ARM: tegra: power: Use runnable threads average for hotplug

Sample scheduler runnable threads average in auto-hotplug work
function and use it to determine the auto-hotplug target for number
of on-line cores. Use cpu up delay as sampling period, and enforce
down delay by checking last cpu configuration change time stamp.

Bug 958978

Change-Id: I4280a11d39914687e6ffaa6f38df594d10aedaa9
Signed-off-by: Alex Frid <afrid@nvidia.com>
(cherry picked from commit 507e2ef5e4f09b23de2e924003dba259d3c8bc3c)
Reviewed-on: http://git-master/r/111638
Reviewed-by: Sai Gurrappadi <sgurrappadi@nvidia.com>
Tested-by: Sai Gurrappadi <sgurrappadi@nvidia.com>
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Peter Boonstoppel <pboonstoppel@nvidia.com>
Reviewed-by: Yu-Huan Hsu <yhsu@nvidia.com>

1 files changed, 39 insertions, 8 deletions

diff --git a/arch/arm/mach-tegra/cpu-tegra3.c b/arch/arm/mach-tegra/cpu-tegra3.c
index 5b9ac4f770a2..01b954625e54 100644
--- a/arch/arm/mach-tegra/cpu-tegra3.c
+++ b/arch/arm/mach-tegra/cpu-tegra3.c
@@ -73,6 +73,8 @@ static struct clk *cpu_clk;
 static struct clk *cpu_g_clk;
 static struct clk *cpu_lp_clk;
 
+static unsigned long last_change_time;
+
 static struct {
 	cputime64_t time_up_total;
 	u64 last_update;
@@ -186,6 +188,14 @@ enum {
 	TEGRA_CPU_SPEED_SKEWED,
 };
 
+#define NR_FSHIFT	2
+static unsigned int nr_run_thresholds[] = {
+/*      1,  2,  3,  4 - on-line cpus target */
+	5,  9, 13, UINT_MAX /* avg run threads * 4 (e.g., 9 = 2.25 threads) */
+};
+static unsigned int nr_run_hysteresis = 2;	/* 0.5 thread */
+static unsigned int nr_run_last;
+
 static noinline int tegra_cpu_speed_balance(void)
 {
 	unsigned long highest_speed = tegra_cpu_highest_speed();
@@ -194,17 +204,36 @@ static noinline int tegra_cpu_speed_balance(void)
 	unsigned int nr_cpus = num_online_cpus();
 	unsigned int max_cpus = pm_qos_request(PM_QOS_MAX_ONLINE_CPUS) ? : 4;
 	unsigned int min_cpus = pm_qos_request(PM_QOS_MIN_ONLINE_CPUS);
+	unsigned int avg_nr_run = avg_nr_running();
+	unsigned int nr_run;
+
+	/* Evaluate:
+	 * - distribution of freq targets for already on-lined CPUs
+	 * - average number of runnable threads
+	 * - effective MIPS available within EDP frequency limits,
+	 * and return:
+	 * TEGRA_CPU_SPEED_BALANCED to bring one more CPU core on-line
+	 * TEGRA_CPU_SPEED_BIASED to keep CPU core composition unchanged
+	 * TEGRA_CPU_SPEED_SKEWED to remove CPU core off-line
+	 */
+	for (nr_run = 1; nr_run < ARRAY_SIZE(nr_run_thresholds); nr_run++) {
+		unsigned int nr_threshold = nr_run_thresholds[nr_run - 1];
+		if (nr_run_last <= nr_run)
+			nr_threshold += nr_run_hysteresis;
+		if (avg_nr_run <= (nr_threshold << (FSHIFT - NR_FSHIFT)))
+			break;
+	}
+	nr_run_last = nr_run;
 
-	/* balanced: freq targets for all CPUs are above 50% of highest speed
-	   biased: freq target for at least one CPU is below 50% threshold
-	   skewed: freq targets for at least 2 CPUs are below 25% threshold */
 	if (((tegra_count_slow_cpus(skewed_speed) >= 2) ||
+	     (nr_run < nr_cpus) ||
 	     tegra_cpu_edp_favor_down(nr_cpus, mp_overhead) ||
 	     (highest_speed <= idle_bottom_freq) || (nr_cpus > max_cpus)) &&
 	    (nr_cpus > min_cpus))
 		return TEGRA_CPU_SPEED_SKEWED;
 
 	if (((tegra_count_slow_cpus(balanced_speed) >= 1) ||
+	     (nr_run <= nr_cpus) ||
 	     (!tegra_cpu_edp_favor_up(nr_cpus, mp_overhead)) ||
 	     (highest_speed <= idle_bottom_freq) || (nr_cpus == max_cpus)) &&
 	    (nr_cpus >= min_cpus))
@@ -218,7 +247,6 @@ static void tegra_auto_hotplug_work_func(struct work_struct *work)
 	bool up = false;
 	unsigned int cpu = nr_cpu_ids;
 	unsigned long now = jiffies;
-	static unsigned long last_change_time;
 
 	mutex_lock(tegra3_cpu_lock);
 
@@ -231,7 +259,8 @@ static void tegra_auto_hotplug_work_func(struct work_struct *work)
 		if (cpu < nr_cpu_ids) {
 			up = false;
 		} else if (!is_lp_cluster() && !no_lp &&
-			   !pm_qos_request(PM_QOS_MIN_ONLINE_CPUS)) {
+			   !pm_qos_request(PM_QOS_MIN_ONLINE_CPUS) &&
+			   ((now - last_change_time) >= down_delay)) {
 			if(!clk_set_parent(cpu_clk, cpu_lp_clk)) {
 				hp_stats_update(CONFIG_NR_CPUS, true);
 				hp_stats_update(0, false);
@@ -241,11 +270,12 @@ static void tegra_auto_hotplug_work_func(struct work_struct *work)
 			}
 		}
 		queue_delayed_work(
-			hotplug_wq, &hotplug_work, down_delay);
+			hotplug_wq, &hotplug_work, up2gn_delay);
 		break;
 	case TEGRA_HP_UP:
 		if (is_lp_cluster() && !no_lp) {
 			if(!clk_set_parent(cpu_clk, cpu_g_clk)) {
+				last_change_time = now;
 				hp_stats_update(CONFIG_NR_CPUS, false);
 				hp_stats_update(0, true);
 				/* catch-up with governor target speed */
@@ -307,6 +337,7 @@ static int min_cpus_notify(struct notifier_block *nb, unsigned long n, void *p)
 		tegra_update_cpu_speed(speed);
 
 		if (!clk_set_parent(cpu_clk, cpu_g_clk)) {
+			last_change_time = jiffies;
 			hp_stats_update(CONFIG_NR_CPUS, false);
 			hp_stats_update(0, true);
 		}
@@ -372,7 +403,7 @@ void tegra_auto_hotplug_governor(unsigned int cpu_freq, bool suspend)
 		} else if (cpu_freq <= bottom_freq) {
 			hp_state = TEGRA_HP_DOWN;
 			queue_delayed_work(
-				hotplug_wq, &hotplug_work, down_delay);
+				hotplug_wq, &hotplug_work, up_delay);
 		}
 		break;
 	case TEGRA_HP_DOWN:
@@ -388,7 +419,7 @@ void tegra_auto_hotplug_governor(unsigned int cpu_freq, bool suspend)
 		if (cpu_freq <= bottom_freq) {
 			hp_state = TEGRA_HP_DOWN;
 			queue_delayed_work(
-				hotplug_wq, &hotplug_work, down_delay);
+				hotplug_wq, &hotplug_work, up_delay);
 		} else if (cpu_freq <= top_freq) {
 			hp_state = TEGRA_HP_IDLE;
 		}