sched: Introduce Window Assisted Load Tracking (WALT)

use a window based view of time in order to track task
demand and CPU utilization in the scheduler.

Window Assisted Load Tracking (WALT) implementation credits:
 Srivatsa Vaddagiri, Steve Muckle, Syed Rameez Mustafa, Joonwoo Park,
 Pavan Kumar Kondeti, Olav Haugan

2016-03-06: Integration with EAS/refactoring by Vikram Mulukutla
            and Todd Kjos

Change-Id: I21408236836625d4e7d7de1843d20ed5ff36c708

Includes fixes for issues:

eas/walt: Use walt_ktime_clock() instead of ktime_get_ns() to avoid a
race resulting in watchdog resets
BUG: 29353986
Change-Id: Ic1820e22a136f7c7ebd6f42e15f14d470f6bbbdb

Handle walt accounting anomoly during resume

During resume, there is a corner case where on wakeup, a task's
prev_runnable_sum can go negative. This is a workaround that
fixes the condition and warns (instead of crashing).

BUG: 29464099
Change-Id: I173e7874324b31a3584435530281708145773508

Signed-off-by: Todd Kjos <tkjos@google.com>
Signed-off-by: Srinath Sridharan <srinathsr@google.com>
Signed-off-by: Juri Lelli <juri.lelli@arm.com>
[jstultz: fwdported to 4.4]
Signed-off-by: John Stultz <john.stultz@linaro.org>

include/linux/sched.h

@@ -317,6 +317,15 @@ extern char ___assert_task_state[1 - 2*!!(
 /* Task command name length */
 #define TASK_COMM_LEN 16
 
+enum task_event {
+	PUT_PREV_TASK   = 0,
+	PICK_NEXT_TASK  = 1,
+	TASK_WAKE       = 2,
+	TASK_MIGRATE    = 3,
+	TASK_UPDATE     = 4,
+	IRQ_UPDATE	= 5,
+};
+
 #include <linux/spinlock.h>
 
 /*
@@ -1276,6 +1285,41 @@ struct sched_statistics {
 };
 #endif
 
+#ifdef CONFIG_SCHED_WALT
+#define RAVG_HIST_SIZE_MAX  5
+
+/* ravg represents frequency scaled cpu-demand of tasks */
+struct ravg {
+	/*
+	 * 'mark_start' marks the beginning of an event (task waking up, task
+	 * starting to execute, task being preempted) within a window
+	 *
+	 * 'sum' represents how runnable a task has been within current
+	 * window. It incorporates both running time and wait time and is
+	 * frequency scaled.
+	 *
+	 * 'sum_history' keeps track of history of 'sum' seen over previous
+	 * RAVG_HIST_SIZE windows. Windows where task was entirely sleeping are
+	 * ignored.
+	 *
+	 * 'demand' represents maximum sum seen over previous
+	 * sysctl_sched_ravg_hist_size windows. 'demand' could drive frequency
+	 * demand for tasks.
+	 *
+	 * 'curr_window' represents task's contribution to cpu busy time
+	 * statistics (rq->curr_runnable_sum) in current window
+	 *
+	 * 'prev_window' represents task's contribution to cpu busy time
+	 * statistics (rq->prev_runnable_sum) in previous window
+	 */
+	u64 mark_start;
+	u32 sum, demand;
+	u32 sum_history[RAVG_HIST_SIZE_MAX];
+	u32 curr_window, prev_window;
+	u16 active_windows;
+};
+#endif
+
 struct sched_entity {
 	struct load_weight	load;		/* for load-balancing */
 	struct rb_node		run_node;
@@ -1433,6 +1477,15 @@ struct task_struct {
 	const struct sched_class *sched_class;
 	struct sched_entity se;
 	struct sched_rt_entity rt;
+#ifdef CONFIG_SCHED_WALT
+	struct ravg ravg;
+	/*
+	 * 'init_load_pct' represents the initial task load assigned to children
+	 * of this task
+	 */
+	u32 init_load_pct;
+#endif
+
 #ifdef CONFIG_CGROUP_SCHED
 	struct task_group *sched_task_group;
 #endif

include/linux/sched/sysctl.h

@@ -43,6 +43,11 @@ extern unsigned int sysctl_sched_is_big_little;
 extern unsigned int sysctl_sched_sync_hint_enable;
 extern unsigned int sysctl_sched_initial_task_util;
 extern unsigned int sysctl_sched_cstate_aware;
+#ifdef CONFIG_SCHED_WALT
+extern unsigned int sysctl_sched_use_walt_cpu_util;
+extern unsigned int sysctl_sched_use_walt_task_util;
+extern unsigned int sysctl_sched_walt_init_task_load_pct;
+#endif
 
 enum sched_tunable_scaling {
 	SCHED_TUNABLESCALING_NONE,

include/trace/events/sched.h

@@ -937,6 +937,155 @@ TRACE_EVENT(sched_tune_filter,
 		__entry->payoff, __entry->region)
 );
 
+#ifdef CONFIG_SCHED_WALT
+struct rq;
+
+TRACE_EVENT(walt_update_task_ravg,
+
+	TP_PROTO(struct task_struct *p, struct rq *rq, int evt,
+						u64 wallclock, u64 irqtime),
+
+	TP_ARGS(p, rq, evt, wallclock, irqtime),
+
+	TP_STRUCT__entry(
+		__array(	char,	comm,   TASK_COMM_LEN	)
+		__field(	pid_t,	pid			)
+		__field(	pid_t,	cur_pid			)
+		__field(unsigned int,	cur_freq		)
+		__field(	u64,	wallclock		)
+		__field(	u64,	mark_start		)
+		__field(	u64,	delta_m			)
+		__field(	u64,	win_start		)
+		__field(	u64,	delta			)
+		__field(	u64,	irqtime			)
+		__field(        int,    evt			)
+		__field(unsigned int,	demand			)
+		__field(unsigned int,	sum			)
+		__field(	 int,	cpu			)
+		__field(	u64,	cs			)
+		__field(	u64,	ps			)
+		__field(	u32,	curr_window		)
+		__field(	u32,	prev_window		)
+		__field(	u64,	nt_cs			)
+		__field(	u64,	nt_ps			)
+		__field(	u32,	active_windows		)
+	),
+
+	TP_fast_assign(
+		__entry->wallclock      = wallclock;
+		__entry->win_start      = rq->window_start;
+		__entry->delta          = (wallclock - rq->window_start);
+		__entry->evt            = evt;
+		__entry->cpu            = rq->cpu;
+		__entry->cur_pid        = rq->curr->pid;
+		__entry->cur_freq       = rq->cur_freq;
+		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
+		__entry->pid            = p->pid;
+		__entry->mark_start     = p->ravg.mark_start;
+		__entry->delta_m        = (wallclock - p->ravg.mark_start);
+		__entry->demand         = p->ravg.demand;
+		__entry->sum            = p->ravg.sum;
+		__entry->irqtime        = irqtime;
+		__entry->cs             = rq->curr_runnable_sum;
+		__entry->ps             = rq->prev_runnable_sum;
+		__entry->curr_window	= p->ravg.curr_window;
+		__entry->prev_window	= p->ravg.prev_window;
+		__entry->nt_cs		= rq->nt_curr_runnable_sum;
+		__entry->nt_ps		= rq->nt_prev_runnable_sum;
+		__entry->active_windows	= p->ravg.active_windows;
+	),
+
+	TP_printk("wc %llu ws %llu delta %llu event %d cpu %d cur_freq %u cur_pid %d task %d (%s) ms %llu delta %llu demand %u sum %u irqtime %llu"
+		" cs %llu ps %llu cur_window %u prev_window %u nt_cs %llu nt_ps %llu active_wins %u"
+		, __entry->wallclock, __entry->win_start, __entry->delta,
+		__entry->evt, __entry->cpu,
+		__entry->cur_freq, __entry->cur_pid,
+		__entry->pid, __entry->comm, __entry->mark_start,
+		__entry->delta_m, __entry->demand,
+		__entry->sum, __entry->irqtime,
+		__entry->cs, __entry->ps,
+		__entry->curr_window, __entry->prev_window,
+		  __entry->nt_cs, __entry->nt_ps,
+		  __entry->active_windows
+		)
+);
+
+TRACE_EVENT(walt_update_history,
+
+	TP_PROTO(struct rq *rq, struct task_struct *p, u32 runtime, int samples,
+			int evt),
+
+	TP_ARGS(rq, p, runtime, samples, evt),
+
+	TP_STRUCT__entry(
+		__array(	char,	comm,   TASK_COMM_LEN	)
+		__field(	pid_t,	pid			)
+		__field(unsigned int,	runtime			)
+		__field(	 int,	samples			)
+		__field(	 int,	evt			)
+		__field(	 u64,	demand			)
+		__field(unsigned int,	walt_avg		)
+		__field(unsigned int,	pelt_avg		)
+		__array(	 u32,	hist, RAVG_HIST_SIZE_MAX)
+		__field(	 int,	cpu			)
+	),
+
+	TP_fast_assign(
+		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
+		__entry->pid            = p->pid;
+		__entry->runtime        = runtime;
+		__entry->samples        = samples;
+		__entry->evt            = evt;
+		__entry->demand         = p->ravg.demand;
+		__entry->walt_avg = (__entry->demand << 10) / walt_ravg_window,
+		__entry->pelt_avg	= p->se.avg.util_avg;
+		memcpy(__entry->hist, p->ravg.sum_history,
+					RAVG_HIST_SIZE_MAX * sizeof(u32));
+		__entry->cpu            = rq->cpu;
+	),
+
+	TP_printk("%d (%s): runtime %u samples %d event %d demand %llu"
+		" walt %u pelt %u (hist: %u %u %u %u %u) cpu %d",
+		__entry->pid, __entry->comm,
+		__entry->runtime, __entry->samples, __entry->evt,
+		__entry->demand,
+		__entry->walt_avg,
+		__entry->pelt_avg,
+		__entry->hist[0], __entry->hist[1],
+		__entry->hist[2], __entry->hist[3],
+		__entry->hist[4], __entry->cpu)
+);
+
+TRACE_EVENT(walt_migration_update_sum,
+
+	TP_PROTO(struct rq *rq, struct task_struct *p),
+
+	TP_ARGS(rq, p),
+
+	TP_STRUCT__entry(
+		__field(int,		cpu			)
+		__field(int,		pid			)
+		__field(	u64,	cs			)
+		__field(	u64,	ps			)
+		__field(	s64,	nt_cs			)
+		__field(	s64,	nt_ps			)
+	),
+
+	TP_fast_assign(
+		__entry->cpu		= cpu_of(rq);
+		__entry->cs		= rq->curr_runnable_sum;
+		__entry->ps		= rq->prev_runnable_sum;
+		__entry->nt_cs		= (s64)rq->nt_curr_runnable_sum;
+		__entry->nt_ps		= (s64)rq->nt_prev_runnable_sum;
+		__entry->pid		= p->pid;
+	),
+
+	TP_printk("cpu %d: cs %llu ps %llu nt_cs %lld nt_ps %lld pid %d",
+		  __entry->cpu, __entry->cs, __entry->ps,
+		  __entry->nt_cs, __entry->nt_ps, __entry->pid)
+);
+#endif /* CONFIG_SCHED_WALT */
+
 #endif /* CONFIG_SMP */
 
 #endif /* _TRACE_SCHED_H */

init/Kconfig

@@ -392,6 +392,15 @@ config IRQ_TIME_ACCOUNTING
 
 endchoice
 
+config SCHED_WALT
+        bool "Support window based load tracking"
+        depends on SMP
+        help
+        This feature will allow the scheduler to maintain a tunable window
+	based set of metrics for tasks and runqueues. These metrics can be
+	used to guide task placement as well as task frequency requirements
+	for cpufreq governors.
+
 config BSD_PROCESS_ACCT
 	bool "BSD Process Accounting"
 	depends on MULTIUSER

kernel/sched/Makefile

@@ -15,6 +15,7 @@ obj-y += core.o loadavg.o clock.o cputime.o
 obj-y += idle_task.o fair.o rt.o deadline.o stop_task.o
 obj-y += wait.o completion.o idle.o
 obj-$(CONFIG_SMP) += cpupri.o cpudeadline.o energy.o
+obj-$(CONFIG_SCHED_WALT) += walt.o
 obj-$(CONFIG_SCHED_AUTOGROUP) += auto_group.o
 obj-$(CONFIG_SCHEDSTATS) += stats.o
 obj-$(CONFIG_SCHED_DEBUG) += debug.o

kernel/sched/core.c

@@ -89,6 +89,7 @@
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/sched.h>
+#include "walt.h"
 
 DEFINE_MUTEX(sched_domains_mutex);
 DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
@@ -1085,7 +1086,9 @@ static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new
 
 	dequeue_task(rq, p, 0);
 	p->on_rq = TASK_ON_RQ_MIGRATING;
+	double_lock_balance(rq, cpu_rq(new_cpu));
 	set_task_cpu(p, new_cpu);
+	double_unlock_balance(rq, cpu_rq(new_cpu));
 	raw_spin_unlock(&rq->lock);
 
 	rq = cpu_rq(new_cpu);
@@ -1309,6 +1312,8 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
 			p->sched_class->migrate_task_rq(p);
 		p->se.nr_migrations++;
 		perf_event_task_migrate(p);
+
+		walt_fixup_busy_time(p, new_cpu);
 	}
 
 	__set_task_cpu(p, new_cpu);
@@ -1937,6 +1942,10 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 {
 	unsigned long flags;
 	int cpu, success = 0;
+#ifdef CONFIG_SMP
+	struct rq *rq;
+	u64 wallclock;
+#endif
 
 	/*
 	 * If we are going to wake up a thread waiting for CONDITION we
@@ -1994,6 +2003,14 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 	 */
 	smp_rmb();
 
+	rq = cpu_rq(task_cpu(p));
+
+	raw_spin_lock(&rq->lock);
+	wallclock = walt_ktime_clock();
+	walt_update_task_ravg(rq->curr, rq, TASK_UPDATE, wallclock, 0);
+	walt_update_task_ravg(p, rq, TASK_WAKE, wallclock, 0);
+	raw_spin_unlock(&rq->lock);
+
 	p->sched_contributes_to_load = !!task_contributes_to_load(p);
 	p->state = TASK_WAKING;
 
@@ -2001,10 +2018,12 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 		p->sched_class->task_waking(p);
 
 	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
+
 	if (task_cpu(p) != cpu) {
 		wake_flags |= WF_MIGRATED;
 		set_task_cpu(p, cpu);
 	}
+
 #endif /* CONFIG_SMP */
 
 	ttwu_queue(p, cpu);
@@ -2053,8 +2072,13 @@ static void try_to_wake_up_local(struct task_struct *p)
 
 	trace_sched_waking(p);
 
-	if (!task_on_rq_queued(p))
+	if (!task_on_rq_queued(p)) {
+		u64 wallclock = walt_ktime_clock();
+
+		walt_update_task_ravg(rq->curr, rq, TASK_UPDATE, wallclock, 0);
+		walt_update_task_ravg(p, rq, TASK_WAKE, wallclock, 0);
 		ttwu_activate(rq, p, ENQUEUE_WAKEUP);
+	}
 
 	ttwu_do_wakeup(rq, p, 0);
 	ttwu_stat(p, smp_processor_id(), 0);
@@ -2120,6 +2144,7 @@ static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
 	p->se.nr_migrations		= 0;
 	p->se.vruntime			= 0;
 	INIT_LIST_HEAD(&p->se.group_node);
+	walt_init_new_task_load(p);
 
 #ifdef CONFIG_SCHEDSTATS
 	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
@@ -2387,6 +2412,9 @@ void wake_up_new_task(struct task_struct *p)
 	struct rq *rq;
 
 	raw_spin_lock_irqsave(&p->pi_lock, flags);
+
+	walt_init_new_task_load(p);
+
 	/* Initialize new task's runnable average */
 	init_entity_runnable_average(&p->se);
 #ifdef CONFIG_SMP
@@ -2399,6 +2427,7 @@ void wake_up_new_task(struct task_struct *p)
 #endif
 
 	rq = __task_rq_lock(p);
+	walt_mark_task_starting(p);
 	activate_task(rq, p, ENQUEUE_WAKEUP_NEW);
 	p->on_rq = TASK_ON_RQ_QUEUED;
 	trace_sched_wakeup_new(p);
@@ -2948,9 +2977,12 @@ void scheduler_tick(void)
 	sched_clock_tick();
 
 	raw_spin_lock(&rq->lock);
+	walt_set_window_start(rq);
 	update_rq_clock(rq);
 	curr->sched_class->task_tick(rq, curr, 0);
 	update_cpu_load_active(rq);
+	walt_update_task_ravg(rq->curr, rq, TASK_UPDATE,
+			walt_ktime_clock(), 0);
 	calc_global_load_tick(rq);
 	sched_freq_tick(cpu);
 	raw_spin_unlock(&rq->lock);
@@ -3189,6 +3221,7 @@ static void __sched notrace __schedule(bool preempt)
 	unsigned long *switch_count;
 	struct rq *rq;
 	int cpu;
+	u64 wallclock;
 
 	cpu = smp_processor_id();
 	rq = cpu_rq(cpu);
@@ -3250,6 +3283,9 @@ static void __sched notrace __schedule(bool preempt)
 		update_rq_clock(rq);
 
 	next = pick_next_task(rq, prev);
+	wallclock = walt_ktime_clock();
+	walt_update_task_ravg(prev, rq, PUT_PREV_TASK, wallclock, 0);
+	walt_update_task_ravg(next, rq, PICK_NEXT_TASK, wallclock, 0);
 	clear_tsk_need_resched(prev);
 	clear_preempt_need_resched();
 	rq->clock_skip_update = 0;
@@ -5672,6 +5708,9 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
 	switch (action & ~CPU_TASKS_FROZEN) {
 
 	case CPU_UP_PREPARE:
+		raw_spin_lock_irqsave(&rq->lock, flags);
+		walt_set_window_start(rq);
+		raw_spin_unlock_irqrestore(&rq->lock, flags);
 		rq->calc_load_update = calc_load_update;
 		account_reset_rq(rq);
 		break;
@@ -5692,6 +5731,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
 		sched_ttwu_pending();
 		/* Update our root-domain */
 		raw_spin_lock_irqsave(&rq->lock, flags);
+		walt_migrate_sync_cpu(cpu);
 		if (rq->rd) {
 			BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
 			set_rq_offline(rq);
@@ -7536,6 +7576,7 @@ void __init sched_init_smp(void)
 {
 	cpumask_var_t non_isolated_cpus;
 
+	walt_init_cpu_efficiency();
 	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
 	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
 

kernel/sched/fair.c

@@ -30,11 +30,13 @@
 #include <linux/mempolicy.h>
 #include <linux/migrate.h>
 #include <linux/task_work.h>
+#include <linux/module.h>
 
 #include <trace/events/sched.h>
 
 #include "sched.h"
 #include "tune.h"
+#include "walt.h"
 
 /*
  * Targeted preemption latency for CPU-bound tasks:
@@ -56,6 +58,10 @@ unsigned int sysctl_sched_sync_hint_enable = 1;
 unsigned int sysctl_sched_initial_task_util = 0;
 unsigned int sysctl_sched_cstate_aware = 1;
 
+#ifdef CONFIG_SCHED_WALT
+unsigned int sysctl_sched_use_walt_cpu_util = 1;
+unsigned int sysctl_sched_use_walt_task_util = 1;
+#endif
 /*
  * The initial- and re-scaling of tunables is configurable
  * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
@@ -4225,6 +4231,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 		if (cfs_rq_throttled(cfs_rq))
 			break;
 		cfs_rq->h_nr_running++;
+		walt_inc_cfs_cumulative_runnable_avg(cfs_rq, p);
 
 		flags = ENQUEUE_WAKEUP;
 	}
@@ -4232,6 +4239,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 	for_each_sched_entity(se) {
 		cfs_rq = cfs_rq_of(se);
 		cfs_rq->h_nr_running++;
+		walt_inc_cfs_cumulative_runnable_avg(cfs_rq, p);
 
 		if (cfs_rq_throttled(cfs_rq))
 			break;
@@ -4246,6 +4254,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 #ifdef CONFIG_SMP
 
 	if (!se) {
+		walt_inc_cumulative_runnable_avg(rq, p);
 		if (!task_new && !rq->rd->overutilized &&
 		    cpu_overutilized(rq->cpu))
 			rq->rd->overutilized = true;
@@ -4295,6 +4304,7 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 		if (cfs_rq_throttled(cfs_rq))
 			break;
 		cfs_rq->h_nr_running--;
+		walt_dec_cfs_cumulative_runnable_avg(cfs_rq, p);
 
 		/* Don't dequeue parent if it has other entities besides us */
 		if (cfs_rq->load.weight) {
@@ -4315,6 +4325,7 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 	for_each_sched_entity(se) {
 		cfs_rq = cfs_rq_of(se);
 		cfs_rq->h_nr_running--;
+		walt_dec_cfs_cumulative_runnable_avg(cfs_rq, p);
 
 		if (cfs_rq_throttled(cfs_rq))
 			break;
@@ -4329,6 +4340,7 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 #ifdef CONFIG_SMP
 
 	if (!se) {
+		walt_dec_cumulative_runnable_avg(rq, p);
 
 		/*
 		 * We want to potentially trigger a freq switch
@@ -5228,6 +5240,12 @@ static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
 
 static inline unsigned long task_util(struct task_struct *p)
 {
+#ifdef CONFIG_SCHED_WALT
+	if (!walt_disabled && sysctl_sched_use_walt_task_util) {
+		unsigned long demand = p->ravg.demand;
+		return (demand << 10) / walt_ravg_window;
+	}
+#endif
 	return p->se.avg.util_avg;
 }
 
@@ -6620,7 +6638,9 @@ static void detach_task(struct task_struct *p, struct lb_env *env)
 
 	deactivate_task(env->src_rq, p, 0);
 	p->on_rq = TASK_ON_RQ_MIGRATING;
+	double_lock_balance(env->src_rq, env->dst_rq);
 	set_task_cpu(p, env->dst_cpu);
+	double_unlock_balance(env->src_rq, env->dst_rq);
 }
 
 /*

kernel/sched/rt.c

@@ -8,6 +8,8 @@
 #include <linux/slab.h>
 #include <linux/irq_work.h>
 
+#include "walt.h"
+
 int sched_rr_timeslice = RR_TIMESLICE;
 
 static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
@@ -1261,6 +1263,7 @@ enqueue_task_rt(struct rq *rq, struct task_struct *p, int flags)
 		rt_se->timeout = 0;
 
 	enqueue_rt_entity(rt_se, flags & ENQUEUE_HEAD);
+	walt_inc_cumulative_runnable_avg(rq, p);
 
 	if (!task_current(rq, p) && p->nr_cpus_allowed > 1)
 		enqueue_pushable_task(rq, p);
@@ -1272,6 +1275,7 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int flags)
 
 	update_curr_rt(rq);
 	dequeue_rt_entity(rt_se);
+	walt_dec_cumulative_runnable_avg(rq, p);
 
 	dequeue_pushable_task(rq, p);
 }

kernel/sched/sched.h

@@ -410,6 +410,10 @@ struct cfs_rq {
 	struct list_head leaf_cfs_rq_list;
 	struct task_group *tg;	/* group that "owns" this runqueue */
 
+#ifdef CONFIG_SCHED_WALT
+	u64 cumulative_runnable_avg;
+#endif
+
 #ifdef CONFIG_CFS_BANDWIDTH
 	int runtime_enabled;
 	u64 runtime_expires;
@@ -663,6 +667,27 @@ struct rq {
 	u64 max_idle_balance_cost;
 #endif
 
+#ifdef CONFIG_SCHED_WALT
+	/*
+	 * max_freq = user or thermal defined maximum
+	 * max_possible_freq = maximum supported by hardware
+	 */
+	unsigned int cur_freq, max_freq, min_freq, max_possible_freq;
+	struct cpumask freq_domain_cpumask;
+
+	u64 cumulative_runnable_avg;
+	int efficiency; /* Differentiate cpus with different IPC capability */
+	int load_scale_factor;
+	int capacity;
+	int max_possible_capacity;
+	u64 window_start;
+	u64 curr_runnable_sum;
+	u64 prev_runnable_sum;
+	u64 nt_curr_runnable_sum;
+	u64 nt_prev_runnable_sum;
+#endif /* CONFIG_SCHED_WALT */
+
+
 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
 	u64 prev_irq_time;
 #endif
@@ -1513,6 +1538,10 @@ static inline unsigned long capacity_orig_of(int cpu)
 	return cpu_rq(cpu)->cpu_capacity_orig;
 }
 
+extern unsigned int sysctl_sched_use_walt_cpu_util;
+extern unsigned int walt_ravg_window;
+extern unsigned int walt_disabled;
+
 /*
  * cpu_util returns the amount of capacity of a CPU that is used by CFS
  * tasks. The unit of the return value must be the one of capacity so we can
@@ -1544,6 +1573,11 @@ static inline unsigned long __cpu_util(int cpu, int delta)
 	unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
 	unsigned long capacity = capacity_orig_of(cpu);
 
+#ifdef CONFIG_SCHED_WALT
+	if (!walt_disabled && sysctl_sched_use_walt_cpu_util)
+		util = (cpu_rq(cpu)->prev_runnable_sum << SCHED_LOAD_SHIFT) /
+			walt_ravg_window;
+#endif
 	delta += util;
 	if (delta < 0)
 		return 0;

kernel/sched/stop_task.c

@@ -1,4 +1,5 @@
 #include "sched.h"
+#include "walt.h"
 
 /*
  * stop-task scheduling class.
@@ -42,12 +43,14 @@ static void
 enqueue_task_stop(struct rq *rq, struct task_struct *p, int flags)
 {
 	add_nr_running(rq, 1);
+	walt_inc_cumulative_runnable_avg(rq, p);
 }
 
 static void
 dequeue_task_stop(struct rq *rq, struct task_struct *p, int flags)
 {
 	sub_nr_running(rq, 1);
+	walt_dec_cumulative_runnable_avg(rq, p);
 }
 
 static void yield_task_stop(struct rq *rq)

kernel/sched/walt.h

@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2016, The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#ifndef __WALT_H
+#define __WALT_H
+
+#ifdef CONFIG_SCHED_WALT
+
+void walt_update_task_ravg(struct task_struct *p, struct rq *rq, int event,
+		u64 wallclock, u64 irqtime);
+void walt_inc_cumulative_runnable_avg(struct rq *rq, struct task_struct *p);
+void walt_dec_cumulative_runnable_avg(struct rq *rq, struct task_struct *p);
+void walt_inc_cfs_cumulative_runnable_avg(struct cfs_rq *rq,
+		struct task_struct *p);
+void walt_dec_cfs_cumulative_runnable_avg(struct cfs_rq *rq,
+		struct task_struct *p);
+void walt_fixup_busy_time(struct task_struct *p, int new_cpu);
+void walt_init_new_task_load(struct task_struct *p);
+void walt_mark_task_starting(struct task_struct *p);
+void walt_set_window_start(struct rq *rq);
+void walt_migrate_sync_cpu(int cpu);
+void walt_init_cpu_efficiency(void);
+u64 walt_ktime_clock(void);
+
+#else /* CONFIG_SCHED_WALT */
+
+static inline void walt_update_task_ravg(struct task_struct *p, struct rq *rq,
+		int event, u64 wallclock, u64 irqtime) { }
+static inline void walt_inc_cumulative_runnable_avg(struct rq *rq, struct task_struct *p) { }
+static inline void walt_dec_cumulative_runnable_avg(struct rq *rq, struct task_struct *p) { }
+static inline void walt_inc_cfs_cumulative_runnable_avg(struct cfs_rq *rq,
+		struct task_struct *p) { }
+static inline void walt_dec_cfs_cumulative_runnable_avg(struct cfs_rq *rq,
+		struct task_struct *p) { }
+static inline void walt_fixup_busy_time(struct task_struct *p, int new_cpu) { }
+static inline void walt_init_new_task_load(struct task_struct *p) { }
+static inline void walt_mark_task_starting(struct task_struct *p) { }
+static inline void walt_set_window_start(struct rq *rq) { }
+static inline void walt_migrate_sync_cpu(int cpu) { }
+static inline void walt_init_cpu_efficiency(void) { }
+static inline u64 walt_ktime_clock(void) { return 0; }
+
+#endif /* CONFIG_SCHED_WALT */
+
+extern unsigned int walt_disabled;
+
+#endif

kernel/sysctl.c

@@ -311,6 +311,29 @@ static struct ctl_table kern_table[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec,
 	},
+#ifdef CONFIG_SCHED_WALT
+	{
+		.procname	= "sched_use_walt_cpu_util",
+		.data		= &sysctl_sched_use_walt_cpu_util,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "sched_use_walt_task_util",
+		.data		= &sysctl_sched_use_walt_task_util,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "sched_walt_init_task_load_pct",
+		.data		= &sysctl_sched_walt_init_task_load_pct,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+#endif
 	{
 		.procname	= "sched_sync_hint_enable",
 		.data		= &sysctl_sched_sync_hint_enable,