[PATCH v2 openEuler-23.09] sched/fair: Introduce multiple qos level

5 Sep 2023

hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I7YS6M
-------------------------------
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: Zhao Wenhui zhaowenhui8@huawei.com
---
 arch/arm64/configs/openeuler_defconfig |  1 +
 arch/x86/configs/openeuler_defconfig   |  1 +
 include/linux/sched/sysctl.h           |  4 ++
 init/Kconfig                           |  9 ++++
 kernel/sched/core.c                    | 24 ++++++----
 kernel/sched/fair.c                    | 64 ++++++++++++++++++++++++--
 kernel/sched/sched.h                   | 33 +++++++++++++
 kernel/sysctl.c                        |  9 ++++
 8 files changed, 133 insertions(+), 12 deletions(-)

diff --git a/arch/arm64/configs/openeuler_defconfig b/arch/arm64/configs/openeuler_defconfig
index 37dc1012c94d..0203ca36f600 100644
--- a/arch/arm64/configs/openeuler_defconfig
+++ b/arch/arm64/configs/openeuler_defconfig
@@ -177,6 +177,7 @@ CONFIG_CGROUP_PERF=y
 CONFIG_CGROUP_BPF=y
 # CONFIG_CGROUP_MISC is not set
 CONFIG_QOS_SCHED=y
+CONFIG_QOS_SCHED_MULTILEVEL=y
 # CONFIG_CGROUP_DEBUG is not set
 CONFIG_SOCK_CGROUP_DATA=y
 CONFIG_NAMESPACES=y
diff --git a/arch/x86/configs/openeuler_defconfig b/arch/x86/configs/openeuler_defconfig
index d750bf90ce57..3d91106faee5 100644
--- a/arch/x86/configs/openeuler_defconfig
+++ b/arch/x86/configs/openeuler_defconfig
@@ -186,6 +186,7 @@ CONFIG_CGROUP_SCHED=y
 CONFIG_FAIR_GROUP_SCHED=y
 CONFIG_CFS_BANDWIDTH=y
 CONFIG_QOS_SCHED=y
+CONFIG_QOS_SCHED_MULTILEVEL=y
 CONFIG_RT_GROUP_SCHED=y
 CONFIG_SCHED_MM_CID=y
 CONFIG_QOS_SCHED_DYNAMIC_AFFINITY=y
diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
index 28d9be8e4614..3a02a76b08ca 100644
--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -37,4 +37,8 @@ extern unsigned int sysctl_overload_detect_period;
 extern unsigned int sysctl_offline_wait_interval;
 #endif
+#ifdef CONFIG_QOS_SCHED_MULTILEVEL
+extern unsigned int sysctl_qos_level_weights[];
+#endif
+
 #endif /* _LINUX_SCHED_SYSCTL_H */
diff --git a/init/Kconfig b/init/Kconfig
index d96c76143610..0bd8bb67e699 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -996,6 +996,15 @@ config QOS_SCHED
If in doubt, say N.
+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
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 652c06bd546d..238b5b55c38a 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7689,7 +7689,7 @@ 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;
    }
@@ -10356,7 +10356,7 @@ 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);
@@ -10385,7 +10385,7 @@ 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);
@@ -11275,7 +11275,7 @@ 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;
@@ -11297,19 +11297,27 @@ 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();
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index d9af04551788..3280ae7edd9e 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -190,6 +190,23 @@ unsigned int sysctl_offline_wait_interval = 100;  /* in ms */
 static int unthrottle_qos_cfs_rqs(int cpu);
 #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
@@ -3528,6 +3545,9 @@ 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);
@@ -3574,6 +3594,9 @@ 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
if (likely(se->load.weight == shares))
    	return;
@@ -8258,7 +8281,7 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_
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);
@@ -8433,7 +8456,7 @@ 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)) {
    	throttle_qos_cfs_rq(cfs_rq);
@@ -8449,7 +8472,7 @@ 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);
 }
@@ -8462,7 +8485,7 @@ 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));
@@ -8487,6 +8510,39 @@ 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;
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index f4e65a5e3009..a9eecb76dfff 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1213,6 +1213,39 @@ static inline bool is_migration_disabled(struct task_struct *p)
 #endif
 }
+#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_HIGH = 1,
+	QOS_LEVEL_HIGH_EX = 2
+};
+#else
+enum task_qos_level {
+	QOS_LEVEL_OFFLINE = -1,
+	QOS_LEVEL_ONLINE = 0,
+};
+#endif
+
+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
+
 DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
#define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index e9af234bf882..1714abd73f23 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -2068,6 +2068,15 @@ static struct ctl_table kern_table[] = {
    	.extra1		= SYSCTL_ONE_HUNDRED,
    	.extra2		= &one_thousand,
    },
+#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
    {
    	.procname	= "max_rcu_stall_to_panic",
-- 
2.33.0


    

2024

2023

2022

2021

2020

2019

[PATCH v2 openEuler-23.09] sched/fair: Introduce multiple qos level