capacity(i);
 	}
 }
+
+static inline bool dl_bw_visited(int cpu, u64 gen)
+{
+	struct root_domain *rd = cpu_rq(cpu)->rd;
+
+	if (rd->visit_gen == gen)
+		return true;
+
+	rd->visit_gen = gen;
+	return false;
+}
 #else
 static inline struct dl_bw *dl_bw_of(int i)
 {
@@ -112,6 +123,11 @@ static inline unsigned long dl_bw_capacity(int i)
 {
 	return SCHED_CAPACITY_SCALE;
 }
+
+static inline bool dl_bw_visited(int cpu, u64 gen)
+{
+	return false;
+}
 #endif
 
 static inline
@@ -2535,11 +2551,15 @@ const struct sched_class dl_sched_class
 	.update_curr		= update_curr_dl,
 };
 
+/* Used for dl_bw check and update, used under sched_rt_handler()::mutex */
+static u64 dl_generation;
+
 int sched_dl_global_validate(void)
 {
 	u64 runtime = global_rt_runtime();
 	u64 period = global_rt_period();
 	u64 new_bw = to_ratio(period, runtime);
+	u64 gen = ++dl_generation;
 	struct dl_bw *dl_b;
 	int cpu, ret = 0;
 	unsigned long flags;
@@ -2548,13 +2568,13 @@ int sched_dl_global_validate(void)
 	 * Here we want to check the bandwidth not being set to some
 	 * value smaller than the currently allocated bandwidth in
 	 * any of the root_domains.
-	 *
-	 * FIXME: Cycling on all the CPUs is overdoing, but simpler than
-	 * cycling on root_domains... Discussion on different/better
-	 * solutions is welcome!
 	 */
 	for_each_possible_cpu(cpu) {
 		rcu_read_lock_sched();
+
+		if (dl_bw_visited(cpu, gen))
+			goto next;
+
 		dl_b = dl_bw_of(cpu);
 
 		raw_spin_lock_irqsave(&dl_b->lock, flags);
@@ -2562,6 +2582,7 @@ int sched_dl_global_validate(void)
 			ret = -EBUSY;
 		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
 
+next:
 		rcu_read_unlock_sched();
 
 		if (ret)
@@ -2587,6 +2608,7 @@ static void init_dl_rq_bw_ratio(struct dl_rq *dl_rq)
 void sched_dl_do_global(void)
 {
 	u64 new_bw = -1;
+	u64 gen = ++dl_generation;
 	struct dl_bw *dl_b;
 	int cpu;
 	unsigned long flags;
@@ -2597,11 +2619,14 @@ void sched_dl_do_global(void)
 	if (global_rt_runtime() != RUNTIME_INF)
 		new_bw = to_ratio(global_rt_period(), global_rt_runtime());
 
-	/*
-	 * FIXME: As above...
-	 */
 	for_each_possible_cpu(cpu) {
 		rcu_read_lock_sched();
+
+		if (dl_bw_visited(cpu, gen)) {
+			rcu_read_unlock_sched();
+			continue;
+		}
+
 		dl_b = dl_bw_of(cpu);
 
 		raw_spin_lock_irqsave(&dl_b->lock, flags);

kernel/sched/sched.h

@@ -801,6 +801,15 @@ struct root_domain {
 	struct dl_bw		dl_bw;
 	struct cpudl		cpudl;
 
+	/*
+	 * Indicate whether a root_domain's dl_bw has been checked or
+	 * updated. It's monotonously increasing value.
+	 *
+	 * Also, some corner cases, like 'wrap around' is dangerous, but given
+	 * that u64 is 'big enough'. So that shouldn't be a concern.
+	 */
+	u64 visit_gen;
+
 #ifdef HAVE_RT_PUSH_IPI
 	/*
 	 * For IPI pull requests, loop across the rto_mask.

kernel/sched/topology.c

@@ -516,6 +516,7 @@ static int init_rootdomain(struct root_domain *rd)
 	init_irq_work(&rd->rto_push_work, rto_push_irq_work_func);
 #endif
 
+	rd->visit_gen = 0;
 	init_dl_bw(&rd->dl_bw);
 	if (cpudl_init(&rd->cpudl) != 0)
 		goto free_rto_mask;