diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index e8ad3c590a23..61448c19a132 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -246,6 +246,7 @@ struct perf_event;
 #define PERF_PMU_CAP_ITRACE			0x20
 #define PERF_PMU_CAP_HETEROGENEOUS_CPUS		0x40
 #define PERF_PMU_CAP_NO_EXCLUDE			0x80
+#define PERF_PMU_CAP_AUX_OUTPUT			0x100
 
 /**
  * struct pmu - generic performance monitoring unit
@@ -446,6 +447,16 @@ struct pmu {
 	void (*addr_filters_sync)	(struct perf_event *event);
 					/* optional */
 
+	/*
+	 * Check if event can be used for aux_output purposes for
+	 * events of this PMU.
+	 *
+	 * Runs from perf_event_open(). Should return 0 for "no match"
+	 * or non-zero for "match".
+	 */
+	int (*aux_output_match)		(struct perf_event *event);
+					/* optional */
+
 	/*
 	 * Filter events for PMU-specific reasons.
 	 */
@@ -681,6 +692,9 @@ struct perf_event {
 	struct perf_addr_filter_range	*addr_filter_ranges;
 	unsigned long			addr_filters_gen;
 
+	/* for aux_output events */
+	struct perf_event		*aux_event;
+
 	void (*destroy)(struct perf_event *);
 	struct rcu_head			rcu_head;
 
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 7198ddd0c6b1..bb7b271397a6 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -374,7 +374,8 @@ struct perf_event_attr {
 				namespaces     :  1, /* include namespaces data */
 				ksymbol        :  1, /* include ksymbol events */
 				bpf_event      :  1, /* include bpf events */
-				__reserved_1   : 33;
+				aux_output     :  1, /* generate AUX records instead of events */
+				__reserved_1   : 32;
 
 	union {
 		__u32		wakeup_events;	  /* wakeup every n events */
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 0463c1151bae..2aad959e6def 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -1887,6 +1887,89 @@ list_del_event(struct perf_event *event, struct perf_event_context *ctx)
 	ctx->generation++;
 }
 
+static int
+perf_aux_output_match(struct perf_event *event, struct perf_event *aux_event)
+{
+	if (!has_aux(aux_event))
+		return 0;
+
+	if (!event->pmu->aux_output_match)
+		return 0;
+
+	return event->pmu->aux_output_match(aux_event);
+}
+
+static void put_event(struct perf_event *event);
+static void event_sched_out(struct perf_event *event,
+			    struct perf_cpu_context *cpuctx,
+			    struct perf_event_context *ctx);
+
+static void perf_put_aux_event(struct perf_event *event)
+{
+	struct perf_event_context *ctx = event->ctx;
+	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
+	struct perf_event *iter;
+
+	/*
+	 * If event uses aux_event tear down the link
+	 */
+	if (event->aux_event) {
+		iter = event->aux_event;
+		event->aux_event = NULL;
+		put_event(iter);
+		return;
+	}
+
+	/*
+	 * If the event is an aux_event, tear down all links to
+	 * it from other events.
+	 */
+	for_each_sibling_event(iter, event->group_leader) {
+		if (iter->aux_event != event)
+			continue;
+
+		iter->aux_event = NULL;
+		put_event(event);
+
+		/*
+		 * If it's ACTIVE, schedule it out and put it into ERROR
+		 * state so that we don't try to schedule it again. Note
+		 * that perf_event_enable() will clear the ERROR status.
+		 */
+		event_sched_out(iter, cpuctx, ctx);
+		perf_event_set_state(event, PERF_EVENT_STATE_ERROR);
+	}
+}
+
+static int perf_get_aux_event(struct perf_event *event,
+			      struct perf_event *group_leader)
+{
+	/*
+	 * Our group leader must be an aux event if we want to be
+	 * an aux_output. This way, the aux event will precede its
+	 * aux_output events in the group, and therefore will always
+	 * schedule first.
+	 */
+	if (!group_leader)
+		return 0;
+
+	if (!perf_aux_output_match(event, group_leader))
+		return 0;
+
+	if (!atomic_long_inc_not_zero(&group_leader->refcount))
+		return 0;
+
+	/*
+	 * Link aux_outputs to their aux event; this is undone in
+	 * perf_group_detach() by perf_put_aux_event(). When the
+	 * group in torn down, the aux_output events loose their
+	 * link to the aux_event and can't schedule any more.
+	 */
+	event->aux_event = group_leader;
+
+	return 1;
+}
+
 static void perf_group_detach(struct perf_event *event)
 {
 	struct perf_event *sibling, *tmp;
@@ -1902,6 +1985,8 @@ static void perf_group_detach(struct perf_event *event)
 
 	event->attach_state &= ~PERF_ATTACH_GROUP;
 
+	perf_put_aux_event(event);
+
 	/*
 	 * If this is a sibling, remove it from its group.
 	 */
@@ -10426,6 +10511,12 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
 		goto err_ns;
 	}
 
+	if (event->attr.aux_output &&
+	    !(pmu->capabilities & PERF_PMU_CAP_AUX_OUTPUT)) {
+		err = -EOPNOTSUPP;
+		goto err_pmu;
+	}
+
 	err = exclusive_event_init(event);
 	if (err)
 		goto err_pmu;
@@ -11082,6 +11173,8 @@ SYSCALL_DEFINE5(perf_event_open,
 		}
 	}
 
+	if (event->attr.aux_output && !perf_get_aux_event(event, group_leader))
+		goto err_locked;
 
 	/*
 	 * Must be under the same ctx::mutex as perf_install_in_context(),