Merge "hrtimer: reprogram remote CPU's timer hardware" (4a1aae64) · Commits · e / devices / android_kernel_xiaomi_markw

Documentation/scheduler/sched-hmp.txt

+11 −0

Original line number	Diff line number	Diff line
		@@ -1485,6 +1485,17 @@ overcommitted scenario. See notes on sched_spill_nr_run and sched_spill_load for
		how overcommitment threshold is defined and also notes on
		'sched_upmigrate_min_nice' tunable.

		** 7.26 power_aware_timer_migration

		Appears at: /proc/sys/kernel/power_aware_timer_migration

		Default value: 0

		This tunable helps save power by offloading timer activity to cpus of lesser
		power cost (power cluster). Both high-res and low-res timers that are not pinned
		to one cpu are now enqueued on an online CPU found in least shallow C-state in
		the power-cluster. The current CPU is preferred in case of a tie.

		=========================
		8. HMP SCHEDULER TRACE POINTS
		=========================

include/linux/sched/sysctl.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -67,6 +67,7 @@ extern unsigned int sysctl_sched_downmigrate_pct;
		extern int sysctl_sched_upmigrate_min_nice;
		extern unsigned int sysctl_sched_powerband_limit_pct;
		extern unsigned int sysctl_sched_boost;
		extern unsigned int sysctl_power_aware_timer_migration;

		#ifdef CONFIG_SCHED_QHMP
		extern unsigned int sysctl_sched_min_runtime;

kernel/sched/core.c

+69 −2

Original line number	Diff line number	Diff line
		@@ -648,6 +648,64 @@ void resched_cpu(int cpu)

		#ifdef CONFIG_SMP
		#ifdef CONFIG_NO_HZ_COMMON

		#ifdef CONFIG_SCHED_HMP

		__read_mostly unsigned int sysctl_power_aware_timer_migration;

		/*
		* Return the CPU found in shallowest C-state in least power-cost
		* cluster. Prefer the current CPU when there is a tie.
		*/
		static int _get_nohz_timer_target_hmp(void)
		{
		int i, best_cpu = smp_processor_id();
		int min_cost = INT_MAX, min_cstate = INT_MAX;

		if (!sysctl_power_aware_timer_migration)
		return nr_cpu_ids;

		rcu_read_lock();

		for_each_online_cpu(i) {
		struct rq *rq = cpu_rq(i);
		int cpu_cost = power_cost(i, 0);
		int cstate = rq->cstate;

		if (power_delta_exceeded(cpu_cost, min_cost)) {
		if (cpu_cost > min_cost)
		continue;

		best_cpu = i;
		min_cost = cpu_cost;
		min_cstate = cstate;
		continue;
		}

		if (cstate < min_cstate \|\| (i == smp_processor_id() &&
		cstate == min_cstate)) {
		best_cpu = i;
		min_cstate = cstate;
		}
		}
		rcu_read_unlock();

		return best_cpu;
		}

		#else

		static int _get_nohz_timer_target_hmp(void)
		{
		/*
		* sched_enable_hmp = 0 for !CONFIG_SCHED_HMP, which means we should not
		* come here for !CONFIG_SCHED_HMP
		*/
		return raw_smp_processor_id();
		}

		#endif

		/*
		* In the semi idle case, use the nearest busy cpu for migrating timers
		* from an idle cpu. This is good for power-savings.
		@@ -659,10 +717,19 @@ void resched_cpu(int cpu)
		int get_nohz_timer_target(int pinned)
		{
		int cpu = smp_processor_id();
		int i;
		int i, lower_power_cpu;
		struct sched_domain *sd;

		if (pinned \|\| !get_sysctl_timer_migration() \|\| !idle_cpu(cpu))
		if (pinned \|\| !get_sysctl_timer_migration())
		return cpu;

		if (sched_enable_hmp) {
		lower_power_cpu = _get_nohz_timer_target_hmp();
		if (lower_power_cpu < nr_cpu_ids)
		return lower_power_cpu;
		}

		if (!idle_cpu(cpu))
		return cpu;

		rcu_read_lock();

kernel/sched/qhmp_core.c

+69 −2

Original line number	Diff line number	Diff line
		@@ -648,6 +648,64 @@ void resched_cpu(int cpu)

		#ifdef CONFIG_SMP
		#ifdef CONFIG_NO_HZ_COMMON

		#ifdef CONFIG_SCHED_HMP

		__read_mostly unsigned int sysctl_power_aware_timer_migration;

		/*
		* Return the CPU found in shallowest C-state in least power-cost
		* cluster. Prefer the current CPU when there is a tie.
		*/
		static int _get_nohz_timer_target_hmp(void)
		{
		int i, best_cpu = smp_processor_id();
		int min_cost = INT_MAX, min_cstate = INT_MAX;

		if (!sysctl_power_aware_timer_migration)
		return nr_cpu_ids;

		rcu_read_lock();

		for_each_online_cpu(i) {
		struct rq *rq = cpu_rq(i);
		int cpu_cost = power_cost(0, i);
		int cstate = rq->cstate;

		if (power_delta_exceeded(cpu_cost, min_cost)) {
		if (cpu_cost > min_cost)
		continue;

		best_cpu = i;
		min_cost = cpu_cost;
		min_cstate = cstate;
		continue;
		}

		if (cstate < min_cstate \|\| (i == smp_processor_id() &&
		cstate == min_cstate)) {
		best_cpu = i;
		min_cstate = cstate;
		}
		}
		rcu_read_unlock();

		return best_cpu;
		}

		#else

		static int _get_nohz_timer_target_hmp(void)
		{
		/*
		* sched_enable_hmp = 0 for !CONFIG_SCHED_HMP, which means we should not
		* come here for !CONFIG_SCHED_HMP
		*/
		return raw_smp_processor_id();
		}

		#endif

		/*
		* In the semi idle case, use the nearest busy cpu for migrating timers
		* from an idle cpu. This is good for power-savings.
		@@ -659,10 +717,19 @@ void resched_cpu(int cpu)
		int get_nohz_timer_target(int pinned)
		{
		int cpu = smp_processor_id();
		int i;
		int i, lower_power_cpu;
		struct sched_domain *sd;

		if (pinned \|\| !get_sysctl_timer_migration() \|\| !idle_cpu(cpu))
		if (pinned \|\| !get_sysctl_timer_migration())
		return cpu;

		if (sched_enable_hmp) {
		lower_power_cpu = _get_nohz_timer_target_hmp();
		if (lower_power_cpu < nr_cpu_ids)
		return lower_power_cpu;
		}

		if (!idle_cpu(cpu))
		return cpu;

		rcu_read_lock();

kernel/sched/sched.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -934,6 +934,7 @@ extern void sched_account_irqtime(int cpu, struct task_struct *curr,
		unsigned int cpu_temp(int cpu);
		extern unsigned int nr_eligible_big_tasks(int cpu);
		extern void update_up_down_migrate(void);
		extern int power_delta_exceeded(unsigned int cpu_cost, unsigned int base_cost);

		/*
		* 'load' is in reference to "best cpu" at its best frequency.