[PATCH openEuler-23.09 v2 3/7] sched: Adjust wakeup cpu range according CPU util dynamicly

14 Aug 2023

From: tanghui tanghui20@huawei.com
hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I7NS6Y
--------------------------------
Compare taskgroup 'util_avg' in perferred cpu with capacity preferred cpu,
dynamicly adjust cpu range for task wakeup process.
Signed-off-by: tanghui tanghui20@huawei.com
---
 include/linux/sched.h        |   1 +
 include/linux/sched/sysctl.h |   3 +
 kernel/sched/fair.c          | 133 +++++++++++++++++++++++++++++++++++
 kernel/sysctl.c              |  11 +++
 4 files changed, 148 insertions(+)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index ce6208bfb530..238ac254af0d 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1536,6 +1536,7 @@ struct task_struct {
#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
    cpumask_t 			*prefer_cpus;
+	const cpumask_t			*select_cpus;
 #endif
/*
diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
index 5a64582b086b..ede157a678f8 100644
--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -29,4 +29,7 @@ extern int sysctl_numa_balancing_mode;
 #define sysctl_numa_balancing_mode	0
 #endif
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+extern int sysctl_sched_util_low_pct;
+#endif
 #endif /* _LINUX_SCHED_SYSCTL_H */
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 373ff5f55884..b5265b610a7d 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -6704,7 +6704,11 @@ find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this
    	return cpumask_first(sched_group_span(group));
/* Traverse only the allowed CPUs */
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	for_each_cpu_and(i, sched_group_span(group), p->select_cpus) {
+#else
    for_each_cpu_and(i, sched_group_span(group), p->cpus_ptr) {
+#endif
    	struct rq *rq = cpu_rq(i);
if (!sched_core_cookie_match(rq, p))
@@ -6751,7 +6755,11 @@ static inline int find_idlest_cpu(struct sched_domain *sd, struct task_struct *p
 {
    int new_cpu = cpu;
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	if (!cpumask_intersects(sched_domain_span(sd), p->select_cpus))
+#else
    if (!cpumask_intersects(sched_domain_span(sd), p->cpus_ptr))
+#endif
    	return prev_cpu;
/*
@@ -6875,7 +6883,11 @@ static int select_idle_core(struct task_struct *p, int core, struct cpumask *cpu
    	if (!available_idle_cpu(cpu)) {
    		idle = false;
    		if (*idle_cpu == -1) {
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+				if (sched_idle_cpu(cpu) && cpumask_test_cpu(cpu, p->select_cpus)) {
+#else
    			if (sched_idle_cpu(cpu) && cpumask_test_cpu(cpu, p->cpus_ptr)) {
+#endif
    				*idle_cpu = cpu;
    				break;
    			}
@@ -6901,7 +6913,11 @@ static int select_idle_smt(struct task_struct *p, int target)
 {
    int cpu;
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	for_each_cpu_and(cpu, cpu_smt_mask(target), p->select_cpus) {
+#else
    for_each_cpu_and(cpu, cpu_smt_mask(target), p->cpus_ptr) {
+#endif
    	if (cpu == target)
    		continue;
    	if (available_idle_cpu(cpu) || sched_idle_cpu(cpu))
@@ -6949,7 +6965,11 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, bool
    struct sched_domain *this_sd = NULL;
    u64 time = 0;
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	cpumask_and(cpus, sched_domain_span(sd), p->select_cpus);
+#else
    cpumask_and(cpus, sched_domain_span(sd), p->cpus_ptr);
+#endif
if (sched_feat(SIS_PROP) && !has_idle_core) {
    	u64 avg_cost, avg_idle, span_avg;
@@ -7122,6 +7142,9 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
    lockdep_assert_irqs_disabled();
if ((available_idle_cpu(target) || sched_idle_cpu(target)) &&
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	    cpumask_test_cpu(target, p->select_cpus) &&
+#endif
        asym_fits_cpu(task_util, util_min, util_max, target))
    	return target;
@@ -7130,6 +7153,9 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
     */
    if (prev != target && cpus_share_cache(prev, target) &&
        (available_idle_cpu(prev) || sched_idle_cpu(prev)) &&
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	    cpumask_test_cpu(prev, p->select_cpus) &&
+#endif
        asym_fits_cpu(task_util, util_min, util_max, prev))
    	return prev;
@@ -7156,7 +7182,11 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
        recent_used_cpu != target &&
        cpus_share_cache(recent_used_cpu, target) &&
        (available_idle_cpu(recent_used_cpu) || sched_idle_cpu(recent_used_cpu)) &&
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	    cpumask_test_cpu(p->recent_used_cpu, p->select_cpus) &&
+#else
        cpumask_test_cpu(p->recent_used_cpu, p->cpus_ptr) &&
+#endif
        asym_fits_cpu(task_util, util_min, util_max, recent_used_cpu)) {
    	return recent_used_cpu;
    }
@@ -7638,6 +7668,82 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
    return target;
 }
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+/*
+ * Low utilization threshold for CPU
+ *
+ * (default: 85%), units: percentage of CPU utilization)
+ */
+int sysctl_sched_util_low_pct = 85;
+
+static inline bool prefer_cpus_valid(struct task_struct *p)
+{
+	return p->prefer_cpus &&
+	       !cpumask_empty(p->prefer_cpus) &&
+	       !cpumask_equal(p->prefer_cpus, p->cpus_ptr) &&
+	       cpumask_subset(p->prefer_cpus, p->cpus_ptr);
+}
+
+/*
+ * set_task_select_cpus: select the cpu range for task
+ * @p: the task whose available cpu range will to set
+ * @idlest_cpu: the cpu which is the idlest in prefer cpus
+ *
+ * If sum of 'util_avg' among 'prefer_cpus' lower than the percentage
+ * 'sysctl_sched_util_low_pct' of 'prefer_cpus' capacity, select
+ * 'prefer_cpus' range for task, otherwise select 'cpus_ptr' for task.
+ *
+ * The available cpu range set to p->select_cpus. Idlest cpu in preferred cpus
+ * set to @idlest_cpu, which is set to wakeup cpu when fast path wakeup cpu
+ * without p->select_cpus.
+ */
+static void set_task_select_cpus(struct task_struct *p, int *idlest_cpu,
+				 int sd_flag)
+{
+	unsigned long util_avg_sum = 0;
+	unsigned long tg_capacity = 0;
+	long min_util = INT_MIN;
+	struct task_group *tg;
+	long spare;
+	int cpu;
+
+	p->select_cpus = p->cpus_ptr;
+	if (!prefer_cpus_valid(p))
+		return;
+
+	rcu_read_lock();
+	tg = task_group(p);
+	for_each_cpu(cpu, p->prefer_cpus) {
+		if (unlikely(!tg->se[cpu]))
+			continue;
+
+		if (idlest_cpu && available_idle_cpu(cpu)) {
+			*idlest_cpu = cpu;
+		} else if (idlest_cpu) {
+			spare = (long)(capacity_of(cpu) - tg->se[cpu]->avg.util_avg);
+			if (spare > min_util) {
+				min_util = spare;
+				*idlest_cpu = cpu;
+			}
+		}
+
+		if (available_idle_cpu(cpu)) {
+			rcu_read_unlock();
+			p->select_cpus = p->prefer_cpus;
+			return;
+		}
+
+		util_avg_sum += tg->se[cpu]->avg.util_avg;
+		tg_capacity += capacity_of(cpu);
+	}
+	rcu_read_unlock();
+
+	if (tg_capacity > cpumask_weight(p->prefer_cpus) &&
+	    util_avg_sum * 100 <= tg_capacity * sysctl_sched_util_low_pct)
+		p->select_cpus = p->prefer_cpus;
+}
+#endif
+
 /*
  * select_task_rq_fair: Select target runqueue for the waking task in domains
  * that have the relevant SD flag set. In practice, this is SD_BALANCE_WAKE,
@@ -7658,11 +7764,19 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags)
    int want_affine = 0;
    /* SD_flags and WF_flags share the first nibble */
    int sd_flag = wake_flags & 0xF;
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	int idlest_cpu = 0;
+#endif
/*
     * required for stable ->cpus_allowed
     */
    lockdep_assert_held(&p->pi_lock);
+
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	set_task_select_cpus(p, &idlest_cpu, sd_flag);
+#endif
+
    if (wake_flags & WF_TTWU) {
    	record_wakee(p);
@@ -7673,7 +7787,11 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags)
    		new_cpu = prev_cpu;
    	}
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+		want_affine = !wake_wide(p) && cpumask_test_cpu(cpu, p->select_cpus);
+#else
    	want_affine = !wake_wide(p) && cpumask_test_cpu(cpu, p->cpus_ptr);
+#endif
    }
rcu_read_lock();
@@ -7684,7 +7802,13 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags)
    	 */
    	if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
    	    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+			new_cpu = cpu;
+			if (cpu != prev_cpu &&
+			    cpumask_test_cpu(prev_cpu, p->select_cpus))
+#else
    		if (cpu != prev_cpu)
+#endif
    			new_cpu = wake_affine(tmp, p, cpu, prev_cpu, sync);
sd = NULL; /* Prefer wake_affine over balance flags */
@@ -7711,6 +7835,10 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags)
    }
    rcu_read_unlock();
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	if (!cpumask_test_cpu(new_cpu, p->select_cpus))
+		new_cpu = idlest_cpu;
+#endif
    return new_cpu;
 }
@@ -9860,8 +9988,13 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
    	int local_group;
/* Skip over this group if it has no CPUs allowed */
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+		if (!cpumask_intersects(sched_group_span(group),
+					p->select_cpus))
+#else
    	if (!cpumask_intersects(sched_group_span(group),
    				p->cpus_ptr))
+#endif
    		continue;
/* Skip over this group if no cookie matched */
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index bfe53e835524..acc20b417dc8 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -2054,6 +2054,17 @@ static struct ctl_table kern_table[] = {
    	.extra1		= SYSCTL_ONE,
    	.extra2		= SYSCTL_INT_MAX,
    },
+#endif
+#ifdef CONFIG_QOS_SCHED_DYNAMIC_AFFINITY
+	{
+		.procname       = "sched_util_low_pct",
+		.data           = &sysctl_sched_util_low_pct,
+		.maxlen         = sizeof(sysctl_sched_util_low_pct),
+		.mode           = 0644,
+		.proc_handler   = proc_dointvec_minmax,
+		.extra1         = SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE_HUNDRED,
+	},
 #endif
    { }
 };
-- 
2.17.1


    

2024

2023

2022

2021

2020

2019

[PATCH openEuler-23.09 v2 3/7] sched: Adjust wakeup cpu range according CPU util dynamicly