sched: add a knob to prefer the waker CPU for sync wakeups

The current policy has a preference to select an idle CPU in the waker
cluster compared to the waker CPU running only 1 task. By selecting
an idle CPU, it eliminates the chance of waker migrating to a
different CPU after the wakee preempts it. This policy is also not
susceptible to the incorrect "sync" usage i.e the waker does not
goto sleep after waking up the wakee.

However LPM exit latency associated with an idle CPU outweigh the
above benefits on some targets. So add a knob to prefer the waker
CPU having only 1 runnable task over idle CPUs in the waker cluster.

Change-Id: Id974748c07625c1b19112235f426a5d204dfdb33
Signed-off-by: Pavankumar Kondeti <pkondeti@codeaurora.org>

5 files changed, 59 insertions, 8 deletions

diff --git a/Documentation/scheduler/sched-hmp.txt b/Documentation/scheduler/sched-hmp.txt
index 22449aec5558..b400e053e55d 100644
--- a/Documentation/scheduler/sched-hmp.txt
+++ b/Documentation/scheduler/sched-hmp.txt
@@ -1220,6 +1220,23 @@ This tunable is a percentage.  Configure the minimum demand of big sync waker
 task.  Scheduler places small wakee tasks woken up by big sync waker on the
 waker's cluster.
 
+*** 7.19 sched_prefer_sync_wakee_to_waker
+
+Appears at: /proc/sys/kernel/sched_prefer_sync_wakee_to_waker
+
+Default value: 0
+
+The default sync wakee policy has a preference to select an idle CPU in the
+waker cluster compared to the waker CPU running only 1 task. By selecting
+an idle CPU, it eliminates the chance of waker migrating to a different CPU
+after the wakee preempts it. This policy is also not susceptible to the
+incorrect "sync" usage i.e the waker does not goto sleep after waking up
+the wakee.
+
+However LPM exit latency associated with an idle CPU outweigh the above
+benefits on some targets. When this knob is turned on, the waker CPU is
+selected if it has only 1 runnable task.
+
 =========================
 8. HMP SCHEDULER TRACE POINTS
 =========================
diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
index 1f9c2c734b20..861f715a673d 100644
--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -64,6 +64,7 @@ extern unsigned int sysctl_sched_pred_alert_freq;
 extern unsigned int sysctl_sched_freq_aggregate;
 extern unsigned int sysctl_sched_enable_thread_grouping;
 extern unsigned int sysctl_sched_freq_aggregate_threshold_pct;
+extern unsigned int sysctl_sched_prefer_sync_wakee_to_waker;
 
 #else /* CONFIG_SCHED_HMP */
 
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 83da13b5f6b8..0a3d2e45daea 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2590,6 +2590,7 @@ static u32 __compute_runnable_contrib(u64 n)
 #define SBC_FLAG_COST_CSTATE_PREV_CPU_TIE_BREAKER	0x80
 #define SBC_FLAG_CSTATE_LOAD				0x100
 #define SBC_FLAG_BEST_SIBLING				0x200
+#define SBC_FLAG_WAKER_CPU				0x400
 
 /* Cluster selection flag */
 #define SBC_FLAG_COLOC_CLUSTER				0x10000
@@ -3060,6 +3061,15 @@ wake_to_waker_cluster(struct cpu_select_env *env)
 	       task_load(env->p) < sched_small_wakee_task_load;
 }
 
+static inline bool
+bias_to_waker_cpu(struct task_struct *p, int cpu)
+{
+	return sysctl_sched_prefer_sync_wakee_to_waker &&
+	       cpu_rq(cpu)->nr_running == 1 &&
+	       cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) &&
+	       cpu_active(cpu) && !cpu_isolated(cpu);
+}
+
 static inline int
 cluster_allowed(struct task_struct *p, struct sched_cluster *cluster)
 {
@@ -3080,6 +3090,7 @@ static int select_best_cpu(struct task_struct *p, int target, int reason,
 	struct cluster_cpu_stats stats;
 	struct related_thread_group *grp;
 	unsigned int sbc_flag = 0;
+	int cpu = raw_smp_processor_id();
 
 	struct cpu_select_env env = {
 		.p			= p,
@@ -3111,14 +3122,20 @@ static int select_best_cpu(struct task_struct *p, int target, int reason,
 		else
 			env.rtg = grp;
 	} else {
-		cluster = cpu_rq(smp_processor_id())->cluster;
-		if (wake_to_waker_cluster(&env) &&
-		    cluster_allowed(p, cluster)) {
-			env.need_waker_cluster = 1;
-			bitmap_zero(env.candidate_list, NR_CPUS);
-			__set_bit(cluster->id, env.candidate_list);
-			env.sbc_best_cluster_flag = SBC_FLAG_WAKER_CLUSTER;
-
+		cluster = cpu_rq(cpu)->cluster;
+		if (wake_to_waker_cluster(&env)) {
+			if (bias_to_waker_cpu(p, cpu)) {
+				target = cpu;
+				sbc_flag = SBC_FLAG_WAKER_CLUSTER |
+					   SBC_FLAG_WAKER_CPU;
+				goto out;
+			} else if (cluster_allowed(p, cluster)) {
+				env.need_waker_cluster = 1;
+				bitmap_zero(env.candidate_list, NR_CPUS);
+				__set_bit(cluster->id, env.candidate_list);
+				env.sbc_best_cluster_flag =
+							SBC_FLAG_WAKER_CLUSTER;
+			}
 		} else if (bias_to_prev_cpu(&env, &stats)) {
 			sbc_flag = SBC_FLAG_PREV_CPU;
 			goto out;
diff --git a/kernel/sched/hmp.c b/kernel/sched/hmp.c
index 6e1757aa1541..ba5048fe350a 100644
--- a/kernel/sched/hmp.c
+++ b/kernel/sched/hmp.c
@@ -890,6 +890,13 @@ unsigned int __read_mostly sched_spill_load;
 unsigned int __read_mostly sysctl_sched_spill_load_pct = 100;
 
 /*
+ * Prefer the waker CPU for sync wakee task, if the CPU has only 1 runnable
+ * task. This eliminates the LPM exit latency associated with the idle
+ * CPUs in the waker cluster.
+ */
+unsigned int __read_mostly sysctl_sched_prefer_sync_wakee_to_waker;
+
+/*
  * Tasks whose bandwidth consumption on a cpu is more than
  * sched_upmigrate are considered "big" tasks. Big tasks will be
  * considered for "up" migration, i.e migrating to a cpu with better
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index dad3324e7372..cdce7d0f5a0e 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -422,6 +422,15 @@ static struct ctl_table kern_table[] = {
 		.extra2		= &one_hundred,
 	},
 	{
+		.procname	= "sched_prefer_sync_wakee_to_waker",
+		.data		= &sysctl_sched_prefer_sync_wakee_to_waker,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &one,
+	},
+	{
 		.procname       = "sched_enable_thread_grouping",
 		.data           = &sysctl_sched_enable_thread_grouping,
 		.maxlen         = sizeof(unsigned int),