Commit f0f516f1 authored by Yipeng Zou's avatar Yipeng Zou
Browse files

sched: introduce smart grid qos zone 

hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I7ZBSR


CVE: NA

----------------------------------------

Since commit b869720191ec ("sched: smart grid: init sched_grid_qos
structure on QOS purpose") introduced a smart_grid-based QOS
partitioningmechanism, this commit further expands the partitioning
mechanism to implement smart_grid zone.

In the default configuration smart_grid the entire system is divided
into two partitions:

1. Hot zone (performance first)
2. Warm zone (energy consumption priority)

In addition, the smart_grid will dynamically maintain the size of the hot
zone in the current system based on the task load status in the current
partition, which based on commit 65523f55989a ("sched: Introduce smart
grid scheduling strategy for cfs").

 --------        --------        --------
| group0 |      | group1 |      | group2 |
 --------        --------        --------
    |                |              |
    v                v              v
 -------------------------    --------------
|                         |  |              |
|         hot zone        |  |   warm zone  |
|                         |  |              |
 -------------------------   ---------------

Signed-off-by: default avatarYipeng Zou <zouyipeng@huawei.com>
parent ae319ae1

include/linux/sched/grid_qos.h

+21 −0
Original line number Diff line number Diff line
@@ -84,7 +84,28 @@ void sched_grid_qos_free(struct task_struct *p); 


int sched_grid_preferred_interleave_nid(struct mempolicy *policy);

int sched_grid_preferred_nid(int preferred_nid, nodemask_t *nodemask);



enum sg_zone_type {

	SMART_GRID_ZONE_HOT = 0,

	SMART_GRID_ZONE_WARM,

	SMART_GRID_ZONE_NR

};



struct auto_affinity;

struct sched_grid_zone {

	raw_spinlock_t lock;

	struct cpumask cpus[SMART_GRID_ZONE_NR];

	struct list_head af_list_head;	/* struct auto_affinity list head */

};



int __init sched_grid_zone_init(void);

int sched_grid_zone_update(bool is_locked);

int sched_grid_zone_add_af(struct auto_affinity *af);

int sched_grid_zone_del_af(struct auto_affinity *af);

struct cpumask *sched_grid_zone_cpumask(enum sg_zone_type zone);

#else

static inline int __init sched_grid_zone_init(void) { return 0; }



static inline int

sched_grid_preferred_interleave_nid(struct mempolicy *policy)

{

kernel/sched/core.c

+6 −1
Original line number Diff line number Diff line
@@ -23,7 +23,7 @@ 
#include "../workqueue_internal.h"

#include "../../io_uring/io-wq.h"

#include "../smpboot.h"



#include <linux/sched/grid_qos.h>

#include "pelt.h"

#include "smp.h"
@@ -8143,6 +8143,7 @@ void __init sched_init_smp(void) 


	sched_smp_initialized = true;



	sched_grid_zone_init();

	init_auto_affinity(&root_task_group);

}
@@ -9635,6 +9636,10 @@ static int cpu_affinity_stat_show(struct seq_file *sf, void *v) 
	seq_printf(sf, "dcount %d\n", ad->dcount);

	seq_printf(sf, "domain_mask 0x%x\n", ad->domain_mask);

	seq_printf(sf, "curr_level %d\n", ad->curr_level);

	seq_printf(sf, "zone hot %*pbl\n",

			cpumask_pr_args(sched_grid_zone_cpumask(SMART_GRID_ZONE_HOT)));

	seq_printf(sf, "zone warm %*pbl\n",

			cpumask_pr_args(sched_grid_zone_cpumask(SMART_GRID_ZONE_WARM)));

	for (i = 0; i < ad->dcount; i++)

		seq_printf(sf, "sd_level %d, cpu list %*pbl, stay_cnt %llu\n",

			i, cpumask_pr_args(ad->domains[i]),

kernel/sched/fair.c

+7 −0
Original line number Diff line number Diff line
@@ -5874,6 +5874,7 @@ static void affinity_domain_up(struct task_group *tg) 
		if (IS_DOMAIN_SET(level + 1, ad->domain_mask) &&

		    cpumask_weight(ad->domains[level + 1]) > 0) {

			ad->curr_level = level + 1;

			sched_grid_zone_update(false);

			return;

		}

		level++;
@@ -5894,6 +5895,7 @@ static void affinity_domain_down(struct task_group *tg) 


		if (IS_DOMAIN_SET(level - 1, ad->domain_mask)) {

			ad->curr_level = level - 1;

			sched_grid_zone_update(false);

			return;

		}

		level--;
@@ -5959,6 +5961,7 @@ static int tg_update_affinity_domain_down(struct task_group *tg, void *data) 
		}



	}

	sched_grid_zone_update(false);

	raw_spin_unlock_irqrestore(&auto_affi->lock, flags);



	return 0;
@@ -6019,6 +6022,7 @@ void stop_auto_affinity(struct auto_affinity *auto_affi) 
	raw_spin_unlock_irq(&auto_affi->lock);



	smart_grid_usage_dec();

	sched_grid_zone_update(false);

	mutex_unlock(&smart_grid_used_mutex);

}
@@ -6220,6 +6224,8 @@ int init_auto_affinity(struct task_group *tg) 


	auto_affi->tg = tg;

	tg->auto_affinity = auto_affi;

	INIT_LIST_HEAD(&auto_affi->af_list);

	sched_grid_zone_add_af(auto_affi);

	return 0;

}
@@ -6234,6 +6240,7 @@ static void destroy_auto_affinity(struct task_group *tg) 
		smart_grid_usage_dec();



	hrtimer_cancel(&auto_affi->period_timer);

	sched_grid_zone_del_af(auto_affi);

	free_affinity_domains(&auto_affi->ad);



	kfree(tg->auto_affinity);

kernel/sched/grid/qos.c

+83 −0
Original line number Diff line number Diff line
@@ -23,6 +23,7 @@ 
#include <linux/numa.h>

#include <linux/sched/grid_qos.h>

#include "internal.h"

#include <../kernel/sched/sched.h>



static inline int qos_affinity_set(struct task_struct *p)

{
@@ -153,3 +154,85 @@ int sched_grid_preferred_nid(int preferred_nid, nodemask_t *nodemask) 


	return nd;

}



static struct sched_grid_zone sg_zone;



int __init sched_grid_zone_init(void)

{

	int index;



	for (index = 0; index < SMART_GRID_ZONE_NR; index++)

		cpumask_clear(&sg_zone.cpus[index]);



	raw_spin_lock_init(&sg_zone.lock);

	INIT_LIST_HEAD(&sg_zone.af_list_head);

	return 0;

}



int sched_grid_zone_update(bool is_locked)

{

	struct list_head *pos;

	struct auto_affinity *af_pos;

	unsigned long flags;



	if (!is_locked)

		raw_spin_lock_irqsave(&sg_zone.lock, flags);



	cpumask_clear(&sg_zone.cpus[SMART_GRID_ZONE_HOT]);



	list_for_each(pos, &sg_zone.af_list_head) {

		af_pos = list_entry(pos, struct auto_affinity, af_list);



		/* when smart_grid not used we need calculate all task_group */

		/* when smart_grid used we only calculate enabled task_group */

		if (smart_grid_used() && af_pos->mode == 0)

			continue;



		cpumask_or(&sg_zone.cpus[SMART_GRID_ZONE_HOT], &sg_zone.cpus[SMART_GRID_ZONE_HOT],

			   af_pos->ad.domains[af_pos->ad.curr_level]);

	}



	cpumask_complement(&sg_zone.cpus[SMART_GRID_ZONE_WARM],

			   &sg_zone.cpus[SMART_GRID_ZONE_HOT]);



	if (!is_locked)

		raw_spin_unlock_irqrestore(&sg_zone.lock, flags);



	return 0;

}



int sched_grid_zone_add_af(struct auto_affinity *af)

{

	unsigned long flags;



	if (af == NULL)

		return -1;



	raw_spin_lock_irqsave(&sg_zone.lock, flags);

	list_add_tail(&af->af_list, &sg_zone.af_list_head);

	sched_grid_zone_update(true);

	raw_spin_unlock_irqrestore(&sg_zone.lock, flags);

	return 0;

}



int sched_grid_zone_del_af(struct auto_affinity *af)

{

	unsigned long flags;



	if (af == NULL)

		return -1;



	raw_spin_lock_irqsave(&sg_zone.lock, flags);

	list_del(&af->af_list);

	sched_grid_zone_update(true);

	raw_spin_unlock_irqrestore(&sg_zone.lock, flags);

	return 0;

}



struct cpumask *sched_grid_zone_cpumask(enum sg_zone_type zone)

{

	if (zone >= SMART_GRID_ZONE_NR)

		return NULL;



	return &sg_zone.cpus[zone];

}

kernel/sched/sched.h

+1 −0
Original line number Diff line number Diff line
@@ -425,6 +425,7 @@ struct auto_affinity { 
	int			period_active;

	struct affinity_domain	ad;

	struct task_group	*tg;

	struct list_head	af_list;

#endif

};