diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 2b6b43cc0dd5121d8d4f6024f8ec67f862ff3328..529c41fa73c84c969fee8b8326bbd432dbfc872b 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -774,6 +774,9 @@ struct perf_cpu_context {
 #ifdef CONFIG_CGROUP_PERF
 	struct perf_cgroup		*cgrp;
 #endif
+
+	struct list_head		sched_cb_entry;
+	int				sched_cb_usage;
 };
 
 struct perf_output_handle {
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 57aff715039fe42c4b2ea812000b798ec7070cc1..803481cb6cbd57d5b75272011dc9a1497b695480 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2805,13 +2805,26 @@ static void perf_event_context_sched_out(struct task_struct *task, int ctxn,
 	}
 }
 
+static DEFINE_PER_CPU(struct list_head, sched_cb_list);
+
 void perf_sched_cb_dec(struct pmu *pmu)
 {
+	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+
 	this_cpu_dec(perf_sched_cb_usages);
+
+	if (!--cpuctx->sched_cb_usage)
+		list_del(&cpuctx->sched_cb_entry);
 }
 
+
 void perf_sched_cb_inc(struct pmu *pmu)
 {
+	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+
+	if (!cpuctx->sched_cb_usage++)
+		list_add(&cpuctx->sched_cb_entry, this_cpu_ptr(&sched_cb_list));
+
 	this_cpu_inc(perf_sched_cb_usages);
 }
 
@@ -2829,34 +2842,24 @@ static void perf_pmu_sched_task(struct task_struct *prev,
 {
 	struct perf_cpu_context *cpuctx;
 	struct pmu *pmu;
-	unsigned long flags;
 
 	if (prev == next)
 		return;
 
-	local_irq_save(flags);
-
-	rcu_read_lock();
-
-	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		if (pmu->sched_task) {
-			cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
-
-			perf_ctx_lock(cpuctx, cpuctx->task_ctx);
+	list_for_each_entry(cpuctx, this_cpu_ptr(&sched_cb_list), sched_cb_entry) {
+		pmu = cpuctx->unique_pmu; /* software PMUs will not have sched_task */
 
-			perf_pmu_disable(pmu);
+		if (WARN_ON_ONCE(!pmu->sched_task))
+			continue;
 
-			pmu->sched_task(cpuctx->task_ctx, sched_in);
+		perf_ctx_lock(cpuctx, cpuctx->task_ctx);
+		perf_pmu_disable(pmu);
 
-			perf_pmu_enable(pmu);
+		pmu->sched_task(cpuctx->task_ctx, sched_in);
 
-			perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
-		}
+		perf_pmu_enable(pmu);
+		perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
 	}
-
-	rcu_read_unlock();
-
-	local_irq_restore(flags);
 }
 
 static void perf_event_switch(struct task_struct *task,
@@ -10393,6 +10396,8 @@ static void __init perf_event_init_all_cpus(void)
 
 		INIT_LIST_HEAD(&per_cpu(pmu_sb_events.list, cpu));
 		raw_spin_lock_init(&per_cpu(pmu_sb_events.lock, cpu));
+
+		INIT_LIST_HEAD(&per_cpu(sched_cb_list, cpu));
 	}
 }