perf stat: Enable group read of counters

index 450b5fadf8cb..4dd0fcc06db9 100644
@@ -1261,15 +1261,42 @@ void perf_counts_values__scale(struct perf_counts_values *count,
 		*pscaled = scaled;
 }
 
+static int perf_evsel__read_size(struct perf_evsel *evsel)
+{
+	u64 read_format = evsel->attr.read_format;
+	int entry = sizeof(u64); /* value */
+	int size = 0;
+	int nr = 1;
+
+	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
+		size += sizeof(u64);
+
+	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
+		size += sizeof(u64);
+
+	if (read_format & PERF_FORMAT_ID)
+		entry += sizeof(u64);
+
+	if (read_format & PERF_FORMAT_GROUP) {
+		nr = evsel->nr_members;
+		size += sizeof(u64);
+	}
+
+	size += entry * nr;
+	return size;
+}
+
 int perf_evsel__read(struct perf_evsel *evsel, int cpu, int thread,
 		     struct perf_counts_values *count)
 {
+	size_t size = perf_evsel__read_size(evsel);
+
 	memset(count, 0, sizeof(*count));
 
 	if (FD(evsel, cpu, thread) < 0)
 		return -EINVAL;
 
-	if (readn(FD(evsel, cpu, thread), count, sizeof(*count)) <= 0)
+	if (readn(FD(evsel, cpu, thread), count->values, size) <= 0)
 		return -errno;
 
 	return 0;

index 4dd0fcc06db9..89aecf3a35c7 100644
@@ -1302,6 +1302,106 @@ int perf_evsel__read(struct perf_evsel *evsel, int cpu, int thread,
 	return 0;
 }
 
+static int
+perf_evsel__read_one(struct perf_evsel *evsel, int cpu, int thread)
+{
+	struct perf_counts_values *count = perf_counts(evsel->counts, cpu, thread);
+
+	return perf_evsel__read(evsel, cpu, thread, count);
+}
+
+static void
+perf_evsel__set_count(struct perf_evsel *counter, int cpu, int thread,
+		      u64 val, u64 ena, u64 run)
+{
+	struct perf_counts_values *count;
+
+	count = perf_counts(counter->counts, cpu, thread);
+
+	count->val    = val;
+	count->ena    = ena;
+	count->run    = run;
+}
+
+static int
+perf_evsel__process_group_data(struct perf_evsel *leader,
+			       int cpu, int thread, u64 *data)
+{
+	u64 read_format = leader->attr.read_format;
+	struct sample_read_value *v;
+	u64 nr, ena = 0, run = 0, i;
+
+	nr = *data++;
+
+	if (nr != (u64) leader->nr_members)
+		return -EINVAL;
+
+	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
+		ena = *data++;
+
+	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
+		run = *data++;
+
+	v = (struct sample_read_value *) data;
+
+	perf_evsel__set_count(leader, cpu, thread,
+			      v[0].value, ena, run);
+
+	for (i = 1; i < nr; i++) {
+		struct perf_evsel *counter;
+
+		counter = perf_evlist__id2evsel(leader->evlist, v[i].id);
+		if (!counter)
+			return -EINVAL;
+
+		perf_evsel__set_count(counter, cpu, thread,
+				      v[i].value, ena, run);
+	}
+
+	return 0;
+}
+
+static int
+perf_evsel__read_group(struct perf_evsel *leader, int cpu, int thread)
+{
+	struct perf_stat_evsel *ps = leader->priv;
+	u64 read_format = leader->attr.read_format;
+	int size = perf_evsel__read_size(leader);
+	u64 *data = ps->group_data;
+
+	if (!(read_format & PERF_FORMAT_ID))
+		return -EINVAL;
+
+	if (!perf_evsel__is_group_leader(leader))
+		return -EINVAL;
+
+	if (!data) {
+		data = zalloc(size);
+		if (!data)
+			return -ENOMEM;
+
+		ps->group_data = data;
+	}
+
+	if (FD(leader, cpu, thread) < 0)
+		return -EINVAL;
+
+	if (readn(FD(leader, cpu, thread), data, size) <= 0)
+		return -errno;
+
+	return perf_evsel__process_group_data(leader, cpu, thread, data);
+}
+
+int perf_evsel__read_counter(struct perf_evsel *evsel, int cpu, int thread)
+{
+	u64 read_format = evsel->attr.read_format;
+
+	if (read_format & PERF_FORMAT_GROUP)
+		return perf_evsel__read_group(evsel, cpu, thread);
+	else
+		return perf_evsel__read_one(evsel, cpu, thread);
+}
+
 int __perf_evsel__read_on_cpu(struct perf_evsel *evsel,
 			      int cpu, int thread, bool scale)
 {

index fb40ca3c6519..de03c18daaf0 100644
@@ -299,6 +299,8 @@ static inline bool perf_evsel__match2(struct perf_evsel *e1,
 int perf_evsel__read(struct perf_evsel *evsel, int cpu, int thread,
 		     struct perf_counts_values *count);
 
+int perf_evsel__read_counter(struct perf_evsel *evsel, int cpu, int thread);
+
 int __perf_evsel__read_on_cpu(struct perf_evsel *evsel,
 			      int cpu, int thread, bool scale);
 

index 53b9a994a3dc..35e9848734d6 100644
@@ -128,6 +128,10 @@ static int perf_evsel__alloc_stat_priv(struct perf_evsel *evsel)
 
 static void perf_evsel__free_stat_priv(struct perf_evsel *evsel)
 {
+	struct perf_stat_evsel *ps = evsel->priv;
+
+	if (ps)
+		free(ps->group_data);
 	zfree(&evsel->priv);
 }
 

index 7522bf10b03e..eacaf958e19d 100644
@@ -28,8 +28,9 @@ enum perf_stat_evsel_id {
 };
 
 struct perf_stat_evsel {
-	struct stats		res_stats[3];
-	enum perf_stat_evsel_id	id;
+	struct stats		 res_stats[3];
+	enum perf_stat_evsel_id	 id;
+	u64			*group_data;
 };
 
 enum aggr_mode {

index 48ac53b199fc..866da7aa54bf 100644
@@ -213,10 +213,20 @@ static void perf_stat__reset_stats(void)
 static int create_perf_stat_counter(struct perf_evsel *evsel)
 {
 	struct perf_event_attr *attr = &evsel->attr;
+	struct perf_evsel *leader = evsel->leader;
 
-	if (stat_config.scale)
+	if (stat_config.scale) {
 		attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
 				    PERF_FORMAT_TOTAL_TIME_RUNNING;
+	}
+
+	/*
+	 * The event is part of non trivial group, let's enable
+	 * the group read (for leader) and ID retrieval for all
+	 * members.
+	 */
+	if (leader->nr_members > 1)
+		attr->read_format |= PERF_FORMAT_ID|PERF_FORMAT_GROUP;
 
 	attr->inherit = !no_inherit;
 

@@ -333,13 +343,21 @@ static int read_counter(struct perf_evsel *counter)
 			struct perf_counts_values *count;
 
 			count = perf_counts(counter->counts, cpu, thread);
-			if (perf_evsel__read(counter, cpu, thread, count)) {
+
+			/*
+			 * The leader's group read loads data into its group members
+			 * (via perf_evsel__read_counter) and sets threir count->loaded.
+			 */
+			if (!count->loaded &&
+			    perf_evsel__read_counter(counter, cpu, thread)) {
 				counter->counts->scaled = -1;
 				perf_counts(counter->counts, cpu, thread)->ena = 0;
 				perf_counts(counter->counts, cpu, thread)->run = 0;
 				return -1;
 			}
 
+			count->loaded = false;
+
 			if (STAT_RECORD) {
 				if (perf_evsel__write_stat_event(counter, cpu, thread, count)) {
 					pr_err("failed to write stat event\n");

@@ -559,6 +577,11 @@ static int store_counter_ids(struct perf_evsel *counter)
 	return __store_counter_ids(counter, cpus, threads);
 }
 
+static bool perf_evsel__should_store_id(struct perf_evsel *counter)
+{
+	return STAT_RECORD || counter->attr.read_format & PERF_FORMAT_ID;
+}
+
 static int __run_perf_stat(int argc, const char **argv)
 {
 	int interval = stat_config.interval;

@@ -631,7 +654,8 @@ static int __run_perf_stat(int argc, const char **argv)
 		if (l > unit_width)
 			unit_width = l;
 
-		if (STAT_RECORD && store_counter_ids(counter))
+		if (perf_evsel__should_store_id(counter) &&
+		    store_counter_ids(counter))
 			return -1;
 	}
 

index 34d8baaf558a..cb45a6aecf9d 100644
@@ -12,6 +12,7 @@ struct perf_counts_values {
 		};
 		u64 values[3];
 	};
+	bool	loaded;
 };
 
 struct perf_counts {

index 89aecf3a35c7..3735c9e0080d 100644
@@ -49,6 +49,7 @@ static struct {
 	bool clockid_wrong;
 	bool lbr_flags;
 	bool write_backward;
+	bool group_read;
 } perf_missing_features;
 
 static clockid_t clockid;

@@ -1321,6 +1322,7 @@ perf_evsel__set_count(struct perf_evsel *counter, int cpu, int thread,
 	count->val    = val;
 	count->ena    = ena;
 	count->run    = run;
+	count->loaded = true;
 }
 
 static int

@@ -1677,6 +1679,8 @@ int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
 	if (perf_missing_features.lbr_flags)
 		evsel->attr.branch_sample_type &= ~(PERF_SAMPLE_BRANCH_NO_FLAGS |
 				     PERF_SAMPLE_BRANCH_NO_CYCLES);
+	if (perf_missing_features.group_read && evsel->attr.inherit)
+		evsel->attr.read_format &= ~(PERF_FORMAT_GROUP|PERF_FORMAT_ID);
 retry_sample_id:
 	if (perf_missing_features.sample_id_all)
 		evsel->attr.sample_id_all = 0;

@@ -1832,6 +1836,12 @@ int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
 		perf_missing_features.lbr_flags = true;
 		pr_debug2("switching off branch sample type no (cycles/flags)\n");
 		goto fallback_missing_features;
+	} else if (!perf_missing_features.group_read &&
+		    evsel->attr.inherit &&
+		   (evsel->attr.read_format & PERF_FORMAT_GROUP)) {
+		perf_missing_features.group_read = true;
+		pr_debug2("switching off group read\n");
+		goto fallback_missing_features;
 	}
 out_close:
 	do {