[PATCH v4 0/4] perf: Support a new 'percore' event qualifier

[PATCH v4 0/4] perf: Support a new 'percore' event qualifier

[Jin Yao]

The 'percore' event qualifier which sums up the event counts for both
hardware threads in a core. For example,

perf stat -e cpu/event=0,umask=0x3,percore=1/,cpu/event=0,umask=0x3/

In this example, we count the event 'ref-cycles' per-core and per-CPU in
one perf stat command-line.

We can already support per-core counting with --per-core, but it's
often useful to do this together with other metrics that are collected
per CPU (per hardware thread). So this patch series supports this
per-core counting on a event level.

 v4:
 ---
 Add percore qualifier to documantation.
 Rebase to latest perf/core branch

 v3:
 ---
 Simplify the patch: "perf: Add a 'percore' event qualifier"
 Other patches don't have changes.

 v2:
 ---
 1. Change 'coresum' to 'percore'.
 2. Move the aggregate counts printing to a seperate patch.

Jin Yao (4):
  perf: Add a 'percore' event qualifier
  perf stat: Factor out aggregate counts printing
  perf stat: Support 'percore' event qualifier
  perf test: Add a simple test for term 'percore'

 tools/perf/Documentation/perf-list.txt |  12 ++++
 tools/perf/Documentation/perf-stat.txt |   4 ++
 tools/perf/builtin-stat.c              |  21 +++++++
 tools/perf/tests/parse-events.c        |  10 ++-
 tools/perf/util/evsel.c                |   2 +
 tools/perf/util/evsel.h                |   3 +
 tools/perf/util/parse-events.c         |  27 +++++++++
 tools/perf/util/parse-events.h         |   1 +
 tools/perf/util/parse-events.l         |   1 +
 tools/perf/util/stat-display.c         | 107 ++++++++++++++++++++++++---------
 tools/perf/util/stat.c                 |   8 ++-
 11 files changed, 163 insertions(+), 33 deletions(-)

-- 
2.7.4

[PATCH v4 1/4] perf: Add a 'percore' event qualifier

[Jin Yao]

Add a 'percore' event qualifier, like cpu/event=0,umask=0x3,percore=1/,
that sums up the event counts for both hardware threads in a core.

We can already do this with --per-core, but it's often useful to do
this together with other metrics that are collected per hardware thread.
So we need to support this per-core counting on a event level.

This can be implemented in only the user tool, no kernel support needed.

 v4:
 ---
 1. Add Arnaldo's patch which updates the documentation for
    this new qualifier.
 2. Rebase to latest perf/core branch

 v3:
 ---
 Simplify the code according to Jiri's comments.
 Before:
   "return term->val.percore ? true : false;"
 Now:
   "return term->val.percore;"

 v2:
 ---
 Change the qualifier name from 'coresum' to 'percore' according to
 comments from Jiri and Andi.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
---
 tools/perf/Documentation/perf-list.txt | 12 ++++++++++++
 tools/perf/util/evsel.c                |  2 ++
 tools/perf/util/evsel.h                |  3 +++
 tools/perf/util/parse-events.c         | 27 +++++++++++++++++++++++++++
 tools/perf/util/parse-events.h         |  1 +
 tools/perf/util/parse-events.l         |  1 +
 6 files changed, 46 insertions(+)

diff --git a/tools/perf/Documentation/perf-list.txt b/tools/perf/Documentation/perf-list.txt
index 138fb6e..18ed1b0 100644
--- a/tools/perf/Documentation/perf-list.txt
+++ b/tools/perf/Documentation/perf-list.txt
@@ -199,6 +199,18 @@ also be supplied. For example:
 
   perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
 
+EVENT QUALIFIERS:
+
+It is also possible to add extra qualifiers to an event:
+
+percore:
+
+Sums up the event counts for all hardware threads in a core, e.g.:
+
+
+  perf stat -e cpu/event=0,umask=0x3,percore=1/
+
+
 EVENT GROUPS
 ------------
 
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 84cfb9f..99b6143 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -813,6 +813,8 @@ static void apply_config_terms(struct perf_evsel *evsel,
 			break;
 		case PERF_EVSEL__CONFIG_TERM_DRV_CFG:
 			break;
+		case PERF_EVSEL__CONFIG_TERM_PERCORE:
+			break;
 		default:
 			break;
 		}
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 6d190cb..cad54e8 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -50,6 +50,7 @@ enum term_type {
 	PERF_EVSEL__CONFIG_TERM_OVERWRITE,
 	PERF_EVSEL__CONFIG_TERM_DRV_CFG,
 	PERF_EVSEL__CONFIG_TERM_BRANCH,
+	PERF_EVSEL__CONFIG_TERM_PERCORE,
 };
 
 struct perf_evsel_config_term {
@@ -67,6 +68,7 @@ struct perf_evsel_config_term {
 		bool	overwrite;
 		char	*branch;
 		unsigned long max_events;
+		bool	percore;
 	} val;
 	bool weak;
 };
@@ -158,6 +160,7 @@ struct perf_evsel {
 	struct perf_evsel	**metric_events;
 	bool			collect_stat;
 	bool			weak_group;
+	bool			percore;
 	const char		*pmu_name;
 	struct {
 		perf_evsel__sb_cb_t	*cb;
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 4432bfe..cf0b9b8 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -950,6 +950,7 @@ static const char *config_term_names[__PARSE_EVENTS__TERM_TYPE_NR] = {
 	[PARSE_EVENTS__TERM_TYPE_OVERWRITE]		= "overwrite",
 	[PARSE_EVENTS__TERM_TYPE_NOOVERWRITE]		= "no-overwrite",
 	[PARSE_EVENTS__TERM_TYPE_DRV_CFG]		= "driver-config",
+	[PARSE_EVENTS__TERM_TYPE_PERCORE]		= "percore",
 };
 
 static bool config_term_shrinked;
@@ -970,6 +971,7 @@ config_term_avail(int term_type, struct parse_events_error *err)
 	case PARSE_EVENTS__TERM_TYPE_CONFIG2:
 	case PARSE_EVENTS__TERM_TYPE_NAME:
 	case PARSE_EVENTS__TERM_TYPE_SAMPLE_PERIOD:
+	case PARSE_EVENTS__TERM_TYPE_PERCORE:
 		return true;
 	default:
 		if (!err)
@@ -1061,6 +1063,14 @@ do {									   \
 	case PARSE_EVENTS__TERM_TYPE_MAX_EVENTS:
 		CHECK_TYPE_VAL(NUM);
 		break;
+	case PARSE_EVENTS__TERM_TYPE_PERCORE:
+		CHECK_TYPE_VAL(NUM);
+		if ((unsigned int)term->val.num > 1) {
+			err->str = strdup("expected 0 or 1");
+			err->idx = term->err_val;
+			return -EINVAL;
+		}
+		break;
 	default:
 		err->str = strdup("unknown term");
 		err->idx = term->err_term;
@@ -1199,6 +1209,10 @@ do {								\
 		case PARSE_EVENTS__TERM_TYPE_DRV_CFG:
 			ADD_CONFIG_TERM(DRV_CFG, drv_cfg, term->val.str);
 			break;
+		case PARSE_EVENTS__TERM_TYPE_PERCORE:
+			ADD_CONFIG_TERM(PERCORE, percore,
+					term->val.num ? true : false);
+			break;
 		default:
 			break;
 		}
@@ -1260,6 +1274,18 @@ int parse_events_add_tool(struct parse_events_state *parse_state,
 	return add_event_tool(list, &parse_state->idx, tool_event);
 }
 
+static bool config_term_percore(struct list_head *config_terms)
+{
+	struct perf_evsel_config_term *term;
+
+	list_for_each_entry(term, config_terms, list) {
+		if (term->type == PERF_EVSEL__CONFIG_TERM_PERCORE)
+			return term->val.percore;
+	}
+
+	return false;
+}
+
 int parse_events_add_pmu(struct parse_events_state *parse_state,
 			 struct list_head *list, char *name,
 			 struct list_head *head_config,
@@ -1333,6 +1359,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
 		evsel->metric_name = info.metric_name;
 		evsel->pmu_name = name;
 		evsel->use_uncore_alias = use_uncore_alias;
+		evsel->percore = config_term_percore(&evsel->config_terms);
 	}
 
 	return evsel ? 0 : -ENOMEM;
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index a052cd6..f7139e1 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -75,6 +75,7 @@ enum {
 	PARSE_EVENTS__TERM_TYPE_NOOVERWRITE,
 	PARSE_EVENTS__TERM_TYPE_OVERWRITE,
 	PARSE_EVENTS__TERM_TYPE_DRV_CFG,
+	PARSE_EVENTS__TERM_TYPE_PERCORE,
 	__PARSE_EVENTS__TERM_TYPE_NR,
 };
 
diff --git a/tools/perf/util/parse-events.l b/tools/perf/util/parse-events.l
index c54bfe8..ca60988 100644
--- a/tools/perf/util/parse-events.l
+++ b/tools/perf/util/parse-events.l
@@ -283,6 +283,7 @@ inherit			{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_INHERIT); }
 no-inherit		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_NOINHERIT); }
 overwrite		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_OVERWRITE); }
 no-overwrite		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_NOOVERWRITE); }
+percore			{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_PERCORE); }
 ,			{ return ','; }
 "/"			{ BEGIN(INITIAL); return '/'; }
 {name_minus}		{ return str(yyscanner, PE_NAME); }
-- 
2.7.4

[PATCH v4 2/4] perf stat: Factor out aggregate counts printing

[Jin Yao]

Move the aggregate counts printing to a new function
print_counter_aggrdata, which will be used in following
patches.

 v4:
 ---
 Rebase to latest perf/core branch.

 v3:
 ---
 No change

 v2:
 ---
 Create this patch according to Jiri's comments.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
---
 tools/perf/util/stat-display.c | 64 +++++++++++++++++++++++++-----------------
 1 file changed, 39 insertions(+), 25 deletions(-)

diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index 3324f23..f5b4ee7 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -594,6 +594,41 @@ static void aggr_cb(struct perf_stat_config *config,
 	}
 }
 
+static void print_counter_aggrdata(struct perf_stat_config *config,
+				   struct perf_evsel *counter, int s,
+				   char *prefix, bool metric_only,
+				   bool *first)
+{
+	struct aggr_data ad;
+	FILE *output = config->output;
+	u64 ena, run, val;
+	int id, nr;
+	double uval;
+
+	ad.id = id = config->aggr_map->map[s];
+	ad.val = ad.ena = ad.run = 0;
+	ad.nr = 0;
+	if (!collect_data(config, counter, aggr_cb, &ad))
+		return;
+
+	nr = ad.nr;
+	ena = ad.ena;
+	run = ad.run;
+	val = ad.val;
+	if (*first && metric_only) {
+		*first = false;
+		aggr_printout(config, counter, id, nr);
+	}
+	if (prefix && !metric_only)
+		fprintf(output, "%s", prefix);
+
+	uval = val * counter->scale;
+	printout(config, id, nr, counter, uval, prefix,
+		 run, ena, 1.0, &rt_stat);
+	if (!metric_only)
+		fputc('\n', output);
+}
+
 static void print_aggr(struct perf_stat_config *config,
 		       struct perf_evlist *evlist,
 		       char *prefix)
@@ -601,9 +636,7 @@ static void print_aggr(struct perf_stat_config *config,
 	bool metric_only = config->metric_only;
 	FILE *output = config->output;
 	struct perf_evsel *counter;
-	int s, id, nr;
-	double uval;
-	u64 ena, run, val;
+	int s;
 	bool first;
 
 	if (!(config->aggr_map || config->aggr_get_id))
@@ -616,33 +649,14 @@ static void print_aggr(struct perf_stat_config *config,
 	 * Without each counter has its own line.
 	 */
 	for (s = 0; s < config->aggr_map->nr; s++) {
-		struct aggr_data ad;
 		if (prefix && metric_only)
 			fprintf(output, "%s", prefix);
 
-		ad.id = id = config->aggr_map->map[s];
 		first = true;
 		evlist__for_each_entry(evlist, counter) {
-			ad.val = ad.ena = ad.run = 0;
-			ad.nr = 0;
-			if (!collect_data(config, counter, aggr_cb, &ad))
-				continue;
-			nr = ad.nr;
-			ena = ad.ena;
-			run = ad.run;
-			val = ad.val;
-			if (first && metric_only) {
-				first = false;
-				aggr_printout(config, counter, id, nr);
-			}
-			if (prefix && !metric_only)
-				fprintf(output, "%s", prefix);
-
-			uval = val * counter->scale;
-			printout(config, id, nr, counter, uval, prefix,
-				 run, ena, 1.0, &rt_stat);
-			if (!metric_only)
-				fputc('\n', output);
+			print_counter_aggrdata(config, counter, s,
+					       prefix, metric_only,
+					       &first);
 		}
 		if (metric_only)
 			fputc('\n', output);
-- 
2.7.4

[PATCH v4 3/4] perf stat: Support 'percore' event qualifier

[Jin Yao]

With this patch, we can use the 'percore' event qualifier in perf-stat.

root@skl:/tmp# perf stat -e cpu/event=0,umask=0x3,percore=1/,cpu/event=0,umask=0x3/ -a -A -I1000
     1.000773050 S0-C0             98,352,832      cpu/event=0,umask=0x3,percore=1/                                     (50.01%)
     1.000773050 S0-C1            103,763,057      cpu/event=0,umask=0x3,percore=1/                                     (50.02%)
     1.000773050 S0-C2            196,776,995      cpu/event=0,umask=0x3,percore=1/                                     (50.02%)
     1.000773050 S0-C3            176,493,779      cpu/event=0,umask=0x3,percore=1/                                     (50.02%)
     1.000773050 CPU0              47,699,641      cpu/event=0,umask=0x3/                                        (50.02%)
     1.000773050 CPU1              49,052,451      cpu/event=0,umask=0x3/                                        (49.98%)
     1.000773050 CPU2             102,771,422      cpu/event=0,umask=0x3/                                        (49.98%)
     1.000773050 CPU3             100,784,662      cpu/event=0,umask=0x3/                                        (49.98%)
     1.000773050 CPU4              43,171,342      cpu/event=0,umask=0x3/                                        (49.98%)
     1.000773050 CPU5              54,152,158      cpu/event=0,umask=0x3/                                        (49.98%)
     1.000773050 CPU6              93,618,410      cpu/event=0,umask=0x3/                                        (49.98%)
     1.000773050 CPU7              74,477,589      cpu/event=0,umask=0x3/                                        (49.99%)

In this example, we count the event 'ref-cycles' per-core and per-CPU in
one perf stat command-line. From the output, we can see:

  S0-C0 = CPU0 + CPU4
  S0-C1 = CPU1 + CPU5
  S0-C2 = CPU2 + CPU6
  S0-C3 = CPU3 + CPU7

So the result is expected (tiny difference is ignored).

Note that, the 'percore' event qualifier needs to use with option '-A'.

 v4:
 ---
 Rebase to latest perf/core branch.

 v3:
 ---
 No change

 v2:
 ---
 Change 'coresum' to 'percore'.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
---
 tools/perf/Documentation/perf-stat.txt |  4 ++++
 tools/perf/builtin-stat.c              | 21 +++++++++++++++++
 tools/perf/util/stat-display.c         | 43 ++++++++++++++++++++++++++++++----
 tools/perf/util/stat.c                 |  8 ++++---
 4 files changed, 69 insertions(+), 7 deletions(-)

diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index 39c05f8..1e312c2 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -43,6 +43,10 @@ report::
 	  param1 and param2 are defined as formats for the PMU in
 	  /sys/bus/event_source/devices/<pmu>/format/*
 
+	  'percore' is a event qualifier that sums up the event counts for both
+	  hardware threads in a core. For example:
+	  perf stat -A -a -e cpu/event,percore=1/,otherevent ...
+
 	- a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
 	  where M, N, K are numbers (in decimal, hex, octal format).
 	  Acceptable values for each of 'config', 'config1' and 'config2'
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 7f9c4b7..4a79fa9 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -847,6 +847,18 @@ static int perf_stat__get_core_cached(struct perf_stat_config *config,
 	return perf_stat__get_aggr(config, perf_stat__get_core, map, idx);
 }
 
+static bool term_percore_set(void)
+{
+	struct perf_evsel *counter;
+
+	evlist__for_each_entry(evsel_list, counter) {
+		if (counter->percore)
+			return true;
+	}
+
+	return false;
+}
+
 static int perf_stat_init_aggr_mode(void)
 {
 	int nr;
@@ -867,6 +879,15 @@ static int perf_stat_init_aggr_mode(void)
 		stat_config.aggr_get_id = perf_stat__get_core_cached;
 		break;
 	case AGGR_NONE:
+		if (term_percore_set()) {
+			if (cpu_map__build_core_map(evsel_list->cpus,
+						    &stat_config.aggr_map)) {
+				perror("cannot build core map");
+				return -1;
+			}
+			stat_config.aggr_get_id = perf_stat__get_core_cached;
+		}
+		break;
 	case AGGR_GLOBAL:
 	case AGGR_THREAD:
 	case AGGR_UNSET:
diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index f5b4ee7..4c53bae 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -88,9 +88,17 @@ static void aggr_printout(struct perf_stat_config *config,
 			config->csv_sep);
 			break;
 	case AGGR_NONE:
-		fprintf(config->output, "CPU%*d%s",
-			config->csv_output ? 0 : -4,
-			perf_evsel__cpus(evsel)->map[id], config->csv_sep);
+		if (evsel->percore) {
+			fprintf(config->output, "S%d-C%*d%s",
+				cpu_map__id_to_socket(id),
+				config->csv_output ? 0 : -5,
+				cpu_map__id_to_cpu(id), config->csv_sep);
+		} else {
+			fprintf(config->output, "CPU%*d%s ",
+				config->csv_output ? 0 : -5,
+				perf_evsel__cpus(evsel)->map[id],
+				config->csv_sep);
+		}
 		break;
 	case AGGR_THREAD:
 		fprintf(config->output, "%*s-%*d%s",
@@ -1103,6 +1111,30 @@ static void print_footer(struct perf_stat_config *config)
 			"the same PMU. Try reorganizing the group.\n");
 }
 
+static void print_percore(struct perf_stat_config *config,
+			  struct perf_evsel *counter, char *prefix)
+{
+	bool metric_only = config->metric_only;
+	FILE *output = config->output;
+	int s;
+	bool first = true;
+
+	if (!(config->aggr_map || config->aggr_get_id))
+		return;
+
+	for (s = 0; s < config->aggr_map->nr; s++) {
+		if (prefix && metric_only)
+			fprintf(output, "%s", prefix);
+
+		print_counter_aggrdata(config, counter, s,
+				       prefix, metric_only,
+				       &first);
+	}
+
+	if (metric_only)
+		fputc('\n', output);
+}
+
 void
 perf_evlist__print_counters(struct perf_evlist *evlist,
 			    struct perf_stat_config *config,
@@ -1153,7 +1185,10 @@ perf_evlist__print_counters(struct perf_evlist *evlist,
 			print_no_aggr_metric(config, evlist, prefix);
 		else {
 			evlist__for_each_entry(evlist, counter) {
-				print_counter(config, counter, prefix);
+				if (counter->percore)
+					print_percore(config, counter, prefix);
+				else
+					print_counter(config, counter, prefix);
 			}
 		}
 		break;
diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
index 2856cc9..c3115d9 100644
--- a/tools/perf/util/stat.c
+++ b/tools/perf/util/stat.c
@@ -277,9 +277,11 @@ process_counter_values(struct perf_stat_config *config, struct perf_evsel *evsel
 		if (!evsel->snapshot)
 			perf_evsel__compute_deltas(evsel, cpu, thread, count);
 		perf_counts_values__scale(count, config->scale, NULL);
-		if (config->aggr_mode == AGGR_NONE)
-			perf_stat__update_shadow_stats(evsel, count->val, cpu,
-						       &rt_stat);
+		if ((config->aggr_mode == AGGR_NONE) && (!evsel->percore)) {
+			perf_stat__update_shadow_stats(evsel, count->val,
+						       cpu, &rt_stat);
+		}
+
 		if (config->aggr_mode == AGGR_THREAD) {
 			if (config->stats)
 				perf_stat__update_shadow_stats(evsel,
-- 
2.7.4

[PATCH v4 4/4] perf test: Add a simple test for term 'percore'

[Jin Yao]

It's a simple test which just checks if parser works.

 v4:
 ---
 No change

 v3:
 ---
 No change

 v2:
 ---
 Change 'coresum' to 'percore'

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
---
 tools/perf/tests/parse-events.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/tools/perf/tests/parse-events.c b/tools/perf/tests/parse-events.c
index 4a69c07..e59a7bb 100644
--- a/tools/perf/tests/parse-events.c
+++ b/tools/perf/tests/parse-events.c
@@ -605,6 +605,14 @@ static int test__checkterms_simple(struct list_head *terms)
 	TEST_ASSERT_VAL("wrong val", term->val.num == 1);
 	TEST_ASSERT_VAL("wrong config", !strcmp(term->config, "umask"));
 
+	/* percore=1*/
+	term = list_entry(term->list.next, struct parse_events_term, list);
+	TEST_ASSERT_VAL("wrong type term",
+			term->type_term == PARSE_EVENTS__TERM_TYPE_PERCORE);
+	TEST_ASSERT_VAL("wrong type val",
+			term->type_val == PARSE_EVENTS__TERM_TYPE_NUM);
+	TEST_ASSERT_VAL("wrong val", term->val.num == 1);
+
 	return 0;
 }
 
@@ -1734,7 +1742,7 @@ struct terms_test {
 
 static struct terms_test test__terms[] = {
 	[0] = {
-		.str   = "config=10,config1,config2=3,umask=1",
+		.str   = "config=10,config1,config2=3,umask=1,percore=1",
 		.check = test__checkterms_simple,
 	},
 };
-- 
2.7.4

Re: [PATCH v4 0/4] perf: Support a new 'percore' event qualifier

[Jiri Olsa]

On Fri, Apr 12, 2019 at 09:59:46PM +0800, Jin Yao wrote:
> The 'percore' event qualifier which sums up the event counts for both
> hardware threads in a core. For example,
> 
> perf stat -e cpu/event=0,umask=0x3,percore=1/,cpu/event=0,umask=0x3/
> 
> In this example, we count the event 'ref-cycles' per-core and per-CPU in
> one perf stat command-line.
> 
> We can already support per-core counting with --per-core, but it's
> often useful to do this together with other metrics that are collected
> per CPU (per hardware thread). So this patch series supports this
> per-core counting on a event level.
> 
>  v4:
>  ---
>  Add percore qualifier to documantation.
>  Rebase to latest perf/core branch

Acked-by: Jiri Olsa <jolsa@kernel.org>

thanks,
jirka

> 
>  v3:
>  ---
>  Simplify the patch: "perf: Add a 'percore' event qualifier"
>  Other patches don't have changes.
> 
>  v2:
>  ---
>  1. Change 'coresum' to 'percore'.
>  2. Move the aggregate counts printing to a seperate patch.
> 
> Jin Yao (4):
>   perf: Add a 'percore' event qualifier
>   perf stat: Factor out aggregate counts printing
>   perf stat: Support 'percore' event qualifier
>   perf test: Add a simple test for term 'percore'
> 
>  tools/perf/Documentation/perf-list.txt |  12 ++++
>  tools/perf/Documentation/perf-stat.txt |   4 ++
>  tools/perf/builtin-stat.c              |  21 +++++++
>  tools/perf/tests/parse-events.c        |  10 ++-
>  tools/perf/util/evsel.c                |   2 +
>  tools/perf/util/evsel.h                |   3 +
>  tools/perf/util/parse-events.c         |  27 +++++++++
>  tools/perf/util/parse-events.h         |   1 +
>  tools/perf/util/parse-events.l         |   1 +
>  tools/perf/util/stat-display.c         | 107 ++++++++++++++++++++++++---------
>  tools/perf/util/stat.c                 |   8 ++-
>  11 files changed, 163 insertions(+), 33 deletions(-)
> 
> -- 
> 2.7.4
> 

Re: [PATCH v4 3/4] perf stat: Support 'percore' event qualifier

[Ravi Bangoria]

On 4/12/19 7:29 PM, Jin Yao wrote:
> diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
> index 39c05f8..1e312c2 100644
> --- a/tools/perf/Documentation/perf-stat.txt
> +++ b/tools/perf/Documentation/perf-stat.txt
> @@ -43,6 +43,10 @@ report::
>  	  param1 and param2 are defined as formats for the PMU in
>  	  /sys/bus/event_source/devices/<pmu>/format/*
>  
> +	  'percore' is a event qualifier that sums up the event counts for both
> +	  hardware threads in a core.

s/both/all/  :

  $ lscpu | grep Thread
  Thread(s) per core:    4


Apart from that, for the series:
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>

Re: [PATCH v4 3/4] perf stat: Support 'percore' event qualifier

[Arnaldo Carvalho de Melo]

Em Mon, Apr 15, 2019 at 08:09:09PM +0530, Ravi Bangoria escreveu:
> 
> 
> On 4/12/19 7:29 PM, Jin Yao wrote:
> > diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
> > index 39c05f8..1e312c2 100644
> > --- a/tools/perf/Documentation/perf-stat.txt
> > +++ b/tools/perf/Documentation/perf-stat.txt
> > @@ -43,6 +43,10 @@ report::
> >  	  param1 and param2 are defined as formats for the PMU in
> >  	  /sys/bus/event_source/devices/<pmu>/format/*
> >  
> > +	  'percore' is a event qualifier that sums up the event counts for both
> > +	  hardware threads in a core.
> 
> s/both/all/  :
> 
>   $ lscpu | grep Thread
>   Thread(s) per core:    4
> 
> 
> Apart from that, for the series:
> Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>

Yeah I had fixed up that already, fixed up the patch description this
time around, "both" was being used there as well, changed to "all".

Thanks, added your Tested-by tag, appreciated.

- Arnaldo

Re: [PATCH v4 3/4] perf stat: Support 'percore' event qualifier

[Jin, Yao]

On 4/16/2019 3:26 AM, Arnaldo Carvalho de Melo wrote:
> Em Mon, Apr 15, 2019 at 08:09:09PM +0530, Ravi Bangoria escreveu:
>>
>>
>> On 4/12/19 7:29 PM, Jin Yao wrote:
>>> diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
>>> index 39c05f8..1e312c2 100644
>>> --- a/tools/perf/Documentation/perf-stat.txt
>>> +++ b/tools/perf/Documentation/perf-stat.txt
>>> @@ -43,6 +43,10 @@ report::
>>>   	  param1 and param2 are defined as formats for the PMU in
>>>   	  /sys/bus/event_source/devices/<pmu>/format/*
>>>   
>>> +	  'percore' is a event qualifier that sums up the event counts for both
>>> +	  hardware threads in a core.
>>
>> s/both/all/  :
>>
>>    $ lscpu | grep Thread
>>    Thread(s) per core:    4
>>
>>
>> Apart from that, for the series:
>> Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
> 
> Yeah I had fixed up that already, fixed up the patch description this
> time around, "both" was being used there as well, changed to "all".
> 
> Thanks, added your Tested-by tag, appreciated.
> 
> - Arnaldo
> 

Thanks Ravi, Thanks Arnaldo!

Re: [PATCH v4 0/4] perf: Support a new 'percore' event qualifier

[Arnaldo Carvalho de Melo]

Em Fri, Apr 12, 2019 at 09:59:46PM +0800, Jin Yao escreveu:
> The 'percore' event qualifier which sums up the event counts for both
> hardware threads in a core. For example,
> 
> perf stat -e cpu/event=0,umask=0x3,percore=1/,cpu/event=0,umask=0x3/
> 
> In this example, we count the event 'ref-cycles' per-core and per-CPU in
> one perf stat command-line.
> 
> We can already support per-core counting with --per-core, but it's
> often useful to do this together with other metrics that are collected
> per CPU (per hardware thread). So this patch series supports this
> per-core counting on a event level.
> 
>  v4:
>  ---
>  Add percore qualifier to documantation.
>  Rebase to latest perf/core branch

Thanks, applied.

- Arnaldo

[tip:perf/core] perf tools: Add a 'percore' event qualifier

[tip-bot for Jin Yao]

Commit-ID:  064b4e82aa1633c27c383cc686b87ced57e072d1
Gitweb:     https://git.kernel.org/tip/064b4e82aa1633c27c383cc686b87ced57e072d1
Author:     Jin Yao <yao.jin@linux.intel.com>
AuthorDate: Fri, 12 Apr 2019 21:59:47 +0800
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Thu, 16 May 2019 14:17:24 -0300

perf tools: Add a 'percore' event qualifier

Add a 'percore' event qualifier, like cpu/event=0,umask=0x3,percore=1/,
that sums up the event counts for both hardware threads in a core.

We can already do this with --per-core, but it's often useful to do
this together with other metrics that are collected per hardware thread.
So we need to support this per-core counting on a event level.

This can be implemented in only the user tool, no kernel support needed.

 v4:
 ---
 1. Add Arnaldo's patch which updates the documentation for
    this new qualifier.
 2. Rebase to latest perf/core branch

 v3:
 ---
 Simplify the code according to Jiri's comments.
 Before:
   "return term->val.percore ? true : false;"
 Now:
   "return term->val.percore;"

 v2:
 ---
 Change the qualifier name from 'coresum' to 'percore' according to
 comments from Jiri and Andi.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1555077590-27664-2-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/Documentation/perf-list.txt | 12 ++++++++++++
 tools/perf/util/evsel.c                |  2 ++
 tools/perf/util/evsel.h                |  3 +++
 tools/perf/util/parse-events.c         | 27 +++++++++++++++++++++++++++
 tools/perf/util/parse-events.h         |  1 +
 tools/perf/util/parse-events.l         |  1 +
 6 files changed, 46 insertions(+)

diff --git a/tools/perf/Documentation/perf-list.txt b/tools/perf/Documentation/perf-list.txt
index 138fb6e94b3c..18ed1b0fceb3 100644
--- a/tools/perf/Documentation/perf-list.txt
+++ b/tools/perf/Documentation/perf-list.txt
@@ -199,6 +199,18 @@ also be supplied. For example:
 
   perf stat -C 0 -e 'hv_gpci/dtbp_ptitc,phys_processor_idx=0x2/' ...
 
+EVENT QUALIFIERS:
+
+It is also possible to add extra qualifiers to an event:
+
+percore:
+
+Sums up the event counts for all hardware threads in a core, e.g.:
+
+
+  perf stat -e cpu/event=0,umask=0x3,percore=1/
+
+
 EVENT GROUPS
 ------------
 
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index a10cf4cde920..a6f572a40deb 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -813,6 +813,8 @@ static void apply_config_terms(struct perf_evsel *evsel,
 			break;
 		case PERF_EVSEL__CONFIG_TERM_DRV_CFG:
 			break;
+		case PERF_EVSEL__CONFIG_TERM_PERCORE:
+			break;
 		default:
 			break;
 		}
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 6d190cbf1070..cad54e8ba522 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -50,6 +50,7 @@ enum term_type {
 	PERF_EVSEL__CONFIG_TERM_OVERWRITE,
 	PERF_EVSEL__CONFIG_TERM_DRV_CFG,
 	PERF_EVSEL__CONFIG_TERM_BRANCH,
+	PERF_EVSEL__CONFIG_TERM_PERCORE,
 };
 
 struct perf_evsel_config_term {
@@ -67,6 +68,7 @@ struct perf_evsel_config_term {
 		bool	overwrite;
 		char	*branch;
 		unsigned long max_events;
+		bool	percore;
 	} val;
 	bool weak;
 };
@@ -158,6 +160,7 @@ struct perf_evsel {
 	struct perf_evsel	**metric_events;
 	bool			collect_stat;
 	bool			weak_group;
+	bool			percore;
 	const char		*pmu_name;
 	struct {
 		perf_evsel__sb_cb_t	*cb;
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 4432bfe039fd..cf0b9b81c5aa 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -950,6 +950,7 @@ static const char *config_term_names[__PARSE_EVENTS__TERM_TYPE_NR] = {
 	[PARSE_EVENTS__TERM_TYPE_OVERWRITE]		= "overwrite",
 	[PARSE_EVENTS__TERM_TYPE_NOOVERWRITE]		= "no-overwrite",
 	[PARSE_EVENTS__TERM_TYPE_DRV_CFG]		= "driver-config",
+	[PARSE_EVENTS__TERM_TYPE_PERCORE]		= "percore",
 };
 
 static bool config_term_shrinked;
@@ -970,6 +971,7 @@ config_term_avail(int term_type, struct parse_events_error *err)
 	case PARSE_EVENTS__TERM_TYPE_CONFIG2:
 	case PARSE_EVENTS__TERM_TYPE_NAME:
 	case PARSE_EVENTS__TERM_TYPE_SAMPLE_PERIOD:
+	case PARSE_EVENTS__TERM_TYPE_PERCORE:
 		return true;
 	default:
 		if (!err)
@@ -1061,6 +1063,14 @@ do {									   \
 	case PARSE_EVENTS__TERM_TYPE_MAX_EVENTS:
 		CHECK_TYPE_VAL(NUM);
 		break;
+	case PARSE_EVENTS__TERM_TYPE_PERCORE:
+		CHECK_TYPE_VAL(NUM);
+		if ((unsigned int)term->val.num > 1) {
+			err->str = strdup("expected 0 or 1");
+			err->idx = term->err_val;
+			return -EINVAL;
+		}
+		break;
 	default:
 		err->str = strdup("unknown term");
 		err->idx = term->err_term;
@@ -1199,6 +1209,10 @@ do {								\
 		case PARSE_EVENTS__TERM_TYPE_DRV_CFG:
 			ADD_CONFIG_TERM(DRV_CFG, drv_cfg, term->val.str);
 			break;
+		case PARSE_EVENTS__TERM_TYPE_PERCORE:
+			ADD_CONFIG_TERM(PERCORE, percore,
+					term->val.num ? true : false);
+			break;
 		default:
 			break;
 		}
@@ -1260,6 +1274,18 @@ int parse_events_add_tool(struct parse_events_state *parse_state,
 	return add_event_tool(list, &parse_state->idx, tool_event);
 }
 
+static bool config_term_percore(struct list_head *config_terms)
+{
+	struct perf_evsel_config_term *term;
+
+	list_for_each_entry(term, config_terms, list) {
+		if (term->type == PERF_EVSEL__CONFIG_TERM_PERCORE)
+			return term->val.percore;
+	}
+
+	return false;
+}
+
 int parse_events_add_pmu(struct parse_events_state *parse_state,
 			 struct list_head *list, char *name,
 			 struct list_head *head_config,
@@ -1333,6 +1359,7 @@ int parse_events_add_pmu(struct parse_events_state *parse_state,
 		evsel->metric_name = info.metric_name;
 		evsel->pmu_name = name;
 		evsel->use_uncore_alias = use_uncore_alias;
+		evsel->percore = config_term_percore(&evsel->config_terms);
 	}
 
 	return evsel ? 0 : -ENOMEM;
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index a052cd6ac63e..f7139e1a2fd3 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -75,6 +75,7 @@ enum {
 	PARSE_EVENTS__TERM_TYPE_NOOVERWRITE,
 	PARSE_EVENTS__TERM_TYPE_OVERWRITE,
 	PARSE_EVENTS__TERM_TYPE_DRV_CFG,
+	PARSE_EVENTS__TERM_TYPE_PERCORE,
 	__PARSE_EVENTS__TERM_TYPE_NR,
 };
 
diff --git a/tools/perf/util/parse-events.l b/tools/perf/util/parse-events.l
index c54bfe88626c..ca6098874fe2 100644
--- a/tools/perf/util/parse-events.l
+++ b/tools/perf/util/parse-events.l
@@ -283,6 +283,7 @@ inherit			{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_INHERIT); }
 no-inherit		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_NOINHERIT); }
 overwrite		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_OVERWRITE); }
 no-overwrite		{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_NOOVERWRITE); }
+percore			{ return term(yyscanner, PARSE_EVENTS__TERM_TYPE_PERCORE); }
 ,			{ return ','; }
 "/"			{ BEGIN(INITIAL); return '/'; }
 {name_minus}		{ return str(yyscanner, PE_NAME); }

[tip:perf/core] perf stat: Factor out aggregate counts printing

[tip-bot for Jin Yao]

Commit-ID:  40480a8136700d678dc07222c4d7287c89d0c04d
Gitweb:     https://git.kernel.org/tip/40480a8136700d678dc07222c4d7287c89d0c04d
Author:     Jin Yao <yao.jin@linux.intel.com>
AuthorDate: Fri, 12 Apr 2019 21:59:48 +0800
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Thu, 16 May 2019 14:17:24 -0300

perf stat: Factor out aggregate counts printing

Move the aggregate counts printing to a new function
print_counter_aggrdata, which will be used in following patches.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1555077590-27664-3-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/stat-display.c | 64 +++++++++++++++++++++++++-----------------
 1 file changed, 39 insertions(+), 25 deletions(-)

diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index 3324f23c7efc..f5b4ee79568c 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -594,6 +594,41 @@ static void aggr_cb(struct perf_stat_config *config,
 	}
 }
 
+static void print_counter_aggrdata(struct perf_stat_config *config,
+				   struct perf_evsel *counter, int s,
+				   char *prefix, bool metric_only,
+				   bool *first)
+{
+	struct aggr_data ad;
+	FILE *output = config->output;
+	u64 ena, run, val;
+	int id, nr;
+	double uval;
+
+	ad.id = id = config->aggr_map->map[s];
+	ad.val = ad.ena = ad.run = 0;
+	ad.nr = 0;
+	if (!collect_data(config, counter, aggr_cb, &ad))
+		return;
+
+	nr = ad.nr;
+	ena = ad.ena;
+	run = ad.run;
+	val = ad.val;
+	if (*first && metric_only) {
+		*first = false;
+		aggr_printout(config, counter, id, nr);
+	}
+	if (prefix && !metric_only)
+		fprintf(output, "%s", prefix);
+
+	uval = val * counter->scale;
+	printout(config, id, nr, counter, uval, prefix,
+		 run, ena, 1.0, &rt_stat);
+	if (!metric_only)
+		fputc('\n', output);
+}
+
 static void print_aggr(struct perf_stat_config *config,
 		       struct perf_evlist *evlist,
 		       char *prefix)
@@ -601,9 +636,7 @@ static void print_aggr(struct perf_stat_config *config,
 	bool metric_only = config->metric_only;
 	FILE *output = config->output;
 	struct perf_evsel *counter;
-	int s, id, nr;
-	double uval;
-	u64 ena, run, val;
+	int s;
 	bool first;
 
 	if (!(config->aggr_map || config->aggr_get_id))
@@ -616,33 +649,14 @@ static void print_aggr(struct perf_stat_config *config,
 	 * Without each counter has its own line.
 	 */
 	for (s = 0; s < config->aggr_map->nr; s++) {
-		struct aggr_data ad;
 		if (prefix && metric_only)
 			fprintf(output, "%s", prefix);
 
-		ad.id = id = config->aggr_map->map[s];
 		first = true;
 		evlist__for_each_entry(evlist, counter) {
-			ad.val = ad.ena = ad.run = 0;
-			ad.nr = 0;
-			if (!collect_data(config, counter, aggr_cb, &ad))
-				continue;
-			nr = ad.nr;
-			ena = ad.ena;
-			run = ad.run;
-			val = ad.val;
-			if (first && metric_only) {
-				first = false;
-				aggr_printout(config, counter, id, nr);
-			}
-			if (prefix && !metric_only)
-				fprintf(output, "%s", prefix);
-
-			uval = val * counter->scale;
-			printout(config, id, nr, counter, uval, prefix,
-				 run, ena, 1.0, &rt_stat);
-			if (!metric_only)
-				fputc('\n', output);
+			print_counter_aggrdata(config, counter, s,
+					       prefix, metric_only,
+					       &first);
 		}
 		if (metric_only)
 			fputc('\n', output);

[tip:perf/core] perf stat: Support 'percore' event qualifier

[tip-bot for Jin Yao]

Commit-ID:  4fc4d8dfa056dfd48afe73b9ea3b7570ceb80b9c
Gitweb:     https://git.kernel.org/tip/4fc4d8dfa056dfd48afe73b9ea3b7570ceb80b9c
Author:     Jin Yao <yao.jin@linux.intel.com>
AuthorDate: Fri, 12 Apr 2019 21:59:49 +0800
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Thu, 16 May 2019 14:17:24 -0300

perf stat: Support 'percore' event qualifier

With this patch, we can use the 'percore' event qualifier in perf-stat.

  root@skl:/tmp# perf stat -e cpu/event=0,umask=0x3,percore=1/,cpu/event=0,umask=0x3/ -a -A -I1000
    1.000773050 S0-C0   98,352,832 cpu/event=0,umask=0x3,percore=1/  (50.01%)
    1.000773050 S0-C1  103,763,057 cpu/event=0,umask=0x3,percore=1/  (50.02%)
    1.000773050 S0-C2  196,776,995 cpu/event=0,umask=0x3,percore=1/  (50.02%)
    1.000773050 S0-C3  176,493,779 cpu/event=0,umask=0x3,percore=1/  (50.02%)
    1.000773050 CPU0    47,699,641 cpu/event=0,umask=0x3/            (50.02%)
    1.000773050 CPU1    49,052,451 cpu/event=0,umask=0x3/            (49.98%)
    1.000773050 CPU2   102,771,422 cpu/event=0,umask=0x3/            (49.98%)
    1.000773050 CPU3   100,784,662 cpu/event=0,umask=0x3/            (49.98%)
    1.000773050 CPU4    43,171,342 cpu/event=0,umask=0x3/            (49.98%)
    1.000773050 CPU5    54,152,158 cpu/event=0,umask=0x3/            (49.98%)
    1.000773050 CPU6    93,618,410 cpu/event=0,umask=0x3/            (49.98%)
    1.000773050 CPU7    74,477,589 cpu/event=0,umask=0x3/            (49.99%)

In this example, we count the event 'ref-cycles' per-core and per-CPU in
one perf stat command-line. From the output, we can see:

  S0-C0 = CPU0 + CPU4
  S0-C1 = CPU1 + CPU5
  S0-C2 = CPU2 + CPU6
  S0-C3 = CPU3 + CPU7

So the result is expected (tiny difference is ignored).

Note that, the 'percore' event qualifier needs to use with option '-A'.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1555077590-27664-4-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/Documentation/perf-stat.txt |  4 ++++
 tools/perf/builtin-stat.c              | 21 +++++++++++++++++
 tools/perf/util/stat-display.c         | 43 ++++++++++++++++++++++++++++++----
 tools/perf/util/stat.c                 |  8 ++++---
 4 files changed, 69 insertions(+), 7 deletions(-)

diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index 39c05f89104e..1e312c2672e4 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -43,6 +43,10 @@ report::
 	  param1 and param2 are defined as formats for the PMU in
 	  /sys/bus/event_source/devices/<pmu>/format/*
 
+	  'percore' is a event qualifier that sums up the event counts for both
+	  hardware threads in a core. For example:
+	  perf stat -A -a -e cpu/event,percore=1/,otherevent ...
+
 	- a symbolically formed event like 'pmu/config=M,config1=N,config2=K/'
 	  where M, N, K are numbers (in decimal, hex, octal format).
 	  Acceptable values for each of 'config', 'config1' and 'config2'
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index a3c060878faa..24b8e690fb69 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -847,6 +847,18 @@ static int perf_stat__get_core_cached(struct perf_stat_config *config,
 	return perf_stat__get_aggr(config, perf_stat__get_core, map, idx);
 }
 
+static bool term_percore_set(void)
+{
+	struct perf_evsel *counter;
+
+	evlist__for_each_entry(evsel_list, counter) {
+		if (counter->percore)
+			return true;
+	}
+
+	return false;
+}
+
 static int perf_stat_init_aggr_mode(void)
 {
 	int nr;
@@ -867,6 +879,15 @@ static int perf_stat_init_aggr_mode(void)
 		stat_config.aggr_get_id = perf_stat__get_core_cached;
 		break;
 	case AGGR_NONE:
+		if (term_percore_set()) {
+			if (cpu_map__build_core_map(evsel_list->cpus,
+						    &stat_config.aggr_map)) {
+				perror("cannot build core map");
+				return -1;
+			}
+			stat_config.aggr_get_id = perf_stat__get_core_cached;
+		}
+		break;
 	case AGGR_GLOBAL:
 	case AGGR_THREAD:
 	case AGGR_UNSET:
diff --git a/tools/perf/util/stat-display.c b/tools/perf/util/stat-display.c
index f5b4ee79568c..4c53bae5644b 100644
--- a/tools/perf/util/stat-display.c
+++ b/tools/perf/util/stat-display.c
@@ -88,9 +88,17 @@ static void aggr_printout(struct perf_stat_config *config,
 			config->csv_sep);
 			break;
 	case AGGR_NONE:
-		fprintf(config->output, "CPU%*d%s",
-			config->csv_output ? 0 : -4,
-			perf_evsel__cpus(evsel)->map[id], config->csv_sep);
+		if (evsel->percore) {
+			fprintf(config->output, "S%d-C%*d%s",
+				cpu_map__id_to_socket(id),
+				config->csv_output ? 0 : -5,
+				cpu_map__id_to_cpu(id), config->csv_sep);
+		} else {
+			fprintf(config->output, "CPU%*d%s ",
+				config->csv_output ? 0 : -5,
+				perf_evsel__cpus(evsel)->map[id],
+				config->csv_sep);
+		}
 		break;
 	case AGGR_THREAD:
 		fprintf(config->output, "%*s-%*d%s",
@@ -1103,6 +1111,30 @@ static void print_footer(struct perf_stat_config *config)
 			"the same PMU. Try reorganizing the group.\n");
 }
 
+static void print_percore(struct perf_stat_config *config,
+			  struct perf_evsel *counter, char *prefix)
+{
+	bool metric_only = config->metric_only;
+	FILE *output = config->output;
+	int s;
+	bool first = true;
+
+	if (!(config->aggr_map || config->aggr_get_id))
+		return;
+
+	for (s = 0; s < config->aggr_map->nr; s++) {
+		if (prefix && metric_only)
+			fprintf(output, "%s", prefix);
+
+		print_counter_aggrdata(config, counter, s,
+				       prefix, metric_only,
+				       &first);
+	}
+
+	if (metric_only)
+		fputc('\n', output);
+}
+
 void
 perf_evlist__print_counters(struct perf_evlist *evlist,
 			    struct perf_stat_config *config,
@@ -1153,7 +1185,10 @@ perf_evlist__print_counters(struct perf_evlist *evlist,
 			print_no_aggr_metric(config, evlist, prefix);
 		else {
 			evlist__for_each_entry(evlist, counter) {
-				print_counter(config, counter, prefix);
+				if (counter->percore)
+					print_percore(config, counter, prefix);
+				else
+					print_counter(config, counter, prefix);
 			}
 		}
 		break;
diff --git a/tools/perf/util/stat.c b/tools/perf/util/stat.c
index 2856cc9d5a31..c3115d939b0b 100644
--- a/tools/perf/util/stat.c
+++ b/tools/perf/util/stat.c
@@ -277,9 +277,11 @@ process_counter_values(struct perf_stat_config *config, struct perf_evsel *evsel
 		if (!evsel->snapshot)
 			perf_evsel__compute_deltas(evsel, cpu, thread, count);
 		perf_counts_values__scale(count, config->scale, NULL);
-		if (config->aggr_mode == AGGR_NONE)
-			perf_stat__update_shadow_stats(evsel, count->val, cpu,
-						       &rt_stat);
+		if ((config->aggr_mode == AGGR_NONE) && (!evsel->percore)) {
+			perf_stat__update_shadow_stats(evsel, count->val,
+						       cpu, &rt_stat);
+		}
+
 		if (config->aggr_mode == AGGR_THREAD) {
 			if (config->stats)
 				perf_stat__update_shadow_stats(evsel,