Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit ff3a4812 authored by Satya Durga Srinivasu Prabhala's avatar Satya Durga Srinivasu Prabhala
Browse files

sched/walt: improve the scheduler 



This change is for general scheduler improvement.

Change-Id: Ie162a57537bb9ada66a4254d606e17d54b7a3a49
Signed-off-by: default avatarSatya Durga Srinivasu Prabhala <satyap@codeaurora.org>
[pkondeti@codeaurora.org: code refactoring and implemented freq to load
calculations.]
Signed-off-by: default avatarPavankumar Kondeti <pkondeti@codeaurora.org>
parent 20768179

include/linux/sched/sysctl.h

+7 −0
Original line number Diff line number Diff line
@@ -41,6 +41,7 @@ extern unsigned int sysctl_sched_group_upmigrate_pct; 
extern unsigned int sysctl_sched_group_downmigrate_pct;

extern unsigned int sysctl_sched_walt_rotate_big_tasks;

extern unsigned int sysctl_sched_min_task_util_for_boost_colocation;

extern unsigned int sysctl_sched_little_cluster_coloc_fmin_khz;



extern int

walt_proc_update_handler(struct ctl_table *table, int write,
@@ -122,4 +123,10 @@ extern int sysctl_schedstats(struct ctl_table *table, int write, 
				 void __user *buffer, size_t *lenp,

				 loff_t *ppos);



#ifdef CONFIG_SCHED_WALT

extern int sched_little_cluster_coloc_fmin_khz_handler(struct ctl_table *table,

					int write, void __user *buffer,

					size_t *lenp, loff_t *ppos);

#endif



#endif /* _LINUX_SCHED_SYSCTL_H */

include/trace/events/walt.h

+18 −4
Original line number Diff line number Diff line
@@ -493,8 +493,14 @@ DEFINE_EVENT(sched_cpu_load, sched_cpu_load_lb, 


TRACE_EVENT(sched_load_to_gov,



	TP_PROTO(struct rq *rq, u64 aggr_grp_load, u32 tt_load, u64 freq_aggr_thresh, u64 load, int policy, int big_task_rotation),

	TP_ARGS(rq, aggr_grp_load, tt_load, freq_aggr_thresh, load, policy, big_task_rotation),

	TP_PROTO(struct rq *rq, u64 aggr_grp_load, u32 tt_load,

		u64 freq_aggr_thresh, u64 load, int policy,

		int big_task_rotation,

		unsigned int sysctl_sched_little_cluster_coloc_fmin_khz,

		u64 coloc_boost_load),

	TP_ARGS(rq, aggr_grp_load, tt_load, freq_aggr_thresh, load, policy,

		big_task_rotation, sysctl_sched_little_cluster_coloc_fmin_khz,

		coloc_boost_load),



	TP_STRUCT__entry(

		__field(	int,	cpu			)
@@ -510,6 +516,9 @@ TRACE_EVENT(sched_load_to_gov, 
		__field(	u64,	pl			)

		__field(	u64,    load			)

		__field(	int,    big_task_rotation	)

		__field(unsigned int,

				sysctl_sched_little_cluster_coloc_fmin_khz)

		__field(u64,	coloc_boost_load)

	),



	TP_fast_assign(
@@ -526,13 +535,18 @@ TRACE_EVENT(sched_load_to_gov, 
		__entry->pl		= rq->walt_stats.pred_demands_sum;

		__entry->load		= load;

		__entry->big_task_rotation = big_task_rotation;

		__entry->sysctl_sched_little_cluster_coloc_fmin_khz =

				sysctl_sched_little_cluster_coloc_fmin_khz;

		__entry->coloc_boost_load = coloc_boost_load;

	),



	TP_printk("cpu=%d policy=%d ed_task_pid=%d aggr_grp_load=%llu freq_aggr_thresh=%llu tt_load=%llu rq_ps=%llu grp_rq_ps=%llu nt_ps=%llu grp_nt_ps=%llu pl=%llu load=%llu big_task_rotation=%d",

	TP_printk("cpu=%d policy=%d ed_task_pid=%d aggr_grp_load=%llu freq_aggr_thresh=%llu tt_load=%llu rq_ps=%llu grp_rq_ps=%llu nt_ps=%llu grp_nt_ps=%llu pl=%llu load=%llu big_task_rotation=%d sysctl_sched_little_cluster_coloc_fmin_khz=%u coloc_boost_load=%llu",

		__entry->cpu, __entry->policy, __entry->ed_task_pid,

		__entry->aggr_grp_load, __entry->freq_aggr_thresh,

		__entry->tt_load, __entry->rq_ps, __entry->grp_rq_ps,

		__entry->nt_ps, __entry->grp_nt_ps, __entry->pl, __entry->load,

		__entry->big_task_rotation)

		__entry->big_task_rotation,

		__entry->sysctl_sched_little_cluster_coloc_fmin_khz,

		__entry->coloc_boost_load)

);

#endif

kernel/sched/energy.c

+2 −0
Original line number Diff line number Diff line
@@ -274,6 +274,8 @@ static int sched_energy_probe(struct platform_device *pdev) 


	kfree(max_frequencies);



	walt_map_freq_to_load();



	dev_info(&pdev->dev, "Sched-energy-costs capacity updated\n");

	return 0;

kernel/sched/sched.h

+9 −0
Original line number Diff line number Diff line
@@ -108,6 +108,7 @@ struct sched_cluster { 
	int notifier_sent;

	bool wake_up_idle;

	u64 aggr_grp_load;

	u64 coloc_boost_load;

};



extern unsigned int sched_disable_window_stats;
@@ -2922,6 +2923,13 @@ static inline enum sched_boost_policy task_boost_policy(struct task_struct *p) 
	return policy;

}



extern void walt_map_freq_to_load(void);



static inline bool is_min_capacity_cluster(struct sched_cluster *cluster)

{

	return is_min_capacity_cpu(cluster_first_cpu(cluster));

}



#else	/* CONFIG_SCHED_WALT */



struct walt_sched_stats;
@@ -3069,6 +3077,7 @@ static inline unsigned int power_cost(int cpu, u64 demand) 
#endif



static inline void note_task_waking(struct task_struct *p, u64 wallclock) { }

static inline void walt_map_freq_to_load(void) { }

#endif	/* CONFIG_SCHED_WALT */



static inline bool energy_aware(void)

kernel/sched/walt.c

+76 −1
Original line number Diff line number Diff line
@@ -465,6 +465,7 @@ u64 freq_policy_load(struct rq *rq) 
	struct sched_cluster *cluster = rq->cluster;

	u64 aggr_grp_load = cluster->aggr_grp_load;

	u64 load, tt_load = 0;

	u64 coloc_boost_load = cluster->coloc_boost_load;



	if (rq->ed_task != NULL) {

		load = sched_ravg_window;
@@ -476,6 +477,9 @@ u64 freq_policy_load(struct rq *rq) 
	else

		load = rq->prev_runnable_sum + rq->grp_time.prev_runnable_sum;



	if (coloc_boost_load)

		load = max_t(u64, load, coloc_boost_load);



	tt_load = top_task_load(rq);

	switch (reporting_policy) {

	case FREQ_REPORT_MAX_CPU_LOAD_TOP_TASK:
@@ -492,7 +496,9 @@ u64 freq_policy_load(struct rq *rq) 


done:

	trace_sched_load_to_gov(rq, aggr_grp_load, tt_load, freq_aggr_thresh,

				load, reporting_policy, walt_rotation_enabled);

				load, reporting_policy, walt_rotation_enabled,

				sysctl_sched_little_cluster_coloc_fmin_khz,

				coloc_boost_load);

	return load;

}
@@ -2280,6 +2286,7 @@ struct sched_cluster init_cluster = { 
	.notifier_sent		=	0,

	.wake_up_idle		=	0,

	.aggr_grp_load		=	0,

	.coloc_boost_load	=	0,

};



void init_clusters(void)
@@ -3063,6 +3070,71 @@ static void transfer_busy_time(struct rq *rq, struct related_thread_group *grp, 
	BUG_ON((s64)*src_nt_prev_runnable_sum < 0);

}



/* Set to 1GHz by default */

unsigned int sysctl_sched_little_cluster_coloc_fmin_khz = 1000000;

static u64 coloc_boost_load;



void walt_map_freq_to_load(void)

{

	struct sched_cluster *cluster;



	for_each_sched_cluster(cluster) {

		if (is_min_capacity_cluster(cluster)) {

			int fcpu = cluster_first_cpu(cluster);



			coloc_boost_load = div64_u64(

				((u64)sched_ravg_window *

				arch_scale_cpu_capacity(NULL, fcpu) *

				sysctl_sched_little_cluster_coloc_fmin_khz),

				(u64)1024 * cpu_max_possible_freq(fcpu));

			coloc_boost_load = div64_u64(coloc_boost_load << 2, 5);

			break;

		}

	}

}



static void walt_update_coloc_boost_load(void)

{

	struct related_thread_group *grp;

	struct sched_cluster *cluster;



	if (!sysctl_sched_little_cluster_coloc_fmin_khz ||

			sysctl_sched_boost == CONSERVATIVE_BOOST)

		return;



	grp = lookup_related_thread_group(DEFAULT_CGROUP_COLOC_ID);

	if (!grp || !grp->preferred_cluster ||

			is_min_capacity_cluster(grp->preferred_cluster))

		return;



	for_each_sched_cluster(cluster) {

		if (is_min_capacity_cluster(cluster)) {

			cluster->coloc_boost_load = coloc_boost_load;

			break;

		}

	}

}



int sched_little_cluster_coloc_fmin_khz_handler(struct ctl_table *table,

				int write, void __user *buffer, size_t *lenp,

				loff_t *ppos)

{

	int ret;

	static DEFINE_MUTEX(mutex);



	mutex_lock(&mutex);



	ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);

	if (ret || !write)

		goto done;



	walt_map_freq_to_load();



done:

	mutex_unlock(&mutex);

	return ret;

}



/*

 * Runs in hard-irq context. This should ideally run just after the latest

 * window roll-over.
@@ -3100,10 +3172,13 @@ void walt_irq_work(struct irq_work *irq_work) 
		}



		cluster->aggr_grp_load = aggr_grp_load;

		cluster->coloc_boost_load = 0;



		raw_spin_unlock(&cluster->load_lock);

	}



	walt_update_coloc_boost_load();



	for_each_sched_cluster(cluster) {

		cpumask_t cluster_online_cpus;

		unsigned int num_cpus, i = 1;