Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 0ce90475 authored by Peter Zijlstra's avatar Peter Zijlstra Committed by Ingo Molnar
Browse files

sched/fair: Add some serialization to the sched_domain load-balance walk



Since the sched_domain walk is completely unserialized (!SD_SERIALIZE)
it is possible that multiple cpus in the group get elected to do the
next level. Avoid this by adding some serialization.

Signed-off-by: default avatarPeter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-vqh9ai6s0ewmeakjz80w4qz6@git.kernel.org


Signed-off-by: default avatarIngo Molnar <mingo@kernel.org>
parent c22402a2
Loading
Loading
Loading
Loading
+1 −0
Original line number Original line Diff line number Diff line
@@ -927,6 +927,7 @@ struct sched_group_power {
struct sched_group {
struct sched_group {
	struct sched_group *next;	/* Must be a circular list */
	struct sched_group *next;	/* Must be a circular list */
	atomic_t ref;
	atomic_t ref;
	int balance_cpu;


	unsigned int group_weight;
	unsigned int group_weight;
	struct sched_group_power *sgp;
	struct sched_group_power *sgp;
+2 −0
Original line number Original line Diff line number Diff line
@@ -6060,6 +6060,7 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu)


		sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span));
		sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span));
		atomic_inc(&sg->sgp->ref);
		atomic_inc(&sg->sgp->ref);
		sg->balance_cpu = -1;


		if (cpumask_test_cpu(cpu, sg_span))
		if (cpumask_test_cpu(cpu, sg_span))
			groups = sg;
			groups = sg;
@@ -6135,6 +6136,7 @@ build_sched_groups(struct sched_domain *sd, int cpu)


		cpumask_clear(sched_group_cpus(sg));
		cpumask_clear(sched_group_cpus(sg));
		sg->sgp->power = 0;
		sg->sgp->power = 0;
		sg->balance_cpu = -1;


		for_each_cpu(j, span) {
		for_each_cpu(j, span) {
			if (get_group(j, sdd, NULL) != group)
			if (get_group(j, sdd, NULL) != group)
+7 −2
Original line number Original line Diff line number Diff line
@@ -3828,7 +3828,8 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
	 */
	 */
	if (local_group) {
	if (local_group) {
		if (idle != CPU_NEWLY_IDLE) {
		if (idle != CPU_NEWLY_IDLE) {
			if (balance_cpu != this_cpu) {
			if (balance_cpu != this_cpu ||
			    cmpxchg(&group->balance_cpu, -1, balance_cpu) != -1) {
				*balance = 0;
				*balance = 0;
				return;
				return;
			}
			}
@@ -4929,7 +4930,7 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
	int balance = 1;
	int balance = 1;
	struct rq *rq = cpu_rq(cpu);
	struct rq *rq = cpu_rq(cpu);
	unsigned long interval;
	unsigned long interval;
	struct sched_domain *sd;
	struct sched_domain *sd, *last = NULL;
	/* Earliest time when we have to do rebalance again */
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
	int update_next_balance = 0;
@@ -4939,6 +4940,7 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)


	rcu_read_lock();
	rcu_read_lock();
	for_each_domain(cpu, sd) {
	for_each_domain(cpu, sd) {
		last = sd;
		if (!(sd->flags & SD_LOAD_BALANCE))
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;
			continue;


@@ -4983,6 +4985,9 @@ out:
		if (!balance)
		if (!balance)
			break;
			break;
	}
	}
	for (sd = last; sd; sd = sd->child)
		(void)cmpxchg(&sd->groups->balance_cpu, cpu, -1);

	rcu_read_unlock();
	rcu_read_unlock();


	/*
	/*