perf stat: Add mean, min, max and last --tpebs-mode options

Add command line configuration option for how retirement latency
events are combined.

The default "mean" gives the average of retirement latency.

"min" or "max" give the smallest or largest retirment latency times
respectively.

"last" uses the last retirment latency sample's time.

Committer notes:

Enclose parse_tpebs_mode() under HAVE_ARCH_X86_64_SUPPORT to match the
ifdef block where it is used, fixing the build in systems like:

  20     5.60 debian:experimental-x-mips    : FAIL gcc version 14.2.0 (Debian 14.2.0-1)
    builtin-stat.c:2330:12: error: 'parse_tpebs_mode' defined but not used [-Werror=unused-function]
     2330 | static int parse_tpebs_mode(const struct option *opt, const char *str,
          |            ^~~~~~~~~~~~~~~~

Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: Weilin Wang <weilin.wang@intel.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexandre Torgue <alexandre.torgue@foss.st.com>
Cc: Andreas Färber <afaerber@suse.de>
Cc: Caleb Biggers <caleb.biggers@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Maxime Coquelin <mcoquelin.stm32@gmail.com>
Cc: Perry Taylor <perry.taylor@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Falcon <thomas.falcon@intel.com>
Link: https://lore.kernel.org/r/20250414174134.3095492-15-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/perf/Documentation/perf-stat.txt

@@ -506,6 +506,13 @@ this option is not set. The TPEBS hardware feature starts from Intel Granite
 Rapids microarchitecture. This option only exists in X86_64 and is meaningful on
 Intel platforms with TPEBS feature.
 
+--tpebs-mode=[mean|min|max|last]::
+Set how retirement latency events have their sample times
+combined. The default "mean" gives the average of retirement
+latency. "min" or "max" give the smallest or largest retirment latency
+times respectively. "last" uses the last retirment latency sample's
+time.
+
 --td-level::
 Print the top-down statistics that equal the input level. It allows
 users to print the interested top-down metrics level instead of the

tools/perf/builtin-stat.c

@@ -2327,6 +2327,32 @@ static void setup_system_wide(int forks)
 	}
 }
 
+#ifdef HAVE_ARCH_X86_64_SUPPORT
+static int parse_tpebs_mode(const struct option *opt, const char *str,
+			    int unset __maybe_unused)
+{
+	enum tpebs_mode *mode = opt->value;
+
+	if (!strcasecmp("mean", str)) {
+		*mode = TPEBS_MODE__MEAN;
+		return 0;
+	}
+	if (!strcasecmp("min", str)) {
+		*mode = TPEBS_MODE__MIN;
+		return 0;
+	}
+	if (!strcasecmp("max", str)) {
+		*mode = TPEBS_MODE__MAX;
+		return 0;
+	}
+	if (!strcasecmp("last", str)) {
+		*mode = TPEBS_MODE__LAST;
+		return 0;
+	}
+	return -1;
+}
+#endif // HAVE_ARCH_X86_64_SUPPORT
+
 int cmd_stat(int argc, const char **argv)
 {
 	struct opt_aggr_mode opt_mode = {};
@@ -2431,6 +2457,9 @@ int cmd_stat(int argc, const char **argv)
 #ifdef HAVE_ARCH_X86_64_SUPPORT
 		OPT_BOOLEAN(0, "record-tpebs", &tpebs_recording,
 			"enable recording for tpebs when retire_latency required"),
+		OPT_CALLBACK(0, "tpebs-mode", &tpebs_mode, "tpebs-mode",
+			"Mode of TPEBS recording: mean, min or max",
+			parse_tpebs_mode),
 #endif
 		OPT_UINTEGER(0, "td-level", &stat_config.topdown_level,
 			"Set the metrics level for the top-down statistics (0: max level)"),

tools/perf/util/intel-tpebs.c

@@ -31,6 +31,7 @@
 #define PERF_DATA		"-"
 
 bool tpebs_recording;
+enum tpebs_mode tpebs_mode;
 static LIST_HEAD(tpebs_results);
 static pthread_t tpebs_reader_thread;
 static struct child_process tpebs_cmd;
@@ -45,6 +46,8 @@ struct tpebs_retire_lat {
 	char *event;
 	/** @stats: Recorded retirement latency stats. */
 	struct stats stats;
+	/** @last: Last retirement latency read. */
+	uint64_t last;
 	/* Has the event been sent to perf record? */
 	bool started;
 };
@@ -142,6 +145,7 @@ static int process_sample_event(const struct perf_tool *tool __maybe_unused,
 	 * latency value will be used. Save the number of samples and the sum of
 	 * retire latency value for each event.
 	 */
+	t->last = sample->retire_lat;
 	update_stats(&t->stats, sample->retire_lat);
 	mutex_unlock(tpebs_mtx_get());
 	return 0;
@@ -517,7 +521,21 @@ int evsel__tpebs_read(struct evsel *evsel, int cpu_map_idx, int thread)
 			return ret;
 		mutex_lock(tpebs_mtx_get());
 	}
-	val = rint(t->stats.mean);
+	switch (tpebs_mode) {
+	case TPEBS_MODE__MIN:
+		val = rint(t->stats.min);
+		break;
+	case TPEBS_MODE__MAX:
+		val = rint(t->stats.max);
+		break;
+	case TPEBS_MODE__LAST:
+		val = t->last;
+		break;
+	default:
+	case TPEBS_MODE__MEAN:
+		val = rint(t->stats.mean);
+		break;
+	}
 	mutex_unlock(tpebs_mtx_get());
 
 	if (old_count) {

tools/perf/util/intel-tpebs.h

@@ -8,7 +8,15 @@
 struct evlist;
 struct evsel;
 
+enum tpebs_mode {
+	TPEBS_MODE__MEAN,
+	TPEBS_MODE__MIN,
+	TPEBS_MODE__MAX,
+	TPEBS_MODE__LAST,
+};
+
 extern bool tpebs_recording;
+extern enum tpebs_mode tpebs_mode;
 
 int evsel__tpebs_open(struct evsel *evsel);
 void evsel__tpebs_close(struct evsel *evsel);