diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 22dd05d..f5b6137 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -185,18 +185,18 @@
 
 static void perf_record__open(struct perf_record *rec)
 {
-	struct perf_evsel *pos, *first;
+	struct perf_evsel *pos;
 	struct perf_evlist *evlist = rec->evlist;
 	struct perf_session *session = rec->session;
 	struct perf_record_opts *opts = &rec->opts;
 
-	first = list_entry(evlist->entries.next, struct perf_evsel, node);
-
 	perf_evlist__config_attrs(evlist, opts);
 
+	if (opts->group)
+		perf_evlist__group(evlist);
+
 	list_for_each_entry(pos, &evlist->entries, node) {
 		struct perf_event_attr *attr = &pos->attr;
-		struct xyarray *group_fd = NULL;
 		/*
 		 * Check if parse_single_tracepoint_event has already asked for
 		 * PERF_SAMPLE_TIME.
@@ -211,16 +211,13 @@
 		 */
 		bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;
 
-		if (opts->group && pos != first)
-			group_fd = first->fd;
 fallback_missing_features:
 		if (opts->exclude_guest_missing)
 			attr->exclude_guest = attr->exclude_host = 0;
 retry_sample_id:
 		attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1;
 try_again:
-		if (perf_evsel__open(pos, evlist->cpus, evlist->threads,
-				     opts->group, group_fd) < 0) {
+		if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) {
 			int err = errno;
 
 			if (err == EPERM || err == EACCES) {
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 861f0ae..23908a8 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -281,13 +281,9 @@
 				    struct perf_evsel *first)
 {
 	struct perf_event_attr *attr = &evsel->attr;
-	struct xyarray *group_fd = NULL;
 	bool exclude_guest_missing = false;
 	int ret;
 
-	if (group && evsel != first)
-		group_fd = first->fd;
-
 	if (scale)
 		attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
 				    PERF_FORMAT_TOTAL_TIME_RUNNING;
@@ -299,8 +295,7 @@
 		evsel->attr.exclude_guest = evsel->attr.exclude_host = 0;
 
 	if (perf_target__has_cpu(&target)) {
-		ret = perf_evsel__open_per_cpu(evsel, evsel_list->cpus,
-					       group, group_fd);
+		ret = perf_evsel__open_per_cpu(evsel, evsel_list->cpus);
 		if (ret)
 			goto check_ret;
 		return 0;
@@ -311,8 +306,7 @@
 		attr->enable_on_exec = 1;
 	}
 
-	ret = perf_evsel__open_per_thread(evsel, evsel_list->threads,
-					  group, group_fd);
+	ret = perf_evsel__open_per_thread(evsel, evsel_list->threads);
 	if (!ret)
 		return 0;
 	/* fall through */
@@ -483,6 +477,9 @@
 		close(child_ready_pipe[0]);
 	}
 
+	if (group)
+		perf_evlist__group(evsel_list);
+
 	first = list_entry(evsel_list->entries.next, struct perf_evsel, node);
 
 	list_for_each_entry(counter, &evsel_list->entries, node) {
diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c
index 1d592f5..9a479b6 100644
--- a/tools/perf/builtin-test.c
+++ b/tools/perf/builtin-test.c
@@ -294,7 +294,7 @@
 		goto out_thread_map_delete;
 	}
 
-	if (perf_evsel__open_per_thread(evsel, threads, false, NULL) < 0) {
+	if (perf_evsel__open_per_thread(evsel, threads) < 0) {
 		pr_debug("failed to open counter: %s, "
 			 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
 			 strerror(errno));
@@ -369,7 +369,7 @@
 		goto out_thread_map_delete;
 	}
 
-	if (perf_evsel__open(evsel, cpus, threads, false, NULL) < 0) {
+	if (perf_evsel__open(evsel, cpus, threads) < 0) {
 		pr_debug("failed to open counter: %s, "
 			 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
 			 strerror(errno));
@@ -533,7 +533,7 @@
 
 		perf_evlist__add(evlist, evsels[i]);
 
-		if (perf_evsel__open(evsels[i], cpus, threads, false, NULL) < 0) {
+		if (perf_evsel__open(evsels[i], cpus, threads) < 0) {
 			pr_debug("failed to open counter: %s, "
 				 "tweak /proc/sys/kernel/perf_event_paranoid?\n",
 				 strerror(errno));
@@ -737,7 +737,7 @@
 	 * Call sys_perf_event_open on all the fds on all the evsels,
 	 * grouping them if asked to.
 	 */
-	err = perf_evlist__open(evlist, opts.group);
+	err = perf_evlist__open(evlist);
 	if (err < 0) {
 		pr_debug("perf_evlist__open: %s\n", strerror(errno));
 		goto out_delete_evlist;
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index e45a1ba..392d219 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -886,17 +886,14 @@
 
 static void perf_top__start_counters(struct perf_top *top)
 {
-	struct perf_evsel *counter, *first;
+	struct perf_evsel *counter;
 	struct perf_evlist *evlist = top->evlist;
 
-	first = list_entry(evlist->entries.next, struct perf_evsel, node);
+	if (top->group)
+		perf_evlist__group(evlist);
 
 	list_for_each_entry(counter, &evlist->entries, node) {
 		struct perf_event_attr *attr = &counter->attr;
-		struct xyarray *group_fd = NULL;
-
-		if (top->group && counter != first)
-			group_fd = first->fd;
 
 		attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
 
@@ -927,8 +924,7 @@
 		attr->sample_id_all = top->sample_id_all_missing ? 0 : 1;
 try_again:
 		if (perf_evsel__open(counter, top->evlist->cpus,
-				     top->evlist->threads, top->group,
-				     group_fd) < 0) {
+				     top->evlist->threads) < 0) {
 			int err = errno;
 
 			if (err == EPERM || err == EACCES) {
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 9b38681..feffee3 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -108,6 +108,12 @@
 	evlist->nr_entries += nr_entries;
 }
 
+void perf_evlist__group(struct perf_evlist *evlist)
+{
+	if (evlist->nr_entries)
+		parse_events__set_leader(&evlist->entries);
+}
+
 int perf_evlist__add_default(struct perf_evlist *evlist)
 {
 	struct perf_event_attr attr = {
@@ -757,21 +763,13 @@
 	evlist->selected = evsel;
 }
 
-int perf_evlist__open(struct perf_evlist *evlist, bool group)
+int perf_evlist__open(struct perf_evlist *evlist)
 {
-	struct perf_evsel *evsel, *first;
+	struct perf_evsel *evsel;
 	int err, ncpus, nthreads;
 
-	first = list_entry(evlist->entries.next, struct perf_evsel, node);
-
 	list_for_each_entry(evsel, &evlist->entries, node) {
-		struct xyarray *group_fd = NULL;
-
-		if (group && evsel != first)
-			group_fd = first->fd;
-
-		err = perf_evsel__open(evsel, evlist->cpus, evlist->threads,
-				       group, group_fd);
+		err = perf_evsel__open(evsel, evlist->cpus, evlist->threads);
 		if (err < 0)
 			goto out_err;
 	}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 528c1ac..a19ccd7 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -85,7 +85,7 @@
 
 union perf_event *perf_evlist__mmap_read(struct perf_evlist *self, int idx);
 
-int perf_evlist__open(struct perf_evlist *evlist, bool group);
+int perf_evlist__open(struct perf_evlist *evlist);
 
 void perf_evlist__config_attrs(struct perf_evlist *evlist,
 			       struct perf_record_opts *opts);
@@ -132,4 +132,5 @@
 				   struct list_head *list,
 				   int nr_entries);
 
+void perf_evlist__group(struct perf_evlist *evlist);
 #endif /* __PERF_EVLIST_H */
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 9c54e8f..f5b68e7 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -21,7 +21,6 @@
 #include "perf_regs.h"
 
 #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
-#define GROUP_FD(group_fd, cpu) (*(int *)xyarray__entry(group_fd, cpu, 0))
 
 static int __perf_evsel__sample_size(u64 sample_type)
 {
@@ -493,6 +492,7 @@
 {
 	perf_evsel__exit(evsel);
 	close_cgroup(evsel->cgrp);
+	free(evsel->group_name);
 	free(evsel->name);
 	free(evsel);
 }
@@ -568,9 +568,28 @@
 	return 0;
 }
 
+static int get_group_fd(struct perf_evsel *evsel, int cpu, int thread)
+{
+	struct perf_evsel *leader = evsel->leader;
+	int fd;
+
+	if (!leader)
+		return -1;
+
+	/*
+	 * Leader must be already processed/open,
+	 * if not it's a bug.
+	 */
+	BUG_ON(!leader->fd);
+
+	fd = FD(leader, cpu, thread);
+	BUG_ON(fd == -1);
+
+	return fd;
+}
+
 static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
-			      struct thread_map *threads, bool group,
-			      struct xyarray *group_fds)
+			      struct thread_map *threads)
 {
 	int cpu, thread;
 	unsigned long flags = 0;
@@ -586,13 +605,15 @@
 	}
 
 	for (cpu = 0; cpu < cpus->nr; cpu++) {
-		int group_fd = group_fds ? GROUP_FD(group_fds, cpu) : -1;
 
 		for (thread = 0; thread < threads->nr; thread++) {
+			int group_fd;
 
 			if (!evsel->cgrp)
 				pid = threads->map[thread];
 
+			group_fd = get_group_fd(evsel, cpu, thread);
+
 			FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
 								     pid,
 								     cpus->map[cpu],
@@ -602,8 +623,9 @@
 				goto out_close;
 			}
 
-			if (group && group_fd == -1)
-				group_fd = FD(evsel, cpu, thread);
+			pr_debug("event cpu %d, thread %d, fd %d, group %d\n",
+				 cpu, pid, FD(evsel, cpu, thread),
+				 group_fd);
 		}
 	}
 
@@ -647,8 +669,7 @@
 };
 
 int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
-		     struct thread_map *threads, bool group,
-		     struct xyarray *group_fd)
+		     struct thread_map *threads)
 {
 	if (cpus == NULL) {
 		/* Work around old compiler warnings about strict aliasing */
@@ -658,23 +679,19 @@
 	if (threads == NULL)
 		threads = &empty_thread_map.map;
 
-	return __perf_evsel__open(evsel, cpus, threads, group, group_fd);
+	return __perf_evsel__open(evsel, cpus, threads);
 }
 
 int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
-			     struct cpu_map *cpus, bool group,
-			     struct xyarray *group_fd)
+			     struct cpu_map *cpus)
 {
-	return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group,
-				  group_fd);
+	return __perf_evsel__open(evsel, cpus, &empty_thread_map.map);
 }
 
 int perf_evsel__open_per_thread(struct perf_evsel *evsel,
-				struct thread_map *threads, bool group,
-				struct xyarray *group_fd)
+				struct thread_map *threads)
 {
-	return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group,
-				  group_fd);
+	return __perf_evsel__open(evsel, &empty_cpu_map.map, threads);
 }
 
 static int perf_event__parse_id_sample(const union perf_event *event, u64 type,
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 6a258c9..c411b42 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -70,6 +70,8 @@
 	bool 			supported;
 	/* parse modifier helper */
 	int			exclude_GH;
+	struct perf_evsel	*leader;
+	char			*group_name;
 };
 
 struct cpu_map;
@@ -109,14 +111,11 @@
 void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads);
 
 int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
-			     struct cpu_map *cpus, bool group,
-			     struct xyarray *group_fds);
+			     struct cpu_map *cpus);
 int perf_evsel__open_per_thread(struct perf_evsel *evsel,
-				struct thread_map *threads, bool group,
-				struct xyarray *group_fds);
+				struct thread_map *threads);
 int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
-		     struct thread_map *threads, bool group,
-		     struct xyarray *group_fds);
+		     struct thread_map *threads);
 void perf_evsel__close(struct perf_evsel *evsel, int ncpus, int nthreads);
 
 #define perf_evsel__match(evsel, t, c)		\
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 4364575..f6453cd 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -611,14 +611,32 @@
 			 pmu_event_name(head_config));
 }
 
-int parse_events__modifier_group(struct list_head *list __used,
-				 char *event_mod __used)
+struct perf_evsel *parse_events__set_leader(struct list_head *list)
 {
-	return 0;
+	struct perf_evsel *evsel, *leader;
+
+	leader = list_entry(list->next, struct perf_evsel, node);
+	leader->leader = NULL;
+
+	list_for_each_entry(evsel, list, node)
+		if (evsel != leader)
+			evsel->leader = leader;
+
+	return leader;
 }
 
-void parse_events__group(char *name __used, struct list_head *list __used)
+int parse_events__modifier_group(struct list_head *list,
+				 char *event_mod)
 {
+	return parse_events__modifier_event(list, event_mod, true);
+}
+
+void parse_events__group(char *name, struct list_head *list)
+{
+	struct perf_evsel *leader;
+
+	leader = parse_events__set_leader(list);
+	leader->group_name = name ? strdup(name) : NULL;
 }
 
 void parse_events_update_lists(struct list_head *list_event,
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index 75a6800..e1a184c 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -92,6 +92,7 @@
 				void *ptr, char *type);
 int parse_events_add_pmu(struct list_head **list, int *idx,
 			 char *pmu , struct list_head *head_config);
+struct perf_evsel *parse_events__set_leader(struct list_head *list);
 void parse_events__group(char *name, struct list_head *list);
 void parse_events_update_lists(struct list_head *list_event,
 			       struct list_head *list_all);
diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c
index 0688bfb..f5bba4b8 100644
--- a/tools/perf/util/python.c
+++ b/tools/perf/util/python.c
@@ -627,7 +627,7 @@
 	 * This will group just the fds for this single evsel, to group
 	 * multiple events, use evlist.open().
 	 */
-	if (perf_evsel__open(evsel, cpus, threads, group, NULL) < 0) {
+	if (perf_evsel__open(evsel, cpus, threads) < 0) {
 		PyErr_SetFromErrno(PyExc_OSError);
 		return NULL;
 	}
@@ -824,7 +824,10 @@
 	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|OOii", kwlist, &group))
 		return NULL;
 
-	if (perf_evlist__open(evlist, group) < 0) {
+	if (group)
+		perf_evlist__group(evlist);
+
+	if (perf_evlist__open(evlist) < 0) {
 		PyErr_SetFromErrno(PyExc_OSError);
 		return NULL;
 	}