[RFC PATCH 2/2] sched/fair: Export a param to control the traverse len when select idle cpu.

["Kenan.Liu" <Kenan.Liu@linux.alibaba.com>]

From: "Kenan.Liu" <Kenan.Liu@linux.alibaba.com>

The variable 'nr' decides the length of traverse when we try to find an
idle cpu in function select_idle_cpu(). A fixed value such as 4 may not
perform well in all scenes and may lead to un-acceptable overhead. Export
two sysctl parameters to enable adjustments.

Signed-off-by: Kenan.Liu <Kenan.Liu@linux.alibaba.com>
Signed-off-by: Ben Luo <luoben@linux.alibaba.com>
---
 kernel/sched/fair.c | 65 +++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 63 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index ad7c93f..e10de3b 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -125,6 +125,9 @@
 static unsigned int normalized_sysctl_sched_wakeup_granularity	= 1000000UL;
 
 const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
+unsigned int __read_mostly sysctl_sched_idle_search_nr_default = 4;
+unsigned int __read_mostly sysctl_sched_idle_search_nr_threshold = 4;
+
 static bool smt_neighbour_topo;
 static bool core_smt_topo_detect;
 static unsigned int smt_nr_cpu = 2;
@@ -207,6 +210,50 @@ int __weak arch_asym_cpu_priority(int cpu)
 #endif
 
 #ifdef CONFIG_SYSCTL
+static int sched_set_idle_search_nr_default(struct ctl_table *table, int write, void *buffer,
+		size_t *lenp, loff_t *ppos)
+{
+	static DEFINE_MUTEX(mutex);
+	unsigned int old_nr;
+	int ret;
+
+	mutex_lock(&mutex);
+	old_nr = sysctl_sched_idle_search_nr_default;
+	ret = proc_douintvec(table, write, buffer, lenp, ppos);
+	if (!ret && write) {
+		if (sysctl_sched_idle_search_nr_default == 0) {
+			sysctl_sched_idle_search_nr_default = old_nr;
+			mutex_unlock(&mutex);
+			return -EINVAL;
+		}
+	}
+
+	mutex_unlock(&mutex);
+	return ret;
+}
+
+static int sched_set_idle_search_nr_threshold(struct ctl_table *table, int write, void *buffer,
+		size_t *lenp, loff_t *ppos)
+{
+	static DEFINE_MUTEX(mutex);
+	unsigned int old_threshold;
+	int ret;
+
+	mutex_lock(&mutex);
+	old_threshold = sysctl_sched_idle_search_nr_threshold;
+	ret = proc_douintvec(table, write, buffer, lenp, ppos);
+	if (!ret && write) {
+		if (sysctl_sched_idle_search_nr_threshold == 0) {
+			sysctl_sched_idle_search_nr_threshold = old_threshold;
+			mutex_unlock(&mutex);
+			return -EINVAL;
+		}
+	}
+
+	mutex_unlock(&mutex);
+	return ret;
+}
+
 static struct ctl_table sched_fair_sysctls[] = {
 	{
 		.procname       = "sched_child_runs_first",
@@ -235,6 +282,20 @@ int __weak arch_asym_cpu_priority(int cpu)
 		.extra1		= SYSCTL_ZERO,
 	},
 #endif /* CONFIG_NUMA_BALANCING */
+	{
+		.procname	= "sched_cfs_idle_search_nr_default",
+		.data		= &sysctl_sched_idle_search_nr_default,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= sched_set_idle_search_nr_default,
+	},
+	{
+		.procname	= "sched_cfs_idle_search_nr_threshold",
+		.data		= &sysctl_sched_idle_search_nr_threshold,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= sched_set_idle_search_nr_threshold,
+	},
 	{}
 };
 
@@ -7027,10 +7088,10 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, bool
 		avg_cost = this_sd->avg_scan_cost + 1;
 
 		span_avg = sd->span_weight * avg_idle;
-		if (span_avg > 4*avg_cost)
+		if (span_avg > sysctl_sched_idle_search_nr_threshold * avg_cost)
 			nr = div_u64(span_avg, avg_cost);
 		else
-			nr = 4;
+			nr = sysctl_sched_idle_search_nr_default;
 
 		time = cpu_clock(this);
 	}
-- 
1.8.3.1