diff --git a/kernel/events/core.c b/kernel/events/core.c
index d640a8b..afdd9e1 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -902,13 +902,23 @@
 	}
 }
 
-static void unclone_ctx(struct perf_event_context *ctx)
+/*
+ * This must be done under the ctx->lock, such as to serialize against
+ * context_equiv(), therefore we cannot call put_ctx() since that might end up
+ * calling scheduler related locks and ctx->lock nests inside those.
+ */
+static __must_check struct perf_event_context *
+unclone_ctx(struct perf_event_context *ctx)
 {
-	if (ctx->parent_ctx) {
-		put_ctx(ctx->parent_ctx);
+	struct perf_event_context *parent_ctx = ctx->parent_ctx;
+
+	lockdep_assert_held(&ctx->lock);
+
+	if (parent_ctx)
 		ctx->parent_ctx = NULL;
-	}
 	ctx->generation++;
+
+	return parent_ctx;
 }
 
 static u32 perf_event_pid(struct perf_event *event, struct task_struct *p)
@@ -2210,6 +2220,9 @@
 static int context_equiv(struct perf_event_context *ctx1,
 			 struct perf_event_context *ctx2)
 {
+	lockdep_assert_held(&ctx1->lock);
+	lockdep_assert_held(&ctx2->lock);
+
 	/* Pinning disables the swap optimization */
 	if (ctx1->pin_count || ctx2->pin_count)
 		return 0;
@@ -2943,6 +2956,7 @@
  */
 static void perf_event_enable_on_exec(struct perf_event_context *ctx)
 {
+	struct perf_event_context *clone_ctx = NULL;
 	struct perf_event *event;
 	unsigned long flags;
 	int enabled = 0;
@@ -2974,7 +2988,7 @@
 	 * Unclone this context if we enabled any event.
 	 */
 	if (enabled)
-		unclone_ctx(ctx);
+		clone_ctx = unclone_ctx(ctx);
 
 	raw_spin_unlock(&ctx->lock);
 
@@ -2984,6 +2998,9 @@
 	perf_event_context_sched_in(ctx, ctx->task);
 out:
 	local_irq_restore(flags);
+
+	if (clone_ctx)
+		put_ctx(clone_ctx);
 }
 
 void perf_event_exec(void)
@@ -3135,7 +3152,7 @@
 static struct perf_event_context *
 find_get_context(struct pmu *pmu, struct task_struct *task, int cpu)
 {
-	struct perf_event_context *ctx;
+	struct perf_event_context *ctx, *clone_ctx = NULL;
 	struct perf_cpu_context *cpuctx;
 	unsigned long flags;
 	int ctxn, err;
@@ -3169,9 +3186,12 @@
 retry:
 	ctx = perf_lock_task_context(task, ctxn, &flags);
 	if (ctx) {
-		unclone_ctx(ctx);
+		clone_ctx = unclone_ctx(ctx);
 		++ctx->pin_count;
 		raw_spin_unlock_irqrestore(&ctx->lock, flags);
+
+		if (clone_ctx)
+			put_ctx(clone_ctx);
 	} else {
 		ctx = alloc_perf_context(pmu, task);
 		err = -ENOMEM;
@@ -7523,7 +7543,7 @@
 static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
 {
 	struct perf_event *child_event, *next;
-	struct perf_event_context *child_ctx, *parent_ctx;
+	struct perf_event_context *child_ctx, *clone_ctx = NULL;
 	unsigned long flags;
 
 	if (likely(!child->perf_event_ctxp[ctxn])) {
@@ -7550,28 +7570,16 @@
 	child->perf_event_ctxp[ctxn] = NULL;
 
 	/*
-	 * In order to avoid freeing: child_ctx->parent_ctx->task
-	 * under perf_event_context::lock, grab another reference.
-	 */
-	parent_ctx = child_ctx->parent_ctx;
-	if (parent_ctx)
-		get_ctx(parent_ctx);
-
-	/*
 	 * If this context is a clone; unclone it so it can't get
 	 * swapped to another process while we're removing all
 	 * the events from it.
 	 */
-	unclone_ctx(child_ctx);
+	clone_ctx = unclone_ctx(child_ctx);
 	update_context_time(child_ctx);
 	raw_spin_unlock_irqrestore(&child_ctx->lock, flags);
 
-	/*
-	 * Now that we no longer hold perf_event_context::lock, drop
-	 * our extra child_ctx->parent_ctx reference.
-	 */
-	if (parent_ctx)
-		put_ctx(parent_ctx);
+	if (clone_ctx)
+		put_ctx(clone_ctx);
 
 	/*
 	 * Report the task dead after unscheduling the events so that we