Commit 0b620bf6 authored by Zhao Wenhui's avatar Zhao Wenhui
Browse files

sched/fair: Introduce multiple qos level 

hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I8S4JH



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

Expand qos_level from {-1,0} to [-2, 2], to distinguish the tasks expected
to be with extremely high or low priority level. Using qos_level_weight
to reweight the shares when calculating group's weight. Meanwhile,
set offline task's schedule policy to SCHED_IDLE so that it can be
preempted at check_preempt_wakeup.

Signed-off-by: default avatarZhao Wenhui <zhaowenhui8@huawei.com>
parent 45c76c86

init/Kconfig

+9 −0
Original line number Diff line number Diff line
@@ -1049,6 +1049,15 @@ config QOS_SCHED_PRIO_LB 
	  for Qos scheduler, which prefer migrating online tasks

	  and migrating offline tasks secondly between CPUs.



config QOS_SCHED_MULTILEVEL

	bool "Multiple qos level task scheduling"

	depends on QOS_SCHED

	default n

	help

	  This feature enable multiple qos level on task scheduling.

	  Expand the qos_level to [-2,2] to distinguish the tasks expected

	  to be with extremely high or low priority level.



config FAIR_GROUP_SCHED

	bool "Group scheduling for SCHED_OTHER"

	depends on CGROUP_SCHED

kernel/sched/core.c

+16 −8
Original line number Diff line number Diff line
@@ -7764,7 +7764,7 @@ static int __sched_setscheduler(struct task_struct *p, 
	 * other than SCHED_IDLE, the online task preemption and cpu resource

	 * isolation will be invalid, so return -EINVAL in this case.

	 */

	if (unlikely(task_group(p)->qos_level == -1 && !idle_policy(policy))) {

	if (unlikely(is_offline_level(task_group(p)->qos_level) && !idle_policy(policy))) {

		retval = -EINVAL;

		goto unlock;

	}
@@ -10413,7 +10413,7 @@ static void sched_change_qos_group(struct task_struct *tsk, struct task_group *t 
	 */

	if (!(tsk->flags & PF_EXITING) &&

	    !task_group_is_autogroup(tg) &&

	    (tg->qos_level == -1)) {

	    (is_offline_level(tg->qos_level))) {

		attr.sched_priority = 0;

		attr.sched_policy = SCHED_IDLE;

		attr.sched_nice = PRIO_TO_NICE(tsk->static_prio);
@@ -10442,7 +10442,7 @@ void sched_move_offline_task(struct task_struct *p) 
{

	struct offline_args *args;



	if (unlikely(task_group(p)->qos_level != -1))

	if (unlikely(!is_offline_level(task_group(p)->qos_level)))

		return;



	args = kmalloc(sizeof(struct offline_args), GFP_ATOMIC);
@@ -11339,7 +11339,7 @@ static int tg_change_scheduler(struct task_group *tg, void *data) 
	struct cgroup_subsys_state *css = &tg->css;



	tg->qos_level = qos_level;

	if (qos_level == -1)

	if (is_offline_level(qos_level))

		policy = SCHED_IDLE;

	else

		policy = SCHED_NORMAL;
@@ -11361,19 +11361,27 @@ static int cpu_qos_write(struct cgroup_subsys_state *css, 
	if (!tg->se[0])

		return -EINVAL;



	if (qos_level != -1 && qos_level != 0)

#ifdef CONFIG_QOS_SCHED_MULTILEVEL

	if (qos_level > QOS_LEVEL_HIGH_EX || qos_level < QOS_LEVEL_OFFLINE_EX)

#else

	if (qos_level != QOS_LEVEL_OFFLINE && qos_level != QOS_LEVEL_ONLINE)

#endif

		return -EINVAL;



	if (tg->qos_level == qos_level)

		goto done;



	if (tg->qos_level == -1 && qos_level == 0)

#ifdef CONFIG_QOS_SCHED_MULTILEVEL

	if (!is_normal_level(tg->qos_level))

#else

	if (tg->qos_level == QOS_LEVEL_OFFLINE && qos_level == QOS_LEVEL_ONLINE)

#endif

		return -EINVAL;



	cpus_read_lock();

	if (qos_level == -1)

	if (is_offline_level(qos_level))

		cfs_bandwidth_usage_inc();

	else

	else if (is_offline_level(tg->qos_level) && !is_offline_level(qos_level))

		cfs_bandwidth_usage_dec();

	cpus_read_unlock();

kernel/sched/fair.c

+70 −4
Original line number Diff line number Diff line
@@ -164,6 +164,23 @@ static DEFINE_PER_CPU(int, qos_smt_status); 
unsigned int sysctl_sched_prio_load_balance_enabled;

#endif



#ifdef CONFIG_QOS_SCHED_MULTILEVEL

#define QOS_LEVEL_WEIGHT_OFFLINE_EX	1

#define QOS_LEVEL_WEIGHT_OFFLINE	10

#define QOS_LEVEL_WEIGHT_ONLINE		100

#define QOS_LEVEL_WEIGHT_HIGH		1000

#define QOS_LEVEL_WEIGHT_HIGH_EX	10000



unsigned int sysctl_qos_level_weights[5] = {

	QOS_LEVEL_WEIGHT_OFFLINE_EX,

	QOS_LEVEL_WEIGHT_OFFLINE,

	QOS_LEVEL_WEIGHT_ONLINE,

	QOS_LEVEL_WEIGHT_HIGH,

	QOS_LEVEL_WEIGHT_HIGH_EX,

};

static long qos_reweight(long shares, struct task_group *tg);

#endif



#ifdef CONFIG_CFS_BANDWIDTH

/*

 * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
@@ -262,6 +279,15 @@ static struct ctl_table sched_fair_sysctls[] = { 
		.extra1		= SYSCTL_ZERO,

		.extra2		= SYSCTL_ONE,

	},

#endif

#ifdef CONFIG_QOS_SCHED_MULTILEVEL

	{

		.procname	= "qos_level_weights",

		.data		= &sysctl_qos_level_weights,

		.maxlen		= 5*sizeof(int),

		.mode		= 0644,

		.proc_handler	= proc_dointvec,

	},

#endif

	{}

};
@@ -3989,6 +4015,9 @@ static long calc_group_shares(struct cfs_rq *cfs_rq) 
	struct task_group *tg = cfs_rq->tg;



	tg_shares = READ_ONCE(tg->shares);

#ifdef CONFIG_QOS_SCHED_MULTILEVEL

	tg_shares = qos_reweight(tg_shares, tg);

#endif



	load = max(scale_load_down(cfs_rq->load.weight), cfs_rq->avg.load_avg);
@@ -4035,6 +4064,10 @@ static void update_cfs_group(struct sched_entity *se) 


#ifndef CONFIG_SMP

	shares = READ_ONCE(gcfs_rq->tg->shares);

#ifdef CONFIG_QOS_SCHED_MULTILEVEL

	shares = qos_reweight(shares, gcfs_rq->tg);

#endif



#else

	shares = calc_group_shares(gcfs_rq);

#endif
@@ -8678,7 +8711,7 @@ static inline void cancel_qos_timer(int cpu) 


static inline bool is_offline_task(struct task_struct *p)

{

	return task_group(p)->qos_level == -1;

	return task_group(p)->qos_level < QOS_LEVEL_ONLINE;

}



static void start_qos_hrtimer(int cpu);
@@ -8878,7 +8911,7 @@ static bool check_qos_cfs_rq(struct cfs_rq *cfs_rq) 
	if (unlikely(__this_cpu_read(qos_cpu_overload)))

		return false;



	if (unlikely(cfs_rq && cfs_rq->tg->qos_level < 0 &&

	if (unlikely(cfs_rq && is_offline_level(cfs_rq->tg->qos_level) &&

		     !sched_idle_cpu(smp_processor_id()) &&

		     cfs_rq->h_nr_running == cfs_rq->idle_h_nr_running)) {
@@ -8898,7 +8931,7 @@ static inline void unthrottle_qos_sched_group(struct cfs_rq *cfs_rq) 
	struct rq_flags rf;



	rq_lock_irqsave(rq, &rf);

	if (cfs_rq->tg->qos_level == -1 && cfs_rq_throttled(cfs_rq))

	if (is_offline_level(cfs_rq->tg->qos_level) && cfs_rq_throttled(cfs_rq))

		unthrottle_qos_cfs_rq(cfs_rq);

	rq_unlock_irqrestore(rq, &rf);

}
@@ -8911,7 +8944,7 @@ void sched_qos_offline_wait(void) 
		rcu_read_lock();

		qos_level = task_group(current)->qos_level;

		rcu_read_unlock();

		if (qos_level != -1 || fatal_signal_pending(current))

		if (!is_offline_level(qos_level) || fatal_signal_pending(current))

			break;



		schedule_timeout_killable(msecs_to_jiffies(sysctl_offline_wait_interval));
@@ -8941,6 +8974,39 @@ static enum hrtimer_restart qos_overload_timer_handler(struct hrtimer *timer) 
	return HRTIMER_NORESTART;

}



#ifdef CONFIG_QOS_SCHED_MULTILEVEL

static long qos_reweight(long shares, struct task_group *tg)

{

	long qos_weight = 100;

	long div = 100;

	long scale_shares;



	switch (tg->qos_level) {

	case QOS_LEVEL_OFFLINE_EX:

		qos_weight = sysctl_qos_level_weights[0];

		break;

	case QOS_LEVEL_OFFLINE:

		qos_weight = sysctl_qos_level_weights[1];

		break;

	case QOS_LEVEL_ONLINE:

		qos_weight = sysctl_qos_level_weights[2];

		break;

	case QOS_LEVEL_HIGH:

		qos_weight = sysctl_qos_level_weights[3];

		break;

	case QOS_LEVEL_HIGH_EX:

		qos_weight = sysctl_qos_level_weights[4];

		break;

	}

	if (qos_weight > LONG_MAX / shares)

		scale_shares = LONG_MAX / div;

	else

		scale_shares = shares * qos_weight / div;

	scale_shares = clamp_t(long, scale_shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));

	return scale_shares;

}

#endif



static void start_qos_hrtimer(int cpu)

{

	ktime_t time;

kernel/sched/sched.h

+25 −1
Original line number Diff line number Diff line
@@ -1440,11 +1440,20 @@ do { \ 
} while (0)



#ifdef CONFIG_QOS_SCHED

#ifdef CONFIG_QOS_SCHED_MULTILEVEL

enum task_qos_level {

	QOS_LEVEL_OFFLINE_EX = -2,

	QOS_LEVEL_OFFLINE = -1,

	QOS_LEVEL_ONLINE = 0,

	QOS_LEVEL_MAX

	QOS_LEVEL_HIGH = 1,

	QOS_LEVEL_HIGH_EX = 2

};

#else

enum task_qos_level {

	QOS_LEVEL_OFFLINE = -1,

	QOS_LEVEL_ONLINE = 0,

};

#endif

void init_qos_hrtimer(int cpu);

#endif
@@ -3323,6 +3332,21 @@ static inline int qos_idle_policy(int policy) 
{

	return policy == QOS_LEVEL_OFFLINE;

}



static inline int is_high_level(long qos_level)

{

	return qos_level > QOS_LEVEL_ONLINE;

}



static inline int is_normal_level(long qos_level)

{

	return qos_level == QOS_LEVEL_ONLINE;

}



static inline int is_offline_level(long qos_level)

{

	return qos_level < QOS_LEVEL_ONLINE;

}

#endif



#ifdef CONFIG_QOS_SCHED_SMT_EXPELLER