[PATCH v2 08/22] sched/fair: Use IPC class to pick the busiest group

[Ricardo Neri]

As it iterates, update_sd_pick_busiest() keeps on selecting as busiest
sched groups of identical priority. Since both groups have the same
priority, either group is a good choice. The IPCC score of the tasks
placed a sched group can break this tie.

Pick as busiest the sched group that yields a higher IPCC score after
load balancing.

Cc: Ben Segall <bsegall@xxxxxxxxxx>
Cc: Daniel Bristot de Oliveira <bristot@xxxxxxxxxx>
Cc: Dietmar Eggemann <dietmar.eggemann@xxxxxxx>
Cc: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
Cc: Len Brown <len.brown@xxxxxxxxx>
Cc: Mel Gorman <mgorman@xxxxxxx>
Cc: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
Cc: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
Cc: Tim C. Chen <tim.c.chen@xxxxxxxxx>
Cc: Valentin Schneider <vschneid@xxxxxxxxxx>
Cc: x86@xxxxxxxxxx
Cc: linux-pm@xxxxxxxxxxxxxxx
Cc: linux-kernel@xxxxxxxxxxxxxxx
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
---
Changes since v1:
 * Added a comment to clarify why sched_asym_prefer() needs a tie breaker
   only in update_sd_pick_busiest(). (PeterZ)
 * Renamed functions for accuracy:
   sched_asym_class_prefer() >> sched_asym_ipcc_prefer()
   sched_asym_class_pick() >> sched_asym_ipcc_pick()
 * Reworded commit message for clarity.
---
 kernel/sched/fair.c | 75 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 75 insertions(+)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index e333f9623b3a..e8b181c31842 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9176,6 +9176,63 @@ static void update_sg_lb_stats_scores(struct sg_lb_ipcc_stats *sgcs,
 	sgs->ipcc_score_before = before;
 }
 
+/**
+ * sched_asym_ipcc_prefer - Select a sched group based on its IPCC score
+ * @a:	Load balancing statistics of @sg_a
+ * @b:	Load balancing statistics of @sg_b
+ *
+ * Returns: true if preferring @a has a higher IPCC score than @b after
+ * balancing load. Returns false otherwise.
+ */
+static bool sched_asym_ipcc_prefer(struct sg_lb_stats *a,
+				   struct sg_lb_stats *b)
+{
+	if (!sched_ipcc_enabled())
+		return false;
+
+	/* @a increases overall throughput after load balance. */
+	if (a->ipcc_score_after > b->ipcc_score_after)
+		return true;
+
+	/*
+	 * If @a and @b yield the same overall throughput, pick @a if
+	 * its current throughput is lower than that of @b.
+	 */
+	if (a->ipcc_score_after == b->ipcc_score_after)
+		return a->ipcc_score_before < b->ipcc_score_before;
+
+	return false;
+}
+
+/**
+ * sched_asym_ipcc_pick - Select a sched group based on its IPCC score
+ * @a:		A scheduling group
+ * @b:		A second scheduling group
+ * @a_stats:	Load balancing statistics of @a
+ * @b_stats:	Load balancing statistics of @b
+ *
+ * Returns: true if @a has the same priority and @a has tasks with IPCC classes
+ * that yield higher overall throughput after load balance.
+ * Returns false otherwise.
+ */
+static bool sched_asym_ipcc_pick(struct sched_group *a,
+				 struct sched_group *b,
+				 struct sg_lb_stats *a_stats,
+				 struct sg_lb_stats *b_stats)
+{
+	/*
+	 * Only use the class-specific preference selection if both sched
+	 * groups have the same priority. We are not looking at a specific
+	 * CPU. We do not care about the idle state of the groups'
+	 * preferred CPU.
+	 */
+	if (arch_asym_cpu_priority(a->asym_prefer_cpu, false) !=
+	    arch_asym_cpu_priority(b->asym_prefer_cpu, false))
+		return false;
+
+	return sched_asym_ipcc_prefer(a_stats, b_stats);
+}
+
 #else /* CONFIG_IPC_CLASSES */
 static void update_sg_lb_ipcc_stats(struct sg_lb_ipcc_stats *sgcs,
 				    struct rq *rq)
@@ -9193,6 +9250,14 @@ static void update_sg_lb_stats_scores(struct sg_lb_ipcc_stats *sgcs,
 {
 }
 
+static bool sched_asym_ipcc_pick(struct sched_group *a,
+				 struct sched_group *b,
+				 struct sg_lb_stats *a_stats,
+				 struct sg_lb_stats *b_stats)
+{
+	return false;
+}
+
 #endif /* CONFIG_IPC_CLASSES */
 
 /**
@@ -9452,6 +9517,16 @@ static bool update_sd_pick_busiest(struct lb_env *env,
 				      sds->busiest->asym_prefer_cpu,
 				      false))
 			return false;
+
+		/*
+		 * Unlike other callers of sched_asym_prefer(), here both @sg
+		 * and @sds::busiest have tasks running. When they have equal
+		 * priority, their IPC class scores can be used to select a
+		 * better busiest.
+		 */
+		if (sched_asym_ipcc_pick(sds->busiest, sg, &sds->busiest_stat, sgs))
+			return false;
+
 		break;
 
 	case group_misfit_task:
-- 
2.25.1